THESE - univ-perp.fr

UNIVERSITE LOUIS PASTEUR CENTRE NATIONAL

STRASBOURG I DE LA RECHERCHE SCIENTIFIQUE Centre de Géochimie de la Surface

UFR DES SCIENCES DE LA VIE ET DE LA TERRE INSTITUT DE GEOLOGIE

THESE

présentée pour obtenir le titre de

Docteur de l'Université Louis Pasteur de Strasbourg Label "Doctorat Européen"

mention : géochimie

par

Wolfgang LUDWIG

CONTINENTAL EROSION AND RIVER TRANSPORT OF ORGANIC CARBON TO THE WORLD'S OCEANS

Soutenue publiquement le 6 Septembre 1996 devant la commission d'examen:

MM. Gerd ESSER, Rapporteur externe Joël HUMBERT, Rapporteur interne Stephan KEMPE, Président du jury Michel MEYBECK, Membre invité Jean Luc PROBST, Directeur de thèse Des WALLING, Rapporteur externe

I

ACKNOWLEDGEMENTS

REMERCIEMENTS

DANKSAGUNG

The scientific work that is presented on the following pages could not have been realized without the help of many persons. At least a few of them have to be mentioned here. The great number of persons who contributed to this work by supplying ideas or practical help makes it impossible for me to include all of them. This does not mean, of course, that those who are not mentioned by name in the following do not also merit my sincere acknowledgements.

First of all, I want to thank Jean-Luc Probst for all the efforts and discussions he dedicated to my studies. He was a very good director for me, and when I select this brief formulation to describe the important role he had in the realization of this work, it is because I think this is the best compliment one can make to his PhD director. Also his wife Anne should be mentioned here, who's office is just a few meters away from the office of Jean-Luc. While my discussions with Jean-Luc mainly focused on the global scale, i.e. how to find a few simple equations for a huge number of complex phenomenons, allowed me the work on the Strengbach catchment with Anne also to include a very local scale into my studies. This represents certainly an enrichment for the following pages.

I'm very grateful to Gerd Esser, Joël Humbert, Stefan Kempe, Michel Meybeck, and Des Walling for having accepted the task of the scientific examination of my thesis. It was probably impossible to assemble in a thesis jury a higher degree of competence and knowledge with respect to the different aspects that are treated in the manuscript. Having had this jury was a great honour for me. Stefan Kempe also has to be acknowledged here because he made all SCOPE/ UNEP river data available to me during the first three month of my thesis when I was still in Hamburg. He was already one of my teachers during my master studies in geology and geochemistry at the university of Hamburg, and his interesting lessons were one of the reasons that aroused my interest in the organic carbon transport by rivers.

Also many other persons supported my work by supplying data. Paul Reich sent me the global soil carbon map created at the Soil Conservation Service of the United States Department of Agriculture. This data set was a useful tool to improve the modelling for global DOC fluxes. James Miller from the Institute of Marine and Coastal Science at Rutgers University (New Jersey) supplied two global river routing files. Due to these files, the GEM-CO2 and GEM-Corg outputs could also be used to calculate the river carbon inputs to the oceans at a grid point scale. The Auckland Regional Authorities in New Zealand authorized me the utilization of the still unpublished carbon data for the Waikato River, and Rob Hart from the University of Natal in South Africa sent me additional data for the Orange River. Henri Etcheber (University of Bordeaux) supplied DOC and POC data for the Garonne River, Joe Boeglin (ORSTOM/ CGS) supplied some data for the upper Niger River, and Didier Orange (ORSTOM) sent me discharge data for the Ubangi River. To all of them I own my sincere acknowledgements. Their data were very helpful in the realisation of my studies.

Speaking about the persons who contributed data and active support to my work is naturally not possible without also to mention Philippe Amiotte-Suchet. He started in Strasbourg his PhD on the river fluxes of inorganic carbon two years before I began my thesis, and his experience and advice was a great help for me in developing my modelling approaches on the fluxes of organic carbon. Without

II

the data and results of Philippe, the last chapter of my manuscript, which represents a synthesis of the role of continental erosion in the global carbon cycle, could not have been written. Also Guy Munhoven must be mentioned here, who is actually finishing his PhD at the University of Liège in Belgium. During his one year stay at the CGS in Strasbourg he digitized the UNESCO runoff maps, which were a very useful tool for my chapter on the controlling factors of continental runoff (chapter III). I'll never forget the famous "amoeba-subroutines" he applied to find out the map projections. Since that time, artificial intelligence became a new meaning for me.

Finally, research activities need not only scientific but also financial support. I thank the French research and education ministry for giving me a three years PhD fellowship, and I'm very grateful to the Société de Sécours des Amis de la Science in Paris for the attribution of a nine-month grant. This grant was very helpful for me in the final phase of my PhD when I was writing down the manuscript. Thank you also very much, Helène Paquet, for support when I was applying for this grant.

Si je passe maintenant dans la langue française, c'est pour m'adresser aux gens qui travaillent au Centre de Géochimie de la Surface et à l'Institut de Géologie. En faisant le pas de l'Allemagne vers la France, leur chaleureux accueil et leur soutien actif pour les petits problèmes du quotidien m'ont été d'une grande aide après mon arrivé à Strasbourg. Maintenant, au bout de presque quatre ans, je peux dire que j'ai fait la connaissance d'un esprit d'ouverture et de convivialité qui me sont devenu très précieux. Grâce à cette expérience, la France c'est montrée pour moi de son meilleur côté et j'ai presque l'impression d'avoir gagné une deuxième patrie.

Merci d'abord à Alain Clément, notre informaticien à bord, pour une disponibilité quasi-permanente en ce qui concerne les problèmes informatiques. Inutile de dire que ses interventions dans mes batailles avec les ordinateurs de l'Institut étaient toujours d'une efficacité impressionnante. Un grand merci également à Marie-Camille Adloff pour son aide concernant les questions administratives. Je crois que, contrairement aux attributs qu'on fait souvent à mes compatriotes, j'étais souvent bien en retard pour lui donner quoi que ce soit. Néanmoins, je suis toujours tombé sur une gentillesse extraordinaire. Merci aussi à Gérard Krempp pour son aide avec les analyses de COD sur le Strengbach. Lors de ma thèse, mes interventions dans les laboratoires n'étaient pas très fréquentes, mais j'ai toujours bien aimé discuter avec lui devant l'analyseur de COD sur les petits secrets de la machine. Ma reconnaissance va également à Joe Boeglin, François Chabaux, Christine Destrigneville, Bertrand Fritz, Yvette Hartmeier, Daniel Jeannette, Anne-Marie Karpoff, Betty Kiefel, Jean-Claude Leprun, Louis Martinez, Daniel Million, Christine Mosser, Pascal Podwojewski, Claude Roquin, André Schaaf, Peter Stille, Yves Tardy, Daniel Tisserant et tous les autres chercheurs et techniciens qui ont montré un grand intérêt à mon travail et qui avaient toujours quelques mots sympathiques pour moi dans les couloirs de l'institut. Merci à tous.

Bien évidemment, parler de l'Institut n'est pas possible sans parler des nombreux chercheurs non-permanents qui y sont (ou étaient) de passage, soit pour une thèse, soit pour un post-doc ou pour un stage. Une grande partie de l'hospitalité et de la bonne ambiance qui règne entre les murs de la rue Blessig sont dues à leur présence. Je remercie d'abord Nathalie Fillion et Benoît Momboisse pour m'avoir supporté dans notre bureau commun. Je remercie ensuite Isabelle Bourasseau, Pascale Dejardin, Khadija Semhi et Frédéric Vitali qui étaient avec moi dans la même équipe "fin de rédaction". Grâce à eux, le moral n'est jamais descendu trop bas et le combat nocturne contre le papier (soutenu de quelques m3 de café) nous a certainement soudé un peu plus ensemble. Merci à Christopher Swezey pour ses nombreuses corrections de mon anglais (si le texte des pages suivantes est encore bourré des fautes, c'est parce qu'il est reparti aux Etats Unis depuis longtemps) et à Eric Delaitre, Magali Geiller, Amalia Jimenez, Arno Lassin, Marie-Claire Pierret, Laurent Richard et Aude Tricca qui m'ont beaucoup aidé pour la préparation de la soutenance. Un grand merci également à Yves Mongodin qui m'a sauvé en faisant le "shooting" de mes diapositives pour la soutenance. Et il y a encore Sophie Belin, Laurent Eisenlohr, Frédéric Gérard, Claude Hammecker, Naoko Kosaka, Laurent Pourcelot, Jean-Pierre Sizun, Mark Steinmann, Theo Toulkeridis, Lucia Valencia, Horst Zwingmann, .... A part tout leur soutien que j'ai eu dans le cadre de mon travail, j'ai surtout aussi passé des bons moments avec eux. Peut-être un jour j'aurai oublié la quantité de carbone organique qui s'en va du

III

bassin de l'Amazone. Mais je crois que je me rappellerai toujours ce qui était le blue monday au Café des Anges.

Da ist dann ja noch meine Muttersprache. Ich danke vor allem meinen Eltern für ihre tatkräftige Unterstützung, die sie mir in all meinen langen Studienjahren zukommen haben lassen. Es mag ihnen vielleicht nicht immer klar gewesen sein, warum man soviel Zeit, Geld und Energie investiert, nur um zu wissen, was jedes Jahr an Kohlestoff die Flüsse hinunterfliesst. Trotzdem haben sie immer zu mir gehalten. Auch meinem Onkel Helmut Wolbert möchte ich hiermit noch einmal für seine tat- und vor allem wortkräftige Unterstützung während meiner Doktorfeier danken.

Und schliesslich: Wenn es mir ein leichtes war, nach Frankreich zu gehen, um dort meine Doktorarbeit durchzuführen, so war dies nicht nur der Fall, weil heutzutage der Europagedanke in aller Köpfe ist. Danke Valérie, dass du mir brereits vor vielen Jahren dieses Land zugänglich gemacht hast. Und danke auch für vieles mehr.

V

TABLE OF CONTENTS

INTRODUCTION

English Version .................................................................................... 1 Version Française ................................................................................. 5

CHAPTER I THE STATE OF ART: CASE STUDIES ON RIVER FLUXES OF ORGANIC CARBON

1.1. Introduction ............................................................................................. 9 1.2. A Small Mountainous Watershed in Temperate Region: The Strengbach Catchment..................................................................................... 11

1.2.1. The May 1994 Storm Event ...................................................... 12 1.2.2. Permanent DOC Monitoring ..................................................... 15

1.3. A Medium Size Temperate River: The Garonne River Study................ 19 1.3.1. Hydrology.................................................................................. 19 1.3.2. Dissolved Organic Carbon ........................................................ 20 1.3.3. Particulate Organic Carbon and Total Suspended Solids.......... 21 1.3.4. Autochthonous Carbon.............................................................. 24

1.4. Major World Rivers ................................................................................ 25 1.4.1. Boreal Climates ......................................................................... 25

1.4.1.1. The Mackenzie River................................................... 25 1.4.2. Temperate Climates................................................................... 27

1.4.2.1. The St. Lawrence River ............................................... 27 1.4.2.2. The Waikato River....................................................... 31

1.4.3. Tropical Wet Climates............................................................... 34 1.4.3.1. The Orinoco River ....................................................... 34 1.4.3.2. The Paraná River ......................................................... 36 1.4.3.3. The Ubangi River (Zaire Basin) .................................. 38

1.4.4. Tropical Dry Climates ............................................................... 40 1.4.4.1. The Niger River ........................................................... 40 1.4.4.2. The Orange River......................................................... 44 1.4.4.3. The Indus River ........................................................... 47

1.5. Conclusions ............................................................................................. 49

VI

CHAPTER II METHODICAL ASPECTS: TOWARDS A QUANTIFICATION OF RIVER BASIN CHARACTERISTICS AND THEIR INFLUENCE ON THE FLUVIAL MATTER TRANSPORT TO THE OCEANS

2.1. Introduction............................................................................................. 53 2.2. Data and Methods ................................................................................... 54

2.2.1. River Basins .............................................................................. 54 2.2.2. Environmental Data Sets........................................................... 57

2.2.2.1. Hydroclimatic Parameters ........................................... 57 2.2.2.2. Biological Parameters.................................................. 57 2.2.2.3. Geomorphological and Lithological Parameters......... 57 2.2.2.4. Other Parameters ......................................................... 58

2.2.3. Empirical Modelling ................................................................. 58 2.3. Climatic Classification............................................................................ 64

2.3.1. The Holdridge Life Zone Classification ................................... 64 2.3.2. Climatic Classification and Climatic Variability of the River Basins ......................................................................................... 64

2.4. Seasonal Variability ................................................................................ 68 CHAPTER III CONTINENTAL RUNOFF AND ITS RELATION TO CLIMATE AND MORPHOLOGY


3.2.1. Runoff and Precipitation Data................................................... 74 3.2.2. Other Data Sets.......................................................................... 76 3.2.3. Statistics..................................................................................... 76

3.3. Validation of the UNESCO Runoff Map................................................ 78 3.3.1. Comparison with Literature Estimates...................................... 78 3.3.2. Determination of Regional and Global Figures for Continental Runoff ............................................................................... 79

3.4. Key Parameters for Continental Runoff ................................................. 82 3.4.1. Limiting Conditions .................................................................. 82 3.4.2. Potential Evapotranspiration ..................................................... 84 3.4.3. Precipitation .............................................................................. 86 3.4.4. Best Estimates ........................................................................... 88

3.5. Determination of the Controlling Factors for Continental Runoff ......... 90 3.5.1. Basin Averages.......................................................................... 90 3.5.2. Grid Point Averages .................................................................. 91

3.5.2.1. Non-Polar Climate Types ............................................ 91 3.5.2.2. Polar Climate Types .................................................... 96

3.5.3. Extrapolation to the Global Scale ............................................. 97 3.6. Comparison with other Models............................................................... 98

VII

3.6.1. Water Budget Models (Bucket Models).................................... 98 3.6.2. General Circulation Models ...................................................... 100

3.7. Conclusions ............................................................................................. 102 CHAPTER IV HOW TO PREDICT RIVER SEDIMENT DISCHARGES TO THE OCEANS

4.1. Introduction ............................................................................................. 105 4.2. Data and Methods.................................................................................... 106

4.2.1. Sediment Fluxes ........................................................................ 106 4.2.2. Other Data and Statistics ........................................................... 107

4.3. Previous Models for Sediment Yield Prediction..................................... 107 4.3.1. Relationships with Precipitation................................................ 108 4.3.2. Relationships with Climate........................................................ 109 4.3.3. Relationships with Basin Elevation and Morphology............... 110 4.3.4. Combined Parameter Models .................................................... 112

4.4. A New Modelling of the Climatic, Morphological and Lithological Control of River Sediment Yields ................................................................... 113

4.4.1. Identification of the Controlling Parameters ............................. 113 4.4.2. Climatic Particularities .............................................................. 115 4.4.3. Influence of Lithology............................................................... 120 4.4.4. Influence of General Basin Characteristics ............................... 121

4.4.4.1. Basin Area and Hypsometry ........................................ 122 4.4.4.2. Orogeny ....................................................................... 126 4.4.4.3. Land Use ...................................................................... 128

4.5. A Global Map of River Sediment Yields from the Continents............... 129 4.6. Conclusions ............................................................................................. 135

CHAPTER V PREDICTING THE OCEANIC INPUT OF ORGANIC CARBON BY CONTINENTAL EROSION: FROM THE BASIN SCALE TO THE GLOBAL SCALE

5.1. Introduction ............................................................................................. 137 5.2. Data and Methods.................................................................................... 138

5.2.1. River Fluxes of DOC and POC ................................................. 138 5.2.2. Environmental Data Sets and Empirical Modelling.................. 141

5.3. Factors Controlling DOC Fluxes ............................................................ 141 5.4. Factors Controlling POC Fluxes ............................................................. 144 5.5. Extrapolation to the Global Scale ........................................................... 147 5.6. Conclusions ............................................................................................. 153

VIII

CHAPTER VI ATMOSPHERIC CO2 CONSUMPTION AND RIVER CARBON INPUTS TO THE OCEANS BY CONTINENTAL EROSION: PRESENT-DAY FLUXES AND IMPLICATIONS FOR THE LAST GLACIAL MAXIMUM


6.2.1. Modelling of Inorganic Carbon Fluxes..................................... 156 6.2.2. Environmental Data Sets and Empirical Modelling ................. 157 6.2.3. River Routing Files ................................................................... 159 6.2.4. Determination of the LGM Boundary Conditions .................... 159

6.3. River Carbon Fluxes in the Present-Day Global Carbon Cycle ............. 160 6.3.1. Spatial Variability of the Atmospheric CO2 Consumption by Continental Erosion......................................................................... 160 6.3.2. Determining Local Inputs of Alkalinity, DOC, and POC to the Oceans ........................................................................................ 166 6.3.3. Fate of The River Carbon in the Oceans................................... 170 6.3.4. Influence of Climate on the Consumption of Atmospheric CO2 by Rock Weathering .................................................................... 172

6.4. Implications for the Atmospheric CO2 Consumption by Continental Erosion During the Last Glacial Maximum ................................ 176

6.4.1. Application of GEM-CO2 and GEM-Corg to LGM Conditions ............................................................................................ 176 6.4.2. Other Factors Related to Continental Erosion .......................... 180

6.5. Conclusions............................................................................................. 181 GENERAL CONCLUSIONS

English Version .................................................................................... 183 Version Française ................................................................................. 187

REFERENCES......................................................................................................... 193 LIST OF FIGURES ................................................................................................. A-I LIST OF TABLES................................................................................................... A-V

INTRODUCTION

English Version

From the very beginning, rivers played a vital role in man's history. The rivers Tigris/ Euphrates, and later also the Nile, provided the water and fertile alluvial lands which allowed the development of the first sophisticated cultures from a social and economical point of view. Later, rivers became also an important transport medium for the trade of goods with other peoples and countries. They formed the natural pathways to explore the unknown parts of the continents and guided peoples that were searching for new living spaces. At the same time, rivers were selected as the frontiers of countries, being thus in many cases also the places for wars and other conflicts. Modern civilisation broad along new utilisations, as for example the use of rivers for electrical power generation, or, being more and more problematic for the environment, their use as a world-wide canalisation system in order to get rid of the tremendous masses of waste which are produced in the growing metropoles.

See the rivers how they run. Changeless towards a changeless sea. These words of the English writer Charles Kingsley (1819-1875) may express the fascination that rivers exerted on many poets, musicians, or painters, who saw them as a symbol of the persistence of nature as against the transitoriness of life. It is the same persistence that attracts the interests of geochemists on rivers. In fact, the run of a river is not as changeless as it may seem for the observer standing on the river's bank. It is because of the permanent current of rivers that deep valleys have been cut into the landscape, or that big mountains which were formed in former geological times have been completely eroded. On a global scale, billions of tons of dissolved and particulate matter are taken off the continents and washed out to the oceans every year by rivers, which represent a non-negligible quantity for the global cycling of many elements and substances on Earth. For example, the riverine inputs of nutrients to the coastal zones are important in order to maintain the biological fertility in these zones. Moreover, the importance of river fluxes in global cycles even increased in as much as man has nowadays achieved the capacity to alter these fluxes. Cutting off the rain forest in the tropics is supposed to severely enhance erosion and river fluxes in these regions. At other places, artificial river damming retains much of the transported matter, and drastically decreases the fluxes. In many rivers which drain highly populated regions, the contributions of industrial and agricultural wastes surmount by far the contributions of the natural sources, with the possible consequences of catastrophic plankton blooms or increasing rates of fish diseases in the coastal seas.

The scientific goal of the here-presented thesis is a global and regional quantification of the amount of organic carbon that is discharged to the oceans every year by rivers in order to evaluate the role of this transport pathway within the global carbon cycle. Rising atmospheric CO2 concentrations mainly related to industrial fossil fuel combustion are expected to lead to significant global climatic changes during the coming decades. After 30 years of measurements in the atmosphere and in the oceans, the global atmospheric CO2 budget is still surprisingly uncertain. Via the vegetation cycle, the erosion of organic matter and its successive transport to the oceans by rivers is a sink of atmospheric CO2. Together with the atmospheric CO2 that is consumed by chemical rock weathering, this sink has been estimated to represent globally about 0.5 to 0.6 GtC (gigatons of carbon) per year, with about 35% that can be attributed to dissolved organic carbon, 20% to particulate organic carbon, and 45% to the consumption of atmospheric CO2 by rock weathering. Adding the carbon originating from lithological sources, the total amount of carbon which is discharged to the oceans by rivers may be as great as 1 GtC/yr. This is a non-negligible quantity in the present-day global carbon cycle. (Fig. 1).

1

Up to now, it was difficult to evaluate the role of the fluvial transport pathway in this context more in detail because of the lack of sufficient river data world-wide. The recent work of Amiotte-Suchet [1995] carried out at the Centre de Géochimie de la Surface (CGS) provided already a modelling for the prediction of the river fluxes of inorganic carbon, but still little is known about the fluxes of organic carbon. The purpose of this study is therefore to develop a modelling tool based upon the major controlling factors for the river fluxes of organic carbon which allows to respond to this data need. This approach concerns in about equal parts the river fluxes of dissolved and of particulate organic carbon. Because both fluxes are strongly coupled to the fluxes of water and of sediments, such an investigation is not possible without a detailed examination of the major controls for these two key parameters at the global scale. Many of the relationships established in this study can therefore also be applied to other scientific questions related to the erosion and transport of matter from the continents to the oceans by rivers.

ATMOSPHERE

LITHOSPHERE

OCEANS

VEGETATION

SOILS

7.0

1.5

+3

a

b

c d e fg h

p

sedimentary rocksfossil fuels

(i,j,k)1

(l,m,n)

Rivers

"Missing Carbon"

o

+2

2

Fig. 1 Generalized scheme of the present-day global carbon cycle according to various literature estimates (after Amiotte-Suchet [1995]). All fluxes are in GtC/yr: a, gross primary production (105);b, litterfall (55); c, respiration of living biomass (50); d, respiration of soil organic matter (54.5); e, fossil fuel combustion (6); f, deforestation (1.5); g, oceanic uptake (92); h, oceanic degassing (90.5); i,river flux of dissolved organic carbon (DOC) (0.2); j, river flux of particulate organic carbon (POC)from the soils (0.1); k, river flux of dissolved inorganic carbon (DIC) consumed by chemical rockweathering (0.25); l, river DIC flux from carbonate dissolution (0.15); m, river flux of particulateinorganic carbon (PIC) (0.15); n, river POC flux as fossil carbon from sedimentary rocks (0.1); o,carbon sedimentation (0.5, with 0.1 as organic carbon and 0.4 as inorganic carbon); p, volcanism(0.1). For a detailed review of the corresponding literature sources, see Amiotte-Suchet [1995].

Organic matter in rivers covers a broad spectrum of substances, ranging in size from free monomers via macromolecules and colloids to large particles and living animals (Fig. 2). For practical purposes, this reservoir is commonly divided into the fraction that is greater than 0.45 µm in size, and the one that is smaller than this limit. Filtration removes the particulate organic matter (POM, or POC when only considering the particulate organic matter's carbon), such as zooplankton, algae, bacteria, and detrital organic matter from the soils and plants from the remainder dissolved organic matter (DOM, or, respectively, DOC). So far no more than about one third of DOM in natural waters can be chemically identified as carbohydrates, amino acids, hydrocarbons, fatty acids, and phenolic compounds, while the bulk of DOM is present at the molecular level as complex polymeric organic

2

acids, called humic substances, or Gelbstoff (= 'yellow substance' because of the often yellow colour of these substances - Kalle [1966]).

10-1 1 10-1 0 10-9 10-8 10-7 10-6 10-5 10-4 10-3

Zooplankton

Phytoplankton

Bacteria

Viruses

Clay-humic-metal-complexes Hydr. A

HuA

FuA

FA

CH

AA

HC

sizeofwatermolecules

general colloid range

0.45 micronboundary

D O C P O C

meters

Fig. 2 Continuum of dissolved and particulate organic carbon in natural waters (from Thurman[1985]). Abbreviations: FuA, Fulvic Acid; HuA, Humic Acid; Hydr. A, Hydrophillic Acids; FA, FattyAcids; CH, Carbohydrates; AA, Amino Acids; HC, Hydrocarbons.

Beside the classifications based upon its size and its chemical nature, organic matter in rivers has been further classified with respect to its source. Here, one distinguishes allochthonous organic carbon, that is all carbon originating from land (or at least from outside the aquatic system), and autochthonous organic carbon, that is the carbon that is biosynthesized within the river itself and the adjacent lakes (mainly via the photosynthesis pathway). This classification is applied both for DOC and POC. Chemically, the difference between both sources is often followed in a way that autochthonous carbon consists in greater shares of labile substances such as carbohydrates or proteins because of the fresher character of this material (Spitzy and Leenheer [1991], Degens and Ittekkot [1985]). For allochthonous carbon, this is only the case as far as the material stems from fresh vascular plant matter. The part of the allochthonous carbon originating from the soils has normally decomposed for a longer period of time, consisting thus in greater shares of more refractory substances such as humic compounds. These classifications have to be kept in mind when reading the different aspects of riverine organic matter fluxes treated on the following pages.

Up to now, a considerable body of data on the fluxes of DOC and POC has been acquired during many field studies. Chapter I reviews some of them exemplary for the different climatic regions of the world. The river case studies mainly inform about the seasonal and interannual variations of river DOC and POC fluxes. They allow not only to understand the seasonal behaviour of the organic matter export from river basins under different hydroclimatic and geomorphological conditions, but also a comparison of the carbon fluxes is possible when a different temporal and spatial scaling is applied. The presented case studies comprise major world rivers which were normally sampled within monthly time intervals, a medium size river that was sampled in a daily to weekly resolution, and a small mountainous watershed that has been sampled up to a time resolution of about half an hour during peak discharges.

3

In order to investigate the controls of mean annual river fluxes observed for different rivers, a detailed characterisation of the environmental patterns of the river basins is needed. In this study, this was done by extracting the hydroclimatic, biological and geomorphological characteristics from a large number of global environmental data sets using the digitized basin contours. The purpose of chapter II is to present these data, and the way they were used for the statistics. On the whole, 60 river basins were included in this work, which cover together about 50% of the continental area that is drained to the oceans.

Chapter III is investigating continental runoff and its relation to climate and morphology. Based upon a digitized and gridded version of the UNESCO runoff maps (Korzoun et al. [1977]), and long-term literature runoff estimates for the 60 river basins, an approach for a refinement of global and regional runoff figures is developed. Then, multiple correlation statistics are applied in order to identify the most significant controlling factors at the global scale, and an empirical modelling for continental runoff is proposed.

River sediment fluxes are the key parameter for the river transport of particulate organic carbon to the oceans. The major controls for this key parameter are examined in chapter IV. It is shown that sediment yields (sediment fluxes divided by basin area) can be best correlated by forming the product of different hydroclimatic, geomorphological, and lithological factors, and a new modelling to predict sediment yields at the global scale is presented. This modelling is then applied to the total continents in order to produce a map of the regional variability of river sediment yields, which is in good agreement with field data.

After having established the major controls for the river fluxes of water and of sediments, chapter V returns to river organic carbon fluxes. For 31 of the investigated river basins, mean annual DOC and POC fluxes can be estimated from the literature. While most of the previous studies on the fluvial organic matter transport mainly tried to explain the seasonal variability of the fluxes in relation to the river hydrograph, it is shown in this chapter that it is also possible to relate the variability of mean annual DOC and POC fluxes with the environmental variability of the corresponding river basins. The most significant empirical regression models are established and applied to the total continental area on the basis of the corresponding data sets. Consequently, two global maps showing respectively the spatial distribution of river DOC and POC fluxes from the continents can be proposed in this chapter, and the regional particularities of the carbon fluxes are discussed in detail.

Chapter VI attempts a synthesis and evaluates the role of continental erosion within the global carbon cycle. An important element is here the linking of the modelling of the river fluxes of organic carbon with a global modelling of the river fluxes of inorganic carbon, which has been developed in the framework of the PhD thesis of Amiotte-Suchet [1995]. Since both inorganic and organic carbon fluxes represent a permanent transfer of carbon from the atmosphere to the oceans, a detailed picture of the consumption of atmospheric CO2 by continental erosion can thus be drawn. Furthermore, the transport of this carbon is followed by coupling the two modelling approaches with a global river routing scheme, allowing also a prediction of the local inputs of the different carbon species into the oceanic system. Finally, the overall empirical relationships established and used in this work are applied to a last glacial maximum (18000 yrs b.p.) scenario defined by general circulation model run, and the consequences for the river carbon fluxes are discussed.

Note that many of the here presented data and results have been subject of several publications: Frakes et al. [1994] (chapter II), Ludwig and Probst [1996] (chapter IV), Ludwig et al. [1996], Probst et al. [1994a] (chapter V), Ludwig et al. (a), in press, and in Ludwig et al. (b), in press (chapter VI). In these publications, sometimes different data sets were used to run the models, or different literature values were used for the regressions. This can have as a consequence that certain of the values are slightly different from the ones given in this work, although the general findings are naturally the same. In the here presented thesis, much care was taken to calculate all values on the basis of the same datasets and literature values. As far as specific river fluxes are given, they refer all to the same basin areas and discharge values.

4

Finally one has also to mention that this work was realized within the CEC research program ESCOBA (European Study on Carbon in the Ocean, the Biosphere, and the Atmosphere) - Terrestrial Biosphere. Many of the presented and utilized river data have been acquired in a number of international and national research programs, such as the international SCOPE/ UNEP program Transport of Carbon and Minerals in Major World Rivers carried out under the leadership of the Geological/Palaeontological Institute of the University of Hamburg, or the French research programs PEGI/ GBF (Programme Environnement Géosphère Intertropicale/ opération Grands Bassins Fluviaux) created by INSU-CNRS/ ORSTOM, and DBT II (Dynamique et Bilan de la Terre) / Fleuves et Erosion / ONT (Observatoire National de Terrain) Garonne created by INSU-CNRS. The monitoring of the Strengbach catchment in the French Vosges mountains was funded by the French research program DEFORPA (Déperissement des Forêts et Pollution Atmosphérique) until 1991, and it is actually financed within the STEP/ ENCORE program of the CEC, by the French ministries of Research and of Environment, and by the Région Alsace.

Version Française

Depuis les temps les plus reculés, les fleuves ont joué un rôle essentiel dans l'histoire de l'homme. Le Tigre, l'Euphrate et plus tard le Nil fournissaient les eaux et terres alluviales fertiles et permirent ainsi le développement des premières cultures sophistiquées socialement et économiquement. Plus tard les fleuves devinrent également un important moyen de transport pour les échanges de biens avec d'autres peuples. Ils constituèrent les voies naturelles pour l'exploration de terres encore inconnues ou guidèrent les hommes en quête de nouveaux espaces pour vivre. En même temps, l'on choisit les fleuves pour marquer les frontières entre les pays et ils furent ainsi souvent le cadre de nombreux conflits. La civilisation moderne a élargi leur utilisation, ainsi par exemple l'utilisation des fleuves pour la production d'électricité, ou bien - ce qui est de plus en plus grave pour l'environnement - leur utilisation dans le monde entier comme système de canalisation permettant de se débarrasser des énormes masses de déchets générés par les métropoles grandissantes.

See the rivers how they run. Changeless towards a changeless sea. Ces mots de l'écrivain anglais Charles Kingsley (1819-1875) expriment la fascination que les fleuves ont exercée sur de nombreux poètes, musiciens et peintres qui voyaient en eux le symbole de la permanence de la nature par opposition à la vie éphémère. C'est cette permanence qui attire également l'intérêt des géochimistes pour les fleuves. En fait, le cours d'un fleuve n'est pas aussi immuable que l'observateur debout sur la rive pourrait l'imaginer. C'est le cours continu des fleuves qui a creusé dans le paysage des vallées sans fonds ou bien, au contraire, érodé complètement de hautes montagnes formées lors des périodes géologiques précédentes. A l'échelle globale, des milliards de tonnes de matière dissoute ou particulaire sont arrachées chaque année des continents et entraînées vers les océans par les fleuves, ce qui représente une quantité non-négligeable pour le cycle global de nombreux éléments et substances sur la terre. Par exemple, l'apport fluvial en sels nutritifs aux zones côtières est important pour maintenir la fertilité biologique dans ces zones. L'importance des transports fluviaux dans les cycles globaux a même d'autant plus augmenté que l'homme a atteint la capacité de perturber ces transports. La destruction des forêts pluviales dans les tropiques augmente considérablement l'érosion et les transports fluviaux dans ces régions. En d'autres lieux les nombreux barrages artificiels retiennent une grande partie des matières transportées, entraînant ainsi une diminution drastique des flux. Pour beaucoup de fleuves drainant des régions très peuplées, les apports en déchets industriels et agricoles dépassent de loin les apports provenant des sources naturelles, pouvant ainsi entraîner un développement catastrophique du plancton ou une augmentation des maladies des poissons vivant à proximité des côtes.

L'objectif scientifique de cette thèse est une quantification globale et régionale de la matière organique transportée chaque année par les fleuves aux océans afin d'évaluer le rôle de ce transport dans le cycle global du carbone. Des concentrations croissantes en CO2 dans l'atmosphère liées à la combustion industrielle des carburants fossiles sont susceptibles d'entraîner des changements climatiques significatifs au cours des prochaines décennies. Après 30 ans de mesures dans

5

l'atmosphère et dans les océans, le budget atmosphérique global en CO2 est toujours étonnamment incertain. Via le cycle de la végétation, l'érosion de la matière organique et ses transferts successifs vers les océans par les fleuves est un puits permanent de CO2 atmosphérique. Avec le CO2 atmosphérique qui est consommé par l'altération chimique des roches, ce puits a été estimé à une valeur globale de 0,5 à 0,6 GtC (gigatonnes de carbone) par an, dont environ 35% peuvent être attribuées au carbone organique dissous, 20% au carbone organique particulaire et 45% à la consommation de CO2 atmosphérique par l'altération des roches. Si l'on ajoute encore le carbone d'origine lithologique, la quantité totale de carbone qui est transportée chaque année vers les océans par les fleuves peut aller jusqu'à une valeur de 1 GtC/an, ce qui représente une quantité non-négligeable dans le cycle global de carbone à l'heure actuel (Fig. 1).

Jusqu'à ce jour, il était cependant difficile d'évaluer le rôle du transport fluvial plus en détail dans ce contexte par manque de données suffisantes sur les fleuves à l'échelle globale. Le travail récent de Amiotte-Suchet [1995] qui a été effectué au Centre de Géochimie de la Surface (CGS) a déjà fourni une modélisation des flux de carbone inorganique à l'échelle globale, mais l'on sait toujours peu sur les flux de la matière organique. Le propos de cette étude est ainsi de développer un outil de modélisation basé sur les principaux facteurs de contrôle des transports fluviaux de carbone organique, qui permettrait de répondre à ce besoin en données. L'approche concerne de manière à peu près équivalente les flux de carbone organique dissous et les flux de carbone organique particulaire. Ces flux étant étroitement associés aux flux d'eaux et de sédiments, une telle étude n'est possible qu'avec une investigation détaillée des principaux facteurs de contrôle de ces deux paramètres clés à l'échelle globale. Beaucoup de relations établies dans cette étude peuvent être appliquées à d'autres problèmes scientifiques concernant l'érosion et les transports fluviaux de matière des continents vers les océans.

La matière organique dans les fleuves couvre un large spectre de substances, qui varient en taille en partant des monomères libres, via les macromolécules et les colloïdes jusqu'aux particules larges et aux animaux vivants (Fig. 2). Pour des raisons pratiques, ce réservoir est généralement divisé en deux fractions: l'une concernant les tailles supérieures à 0,45 μm, l'autre les tailles inférieures à 0,45 μm. La matière organique particulaire (MOP, ou COP si l'on considère uniquement le carbone organique particulaire) - telles le zooplancton, les algues, bactéries et les matières organiques détritiques provenant des sols et des plantes - est séparé par filtration de la matière organique dissoute (MOD, ou respectivement, COD) restante en solution. Actuellement, l'on ne peut identifier chimiquement qu'environ un tiers du MOD dans les eaux naturelles. Cette partie identifiable est constituée des carbohydrates, des acides aminés, des hydrocarbures, des acides gras et les composés phénoliques. La partie non-identifiable du MOD est constituée par des acides organiques polymères complexes, appelés substances humiques ou Gelbstoff (= 'substance jaune' à cause de la couleur jaune qu'ont souvent les acides humiques en solution - Kalle [1966]).

A côté des classifications par taille et selon la nature chimique, la matière organique dans les fleuves a en outre été classée en fonction de son origine. Ici, l'on distingue le carbone organique allochtone, c'est-à-dire le carbone issu de la biosphère terrestre (ou hors du système aquatique), du carbone organique autochtone, c'est-à-dire le carbone synthétisé dans les eaux de fleuve et des lacs attenants (principalement par photosynthèse). Cette classification est appliquée aussi bien pour le COD que pour le COP. Chimiquement, les deux sources se distinguent souvent par le fait que le carbone autochtone contient en grande partie des substances labiles comme les carbohydrates ou les protéines à cause du caractère 'plus frais' de cette matière (Spitzy et Leenheer [1991], Degens et Ittekkot [1985]). Pour le carbone allochtone, ceci est uniquement le cas en ce qui concerne la matière organique provenant des plantes vasculaires. La matière organique allochtone issue des sols est normalement décomposée depuis longtemps et consiste ainsi pour une grande part en substances plus réfractaires, tels que les composés humiques. Ces classifications doivent être gardées à l'esprit dans les pages suivantes lorsque seront traités les différents aspects des transports fluviaux de matière organique.

A ce jour, un nombre considérable de données sur les flux de COD et COP a été accumulé lors d'études de terrain. Le chapitre I reprend quelques-unes de ces études à titre d'exemple pour les différentes régions climatiques du monde. Les cas d'études révèlent essentiellement les variations

6

saisonnières et interannuelles des flux de COD et COP. Elles permettent non seulement de comprendre le comportement saisonnier de l'exportation de la matière organique à partir des bassins versants caractérisés par des conditions climatiques et géomorphologiques différentes, mais également de comparer les flux de la matière organique si on applique un changement d'échelle temporelle et spatiale. Les cas d'études présentés incluent les grands fleuves mondiaux échantillonnés mensuellement, un fleuve de taille moyenne échantillonné à intervalle d'une semaine à un jour et un petit cours d'eau montagneux qui a été prélevé jusqu'à une résolution d'une demi heure pendant des crues.

Afin de rechercher les contrôles de flux moyens annuels de carbone qui sont observés pour différents bassins, il est nécessaire de déterminer les caractéristiques environnementales de ces bassins. Ceci a été réalisé dans cette étude en extrayant les caractéristiques hydroclimatiques, biologiques et géomorphologiques d'un grande nombre de banques de donnés globales à partir des contours digitalisés des bassins. Le propos du chapitre II est de présenter ces données ainsi que la manière dont elles ont été utilisées pour les analyses statistiques. Au total, 60 bassins versants ont été inclus dans cette étude. Ils représentent ensemble à peu près 50% de la surface continentale drainée vers les océans.

Le chapitre III est consacré à l'investigation du drainage continental et à sa relation avec le climat et la morphologie. A partir des versions digitalisées des cartes de drainage de l'UNESCO (Korzoun et al. [1977]) et des estimations de l'écoulement moyen pour les 60 bassins provenant de la littérature, une approche a été développé pour quantifier le drainage continental à l'échelle globale et régionale. Ensuite, des procédures de régressions multiples ont été appliquées pour identifier les facteurs de contrôle à l'échelle globale et une modélisation empirique pour le drainage continental est proposée.

Le flux de sédiments constitue le paramètre clé pour le transport fluvial de carbone organique particulaire vers les océans. Les principaux contrôles de ce paramètre clé sont examinés dans le chapitre IV. Il est montré que les flux de sédiments spécifiques (flux de sédiments rapportés à la surface du bassin) peuvent être le mieux corrélés avec des produits des facteurs hydroclimatiques, géomorphologiques et lithologiques et une nouvelle modélisation empirique pour prédire les flux de sédiments spécifiques à l'échelle globale est présentée. En appliquant cette modélisation empirique à la totalité de la surface continentale, une carte globale de la variabilité des flux de sédiments spécifiques est ainsi proposée qui est bon accord avec des observations de terrain.

Après avoir établi les principaux facteurs de contrôle pour les transports fluviaux d'eau et de sédiments, le chapitre V revient aux flux de carbone organique. Pour 31 des bassins versants étudiés dans cette étude, on peut estimer des flux annuels moyens de COD et de COP. Alors que la plupart des précédentes études sur le transport de la matière organique dans les fleuves ont tenté essentiellement d'expliquer la variation saisonnière des flux par rapport à la variation du débit, il est montré qu'il est également possible de rapporter la variabilité des flux de COD et de COP annuels moyens à la variabilité environnementale des bassins versants correspondants. Les modèles de régression les plus significatifs sont établis et appliqués à l'ensemble de la surface continentale totale à partir des paramètres environnementales concernés. Par conséquence, deux cartes globales représentant respectivement la distribution spatiale des flux spécifiques de COD et de COP sur les continents sont ainsi proposées, en discutant en détail les particularités régionales.

Le chapitre VI tente de faire une synthèse et évalue le rôle de l'érosion continentale dans le cycle global du carbone. Un élément important est ici le couplage de la modélisation empirique pour les transports fluviaux de carbone organique avec la modélisation empirique pour les transports fluviaux de carbone inorganique. La dernière a été développée dans le cadre de la thèse de Amiotte-Suchet [1995]. Ceci permet de donner une image détaillée de la consommation de CO2 atmosphérique sur les continents, car également les transports fluviaux de carbone organique comme les transports de carbone inorganique représentent un transfert permanent de carbone atmosphérique vers les océans. De plus, les deux approches de modélisation ont été couplées à une procédure de routage des eaux

7

continentales vers les océans, permettant ainsi une régionalisation des apports de carbone sous ses différents formes aux océans. Enfin, l'ensemble des relations empiriques présentées dans cette étude sont appliquées à un scénario climatique pour le dernier maximum glaciale (18000 ans b.p.) et les conséquences pour les transports fluviaux de carbone sont discutées.

Nous noterons que beaucoup des données et résultats présentés ici ont été le sujet de plusieurs publications: Frakes et al. [1994] (chapitre II), Ludwig and Probst [1996] (chapitre IV), Ludwig et al. [1996], Probst et al. [1994a] (chapitre V), Ludwig et al. (a), sous presse et dans Ludwig et al. (b), sous presse (chapitre VI). Dans ces publications, les modèles empiriques ont pu être appliqués à d'autres banques de données environnementales où d'autres valeurs de la littérature ont pu être utilisées pour déterminer les modèles de régression. Il est possible par conséquent que certaines valeurs présentées ici diffèrent légèrement de celles données dans les publications. Ceci ne modifie cependant en rien les conclusions tirées dans cette étude. Un soin particulier a été pris que toutes les valeurs soient calculées sur la base des mêmes banques de données et des mêmes valeurs de littérature. Lorsque sont indiqués des flux, ils se réfèrent toujours aux mêmes valeurs de surface des bassins ou aux mêmes valeurs de drainage.

Finalement il faut aussi mentionner que ce travail a été réalisé dans le cadre du programme de recherche de la CEE ESCOBA (European Study on Carbon in the Ocean, the Biosphere, and the Atmosphere) -Terrestrial Biosphere. Beaucoup des données présentées et utilisées dans cette étude ont été acquises lors de plusieurs programmes de recherche nationaux et internationaux, comme le programme international SCOPE/ UNEP Transport of Carbon and Minerals in Major World Rivers (qui a été effectué sous la direction du Geological/Palaeontological Institute de l'Université de Hambourg) ou les programmes français PEGI/ GBF (Programme Environnement Géosphère Intertropicale/ opération Grands Bassins Fluviaux) de l'INSU-CNRS/ ORSTOM et DBT II (Dynamique et Bilan de la Terre) / Fleuves et Erosion / ONT (Observatoire National de Terrain) Garonne de l'INSU-CNRS. Les recherches dans le bassin versant du Strengbach dans les Vosges ont été financées par le programme français DEFORPA (Dépérissement des Forêts et Pollution Atmosphérique) jusqu'en 1991, et après dans le cadre du programme STEP/ ENCORE de la CEE, par un support financier du Ministère de la Recherche et de l'Environnement et par un financement de la Région Alsace.

8

9

CHAPTER I

THE STATE OF ART: CASE STUDIES ON RIVER FLUXES OF ORGANIC CARBON

1.1. Introduction

Up to now, a considerable body of data on river fluxes of dissolved organic carbon (DOC) and particulate organic carbon (POC) has been acquired during many field studies. It is the purpose of this chapter to review many of these studies, and to attempt a synthesis of their major findings. The river case studies inform both about the variability of annual DOC and POC fluxes in different climatic regions of the world, as well as about their seasonal and interannual variability. It is mainly the latter aspect that is discussed here. The variability and major controls of mean annual DOC and POC fluxes is subject of chapter V. On the whole, eleven river case studies are presented (Fig. 3). A review of these studies is important, also because it allows to become familiar with the major problems related to the determination of average river fluxes, which has to be kept in mind when using the results of the modelling approaches that will be developed in the following chapters.

Many of the presented carbon data come from the SCOPE/ UNEP program Transport of Carbon and Minerals in Major World Rivers that was carried out under the leadership of the Geological/Palaeontological Institute of the University of Hamburg (henceforth the SCOPE program). Most of the SCOPE data were published (e.g., Degens [1982], Degens et al. [1983], [1985], [1987], [1988], [1991a], Kempe et al. [1993]), but in the cases where there exist unpublished data (e.g., the Waikato River in New Zealand), I included them in this study with the agreement of the scientist who carried out the field work. Other data were collected at the Centre de Géochimie de la Surface (CGS) within the framework of several research programs and public fundings (INSU-CNRS-ORSTOM: PIRAT and PEGI-GBF, INSU-CNRS: DBT - fleuves et érosion/ DBT-ONT Garonne, Ministère d'Environnement, Région Alsace), or were taken from the literature. The presented studies were selected in a way that they allow also a comparison of the carbon fluxes when a different temporal and spatial scaling is applied: they comprise a small French watershed in the Vosges mountains (the Strengbach catchment) that has been simultaneously sampled at several sites in the basin in a time resolution of about half an hour during storm events, a medium size temperate river (the Garonne River in France) that was sampled in a weekly resolution both in its headwater region and in its lower course, and numerous major world rivers (i.e. the rivers investigated within the SCOPE program) that were normally sampled in monthly time intervals close to the river mouths.

The data are presented in a uniform way in order to facilitate a comparison of the studies. Whenever this was possible, additional information on the general hydrologic regimes of the rivers is supplied. In the discussion, special emphasis is given on the evolution of the riverine DOC and POC concentrations in relation to the variations of the river discharge. This shows how important seasonality is for the organic carbon export from the drainage basins, but it may also help to identify

10

-150

-100

-50

0 5

010

015

0

-50

0

50

-50

0 50

-150

-100

-50

0 5

010

015

0

.

1

6

2

1

3 4

5

7 8 9

10

11

F

ig. 3

Dis

trib

utio

n of

the

elev

en r

iver

cas

e st

udie

s th

at a

re p

rese

nted

in th

e te

xt: 1

, Mac

kenz

ie; 2

, St.

Law

renc

e; 3

, Ori

noco

; 4, P

aran

á; 5

, Str

engb

ach

broo

k (V

osge

s m

ount

ains

, Fra

nce)

; 6,

Gar

onne

; 7,

Nig

er;

8, Z

aire

(m

ainl

y on

e of

its

maj

or tr

ibut

arie

s, th

e U

bang

i, is

dis

cuss

ed h

ere)

; 9,

Ora

nge;

10,

Ind

us;

11,

Wai

kato

.

11

the contributions of the different carbon reservoirs (vegetation, soils, living and dead aquatic biomass) to the annual carbon fluxes. We will see in the following that the concentrations of particulate organic carbon are generally strongly coupled to the concentrations of total suspended solids (TSS). In all cases where this parameter has been measured, it is shown together with the organic carbon data. When available, also the C/N ratio is presented, that is the ratio of organic carbon to organic nitrogen in the total suspended solids. The C/N ratio can indicate the source of the particulate organic carbon in rivers. Values around 10 are typical for soil organic matter (FAO [1971-81]). High ratios from 15 to more than 50 imply unaltered vascular plant debris (Hedges et al. [1986]), while it has been proposed that low ratios of 6 to 10 are due to organic matter derived from primary production in the river (e.g., Pocklington and Tan [1987], Depetris and Kempe [1993]). However, such an assignment of low C/N ratios to in-situ primary production in rivers is probably less evident. Also in soils, C/N ratios can decrease to that low values in the lower horizons (FAO [1971-81]) where the age of the organic matter normally increases. It is, for example, a striking feature that the C/N ratios in the Huanghe River are close to 8 (Zhang et al. [1992]). This river is known to have among the highest TSS concentrations of all world rivers (see chapter IV), which should not allow the development of in-situ production in the river.

In one case, the St. Lawrence River study, also δ13C isotope data are available in order to characterize the nature of the particulate organic matter.

1.2. A Small Mountainous Watershed in Temperate Region: The Strengbach Catchment

The Strengbach catchment is located at Aubure on the eastern side of the Vosges mountains, 58 km south-west from Strasbourg (France). Its elevation ranges from 883 m at the outlet to 1146 m at the top of the catchment. The area is about 80 ha. Norway spruce is the abundant vegetation type (2/3 of the area), and mixed white fir and beech cover the rest. The Strengbach catchment is monitored since the end of 1985 in order to assess the effects of acid rain on forest decline and water quality by the mean of calculating hydrochemical budgets. This monitoring was funded by the French research

SS1 SS2

SS3 SS4

900m

950m

1000m

1050m

1050m

1100m

1100m

0 200m

N

Site of theStrengbach catchmentin France

Fig. 4 Map of the sampling sites in the Strengbach catchment at Aubure. SS1, upper brook course; SS2 small side-brooklet draining the north-eastern slopes of the catchment; SS3, waters coming out of a zone characterized by a quasi-permanent water saturation of the soils; SS4, brook gauging station at the outlet of the catchment.

12

program DEFORPA until 1991, and it is actually financed within the STEP/ ENCORE program of the CEC, by the French ministries of Research and of Environment, and by the Région Alsace. Within the framework of the French research program DBT II "Fleuves et Erosion", the hydrochemistry of the catchment waters have been investigated with a high resolution sampling strategy (simultaneous sampling of the catchment waters at several sites in time steps of about half an hour) during peak discharges provoked by storm rainfall (Ladouche et al. [1995]). More details on the monitoring of the hydrochemistry of the Strengbach catchment can be found, for example, in Probst (A.) et al. [1987], [1990], [1992], and Viville et al. [1988].

Dissolved organic carbon is measured in the stream and rain waters since 1993. Analyses were done at the CGS with a Shimadzu 5000 TOC carbon analyser, after filtration of the samples with 0.45 μm Millipore membrane filters. In the following, I present data of a storm event during two days in May 1994, as well as data of the permanent DOC monitoring, which cover two complete seasonal cycles from 1993 to 1995. The corresponding discharge values were measured by D. Viville and co-workers (CEREG/ULP-CNRS, Strasbourg).

1.2.1. The May 1994 Storm Event

Figure 4 shows the locations of four of the principal sampling sites in the Strengbach catchment, which were sampled for DOC during the May storm event (18/05 to 20/05 1994). These sites are the upper brook course (SS1), a small side-brooklet that receives its waters from the north-eastern slopes of the catchment (SS2), the waters coming out of a zone that is characterized by a quasi-permanent water saturation of the soils (SS3), and the lower course at the main gauging station (SS4) at the outlet of the catchment. The evolution of the DOC concentrations (cDOC) during the May storm event in relation to discharge (Q) can be seen in Figure 5, together with the evolution of the concentrations of TSS (cTSS).

Measured DOC concentrations in the catchment range from 0.8 mg/l (SS1) to 9.6 mg/l (SS3). They were greatest in the waters draining the water saturated soil horizon at SS3 (range: 6.8 to 9.6 mg/l; average: 8.1 mg/l), and lowest in the upper course of the brook at SS1 (range: 0.8 to 3.1 mg/l; average: 1.6 mg/l). At the outlet (SS4), cDOC varied between 1.5 and 4.8 mg/l, with a mean value of 3.5 mg/l. The average water discharge was 7.6 l/sec, 1.6 l/sec, 3.1 l/sec, and 20.0 l/sec at SS1, SS2, SS3, and SS4, respectively. The rise of the waters at the outlet started in the early afternoon of the 18th (about 14:30 h). Q achieved its maximum (Qmax1) around midnight (00:15 h), dropped down to a minimum (Qmin) at 03:00 h, and rised then again to a second but smaller maximum (Qmax2) in the early morning of the 19th (05:30 h). The high-water stage ended at about 10:00 h in the morning, and discharge declined slowly during the rest of the day, although the discharge never dropped down to the initial value. At the other sites, discharge followed more or less this evolution (Ladouche et al. [1995]).

The concentrations of dissolved organic carbon show a clear relationship with Q at all four sampling sites. At SS2 and SS4, cDOC increased and decreased in parallel with Q over the complete high-water period. For SS1, the sampling frequency was much lower than at SS2 and SS4, but the existing data imply that the parallel evolution of cDOC with Q also was the case at this site. It is interesting to note that at SS2, Qmax2 provoked an increase of cDOC to values that are nearly as high as those during Qmax1, although this second maximum reached only about two thirds of the first maximum in terms of water discharge. At SS3, the evolution is different compared to the other sites. Here, cDOC increased in parallel with Q at the beginning of the water rise, and reached maximum values about 6.75 h before Qmax1. For the following 2 h, the concentrations were still at an elevated level, but then they decreased rapidly to a value below the initial value. Qmax2 had obviously no marked influence on the concentrations, since cDOC decreased more or less continuously until the end of the high-water stage.

13

POC was not measured in the Strengbach Brook, but TSS. Because in many case studies (see the following of this chapter), the concentrations of POC closely follow those of TSS, I present here also TSS data. This allows to make an idea how the export of particulate organic carbon may have been during the investigated storm event.

Data on the concentrations of TSS are more sparse than data on cDOC, except for the outlet station (SS4), where the data coverage is good over the complete high-water period. Nevertheless, it can be seen from Figure 5 that at all sampling sites a sharp cTSS maximum occurred between 19:15 h and 19:30 h, about 5 h before Qmax1. Then, cTSS decreased rapidly. Only at SS3 and SS4, a much smaller second cTSS maximum is found at Qmax1, and, in the case of SS4, a slight increase of cTSS can also be observed at the rising limb of Qmax2. At the outlet, cTSS ranged from 0.1 to 90 mg/l over the sampling period, with a mean value of 18.5 mg.

0

3

6

9

12

15

0

1

2

3

4

0

1

2

3

4

1

2

3

4

5

0

2

4

6

8

10

6

7

8

9

10

0 12 24 36 48 60

0

10

20

30

40

50

1

2

3

4

5

6

Q

Q

Q

Q

DOC

DOC

SS1

DOC

SS2

SS3

DOC

SS4

Hours (Begin: 18/05/94)

0

3

6

9

12

15

0

3

6

9

12

15

0

1

2

3

4

0

20

40

60

80

0

2

4

6

8

10

0

50

100

150

200

0 12 24 36 48 60

0

10

20

30

40

50

0

20

40

60

80

100

Q

Q

Q

Q

TSS

TSS

SS4

TSS

SS3

TSS

SS2

SS1


Fig. 5 Evolution of DOC (left column) and TSS (right column) concentrations in relation to discharge at four sampling sites in the Strengbach catchment during a storm event in May 1994. Note that the y-axis to the left depict discharge (l/sec), while the y-axis to the right depict DOC (mg/l) and TSS (mg/l) concentrations, respectively. For further explanations, see text.

The high-resolution DOC sampling and Q measurements in the catchment allow a precise calculation of the fluxes of dissolved organic carbon (FDOC) at the corresponding sampling sites. SS4 includes the waters that passed at the three other sampling stations, and it can thus be determined to what extent the waters at SS1, SS2, and SS3 have contributed to the total FDOC measured at SS4. Such a calculation requires an interpolation of values for uniform time slices, because the data coverage is not identical at the sampling sites. For drainage, I did this by linear interpolation in 7.5 min time intervals (Qi, with i = 0.125 h, 0.250 h, 0.375 h, ...) between the two closest Q measurements (Qlow1, Qup1) at each site. Note that for all flux calculations, I consider the high-water period to begin the 18/05 at 12:00 h, and to end the 19/05 at 24:00 h.

14

cDOC was interpolated in the same way as Q (resulting in cDOCi), except at SS1. At this station, DOC data are too sparse and the interpolation was also coupled to the evolution of Q. This means that I first determined cDOCi between the two closest DOC data points (cDOClow2, cDOCup2) by linear interpolation as described above. Then I repeated the interpolation for Q, taking this time the Q limits corresponding to the cDOC values (Qlow2, Qup2). This results in Qi2, which are not identical to Qi because there can exist additional Q measurements between Qlow2 and Qup2. Finally, cDOCi were corrected (cDOCic) according to:

cDOCic = cDOCi x Qi / Qi2 (1)

The evolution of the interpolated cDOC values in comparison with the measured values at SS1 can be seen in Figure 5 (note that at this station, the straight line is not linearly linking the DOC values). This interpolation technique can be justified because we have seen that at all 4 sampling sites the evolution of cDOC was obviously linked to the evolution of Q. Keep in mind, however, that the calculated FDOC values at SS1 are less sure compared to the three other stations.

12 18 24 30 36 42 48

0

20

40

60

80

100

FluxDOC(gC)


SS1

SS2

SS3

"missing carbon"(SS4-SS3-SS2-SS1)

Fig. 6 Fluxes of dissolved organic carbon at four sampling sites in the Strengbach catchment duringa storm event in May 1994.

Figure 6 depicts the corresponding fluxes of dissolved organic carbon at the four sampling sites during the high-water period. On the whole, 8782 gC were exported from the catchment. 23.8% of this carbon can be attributed to SS1, 7.8% to SS2, and 33.3% to SS3. The "missing carbon", which is here the carbon that entered the brook through non-point runoff in addition to the contributions of SS1, SS2, and SS3 is thus 3082 gC, or 35.1% of the total DOC export. On the basis of the extrapolated Q and cDOC values, it is found that the average water discharge during the high-water period was 7.6 l/sec, 1.5 l/sec, 3.0 l/sec, and 18.9 l/sec for SS1, SS2, SS3, and SS4, respectively (note that these values vary slightly from the above given averages because they are time-weighted). The corresponding discharge-weighted mean DOC concentrations are 2.1 mg/l, 3.5 mg/l, 7.6 mg/l, and 3.6 mg/l at the four stations, respectively. For the missing contribution of water and of carbon to the brook that could not be monitored within the catchment, the average discharge is 6.9 l/sec, and the corresponding discharge-weighted cDOC is 3.5 mg/l. It is interesting to note that this concentration corresponds nearly exactly to the concentration found for SS2. It is therefore possible that the missing carbon had the same origin than the waters at SS2 (mainly the north-eastern slopes of the catchment). It probably entered the brook in a more diffuse way in the vicinity of SS2. However, it is also possible

15

that the contribution of this carbon results from a mixture of waters corresponding to those of SS1 and of SS3 (Ladouche et al. [1995]).

The study of cDOC, Q, and FDOC at the different sampling sites reveals that the export of dissolved organic carbon from the Strengbach catchment during the storm event was not a uniform process. Two major types of carbon mobilization can be observed. One type is found for the waters draining the zone of the catchment that is characterized by a quasi-permanent water saturation of the soils. Here, DOC concentrations are generally much higher than in the rest of the catchment, and an important part of the overall DOC flux from the catchment is contributed from this zone. The onset of precipitation provokes still an increase of cDOC, indicating that with the new water supply, additional carbon is leached, and/or soil waters that are rich in DOC are purged out of the reservoir. However, this concentration increase ends rapidly, and then cDOC continuously drop down. Already about 8 to 10 h after the beginning of the peak discharge, cDOC felt below the initial value, and the dilution of the soil waters with the water supplied by precipitation (which are poor in DOC - A. Probst, pers. commun.) dominates the leaching processes. Consequently, it can be calculated that in the first phase of the high-water period (before Qmin), SS3 contributed about 39% to the total FDOC measured at SS4, while in the second phase (after Qmin), this was only about 29%. Because of the throughout higher concentrations at SS3 compared to the other stations, it is not unlikely that a considerable part of the carbon that was mobilized from the water saturated zone has already been leached from the soils in the days or weeks preceding the investigation period.

For the rest of the dissolved organic carbon that was transported out of the catchment, the leaching of the vegetation and soil reservoirs occurred in parallel with the water supplied by precipitation, since cDOC evolved in phase with Q. This is the second type of carbon mobilization that can be observed in the Strengbach catchment. A leaching of soil and vegetation organic matter during the time preceding the storm event was here probably not possible, because of a water deficiency in the corresponding horizons. May be due to shorter reaction times for leaching, cDOC are generally lower here. It is also possible that the nature of the dissolved organic matter is somewhat different compared to the waters at SS3. At SS3, the waters may contain a higher share of more refractory substances, which need a longer time to get into solution. This is, however, speculative, and should be verified by additional geochemical analyses. Another important questions in this context are the residence time and the routing of the water in the catchment. There are actually two PhD thesis carried out in order to investigate the contributions of the different hydrological reservoirs: Idir [in preparation] on the basis of the major element composition (CGS), and Ladouche [in preparation] on the basis of stable isotopes (UPMC, Paris VI).

The total flux of sediments (FTSS) at the outlet during the high-water period is calculated to be 31 kg (or 38.8 t km-2 yr-1 when extrapolated to an annual basis). About 76% of this flux occurred during the first phase of the high-water stage. This points out the important role of initial precipitation for the sediment mobilization, which probably washes out most of the detachable particles. In later stages, when the soils are humid, the binding of the superficial soil particles within the entire soil complex may also be reinforced. The low sampling density especially at SS1 makes it impossible to decompose the TSS flux at SS4 with respect to the contributions from the other sampling sites, but it becomes evident from Figure 5 that the initial sediment flushing occurred at all stations.

1.2.2. Permanent DOC Monitoring

From October 1993 to September 1995, two complete seasonal cycles were monitored in the Strengbach Brook at SS4. The evolution of cDOC in relation to Q during this period is shown in the Figures 7a and b. On the whole, 282 samples were analysed for DOC. Concentrations varied between 1.0 mg/l and 4.8 mg/l in the first year (yr1), and between 0.9 mg/l and 8.5 mg/l in the second year (yr2). The discharge weighted cDOC averages are 1.7 mg/l (yr1) and 1.4 mg/l (yr2), respectively. Note that these values were calculated by weighting cDOC with the water volumes that have been

16

0

30

60

90

120

150

180

210

240

0

1

2

3

4

5

6

7

8

9

Q

DOC

Days 93/94 (Begin: Oct. 93

Q(l/sec)

May 1994storm event

1. year

0

30

60

90

120

150

180

210

240

0

1

2

3

4

5

6

7

8

9

Q

DOC

0 30 60 90 120 150 180 210 240 270 300 330 360O N D J F M A M J J A S

Days 94/95 (Begin: Oct. 94

Q(l/sec)

2. year

Fig 7a Evolution of the concentrations of dissolved organic carbon in relation to discharge in theStrengbach Brook at SS4. The investigation period starts in October 1993 and ends in September1995.

discharged between two samples, and not with the instantaneous Q values. This is necessary because the sampling intervals could be very short during storm events (see above). Otherwise these 'spots' would have been over-represented in the samples. For the average of the two years, the corresponding mean drainage intensity was 938 mm, and the mean specific DOC flux was 1.46 t km-2 yr-1. Note that when extrapolated to one year, the measured DOC export during the May 1994 storm event would correspond to a specific FDOC of 2.67 t km-2 yr-1.

Concentrations of dissolved organic carbon in the Strengbach Brook were generally more elevated in summer than in winter. This has also been reported in other studies investigating DOC in stream waters of small catchments, as, for example, in the study of Grieve [1991] on catchments with

17

peaty soils in Scotland, in the study of Moore [1989] on forested catchments in New Zealand, and in the study of Moore and Jackson [1989] on wetland catchments in New Zealand. However, more striking than seasonal variability of the mean DOC concentrations in the Strengbach Brook is the close coupling of cDOC to Q. cDOC normally increased with increasing discharge. Moreover, for similar discharges, cDOC were higher on the rising limb of a discharge peak than on the falling limb (in the literature this effect has been commonly referred to as a so-called "flushing effect"), although exceptions to this trend occurred (see below).

60 90 120 150

0

20

40

60

80

100

120

140

160

180

0

1

2

3

4

5

6

Q

DOC

Dec. 93 Jan. 94 Feb. 94

Days 93/94 (Begin: Oct. 93)

Q(l/sec)

DOC(mg/l)

Fig 7b Two details of Fig. 7a: Three successive high-water periods in winter 1993/94 (uppergraph), and a storm event insummer 1995 (graph to the right).

294 295 296 297

0

15

30

45

60

75

90

0

1

2

3

4

5

6

7

8

9

Q

DOC

Days 94/95 (Begin: Oct. 94)

Q(l/sec)

DOC(mg/l)

In the above mentioned literature examples, a positive correlation of cDOC and Q was found to be a characteristic feature of the streams draining forested catchments, whereas such a correlation was found to be absent (or to be even negative) in streams draining wetlands (Moore and Jackson [1989], and references therein). In agreement with the Strengbach data, Moore [1989] reported also a flushing effect for forested catchments that were sampled during storm events. Grieve [1991] reported for catchments with peaty soils that maximum cDOC could occur shortly after maximum discharge. In the case of the wetland catchments, even a reverse flushing effect could be observed during storm events, which means that cDOC are more elevated during the falling water and/or at the end of a storm event (Moore and Jackson [1989]). It has to be mentioned, however, that the average DOC concentrations in the here discussed literature examples were throughout higher than in the Strengbach Brook.

18

Discharge-weighted mean cDOC varied between 5.8 and 12.1 mgl/ in the study of Moore [1989], between 8.8 and 10.8 mgl/ for the study of Grieve [1991], and between 29.9 and 38.0 mg/l in the study of Moore and Jackson [1989]. Discharge-weighted DOC concentrations that were in about the same range of those of the Strengbach Brook have been found, for example, by McDowell and Wood [1984] for the Bear Brook catchment (a small catchment in the Hubbard Brook Experimental Forest, New Hampshire), but few is reported in this study about the evolution of cDOC in relation to Q. This broad concentration range implies that factors such as climate, morphology, and the characteristics of the soils and/ or the vegetation cover may have a great influence on concentrations of DOC in stream waters. For example, it is likely that in the above cited literature examples the organic carbon in the soils has a great influence on the reported DOC concentrations. One can expect that the amount of carbon in the soils of these catchments considerably increases in the same sense as the reported mean cDOC increase.

The flushing effect in the Strengbach Brook is a prominent feature of the discharge peaks in winter, when the brook has normally its maximum discharge over the year. This can be clearly seen for the three successive high-water peaks encountered in winter 1993/94 (Fig. 7b). For all three peaks, cDOC increased with the onset of the rising water, but reached maximum concentrations already before the maxima of Q occurred, and then cDOC decreased rapidly. In the case of the first and the third peak in winter 1993/94, there appeared a small cDOC maximum just at the beginning of the water rise. Also during the high-water periods in winter and early spring of 1994/95, a flushing effect is found, even if less pronounced than in the preceding year (Fig. 7a). During winter, the patterns of cDOC and Q found at the outlet of the catchment are hence similar to the patterns that have been observed during the May 1994 storm event at SS3. By looking in detail at this event, we have already seen that the water saturated zone contributed a considerable part to the total FDOC exported from the catchment. Its relative contribution to the fluxes of DOC in winter may be even greater.

Table 1. Monthly export of dissolved organic carbon from the Strengbach catchment(October 1993 to September 1995).

___________________________________________________________________________________________________

1993/94 1994/95 ________________________________________________ ________________________________________________ Month Q FDOC cDOC Month Q FDOC cDOC

(103 m3) (kg C) (mg/l) (1) (103 m3) (kg C) (mg/l) (1) ___________________________________________________________________________________________________ October 81 182 2.2 October 17 29 1.7 November 38 51 1.3 November 14 24 1.7 December 158 293 1.9 December 57 89 1.6 January 183 275 1.5 January 145 233 1.6 February 59 81 1.4 February 118 131 1.1 March 89 123 1.4 March 101 117 1.2 April 78 103 1.3 April 75 92 1.2 May 56 99 1.8 May 38 75 2.0 June 45 75 1.7 June 61 88 1.4 July 14 37 2.7 July 14 31 2.2 August 12 26 2.1 August 10 19 2.0 September 14 25 1.8 September 21 36 1.7

Total 828 1370 1.7 Total 672 965 1.4 ___________________________________________________________________________________________________

(1) discharge-weighted

During discharge peaks in summer no typical flushing effect can be seen. On the contrary, cDOC tend to be more elevated on the falling limb of discharge peaks than on the rising limb, for similar discharges. This becomes the most evident during a storm event in July 1995 that has been sampled and analysed in a high resolution (Fig. 7b). Here, Q rised very quickly, and dropped then down close to its initial value within about 24 h. cDOC increased together with Q, but remained at elevated levels for at least about 2 days.

19

The controls and processes determining the export of DOC from the Strengbach catchment in summer may therefore be somewhat different than in winter. It is possible that the higher temperatures in summer accelerate soil leaching, which is consistent with the generally higher cDOC values found in this season. Higher cDOC can also be partly the effect of a concentration of the waters in the catchment through the enhanced evapotranspiration in summer. Keep in mind, however, that the summer situation is only of minor importance for the annual DOC export from the Strengbach catchment. Clearly, this can be seen in Table 1. The period from June to September contributed in the first year only about 12% to the annual FDOC. In the second year, this was about 18%.

It is interesting to note from Table 1 that for the first year (1993/94), the average discharge-weighted cDOC value is higher than for the second year, although the second year (1994/95) was dryer than the first one. Also this can point out the important role of the water saturated zone (SS3) in the catchment for the annual FDOC. Humid years have the effect that the extension of the water saturated zone increases (Latron [1990]), and greater amounts of soil carbon come under its influence. This should be especially important for the upper soil horizons, which can be expected to contribute much more DOC to the stream waters than the lower horizons. cDOC in soil waters normally decreases drastically in the lower horizons (Moore [1989], McDowell and Wood [1984]), except in wetlands (Moore and Jackson [1989]). In the rest of the catchment, morphology is often steep and the soils are generally well-drained (Latron [1990], El Gh'mari [1995]). Precipitation can thus rapidly percolate to the lower horizons. Soil leaching may here only be important in conjunction with elevated temperatures, which may accelerate the soil leaching. This can also explain why on average, the measured DOC concentrations in the Strengbach Brook are quite low.

1.3. A Medium Size Temperate River: The Garonne River Study

We change now to a larger scale and look at a medium size temperate river, the Garonne in France. Concentrations of DOC and POC in this river have been determined from 1989 to 1992 at two different stations: Portet in the upstream part of the basin (in the South of Toulouse), and La Réole in the downstream part of the basin (about 30 km downstream of Mas d'Agenais). The corresponding catchment areas are 9980 km2 and 52000 km2, respectively (Probst [1983]). Both stations were sampled in about weekly time intervals. The investigation was carried out by H. Etcheber and co-workers from the University of Bordeaux (Departement de Géologie et Géomorphologie), where also the carbon analyses were done (Veyssy et al, in press). The study was financed by the French research program DBT Fleuves et Erosion, ONT Garonne (INSU-CNRS).

1.3.1. Hydrology

The hydrology of the sampling period was characterized by exceptional dry conditions (Fig. 8a and 8b). From October 1989 to September 1992, three complete seasonal cycles were sampled. The average discharge at La Réole for this period was 403 m3/sec. All here cited annual discharge values are calculated with the daily means that were monitored by AFBAG (Agence Financière du Bassin Adour-Garonne). The first year (yr1) was the driest one with an average discharge of only 260 m3/sec, followed by the second year (yr2, 444 m3/sec), and by the third year (yr3, 504 m3/sec). For comparison, the average discharge at La Réole was about 637 m3/sec for the preceding 1977 to 1989 period (also starting in October and ending in September), which was more humid than on average. Probst and Tardy [1985] calculated the 1910 to 1980 average for the Garonne to be 600 m3/sec. Portet, the upstream station, contributed about one third of the total discharge at La Réole during the investigated period (35%, 34%, and 29% for yr1, yr2, and yr3, respectively). The contribution of the upstream station is more important during dry years than during wet ones (Semhi [1996]). Figure 8b shows that principal peak discharges at La Réole occurred also at Portet, although this was not always the case. Generally, there is a time lag of one to two days in peak discharges between the two stations.

20

A detailed study on the hydrology and the transport of dissolved and particulate matter of the Garonne River during the 1989 to 1992 period has been done by Semhi [1996].

77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92

0

500

1000

1500

2000

2500

Fig. 8a Mean monthly discharge of the Garonne River from 1977 to 1992 at La Réole (Data from AFBAG). The light grey signature shows the three investigated years.

Fig. 8b Comparison of thedaily discharges (m3/sec) at La Réole and at Portetduring the investigation period (Data from AFBAG).

0

500

1000

1500

2000

2500

3000La Réole Portet

Oct. 89 to Sept. 90 Oct. 90 to Sept. 91 Oct. 91 to Sept. 92

4790 m 3/sec

(12 /06 /92)

1.3.2. Dissolved Organic Carbon

The evolution of the concentrations of DOC, POC, and TSS in relation to discharge is shown in Figure 9a for Portet, and in Figure 9b for La Réole. Additionally, the percentage of organic carbon (POC%) in the total suspended solids is shown in both figures. For the reason of presentation facilities, only the Q values corresponding to the sampling times (instantaneous Q values) are shown, and not the daily values. The sampling density was high, and the general evolution of the river hydrographs are well represented. It has to be mentioned, however, that the maximum discharge peak for the investigation period, which occurred in June 1992 (Fig. 8b), was not sampled on its top.

Measured cDOC in the Garonne River range from 0.5 mg/l to 47.9 mg/l. Concentrations were generally higher in the waters of the upstream station than in the waters of the downstream station. Discharge-weighted mean annual DOC concentrations are 3.6 mg/l (yr1), 3.8 mg/l (yr2), and 7.7 mg/l (yr3) at Portet, and 2.9 mg/l, 2.7 mg/l, and 3.3 mg/l at La Réole, respectively. Note that all discharge-weighted means for the Garonne cited in this study (as well as for the other rivers that will be presented in the following) were calculated with instantaneous Q values (see also chapter V). Higher DOC concentrations in the upstream waters of a river compared to the downstream waters have also been observed elsewhere, e.g. in the Mississippi River (Barber et al. [1995]). This can be explained with the assumption that in the downstream river parts, considerable amounts of water may be contributed by ground waters, which are poor in DOC (Spitzy [1988]). Another possibility is that a part of DOC is mineralised during the downstream transport. On the basis of the discharge-weighted mean annual cDOC and the annual water discharge given above, the total DOC flux at La Réole was 115000

21

tC for the sampling period (yr1, 23900 tC; yr2, 38200 tC; yr3, 52900 tC). About 55% of this flux was contributed by the waters upstream Portet (yr1, 43.7%; yr2, 47.3%; yr3, 66.9%), if one supposes that no carbon was lost during transport.

At both stations, highest cDOC can be observed during peak discharges, or in the vicinity of peak discharges. A simultaneous evolution of Q and cDOC was the case, for example, in February 1990 or in May 1991 at La Réole, or in May 1991 at Portet. It is mainly found at Portet that maximum cDOC occurred also before (e.g., April/ May 1990) or after major discharge peaks (e.g., May/June 1991, or April/May 1992), normally situated on smaller Q peaks that followed the major peak (these smaller Q peaks cannot always be well seen in Fig. 9b because no samples were taken on their tops). Very striking are the high DOC concentrations encountered in spring and in late summer of 1992 at Portet. The maximum cDOC value achieved nearly 50 mg/l, which is very rarely found in river waters (see other studies presented in this chapter). One may think about whether it is possible that the samples were contaminated, but the trend of elevated concentrations is persistent over several samples, which is an argument against contamination. Moreover, also downstream at La Réole, greatest cDOC were measured in the same period, although only the cDOC peaks preceding the major Q peaks are in phase at both stations. A first peak of high cDOC that occurred earlier at Portet (February/ March) cannot be seen at La Réole. During the following high waters, DOC concentrations were elevated at both stations.

In June of the same year, when the Garonne had its greatest discharge of the sampling period (as mentioned above, neither at Portet nor at La Réole, the top of this discharge was sampled - see Fig. 8b), the evolutions at both stations were different. At La Réole, cDOC were in phase with Q, as it is here the usual case. At Portet, cDOC continuously dropped down during the high-water stage. Only at the end of the high-water stage (end of August/ September), again quite high cDOC were encountered. The fact that at Portet often greater cDOC occur together with a second but smaller Q peak that follows the major Q peak could be explained by an influence of the water level in the river on the water level in the soils of the alluvial plains. Soil levels which are normally water limited may be water recharged by the first rise of the hydrograph. These waters can thus leach additional carbon, which is then washed out with the falling water level and/or during a second rise of Q. Why, however, DOC concentrations rised that high at Portet at the end of summer 1992 is difficult to explain on the basis of the information that was available for the here presented study.

1.3.3. Particulate Organic Carbon and Total Suspended Solids

The concentrations of total suspended solids as well as the concentrations of particulate organic carbon (cPOC) were strongly coupled to the evolution of the river hydrograph. Both cTSS and cPOC were normally at low levels over most of the time, but they sharply increased with the onset of peak discharges, and during peak discharges. At both stations, this behaviour was similar. Often, maximum concentrations preceded the maximum of Q, and the concentrations dropped earlier down than the Q values decreased. This has been already found in the Strengbach catchment, and underlines again the quick flushing out of particles in drainage basins with initial rains. Looking at the evolution of POC%, a seasonal evolution can be recognized. At Portet, great POC% values are typical for the winter situation, while low values are typical in summer. Such a behaviour may be linked to the seasonality of the vegetation cycle. Litterfall occurs in late autumn, and in winter more organic particles should be on top of the soils than in summer. These particles are then successively washed out of the basin. At La Réole, the POC% evolution is more diffuse and high values can be also found in summer, indicating that here may also other factors be important (see below). With respect to the POC% evolution, another interesting feature can be observed. Each time when there is a discharge peak, the POC% values show a clear incision towards low values in their evolution, i.e. the mineral content of the suspended solids suddenly became greater. Related to the strong precipitation that often provokes

22

0

200

400

600

800

1000

0

250

500

750

1000

1250

1500

0

5

10

15

20

25

POC%

Q

DOC

TSS

POC

Days 1989

0

0

10

20

30

0

10

20

30

POC%

Q

DOC

TSS

POC

Days 1990

0

200

400

600

800

1000

0

250

500

750

1000

1250

1500

0

5

10

15

20

25

POC%

Q

DOC

TSS

POC

0 30 60 90 120 150 180 210 240 270 300 330 360 J F M A M J J A S O N D

Days 1991

0

0

10

20

30

0

10

20

30

0 30 60 90 120 150 180 210 240 270 300 330 360

POC%

Q

DOC

TSS

POC

47.9

T S S : 2 76 2

T S S : 5 052

P O C : 5 8.6

J F M A M J J A S O N DDays 1992

Fig. 9a Evolution of DOC, POC, and TSS concentrations in relation to discharge in the Garonne River from 1989 to 1992. The station is Portet in the upstream part of the basin. Note that the y-axis to the left depict discharge (m3/sec), while the y-axis to the right depict DOC (mg/l), POC (mg/l), and TSS (mg/l) concentrations, respectively. POC% is the percentage of organic carbon in the totalsuspended solids. the rise of Q, it is possible that gully and riverbed erosion were more important during peak discharges, leading to a greater share of mineral matter in the eroded particles. More likely is, however, that due to the increasing transport velocity of the river, a resuspension of mineral matter occurred that was previously deposited on the river floor.

Measured cTSS in the Garonne River varied from 1.3 mg/l to 5052 mg/l in the sampling period, and measured cPOC from 0.2 to 59 mg/l. Discharge-weighted annual mean cTSS are 196 mg/l (yr1),

23

0

500

1000

1500

2000

2500

3000

0200400600800

10001200

0

5

10

15

20

25

POC%

Q

DOC

TSS

POC

Days 1989

2

4

6

8

0

5

10

15

20

POC%

Q

DOC

TSS

POC

Days 1990

0

500

1000

1500

2000

2500

3000

0200400600800

10001200

0 30 60 90 120 150 180 210 240 270 300 330 360

0

5

10

15

20

25

POC%

Q

DOC

TSS

POC


2

4

6

8

0

5

10

15

20

0 30 60 90 120 150 180 210 240 270 300 330 360

POC%

Q

DOC

TSS

POC


Fig. 9b Evolution of DOC, POC, and TSS concentrations in relation to discharge in the GaronneRiver from 1989 to 1992. The station is La Réole in the downstream part of the basin. Note that the y-axis to the left depict discharge (m3/sec), while the y-axis to the right depict DOC (mg/l), POC (mg/l), and TSS (mg/l) concentrations, respectively. POC% is the percentage of organic carbon in the totalsuspended solids. 114 mg/l (yr2), and 288 mg/l (yr3) at Portet, and 156 mg/l (yr1), 156 mg/l (yr2), and 85 mg/l (yr3) at La Réole. For cPOC, this is 4.7 mg/l, 3.8 mg/l, and 4.8 mg/l at Portet, and 5.7 mg/l, 4.8 mg/l, and 3.7 mg/l at La Réole, respectively. Note, however, that in the case of TSS and POC it can be misleading to simply multiply these concentrations with the above given average discharge values in order to calculate annual TSS and POC fluxes. The generally observed sharp increase of cTSS and cPOC with the rise of Q together with the subsequent rapid decrease makes it very important to apply short sampling intervals during peak discharges for the calculation of realistic fluxes. For example, the great discharge peak that occurred in June 1992 (Fig. 8b) was not sampled, but it is important to know here

24

the concentrations for the flux calculations. When used for the calculation of TSS fluxes, mean annual discharge-weighted concentrations calculated on the basis of the instantaneous discharge and concentration values can considerably overestimate the real values (Probst [1983]). Semhi [1996] determined in a more sophisticated way (by weighting the concentrations with the water volumes that passed between two samples) the average FTSS and FPOC at La Réole to be 886840 t/yr and 34632 t/yr for the 1989-1992 period, respectively. Using the above given discharge-weighted mean concentrations and discharge values, the resulting fluxes would be about 85% (FTSS) and 66% (FPOC) greater (whereas the FDOC values given above is very close to the value found by Semhi [1996]). These problems of calculating realistic FTSS and FPOC have to be kept in mind also when looking at the cTSS and cPOC values of the other case studies that are discussed in the following (for this problem, see also chapter IV and V).

1.3.4. Autochthonous Carbon

There is another interesting feature that can be observed in the here presented data. The concentrations of dissolved organic carbon show an elevated level from July to October, where Q is lowest over the year. This is the case both for 1990 and for 1991, and it is also indicated by the incomplete data series that exist for 1989. At Portet, the upstream station, such an effect is absent. At the same time, cPOC during 1990 are more elevated relative to cTSS, and the POC% values are great (which is also the case for 1991). Normally, rather low POC% values can be expected at the end of summer and the begin of autumn, as discussed above. The sum of these observations is an indication that during the July to October periods, the share of autochthonous carbon increased in the downstream part of the Garonne River because of plankton blooms.

Assuming that the allochthonous DOC in the Garonne River should have evolved in parallel with Q, as it is the case at Portet, the additional autochthonous contributions to the annual DOC fluxes can be estimated. To do so, I fixed all DOC concentrations with the beginning of July to a value of 1.50 mg/l, supposing that this should have been approximately the allochthonous DOC concentration. This correction was maintained until the first increases of Q that occurred at the end of October or at the beginning of November. Then the this way corrected discharge-weighted mean annual cDOC were calculated for the three years of the investigation period, and the corresponding FDOC. The differences between these values and the above given FDOC can be considered to represent the autochthonous share in the total DOC fluxes. This yields autochthonous contributions to the corresponding mean annual DOC fluxes of 5.8% for the first year, 1.8% for the second year, and 1.5% for the third year. Apparently, autochthonous DOC was enhanced by the very dry conditions characterizing 1989/90. For particulate organic carbon, it is more difficult to estimate the contribution of in-situ production to total POC in the river. Because of the strong linking of cPOC to peak discharges one can suppose that this contribution should be even less than for DOC. Generally, it is indicated that autochthonous organic carbon was of very small importance for the organic matter export from the Garonne River basin during the three investigated years. Moreover, considering the fact that dryness seems to enhance the autochthonous contribution, it can be concluded that the organic matter export should be even more dominated by allochthonous carbon during average years.

However, one has naturally to mention here that the above discussed method is only an indirect way to estimate the ratio of allochthonous to autochthonous carbon in the total organic matter fluxes of the Garonne River. Further investigations based upon chlorophyll measurements or the use of other tracers for riverine in-situ production should be done to confirm this. Etcheber [1986] did a detailed chemical and physical characterization of POC at La Réole during a seasonal cycle in 1980. He found that autochthonous POC was the dominant fraction from April to September (corresponding to rather low water levels of the river), while allochthonous POC was dominant during the rest of the year (corresponding to rather high water levels). This is in agreement with the here presented results from a qualitative, but not from a quantitative point of view. When coupling the findings of Etcheber [1986] with the corresponding water fluxes during that time, it is indicated that autochthonous POC may contribute about one third of the total POC flux. This is considerably more than estimated above.

25

1.4. Major World Rivers

In the following, organic carbon data for major world rivers are presented. If not mentioned otherwise, all rivers have been investigated within the SCOPE program (see introduction), and the carbon analyses were done at the Geological/Palaeontological University of Hamburg. Methodical details are described in Michaelis and Ittekkot [1982]. The rivers were normally sampled close to their mouths in monthly time intervals for one to four years. The data reflect the sum of the hydroclimatic, biological, and geomorphological variability in the basins, which can be great (see chapter II). For this reason, it is more difficult to relate the observed patterns to certain processes which may be at their origin compared to small catchments or small river basins. Nevertheless, we will see that many observations already presented on the previous pages can also be found at the scale of large world rivers.

1.4.1. Boreal Climates

1.4.1.1. The Mackenzie River

Fig. 10 Drainage intensity in the Mackenzie Basin (after Korzoun et al. [1977] - see chapter III). SS is the sampling station for the organic carbon data, GS is the water gauging station at Norman Wells,and M is the river mouth.

The Mackenzie River is here discussed as an example for the organic matter transport in rivers of the boreal climates. The presented data are published in Telang [1985], and Telang et al. [1982], [1983], [1991], together with more information on the Mackenzie River study and additional data on the water chemistry of the river. A map of the drainage intensity in the Mackenzie Basin is shown in Figure 10.

The river was sampled between May 1981 and October 1983, with an interruption of 4 month from June to September 1982. The sampling station was situated above the confluence of the Arctic

26

Red River (Fig. 10), about 350 km downstream Norman Wells, one of the principal gauging stations of the river. Figure 11 depicts the evolution of Q, cDOC, cTSS, cPOC and C/N for the sampling period. Keep in mind that during the entire study, no complete discharge peak was sampled (Fig. 12). The Mackenzie has normally a sharp increase of discharge at the end of spring (May/June) as a consequence of snow melt in spring. The sampling started on top of the discharge maximum in 1981, but the two following discharge peaks that occurred in 1982 and in 1983 are lacking in the record. It is indicated from Figure 12 that the 1981 discharge maximum was greater than on average.

0

5000

10000

15000

20000

25000

30000

0

200

400

600

800

0 30 60 90 120 150 180 210 240 270 300 330 360

5

10

15

20

25

30

35

Q

DOC

TSS

POC

C/N

M J J A S O N D J J M ADays 81/82 (Begin: May 81)

M

2

4

6

8

10

0

5

10

15

20

25

0 30 60 90 120 150 180 210 240 270 300 330 360

Q

DOC

C/N

TSS

POC

O N D J F M A M J J A SDays 82/83 (Begin: Oct. 82)

O

Fig. 11 Evolution of Q, DOC, TSS, POC, and C/N in the Mackenzie River from 1981 to 1983. The y-axis to the left depict discharge (m3/sec), TSS (mg/l) concentration, and C/N, while the y-axis to the right depict DOC (mg/l) and POC (mg/l) concentrations, respectively. Note that there is an interruption of about 4 month between the two years.

cDOC varied between 2.8 mg/l and 9.6 mg/l for the first year (May 81 to May 82), and between 3.2 mg/l and 6.6 mg/l in the second year (Oct. 82 to Oct. 83). The discharge-weighted annual means are 5.9 mg/l and 3.9 mg/l, respectively. cDOC were highest on top and on the falling limb of the discharge peak in 1981. Then they decreased continuously, and reached lowest values in August, about 3 month after the maximum discharge. Already before Q dropped down to the low river flow that is typical during winter, cDOC rised again to a second but smaller maximum. Also the concentrations of cations and anions increased at this stage (Telang et al. [1983]). It is possible that here occurred an input to the river from the soil and groundwater reservoirs which have been filled up with water during the snow melting period. The incomplete data series of the second year are generally in good agreement with the above described evolution. It is worth to be noted that the small cDOC maximum which appeared in the first year in November can be found again in the second year, although nearly one month later (in both cases air temperatures were here already far below zero).

The concentrations of particulate organic carbon were strongly coupled to the concentrations of total suspended solids. cPOC varied between 0.2 mg/l and 22.2 mg/l in the first year, and between 0.2 mg/l and 8.2 mg/l in the second year. For cTSS, the ranges are 2.5 mg/l to 680 mg/l (yr1) and 1.4 mg/l to 326 mg/l (yr2), respectively. Discharge-weighted means are 8.8 mg/l (yr1) and 2.3 mg/l (yr2) for cPOC, and 281 mg/l (yr1) and 84 mg/l (yr2) for cTSS. Highest cPOC and cTSS values occurred on the falling limb of the discharge peak in summer 1981. Smaller maxima appeared in November 1981 and

27

in August 1982, without any evident relationship to discharge. C/N ratios were normally elevated in summer and tended to low values in winter. Note that the C/N ratios increased to values of more than 25, which is quite high in comparison to the values found in other rivers (see below). It is possible that the snowmelt runoff washed great amounts of litter debris into the river. Another explanation for the high C/N ratios is that the organic matter in the upper soil horizons in the Mackenzie basin may decompose more slowly compared to warmer climates, and the organic matter reflects thus more the composition of fresh vascular plant material.

0

5000

10000

15000

20000

25000

30000

73 74 75 76 77 78 79 80 81 82 83 84

Instantaneous Dischargeat Sampling Station

Fig. 12 Mean monthly discharge (m3/sec) of the Mackenzie River at Norman Wells since 1973 (Data from Global Runoff Data Center Koblenz [1991]). For comparison, the instantaneous discharge corresponding to the sampling times is shown. Note that the sampling station was about 350 km downstream of Norman Wells.

For the second year, the discharge-weighted mean annual concentrations of DOC, POC, and TSS are throughout lower than for the first year. This is certainly because the high-water period could not be sampled in the second year, and it points out the great influence of drainage for organic matter export from the basin. An important question that remained unanswered in the Mackenzie study is how the concentrations evolved during the rising limb of the hydrograph. This may especially concern the transport of POC and of TSS. As it can be seen in many of the studies presented in this chapter, there occurs often a sharp rise of the concentrations at early stages of the discharge increase. If this was also the case for the Mackenzie, then the here calculated discharge-weighted POC and TSS concentrations should be considerably lower than the real values.

1.4.2. Temperate Climates

1.4.2.1. The St. Lawrence River

The St. Lawrence River has been sampled for the longest time within the SCOPE program: from September 1981 to August 1985, four complete seasonal cycles were monitored in about be-weekly sampling intervals. The sampling station was upstream of Quebec City, 30 to 40 km above the saltwater intrusion from the Gulf of St. Lawrence. More information on the St. Lawrence River study can be found in Pocklington [1982], [1985], Pocklington and Tan [1983], [1987], Tan [1987], and Telang et al. [1991].

The instantaneous discharge values corresponding to all samples were not available for this study, and I present here only the monthly averages (Fig. 13), as they were published in Telang et al. [1991]. Additionally, I included in Figure 13 the δ13C isotope signature in the particulate matter of the samples. Also for this parameter the monthly averages of the values given in Tan [1987] were calculated.

28

20

30

40

50

60

0

10

20

30

40

6

8

10

12

14

Q

DOC

TSS

POC

C/N

d1 3 C

Sept. 81 to Aug. 82

2

3

4

5

6

0.0

0.5

1.0

1.5

2.0

2.5

-28

-27

-26

-25

-24

-23

-22

Q

DOC

TSS

POC

C/N

d1 3C

Sept. 82 to Aug. 83

20

30

40

50

60

0

10

20

30

40

6

8

10

12

14

Q

DOC

TSS

POC

C/N

d1 3 C

S O N D J F M A M J J A

Sept. 83 to Aug. 84

2

3

4

5

6

0.0

0.5

1.0

1.5

2.0

2.5

-28

-27

-26

-25

-24

-23

-22

TSS

POC

C/N

d1 3C

Q

DOC

S O N D J F M A M J J A

Sept. 84 to Aug. 85 Fig. 13 Evolution of Q, DOC, TSS, POC, C/N, and δ13C in the St. Lawrence River from 1981 to 1985. Note that the y-axis to the left depict discharge (km3/month), TSS (mg/l) concentration, and C/N, while the y-axis to the right depict DOC (mg/l) concentration, POC (mg/l) concentration, and δ13C (‰), respectively.

Total freshwater discharge of the St. Lawrence varied not much during the sampling period. It was 406.4 km3, 408.6 km3, 419.2 km3, and 418.0 km3 for the four successive years, and thus only about 2-3% greater in the second half of the sampling period than in the first half. Maximum discharge occurred in spring (April/ May). Monthly mean DOC concentrations varied in the range from 2.7 mg/l to 5.4 mg/l. The discharge-weighted annual means are 4.2 mg/l (yr1), 3.9 mg/l (yr2), 3.9 mg/l (yr3), and 3.1 mg/l (yr4), respectively (note that all discharge-weighted concentrations for the St. Lawrence were calculated with the monthly means). Contrary to the other case studies that have been presented

29

on the previous pages, no significant correlation of cDOC with discharge can be found. The only consistency in the sequence of cDOC is that the values were generally lower in summer than in winter and spring. Highest concentrations occurred in autumn.

73 74 75 76 77 78 79 80 81 82 83 84 85

3000

6000

9000

12000

15000

18000

21000

J F M A M J J A S O N D 20

40

60

80

100Québec City

Cornwall

Q%-upstream

Fig. 14 Mean monthly discharge (m3/sec) of the St. Lawrence River at Cornwall and at Québec City(sampling station). Data at Cornwall are from Global Runoff Data Center Koblenz [1991], and data at Québec City are from Telang et al. [1991] The small graph in the upper left corner ("Q%-upstream") depicts the mean monthly water contribution from Cornwall to the water discharge atQuébec City in percent (averaged from 1982 to 1984).

Monthly mean concentrations of POC ranged from 0.3 mg/l to 2.0 mg/l, and those of TSS from 4.3 mg/l to 31.5 mg/l. The discharge-weighted annual means are 0.6 mg/l (yr1), 0.7 mg/l (yr2), 0.9 mg/l (yr3), and 0.9 mg/l (yr4) for POC, and 9.2 mg/l (yr1), 13.1 mg/l (yr2), 15.9 mg/l (yr3), and 11.8 mg/l (yr4) for TSS, respectively. With these values, the St. Lawrence River has the lowest concentrations of suspended matter of all case studies that are presented in this chapter. Again, cPOC and cTSS are closely correlated. The pattern of variation in concentration of both species was consistent with respect to the season: low in winter, high in spring, low in summer, and high again in autumn. cPOC and cTSS were significantly higher during the months of greatest discharge, and lower during the months of lowest discharge. In the first year, a slight flushing effect can be observed for cPOC and cTSS at the beginning of the rise of discharge in spring. But generally, both concentrations followed more continuously the evolution of Q. Note, however, that the maximum concentrations during the sampling period occurred for TSS in autumn 1983, and for POC in autumn 1994 on small discharge peaks that are typical during this season, but not on the major discharge peaks in spring (see below).

The ratio of carbon to nitrogen in the particulate organic matter ranged from 8.5 to 12.7. It was consistently lowest in July and greatest in November. This feature may simply reflect the vegetation cycle in the St. Lawrence basin, because litterfall in autumn increases the amount of relative fresh organic debris on top of the soils, which may increase the C/N ratios. It is also possible that a part of the low C/N values in summer can be related to primary production in the river waters. The stable carbon isotope signature in the particulate organic matter showed more negative values in spring when Q was greatest (-26‰ to -27‰), and less negative values in late summer and in early autumn when Q was lowest (-25‰ to >-23‰). This pattern of change in δ13C is consistent over the entire sampling period, and the discharge-weighted annual means are very close for all investigated years. Pocklington

30

and Tan [1987] supposed the seasonal pattern of δ13C and C/N to reflect the various contributions of allochthonous and autochthonous carbon sources in the St. Lawrence River. They interpreted great δ13C values and low C/N ratios to be due to in-situ primary production, being a superimposed organic carbon contribution in summer on the generally more dominant contribution of allochthonous carbon to the river waters. However, Mariotti et al. [1991] (and references therein) reported that lacustrian and riverine phytoplankton can be particularly impoverished in δ13C up to -40‰, which is in disagreement with such an interpretation.

J F M A M J J A S O N D

40

50

60

70

80

-28

-27

-26

-25

-24

-23

Q%-upstream

d C 13

Fig. 15 Comparison of the monthly water contribution from Cornwall (Q%-upstream - see Fig. 14) in percent (left y-axis) with the mean monthly isotope signature in permille (right y-axis) of suspended matter in the waters at Québec City (averaged from Fig. 13).

I find in this study that the seasonal pattern of the δ13C carbon isotope signature in the particulate organic matter of the St. Lawrence River cannot be well understood without looking more in detail at the hydrological patterns of the river. Figure 14 depicts its monthly discharge both at Cornwall and at Québec City, the sampling station. There join no important tributaries to the river between the Ontario Lake and Cornwall, and the Cornwall gauging station reflects hence the water contribution from the great lakes in the upstream part of the basin ("Q%-upstream" in Fig. 14). Further downstream at Québec city, however, the mean annual discharge increases about 40% through tributary contributions from the downstream part of the basin. Seasonally, both stations have their discharge maximum in spring and a smaller maximum at the end of autumn. But it is a striking feature that in spring the downstream basin part contributes much more to the total discharge than during the other seasons. The sampling station at Québec City is thus characterized by a variable mixing of water masses over the year. As it can be seen from Figure 15, the δ13C carbon isotope signature in the particulate organic matter essentially reflects this mixing. The throughout low values indicate that POC in the St. Lawrence River mainly originates from organic matter derived from C3 plants (e.g., Pocklington and Tan [1983], Bird et al. [1992]), which can also be expected in this climate. But it is not unlikely that the abundant vegetation types in the upstream and downstream basin parts may have somewhat different δ13C signatures, which is then also reflected by the waters draining these basin parts. The explanation of the carbon isotope data in the St. Lawrence by a simple mixing of water masses with different carbon isotope signatures is consistent with the fact that the discharge-weighted annual δ13C averages are nearly identical for the three years where this parameter has been investigated. Also the mean annual water contributions from the upstream and downstream parts of the basins remained constant.

The variable mixing of the waters from upstream and downstream the basin should naturally also have an influence on the cDOC and cPOC patterns that were observed at Québec City. Because of

31

the great water contribution to the discharge peak in spring, one can suppose that the evolution of cDOC and cPOC during this season mainly reflects the carbon transport in the downstream waters. This is not necessarily the case in autumn and in winter, when most of the river water comes from upstream. It is interesting that the autumn discharge peak often seems to have a greater impact on the organic carbon concentrations than the spring discharge peak, although the latter is much greater in its amplitude. This can be seen for DOC in the first year, the second year, and the third year, and for POC in the third year and the fourth year. This is probably related to the hydrology of the great lakes in the upstream part of the St. Lawrence basin. In lakes of the temperate climates, the thermal stratification of the water column that established during summer breaks normally down in autumn with the beginning of the first storms. This remixes the lake water and may mobilize additional DOC that accumulated in the deeper layers or it may mobilize POC through the resuspension of particles.

Such hypothesis should naturally be confirmed by further investigations. Nevertheless, keep in mind that the St. Lawrence River study is probably a good example for a river where the interference of different hydrological patterns (dominated by the great lakes upstream and by spring discharge downstream) makes it difficult to interpret the signal of the organic carbon fluxes recorded at the river mouth.

1.4.2.2. The Waikato River

From November 1981 to October 1984, the Waikato River in New Zealand was investigated during three years within the SCOPE program. The sampling station was at Mercer, 20-30 km upstream the river mouth. Field work was carried out by the Auckland Regional Authority (ARA). ARA monitored the quality of the river water at this station already since 1969 with the purpose to use the river as a supplementary source of potable water for Auckland city. The here presented carbon data have not yet been published, but ARA kindly agreed their use in this study.

0

100

200

300

400

500

600

Ngaruawahia Mercer

76 77 78 79 80 81 82 83 84

Fig. 16 Mean monthly discharge (m3/sec) of the Waikato River at Ngaruawahia in comparison with discharge at Mercer during the sampling period. The latter are the instantaneous values which correspond to the sampling times

The mean monthly discharge (1976-1984) values of the Waikato River at Ngaruawahia are shown in Figure 16, together with the instantaneous Q values at Mercer corresponding to the sampling times. Ngaruawahia is about 50 km upstream Mercer, and three tributaries join the Waikato between these two stations. On the basis of the instantaneous Q values, the average river discharge at Mercer can be estimated to be 263 m3/sec for the first year, 214 m3/sec for the second year, and 308 m3/sec for the third year (each year starting in November and ending in October). Note that the main discharge peak of the river (which normally occurs in southern spring) was very low in 1982 compared to the 1976-1984 period. The Waikato passes on its way from upstream to downstream through a series of eight hydroelectric power stations, each with its associated dam and lake structure which exert a

32

regulatory effect on the river flow. The discharge pattern between lakes is therefore complex, and is largely dependent on demands for power generation. Large scale fluctuations due to tributary inflows

100

200

300

400

500

0

10

20

30

40

4

6

8

10

12

14

C/N

Q

DOC

TSS

POC

Days 81/82 (Begin: Nov. 81)

0

5

10

15

20

25

0

1

2

3

4

0 30 60 90 120 150 180 210 240 270 300 330 360

C/N

Q

DOC

TSS

POC

N D J F M A M J J A S ODays 82/83 (Begin: Nov. 82)

100

200

300

400

500

0

5

10

15

20

25

0

10

20

30

40

0

1

2

3

4

0 30 60 90 120 150 180 210 240 270 300 330 360

4

6

8

10

12

14

C/N

Q

DOC

TSS

POC

N D J F M A M J J A S ODays 83/84 (Begin: Nov. 83)

Fig. 17 Evolution of Q, DOC, TSS, POC,and C/N in the Waikato River fromNovember 1981 to October 1984. Note thatthe y-axis to the left depict discharge(km3/sec), TSS (mg/l) concentration, andC/N, while the y-axis to the right depictDOC (mg/l) and POC (mg/l) concentration,respectively.

are dampened out, and it is only in the lower river reach beyond Karapiro (corresponding to about 2/3 of the basin area) that the more usual river flow characteristics are observed. A detailed description of the hydrology and water chemistry of the Waikato River from upstream to downstream can be found in Bryers [1985].

The concentrations of dissolved organic carbon at Mercer varied from 1.4 mg/l to 24.6 mg/l during the sampling period (Fig. 17). The discharge-weighted annual cDOC averages are 7.3 mg/l, 6.8 mg/l, and 3.4 mg/l for the three successive years, respectively. No relationship between cDOC and Q can be observed, but a distinct seasonal pattern is prominent: both in 1983 and in 1984, a sharp increase of cDOC occurred in summer (February). In 1982, no sample was taken in February. The

33

summer cDOC maximum was much greater in 1983 than in 1984. Note that the 1983 maximum followed the unusual low spring discharge in 1982. Already in the case of the Garonne River, indications were found that dry years may lead to an enhancement of the autochthonous share in riverine DOC during summer. It is therefore likely that the February DOC peaks in the Waikato are due to increased primary production and, related to this, the decomposition of in-situ produced organic matter in the river waters.

Assuming that it is only in February that autochthonous DOC is important in the river, the contribution of this carbon with respect to the total amount of DOC that is exported from the basin can be estimated. To do so, I fixed the contribution of allochthonous DOC in February to the value that results from a linear interpolation between the January and the March cDOC values. The difference between this interpolated value and the measured value in February should reflect the autochthonous DOC. With this first-order-approximation, it can be calculated that in the second year, autochthonous DOC contributed about 17% to the annual FDOC, while this was about 16% in the third year.

The concentrations of particulate matter in the Waikato were low during the sampling period: cTSS varied between 4.4 mg/l and 29.6 mg/l, and cPOC between 0.4 mg/l and 3.5 mg/l. The discharge-weighted means are rather constant for both parameters (10.5 mg/l, 13.1 mg/l, and 15.0 mg/l for cTSS, and 1.4 mg/l, 1.5 mg/l, and 1.3 mg/l for cPOC for the three successive years, respectively). cTSS and cPOC are strongly correlated, but as in the case of cDOC, no relationship with Q can be seen. Only in the first year, cTSS and cPOC decreased in parallel with Q on the falling limb of the great discharge peak in spring 1981 (Fig. 16 and 17). For the rest of the sampling period, highest cTSS and cPOC values occurred between the end of spring and the beginning of autumn when Q was low.

The C/N ratios show only in the first year the typical feature of low values in summer and elevated values at the end of autumn and in winter, as this has been found in the St. Lawrence River. In the two following years, C/N values were still elevated in winter, but great values are also found in summer. Supposing that the summer periods might have been influenced by in-situ production in the river and its storage-lakes, this is not in agreement with the assumption that in-situ production should rather lead to low C/N values (see above). Keep in mind, however, that low C/N values should be typical for phytoplankton only. Especially in the marine environment, low C/N values have been found to be a useful indicator for the contribution of phytoplankton-derived organic matter in surface sediments or sediment traps in comparison to the contribution from land-derived organic matter (e.g., Hedges et al. [1988]). In rivers, periods with enhanced in-situ production may also be accompanied by an abundance of macrophytes, which may have C/N ratios close to those of terrestrial land plants.

An important question in the case of the Waikato River study is to what extent the discharge patterns recorded at the sampling station are representative for the hydrology of the entire drainage basin. As mentioned above, the natural discharge patterns of the river are regulated by man through the eight dams that have been constructed along the river course. At first sight, no clear relationships can be detected between discharge and the concentrations of organic matter in the river, at least as far as the seasonal patterns are concerned. Looking at the interannual variations, however, it is striking that cDOC decreased more or less continuously from the end of 1981 to the end of 1984, neglecting here of course the marked cDOC peaks in February. Probably the great discharge peak in spring 1981 had a considerable effect on the mobilization of DOC in the basin, but this effect was retarded by the artificial discharge regulation through the damming of the Waikato (see also below the Orange River study).

1.4.3. Tropical Wet Climates

1.4.3.1. The Orinoco River

Among the rivers of the tropical wet climate, it is the Orinoco River that has been investigated the most intensively with respect to its load of carbon and other major elements. Within the SCOPE

34

program, the Orinoco was sampled from February 1981 to April 1984 (with an interruption from August 1982 to April 1983) together with one of its major tributaries, the Caroní River (Németh et al. [1982], Paolini et al. [1983], [1987]). Additional data for many of its tributaries are given in Depetris and Paolini [1991], although only corresponding to the rainy season. In addition to this study, the Orinoco River was investigated for at least three successive seasonal cycles from 1982 to 1985 by Lewis and Saunders [1989], accompanied by detailed studies of two of its tributaries: the Caura River (Lewis et al. [1986], [1987]) and the Apure River (Saunders and Lewis [1988]).

The sampling station within the SCOPE study was Ciudad Bolivar. Lewis and Saunders [1989] sampled the river about 200 km further downstream at Barrancas, just above the Orinoco delta. The seasonal patterns of the organic matter transport at Ciudad Bolivar is shown in Figure 18 for the first half of the SCOPE study. For comparison, Figure 19 depicts the same parameters but from the study of Lewis and Saunders [1989]. Note that the latter graph shows the average seasonal patterns determined by averaging the values of corresponding month in the investigation period. This exerts naturally a smoothing effect upon the values. Nevertheless, it can be seen that both field studies found seasonal patterns that are very similar. Because the study of Lewis and Saunders [1989] covers the longer time period and comprises a greater percentage of the entire basin, all values (and basic information on the general characteristics of the river) which are given in the following are taken from this study.

An important element characterizing the hydrology of the Orinoco River is the interaction with its extensive floodplain. Discharge is generally at its minimum in March/ April and starts then to rise. The Orinoco enters its floodplain typically during late May or June and reaches its maximum discharge in August/ September. It is normally during November when the river leaves the floodplain. This typical feature was abundant over the entire study period, although the discharge was 10 to 15% greater for the first two years than for the second two years. The mean discharge for the entire study was about 36000 m3/sec, which corresponds to the long-term average (see also chapter III).

The discharge-weighted mean DOC concentration for the entire sampling period is 4.4 mg/l. cDOC were generally lowest in early spring when also discharge was at its minimum. Highest cDOC occurred on the rising limb of the summer discharge peaks, and cDOC decreased then already before the decrease of Q began (flushing effect). Within individual years, this decrease of cDOC could begin at earlier or later stages during the water rise, and on average cDOC remained elevated over most of the high-water period (Fig. 19), mainly corresponding to the time interval when the river was in contact with its floodplain. This behaviour is indicating an accumulation of soluble organic carbon in the soil and vegetation reservoirs during the season of low water, when flushing of the system is minimal. The onset of seasonal runoff then removes this soluble organic matter. The increase of cDOC with the increase of Q is in contrast to the behaviour of most dissolved inorganic solids, which are diluted through the water rise. Lewis and Saunders [1989] suggested therefore also that the observed flushing of DOC could be partly determined by dilution of inorganic solids, which may affect the solubility of organic matter.

The discharge-weighted mean concentrations of TSS and of POC are 80 mg/l and 1.4 mg/l, respectively (again calculated as average of the entire study period). cTSS and cPOC are strongly correlated. A clear flushing effect is observed for the transport of suspended matter. The initial abrupt increase in concentration just as the river level begins to rise can be accounted for jointly by the increase in mechanical erosion rates which accompanies increased runoff, and by the rise of the water level within the channel, which causes resuspension of fine material that was previously deposited during the interval of declining discharges. Flushing of TSS is a common feature of most of the here presented case studies. But it is interesting to note that in addition to the flushing effect a second rise of the concentrations of particulate material occurred in the Orinoco during the falling water stage of each hydrologic cycle. This may reflect the separation of the river from the floodplain, which interrupts the sedimentation losses to the floodplain, and returns some of the stored material to the channel (Meade et al. [1983]).

35

0

20

40

60

80

0 30 60 90 120 150 180 210 240 270 300 330 360

0

50

100

150

200

Q

DOC

TSS

POC

J F M A M J J A S O N D Days 81

0

5

10

15

0 30 60 90 120 150 180 210 240 270 300 330 360

0

1

2

3

Q

DOC

TSS

POC


Fig. 18 Evolution of Q, DOC, TSS, and POC in the Orinoco River from February 1981 to July 1982.The sampling station was Ciudad Bolivar (data from: Depetris and Paolini [1991]). Note that the y-axis to the left depict discharge (103 m3/sec) and TSS (mg/l) concentration, while the y-axis to the right depict DOC (mg/l) and POC (mg/l) concentration, respectively.

0

20

40

60

80

0.0

2.5

5.0

7.5

10.0

0

50

100

150

200

0

1

2

3

Q

DOC

TSS

POC

J F M A M J J A S O N DAverage 1982 to 1985

Fig. 19 Average seasonal patterns of Q, DOC, TSS, and POC in the Orinoco River at Barrancas (adapted from Lewis and Saunders [1989], Fig. 9, p. 225). Note that the y-axis to the left depict discharge (103 m3/sec) and TSS (mg/l) concentration, while the y-axis to the right depict DOC (mg/l) and POC (mg/l) concentration, respectively.

Within the Orinoco study also the portion of living organic matter has been determined in the total POC flux. Lewis and Saunders [1989] found that only 0.15% of total FPOC can be attributed to phytoplankton, 0.02% to zooplankton, and 1.7% to bacteria. The low transport of phytoplankton in particular reflects the tendency of the floodplain to retain phytoplankton biomass, and the failure of phytoplankton to grow in transit in the channel because of low light exposure in the deep turbid water column.

1.4.3.2. The Paraná River

The Paraná River was sampled between March 1981 and November 1984 near the cities of Santa Fe and Paraná, about 600 km above the mouth. Samples were collected from below the surface in the central part of the main channel. Additional water samples from the river and its major tributaries were collected during July-August 1985 over a 1300 km long reach, starting 1930 km upstream the mouth at the Iguazú River and continuing along the mainstream as far downstream as Santa Fe. A great number of publications resulted from the Paraná case study: Depetris and Lenardón [1982], [1983], Depetris and Cascante [1985], [1987], Depetris and Kempe [1990], [1993], Kempe and Depetris [1990], [1992], and Depetris and Paolini [1991].

36

1981 1982 1983

0

2

4

6

8

Begin ENSO-affected periodMean Hydrograph

Fig. 20 Paraná's water heights (m) measured in the city of Paraná from March 1981 to April 1984. The mean seasonal river hydrograph at this station is shown for comparison (figure taken from Depetris and Cascante [1985]).

In the lower Paraná at Santa Fe, the river exhibits normally highest discharges during January and February, and lowest discharges during July and August, i.e. in the southern winter (Fig. 20). During the sampling period, however, the river was affected by an extraordinary flood caused by the 1982 El Niño/ Southern Oscillation (ENSO) climatic anomaly (May 1982 to June 1983). Above-average discharges were recorded in the Paraná between January 1982 and March 1984. The unusual wet conditions over South America started thus a few month earlier than the heating up of the Pacific and outlasted the end of ENSO by eight month (Fig. 20). Because the stage to volume relation is not very well defined for these exceptionally high stages, the total discharge can only be estimated. It probably amounted to 1400 km3 between June 1982 and April 1984, which is roughly an increase of 75% over the long-term average (Depetris and Kempe [1990]). The valley of the lower Paraná was completely flooded, including all its levees, channels and ponds, turning it into a featureless river 30 km wide. For the reason of this exceptional hydrological conditions, in the following only the pre-ENSO (Mar. 81 to May 82) and the ENSO (Jun. 82 to Feb. 84) periods are compared, and not the different seasonal cycles of the sampling period.

The discharge-weighted mean concentrations of total suspended solids decreased by about 50% during the ENSO event compared to the pre-ENSO period (from 142 mg/l to 73 mg/l). Estimating the corresponding water discharges to be 770 km3 and 440 km3 when related to an annual basis (Depetris and Kempe [1990]), the fluxes of TSS nearly stayed at about the same level. The ENSO produced flood did not trigger a flushing-out of sediments as it might be expected intuitively. This can be related to the significant widening of the channel during the flooding, causing a decrease in current velocity and hence in carrying capacity. During the sampling period, cTSS showed several prominent peaks (Fig. 21). The first peak (April 1981) coincided with the draining of the floodplain. Then, cTSS remained low during the following low-water stage. The concentrations increased again on the rising limb of the hydrograph, but rapidly dropped down thereafter, showing thus a typical flushing effect. The highest measured cTSS (500 mg/l) occurred, however, at the end of the record (March-April 1984), i.e. during the receding stage when the floodplain was about to be totally drained and the river had returned to its main channel. Already in the case of the Orinoco, we have seen that cTSS can rise during the retreat of the river from its floodplain. In the Paraná, this second increase of cTSS at the end of the high-water period was even more pronounced than the first rise at the beginning of the high-water period. It is, of course, not clear whether this would be also the case for more usual conditions, or whether this feature was only found because of the exceptional flooding.

37

01234567

0

100

200

300400

500

0102030

?

1981 1982 1983

ENSO - induced floodpre - ENSO period

C/N

C/N

POC (mg/l)

TSS (mg/l)

DOC (mg/l)

Gauge Height (m)

Fig. 21 Variation of Q, DOC, TSS, POC, and C/N in the Paraná River from March 1981 to April 1984 (figure taken from Depetris and Kempe [1993]).

The discharge-weighted mean POC concentrations decreased in the ENSO period to nearly one third of the value of the pre-ENSO period (from 3.5 mg/l to 1.3 mg/l). This decrease is more pronounced than for TSS, and the total POC load also decreased to about 60% of its previous value. The evolution of cPOC followed more or less the evolution of cTSS, but correlation between both parameters is only weak (Fig. 21), in contrast to most of the here presented studies. Consequently, POC% varied in the wide range of 0.4 to 33.0%. Especially the flushing effect at the beginning of the

38

high-water period was much more pronounced for cPOC than for cTSS. The C/N ratios showed in the pre-ENSO period more elevated values in autumn and in winter compared to the values in spring and in summer. During ENSO, C/N values were generally lower than in the pre-ENSO period (the discharge-weighted means for both periods are 11.5 and 9.8, respectively), with the tendency of still decreasing values towards the end of the ENSO triggered flood. An exception are the two smaller cPOC peaks that appeared in July 1982 and in May 1983 during the ENSO flood: they were accompanied by a sharp increase of the C/N ratios. The general evolution of the C/N ratios is indicating an ongoing washing out of relative fresh vascular plant debris from the basin, together with a possible increase of the autochthonous part in the riverine POC.

Dissolved organic carbon concentrations were found to be highly variable in the Paraná River (Fig. 21), ranging from 3.6 mg/l to 29.4 mg/l in the entire sampling period. The discharge-weighted mean cDOC increased from 6.1 mg/l (pre-ENSO) to 10.1 mg/l (ENSO), and the mass transport during ENSO increased therefore by a factor of 2.6. No relationship of cDOC with Q can be detected. Very striking are the sharp cDOC peaks that occurred several times during ENSO, leading to values that are among the highest reported from large rivers world-wide. Such sharp peaks were absent during the pre-ENSO period. For me it is not unlikely that these high cDOC peaks were related to the exceptional inundation of the floodplain. Boggy ponds and bogs might have been formed that were sometimes isolated from the main river channel, getting thus enriched in DOC through the rapid decomposition of organic matter of the inundated vegetation and soil pools. When these waters became again in contact with the main channel, they may have instantaneously delivered great amounts of DOC to the river, leading to the observed cDOC pattern.

1.4.3.3. The Ubangi River (Zaire Basin)

The Zaire (or Congo) River is the second largest river of the world after the Amazon River (see chapter III). Up to now, no long-term surveys of the transport patterns of organic matter comparable to the here presented studies were carried out in these two rivers. Only for the Amazon, DOC and POC was studied in several samples that were taken over a 1800 km long reach of the river on eight cruises (1982-1984) at different stages of the river hydrograph (Richey et al. [1990]). One has to mention here that both for the Zaire and for the Amazon rivers it may be more difficult to interpret the seasonal patterns recorded at the river mouths compared to other major world rivers, without knowing the contributions from their tributaries. A part of the tributaries originate from North of the equator, while the other part originate from South of the equator, having thus opposite seasonal discharge patterns (Nkounkou [1989], Nkounkou and Probst [1987]). For this reason, the signals encountered in the main channels represent to some extent seasonal averages.

Looking at the organic carbon load in the tributaries should be a better mean to detect the seasonal transport patterns. I present in the following DOC data for the Ubangi River, the second largest tributary of the Zaire. The Ubangi was investigated within the French research programs PIRAT-GBF until 1990 and PEGI-GBF (both INSU-CNRS) since 1990, in collaboration with the French research organisation ORSTOM. Water samples were collected since 1988 at Bangui, and analysed for DOC since 1992 at the CGS in Strasbourg after shipping them to France. The here presented data cover the three seasonal cycles from July 1992 to July 1995. Didier Orange kindly supplied the corresponding discharge data (Orange et al. [1995]). Additional information on the Ubangi River and its major element transport can be found, for example, in Olivry et al. [1988], in Probst et al. [1992] and [1994b], or in Orange et al. [1996].

The Ubangi drains the northern part of the Zaire Basin, where drainage intensity is lower than in the central part of the basin (Fig. 22). The mean water discharge of the Ubangi is 4080 m3/sec, about 1/10 of the value that is found for the Zaire River at Brazzaville (40900 m3/sec; data from Olivry et al. [1988]). The corresponding basin areas are 0.49 x 106 km2 and 3.5 x 106 km2, respectively. This is leading to a mean drainage intensity of 270 mm for the Ubangi Basin, compared to 370 mm for the entire Zaire Basin. The distribution of vegetation follows the climatic patterns. Dense tropical

39

rainforest is abundant in the central part of the Zaire Basin. Going further North, the forest becomes more open, being also partly mixed with woody savanna. Because of the above mentioned reasons, the variability of the Ubangi's hydrograph is much greater than the variability of the Zaire's hydrograph, which is true both with respect to seasonal and interannual variability (Fig. 23). The water level in the Ubangi is lowest in spring (March) and highest in autumn (October).

Fig. 22 Drainage intensity inthe Zaire (Congo) Basin (afterKorzoun et al. [1977]). SS is the sampling station for theUbangi tributary at Bangui, GSis the principal gauging stationof the Zaire River atBrazzaville, and M is the rivermouth.

0

5

10

15

20

25

J F M A M J J A S O N D0

5

10

15

20

Ubangi at Bangui

mean annual Q: 4257 m /sec3

(1936 to 1975 average)

Zaire (Congo) at Brazzaville

mean annual Q: 41117 m /sec 3

(1971 to 1983 average)

Percent of mean annual Q

Percent of mean annual Q

Fig. 23 Comparison of the meanhydrographs of the Ubangi River at Banguiand of the Zaire River at Brazzaville (Datafrom Global Runoff Data Center Koblenz[1991]). Note that at Brazzaville, thehydrograph of the Zaire River is lessvariable because of the confluence oftributaries from the two hemispheres, whichhave opposite seasonal discharge patterns.

During the sampling period, the average discharge of the Ubangi was clearly below the long-term average. For the three successive years (starting in July 1992 and ending in June 1995), mean Q was 3040 m3/sec, 2750 m3/sec, and 3320 m3/sec, respectively. The concentrations of DOC varied between 1.6 mg/l and 9.1 mg/l over the entire period, and the discharge-weighted annual means of cDOC are 4.7 mg/l (yr1), 4.5 mg/l (yr2), and 5.7 mg/l (yr3). Generally, cDOC show a good correlation with Q. Lowest cDOC occurred during the low-water stage, and highest cDOC during the high-water stage. A slight flushing effect can be observed, but cDOC stayed at elevated levels throughout the high-water period. The general cDOC evolution compares well, for example, with the observations found for the Orinoco River (see above).

40

0

2000

4000

6000

8000

10000Q

DOC

Days 92/93 (Begin: Jul. 92)

0 30 60 90 120 150 180 210 240 270 300 330 360

2

4

6

8

10Q

DOC

J A S O N D J F M A M JDays 93/94 (Begin: Jul. 93)

0 30 60 90 120 150 180 210 240 270 300 330 360

0

2000

4000

6000

8000

10000

2

4

6

8

10Q

DOC

J A S O N D J F M A M JDays 94/95 (Begin: Jul. 94)

Fig. 24 Evolution of DOC concentration(mg/l, right y-axis) in relation to discharge(m3/sec, left y-axis) in the Ubangi from July1992 to July 1995. Note that the Q valuesare daily means (Orange et al. [1995]).

There is, however, an interesting additional observation that can be made for the Ubangi. As shown by the daily discharge values (Fig. 24), the water level in autumn was not continuously peaking, but normally two peaks occurred on top of the high-water period: one in late September, and one in November. The latter achieved the higher water level. This two peaks are found in the first and in the third year, whereas in the second year the autumn discharge reached only a much lower water level, and the feature of two peaks was absent. cDOC, although generally decreasing after the initial rise of the waters, show the tendency to increase again during the second discharge maximum. This increase is very striking in the third year following the dry 1993/94 period, and resulted in the highest cDOC value that was measured in the Ubangi. It is possible that the marked rise of cDOC during the second discharge peak was related to the interaction of the river with the soil and vegetation pools in the surrounding alluvial plains, allowing during high water levels a leaching of organic carbon which is not possible during lower water levels. Since the second year was very dry, this leaching was here not possible. Consequently, organic carbon could accumulate, leading to a high cDOC peak in the following year. Another possibility is naturally also that the two peaks on top of the Ubangi's high-water stage represent different water masses from further upstream with different DOC signatures. Especially the contribution from the Uele tributary may be rich in DOC because this river drains the part of the Ubangi basin where tropical forest is the most abundant (D. Orange, pers. commun.).

1.4.4. Tropical Dry Climates

1.4.4.1. The Niger River

As this is found for many rivers under tropical dry climate (see chapter II), the drainage basin of the Niger is characterized by large contrasts. Drainage intensity is great both in the headwater region and in the downstream part of the basin, but in between, very dry conditions govern the basin (Fig. 25). The Niger looses about one third of its water by evaporation when the river passes through this intermediate part (see Table 2 and chapter III). Within the SCOPE program, the river was monitored during one seasonal cycle from May 1980 to April 1981 at Lokoja, above the confluence of its major tributary, the Benoue River. Additional samples from upstream this station and from one of the river's upstream tributaries were analysed as well. Results and other basic information of the Niger case study can be found in Martins [1982], [1983], and Martins and Probst [1991]. In addition to the SCOPE study, I present in the following also data from samples that have been collected in the uppermost basin part at Bamako. These data comprise the period from December 1991 to July 1993 (Boeglin and Probst [1996]). The Niger was sampled at this station by J. Boeglin (ORSTOM). DOC analyses were made at the CGS in Strasbourg.

41

Fig. 25 Drainage intensity in the Niger Basin (after Korzoun et al. [1977]). SS1, Bamako/ Koulikoro (the upstream sampling station); GS, gauging station at Niamey; SS2, Jebba; SS3, Confluence of theKaduna tributary; SS4, Lokoja (the principal sampling station within the SCOPE study); M, rivermouth.

When estimated from the instantaneous Q values corresponding to the sampling times (Fig. 26), mean discharge at Lokoja was 5182 m3/sec during the investigation period. This is only about 6% greater than the 1970-1980 average (Table 2). By far most of this water comes from the downstream part of the basin. At Niamey (a river gauging station that represents the upstream basin part including the very dry basin part - see Fig. 25), the average discharge during the investigation period was only 661 m3/sec (all sources for the here mentioned discharge values are referenced in Table 2). The water contribution from downstream Niamey to the total discharge at Lokoja is thus 87%. Note that the basin part upstream Niamey maintains nearly all of the Niger's discharge in winter (December/ January). In early autumn (September/ October) when the rivers has its marked discharge peak, the water contribution from the downstream part is even much greater than 87%.

Table 2. Mean annual runoff of the Niger River from upstream to downstream. ______________________________________________________________________________________________________________

Station Longitude Latitude Mean Runoff (1) Period of complete Source (m3/sec) Record years ______________________________________________________________________________________________________________

Koulikoro 7° 19' W 12° 33' N 1425 1907-1988 82 Global Runoff Data Center Koblenz [1991] Mopti 4° 12' W 14° 30 N 1136 1922-1975 34 Global Runoff Data Center Koblenz [1991] Dire 3° 23' W 16° 16' N 1104 1924-1979 55 Global Runoff Data Center Koblenz [1991]

Niamey 2° 48' E 13° 29' N 898 1929-1991 49 Global Runoff Data Center Koblenz [1991] Gaya 3° 17' E 11° 31' N 1153 1952-1990 23 Global Runoff Data Center Koblenz [1991] Jebba 4° 49' E 9° 11' N 1044 1977 1 Global Runoff Data Center Koblenz [1991]

Lokoja 6° 44' E 7° 49' N 4886 1970-1980 11 Martins [1982], Martins & Probst [1991] ______________________________________________________________________________________________________________

(1) only complete years were used to calculate the value

DOC concentrations at Lokoja ranged between 2.1 mg/l and 6.6 mg/l (Fig. 26). The discharge-weighted annual mean is 3.7 mg/l. cDOC rised to high values at the very beginning of the ascending hydrograph, and decreased then rapidly. The maximum cDOC occurred already 2-3 month before the

42

maximum of Q, and the flushing out of DOC is therefore more pronounced compared to the other presented studies in this chapter where a flushing effect can be observed. However, it is not clear whether this effect is not at least also partly related to the water mixing from the different tributaries of the river, which may have different DOC loads and discharge patterns. The additional samples that were taken from upstream Lokoja at Jebba had relative low cDOC values (Fig. 27). This station can be considered to represent still mainly the Niger's waters from far upstream. In 1977, the only year where I found discharge values for this station (Table 2), the mean annual discharge was here only about 20% greater than at Niamey (although this value may only allow a rough estimate, not at least

0

5000

10000

15000

20000

25000

2

4

6

8

0

50

100

150

200

0

2

4

6

8

10

0 30 60 90 120 150 180 210 240 270 300 330 360

4

8

12

16

20

Q

DOC

TSS

POC

C/N

M J J A S O N D J F M ADays 80/81 (Begin: May 80)

Fig. 26 Evolution of Q, DOC, TSS, POC, and C/N in the Niger River at Lokoja (May 1980 to April 1981). Note that the y-axis to the left depict discharge (m3/sec), TSS (mg/l) concentrations, and C/N, while the y-axis to the right depict DOC (mg/l) and POC (mg/l) concentrations, respectively.

because of the damming of the Niger between these stations, resulting in an artificial water regulation). Note that cDOC at Lokoja are nearly identical to those at Jebba in winter, indicating that the dissolved organic matter that is discharged by the Niger during this season exclusively originates from upstream. Further downstream, between Jebba and Lokoja, most of the Niger's DOC enters the river. Here the mean annual discharge of the Niger increases about a factor of 4 to 5. The cDOC values in the Kaduna tributary which are especially elevated in summer confirm that also cDOC generally increase in this reach.

The TSS concentrations at Lokoja ranged between 11.5 mg/l and 135 mg/l, and those of POC between 1.0 mg/l and 4.6 mg/l. Both parameters are well correlated. The discharge-weighted annual means are 78 mg/l for cTSS and 2.7 mg/l for cPOC, respectively. As in the case of cDOC, cTSS and cPOC show a flushing effect in summer (Fig. 26). Also at the end of autumn and the beginning of winter, still somewhat elevated values were found. The cPOC patterns at Jebba and in the Kaduna

43

tributary are similar to the cDOC patterns (Fig. 27), leading to the conclusion that also for POC most of the Niger's load during the high-water period in late spring to early autumn originated from the

9

2

4

6

8 JebbaKadunaLokoja

DOC

M J J A S O N D J F M A M

Month 1980/81

0

3

6

9

12JebbaKadunaLokoja

POC

M J J A S O N D J F M A M

Month 1980/81

Fig. 27 DOC and POC concentrations (mg/l) during 1980/81 in the Niger River at Jebba, at Lokoja,and in the Kaduna River, one of the Niger's tributaries. For the locations of the sampling sites, seeFig. 25.

downstream basin part. In the Kaduna tributary, the differences between cPOC in summer and in winter are even much more pronounced than for cDOC. This can probably explain the evolution of the C/N ratios at Lokoja during the investigation period. They show highest values in late spring and early summer, and decrease then more or less continuously towards late winter. The high summer values may reflect a different signature of POC coming from the downstream basin part during that time, compared to the signature of the upstream POC that dominates in winter. Keep in mind that in the downstream basin part discharge patterns show a very strong seasonality. It is therefore possible that POC contains here a greater share of relative fresh vascular plant material that is washed out during peak discharges, leading generally to greater C/N values. The evolution of the C/N ratios is hence rather coupled to discharge than to the vegetation cycle, as this was observed, for example, in the St. Lawrence River. A coupling to the vegetation cycle is also less to be expected, since in tropical climates litterfall is not very important and does not underlie that distinct seasonality as in temperate climates.

From Figure 27, the mean annual concentrations of DOC and of POC at Jebba can be estimated to be about 2.5 mg/l and 1.8 mg/l, respectively (the calculation of discharge-weighted means are not possible, because the Q values corresponding to the sampling period are not known). Assuming that these concentrations represent at Lokoja the organic carbon coming from the upstream basin part, the additional organic carbon that entered the Niger between Jebba and Lokoja should account for about 85% to 88% of the total organic carbon flux measured at Lokoja, depending on whether one takes a factor of 4 or of 5 for the increase of discharge between both stations (these percentages are nearly identical for DOC and for POC). This underlines the important role of the downstream part of the Niger for the total organic carbon export from the basin.

Finally, it is also interesting to look at the organic carbon transport at Bamako in the uppermost part of the Niger Basin. The corresponding gauging station is Koulikoro. At Bamako, only Q, cDOC and cTSS have been monitored (Fig. 28). cDOC varied between 1.4 mg/l and 4.0 mg/l during the period from May 1992 to April 1993, and the discharge-weighted mean cDOC is 2.2 mg/l. For cTSS, the values ranged from 1.8 mg/l to 61 mg/l, with a discharge-weighted average of 24.3 mg/l. The mean discharge for this period was 734 m3/sec, which is nearly half the value of the 1907 to 1988 average of 1425 m3/sec for this station (Table 2). In this part of Africa, climate underlies distinct long-term variations (see also chapter III). It is interesting to note that cDOC show also at Bamako a typical flushing effect, with a maximum value that occurred about two month before the discharge maximum at the beginning of October. A second but smaller cDOC maximum occurred both in April 1992 and 1993, without any visible relationship to the river's hydrograph. Note also that the discharge-weighted mean cDOC is quite close to the value found at Jebba during the SCOPE study, confirming to some extent the generally lower cDOC values in the upper part of the Niger basin. The discharge-weighted

44

mean annual cTSS at Bamako is about one third of the value that was found at Lokoja during the SCOPE study.

0

1000

2000

3000

4000

0 30 60 90 120 150 180 210 240 270 300 330 360

0

20

40

60

80

100

TSS

Q

DOC

D J F M A M J J A S O NDays 91/92 (Begin: Dec. 91)

1

2

3

4

5

6

0 30 60 90 120 150 180 210 240 270 300 330 360

TSS

Q

DOC

D J F M A M J J A S O NDays 92/93 (Begin: Dec. 92)

Fig. 28 Evolution of Q, DOC, and TSS in the Niger River at Bamako (December 1991 to July 1993).Note that the y-axis to the left depict discharge (m3/sec) and TSS (mg/l) concentration, while the y-axis to the right depict DOC (mg/l) concentration, respectively.

1.4.4.2. The Orange River

The Orange River was sampled during three complete seasonal cycles from August 1981 to December 1984 in its upper basin part, which accounts for by far most of the river's runoff (Fig. 29). The sampling station was the outlet of Le Roux Dam, about 150 km upstream of the confluence of the Vaal River which drains the north-eastern part of the basin. The here presented data comprise the three years from 1982 to 1984. They were mostly published in Hart [1982], [1983], [1985], and [1987], except the carbon data from the samples collected in the second half of the investigation period. Prof. R.C. Hart kindly agreed also the use of these data in this study.

As in the case of the Waikato (see above), the hydrology of the Orange River is nowadays strongly regulated by impoundments. In the upper Orange, two dams were closed during the seventies: Le Roux Dam (1976), the sampling station, and further upstream, Verwoerd Dam (1970). The storage capacity of these two artificial lakes is approximately equivalent to the mean annual runoff originating from the catchments upstream these reservoirs (cited in Hart [1987]). Figure 30 shows the effect of the water regulation on the runoff patterns during the investigation period. It compares the water flow at Le Roux Dam with the water flow at Aliwal North, a gauging station which is situated upstream Verwoerd Dam (Fig. 29). This station reflects the natural flow patterns in the uppermost Orange Basin. Due to the water regulation, peak discharges in the uppermost basin part were completely dampened at Le Roux Dam. The total water flux during the investigation period was not much different at both stations, but note that also the interannual variations were dampened by the water regulation. In dry years following on wet ones, relative more water is released from the lakes than enters the lakes (1982/83), while the opposite is the case for wet years following on dry ones (1983/84).

Because of the above described effects of the impoundments on the river's natural discharge patterns, it is not surprising that neither DOC nor POC show a clear relationship with Q (Fig. 31). cDOC varied during the investigation period between 1.7 mg/l and 4.4 mg/l. The discharge-weighted annual means are quite constant with 2.3 mg/l, 2.6 mg/l, and 2.8 mg/l for the three successive years,

45

Fig. 29 Drainage intensity inthe Orange basin (afterKorzoun et al. [1977]). SS isthe sampling station at theoutlet of Le Roux Dam, GS isthe gauging station at AliwalNorth (upstream VerhoerdDam), and M is the rivermouth.

Fig. 30 Runoff (m3/sec) at the outlet of Le Roux dam (SS -instantaneous valuescorresponding to the samplingtimes in the SCOPE study) incomparison with runoff atAliwal North (GS - monthly means from Global RunoffData Center Koblenz [1991]) during the investigation period.Numerical values show theannual means.

1982 1983 1984

0

100

200

300

400

500

SS

GS

GS: 112 m /sec3

SS: 112 m /sec3 GS: 53 m /sec3

SS: 78 m /sec3 GS: 79 m /sec3

SS: 72 m /sec3

respectively. But the values increase throughout. During certain periods it is found that the evolution of cDOC may be better understood if looking at the upstream evolution of the river's hydrograph than if looking at the hydrograph at Le Roux Dam. Note, for example, that the highest cDOC values in December 1983 falls together with the rising limb of the discharge peak encountered during that time at Aliwal North (Fig. 30), indicating thus a flushing effect than may be still seen at Le Roux Dam. cPOC varied between 0.5 mg/l and 2.1 mg/l. Also for cPOC, the discharge-weighted annual means increase constantly for the three investigated years: 0.9 mg/l (yr1), 1.1 mg/l (yr2), and 1.2 mg/l (yr3). The corresponding discharge-weighted mean annual cTSS are 61 mg/l (yr1), 37.5 mg/l (yr2), and 37.8 mg/l (yr3), respectively. They do not follow this increase. Consequently, POC% was nearly permanently increasing during the investigation period, except in the second half of 1984. Peaks in cPOC often occurred in late southern summer and in early autumn, normally also accompanied by elevated cDOC values. Probably during this season the autochthonous share in the river's organic matter increased. The fact that such an increase was absent in 1982 is in agreement with the greater discharge in this year and in the preceding year. In 1981, the mean average discharge was 172 m3/sec at Aliwal North, which is more than 50% greater than in 1982.

From 1981 to 1983, the average discharge at Aliwal North decreased to about one third of its initial value (172 m3/sec to 53 m3/sec), and started then to rise again in 1984. This can explain why cTSS at Le Roux Dam was nearly constantly decreasing in 1982 and 1983. In 1984, again higher cTSS values

46

0

50

100

150

200

250

300

0

20

40

60

80

100

0

4

8

12

16

20

C/N

Q

DOC

TSS

POC

Days 82

0

1

2

3

4

5

0

1

2

3

0 30 60 90 120 150 180 210 240 270 300 330 360

C/N

Q

DOC

TSS

POC

J F M A M J J .A S O N D Days 83

0

50

100

150

200

250

300

0

1

2

3

4

5

0

20

40

60

80

100

0

1

2

3

0 30 60 90 120 150 180 210 240 270 300 330 360

0

4

8

12

16

20

C/N

Q

DOC

TSS

POC

J F M A M J J .A S O N D Days 84

Fig. 31 Evolution of Q, DOC, TSS, POC,and C/N in the Orange River at Le RouxDam from 1982 to 1984. Note that the y-axisto the left depict discharge (m3/sec), TSS(mg/l) concentration, and C/N, while the y-axis to the right depict DOC (mg/l) andPOC (mg/l) concentration, respectively.

were found, but this is not reflected by the discharge-weighted annual mean calculated for 1984 (which is the same than for 1983), since the rise of cTSS occurred during low discharge. The evolution of cTSS in the Orange River may be a good example for a dephasing of the concentration patterns from the discharge patterns related to the artificial water regulation. There is an agreement on the level of the interannual variations, but not on the level of the seasonal variations.

1.4.4.3. The Indus River

Finally I present here the case study for the Indus River that was undertaken within the SCOPE program. The Indus was sampled at Kotri Barrage from 1981 to 1984 (Fig. 32). During 1981, sampling density was lowest, and in the following only the three years from 1982 to 1984 are shown. Note, however, that also in 1982 only about 59% of the river's total discharge in that year could be monitored in the SCOPE study, whereas this was about 81% and 95% in the following years (Arain

47

[1987]). Data and basic information on the Indus River case study can be found in Arain and Khuhawar [1982], and in Arain [1985], [1987], and [1988].

Fig. 32 Drainage intensity in the Indus Basin (after Korzoun et al. [1977]). SS, sampling station at Kotri Barrage; GS1, gauging station at Attock; GS2, gauging station at Akhnoor; GS3, gauging station at Panjnad; M, river mouth.

Fig. 33 Mean monthly discharge (m3/sec) at four gauging stations in the Indus Basin (1976-1979 average). Data are from Global Runoff Data Center Koblenz [1991]. The numbers in brackets give the annual means, and the arrows show the flow direction from upstream to downstream.

0

2000

4000

6000

8000

10000

12000


0

2000

4000

6000

8000

10000

12000

14000

16000

18000


Panjnad (1162 m /sec)3

Kotri (2396 m /sec) 3

Attock (3535 m /sec) 3

Akhnoor (848 m /sec) 3

The hydrology of the Indus is characterized by an extreme seasonality. Looking at its mean hydrograph for the 1976 to 1979 period at Kotri Barrage (Fig. 33), it can be seen that there is a factor of about 50 between the minimum (February) and the maximum discharge (August). On average, more than 40% of the total annual water discharge runs off only during August, and about 77% in the period from July to September. In the headwater regions, where nearly all of the river's water comes from, the seasonality of the discharge patterns in the Indus and its tributaries is less pronounced. The maximum discharge peak occurs here in July and precedes thus the downstream discharge maximum with one month. There is also another important feature emerging from Figure 33 that should be outlined: the water amount that arrives at Korti Barrage is much lower than the amount of runoff that is generated in

48

0

2000

4000

6000

8000

0

500

1000

1500

2000

2500

0

5

10

15

20

25

30

C/N

TSS

POC

Q

DOC

Days 82

0

5

10

15

20

0

5

10

15

20

25

0 30 60 90 120 150 180 210 240 270 300 330 360

C/N

TSS

POC

Q

DOC


0

2000

4000

6000

8000

0

5

10

15

20

25

0

500

1000

1500

2000

2500

0

5

10

15

20

25

0 30 60 90 120 150 180 210 240 270 300 330 360

0

5

10

15

20

25

30

C/N

TSS

POC

Q

DOC


Fig. 34 Evolution of Q, DOC, TSS, POC,and C/N in the Indus River at Kotri Barragefrom 1982 to 1984. Note that the y-axis tothe left depict discharge (m3/sec), TSS (mg/l)concentration, and C/N, while the y-axis tothe right depict DOC (mg/l) and POC (mg/l)concentration, respectively.

the headwater regions. In the 1976-79 period, the Indus and its tributaries lost nearly half of its water on the way from upstream to downstream.

The average water discharge of the river at the sampling station was 549 m3/sec, 1836 m3/sec, and 1347 m3/sec (Arain [1987]) for the three successive years from 1982 to 1984, respectively. This was throughout lower than the 1976-79 average (Fig. 33), and much lower than the long-term average of the river, which is commonly cited with about 7600 m3/sec (see chapter III). Figure 34 shows that the great seasonal variability of discharge had also strong effects on the organic matter concentrations in the river waters. cDOC varied between 2.4 mg/l and 22.0 mg/l, with discharge-weighted annual means of 12.4 mg/l (yr1), 16.1 mg/l (yr2), and 13.2 mg/l (yr3), respectively. The concentrations of dissolved organic carbon show a very distinct flushing effect, leading to a threefold to fourfold increase of cDOC on the rising limb of the summer discharge peak. This can be seen best in 1984, but also the more sparse data sampled in summer during the other years indicate that the evolution was

49

similar here. Among the river studies within the SCOPE program, the marked flushing of DOC makes the Indus River to one of the rivers with the highest discharge-weighted mean cDOC.

The high average DOC concentrations in the Indus may also have another reason. Keep in mind that the river looses a great amount of water on its way downstream through evaporation (another part may be lost through irrigation in agriculture). This should exert a concentration effect on the dissolved solids in the river water. Already with the Niger study I presented a river that looses a considerable amount of water through evaporation, but this takes place in the upper Niger Basin, while most of the DOC in the Niger originates from the lower basin part (see above). Contrary to that, one can expect that in the Indus most of the DOC comes from upstream because the downstream basin part is mainly desert (Fig. 32). A concentration of the river's water through evaporation could also explain why the Indus has generally very high concentrations of total dissolved solids (TDS) in its downstream part compared to, for example, the Ganges and the Brahmaputra rivers (as noted by Subramanian and Ittekkot [1991]). In the upstream waters of the Indus, TDS concentrations are only as half as high and compare well with those of the Ganges and the Brahmaputra rivers (Subramanian and Ittekkot [1991]).

The concentrations of POC were strongly coupled to those of TSS. From 1982 to 1984 the former ranged between 0.4 mg/l and 23.3, and the latter ranged between 14.8 mg/l and 2434 mg/l. The discharge-weighted annual means are 19.9 mg/l (yr1), 11.4 mg/l (yr2), and 5.9 mg/l (yr3) for cPOC, and 1168 mg/l (yr1), 712 mg/l (yr2), and 1809 mg/l (yr3) for cTSS, respectively. Similar to the cDOC values, cPOC and cTSS show in 1984 a flushing effect, but the values remained elevated throughout the high-water period and did not quickly drop down after the initial water rise, as this can often be observed for many other rivers (see above). Also in 1983, greatest values occurred during the high-water period, but it cannot be said when the rise of cTSS began since no sample was taken during the main rise of the hydrograph. Note that the C/N ratios were quite high on average. They were low in winter and then clearly increased towards the high-water period. This is in agreement with the findings for the Niger River. The strong seasonality of precipitation and of runoff which is generally found in the semi-arid climates probably takes away great amounts of relative fresh vascular plant material. In the case of the Indus River, it is also possible that the very low water flow in winter may also have allowed some plankton growth in the waters.

1.5. Conclusions

Eleven case studies on the seasonal and interannual variability of dissolved and particulate organic carbon in rivers were presented on the previous pages. Table 3 gives a summary of these studies. The watersheds of the rivers represent a broad range of the hydroclimatic, biological and geomorphological characteristics found on Earth, while the investigation periods reflect the great interannual and long-term variations that can be found for river discharges. Some rivers were sampled during average hydrological conditions (e.g., the St. Lawrence River), others during very dry years (e.g., the Garonne or Indus rivers), and again others during exceptional flooding (the Paraná River). The broad interannual variability makes it sometimes difficult to attribute mean carbon fluxes to the different rivers, which represent typical long-term averages. This has to be kept in mind when using later in this study the here presented organic carbon data for modelling purposes at the global scale (chapter V).

Already when looking at a small brook catchment in the French Vosges mountains we have seen that the organic carbon mobilization is governed by different processes, leading to different signatures in the surface waters draining the catchment. Consequently, when changing to the scale of large rivers, which comprise often a high degree of heterogeneity in their basins, it is not possible to definitively relate the patterns observed in the downstream river waters to certain processes which may be at their origin. Nevertheless, despite a large number of particularities that are characterizing the individual case studies, many common features emerged in these studies. Generally, it has to be pointed out that

50

the fluxes of organic matter are closely linked to the fluxes of water and of sediments in rivers. For this reason the global controls of these two parameters will be investigated in detail in the following two chapters.

Table 3. Summary of the river case studies presented in the text.

______________________________________________________________________________________________________________ Discharge-Weighted (dw) Average

River/ Study __________________________________ Major Characteristics Brook Period Q DOC POC TSS POC% C/N

(m3/sec) (mg/l) (mg/l) (mg/l) ______________________________________________________________________________________________________________ Strengbach 1989-92 24 1.6 - - - - About weekly sampling, with high resolution sampling

(2 yrs) x 10-6 during storm events at different sites in the catchment.

Garonne 1989-92 400 2.8 3.1 87 3.6 - Exceptional dry conditions. Comparison of upstream and (3 yrs) downstream stations. Plankton development in summer.

Mackenzie 1981-83 7470 4.9 5.0 171 3.0 20.4 Strong seasonality of discharge patterns. Insufficient (2 yrs) sampling during discharge peaks.

St. Lawrence 1981-85 13100 3.8 0.8 13 6.0 10.3 Determination of δ13C in POC (dw-mean: -25.4). Mix- (4 yrs) ing of organic matter from upstream and downstream.

Waikato 1981-84 260 5.5 1.4 13 10.1 9.3 Discharge patterns strongly influenced by artificial river (3 yrs) damming. Plankton development in summer.

Orinoco 1982-85 36000 4.4 1.5 80 1.9 7.8 Flushing of DOC/ POC during the water rise. Remobili- (3 yrs) zation of particles when the river drains the floodplain.

Paraná 1981-84 20530 8.7 2.1 58 3.6 10.4 Exceptional flooding triggered by ENSO. Storage of (3 yrs) particulate matter on the floodplain during the flood.

Ubangi 1992-95 6900 5.0 - - - - Exceptional dry conditions. Flushing of DOC during the (3 yrs) water rise.

Niger 1980-81 5180 3.7 2.6 79 3.3 13.1 Typical flushing effect for DOC, POC, TSS. Important (1 yr) contribution of organic matter from downstream.

Orange 1982-84 90 2.5 1.1 49 2.2 11.0 Discharge patterns strongly influenced by artificial river (3 yrs) damming. Only the upper basin was monitored.

Indus 1982-84 1240 14.4 8.8 1920 0.5 22.0 Exceptional dry conditions. Very strong flushing effect. (3 yrs) Concentration of the waters during transport.

______________________________________________________________________________________________________________

With respect to the seasonal patterns, it can be noticed that concentrations of dissolved organic carbon were normally well correlated with discharge. Generally, DOC concentrations increased with increasing discharge, although the relationships can be better described as a function of specific hydrologic phases rather than as a smooth function of discharge over the complete hydrograph. In many cases, cDOC values showed a so-called flushing effect, i.e. a rapid increase of the concentrations on the rising limb of the hydrograph, followed by a subsequent decrease of the concentrations already before the decrease of discharge began. The decrease of cDOC could begin at earlier or later stages of the water rise, leading either to a pattern of a marked peaking of cDOC considerably before the major high-water stages (e.g., the Niger River), or to a pattern of cDOC that remained elevated over most of the high-water period (e.g., the Orinoco, Ubangi, and Indus rivers). When this is compared to the observations made for the Strengbach catchment (Vosges), it is suggested that the former pattern is typical for a rapid purging of DOC that mainly accumulated previously, while the latter pattern is typical for a more permanent leaching of DOC, probably also because greater amounts of carbon are available for leaching processes.

The interannual variability of the DOC fluxes is more difficult to generalize. Normally only very few years have been investigated, but also here exists rather a tendency of greater discharge-weighted mean annual DOC concentrations in humid years compared to dry years (e.g., the Strengbach Brook, the Ubangi and Paraná rivers). A very striking example is the Paraná River when it was affected by an extraordinary flooding triggered by ENSO. The average discharge of the river was

51

here about 75% greater than during the pre-ENSO period, while the corresponding FDOC increased by a factor of 2.6 during ENSO. The seasonal patterns, however, were characterized by an extreme variability of cDOC during ENSO. During several times, a rapid increase of cDOC to very high values was encountered, indicating a sudden delivery of great amounts of DOC to the river waters. Also in the case of the Garonne River (or, but to a lesser extent, of the Ubangi), such a sudden release of DOC was found. This shows that the mobilization of DOC in river basins is not always a continuous process, but can also occur through sporadic events. For an understanding of such events, probably more should be known about the interaction of the river water level with the water levels in the surrounding soils of the alluvial plains, and/or the interaction with the floodplain. In the case of the Paraná, for example, it is possible that during the vast inundation boggy ponds and bogs might have been formed on the floodplain that were temporarily isolated from the main river waters. When these waters became again in contact with the main channel, they may have instantaneously delivered great amounts of DOC to the river.

The concentrations of particulate organic carbon were closely coupled to those of total suspended solids. Also for cPOC and cTSS the dominant seasonal pattern is the flushing effect, normally even more pronounced than in the case of DOC. This underlines the important role of peak discharges for the mobilization and transport of POC and TSS in river basins. It should be noted here, however, that the seasonal patterns can also be influenced by an alternation of mobilization and sedimentation processes taking place in the river. This concerns especially rivers with a large floodplain, as, for example, the Orinoco or the Paraná rivers. A typical feature for these two rivers was an additional increase of cPOC and cTSS on the falling limb of the hydrograph when the river drained the floodplain, indicating a remobilization of particulate matter that was deposited on the floodplain during the high-water stage. Another example for an interaction of mobilization and sedimentation processes was found for the Garonne River. The short sampling intervals allowed here to detect a significant decrease of the POC% values during sharp discharge peaks, suggesting a sudden resuspension of mineral matter from the river floor during that time. POC% normally increased again to about the initial values at the end of the discharge peak.

Sedimentation processes can also influence the interannual variability of FPOC and FTSS. The exceptional flood in the Paraná is again a striking example for this. Despite the considerable increase of the water discharge during ENSO compared to the pre-ENSO period remained the corresponding TSS flux at about the same level, and the corresponding POC flux even decreased to about 60% of its previous value. This can be attributed to a trapping of sediments on the floodplain due to the large increase of the size of the floodplain. Note, however, that the highest TSS concentrations in the Paraná were observed at the end of the exceptional flooding, and not at its beginning. Probably a great portion of the sediments that were deposited on the floodplain returned to the river during its retreat from the floodplain. Long-term observations are needed in order to estimate the effect of sedimentation for the fluxes of particulate matter in rivers (see also chapter IV).

The C/N ratios in the particulate matter showed different seasonal patterns depending on climate and the hydrological regime of the river. In the temperate climate, there exists a tendency to relative great values in late autumn/ early winter and to relative low values in summer (e.g. the St. Lawrence River). This may reflect the seasonality of the vegetation cycle. Late autumn is the period where litterfall occurs, which delivers great amounts of relative fresh organic detritus to the top of the soils. In the upstream part of the Garonne River such an hypothesis is indirectly confirmed by the POC% values. Here a similar evolution of relative great values in late autumn/ early winter and to relative low values in summer could be observed, which is probably also related to litterfall. In the tropical dry climate, a seasonal evolution of the C/N ratios was absent. Litterfall is less important in this climate and does not show a distinct seasonal pattern. In the tropical dry climate, the C/N ratios were positively correlated with the evolution of drainage, that is relative high values occurred during the high-water period and relative low values occurred during the dry period (e.g., the Indus and Niger rivers). Note that in this climate often a rapid but intensive flushing of the water discharge is observed, provoked by strong precipitation such as, for example, the monsoon rains in the Indus Basin. Probably these rains carry away relative great amounts of vascular plant material. In the Mackenzie River,

52

which represents the boreal climate type, the C/N ratios showed also the tendency to increase with increasing discharge. Also the hydrograph of the Mackenzie shows a very strong seasonality, but in this case it is related to snow melting.

The trend to relative low C/N ratios that occurred during the low water stages could also be related to an enhancement of the autochthonous contribution to the total riverine POC. In the discussed case studies, however, it was rather the evolution of cDOC that gave indications for the existence of an enhancement of primary production in the river waters. The evolution of the C/N ratios was not always conclusive in this respect. In some rivers (e.g. the Garonne and the Waikato rivers) it could be observed that cDOC increased considerably during summer/ early autumn when the water discharge was lowest. In the Garonne River the cDOC increase was only observed in the downstream basin part, and also accompanied by an increase of POC% in the particulate matter. It is interesting to note that this effect of enhanced DOC concentrations during summer was generally more pronounced in dry years than in wet years. This is in agreement with the assumption that primary production in river waters should mainly develop when turbulence and turbidity of the waters are minimal, while insolation is maximal. If it is true that these summer DOC peaks may be a mean to quantify somehow autochthonous carbon production in rivers, one has also to mention that the contribution of this carbon to the total organic carbon flux was quite low in the investigated rivers.

Further investigations on the transport of dissolved and particulate organic carbon in large river systems should be designed in a way that the hydrological particularities of these river systems are taken into account. In many of the here discussed case studies, more information on the organic carbon transport in the tributaries would have been needed to better understand the signals that were recorded at the river mouths. A good example for this is the St. Lawrence River. It is likely that in this case the organic carbon mobilization is driven by quite different hydrological patterns, leading to an interference of the signals encountered at the river mouth. Another problem is the existence of great dams in the river basins. Note that among the studies where it was probably the most difficult to detect any relationship between drainage and the seasonal patterns of the here discussed parameters are the Orange and the Waikato rivers, which were strongly influenced by man-created impoundments. At least in the case of the upper Orange River, it could be shown that the natural discharge patterns of the river were completely dampened by an artificial water regulation.

Finally, also more information on the organic carbon fluxes in small catchments is needed. In the literature, only relative few studies can be found which focused on the concentrations and transport of DOC in small watersheds. Even much less is known about the transport of POC in small watersheds.

53

CHAPTER II

METHODICAL ASPECTS: TOWARDS A QUANTIFICATION OF RIVER BASIN

CHARACTERISTICS AND THEIR INFLUENCE ON THE FLUVIAL MATTER TRANSPORT TO THE

OCEANS

2.1. Introduction

The waters transported by rivers witness the variety of physical, chemical, and biological processes taking place in river drainage basins. These processes may be the mechanical erosion of the oupcropping rocks in the headwater regions, the exchange of nutrients and carbon on the floodplain, or the storage of sediments and pollutants in internal reservoirs such as lakes: the effects of all of them are reflected in the composition of the waters and sediments before they are discharged to the oceans. Many of theses processes interfere or even compensate each other, such as, for example, mechanical erosion and sedimentation do. Analysing the water and sediment composition allows an estimation of the total amount of matter that is exported from a particular river basin when the samples are taken close to the river mouth and in regular time steps (at least for one year, and following the major changes of the river hydrograph). But it allows not to identify the importance of each of the processes that may have contributed to a measured flux. Therefore it is difficult to extrapolate the results to other river basins where no measurements exist.

It is the sum of measurements for a large number of rivers that can lead to the identification of the controlling factors for river fluxes when the data are interpreted on the background of the environmental characteristics of the river basins. This is the principal approach followed in this study. Such an approach is depending on the availability of environmental parameters, and on the possibility to determine them in the same way for all rivers. In this work, I make use of a large number of environmental data sets that have been developed in recent years in different disciplines for global scale research. Other data sets were developed in the framework of this study, or during previous studies carried out at the Centre de Géochimie de la Surface (CGS). The advantage of these data sets is that they include values for the overall continental area in a certain spatial grid point resolution. Once an empirical relationship has been established on the basis of the data, it can be easily extrapolated to larger scales in order to establish regional and global budgets.

The purpose of this chapter is to present the data sets I considered, and the way they were used in this study. This information is the basis for many results and calculations that are presented in the following chapters. The general proceeding is that the river fluxes were taken from the literature, and the climatic, biological, and geomorphological characteristics of the river basin were extracted from the data sets using the digitized river basin contours. Normally, one mean surface weighted value was calculated for each basin and each parameter. Then, statistical analyses were applied to identify the

54

main controlling parameters, or parameter combinations. Only the environmental data sets are described here. The literature river fluxes are presented in the corresponding chapters.

One problem with such a proceeding is the huge difference in size between the river basins for which measurements exist. For this reason, Meybeck [1993b] proposed only to use the measurements for large tributaries instead of the measurement for the main rivers in the case of big river systems, such as the Amazon River. In many field studies, however, only the main stream has been sampled, and it will still take a long time until sufficient tributary data are available for the major river systems on Earth. Moreover, one has to keep in mind that basin size itself is also a parameter that can influence river fluxes. A decomposition of river basins into smaller subunits is certainly a valuable tool to determine the major controls for many river fluxes, but this proceeding may not detect processes that alter the river water composition during transport.

Not at least because of the problem of the basin size differences, I present in the following also a simple climatic classification based upon the Holdridge Life Zone Scheme (Holdridge [1947]) that was throughout applied in this study. This classification allows to group the river basins according to their average climatic situation, but also to determine their climatic variability by calculating the percentage each climate type occupies in each basin. The climatic variability of the river basins supplies an important information for the interpretation of the results in a study like this. For certain rivers, the average environmental basin characteristics may be helpful to identify the controlling parameters for observed river fluxes, but the average values may be misleading if the river hydrology and the related fluxes are strongly influenced by a particular region in the drainage basin that has quite different characteristics in comparison to the rest of the basin area.

The application of a climatic classification has also several other advantages. First, it allows to test the representativity of the investigated river basins. Before extrapolating statistical results obtained from a set of river data world-wide, one has to ensure that the selected rivers represent the major ecosystems on Earth. Another advantage is that such a classification makes it possible to calculate not only regional but also 'climatic' budgets, which may allow a first-order-estimate of river fluxes for different scenarios of climate change. Finally, regression analyses can be done within different climatic subgroups of the river basins in order to test whether major differences occur. These differences may reveal the importance of additional controlling factors that are not expressed by the parameters tested in this study, but that belong also to climate in the broad sense of the term.

2.2. Data and Methods

2.2.1. River Basins

For this study, the data set of Pinet and Souriau [1988], which consists of a set of digitized contours of major world river basins, was available. I extended it by the further digitalisation of several additional river basins to a set of 60 drainage basin contours. The rivers are listed in Table 4, together with their basin areas. The latter were calculated in a 0.5° x 0.5° longitude/ latitude grid point resolution by the use of the contour lines. The world-wide distribution of the basins is shown in Figure 35a. Moreover, the contours of the endoreic parts of the continents and the continental divides with respect to the different ocean basins were added to the data set (Fig. 35b). This allows also budget calculations for the fluvial matter input to each ocean. The Atlantic was further divided into the North Atlantic and into the South Atlantic. Note that the input from the Amazon Basin is attributed to the North Atlantic.

55

13

32

529

2

14

2434

4249

5651

485047

5845

6023

31 18

57

35

46

7

8

39

33 27

4122 10

4420

30

29

5553

5915

1119

40 36

435

281

1737

3 46

25

1626

5438

1221

__ __

F

ig. 3

5a W

orld

-wid

e di

stri

butio

n of

the

rive

r bas

ins i

nves

tigat

ed in

this

stud

y. T

he ri

ver n

ames

cor

resp

ondi

ng to

the

num

bers

are

list

ed in

Tab

le 4

.

56

Table 4. Digitized river basins and their basin areas. ______________________________________________________________________________________________________________

River Basin Basin Area River Basin Basin Area River Basin Basin Area # (106 km2) # (106 km2) # (106 km2)

______________________________________________________________________________________________________________ Amazon 1 5.903 Yukon 21 0.843 Liao He 41 0.188

Zaire 2 3.704 Huanghe 22 0.823 Rufiji 42 0.180 Mississippi 3 3.246 Danube 23 0.773 Rio Negro (Argentine) 43 0.175

Ob 4 3.109 Orange 24 0.716 Hungho 44 0.159 Paraná 5 2.868 Colorado 25 0.708 Rhine 45 0.156 Yenisei 6 2.567 Columbia 26 0.664 Brazos 46 0.127

Lena 7 2.465 Kolyma 27 0.659 Loire 47 0.107 Amur 8 1.926 Sao Francisco 28 0.621 Rhône 48 0.097 Nile 9 1.874 Si Kiang 29 0.464 Tana 49 0.089

Changjiang 10 1.822 Irrawaddy 30 0.419 Garonne (1) 50 0.079 Ganges/Brahmaputra 11 1.656 Don 31 0.413 Po 51 0.067

Mackenzie 12 1.615 Senegal 32 0.369 Gambia 52 0.063 Niger 13 1.540 Indagirka 33 0.358 Fly 53 0.058

Zambesi 14 1.413 Limpopo 34 0.344 Susitna 54 0.057 Murray 15 1.131 North Dvina 35 0.329 Purari 55 0.040

St. Lawrence 16 1.114 Godavari 36 0.311 Tiber 56 0.016 Orinoco 17 1.026 Magdalena 37 0.285 Rioni 57 0.016

Tigris/Euphrates 18 0.927 Fraser 38 0.248 Severn 58 0.013 Indus 19 0.912 Yana 39 0.243 Waikato 59 0.012

Mekong 20 0.864 Mahandi 40 0.190 Ems 60 0.009 ______________________________________________________________________________________________________________

(1) incl. Dordogne

Together, the 60 drainage basins cover about 50% of the total exoreic continental area. Taking

only the 10 greatest of them makes already more than 25%. Keep in mind that the greatest basin (Amazon) has more than 600 times the area of the smallest basin (Ems). The basin areas calculated with the contour lines compare well with the values found in the literature. Discrepancies arise only for a few river basins that border to very dry regions, where it is difficult to define the basin limits (e.g. the Nile or the Niger basins).

SA

Me

In

In

Me

NA

SA

Pa

An

Pa

Pa

Ar Ar

NA

NA

Ar Ar

In

Pa

Pa

In

NA

Fig. 35b Continental divides for the different oceans. Abbreviations: Ar, Arctic Ocean; An, Parts of the oceans below 60° South; In, Indian Ocean; Me, Mediterranean Sea; NA, North Atlantic; Pa, Pacific Ocean; SA, South Atlantic. The filled parts of the continents are endoreic.

57

2.2.2. Environmental Data Sets

The above presented basin contours were used to extract the environmental basin characteristics from the data sets which are described below. For each basin and each parameter, one area-weighted mean value was calculated. A table with many of the determined parameter averages is given in chapter V (Tab. 21).

2.2.2.1. Hydroclimatic Parameters

For temperature, I used the mean annual and mean monthly temperatures (MT and AT, respectively) of Legates and Willmott [1992]. These data are available on CD-ROM as spatially interpolated gridded data sets with a resolution of 0.5° x 0.5° longitude/ latitude. Data for the mean annual precipitation total (APPT) and the mean annual runoff (Q) were taken from the Atlas of World Water Balance published by Korzoun et al. [1977]. The corresponding maps were digitized and gridded in the framework of this and another study (Munhoven and Probst [1995]) carried out at the CGS. More details about these data sets and the digitalisation procedure can be found in chapter III. Figure 36 depicts the global distribution of drainage intensity on the continents.

2.2.2.2. Biological Parameters

Information on vegetation was derived from the gridded vegetation map of Olson (Olson et al. [1983], [1985]), which exists in a grid point resolution of 0.5° x 0.5° longitude/ latitude. I used the version that is distributed via ftp by the Carbon Dioxide Information Analysis Center (CDIAC), Oak Ridge, Tennessee. The vegetation units were transferred to values for biomass (VegC) and net primary production (NPP) according to the keys given in the original publications. The data sets result in a global biomass of 575 GtC (gigatons of carbon) and a global NPP of 53 GtC/yr, which is close to other estimates (e.g., Bolin et al. [1979]: 592 GtC for biomass and 63 GtC for NPP; Esser [1991]: 650 GtC for biomass and 45 GtC for NPP). As further vegetation parameter, I derived also the mean forest ratio (ForR) from the Olson dataset following an assignment proposed by Claussen et al. [1994]. This parameter indicates the proportion of forest in the vegetation units.

The mean organic carbon content in the soils (SoilC) was extracted from a global data set developed in a 10' x 10' longitude/ latitude resolution at the Soil Conservation Service of the United States Department of Agriculture (USDA-SCS). They kindly made it available to me for this study. I averaged it to a resolution of 0.5° x 0.5° longitude/ latitude. In this data set, the average amount of soil carbon was empirically determined for each soil type to the suborder level in soil taxonomy on the basis of a database maintained at the USDA-SCS, and then extrapolated to the global scale using the FAO soil maps of the world (FAO [1971-81]). It is assumed that all organic carbon in soils is concentrated in the first meter of the soil profile. The USDA-SCS database is, for example, described in Eswaran et al. [1993] and in Kern [1994]. From the resulting SoilC map (Fig. 37), I calculated the global amount of carbon stored in the soils to be 1470 GtC, which is in agreement with other estimates (e.g., Post [1993]: 1300 Gt C without wetlands; Eswaran et al. [1993]: 1576 GtC).

2.2.2.3. Geomorphological and Lithological Parameters

For the characterization of morphology, I took the mean modal elevation (Elev) of the Fleet Numerical Oceanography Center (FNOC) that is available on CD-ROM (FNOC [1992]), and the mean grid point slope (Slope) of Moore and Mark [1986] which was supplied from the U.S. Geological Survey, Menlo Park, California. The first data set exists in a spatial resolution of 10' x 10' longitude/ latitude, and the second one exists in a resolution of 5' x 5' longitude/ latitude. I averaged both data sets to a resolution of 0.5° x 0.5° longitude/ latitude. The global Elev and Slope maps are shown in the Figures 38 and 39, respectively.

58

As additional soil parameters, I included the data sets of Webb et al. [1991] in this study. These data sets contain information on the soil texture and the soil profile thickness (SoilH). They were originally created to improve land surface characteristics in general circulation models (Webb et al. [1993]), and are distributed in a spatial resolution of 1° x 1° longitude/ latitude on CD-ROM (Webb et al. [1992]). I linearly interpolated them to a 0.5° x 0.5° longitude/ latitude resolution. However, spatial resolution may be misleading if one takes it as a measure for the precision of the data. One has to know that the data were created by extrapolating the selected soil profiles representing each soil type in the FAO soil maps over the continents, following the FAO soil map digitized by Zobler [1986]. Local and regional differences (for example of SoilH) that are related to climatic and morphological variability are thus not taken into account here. I used this data set also to derive an index for soil erodibility (SoilT) based upon average soil texture. The average soil composition of each grid element was plotted in a soil texture triangle (% clay, % silt, % sand) and a three-point scale ranging from 1 (slightly erodible) to 3 (highly erodible) was assigned depending on the position of the grid element in the triangle (see Fig. 40). The assignment was made according to CORINE [1992].

Information about lithology was taken from two global data sets that were created at the CGS during previous studies. One lithological map classifying the major rock types on the continents according to their resistance to mechanical erosion was created by Dubus [1989]. He assigned a numerical index (LithMI) to each rock type. LithMI ranges from 1 to 40, with 1 = plutonic and metamorphic rocks, 2 = volcanic rocks, 4 = consolidated sedimentary rocks, 10 = different rock types in folded zones, 32 = non-consolidated sedimentary rocks, and 40 = recent alluvials. The map is also described in Probst [1992]. A second global lithological map was developed by Amiotte-Suchet [1995] in order to predict the consumption of atmospheric CO2 by rock weathering (Amiotte-Suchet and Probst [1995]). To the rock types in this map, a numerical index characterizing the resistance to chemical erosion (LithCI) was assigned (see chapter VI). Both maps were created in a 1° x 1° longitude/ latitude resolution, and I linearly interpolated them to a 0.5° x 0.5° resolution.

2.2.2.4. Other Parameters

Mean population density (PopD) and the percentage of cultivated area (CultA) in each basin were used as parameters to illustrate human's impact in the river basins. The first parameter was taken from a data set developed at the Institut für Energieforschung in Graz, Austria (Ahamer et al. [1992]). The latter parameter was calculated from the Olson vegetation map (see above) by grouping all agricultural vegetation types together and calculating the percentage of the area that they cover with respect to the total basin area.

Finally, I have also to mention here that I followed in this study the Olson vegetation map in order to define which of the Earth's 0.5° x 0.5° grid points belong to the continents, and which belong to the oceans. Also the extension of the polar ice sheets was taken from this data set.

2.2.3. Empirical Modelling

In the multiple regression analysis, the best regression model was selected by testing all possible combinations of a set of independent variables against a dependent variable. The best model is retained following the adjusted correlation coefficient and the statistical Cp coefficient of Mallows (Mallows [1973]). Cp is a measure of the total squared error for a subset model containing p independent variables. The model with Cp closest to p + 1 can be considered as the best model to predict the dependent variable. This procedure suffers from difficulties in the parameter selection,

59

Fig. 36 Drainage intensity on the continents (after Korzoun et al. [1977]). Hatched lines show the endoreic parts of the continents.

Fig. 37 Global distribution of organic carbon in the soils (data from USDA-SCS, see text). Ice-covered regions are omitted.

61

Fig. 38 Mean modal elevation of the continents (FNOC [1992]).

Fig. 39 Mean surface slope on the continents (data from Moore and Mark [1986]). The values on the Antarctic continent are erroneous.

63

because some parameters can show strong correlation among each other. For example, there is generally a strong correlation between precipitation and runoff at the global scale (e.g., Tardy et al. [1989]), which can make the introduction of both variables into one model problematic. The simultaneous use of all parameters in the regression analyses was often not possible. Therefore I grouped the parameters in all possible combinations in a way that minimized the effects of multicollinearity. Then all parameter combinations were tested individually to identify the best models. Multicollinearity was detected by following the suggestions of SAS [1986] that in a multiple regression model the variance inflation factor (VIF) of each independent variable should be lower than the square of the correlation coefficient of the regression. VIF is defined as 1/(1-Ri

2), where Ri2 is the

coefficient of determination of the regression of the i-th independent variable on all other independent variables.

100% Sand (S) 100% Silt (Z)

100% Clay (C)

C

SCZC

SCLCL ZCL

LSL ZL

Z LSS

High erodibility

Moderate erodibility

Low erodibility

Fig. 40 Textural classes for soil erodibility assessment (figure from CORINE [1992]).

In this work, I tried to find the best possible regression models in conjunction with a maximal number of rives rather than to obtain the greatest correlation coefficients. A more rigorous selection of the data by excluding rivers may increase the statistical significance of the results, while, on the other hand, this holds the risk to bias the results by omitting certain ecosystems in the regressions. Nevertheless, in some cases, the omission of one or two rivers from the regressions could considerably improve the regression coefficients without changing the significance of the retained parameters in the models. These rivers may be an exception of the general trend, or the data may be misleading in understanding the factors that control the organic carbon fluxes. If I found indications that this is true, then I excluded them from the calculations and I used the equations with the more significant coefficients for the modelling. These exceptions are always discussed in the text.

Note also that in all equations for regression models given in this study, the intercept is only shown when it was at lest significant with P < 0.1. Otherwise the regression was forced to pass through the origin and the intercept is omitted.

64

2.3. Climatic Classification

2.3.1. The Holdridge Life Zone Classification

The Holdridge Life Zone diagram was proposed to relate the character of natural vegetation associations to climatic indices (Holdridge [1947], [1964]). Life zones are depicted by a series of hexagons formed by intersecting intervals of climate variables on logarithmic axes in a triangular coordinate system (Fig. 41a). Two variables, the average biotemperature (ABT) and the mean annual precipitation total (APPT), uniquely determine the classification. ABT is the average annual temperature with negative temperatures set to zero. The third variable in the diagram, the annual potential evapotranspiration ratio (APETR), is the ratio of mean annual potential evapotranspiration (APE) over mean annual precipitation. Since APE is calculated as a linear function of biotemperature (the constant of proportionality is 58.93), APETR depends on the two primary variables ABT and APPT. The applicability of the approach to derive APE from ABT is discussed in chapter III. Zones of equal APPT are parallel to the left triangle side, zones of equal APETR to the right triangle side, and zones of equal ABT to its base. One additional division in the Holdridge system is based upon the occurrence of killing frost. This division is along the critical temperature line of ABT = 17°C that divides the hexagons between 12 and 24 °C into warm temperate and subtropical zones (Fig. 41a).

Holdridge and co-workers extensively tested the classification in the tropics. Subsequently, Holdridge Life Zone maps have been developed for a number of regions, for example by Sawer and Lindsey [1963] (eastern and central USA), by Ewel and Whitmore [1973] (Puerto Rico), or by Tosi [1983] (Brazil). Although this classification cannot account for all aspects of natural vegetation distributions, it has been shown that it produces a reasonable representation of present-day ecosystems. For this reason, and because of its easy applicability, the Holdridge Triangle has also been used for the modelling of vegetation distributions coupled to general circulation models (e.g., Emanuel et al. [1985], Prentice [1990], Henderson-Sellers [1993]).

In this study, the Holdridge Triangle was used to determine the average climatic situation of the investigated river basins, as well as their climatic variability (see below). It was also used as a tool to investigate for many of the environmental parameters their relationships with climate. ABT was calculated by setting negative monthly temperature values to zero, and then forming the annual mean. For all continental grid points in a 0.5° x 0.5° grid point resolution the position within the triangle was determined. Grid points falling outside the triangle were assigned to the hexagons being nearest to them.

It has been proposed to rise the limiting temperature in the Holdridge Triangle between the warm temperate and subtropical zones to a value of ABT = 21°C (Emanuel et al. [1985], Henderson-Sellers [1993]) in order to better reflect observed vegetation distributions. The data I used show that globally differences between ABT and AT are negligible above a value of AT = 15°C (Fig. 42), which means that frost effects cannot be detected any more on the basis of the monthly values. This does, of course, say nothing about possible diurnal, daily, or weekly variations, which also can affect vegetation distributions. However, I assume that such short-time variations may be of smaller importance for hydroclimatic relationships as investigated in this study, and I remained the limiting frost temperature in the Holdridge Triangle untouched.

2.3.2. Climatic Classification and Climatic Variability of the River Basins

Figure 43 illustrates the climatic variability for the 0.5° x 0.5° grid elements of the Mackenzie Basin in comparison with the Indus Basin in the Holdridge Triangle: the Mackenzie is much more

65

Tro

pica

l

Sub

trop

ical

War

m T

empe

rate

Coo

l Tem

pera

te

Bor

eal

Sub

pola

r

Pol

ar

Des

ert

Ste

ppe

Des

ert

Des

ert

Dry

Tund

ra

Moi

st

Tund

ra

Wet

Tund

ra

Rai

n

Tund

ra

Rai

nFo

rest

Wet

Fore

stM

oist

Fore

stD

ryB

ush

Rai

n

Fore

st

Wet

Fore

st

Moi

st

Fore

st

Des

ert

Bush

Rai

n

Rai

nFo

rest

Fore

st

Wet

Wet

Fore

st

Fore

st

Moi

st

Moi

st

Fore

st

Fore

st

Dry

Dry

Fore

st

Fore

st

Thor

n

Thor

nW

oodl

and

Ste

ppe

Des

ert

Bus

h

Des

ert

Bus

h

Des

ert

Des

ert

Rai

nFo

rest

Wet

Fore

stM

oist

Fore

stD

ryFo

rest

Ver

yD

ry F

ores

tTh

orn

Woo

dlan

dD

eser

tBu

sh

80

00

40

00

20

00

10

00

50

0

25

0

12

5

62

.50

.12

5

0.2

5

0.5

0

1.0

0

8.0

0

2.0

0

4.0

0

16

.00

32

.00

1.5

3 6 12

17

24

5 5

5 5

5 5

5 5

5

3 4

4

44

6

6

6 6

8 8

8 8

8 8 7

7 7

7 7

7

3 3 3

3 3

3 3

1,2

1,2

1,2

1,2

1,2

1,2

(b)

(a)

AB

T (

°C)

AP

PT

(m

m)

AP

ET

R

F

ig. 4

1 T

he H

oldr

idge

Life

Zon

e Tr

iang

le (a

) and

the

upon

bas

ed c

limat

ic c

lass

ifica

tion

appl

ied

in th

is s

tudy

(b):

1, P

olar

, with

ice;

2, P

olar

, with

out i

ce;

3, D

eser

t; 4,

Tem

pera

te D

ry; 5

, Tun

dra

and

Taig

a; 6

, Tro

pica

l Dry

; 7, T

empe

rate

Wet

; 8, T

ropi

cal W

et. F

or fu

rthe

r exp

lana

tions

, see

text

.

66

4 6 8 10 12 14 16 18

1

2 Std.-DeviationAverage

ABT/ AT

AT (°C)

Fig. 42 Ratio of ABT over AT versus AT, as determined for 2° temperature slices. Calculations were done on the basis of all continental grid points falling into the corresponding temperature ranges.

homogeneous in comparison to the Indus, even if the basin of the Indus is nearly as twice as great. In order to quantify the climatic variability of the investigated river basins, I simplified the Holdridge diagram further to an eight fold classification (Fig. 41b) and calculated for each basin the percentage of the total basin area falling in each class, as well as the average climate for the entire basin. The results are shown in Table 5. The following classes were created, using the hexagon lines and not the straight intersect lines as boundary lines: [1] polar, with ice: ABT < 1.5 °C and under permanent ice cover; [2] polar, without ice: ABT < 1.5 °C and without permanent ice cover; [3] deserts: bioclimates

TropicalSubtropicalWarm Temperate

Cool Temperate

Boreal

Subpolar

Polar

ABT (°C)

APETR APPT (mm)

MackenzieIndus

8000

4000

2000

1000

500

250

125

62.50.125

0.25

0.50

1.00

8.00

2.00

4.00

16.00

32.00

1.5

3

6

12

17

24

Fig. 43 Distribution of all 0.5° x 0.5° longitude/latitude grid elements of the Mackenzie Basin and of the Indus Basin in the Holdridge Triangle.

67

Table 5. Abundance of the different climate types in the investigated river basins (in % of total basinarea).

______________________________________________________________________________________________________________ Polar, Polar, Tundra & Temperate Temperate Tropical Tropical Average with ice without ice Taiga Dry Wet Dry Wet Desert Climate [1] [2] [5] [4] [7] [6] [8] [3] Type

______________________________________________________________________________________________________________ Amazon -- -- -- 2.3 3.4 7.1 87.2 -- [8]

Zaire -- -- -- -- -- 16.8 83.2 -- [8] Mississippi -- -- 1.9 31.3 57.8 1.1 7.9 -- [7]

Ob -- -- 62.3 22.9 14.8 -- -- -- [5] Paraná -- -- -- 6.7 2.1 42.5 48.7 -- [8] Yenisei -- -- 97.0 0.8 2.2 -- -- -- [5]

Lena -- 1.1 98.9 -- -- -- -- -- [5] Amur -- -- 41.8 11.6 46.6 -- -- -- [7] Nile -- -- -- -- 1.5 67.6 17.9 13.0 [6]

Changjiang -- -- 5.4 9.0 55.0 6.6 24.0 -- [7] Ganges/Brahmaputra -- -- -- 8.3 7.7 43.7 40.3 -- [8]

Mackenzie -- -- 91.9 -- 8.1 -- -- -- [5] Niger -- -- -- -- -- 72.8 18.2 9.0 [6]

Zambesi -- -- -- -- -- 82.5 17.5 -- [6] Murray -- -- -- 34.9 3.1 51.4 -- 10.6 [6]

St. Lawrence -- -- 10.4 -- 89.6 -- -- -- [7] Orinoco -- -- -- -- -- 30.7 69.3 -- [8]

Tigris/Euphrates -- -- -- 14.7 2.2 20.5 -- 62.6 [6] Indus -- -- -- 7.4 18.5 24.9 7.5 41.7 [6]

Mekong -- -- -- -- 12.0 39.3 48.7 -- [8] Yukon -- -- 100.0 -- -- -- -- -- [5]

Huanghe -- -- 10.2 75.1 14.1 -- -- 0.6 [4] Danube -- -- 0.6 6.2 93.2 -- -- -- [7] Orange -- -- -- 26.8 8.8 27.8 -- 36.6 [6]

Colorado -- -- 5.0 53.0 11.6 16.9 -- 13.5 [4] Columbia -- -- 6.7 30.9 62.4 -- -- -- [7] Kolyma -- 0.6 99.4 -- -- -- -- -- [5]

Sao Francisco -- -- -- -- -- 65.8 34.2 -- [6] Si Kiang -- -- -- -- 6.1 3.1 90.8 -- [8]

Irrawaddy -- -- -- -- 3.2 21.2 75.6 -- [8] Don -- -- -- 19.0 81.0 -- -- -- [7]

Senegal -- -- -- -- -- 83.3 3.6 13.1 [6] Indagirka -- -- 100.0 -- -- -- -- -- [5] Limpopo -- -- -- 7.7 4.3 88.0 -- -- [6]

North Dvina -- -- 100.0 -- -- -- -- -- [5] Godavari -- -- -- -- -- 91.3 8.7 -- [6]

Magdalena -- -- -- 2.2 3.2 15.0 79.6 -- [8] Fraser -- -- 27.3 7.9 64.8 -- -- -- [7] Yana -- -- 100.0 -- -- -- -- -- [5]

Mahandi -- -- -- -- -- 90.0 10.0 -- [6] Liao He -- -- 1.3 30.9 67.8 -- -- -- [7] Rufiji -- -- -- -- -- 68.4 31.6 -- [6]

Rio Negro (Argentine) -- -- -- 64.7 16.9 -- -- 18.4 [4] Hungho -- -- -- -- 3.7 -- 96.3 -- [8] Rhine -- -- 6.7 -- 93.3 -- -- -- [7] Brazos -- -- -- 24.1 -- 61.2 14.7 -- [6] Loire -- -- -- -- 100.0 -- -- -- [7] Rhône -- -- 2.2 13.7 84.1 -- -- -- [7] Tana -- -- -- -- -- 100.0 -- -- [6]

Garonne -- -- -- -- 100.0 -- -- -- [7] Po -- -- -- 6.6 93.4 -- -- -- [7]

Gambia -- -- -- -- -- 71.3 28.7 -- [6] Fly -- -- -- -- -- -- 100.0 -- [8]

Susitna -- -- 100.0 -- -- -- -- -- [5] Purari -- -- -- -- -- -- 100.0 -- [8] Tiber -- -- -- -- 100.0 -- -- -- [7] Rioni -- -- -- -- 100.0 -- -- -- [7]

Severn -- -- -- -- 100.0 -- -- -- [7] Waikato -- -- -- -- 100.0 -- -- -- [7]

Ems -- -- -- -- 100.0 -- -- -- [7]

All Rivers -- 0.1 22.9 9.4 16.5 20.0 27.5 3.6 -- World (a) 12.1 3.2 19.6 7.9 13.9 17.9 20.5 4.9 --

______________________________________________________________________________________________________________ (a) without endoreic regions

68

with APETR > 4.0; [4] temperate dry: bioclimates with 6.0 °C < ABT < 17.0 °C and 4.0 > APETR > 1.0; [5] tundra and taiga: bioclimates with 1.5°C < ABT < 6.0 °C; [6] tropical dry: bioclimates with ABT >17 °C and 4.0 > APETR > 1.0; [7] temperate wet: bioclimates with 6.0 °C < ABT < 17.0 °C and APETR < 1.0; [8] tropical wet: bioclimates with ABT > 17 °C and APETR < 1.0. The distribution of the resulting eight classes over the continents is shown in Figure 44. This classification forms also the basis of all climatic considerations that were made in the following chapters. Note that the intersect APETR = 1 divides dry and wet climate types. Dry climate types are hence water limited with respect to their annual means, while there is an excess of water in the wet climate types. Only for the tundra and taiga climate type I did not apply such a distinction. This is because dry regions in this climate type are very rare in nature (only about 0.5% of the total exoreic tundra and taiga area would fall in dry bioclimates). Moreover, in these regions it is quite problematic to derive reliable estimates for APE, as we will see in chapter III.

Table 5 shows that among the investigated rivers, the Indus, the Nile, the Colorado, the Changjiang, and the Ganges/Brahmaputra are typical examples for rivers that combine a great climatic variability in their watersheds. This means that the parameter averages for these rivers, which I calculated from the data sets, are probably less meaningful in comparison to the other basins. The rivers of the high latitudes (Mackenzie, North Dvina, Yukon) are characterized by a relative low climatic variability. A relative low variability is also found for most of the tropical wet rivers (Amazon, Orinoco), and naturally for the small rivers. Table 5 shows also that the investigated river basins stretch over all major climate types, and that the percentage the overall basins cover in each type is comparable to the general representation of each climate type over the continents. Only the polar climate types are underrepresented.

2.4. Seasonal Variability

All climatic parameters presented on the previous pages are annual means. It is, however, also seasonal variability that can influence river fluxes. In order to quantify effects related to seasonal variability, I calculated several additional variables based upon the mean monthly temperature (MT) and precipitation data (MPPT) extracted from the global data sets. In the following I present three of them. Since the precipitation maps of Korzoun et al. [1977] depict only annual precipitation totals, the monthly values were determined by taking the monthly precipitation data of Legates and Willmott [1992] (the data set corrected for rainguage biases was used - see chapter III). The data were transferred into the percentages of the annual totals and then applied to the gridded APPT data of Korzoun et al. [1977] in order to derive monthly values from this data set. A comparison of both precipitation data sets is discussed in chapter III.

Fournier [1960] found that a heterogeneous distribution of precipitation over the year, i.e. relative dry periods followed by relative wet periods, leads to high mechanical erosion rates in river basins. He proposed that the ratio of the square of MPPT over APPT of the month with the greatest precipitation total could be used to quantify this heterogeneity. In this study, I used a slightly modified form of this ratio (Four) according to CORINE [1992]. It was calculated in the following way:

12

Four = ∑ MPPTi 2 / APPT (2)

i = 1

69

Fig. 44 Distribution of the climate types distinguished in this study. Below 1, Polar, with ice; 1-2, Polar, without ice; 2-3, Deserts; 3-4, Temperate Dry; 4-5, Tundra & Taiga; 5-6 Tropical Dry; 6-7, Temperate Wet; above 7, Tropical Wet.

Fig. 45 Global distribution of Four on the continents (calculated as percentage of mean annual precipitation). Four characterizes the variability of precipitation over the year. It is the sum of thesquare of mean monthly precipitation over mean annual precipitation for all 12 month of the year (seetext).

71

MPPTi is the precipitation total for the month i. The units of Four, MPPT, and APPT are mm. Note that Four can thus theoretically vary between 1/12 of APPT (all month have equal precipitation) and APPT (all precipitation falls in a single month).

Analogous to precipitation, I determined also a variable characterizing the variability of temperature over the year (DT, delta T):

12

DT = ∑ (| AT-MTi |) / 12 (3) i = 1

The units of both MTi, the mean temperature for the month i, and of AT are °C. Finally, also combinations of MT and MPPT are possible. A third variable was determined as a dimensionless index expressing aridity (ArIn). According to CORINE [1992], it was calculated as:

12

ArIn = ∑ (2MTi - MPPTi) x k (4) i = 1

The unit of MTi is again °C, and the unit of MPPTi is mm. The factor k in the equation represents the portion of the month during which 2MTi - MPPTi > 0. Since I only worked with monthly values, it is here either 1 or 0.

13.4 12.4 11.0--

17.0 18.6 12.7 11.1 9.2

19.1 12.9 12.9 11.0 9.4 9.7

18.1 12.8 12.9 12.3 10.7 10.9 --17.1 12.6 13.7 14.6 12.1 10.8 10.7

27.7 25.5 20.6 19.5 14.9 10.9 11.1 --

12.0

12.6ice

no ice

Four

(% of APPT)

17.6 13.2 8.7--

12.7 13.2 14.1 10.8 7.7

10.4 10.0 9.3 8.6 7.2 4.1

7.2 7.4 6.0 5.9 6.0 5.0 --5.6 5.2 4.5 2.8 2.2 1.0 1.2

4.6 3.5 3.4 2.4 1.6 0.6 0.6 --

10.9

8.4ice

no ice

DT

(°C)

Fig. 46 Distributions of DT and Four in the Holdridge Triangle. The values in the hexagons represent the average of all 0.5° x 0.5° longitude/latitude continental grid elements that fall into these hexagons. For further explanations, see text.

Figure 45 shows the regional distribution of Four over the continents, while Figure 46 depicts the distributions of DT and Four in the Holdridge Triangle. Note that Four is here in both cases depicted as percentage of APPT, and not in mm. DT is strongly decreasing from the boreal bioclimates to the tropical bioclimates, reflecting the seasonal variation of insolation related to the variable distance to the sun. At the same time, the values are decreasing towards the humid pole in the triangle (lower right corner). For a given latitudinal band, DT decreases towards the coast because of the oceanic influence. Here, also precipitation is normally greater. In the case of Four, there is also a

72

tendency of relative low seasonal variability towards the humid pole, but no latitudinal zonation is found. In the dry climate types, seasonality of precipitation is more distinct, which is especially the case for the tropical dry climate. Regionally, values are great in the South of Africa and in the East of Asia (neglecting here the Sahara region).

73

CHAPTER III

CONTINENTAL RUNOFF AND ITS RELATION TO CLIMATE AND MORPHOLOGY

3.1. Introduction

Continental runoff (Q) is one of the principal components of the Earth's climate system. It is the result of an excess of precipitation over evapotranspiration on land, and the amount of water discharged by rivers counterbalances the net flux of water to the atmosphere over the oceans, where evapotranspiration exceeds precipitation. Because rivers are an important transport medium linking the continental lithosphere and biosphere to the oceans, the knowledge of the amount and spatial variability of continental runoff is essential for the understanding of biogeochemical cycles at the global scale. For a large number of chemical elements, oceanic residence times can only be derived from river fluxes, and even if average concentrations can be determined with rather good precision, flux estimates can only be as good as the estimates of the corresponding water discharge. This is also important because the river inputs of carbon and nutrients to coastal waters play a key role in maintaining the biological productivity and the rapid cycling of organic matter in the coastal zones. Human caused global warming, extensive cultivation and farming, deforestation, and river damming can alter the discharge patterns of rivers, leading, for example, to inundation, enhanced coastline erosion, or a reduction of biological productivity in the coastal waters. All this may result in severe problems for coastal populations. Over two thirds of the world's cities with populations greater than 2 million inhabitants are located in coastal areas, often in the vicinity of highly productive estuaries or coastal wetlands (International Geosphere Biosphere Program (IGBP) [1995]).

On land, growth and distribution of vegetation as well as organic matter decomposition in the soils are strongly controlled by the availability of water in the soils. These processes are important elements of the global carbon cycle, but a full understanding and modelling of them is not possible without detailed information on local and regional water budgets. Reliable average precipitation fields can nowadays be estimated from a wide set of historical rainguage data, whereas evapotranspiration depends on a number of factors which are difficult to assess over spatial scales, such as insolation, air and soil humidity, wind speed, and advection effects. Runoff is still the most likely way to constrain local and regional water budgets.

In spite of the need of reliable figures in many disciplines of global scale research, there is still some uncertainty about global and regional estimates for present-day runoff. Detailed studies were done about 20 years ago by Baumgartner and Reichel [1975], and by Korzoun et al. [1977] (I refer here to the English version of their atlas that has been published by UNESCO. The original version was released 1974 in Russian). Both studies were published in the form of hand drawn maps for runoff, precipitation, and evapotranspiration. Baumgartner and Reichel [1975] calculated a global runoff value of 39700 km3, while Korzoun et al. [1977] estimated global runoff to be 47000 km3, about 20% greater than the first figure. Both studies compiled various sources to find out detailed water budgets over the continents, but they did not say anything about the controlling factors on which runoff depends. Therefore, it is not possible to apply their results to climate change scenarios or paleoclimatic studies.

74

In this chapter, these problems are addressed. I use a digitized and gridded version of the runoff maps of Korzoun et al. [1977] (in the following the UNESCO runoff map) together with the large number of global environmental data sets considered in this study (see chapter II) in order to investigate the empirical relations between continental runoff and the environmental factors that may influence it. First, the UNESCO runoff map is tested and validated against most recent literature runoff estimates for the group of the 60 digitized river basins. Conformities and disagreements with the literature estimates are discussed and an approach for a refinement of global and regional runoff figures is developed. Then, multiple correlation statistics are applied in order to identify the most significant controlling factors for runoff at the global scale, and an empirical modelling for continental runoff is proposed.

In the literature, there are numerous models and modelling approaches currently being used to assess the impacts of climate change on water resources. They range from simple empirical models to highly sophisticated conceptual and process-based parameter models (for a review see, e.g., Leavesley [1994]). It is not the main purposes of this chapter just to add another model to them, but rather to find out whether basic parametrisations and relationships widely used in hydrology can be confirmed by a global-scale approach, or whether there is some evidence for other controlling factors to be important. Hydrological models are normally developed and validated at local scales, where the variability of the considered parameters is often relative low. This can make it difficult to identify the dominant controlling parameters for runoff, and large-scale extrapolations may be biased if globally important factors are overlooked.


3.2.1. Runoff and Precipitation Data

Both the global distributions of runoff and of precipitation were digitized following the contour lines on the original maps created by Korzoun et al. [1977] for each of the Earth's continents. The inversion of these maps, i.e. the conversion of the rectangular cartesian coordinates of the contour lines into a longitude-latitude coordinate system, is a non-linear problem. It has been solved iteratively with a Nelder-Mead downhill simplex method applied to the nodes of the grid, combined with a least squares procedure (Munhoven and Probst [1995]). To minimize errors, each map grid was red in at least twice, generally four times. The contour lines were then interpolated on a regular longitude-latitude 0.25° x 0.25° grid using a Delaunay triangulation method, and reduced to a final version in a 0.5° x 0.5° resolution. This method yields a specific annual runoff value of 332 mm for the total continental area, and a specific annual precipitation value of 817 mm. This is 3% and 2% greater than the estimates given by Korzoun et al. [1977], respectively . For precipitation, there is a similar agreement when looking at the mean values for 10° latitudinal bands (for runoff, the authors do not give the latitudinal distribution). Deviation is also in the range of a few percent only, except for the band stretching over the southern tip of South America, which may be due to differences in the area estimates. A global map of the runoff distribution is shown in chapter II (Fig. 36).

Validation of the UNESCO runoff map was done using the set of the 60 digitized river basins. They allow to calculate average basin values out of the runoff map, and to compare the values with independent literature estimates. Literature estimates can vary considerably, and such a comparison is strongly dependent on the selection of the best estimates. Whenever it was possible I took them from the WMO database (Global Runoff Data Center Koblenz [1991]), where also additional information on the basin area corresponding to the gauging stations and on the observation period is available with the runoff values. Figure 47 shows the distribution of the gauging stations included in the WMO database. The station area is important because the digitized basin contours span over the whole watersheds, while the gauging stations are normally upstream the river mouths and do not always include all tributaries. This generally tends to underestimate real runoff. In the literature often appears

75

the correct value corresponding to the principal gauging station of the river, but attributed to an area estimate for the whole basin, leading to an erroneous figure for the specific runoff. The opposite way is also found, although less frequently: the specific runoff corresponding to a far upstream gauging station is applied to the whole basin area, and absolute runoff may be overestimated if the basin becomes much drier downstream.

Fig. 47 Global distribution of the gauging stations included in the WMO database (Global Runoff Data Center Koblenz [1991]).

In Table 6 the specific runoff values extracted from the UNESCO runoff map are compared with the literature estimates for the 60 river basins. Figures 48a and 48b plot the comparison for the specific and for the absolute runoff values, respectively. Note that all specific runoff values refer to the basin areas that were calculated in this study. As far as the literature estimates were taken from the WMO database, I corrected them in the following way: when the area corresponding to the gauging station is only slightly smaller than the whole basin area, i.e. the station is close to the river mouth, I linearly increased the runoff with the area increase. Otherwise, I followed also the area increase, but this time weighted with the runoff patterns figured out in the UNESCO runoff map for the basin parts upstream and downstream the gauging station. For rivers not represented in the WMO database, or when only a very few number of years are documented, the values were normally taken from the recently published Global River Index (GLORI) database (Milliman et al. [1995]), which is a huge compilation of literature estimates for world-wide river data. In this case it is not clear whether all values really correspond to the complete basin area, but generally there is a good agreement between the GLORI data and the values taken from the WMO database, what is indicating this. Only for a very few rivers I preferred to take the values from other sources. One exception, for example, is the Susitna River, where the GLORI-value is much greater than all precipitation values I calculated for this river, and the smaller value of Meybeck [1984] was selected.

76

Table 6. Comparison of drainage intensities extracted from the UNESCO runoff map with literature estimates.

______________________________________________________________________________________________________________ Q Q Literature Q Q Literature

River Abbre- Map Literature Source River Abbre- Map Literature Source viation (mm) (mm) viation (mm) (mm)

______________________________________________________________________________________________________________ Amazon Amz 1056 1067 a [29], b Don Don 60 66 a [89]

Zaire Zai 392 373 a [13] Senegal Sen 130 68 a [34] Mississippi Mis 206 151 a [15], b Indagirka Ink 171 163 a [46]

Ob Ob 132 134 a [55] Limpopo Lim 77 15 b Paraná Par 234 189 a [78] North Dvina NDi 304 318 a [104] Yenisei Ien 247 229 a [49] Godavari God 380 323 a [72]

Lena Len 189 216 a [50] Magdalena Mag 1127 843 a [7] Amur Amo 195 178 a [52] Fraser Fra 640 396 a [72] Nile Nil 74 48 b Yana Yan 124 134 a [42]

Changjiang Yts 604 510 b Mahandi Mah 564 495 a [6] Ganges/ Brahmaputra GaBra 870 737 b Liao He LHe 96 85 b

Mackenzie Mac 190 167 a [14] Rufiji Ruf 181 173 b Niger Nig 237 130 b Rio Negro (Argentine) RNe 202 165 a [52]

Zambesi Zam 75 71 b Hungho Hun 761 753 b Murray Mur 29 11 b Rhine Rhi 552 462 a [49]

St. Lawrence SLa 408 404 b Brazos Brz 93 54 a [20] Orinoco Ori 1035 1072 b Loire Loi 311 246 a [116]

Tigris/Euphrates TiEu 102 117 a [7] Rhône Rho 622 559 a [58] Indus Ind 269 263 b Tana Tan 68 81 a [42], c

Mekong Mek 689 544 b Garonne Gar 488 359 d Yukon Yuk 236 249 a [8], b Po Po 592 707 a [63]

Huanghe Hua 100 72 b Gambia Gam 551 133 a [14], eDanube Dan 301 259 a [64] Fly Fly 2079 1325 b Orange Org 28 15 b Susitna Sus 432 368 f

Colorado Cdo 37 28 b Purari Pur 2379 2137 b Columbia Col 313 280 a [109] Tiber Tib 601 466 b Kolyma Kol 245 213 a [50] Rioni Rio 1129 966 a [20]

Sao Francisco SFr 207 193 b Severn Sev 439 485 a [17] Si Kiang SKi 891 646 b Waikato Wai 1142 1074 a [8], b

Irrawaddy Irr 1183 1026 b Ems Ems 300 404 g All basins together 388 356

______________________________________________________________________________________________________________

a Global Runoff Data Center Koblenz [1991] (numbers in brackets give the number of years of observations) b Milliman et al. [1995] c Charania [1988] d Probst [1983] e Lesack et al. [1984] f Meybeck [1984]] g Cadée [1987]

3.2.2. Other Data Sets

For comparison with the digitized and gridded precipitation maps of Korzoun et al. [1977] (in the following the UNESCO precipitation map), I also used in this chapter the precipitation data set of Leemans and Cramer [1992], and the two data sets of Legates and Willmott [1992]. The two latter consist in a data set that has been corrected for methodological rainguage underestimation, and a non-corrected data set. This is interesting because also Korzoun et al. [1977] corrected their precipitation data for the effects of methodological rainguage biases. All other data sets used in this chapter are described in chapter II.

3.2.3. Statistics

In order to run the multiple regression analysis on a grid point level, all data sets used in this study were reduced to a resolution of 2.5° x 2.5° longitude/latitude, after eliminating the grid points

77

0 500 1000 1500 2000 2500

0

500

1000

1500

2000

2500

Gam

Fly

Tropical Wet

Temperate Wet

Tundra & Taiga

Tropical Dry

Temperate Dry

y=x

Fig. 48a Plot of the specificdrainage intensities (mm)extracted from the UNESCOrunoff map (y-axis) versusliterature estimates (x-axis) forthe 60 river basins of Table 6.For abbreviations of the rivernames, see Table 6.

Fig. 48b Comparison oftotal runoff (km3/yr) extracted from theUNESCO runoff mapwith literature estimatesfor the 60 river basins ofTable 6. See Table 6 for abbreviations of the rivernames.

1500 1000 500 0 500 1000 1500

ZaiOri

MekSKiHunFlyYts

SLaDanFra

RhoGarLoiRioTib

EmsIndSFrGodMahRufMur

GamBrz

LenMacKolInk

SusHuaCdo

GaBraParIrrMagPur

MisAmoColRhiPoDonLHeWaiSev

NigTiEuZamNilSenOrgTanLimIenObYukNDiYan

RNe

LiteratureUNESCO-Map

Temperate Dry

Tundra & Taiga

Tropical Dry

Temperate Wet

Tropical Wet

Amz ( x 5)

falling into the endoreic parts of the continents. This results in about 3200 grid points for each parameter. Then the grid points were further classified according to the climatic classification described in chapter II. When classified as desert, they were also excluded from the statistics. In deserts, Korzoun et al. [1977] generally fixed the runoff to an arbitrary value of 1 mm. Moreover, the multiple correlation statistics were also done on the basis of the basin averages of the 60 river basins, which allows to compare the results with the results of the grid point statistics. The creation of the

78

UNESCO runoff map necessarily required interpolations over large areas (Korzoun et al. [1977]), and I mainly did this comparison to test whether the significance of the results from the grid points statistics could be influenced by the subjective criteria of the authors guiding their interpolations. However, as discussed in chapter II, also the basin averages are not always very meaningful to determine the major controls for runoff. Both methods are therefore complementary in a certain way.

Details on the applied statistical methods are given in chapter II. In the grid point statistics, it could happen that CP did not fall below (p + 1), which means that all parameters were retained to be significant. In these cases, I normally only retained parameter combinations that increased the correlation coefficient in the regressions with more than 0.01 compared to a model with less variables.

3.3. Validation of the UNESCO Runoff Map

3.3.1. Comparison with Literature Estimates

The comparison in Table 6 reveals that the runoff values extracted from the UNESCO runoff map are generally very close to the literature estimates. This means that the work of Korzoun et al. [1977] gives a realistic picture of the runoff distribution over the continents. The actually available information on river runoff does not much change their findings. However, it is obvious that the UNESCO runoff map tends to overestimate the values rather than to underestimate them. The following equation describes the relation between both values:

QLit = 0.84 x QUNESCO (5)

r2 = 0.94, P < 0.0001, n = 60

QLit is the literature runoff in mm, and QUNESCO is the map derived runoff in mm. P is the significance level, r the correlation coefficient, and n the number of river basins considered in the regression. One can expect that some of the differences between the map and the literature values are due to the fact that they refer to different observation periods. It is not clear what information exactly was available to Korzoun and his collaborators when they created their maps, but there is no evidence that the literature estimates I used for the comparison always refer to the longer observation periods. For example, one of the most striking discrepancies between the values in Table 6 appears for the Gambia River, where the UNESCO runoff map yields an about 4 times greater specific runoff than found in the literature. The literature estimate is taken from Lesack et al. [1984] (estimated from fig. 2, page 818) and corresponds to the observation period of 1953-1980. It has to be noted that from 1968-1980 the runoff of the Gambia dropped down to a value only half as great as for the period from 1953-1967 (351 m3/sec and 179 m3/sec, respectively). This is considerably lowering the overall average. The adjacent Senegal River shows a similar reduction of its average runoff for the same periods (from 908 m3/sec to 485 m3/sec, respectively). But from 1904 to 1953 the Senegal had a mean discharge of 995 m3/sec, and its overall 1903 to 1984 discharge average of 863 m3/sec is very close to its average for the 1953-1967 period (the values were determined from Dümenil et al. [1993], page 36). This is a strong indication that the long-term average of the Gambia should be much greater than the value given by Lesack et al. [1984]. It should be closer to the value of the UNESCO runoff map.

The discrepancy for the Limpopo River is probably related to the same reason. The few data that exist for this river show highly variable runoff patterns. The WMO database includes only some incomplete time series from 1976-1979, corresponding more or less to the literature estimate in Table 6. Especially for rivers of the tropical dry climate, runoff can extremely vary between years and even between decades, and average values only based upon a few years can considerably deviate from long-term averages.

79

Another reason that may account for some of the discrepancies in Table 6 is related to the spatial interpolation of the values. If the density of the station network is low, it is not possible to represent the local patchiness of runoff and precipitation patterns in certain regions with precision details. This has naturally greater effects on small river basins. One example for this effect can be the Fly River. Its literature estimate in Table 6 is only about 60% of the value of the adjacent Purari River, whereas the hydroclimatic characteristics extracted from the data sets considered in this work are very similar for both rivers. This implies that the local climatic variability in this region was smoothed by the spatial interpolation .

If one assumes that the above discussed effects are averaged out over the 60 river basins, equation 5 indicates that the UNESCO runoff map overestimates real runoff with about 16%. This value is calculated on the basis of the specific runoff values, and does not account for the great differences of the basin sizes. If one sums up the absolute runoff values, once for the literature estimates and once for the map values, the overestimation decreases to about 9% (see Tab. 6). The fact that the UNESCO runoff map tends to overestimate the values can be at least partly explained with the above mentioned interpolation effect. Willmott and Legates [1991] showed for the spatial interpolation of precipitation data that a too sparse station network clearly tends to overestimate regional and global budgets. This is mainly related to the fact that dry regions are normally underrepresented in station networks. Although Korzoun et al. [1977] used more than 18000 hydrological stations to create their maps, one has to keep in mind that these stations are probably very unevenly distributed over the globe. A slight overestimation of the values in the UNESCO runoff map due to an insufficient station network is therefore not unlikely.

There is also an other effect contributing to the discrepancy between the UNESCO runoff map values and the literature estimates. Figure 48b indicates that the overestimation of the map seems to be greater in the tropical dry climate than in the other climate types. This can be attributed to direct evaporation of river water when the river passes through hot and dry regions, which can have an important effect for certain rivers (as this has been already shown in chapter I for the Indus River). One extreme case is also the Nile River when it passes in its lower course through the eastern part of the Sahara desert. The WMO database reports for the period of 1973-1984 that the Nile's average runoff decreases with about 55% from the gauging stations Gaafra, Esna, Nag Hammadi, and Assiut (1825 m3/sec, 1671 m3/sec, 1493 m3/sec, and 1175 m3/sec, respectively), all situated downstream in the dessert part of the basin. The coverage of the Nile basin with gauging stations is good and the times series comprise long periods, but his effect alone can explain the discrepancy of the values in Table 6. Another example is found in the WMO dataset for the Niger River: for the period of 1945-1975, its average discharge decreases downstream with about 35% from the stations Koulikoro to Niamey (from 1565 m3/sec to 1011 m3/sec). They are situated in the dry north-eastern part of the basin (for a discussion of the Niger Basin, see also chapter I). Because Korzoun et al. [1977] did not depict negative runoff values in their map, they could not account for this effect: all water that runs off in one part of the basin has finally to end up in the ocean. This leads to an overestimation of the runoff values especially in basins of the dry climates.

3.3.2. Determination of Regional and Global Figures for Continental Runoff

I applied the following method to test if the deviation of the UNESCO runoff map values from the literature estimates in Table 6 may be different for different climates. As shown in chapter II, the climatic classification of the river basins investigated in this study according to their average climatic situation (as this is done, for example, in the Figures 48a and 48b) is not without problems because some basins are very large in size and belong to several climate types, while others stay nearly exclusively within one type. For this reason, I calculated from the UNESCO runoff map the specific drainage intensities for all climatic subunits within the basins. This means that all basin grid points that fall into the same climatic type are taken as one subunit, which does not necessarily mean that this

80

Table 7. Estimated fluxes of river water to the oceans. ____________________________________________________________________________________________________________

Area Runoff _________________________________________ _______________________________ Drainage Area on Land _______________________________ Ocean Ocean Total Total Covered Total Total Originating Volume Surface Exoreic Endoreic by Ice Exoreic Endoreic from Ice (106 km3) (106 km2) (106 km2) (106 km2) (%) (km3/yr) (km3/yr) (%)

____________________________________________________________________________________________________________

Polar, under ice [1] -- -- 14.7 0.0 100.0 2285 0 100.0 Polar, without ice [2] -- -- 3.9 0.0 0.0 761 0 0.0 Tundra & Taiga [5] -- -- 23.8 0.8 2.2 7271 118 4.6 Temperate Dry [4] -- -- 9.6 6.5 0.0 729 196 0.0 Temperate Wet [7] -- -- 16.9 1.6 0.0 7756 313 0.0 Tropical Dry [6] -- -- 21.8 3.7 0.0 3101 87 0.0 Tropical Wet [8] -- -- 24.9 0.2 0.0 22398 31 0.0 Desert [3] -- -- 5.9 18.0 0.0 66 46 0.0

Total -- -- 121.6 30.9 10.0 44367 791 5.8

Africa -- -- 18.3 14.7 0.0 4120 132 0.0 Europe -- -- 9.7 3.0 0.8 3128 315 1.4 North America -- -- 25.0 0.4 7.9 7819 7 8.7 South America -- -- 17.7 0.4 0.0 11150 18 0.0 Asia -- -- 32.5 8.5 0.1 15322 310 0.0 Australia -- -- 4.5 3.9 0.0 773 10 0.0 Antarctis -- -- 13.8 0.0 94.7 2055 0 92.1

Total -- -- 121.6 30.9 10.0 44367 791 5.8

Arctic Ocean 14.7 11.0 17.6 -- 3.3 3364 -- 3.7 North Atlantic 137.5 41.9 28.8 -- 3.8 13952 -- 3.3 South Atlantic 171.8 42.5 17.0 -- 0.0 5074 -- 0.0 Pacific 670.4 167.2 21.1 -- 0.3 13670 -- 1.0 Indian Ocean 263.7 70.1 16.6 -- 0.0 5166 -- 0.0 Mediterranean 3.9 2.7 6.7 -- 0.0 1087 -- 0.0 below 60° South 69.5 20.8 13.8 -- 94.7 2055 -- 92.1

Total 1331.4 356.2 121.6 -- 10.0 44367 -- 5.8

____________________________________________________________________________________________________________

subunit is one geographically connected region. The average specific drainage intensity for an entire basin (QUNESCO) is then the sum of the specific subunit values (Qi), multiplied by the percentage that the units occupy in the basin (ai), divided by 100:

QUNESCO = ( a1Q1 + a2Q2 + a3Q3 + a4Q4 + ... + aiQi ) / 100 (6)

Areas are in percent, drainage intensities in mm, and the indices for the climate types (i) follow the assignment defined in chapter II (they are also shown in Table 7). A multiple regression between the literature estimates (QLit) of all basins and the area weighted drainage intensities of all climatic subunits (aiQi) in these basins calculated with the UNESCO runoff map can then help to identify the importance of each climate type with regard to the deviation between QLit and QUNESCO. It leads to the following relationship:

QLit = ( 0.90 a5Q5 + 0.80 a6Q6 + 0.90 a7Q7 + 0.90 a8Q8 ) / 100 (7)

n = 58, r2 = 0.97, P < 0.0001 for all aiQi

The other climate types (i = 1, 2, 3, 4) were not found to be significant, what can be explained with the generally low runoff contribution from these climates, and/ or the absence of these climate types in the basins. Note that the Gambia and the Fly rivers were omitted in the regression because of their obviously somewhat outstanding values (Fig. 48a). It is interesting to note that for all significant climate types the regression coefficients are 0.9, except for the tropical dry climate. Here, the coefficient is 0.8. The abundance of 0.9 as regression coefficient for most of the climate types indicates that the UNESCO runoff map overestimates on average the real values with about 10%. On

81

the other hand, one can conclude that the smaller regression coefficient in the tropical dry climate additionally represents the loss of water related to the evaporation of river water.

Latitude N

-80 -60 -40 -20 0 20 40 60 80

0

1000

2000

3000

4000

5000

6000Baumgartner & Reichel [1975]

UNESCO runoffmap (corrected)

Fig. 49a Holospheric runoff distribution (km3/yr, y-axis) as determined by Baumgartner and Reichel [1975], and as determined in this study on the basis of the UNESCO runoff map.

Fig. 49b Holospheric runoffdistribution of the UNESCOrunoff map (in % of totalrunoff, y-axis), as it isgenerated on land, and as itenters the oceans. The latterwas created by coupling themap values to the 2° x 2.5°latitude/longitude riverrouting scheme of Miller etal. [1994] - see chapter VI.Abbreviated river names(see Table 6) show the important contribution ofthese rivers in thecorresponding bands.

-80 -60 -40 -20 0 20 40 60 80

0

2

4

6

8

10

12

14

16

18

ContinentsOceans

Par

Zai

Amz

Ori, MekGaBra

Mis, YtsOb, Ien, Len

Latitude N

Although these values only allow a very generalized point of view, they can be used to determine continental runoff budgets with respect to major climates, different continents, and different oceans on the basis of the UNESCO runoff map. The result is shown in Table 7. To calculate the figures, all grid point values were reduced by 10% to account for the above discussed effect of overestimation. Since it can be supposed that this effect is mainly related to the spatial interpolation of the data, the reduction was also applied to the climate types that are not significant in equation 7. In the tropical dry climate, however, 20% was subtracted, taking the additional 10% as an estimate for the evaporation of river water.

This yields a global figure of 44370 km3 of water that is discharged to the oceans every year. The value is almost half way between the Baumgartner and Reichel [1975] estimate, and the original estimate of Korzoun et al. [1977]. It can be seen in Figure 49a that discrepancies between both studies

82

mainly occur between the latitudes from 5° to 25° N, and in the latitudes north of 35° N. Although I corrected the original values of Korzoun et al. [1977], the UNESCO runoff map depicts still throughout greater values. It generally shows more runoff in the tropical regions in southern Asia and in Oceania than the maps of Baumgartner and Reichel [1975], which can partly explain the difference in the 5° to 25° latitudinal band. In this part of the world, river records are sparse and comprise often only short time series. Another feature which may be seen in this latitudinal band has been already mentioned: the great values for the Gambia and Senegal rivers may suggest that the UNESCO runoff map depicts too much runoff in the Southwest of the Sahara desert in Africa. This is, however, difficult to assess because of the distinct long-term variations observed for both rivers in this regions (see above).

Nevertheless, most of the difference between the global runoff estimates of both studies is due to the lower amount of runoff that Baumgartner and Reichel [1975] attributed to the latitudes north of 35° N. Here, the coverage with gauging station is normally good and the literature estimates for the rivers in this region compare well with the values of the UNESCO runoff map. It seems that Baumgartner and Reichel [1975] considerably underestimated runoff in this part of the world. The only region where they found much more runoff than Korzoun et al. [1977] is the Australian continent: the estimate they give in their publication is about three times greater than the value I calculated from the UNESCO runoff map (2394 km3 compared to 773 km3 ). There is no obvious explanation for that. The corresponding drainage areas are similar in both studies.

It is interesting to note from Table 7 that about half of the water that is discharged to the oceans comes from the tropical wet climate type. The next important climates are the temperate wet and the tundra and taiga climates, which contribute about 18% and 16% of global runoff, respectively. South America is the continent with the highest specific runoff. It is about two times greater than for North America or for Europe. Nearly 65% of continental runoff is formed north of the equator, but due to the transport by rivers, more than 75% of it enters the oceans in the Northern Hemisphere (Fig. 49b). From the latter value, again more than half of the water input occurs in the 0° to 25° latitudinal belt. The Mediterranean Sea and the Arctic Ocean receive the greatest specific freshwater inputs with respect to water volumes. Their river water inputs are about 14 and 12 times greater than the inputs into the Pacific and into the Indian Ocean, respectively. The latter represent the oceans with the lowest specific river water input. Also for the North Atlantic, freshwater input is still five times greater than for the Pacific and the Indian Ocean.

3.4. Key Parameters for Continental Runoff

3.4.1. Limiting Conditions

Because water storage on the continents can be considered to be constant for time scales of one to several decades, the water balance in a river basin has to obey the fundamental relationship:

Q = APPT - AAE (8)

where AAE is the annual actual evapotranspiration mean, and APPT is the annual precipitation total. In dry regions, where APPT is less than annual potential evapotranspiration (APE), the maximal amount of water that is lost to the atmosphere by evapotranspiration is limited by:

AAE < APPT (9)

Whereas in wet river regions, where APPT exceeds APE, the upper limit for evapotranspiration is given by:

83

AAE < APE (10)

In river basins, the variation of precipitation and potential evapotranspiration throughout the year may result in variations in the soil moisture content and, consequently, in a value of AAE intermediate between the two limiting conditions of equation 9 and 10.

The first attempts to link AAE to APPT and APE were made in the early years of this century by Schreiber [1904] and by Ol'dekop [1911] on the basis of available measurements of rainfall and runoff for different river basins. Later, similar relationships were proposed on the basis of other data (e.g., Bagrov [1953], Turc [1954], Budyko and Zubenok [1961], Pike [1964]). For example, Pike [1964] proposed as a modification of Turc's [1954] formula that the ratio of actual to potential evapotranspiration obeys the following relationship:

AAE / APE = (APPT / APE) / (1 + (APPT / APE) 2) 1/2 (11)

The fundamental assumption in all the suggested formulae is that the ratio of actual to potential evapotranspiration can be expressed as a function of the ratio of precipitation over potential evapotranspiration. AAE becomes closer to APE as more as APPT exceeds APE.

0 .0 0 .5 1 .0 1 .5 2 .0 2 .5 3 .0 3 .5 4 .0

0 .0

0 .2

0 .4

0 .6

0 .8

1 .0AAE = APE

Forest Tundra

DRY WET

A

BC

A : O l'd e k o p [ 1 9 1 1 ]

B : P ik e [ 1 9 6 4 ]

C : S c h re ib e r [ 1 9 0 4 ]

Des- ert

Step-pe

AAE = APPT

APPT / APE

AAE / APE

Fig. 50 Limiting conditions for annual actual evapotranspiration (AAE). In addition, three empirical relationships linking AAE to APPT and APE are show which have been proposed in the literature. For further explanations, see text (figure from Dooge [1992]).

Figure 50 illustrates the empirical relationships for AAE proposed by Schreiber [1904], by Ol'dekop [1911], and by Pike [1964] together with the limiting conditions of equation 9 and 10. Additionally, the biome classification of Budyko (Budyko [1950], [1974]) is added along the abscissa of the figure. Analogous to the classification of Holdridge [1947] (see chapter II), also Budyko uses the ratio of APE over APPT to relate biomes to climatic parameters. Note that Budyko's separation of the forest and steppe biomes represents the separation of wet and dry climates followed in this study. As we will see below, Figure 50 is a useful tool to compare the different estimates for APPT and APE

84

which can be derived for the 60 river basins of Table 6 on the basis of the water budgets resulting from the runoff data.

3.4.2. Potential Evapotranspiration

Potential evaporation is one of the principal elements determining the water budget in a river basin. A reliable estimation of this parameter is crucial in order to investigate the controlling factors for continental runoff at the global scale. Thornthwaite [1948] pointed out that potential evapotranspiration depends mainly on the net radiational heating of the land surface, and, to a lesser extent, on the surface air wind speed, temperature, and relative humidity. However, he believed that it would take a long time before either enough measurements or sufficiently accurate calculations of the net radiational heating of the Earth's surface could be available. For these reasons, he proposed an empirical approach to determine potential evapotranspiration, relating it uniquely to the surface air temperature according to the following equations:

MPE' = 0, for MT < 0 (12.1)

MPE' = 16 (10 MT/I) a, for 0 < MT < 26.5 (12.2)

MPE' = - 415.85 + 32.24 MT - 0.43 MT 2, for MT > 26.5 (12.3)

MPE' is the non-adjusted mean monthly potential evapotranspiration in mm, MT is the mean monthly air temperature in °C, and I and a are calculated according to:

12

I = ∑ (MT/5) 1.514 (13) 1

a = 6.75 x 10-7 I 3 - 7.7 x 10-5 I 2 + 1.79 x 10-2 I + 0.49 (14)

To account for variable day and month lengths, the effective mean monthly evapotranspiration (MPE) is adjusted to:

MPE = MPE' x (ML/30) x (DL/12) (15)

ML is the length of the month in days, and DL is the duration of daylength in hours. Although it is now nearly 50 years ago that Thornthwaite [1948] proposed his formulae, they are still widely used to determine large-scale terrestrial water budgets (e.g., Willmott et al. [1985], Vorösmarty et al. [1989], Mintz and Serafini [1992]).

Eleven years after the publication of Thornthwaite's approach, Holdridge [1959] proposed a much simpler method to estimate potential evapotranspiration. Also he related APE exclusively to surface temperature, with:

APE = 58.93 x ABT (16)

The unit of APE is mm, and ABT is the mean annual biotemperature in °C. I recall here from chapter II that the latter is calculated as the sum of only the mean monthly temperatures above 0 °C, divided by 12. Holdridge supposed that such a simplification can be made because of the effects of natural vegetation on evapotranspiration. He argued that "local variations in edaphic and atmospheric factors sufficient to cause an appreciable change in either evaporation or transpiration, or in both, are counterbalanced by the different physiognomies of the natural vegetation, developed in the past through evolutionary processes, which bring the actual evapotranspiration into equilibrium with the potential evapotranspiration rate and the moisture available" (Holdridge [1959]).

85

Because of their relative easy applicability, I use in the following both the empirical relationships of Thornthwaite [1948] and of Holdridge [1959] as means to derive estimates for APE. Another commonly used expression for APE was proposed by Penman [1948]. Due to parameter requirements of his formulae that were not available for this study, (e.g., wind speed), I did not test his relationships.

Figure 51 plots mean APE versus mean annual temperature (AT) for the 60 river basins of Table 6, calculated both according to the Thornthwaite formulae (Thw), and according to the Holdridge formula (Hld). One can see that in the basins with AT > 15 °C, Thw and Hld differ not much from each other. The latter is here, of course, a linear function of AT because of the absence of months with MT < 0 in this temperature range. Below 15 °C, however, Thw yields considerably greater values than Hld.

-20 -10 0 10 20 30 40

0

500

1000

1500

2000

APE = Thw, cAPE = Thw

APE = Hld

Fig. 51 Plot of mean annual potential evapotranspiration (APE; in mm, y-axis) versus mean annual temperature (AT; in °C, x-axis) for the 60 river basins of Table 6. APE was both calculated with the empirical formulae of Thornthwaite [1948](Thw), and with the formula of Holdridge [1959] (Hld). Thw,c means that the Thw values were corrected for the effects of permafrost -see text.

The formulae of Thornthwaite [1948] underlie the assumption that soil water is always available for evapotranspiration, which does naturally not take into account freezing of the soil. When the soil is frozen, there is almost no infiltration, and it does not matter whether the supply of water to the soil is an assumed wintertime rainfall or an actual spring time snow melt. In both cases the water will be removed as runoff before the soil has thawed and before the infiltration and evapotranspiration can begin. In order to get an idea to what extent soil freezing could influence APE in the river basins, I also modified Thw in Figure 51 in the way that APE was set to zero for all grid points with AT < - 4 °C, resulting in Thw,c. This temperature limit is often cited to characterize permafrost soils (e.g., Starkel [1988]). It becomes evident from this simple modification that freezing of soil could have an important effect on APE, even if this is, of course, a very simplistic correction because also in permafrost regions, a part of the soil water should evaporate due to a seasonal thawing up of the upper soil layers that can be supposed. On the other hand, also for grid points with AT > - 4 °C evapotranspiration may already be hindered because of freezing effects.

Holdridge [1959] based his approach to derive APE upon the fundamental assumption that the climatic conditions determining potential evapotranspiration should always be reflected by the

86

abundant vegetation. It may explain why his formula yields much values in the low temperature river basins in Figure 51. Natural vegetation should adapt as more as soils underlie freezing effects.

3.4.3. Precipitation

Regardless the type of modelling that is applied, all approaches to predict the amount of water running off a drainage basin will depend on the available precipitation data to run the models. Insufficient or inexact rain and snowfall data will either lead to unrealistic results, or when the data are used to adjust the models, to wrong parametrisations. There exist nowadays a number of gridded precipitation data sets that are widely used in many disciplines in global scale research. In a study like this, one has also to test whether they are consistent with the global runoff distribution, and, of course, also whether the exchange of the data sets can influence the results obtained in this study.

For long-term considerations, the following condition has to be respected as limiting condition:

0 < RR < 1 (17)

RR is the runoff ratio that is defined as the ratio of Q over APPT. If RR is 0, all water that enters the basin via precipitation is lost by evapotranspiration (in this case it makes naturally no sense to speak of a drainage basin), whereas a RR of 1 means that all precipitation ends up as runoff.

As mentioned above, I investigated four different precipitation data sets in this chapter. These are the two data sets of Legates and Willmott [1992] (L&W and L&Wc, respectively) the data set of Leemans and Cramer [1992] (L&C), and the precipitation maps of Korzoun et al. [1977] that were digitized and gridded in this study (K). All data sets exist in a spatial resolution of 0.5° x 0.5° longitude/latitude. Two of the data sets (K and L&Wc) were corrected for a systematic bias of the values related to rainguage measurements, which tend to underestimate real rain and snowfall at the gauge sites (Willmott and Legates [1991]). The degree of underestimation varies with the gauge type and sitting. It mainly depends on the wind field around the gauge, on evaporation from the gauge, and on instrumental errors. Willmott and Legates [1991] reported that annually and spatially averaged underestimates may exceed 40% in the frequently snow-covered northern latitudes, while they decrease to between 10% and 20% in the mid-latitudes. Undercatches of less than 5% are typical in the tropics. These figures agree well with the correction factors that Korzoun et al. [1977] applied to derive their precipitation maps (in their publication, they summarize the applied correction for 10° latitudinal bands).

Because of the correction, K and L&Wc include greater precipitation values on the continents than the two other data sets: they result in global continental averages of 872 mm and 849 mm, respectively. The continental averages is only 780 mm for the L&W, and 776 mm for the L&C data sets. The corresponding area is about 134 x 106 km2. Note that this area is less than the continental area given in Table 7. The discrepancy is due to the fact that the L&C data set includes no values for the glaciated regions in the high latitudes, and for comparison purposes, I related all averages to the continental area which is covered by the L&C data (applied to the total area of Table 7, K, L&Wc, and L&W result in mean values of 817 mm, 834 mm, and 744 mm, respectively).

Figures 52a and 52b compare the four data sets with the data of the UNESCO runoff map on the basis of the holospheric distributions. Calculations were done with the data sets in the reduced 2.5° x 2.5° longitude/latitude resolution because it can be supposed that this may smooth to some extent the differences related to the different interpolation methods (see below). Figure 52a shows that the corrected data sets have greater APPT values than the non-corrected ones over nearly the complete latitudinal range, but differences are greatest in the northern latitudes north of 35 °N (not considering here the southernmost latitudes because of the very small continental area). The K data set depicts throughout greater precipitation, and correlates also much better with runoff than the three other data sets. The latter is not very surprising because this data set was created in the same way as the runoff

87

data set. Spatial interpolations were based upon a linking of station data in a more or less subjective way by drawing isolines on topographic maps (sometimes also by reproducing the patterns taken from already existing maps that were generated in the same way), whereas the other precipitation data sets (L&W, L&Wc, L&C) were created by computing irregularly spaced station networks with automated interpolation algorithms to uniformly spaced grid point data sets (Legates [1989], Legates and Willmott [1990], Leemans and Cramer [1992]).

0

500

1000

1500

2000

2500

0

1

2

3

Latitude N

-60 -40 -20 0 20 40 60 80

0

1

L & WL & W, cL & CK

L & WL & W, cL & CK

L & WL & W, cL & CK

AP

PT

(mm

)A

vera

ge R

unof

f Rat

ioQ

ver

sus

AP

PT

Cor

rela

tion

Coe

ffici

ent

Average Runoff R

atio

0123

0123

0123

Latitude N

-60 -40 -20 0 20 40 60 80

0123

L & W

L & W, c

L & C

K

Fig. 52a Comparison of the holospheric distribution of precipitation and runoff for four different precipitation data sets. For further information, see text.

Fig. 52b Standard deviation of the average runoff ratio in Fig. 52a. For further information, see text.

Both methods may have advantages and disadvantages, and it is beyond of the scope of this study to criticize the work of the authors. Nevertheless, it becomes evident from the runoff ratios that, as more as one goes north, precipitation in the non-corrected data sets are not always sufficient to supply the water that is running off the continents. Respecting equation 17, one can calculate for each data set the minimal amount of required additional precipitation to bring the data in agreement with Q. For the total continents (here I take again only a reference area of about 134 x 106 km2) this additional water demand results in values of 17 mm, 14 mm, 9 mm, and 1 mm for the L&C, L&W, L&Wc, and K data sets, respectively. Relating the first value only to the latitudes north of 30° N, it rises up to 24 mm, corresponding to an area of about 67.5 106 km2, and a specific drainage intensity of 256 mm. In other words: in these latitudes, about 10% of total runoff cannot be supplied by precipitation in the data set of L&C. Bearing in mind that this is really a minimum because runoff ratios close to 1 are very rare in nature, this is a strong argument for the utilisation of rainguage-corrected precipitation data sets in global scale investigations from the runoff point of view.

Finally it is also interesting to look at standard deviation of the runoff ratios (Figure 52b). Due to the better correlation of APPT and Q, standard deviation for the K data set is considerably lower than for the other data sets. Differences between the data sets are small for the tropical latitudes, and

88

increase towards the poles. It can be expected that the much lower standard deviation for the K data set may be at least partly related to the subjective 'by-eye' interpolation of Korzoun et al. [1977]. However, note that the comparison of the two Legates and Willmott [1992] data sets (L&W and L&Wc) reveals that the correction of the data for the rainguage undercatches slightly reduces standard deviation, especially in the 40° to 50° N latitude band. Both data sets were created with the same station network and the same interpolation routines. They generally agree better with the UNESCO runoff map distribution than the L&C dataset. This is also the case for the non-corrected L&W data, although the global precipitation totals of the L&W and L&C data sets are quite similar (see above). The runoff ratios determined with L&C depict a slightly larger standard deviation in the 20° to 40° N latitude bands compared with the runoff ratios determined with the L&W and L&Wc data sets.

3.4.4. Best Estimates

In the previous sections, I presented two methods to derive estimates for APE (Thw, Hld) and four data sets to supply estimates for APPT (K, L&Wc, L&W, L&C). The Figures 53a-h give an overview of the possible combinations of these estimates: since Q is known for the 60 river basins of Table 6, the corresponding AAE values can be calculated for each given APPT value (equation 8), and the basins were plotted in the graph of Figure 50. In the resulting eight plots, only the relationship of Pike [1964] is show (equation 11). The river basins are additionally grouped according to the climatic classification followed in this study. Note, however, that the classification was established with the Holdridge Life Zone Scheme (Holdridge [1947]), using the K precipitation data set (see chapter II). The distinction in dry and wet climates is therefore only binding for the situation in Figure 53b.

The comparison reveals that when Thw was used for APE, most of the river basins that are classified to be tundra and taiga climate (e.g., the Mackenzie, the Lena, or the Northern Dvina) clearly become water limited. The effect is amplified as there is less water supplied by precipitation (going in the direction K, L&Wc, L&W, L&C). A strong water limitation of these basins is in disagreement with both the biome classifications of Budyko (Budyko [1950], [1974]) and of Holdridge (Holdridge [1947]). Also the biological parameter averages I determined for these basins, e.g. the relative great forest ratios and biomass densities (a table giving most of the determined parameter averages for the basins is shown in chapter V), do not support this. Taking Hld for APE agrees better for these basins, but it leads in many cases to the paradoxical situation that AAE becomes greater than APE, at least as far as the rainguage-corrected precipitation data sets are concerned. One has therefore to state that neither Thw nor Hld seems to be a perfect estimate for APE in the investigated river basins. Looking at the situation in the Figures 53a-d, one tends to conclude that best estimates for real APE often lies somewhere in between Thw and Hld, especially for the rivers which are characterized by low temperatures. As we have seen, it is here where differences between Thw and Hld are mainly important.

The situation in the Figures 53e-h has naturally to be looked at with caution because it has been shown that the non-corrected data sets underestimate real precipitation. It is also worth to be noted that in these plots the correlation between the data points and the relationship of Pike [1964] is weaker than in the plots 53a-d. This relationship is representing here a number of other hydrological relationships. One can mention that the greater precipitation also better agrees with them.

In summarizing up what has been said on the previous pages, I selected the K dataset as best estimate of APPT for the multiple regression analyses that will be discussed in the following sections. When compared to Q, effects on the significance of the results related to the spatial interpolation of the data are minimized with this data set. For APE, I normally used Hld, but I tested always also Thw. When there is nothing mentioned in the text, Hld was used.

89

0.0

Tropical WetTropical Dry

Temperate WetTemperate Dry

Tundra & Taiga

APPT/ APE APPT/ APE

0.0 0.5 1.0 1.5 2.0 2.5

0.4

0.8

1.2

1.6

(g )

APPT = L & CAPE = Thw

0.0 0.5 1.0 1.5 2.0 2.5

(h)

APPT = L & CAPE = Hld

(f )

APPT = L & WAPE = Hld

0.4

0.8

1.2

1.6

(e)

APPT = L & WAPE = Thw

0.0

0.4

0.8

1.2

1.6 APPT = L & W, cAPE = Thw

(c ) (d)

APPT = L & W, cAPE = Hld

(b)

APPT = KAPE = Hld

0.0

0.4

0.8

1.2

1.6

(a)

APPT = KAPE = Thw

AAE/ APE

AAE/ APE

AAE/ APE

AAE/ APE

3.0

Fig. 53a-h Combinations of the different estimations for APPT and APE discussed in the text for the 60 river basins of Table 6. For further explanations, see text.

90

3.5. Determination of the Controlling Factors for Continental Runoff

3.5.1. Basin Averages

We have seen that a common point of many hydrological models is that AAE becomes closer to APE, as more as APPT exceeds APE. This means that RR is inversely correlated with the annual potential evapotranspiration rate (APETR), which is defined as the ratio of APE over APPT. Consequently, the relationship of Pike [1964] (equation 11) can also be written as:

RR = 1 - 1 / (1 + (APPT / APE) 2) 0.5 (18)

Figure 54 shows the corresponding plot of RR versus APPT over APE. Among the relationships presented in Figure 50, equation 18 (or 11) was found to fit best with the data in this study. For this reason I used the model of Pike [1964] as an estimate to quantify the average climatic control of RR in a river basin. The theoretical runoff ratio calculated according to this model is abbreviated in the following as RRPIKE. It is important to point out that 'climatic' is here exclusively understood as far as mean annual temperature and precipitation values are concerned. Seasonal variability or other factors such as frequently occurring storm events which are also elements of climate in the commonly used sense are not included. They can, nevertheless, also have an influence on the water budget in a river basin, as we will see below.

APPT / APE0.1 1.0 10.0 100.0

0.0

0.2

0.4

0.6

0.8

1.0

RR

Fig. 54 Runoff ratio (RR) as a function of APPT over APE according to Pike [1964].

Correlation coefficients between RR and a number of potential controlling parameters are shown

in Table 8. The coefficients were determined with the average river basin values. Calculations were made with all basins together, as well as by grouping the river basins with respect to their average climatic situation. Note that in nearly all cases RRPIKE depicts the greatest correlation coefficient. The degree of correlation, however, varies between the climate groups. At the same time, other factors such as basin slope (Slope) or the variability of precipitation distribution over the year (Four) show also quite high correlation, at least in certain climate groups.

91

Table 8. Correlation coefficients between RR and certain parameters, calculated on thebasis of the mean river basin values.

___________________________________________________________________________________________________

Climate No. of Basins APE APETR RRPIKE Slope a) Four a) ___________________________________________________________________________________________________

All 60 Hld -0.51 0.71 0.56 -- Thw -0.42 0.61 0.56 --

Tundra & Taiga 10 Hld -0.66 0.64 0.65 -- Thw -- -- 0.65 --

Temperate Dry 3 Hld -- -- -- -- Thw -- -- -- --

Temperate Wet 19 Hld -0.68 0.66 0.59 -- Thw -0.77 0.76 0.59 --

Temperate (wet & dry) 22 Hld -0.45 0.59 0.45 0.43 Thw -0.56 0.68 0.45 0.43

Tropical Dry 16 Hld -- -- 0.69 -- Thw -- -- 0.69 --

Tropical Wet 12 Hld -0.58 0.54 -- 0.68 Thw -0.53 0.52 -- 0.68

Tropical (wet & dry) 28 Hld -0.41 0.69 0.66 0.58 Thw -0.35 0.66 0.66 0.58

___________________________________________________________________________________________________

Only regressions which are significant with P < 0.1 are shown. a) Slope is the mean basin slope and Four is a modified form of the index of Fournier [1960] characterizing the distribution of precipitation over the year - see chapter II

However, the correlation coefficients in Table 8 can only give indications for the importance of certain controlling factors. The number of considered basins may be too small to perform significant statistics, as this is the case for the dry temperate climate group. Another important shortcoming is the fact that the basin averages are statistically not always very meaningful if the basins incorporate very contrasted basin parts. Table 9 illustrates that heterogeneity can be great in some basins, especially for the rivers of the dry climates. For example, in the Colorado or the Rio Negro basins, only less than 20% of the basin is classified to be wet, but more than 70% of total runoff comes from this area. I recall here from chapter II that for both rivers large basin parts are desert. It is therefore likely that in these cases the calculated basin averages are not very meaningful. The heterogeneity of the basins may also explain why in Table 8 there is no significant correlation between RR and the climatic parameters in the tropical dry climate, even if the river number is relative great.

3.5.2. Grid Point Averages

3.5.2.1. Non-Polar Climate Types

Table 10 corresponds Table 8, but this time calculations were made on the basis of the grid point characteristics, and not on the basis of the river basin averages. Due to the great number of values, all regressions are significant now. Again, RRPIKE depicts the greatest correlation coefficients. For all grid points together, the following equation describes the relationship best:

RR = 0.77 x RRPIKE - 0.01 (19)

n = 1752, r2 = 0.63, P < 0.0001 for RRPIKE, and P < 0.05 for the intercept

92

In equation 19, Hld was used for APE. Taking Thw instead, the regression coefficient rises up to a value of 0.99, but the correlation coefficient drops down to r2 = 0.52. Note that both in Table 10 and in Table 8 the use of Thw instead of Hld only increases the correlation between RR and RRPIKE in the temperate climate types (dry and wet). For all other climate types, Hld yields greater correlation coefficients.

Table 9. Comparison of the distribution of wet and dry basin parts in the river basins of Table 6 with the amount of total runoff originating from these basin parts.

______________________________________________________________________________________________________________ % of Basin % of Runoff % of Basin % of Runoff Av. Area (A) originating Av. Area (A) originating Climate that is from A that is Climate that is from A that is

River (a) ____________ ____________ River ____________ ____________ wet dry wet dry wet dry wet dry______________________________________________________________________________________________________________

Mackenzie 5 100 0 100 0 Magdalena 8 83 17 88 12Yukon 5 100 0 100 0 Don 7 81 19 92 8

St. Lawrence 7 100 0 100 0 Irrawaddy 8 79 21 93 7Waikato 7 100 0 100 0 Ob 5 77 23 97 3Rhine 7 100 0 100 0 Orinoco 8 69 31 92 8Ems 7 100 0 100 0 Columbia 7 69 31 96 4

Garonne 7 100 0 100 0 Liao He 7 69 31 91 9Tiber 7 100 0 100 0 Mississippi 7 68 33 96 4

Northern Dvina 5 100 0 100 0 Mekong 8 61 39 75 25Rioni 7 100 0 100 0 Paraná 8 51 49 81 19Loire 7 100 0 100 0 Ganges/Brahmaputra 8 48 52 81 19

Severn 7 100 0 100 0 Sao Francisco 6 34 66 64 36Susitna 5 100 0 100 0 Rufiji 6 32 68 49 51Yana 5 100 0 100 0 Gambia 6 29 71 53 47

Indagirka 5 100 0 100 0 Indus 6 26 74 72 28Hungho 8 100 0 100 0 Huanghe 4 24 76 42 58

Fly 8 100 0 100 0 Nile 6 19 81 52 48Purari 8 100 0 100 0 Niger 6 18 82 61 39

Kolyma 5 100 0 99 1 Zambesi 6 18 83 31 69Lena 5 100 0 99 1 Rio Negro (Argentine) 4 17 83 73 27

Yenisei 5 99 1 100 0 Colorado 4 17 83 79 21Si Kiang 8 97 3 98 2 Brazos 6 15 85 37 63Danube 7 94 6 100 0 Mahandi 6 10 90 11 89

Po 7 93 7 96 4 Orange 6 9 91 61 39Fraser 7 92 8 98 2 Godavari 6 9 91 17 83

Amazon 8 91 9 97 3 Limpopo 6 4 96 7 93Amur 7 88 12 99 1 Senegal 6 4 96 28 72Rhône 7 86 14 95 5 Murray 6 3 97 42 58

Changjiang 7 84 16 91 9 Tigris/Euphrates 6 2 98 11 89Zaire 8 83 17 91 9 Tana 6 0 100 0 100______________________________________________________________________________________________________________

(a) For the definition of the climate codes, see Table 7.

Because it is the combination of factors that is expected to determine the variation of RR at the global scale rather than one factor alone, I applied then multiple regression statistics to identify additional controlling parameters. Taking again all grid points together, the following parameter combination is found to be the best model for RR on the basis of all parameters considered in this study (see chapter II):

RR = 0.73 RRPIKE + 1.06 Slope + 0.78 x 10-3 Four - 0.82 x 10-4 Elev - 0.10 (20)

n = 1752, r2 = 0.79, P < 0.0001 for all parameters including the intercept

The units are radian for Slope, mm for Four, and m for mean elevation (Elev). RR increases with greater RRPIKE, greater Slope and greater Four, while Elev decreases RR. It is not difficult to explain that Slope is retained in the model. For a given grid element that is plain, runoff is mainly formed when the soils are at their field capacity. Additional water that is supplied by precipitation can no more

93

longer be taken up and the water runs off to rivers. When morphology is steep, however, water can move in the soils to the low elevation sites, where soils locally become saturated and flow over. Runoff can thus occur even if the average field capacity for the whole grid element is not achieved. The appearance of Four in equation 21 can be explained on the same background. When precipitation is uniformly distributed over the year (i.e., Four is smallest for a given APPT), there may be more water available for evapotranspiration in the soils on long-term. Frequent events of high precipitation (e.g., seasonal rain periods or periods with strong thunder-showers) can increase the possibility that soils achieve field capacity on short-term, and a greater amount of water may be lost by runoff compared to a uniform precipitation distribution. Finally it is also interesting that Elev is retained in the regression model. This probably directly reflects the effect of atmospheric pressure on evapotranspiration. Given the same temperature, evapotranspiration is greater at low pressure than at high pressure, and atmospheric pressure decreases with increasing elevation. Both Hld and Thw are exclusively based upon temperature data and do not include this effect.

Table 10. Correlation coefficients between RR and certain parameters, calculated on thebasis of the mean grid point values.

___________________________________________________________________________________________________

Climate No. of Gridpts. APE APETR RRPIKE Slope Four ___________________________________________________________________________________________________

All 1752 Hld -0.64 0.80 0.33 0.20 Thw -0.57 0.72 0.33 0.20

Tundra & Taiga 639 Hld -0.64 0.80 0.34 0.22 Thw -0.55 0.65 0.34 0.22

Temperate Dry 267 Hld -0.33 0.46 0.58 0.38 Thw -0.32 0.51 0.58 0.38

Temperate Wet 232 Hld -0.45 0.60 0.54 0.53 Thw -0.51 0.66 0.54 0.53

Temperate (wet & dry) 499 Hld -0.56 0.76 0.38 0.65 Thw -0.54 0.79 0.38 0.65

Tropical Dry 304 Hld -0.49 0.67 0.42 0.47 Thw -0.43 0.56 0.42 0.47

Tropical Wet 310 Hld -0.71 0.73 0.38 0.64 Thw -0.55 0.59 0.38 0.64

Tropical (wet & dry) 614 Hld -0.65 0.85 0.33 0.73 Thw -0.61 0.79 0.33 0.73

___________________________________________________________________________________________________

All regressions are significant with P < 0.1.

Table 10 shows that when the multiple regression statistics are repeated within the different

climate groups, it is always the combination of RRPIKE, Slope, Four, and Elev that yields the most significant regression model for RR (only in the tundra and taiga climate, Four is not found to be significant, which is discussed below). The corresponding equations are given in Table 11. This is a strong confirmation that among all investigated parameters, these four parameters are the most important controlling factors for RR at the global scale. The use of Hld for APE generally results in greater correlation coefficients than the use of Thw. This is, however, mainly the case for the tundra and taiga climate, while in the other climate types differences are only small. Note also in this context that for the tundra and taiga climate type the use of Thw leads to a positive intercept (equation 21.1). Using Hld, the intercept is negative, as this is the case in all other regression equations. This supports to some extent the above made assumption that real APE may lie here somewhere in between Hld and Thw. But the greater correlation coefficients with Hld suggest that Hld gives more realistic estimates in this particular climate.

94

Table 11. Multiple correlation models for RR in different climate types. All models have the form: RR = a RRPIKE + b Slope + c Four + d Elev +e.

_____________________________________________________________________________________________________________

Climate No. of Correlation Regression Coefficient Equation Gridpts. APE Coefficient ______________________________________________ Number a b c d e (r2) RRPIKE Slope Four Elev Intercept (x 10-3) (x 10-4) _____________________________________________________________________________________________________________

All 1752 Hld 0.79 0.73 1.06 0.78 -0.82 -0.10 (21) Thw 0.63 0.92 1.28 0.13 -0.85 -- (21)

Tundra & Taiga 639 Hld 0.73 0.93 1.12 --

-0.73 -0.20 (21.1) Thw 0.57 0.75 1.43 -- -0.97 0.12 (21.1)

Temperate Dry 267 Hld 0.58 0.32 1.14 1.28 -0.51 -0.08 (21.2) Thw 0.58 0.47 1.09 0.77 -0.41 -0.06 (21.2)

Temperate Wet 232 Hld 0.68 0.69 1.10 1.32 -0.86 -0.16 (21.3) Thw 0.67 0.76 1.11 0.68 -0.58 -0.08 (21.3)

Temperate (wet & dry) 499 Hld 0.78 0.57 1.10 1.26 -0.65 -0.11 (21.4) Thw 0.78 0.76 1.06 0.71 -0.47 -0.09 (21.4)

Tropical Dry 304 Hld 0.61 0.57 1.08 0.53 -1.06 -0.03 (21.5) Thw 0.58 0.41 1.22 0.57 -1.35 -- (21.5)

Tropical Wet 310 Hld 0.70 0.72 0.82 0.78 -1.31 -0.05 (21.6) Thw 0.69 0.62 1.12 0.90 -1.86 -0.04 (21.6)

Tropical (wet & dry) 614 Hld 0.81 0.72 0.91 0.69 -1.18 -0.05 (21.7) Thw 0.80 0.61 1.13 0.84 -1.61 -0.04 (21.7)

_____________________________________________________________________________________________________________

All parameters including the intercept are significant with P < 0.1. Significance for the parameters is normally much greater (P < 0.0001), and smaller significance levels are only observed for the intercept.

The regression coefficients for each parameter vary to some extent between the climate groups. This can be related to additional particularities of the climate types that are not expressed by the parameters used in this study. Consequently, the regression coefficients for Four are generally more variable than for the other parameters, at least as far as the regressions made with Hld for APE are concerned. Four is one of the rare parameters considered in this study that reflect seasonal variability. Moreover, one has to keep in mind that a great Four value may only significantly enhance runoff when the soils are in an intermediate position between strong water limitation and water saturation. Even extreme precipitation may not be sufficient to bring the soils to field capacity when they are very dry. On the other hand, it is also evident that Four should be of small importance if the soils are mostly over the year at field capacity. In this sense it is interesting to note in Table 11 that Four has especially great regression coefficients in the temperate climate types (dry and wet), indicating that in these climate types the soils correspond closest to the above mentioned condition. This is interesting because we will see in chapter IV that also when investigating the major controls for sediment fluxes, it is found that Four seems to be of greater importance in the temperate climate than in the other climates. Finally, one has also to mention here that Four is not found to be significant for the tundra and taiga climate type. This should be related to the fact that for the high-latitude rivers snow melt is a major contributor to river runoff (Kuhl and Miller [1992]). There can exist a considerable time lag between precipitation and the formation of runoff when the water is temporarily stored as snow (or as frozen soil water). The distribution of precipitation over the year is therefore less meaningful in this climate. Note that for the Lena River, for example, there occurs nearly no runoff during half of the year (Dümenil et al. [1993]) because most of the water in the basin is frozen (in this context see also the description of the Mackenzie River case study in chapter I).

95

0.0 0.2 0.4 0.6 0.8

0.0

0.2

0.4

0.6

0.8

Tropical Wet

Temperate Wet

Tundra & Taiga

Tropical Dry

Temperate Dry

y = x

Fig. 55a Comparison of themean runoff ratio predicted inthis study (y-axis) with themean runoff ratio resultingfrom the UNESCO runoff andprecipitation maps (x-axis) forthe 60 river basins of Table 6.

Fig. 55b Comparison oftotal runoff (km3/yr) predicted in this studywith total runoffextracted from theUNESCO runoff map forthe 60 river basins ofTable 6. See Table 6 for abbreviations of the rivernames.

1 5 0 0 1 0 0 0 5 0 0 0 5 0 0 1 0 0 0 1 5 0 0

ZaiOri

MekSKiHunPurYts

SLaDanFra

RhoGarDonRioTib

EmsNigNil

GodZamSenRufLimBrz

LenMacKolInk

SusHuaCdo

GaBraParIrrMagFly

MisAmoColRhiPoLoiLHeWaiSev

IndSFrMahTiEuGamMurOrgTanIenObYukNDiYan

RNeTemperate Dry

Amz (x 5)

Tropical Dry

Tundra & Taiga

Temperate Wet

Tropical Wet

This study UNESCO - Map

It is also worth to be mentioned that the regression coefficients for RRPIKE in Table 11 are smaller in the dry climate types (temperate dry and tropical dry) than in the wet climate types. Probably the relationship of Pike [1964] especially overestimates RR in these climates. On the other

96

hand, correlation coefficients are also lower here. Maybe other factors which were not considered in this study are important for the control of RR in the dry climate types (mainly related to seasonal variability), but it is also possible that this effect simply reflects a lack of data. In dry climates, observations are naturally more sparse, and probably sometimes also less reliable because of the great variability of the runoff and precipitation patterns. It can be further noted from Table 11 that Slope has remarkably constant regression coefficients in nearly all climate types. This can be expected because the geomorphological parameters do not underlie any seasonal variability. It can also indicate that the relationship between Slope and RR is closest to a true linear relationship. Not necessarily all variables retained by linear multiple regression models are also always coupled in a true linear way to the independent variable. Contrary to Slope, regression coefficients for Elev are quite variable. It is remarkable that they are generally greater in the tropical climate types than in the colder climate types. It is therefore also possible that temperature interferes in this relationship.

The only exception for the relative constant regression coefficients for Slope is the tropical wet climate, where the coefficient is considerably lower. One may speculate that this could be related to vegetation, since biomass density is great in the wet tropics. This exception is, however, only prominent as far as the Hld estimate for APE is used in the regressions. All trends that emerge from the comparison of the regression coefficients in Table 11 are only valid as long as estimates for APE are close to reality. To verify these trends, more efforts are needed to determine APE globally.

Nevertheless, even if it is not clear whether the variability of the regression coefficients really corresponds to the above attributed processes, one has to point out that splitting up the regression models with respect to the different climate types slightly increases the correlation between observed RR and predicted RR. I quantified this only for the 60 river basins of Table 6 because for these basins the UNESCO runoff map was validated against literature estimates. All calculations were made this time in a 0.5° x 0.5° longitude/latitude grid point resolution (I recall here that the regressions were calculated with grid points in a 2.5° x 2.5° longitude/latitude resolution). Mean basin RR was determined as the mean of all grid point RRs, and not as mean basin Q over mean basin APPT (which gives naturally not exactly the same values). RRs greater than 1 or smaller than 0 were set to 1 or 0, respectively. The correlation coefficient between the predicted RRs and the RRs resulting from the UNESCO precipitation and runoff maps is r2 = 0.82 when only equation 21 is used. Using the combination of equation 21.1, 21.4, and 21.7 (depending on the climate types to which the grid points belong) increases the correlation coefficient to a value of r2 = 0.85 (Fig. 55a). Note that grid points which fall into the desert climate type were calculated in this case with equation 21.7. The correlation coefficient for the resulting specific runoff values (according to Q = RR x APPT) is r2 = 0.97. The models allow thus a precise prediction of runoff for most of the 60 river basins (Fig. 55b). Applying the combination of the equations 21.1, 21.2, 21.3, 21.5, and 21.6 does not increase the correlation coefficient much. It remains still at r2 = 0.85. It is therefore not clear whether the additional splitting up into dry and the wet climate types really may be an improvement for the modelling.

3.5.2.2. Polar Climate Types

Prediction of continental runoff at the global scale requires also the inclusion of the polar climate types. The two climate types distinguished in this study (polar with ice cover, and polar without ice cover) are nearly absent in the 60 river basins of Table 6. Only for the Kolyma and the Lena rivers very small percentages of the basin are falling into the ice-free polar climate type (see chapter II). Observations in these regions are naturally sparse, and the assignments of Q and APPT in the maps of Korzoun et al. [1977] may be considered to be quite uncertain.

For the ice-covered polar climate type, no other factors than APPT are found to control Q on the basis of the considered parameters in this study. The following equation describes the relationship:

Q = 0.91 x APPT - 5.0 (21.8)

97

n= 1086, r2 = 0.88, P < 0.0001 for APPT, and P < 0.006 for the intercept

Both the units of Q and APPT are mm. Equation 21.8 means that RR has more or less a constant value of about 0.9 in the ice-covered regions. AT is slightly negatively correlated with Q, but this effect is only significant at the ice sheet margins, where Q is greatest. It is difficult to compare ice-sheet runoff with river runoff because ice migrates sometimes over long distances to the margins, and the water often enters the oceans in a solid form by ice-sheet calving. Probably not at least also for this reason it is not possible in this study to discern the role of temperature for runoff is the ice-covered regions.

For the ice-free polar climate type, the following regression model is found to predict RR best:

RR = 0.43 RRPIKE + 0.87 Slope - 0.94 x 10-4 Elev + 0.15 (21.9)

n = 166, r2 = 0.45, P < 0.0001 for all parameters, and P < 0.04 for the intercept

The correlation coefficient is naturally smaller in this climate type than in the other regression models. Note that the number of grid points is also small. As in equation 21.1 (tundra and taiga climate), Four is not found to be significant. It is also interesting to note that the intercept is positive this time, contrary to the regression models in the other climate types (equations 21.1 to 21.7). One may argue that equation 21.9 is thus in an intermediate position between equation 21.1 and 21.8.

3.5.3. Extrapolation to the Global Scale

One may finally apply the above discussed empirical regression models to the total continental area in order to predict runoff at the global scale. I selected here, and in the following of this study (in chapter VI these models will be used again), equations 21.1 (tundra and taiga climate), 21.4 (temperate climates), 21.7 (tropical climates, deserts), 21.8 (polar climate without ice), and 21.9 (polar climate with ice) to be the best regressions models, depending upon the climate type to which the corresponding grid points belong. When calculated in a 0.5° x 0.5° longitude/latitude grid point resolution, the models result in a value of about 46850 km3/yr for total exoreic continental runoff, which is close to the 44370 km3/yr determined from the corrected UNESCO runoff map (Table 7). Also for the holospheric distribution, the empirical approach is quite close to the map values. This can be seen in Figure 56.

It has to be pointed out that the empirically determined global runoff value lies between the corrected UNESCO runoff map value and the uncorrected one (which is close to 50000 km3/yr), but the statistics were done with the uncorrected grid point Q values (the correction method is mainly useful for larger scales; applying it also to more local scales does not necessarily improve the reliability of the values). One has to keep in mind that for the modelling, the precision of the runoff values is strongly depending on the precision of the precipitation data. The correction of the UNESCO runoff map values was mainly set up because it can be supposed that an insufficient number of gauging stations leads generally to an overestimation of the values. The UNESCO precipitation map includes about six times more gauging stations than the UNESCO runoff map (Korzoun et al. [1977]). Hence overestimation of the precipitation values due to spatial interpolation should be smaller. The other reason why I applied a correction of the UNESCO runoff map is due to the evaporation of water when rivers pass through dry regions (this correction was only applied to the tropical dry climate ). It is evident that the empirical method can not include this effect. Note in Figure 56 that the empirical method yields mainly too great values in the tropical latitudes. Including here the evaporation effect would bring the empirical method closer to the corrected runoff values.

98

-80 -60 -40 -20 0 20 40 60 80

0

1000

2000

3000

4000

5000

6000

Empirical (this study)

UNESCO RunoffMap (corrected)

Latitude N

Fig. 56 Compa-rison of theholospheric run-off distribution(km3/yr) betweenthe UNESCOrunoff map(corrected) andthe empiricalapproach devel-loped in thisstudy.

3.6. Comparison with other Models

3.6.1. Water Budget Models (Bucket Models)

The most prevalent parametrisations to determine large scale hydrology are so-called bucket algorithms. These parametrisations highly simplify the hydrologic processes of infiltration, evaporation, and runoff formation. For a given grid element, evaporation is exclusively calculated as a function of potential evapotranspiration and of soil moisture. When soils are at field capacity, evapotranspiration occurs at potential rates, and additional water that is supplied by precipitation runs off to rivers. When soils are water limited, actual evapotranspiration is deduced from potential evapotranspiration by a simple evapotranspiration function, which multiplies potential evapotranspiration with a factor ranging between 1 and 0. This factor normally decreases in an exponential way as more as water limitation of the soils increases. Because of parameter limitations for global scale investigations, many approaches based upon bucket models work with gridded data sets of mean monthly precipitation and temperature only (as used in this study). They often rely to the formulae of Thornthwaite (equations 12 to 15) to determine potential evapotranspiration (e.g., Willmott et al. [1985], Vorösmarty et al. [1989], Mintz and Serafini [1992]). Changes in the average soil moisture over the year is then derived by an iterative procedure: for each grid element, an arbitrary field capacity is assigned (often a general value of 150 mm is used) and the model parametrisations are repeatedly applied to the corresponding precipitation and soil moisture values until a steady state is achieved for successive twelve-month periods. Precipitation is first used to fill up the soils, and then contributes to runoff. Some models include also the seasonal water storage in snow as additional compartment to soil water storage (e.g., Willmott et al. [1985], Vorösmarty et al. [1989]).

For comparison, I present in Figure 57 the mean annual runoff resulting of the water budget model of Willmott et al. [1985] together with the runoff extracted from the UNESCO runoff map for the 60 river basins of Table 6. This water budget model is available at the National Center for Atmospheric Research (Boulder, Colorado) in a spatial resolution of 1° x 1° longitude/latitude. I linearly interpolated it to a 0.5° x 0.5° longitude/latitude resolution. Note that this model uses a

99

precipitation data set that was not corrected for rainguage biases (the APPT and APE data correspond in fact to the situation shown in Figure 53e).

1 5 0 0 1 0 0 0 5 0 0 0 5 0 0 1 0 0 0 1 5 0 0

ZaiOri

MekSKiHunPurYts

SLaDanFra

RhoGarDonRioTib

EmsNigNil

GodZamSenRufLimBrz

LenMacKolInk

SusHuaCdo

GaBraParIrrMagFly

MisAmoColRhiPoLoiLHeWaiSev

IndSFrMahTiEuGamMurOrgTanIenObYukNDiYan

RNe

Tundra & Taiga

Amz (x 5)

Q - UNESCO - MapQ - Willmott et al. [1985]

Temperate Dry

Tropical Dry

Temperate Wet

Tropical Wet

Fig. 57 Comparison oftotal runoff (km3/yr) determined with the water budget model of Willmott et al. [1985]with total runoff extracted from the UNESCO runoff map for the 60 river basins of Table 6. See Table 6 for abbreviations of the river names.

It is remarkable that the water budget model of Willmott et al. [1985] nearly predicts no runoff for the tundra and taiga climate rivers. This is not very surprising since we have seen that in this climate taking Thw for APE gives too great estimates for potential evapotranspiration. This should also result in an overestimation of AAE. The tundra and taiga climate type can therefore be considered to be somewhat outstanding here. But also in the other climate types the model tends to underestimate runoff. Taking all rivers together, the model runoff is about 28 % lower than the UNESCO map runoff (for precipitation, the difference between the two data sets is only 8%). One may speculate that the underestimation of Q is related to the fact that bucket models generally ignore morphology. This study clearly shows that morphology influences the formation of runoff. It is interesting to note that obviously the model fits best in the tropical climate types. In the temperate climates, underestimation is more important. This may be difficult to explain, but one has to keep in mind that the principal tool to tune bucket models in order to agree with observations is the modification of the evapotranspiration function. Since about half of global runoff originates from the tropical wet climate, this function has especially to be tuned to agree in this climate if the model output should not considerably fail with respect to the global runoff value.

In any case, the comparison of the values in Figure 57 reveals that the water balance model of Willmott et al. [1985] is not sufficiently precise to predict runoff at the river basin scale on the continents.

100

3.6.2. General Circulation Models

Bucket models are also commonly used in General Circulation Models (GCMs) to represent land-surface hydrology. Potential evapotranspiration is determined in a more sophisticated way relying on several parameters that were calculated by the models, and soil water budgets are established on much shorter time steps (daily or hourly). Sometimes, additional modifications were done to improve runoff prediction. For example, in the GCM maintained at the NASA/ Goddard Institute for Space Studies (GISS), an improved land-surface parametrisation has been developed (Abramopoulos et al. [1988]), as well as a river routing scheme that simulates the water return flow to the oceans (Miller et al. [1994] - see also chapter VI). The model physically determines soil water movement according to soil hydraulic properties and calculates infiltration and both surface and underground runoff. Its ability to describe global runoff is discussed by Russell and Miller [1990], Kuhl and Miller [1992], Miller et al. [1994], Marengo et al. [1994], and Van Blarcum et al. [1995].

One of the advantages of GCMs is that they can produce global climatologies for scenarios different from present day conditions. The reliability of such predictions is depending on the extent to which GCMs are able to reproduce present day climate. In the following, I briefly try to examine this question with respect to continental runoff. It is naturally not possible in this study to look at the variety of GCMs which are actually existing, and I selected the ECHAM2 model to be exemplary discussed here. This model is taken in chapter VI to derive global climatologies for the last glacial maximum, which are then used to model the river carbon fluxes at that time.

Latitude N

-80 -60 -40 -20 0 20 40 60 80

U N ES C O

E C H A M 2

U N E S C O

E C H A M 2

Latitude N

-80 -60 -40 -20 0 20 40 60 80

0

20

40

60

80

100

120

00

00

00

00

00

00APPT

(km3/yr)

AAE

(km3/yr)

Fig. 58a Holospheric distribution of precipitation and evapotranspiration resulting from the UNESCO precipitation and runoff maps and of the ECHAM2 outputs. Calculations were done in the ECHAM2 grid point resolution (5.625° x 5.625° longitude/latitude).

The ECHAM2 general circulation model is the T21 version (spatial resolution of 5.625° x 5.625° longitude/latitude) of the weather forecast model developed at the European Centre for Medium-Range Weather Forecasts (ECMWF), modified (in HAMburg) for climate simulations and coupled to an ocean general circulation model. Model outputs are available from the Deutsches Klima Rechenzentrum in Hamburg via ftp. The model is based upon the conservation of mass, momentum, and energy. The conservation principles result in a set of primitive equations determining vorticity, divergence, temperature, and humidity, as the most basic prognostic model variables. Other atmospheric quantities like wind, rain or snow are calculated diagnostically from the prognostic variables. The ECHAM2 model is capable of simulating the present climate, although not perfectly. The meridional temperature gradient is underestimated by 2-3 °C, and the resulting winds are on average 20% too weak. The general structure of global wind patterns, however, is simulated quite

101

realistically. Temperature discrepancies are mainly abundant in the Southern Hemisphere, and Northern Hemisphere temperatures generally agree well with observed temperature fields (especially on land). This temperature feature is characteristic for many GCMs (Willmott and Legates [1993]). For more details on the ECHAM2 model, see, for example, Lautenschlager and Herterich [1990] and Lautenschlager [1991].

UNESCO - Map

CGM (ECHAM2)

Fig. 58b Comparison of the spatial runoff distribution (km3/yr) between the UNESCO runoff map (upper figure) and the model output of the ECHAM2 General Circulation Model (lower figure). The UNESCO runoff was averaged on the basis of the ECHAM2 grid point resolution (5.625° x 5.625° longitude/latitude).

Figure 58a compares the holospheric precipitation and evapotranspiration distributions extracted from the UNESCO runoff and precipitation maps with those of the ECHAM2 model. The comparison of runoff distribution is shown in Figure 58b. Taking all grid elements together, ECHAM2 predicts an annual precipitation total that is about 8% lower than the precipitation total of the UNESCO precipitation map. Because the model evapotranspiration is about 19% lower, the resulting ECHAM2 runoff is 10% greater than the value from the UNESCO runoff map. In the model continental runoff is

102

counterbalanced by evaporation over the oceans with a slight excess of runoff (about 4% of total Q). This indicates that ECHAM2 has more or less achieved a steady state with respect to the global water cycle, which is not always the case for other available GCM outputs (Munhoven and Probst [1995]). Mean annual temperature on the continents is calculated by the model to be 10.13 °C. The corresponding value resulting from the Legates and Willmott [1992] temperature data set is 10.08 °C.

The ability of GCMs to predict global runoff patterns cannot be better than the ability to predict precipitation patterns. Compared to observations, none of the presently existing GCMs is perfect. The extent of the precipitation deviation of ECHAM2 that emerges from Figure 58a seems to be in the same range than it is observed for other GCMs, even if each GCM naturally has its own particular features (for a comparison of GCM precipitation fields with observed precipitation fields, see Hulme [1991]). In the case of the ECHAM2 model, this has naturally consequences for the spatial runoff distribution predicted by the model (Figure 58b). For example, the too great runoff found with the ECHAM2 model in the high northern latitudes is certainly related to the too great precipitation calculated by the model in these latitudes. One has to state, however, that also the determination of reliable evapotranspiration fields seems to be problematic. Note that numerous model grid points depict negative runoff values, which does not make any physical sense. Globally, more than 10% of the water evaporating in the model is not supplied by precipitation. The occurrence of negative runoff fields is a common feature in many GCMs, with sometimes much greater amplitudes (Munhoven and Probst [1995]). Not at least for this reason, General Circulation Models are still far away to predict local runoff patterns on Earth with sufficient precision.

3.7. Conclusions

We have seen that the runoff and precipitation maps created by Korzoun et al. [1977] are realistic representations of the present-day water budget on Earth. From their study, global continental runoff can be estimated to be about 44400 km3/yr, which is slightly lower than the value the authors give in their original publication. The discrepancy can mainly be attributed to the spatial interpolation of the map values, which obviously tend to overestimate global and regional estimates, as well as to the loss of water directly evaporating from rivers. This global estimate is, however, still more than 10% greater than the often cited estimate of Baumgartner and Reichel [1975]. It seems that the latter study especially underestimated runoff in the northern latitudes.

Also the precipitation map of Korzoun et al. [1977] yields with an estimated global precipitation total of about 124600 km3/yr a greater value than it results from many of the actually available precipitation data sets. This is because the authors corrected the values for biases related to rainguage undercatches. Their proceeding is indirectly confirmed from the runoff point of view: without correction, precipitation cannot account everywhere for the water running off the continents. Compared, for example, to the precipitation data set of Leemans and Cramer [1992] (which is one of the frequently used precipitation data sets), the mean global precipitation value according to the UNESCO precipitation map is about 10% greater. Again, major discrepancies emerge in the northern latitudes.

Precipitation and potential evapotranspiration are the principal controlling factors for continental runoff at the global scale. The data used in this study confirm a common feature of many hydrologic studies, that is that runoff ratio generally increases as more as precipitation exceeds potential evapotranspiration. It is important to point out, however, that morphology also represents a non-negligible controlling factor for runoff: for similar mean annual climatic conditions, greater slope increases runoff ratio, while greater elevation decreases it. The effect of slope can be related to the fact that a uneven morphology locally favours water saturation of the soils. The effect of elevation may directly have an influence on potential evapotranspiration because atmospheric pressure decreases with increasing elevation, which should enhance evapotranspiration. Hence, elevation is probably

103

retained as potential controlling factor to compensate the fact that potential evapotranspiration was calculated in this study only as a function of temperature. Finally, it is also the seasonal variability of precipitation that influences the runoff ratio. Periods of strong precipitation rise the possibility that soils achieve saturation on short-term, and more water can be lost to runoff compared to a more uniform precipitation distribution.

One of the shortcomings in this study is that potential evapotranspiration is uniquely determined from temperature data. Due to parameter limitations, this is a common practice for global scale investigations, but it may not always yield values which are very close to reality. It is interesting in this context that the widely applied formulae of Thornthwaite [1948] do not necessarily give the best estimates for APE over all climate types. They seem to fit well in the temperate climate types, but they considerably deviate in the tundra and taiga climate type. The simple formula of Holdridge [1959] seem to fit as least as good as the Thornthwaite's formulae for most of the climate types, and even better in the tundra and taiga climate type. This is important because it confirms the applicability of the climatic classification based upon the Holdridge life zone scheme (Holdridge [1947]) that is throughout applied in this study.

Further studies should therefore focus on a refinement of the determination of potential evapotranspiration at the global scale. Another improvement for the modelling of continental runoff could be the availability of global precipitation climatologies with higher temporal resolutions, supplying, for example, weekly or daily values. This would allow to better discern the role of strong precipitation events on runoff formation. Last but not least one should also think about how to introduce morphology in hydrologic models. It could be, for example, an interesting task to test whether the introduction of morphological factors such as grid point slope in simple water bucket models could improve the runoff prediction ability of these models.

One may finally also mention here that it is naturally not only the modelling that should receive attention in further research activities. It has been shown that the evaluation of time series of river runoff data is a useful tool to understand the hydroclimatic fluctuations which occurred on Earth during the last decades of this century (e.g., Probst and Tardy [1987], [1989], Tardy et al. [1994], [1995], Kayser et al. [1990]). The knowledge of these fluctuations and their implications for the climatic evolution in certain regions of the world is important in order to better understand to which extent our climate actually changes due to human impacts, as well as to assess the possible consequences of these changes. Nowadays, an increased number of river data should be available compared to the time when Korzoun et al. [1977] or Baumgartner and Reichel [1975] created their global maps. Further studies may use these river data to compute regional and global maps showing also the temporal variability of continental runoff during this century.

105

CHAPTER IV

HOW TO PREDICT RIVER SEDIMENT DISCHARGES TO THE OCEANS

4.1. Introduction

The mechanical erosion of rocks and the subsequent transport of the eroded materials to the oceans by rivers is one of the principal landscape forming processes on Earth. River sediment fluxes play an important role in various natural geochemical cycles such as the global carbon cycle. Also the transport and cycling of human released contaminants is often strongly coupled to the transport of sediment in rivers, since many contaminants have some degree of associations with the particulate phase in water (Walling and Webb [1985], Leenheer [1991]). Beside their role in natural cycles, scientists mainly became interested in studying sediment fluxes because of their often drastic changes to human impacts in natural ecosystems. Modern changes in drainage basin characteristics, e.g. by river damming, deforestation, or extensive cultivation, may significantly alter natural mechanical erosion rates and river sediment fluxes. Such perturbations can have severe consequences for agriculture, mainly through the loss of fertile soils. But also in the estuaries and coastal zones, where sediments end up, an alteration of the natural river sediment supply can provoke considerable changes of the coastal metabolism, the biogeochemical cycling, or the coastline morphology.

In order to assess the response of river sediment fluxes to regional and global change, an understanding of the main factors that control the fluvial sediment transport to the oceans is needed. Several attempts have been made to analyse, for example, the relationships between mechanical erosion rates and precipitation (Langbein and Schumm [1958], Fournier [1960], Douglas [1967], Wilson [1973], Ohmori [1983]). Other authors considered climate as the dominant controlling factor (Jansson [1982], [1988]), while again others pointed out the great influence of basin elevation and morphology on sediment fluxes (Schumm and Hadley [1961], Ahnert [1970], Pinet and Souriau [1988], Milliman and Syvitski [1992], Summerfield and Hulton [1994]), or they proposed multiple regression models combining various parameters to predict sediment fluxes world-wide (Jansen and Painter [1974], Probst [1992]).

The purpose of this chapter is to determine the main factors and relationships that control river sediment fluxes at the global scale. Many of the previous investigations suffered from strong data limitations, especially with respect to environmental parameters. The large number of global environmental data sets that are used in this study allow not only a very detailed description of the hydroclimatic, biological, and geomorphological characteristics on Earth, but they allow also an easy extrapolation of found relationships over the continents. This is important because one can test whether these relationships agree with the observed variability of river sediment fluxes both at regional and global scales.

Such a proceeding is more problematic for sediment fluxes than, for example, for the fluxes of water as applied in the previous chapter. An extrapolation based upon grid point characteristics assumes that the fluxes which are observed at the river mouths are the sum of the fluxes taking place on the grid point level. The overall basin form, its size, and other general basin characteristics such as

106

the existence of lakes or of floodplains cannot be considered here. It is a well known fact, however, that these factors may affect river sediment yields (sediment flux divided by drainage basin area), mainly by determining the capacity of drainage basins to retain the eroded materials by sedimentation. It is therefore also tested in this chapter whether it is possible to identify the influence of such general basin characteristics on river sediment yields, and, if this is the case, whether it is possible to include such factors into global scale extrapolations.


4.2.1. Sediment Fluxes

Table 12 lists the estimated sediment yields (FTSS) for the 60 river basins that are investigated in this study. With very few exceptions, the values were taken from the recently published Global River Index (GLORI) database (Milliman et al. [1995]). All values were normalised to the runoff and basin size values used in this study. This means that I calculated the mean annual sediment concentrations by taking the sediment fluxes and the runoff values given in the GLORI database,

Table 12. Sediment yields for 60 world rivers (FTSS). The unit is t km-2 yr-1. ______________________________________________________________________________________________________________

River Basin Abrevi- FTSS River Basin Abrevi- FTSS River Basin Abrevi- FTSS ation ation ation ______________________________________________________________________________________________________________

Amazon Amz 203 Yukon Yuk 71 Liao He LHe 218 Zaire Zai 13 Huanghe Hua 1338 Rufiji Ruf 95

Mississippi Mis 123 Danube Dan 107 Rio Negro (Argentine) RNe 72 Ob Ob 5 Orange Org 124 Hungho Hun 815

Paraná Par 32 Colorado Cdo 190 Rhine Rhi 16 Yenisei Ien 5 Columbia Col 17 Brazos Brz 122

Lena Len 8 Kolyma Kol 26 Loire Loi 71 Amur Amo 29 Sao Francisco SFr 10 Rhône Rho 580 Nile Nil 64 Si Kiang SKi 149 Tana Tan 1000

Changjiang Yts 264 Irrawaddy Irr 620 Garonne Gar 77 Ganges/Brahmaputra GaBra 640 Don Don 16 Po Po 277

Mackenzie Mac 51 Senegal Sen 65 Gambia Gam 5 Niger Nig 26 Indagirka Ink 35 Fly Fly 1377

Zambesi Zam 34 Limpopo Lim 96 Susitna Sus 193 Murray Mur 25 North Dvina NDi 12 Purari Pur 2514

St. Lawrence SLa 4 Godavari God 548 Tiber Tib 472 Orinoco Ori 205 Magdalena Mag 773 Rioni Rio 572

Tigris/Euphrates TiEu 255 Fraser Fra 72 Severn Sev 82 Indus Ind 274 Yana Yan 14 Waikato Wai 132

Mekong Mek 185 Mahandi Mah 451 Ems Ems 27 ______________________________________________________________________________________________________________ All data were taken from Milliman et al. [1995], except the data for Ems and Tan, which come from Milliman and Syvitski [1992]. Sediment yields are normalised to the drainage intensities and drainage basin areas used in this study (see text). Note that for most of the rivers, bedload transport has not been measured. On average, bedload transport may account for an additional flux of about 10% of the suspended matter flux (Milliman and Meade [1983]).

multiplied then these concentrations with the runoff literature estimates used in chapter III, and finally divided the resulting fluxes with the drainage basin areas given in chapter II. Such a proceeding is needed to avoid confusion with respect to the other values cited in this study. In by far most of the cases, this does not change much the final values compared to the original values in the GLORI database. Note that always the FTSS values were selected representing natural averages. For some rivers, today's sediment fluxes can be considerably lower because of damming of these rivers. A striking example for this is the Nile River, which discharges nearly no more sediments into the Mediterranean Sea after the construction of the Aswan Dam.

107

The values in Table 12 are best estimates. The numerous problems related to the determination of average sediment yields for river basins must be kept in mind when considering such a data compilation. Because of the temporal sediment storage capacity in river basins, a regular sampling for especially long time periods is needed to get reliable average values, and a high frequency of sampling is particularly important during flood periods. Not all of the given values have been measured under these optimal conditions. Meade and Parker [1985] reported from the investigation of rivers of the United States that in some cases, more than half of the sediment load for the year is likely to be transported in only 5 or 10 days within any individual year. Also Probst and Bazerbachi [1986] found for the Garonne River in France that 50% of the sediment flux measured during a seasonal cycle was discharged within 17 days, with again about half of it within one single day. A similar behaviour can be found for the comparison of year-to-year values. Again for the Garonne River, Semhi [1996] reported that the mean annual sediment yield decreased with about a factor of two during a dry period compared to a humid period. Moreover, there is evidence from the geological record that during catastrophic events, one century flood may wash out more sediments from a basin than all other years of the century together (Milliman et al. [1996]). Especially in arid and semi-arid regions, where water discharge varies markedly both between years and within years, it is of great importance to have long and dense records. The flashiness of water discharges often observed in these regions makes it difficult to obtain observations at the gauging station at rising stages. Therefore data from semi-arid areas might be less reliable than data from other climates, unless data are automatically recorded (Jansson [1988]).

4.2.2. Other Data and Statistics

All other data sets that were used to determine the river basin characteristics are described in chapter II, as well as the applied statistical procedures and other technics. The runoff data set is discussed in chapter III. Note that with regard to climatic distinctions, I consider in the following all continental grid points as 'dry' that fall either into the tropical dry or into the temperate dry climate type, or that are desert. 'Wet' grid points are consequently grid points falling into the other climate types.

4.3. Previous Models for Sediment Yield Prediction

In the following, a review is made of the different approaches that were proposed to predict river sediment yields. All examples giving mechanical denudation rates estimated them from river sediment fluxes, so that the two terms can be considered as synonyms here, which is, of course, not the case in general. In some cases, I applied the proposed relationships to the data sets used in this study, and the resulting amounts of total sediments that should be discharged to the oceans according to the relationships were calculated. This was done in order to get an-order-of-magnitude estimate when the authors did no extrapolations of their findings to the global scale. The calculated figures can then be compared with the estimates for global river sediment fluxes that were made based upon compilations of world-wide river data. Such compilations have been done, for example, by Holeman [1968] or by Milliman and Meade [1983], resulting in global fluxes of about 20 Gt (gigatons, 106 t) and 13.5 Gt (with an upper limit of 16 Gt) per year, respectively. In a more recent work based upon a greater number of river data for small basins, Milliman and Syvitski [1992] supposed that global FTSS may be as great as 20 Gt/yr.

4.3.1. Relationships with Precipitation

Precipitation has been widely held to have a predominant influence on river sediment yields. Several non-linear relationships have been proposed linking the two parameters. Some of them are

108

depicted in Figure 59. Although there is little agreement in detail between these postulated relationships, certain common elements emerge. An initial peak of erosion is observed in semi-arid environments, and a progressive increase in denudation above a mean annual precipitation (APPT) value of about 1000 mm occurs. The existence of such a two-maxima curve has been at least partly related to the role of vegetation. The presence of vegetation greatly reduces the erodibility of surface materials, and the abrupt retardation of mechanical denudation rates above the precipitation value at the first maximum was attributed to a marked increase of vegetation cover (e.g., Langbein and Schumm [1958]). With ongoing precipitation increase, the maximum protective effect of vegetation may be reached, and beyond a certain threshold value, increasing precipitation (and thus also runoff) tends to increase the rates of mechanical erosion.

0 400 800 1200 1600 2000

10

100

1000

Mean Annual Precipitation (mm)

Denudation Rate (mm/kyr)

Langbein and Schumm [1958]

Douglas [1967]

Ohmori [1983]

Wilson [1973]

(extrapolated)

Fig. 59 Proposed relationships between denudation rate and mean annual precipitation (figure from Summerfield [1991]).

Langbein and Schumm [1958] proposed a mathematical formula for their relationship, but normally the curves shown in Figure 59 were obtained by a by-hand-fitting of data points, often with a great scattering of the data around the curve (Wilson [1973]). Since all curves in Figure 59 (in the following the L&S curve for Langbein and Schumm [1958], the D curve for Douglas [1967], the W curve for Wilson [1973], and the O curve for Ohmori [1983]) cover the APPT range of 250 to 1250 mm, I fitted spline interpolations to them in this range. For all fits, the correlation coefficient was r > 0.98. Then the corresponding sediment fluxes were calculated by applying the relationships to the global precipitation data set used in this study (see chapter II). A mean rock density of 2700 kg/m3 was taken for the conversion of denudation rates to weight units (according to Summerfield [1991]). Only the ice-free and exoreic grid points of the continents were considered. This leads to sediment fluxes of 10.8 Gt/yr, 11.5 Gt/yr, 19.3 Gt/yr, and 8.8 Gt/yr for the L&S, D, W, and O relationships, respectively. The 250 to 1250 mm precipitation range contributes about 42% of total precipitation falling on the continents. For the remainder 59%, an estimation of the corresponding FTSS values is, of course, speculative. Taking only the O curve, which is based on the largest set of observations (as mentioned in Summerfield [1991]), an extrapolation of the curve up to 2000 mm (see Fig. 59) yields an additional sediment flux of 23.4 Gt/yr for the 1250 to 2000 mm precipitation range, corresponding to about 27%

109

of total precipitation. Assuming then that the O curve does not further increase above 2000 mm and the denudation rate remains constant, results again in a sediment plus of about 24.4 Gt/yr for the rest of total continental precipitation (the 0 to 250 mm precipitation range can be neglected here). The total amount of sediment that should be discharged to the oceans according to the O curve would therefore sum up to a value of about 57 Gt/yr.

Again, it has to be said that the above made extrapolations are highly debatable and the calculated sediment fluxes should only be considered as order-of-magnitude estimates. Nevertheless, the extrapolations indicate that the proposed sediment yield - precipitation relationships tend to produce much higher global sediment fluxes as this can be confirmed from the world-wide river data compilations (see above), at least as far as they include a second maximum in the above 1000 mm range (all curves except the L&S one). In this context, it is also interesting to mention the approach of Fournier [1960] to predict sediment yields, even if he established individual regressions for different ranges of relief, and this approach can thus already be considered as a combined parameter model (combined parameter models are discussed below). He selected the month of the year where precipitation is greatest and proposed river sediment yields to be correlated with the ratio of the square of this monthly precipitation over the mean annual precipitation. Note that in this study a modified form of his index was used (see chapter II) which is, however, well correlated with the index originally proposed by Fournier [1960]. Fournier's approach underlines the importance of a strong seasonality of precipitation for sediment fluxes, but his index is naturally also highly correlated with total precipitation (see also later in the text Fig. 61). When he extrapolated his models to the total continental area, he calculated an overall sediment flux of more than 64 Gt/yr, which is also far above from what is found with the world-wide river data compilations.

4.3.2. Relationships with Climate

Because precipitation is a central element of climate, the above discussed relationships can be considered to be a special case of sediment yield - climate relationships. Climate, however, is characterized by a multitude of additional parameters, of which temperature and seasonal variability are the most important. Relationships between sediment yields and the climate type in which the sediment fluxes have been measured can be considered as a combined parameter model, where it is not known, which exactly are the factors that have contributed to the made observations. Especially vegetation characteristics strongly depend on climate, whereas geomorphological factors are normally quite independent of it (see also later in the text Fig. 61).

One of the most extensive studies investigating the relationships between sediment yields and climate is the work of Jansson (Jansson [1982], [1988]). She used a climatic classification modified from the Köppen's classification (Köppen [1936]) which is different from the classification applied in this study because the Köppen's classification also distinguishes seasonal features such as the occurrence of dry or wet periods, monsoon rains etc.. Jansson reported great sediment yields in the non-seasonal tropical climate, but relative low values in the tropical climate with a dry season (although a great scatter occurred in nearly all climate classes that were distinguished, which was mainly attributed to differences in morphology and lithology). In contrast, the warm temperate climates with a dry season were found to exhibit rather great values, but not the warm temperate climates without a dry period. Boreal climates were found to have low values, except for climates being boreal as a consequence of altitude rather than of latitude, where sediment yields were usually greater. She did not try to extrapolate her findings to the global scale in order to get an estimate for the total sediment flux to the oceans, probably because of the great scattering of the data. One may state here that her investigations confirm to some extent the trends emerging in the above discussed sediment yield - precipitation relationships: great sediment yields occur both in dry regions (first maxima in the FTSS - APPT curves ?) and in very humid tropical regions (second maxima ?). However, the great scattering in her data indicates that climate alone is probably not a useful criterion to predict river sediment yields.

110

4.3.3. Relationships with Basin Elevation and Morphology

Many studies pointed out the great influence of basin elevation and morphology on river sediment fluxes, but only a few of them proposed mathematical relationships. Among these studies are the one of Pinet and Souriau [1988] and the one of Ahnert [1970]. The former authors found river sediment fluxes to be linearly correlated with mean basin elevation, while the latter author proposed a linear relationship with local relief, which is the difference between maximal and minimal elevation in a given sector. For the study of Ahnert [1970] this is not completely true because he considered both chemical and mechanical denudation rates, even if it can be supposed that by far the larger parts in the values he used should reflect sediment transport. He assumed that any climatic influence, for example through differences in drainage intensity, should mainly change the ratio of chemical to mechanical denudation, but not total denudation rates.

Looking mainly at river sediment yields of large world rivers, Pinet and Souriau [1988] proposed the following two equations to describe mechanical denudation globally:

Ds = 419 x 10-6 Elev - 0.245 (21)

Ds = 61 x 10-6 Elev (22)

Ds is the mean denudation rate in m/kyr, and Elev the mean basin elevation in m. Equation 21 is valid for mechanical denudation taking place in regions related to orogenies younger than 250 Million (Mio) years, while equation 22 is related to orogenies older than 250 Mio years. The considerable intercept in equation 21 was attributed by the authors to continental sedimentation in the basins associated with young orogenies.

In order to apply these equations to the total continental area on the basis of the elevation data used in this study, an identification of the continental regions belonging to younger orogenies is needed. For a first order approximation, I followed the distribution of continental crust ages given by Burchfiel [1983], and adapted it to the continental grid point distribution I used. The result is shown in Figure 60. Then equations 21 and 22 were applied to the corresponding regions (again, only the ice-free, exoreic grid points were taken). This results in a total sediment flux to the oceans of about 16.2 Gt/yr. About 5.2 Gt/yr would originate from the parts of the continents older than 250 Mio yrs (73% of the total area), while 11 Gt/yr would come from the parts younger than 250 Mio yrs (27% of the total area). In the latter regions, however, more than 17.6 Gt/yr would be stored by continental sedimentation according to the negative intercept in equation 1. This is more than the total amount of sediments that is exported to the oceans. Note that for the calculations, this time only a mean rock density of 2500 kg/m3 was taken for the conversion of denudation rates to sediment fluxes following according to Pinet and Souriau [1988].

The so derived global sediment flux estimate is in better agreement with the estimates derived from the world-wide river data compilations than this is the case for the sediment yield - precipitation relationships (see above). However, correlating river sediment yields uniquely with elevation can only hardly be justified on the basis of a causal relationship between both parameters because the flux of sediment at a specific location should be rather a function of the gradient at that point, irrespective of its elevation above mean sea level. This has also been pointed out by Summerfield and Hulton [1994] who did a compilation of sediment yield data for major world rivers together with a detailed characterisation of the morphology of the basins. They reported that, at first sight, the idea that mechanical denudation rates vary as a function of mean elevation would appear to be supported by their data. But they show at the same time that basin elevation is also strongly correlated with other topographic factors which show even greater correlation with sediment yields, such as, for example, basin slope.

111

-150 -100 -50 0 50 100 150

-150 -100 -50 0 50 100 150

-50

0

50

-50

0

50

< 250 Mio yrs

> 250 Mio yrs

Fig. 60 Regions of the continents belonging to orogenies younger than 250 Million years (adapted from Burchfiel [1983]).

For this reason, it is interesting to look at the study of Ahnert [1970], because he related denudation rates to local relief rather than to mean elevation. He proposed the following equation to be the best model for denudation rates:

Ds = 153.5 x 10-6 LR - 10.88 x 10-3 (23)

Ds is again the mean denudation rate in m/kyr, and LR the local relief in m. In his work, Ahnert [1970] showed also that local relief is strongly correlated with slope, but it is difficult to apply his LR - slope relationship to the slope data I used in this study. In order to apply equation 23 to the global scale, I derived from the elevation data a global LR data set by calculating the difference between maximal and minimal elevation in a 10' x 10' longitude/ latitude resolution. This is comparable with the approach of Ahnert [1970]. He derived local relief by taking the values form topographical maps within sectors of 400 km2 each. Note that a 10' x 10' grid element at the equator comprises an area of about 340 km2, which is not much different. For calculation facilities, I averaged then the LR values to a 0.5° x 0.5° longitude/ latitude grid point resolution. With the so created data set, a global sediment flux to the oceans of about 9.3 Gt/yr is found when equation 23 is applied (again a mean rock density of 2500 kg/m3 was taken according to Ahnert [1970]).

The global scale extrapolations of the above presented relationships are naturally depending on the question to what extent the data used to establish the relationship are representative at the global scale, or not. If this is the case, however, one may conclude that a pure coupling of sediment yields to morphological parameters requires either a very different behaviour of mechanical denudation for certain parts of the continents (with very high continental sedimentation rates), or it leads to erosion rates which tend to be somewhat too low compared to the estimates derived from the world-wide river data compilations. This is true in as much the global sediment fluxes calculated with the morphological parameter models should be rather maximum values because they include also the deserts of the continents, where nearly no water runs off.

4.3.4. Combined Parameter Models

Because of indications for both climatic and geomorphological controls, one may finally look at combined parameter models that have been proposed to predict river sediment yields globally. Jansen

112

and Painter [1974] investigated sediment yields for 79 river basins (basin area > 5000 km2) together with 8 potential controlling factors: drainage intensity (Q, mm), basin area (A, km2), mean elevation (Elev, m), Slope (relief length ratio = equivalent to main channel slope, m/km), annual precipitation total (APPT, mm), mean annual temperature (AT, °C), an index characterising the soil protection capacity of vegetation (VegI, deserts = 1 to forest = 4), and an index characterising the softness of lithology (LithMI, 2 = hardly erodible to 6 = easily erodible). For all rivers together, they found the following model to be the best model to predict sediment yields globally (the unit of FTSS is t km-2 yr-1):

log FTSS = 0.100 log Q - 0.314 log A + 0.750 log Elev + 1.104 log Slope + 0.368 log AT + 0.786 log LithMI - 2.324 VegI -2.032 (24)

Jansen and Painter [1974] also grouped the basins according to a fourfold climatic classification and repeated the statistics within each of the groups. The trends in the resulting equations remained similar to those of equation 24. Sediment fluxes increased with increasing runoff, altitude, relief, precipitation, temperature and rock softness, and they decreased with increasing basin area and increasing vegetation protection, even if some prominent exceptions occurred. Based upon their different climatic models, they extrapolated the global sediment flux to be 26.7 Gt/yr.

In a similar study, Probst [1992] proposed on the basis of data for large river basins two multiple regression models to predict sediment fluxes with 4 and 5 parameters, respectively:

ln FTSS = 0.9655 ln Slope + 0.0023 Q + 0.5692 ln APPT - 0.8660 VegI + 1.561 (25)

ln FTSS = 1.028 ln Slope + 0.0365 LithMI + 0.6932 ln APPT + 0.0016 Q - 0.7516 VegI - 72.3 x 10-3 (26)

Variables and units are the same than in equation 24, but this time Slope was calculated in the same way as in the here presented study (but the unit in the above given equations is %), VegI ranges from 0 to 6 (desert to forest), and LithMI is the same index that was also used in this study (I recall here from chapter II that LithMI ranges from 1 to 40, with 1 = plutonic and metamorphic rocks, 2 = volcanic rocks, 4 = consolidated sedimentary rocks, 10 = different rock types in folded zones, 32 = non-consolidated sedimentary rocks, 40 = recent alluvials). When he applied equations 25 and 26 to the corresponding parameter averages calculated for 10° latitudinal bands, he found total sediment fluxes of 22.9 Gt/yr and 21.7 Gt/yr, respectively.

As it can be expected, the global sediment flux estimates derived by the combined parameter models lie in between the estimates obtained from the regression models which are uniquely based on precipitation data, and the models which are uniquely based on morphological data. However, one has to mention here that the above described combined parameter models are probably very scale-depending. The flux estimates have been established on the basis of large scale averages, and it is doubtful whether similar results would be obtained when the equations would be applied to finer spatial scales. Note that many parameters enter the equations in an exponential way.

4.4. A New Modelling of the Climatic, Morphological and Lithological Control of River Sediment Yields

We have seen on the previous pages that proposed models for the prediction of river sediment yields can be highly variable with respect to the retained controlling parameters, as well as with respect to the total sediment fluxes they predict. Table 13 gives a summary of the discussed models. This variability makes the applicability of these models to the overall continental area highly questionable in a study like this. Therefore it is investigated in the following which are the major

113

controlling factors for river sediment yields on the basis of the data used in this study, and whether it is possible to use them for prediction of sediment yields both on regional and global scales.

Table 13. Summary of the different models for sediment yield prediction discussed in the text.

______________________________________________________________________________________________________________ Parameters used Total Sedi-

Study for sediment yield ment flux Remarks prediction (Gt/yr) ______________________________________________________________________________________________________________

Langbein and Schumm [1958] APPT a (10.8) Hand-fitted relationship. The given sediment flux corresponds only to about 42% of global APPT (250 to 1250 mm range).

Douglas [1967] APPT a (11.5) Hand-fitted relationship. The given sediment flux corresponds only to about 42% of global APPT (250 to 1250 mm range).

Wilson [1973] APPT a (19.3) Hand-fitted relationship. The given sediment flux corresponds only to about 42% of global APPT (250 to 1250 mm range).

Ohmori [1983] APPT a 56.6 As above, but the relationship was also extrapolated to the remainder 58% of global APPT (incl. > 1250 mm range)

Fournier [1960] Four, Relief 64.0 Sediment yields were correlated with a seasonality index for precipitation. Different relationships for different relief types.

Pinet and Souriau [1988] Elev, Orogeny a 16.2 2 linear relationships with elevation depending on the orogeny Type type. Important sedimentation (a 17.6 Gt/yr) in young orogenies.

Ahnert [1970] LR a 9.3 Linear relationship with local relief (which is the difference between maximal and minimal elevation in a given sector).

Jansen and Painter [1974] Q, A, Elev, Slope 26.7 Multiple correlation models depending on the climate type to AT, LithMI, VegI which the drainage basins belong. Variable basin sizes.

Probst [1992] - model I Slope, Q, APPT, 22.9 Multiple correlation model on the basis of data from large river VegI basins. Best model including 4 variables.

Probst [1992] - model II Slope, LithMI, APPT 21.7 Multiple correlation model on the basis of data from large river Q, VegI basins. Best model including 5 variables. ______________________________________________________________________________________________________________

Explanations: A, drainage basin area; APPT, annual precipitation total; AT, mean annual temperature; Elev, mean elevation; Four, index characterizing the seasonal variability of precipitation (note that in this study a slightly modified form of this index is used - see chapter II); LithMI, index characterizing the softness of lithology; LR, local relief; Q, drainage intensity; Slope, mean slope; VegI, index characterizing the soil protection capacity of vegetation. For further explanations, see text. a Values calculated in this study.

4.4.1. Identification of the Controlling Parameters

Figure 61 shows a graphically presented correlation matrix between FTSS and most of the environmental parameters that were determined in this study for the 60 river basins of Table 12 (see chapter II). One can see that Q has the strongest correlation with FTSS among all parameters considered. The next strongest correlation with FTSS include APPT, Four, VegC, LithMI, and Slope. Correlation of certain variables with FTSS must not necessarily indicate a causal relationship because multicollinearity exists between various parameters. Strong multicollinearity exists especially between the different hydroclimatic variables, and between the hydroclimatic and the biological variables. The good correlation between Q and VegC may explain, for example, why there is a positive correlation between VegC and FTSS, while one should expect an inverse relationship between both parameters because of the protection of the soils against mechanical erosion by a dense plant cover (as discussed above). Note that there occurs no multicorrelation between the hydroclimatic and the geomorphological parameters.

Linear multiple correlation statistics do not yield a significant increase in the correlation between FTSS and the parameters shown in Figure 61, when compared to the correlation between FTSS ,

114

and Q. However, some improvement is obtained by using the product of certain parameters. The following equation gives the best model to describe the sediment fluxes globally:

FTSS = 0.020 (Q x Slope x Four) (27)

n = 58, r = 0.91, P < 0.0001

+0.66

+0.53

+0.93

- 0.57

+0.75

+0.86

+0.64+0.87

+0.75 +0.58

- 0.61

+0.63

- 0.65

- 0.60

+0.58

+0.67+0.55

C u ltA P opD

AT

Q APPT

ForR

E lev

ElevM

S o ilTI

Four ArIn

S lope

S o ilH

A SoilC

+0.70+0.61

+0.47

+0.29

+0.34

- 0.23- 0.24

+0.33

+0.39

+0.22

VegC

LithMI

LithCI

+0.34

F TSS

Fig. 61 Correlation between sediment yields (FTSS) and different environmental parameters (bold lines; only regressions significant with P < 0.1 are depicted) for the 60 river basins of Table 12, as well as correlation between these environmental parameters (fine lines; only regressions withcorrelation coefficients < - 0.5 and > + 0.5 are depicted). Q, drainage intensity; AT, mean annualtemperature; APPT, mean annual precipitation total; Four, modified Fournier-index; ArIn, aridity index; LithMI and LithCI, indices for the erodibility of the abundant basin lithology with respect to mechanical and chemical erosion, respectively; Elev, mean modal basin elevation; ElevM, maximumbasin elevation; Slope, average basin slope; SoilTI, index for the erodibility of the abundant soil type;SoilH, average soil depth; A, basin area; SoilC, average organic carbon content in the soils; VegC,average biomass density; ForR, mean forest ratio; CultA, percentage of the cultivated area in thebasins; PopD, mean population density in the basins. For details on the parameters, see chapter II.

The units are t km-2 yr-1 for FTSS, mm for Q and Four, and radian for Slope; r is the correlation coefficient, P is the significance level, and n is the number of river basins considered in the equation. Note that the product of Q, Slope, and Four (in the following also P3) was calculated by forming the product of all grid points in the basins, and not as the product of the basin averages. Including LithMI in the parameter product can still increase the correlation coefficient (r = 0.93), but this leads to a significant positive intercept (P < 0.1) in the regression, which makes the model less suitable for sediment yield predictions.

115

In equation 27, I omitted the Huanghe River in China and the Tana River in Kenya from the regression. Both rivers have by far the greatest specific sediment fluxes of all rivers included in this study. The extreme sediment yield found for the Huanghe River is related to one of its tributaries, the Huangfuchuan River. This tributary drains a highly erodible loess-covered terrain, leading to a specific sediment export of more that 50000 t km-2 yr-1 (Summerfield [1991]). Such local particularities cannot be accounted for in a study such as this one. It has been show that it is probably the interaction of the great erodibility of the abundant loess together with its extensive agricultural use that leads to the extremely great sediment fluxes of the Huanghe River (Milliman et al. [1987]). Also the high sediment load of the Tana River seems to be related to a particular, very erodible region in the basin (Charania [1988]). But the literature value for this river may also to be less reliable.

On the basis of the global Q, Slope, and Four data sets, equation 27 results in an average global sediment yield of 139.4 t km-2 yr-1 when applied to a total exoreic continental area of about 106 x 106 km2 (ice-free). The total sediment flux is thus 14.80 Gt/yr. This figure is in good agreement with the above mentioned world-wide river data compilations.

4.4.2. Climatic Particularities

Forming the product of parameters can lead to large differences between the resulting values, which holds the risk that the regressions are strongly influenced by extremes values. I tested therefore whether the above presented model, or similar models, can be confirmed by making subgroups of the river basins. As criterion to form the subgroups, the average climatic situation of the basins was selected (for the definition of the climate types, see chapter II). Figure 62 provides a plot of mean annual sediment concentration (cTSS) versus specific drainage intensity for all investigated rivers in this study. The rivers are additionally classified according to their average climatic situation. Although concentrations scatter over more than three orders of magnitude, one can see that for a given drainage intensity, the rivers in dry climates (white signs) tend to have greater concentrations than the rivers in wet climates (black signs). Consequently, omitting the dry climate rivers from the regression in equation 27 does not significantly change the regression coefficient, and the correlation coefficient still increases (r = 0.93). This is indicating that the above presented parameter product is useful to predict sediment fluxes in the wet climates, while in the dry climates sediment fluxes may be controlled differently. Applied to all wet grid points only, equation 27 yields a global sediment flux of 14.10 Gt/yr, with a corresponding area of 68.9 106 x km2. Note also from Figure 62 that the sediment concentrations of the wet climate rivers have the tendency to increase with increasing drainage intensities. Concentrations tend towards a value of about 1g/l. This concentration has been proposed by Probst and Sigha [1989] to be a limiting value in surface runoff waters. For the dry climate rivers, cTSS rather have the tendency to decrease with increasing drainage intensities. Such a behaviour has also been reported by Probst and Amiotte-Suchet [1992] for rivers of the Maghreb region in North Africa.

Taking only the group of river basins that belong to the dry climates, no clear correlation between the observed sediment fluxes and environmental parameters or parameter combinations of the river basins can be found. This is also less surprising because I have shown in chapter II that the climatic variability can be great for these basins, making the average basin values extracted from the data sets less meaningful. The climatic grouping is based on the average basin situation, but in some basins classified as dry, the river hydrology can be strongly influenced by a particular part of the basin that has quite different environmental characteristics in comparison to the rest of the basin. I recall here from chapter III that, for example, for the Colorado River only about 17% of the basin area can be classified as wet, but about 79% of the total runoff of the Colorado originates from this part of the basin. Another reason that could additionally contribute to generally weaker correlation in the dry climate river group is the already mentioned fact that the literature estimates for these rivers are probably less reliable.

116

Q (mm)

0 500 1000 1500 2000 2500

10

100

1000

10000

Tropical Wet

Temperate WetTropical Dry

Tundra & TaigaTemperate Dry

TSS (mg/l)

Fig. 62 Plot of mean annual sediment concentration versus mean annual drainage intensity for the river basins of Table 12.

Table 14 lists the results when the regression were performed within the individual climate

groups of river basins. The temperate dry and the tropical dry climates were still lumped together as one type because only 2 rivers belong to the temperate dry climate (not considering the Huanghe River), and this number is, of course, too small for meaningful statistical analysis. When parameter products are formed, it is not possible to determine the statistical significance of each parameter in the regressions. For this reason, I tested several parameter combinations. The comparison of the correlation coefficients for each combination can help to estimate whether additional parameters really improve the regression or not. For all 3 wet river groups, great correlation coefficients were found. In the tundra and taiga climate, both the products of Q, Slope, Four, and LithMI (P4) and of Q, Slope, and LithMI (P5) show the greatest correlation with FTSS. Four is probably not a very meaningful parameter in this climate type, because of the temporal storage of water as snow. There can be a considerable time lag between precipitation and mechanical erosion via the generation of runoff. Consequently, taking out Four of the parameter products does not decrease the correlation coefficient. Also in the tropical wet climate type, P4 and P5 yield the greatest correlation coefficients with FTSS. In the temperate wet climate, however, it is the product of Four and Slope (P2) that is best correlated with sediment fluxes, and neither the inclusion of Q, nor of LithMI in any of the parameter combinations can improve the regressions with respect to this correlation. As already discussed in the previous chapter, one can suppose that Four should be especially important when strong precipitation fall on soils that are in an intermediate position between field capacity and extreme water limitation. This can then provoke a short-time overflow of the soils, which may enhance sheet erosion compared to a site with the same annual climatic characteristics, but where precipitation is more uniformly distributed over the year. Moreover, an overflow of the soils when they have been dry before may also facilitate the detachment of the particles from the soil complex. Among the 3 wet climate types, it is probably in the temperate wet climate where the condition that water limitation in the soils coincides with strong precipitation is the most often fulfilled. In the tropical wet climate, for comparison, there should be a greater water excess, and the soils should be on average closer to field capacity throughout the year. Four might therefore be less important in this climate compared to the temperate wet climate, even if Four naturally has greater absolute values on average in the tropical wet climate.

It is worth to look at this aspect more in detail. Already when investigating the major controls for continental runoff in chapter III, Four was found to be more important in the temperate climates than in the other climate types. Figure 63 depicts for the climate types distinguished in this study (except the polar and the desert types) the difference between mean monthly precipitation and mean potential evapotranspiration (MPPT - MPE) over the year, and compares it with the monthly distribution of Four. The values represent the averages and standard deviations calculated on the basis

117

of all continental grid points that fall into the corresponding climate types, except that only the grid points of the northern hemisphere were taken because of the inversion of the seasons in the northern and the southern hemispheres (this is especially important for the tropical climates; the other climate types are underrepresented in the southern hemisphere). Potential evapotranspiration was determined as a function of temperature according to Holdridge [1959] (see chapter III). As a first approximation, the difference of MPPT and MPE can be used as an indicator of the water content in the soils over the year. When the difference is positive, the soils should be close to field capacity, and negative values indicate water limitation of the soils. Keep in mind, however, that the absolute values shown in the plots cannot be taken as a measure for the amount of water limitation, because the water limitation hinders the soil water to evaporate at potential rates.

Table 14. Regression of sediment yield versus several parameters and parameter products. Therivers are grouped according to their average climatic situation.

________________________________________________________________________________________________________

Correlation Coefficient r Number ________________________________________________________________________

Climate of Rivers Q P2 P3 P4 P5 P6

Tundra & Taiga 10 0.65 0.89 0.94 0.98 0.98 0.97 Temperate Wet 19 0.49 0.83 0.75 0.73 0.75 0.76 Tropical Wet 12 0.87 0.73 0.92 0.96 0.96 0.93

Dry Climates 17 0.80 0.55 0.79 0.81 0.55 0.63

Regression Coefficient m (f(x) = m x) Number ________________________________________________________________________

Climate of Rivers Q P2 P3 P4 P5 P6

Tundra & Taiga 10 0.22 13.13 0.037 0.00255 0.125 -- Temperate Wet 19 0.38 16.52 0.020 0.00119 0.114 1.97 Tropical Wet 12 0.83 26.70 0.020 0.00081 0.211 5.17

Dry Climates 17 1.18 25.73 0.088 0.01176 1.252 14.36

________________________________________________________________________________________________________

P2 = (Four x Slope) P3 = (Four x Slope x Q) P4 = (Four x Slope x Q x LithMI) P5 = (Slope x Q x LithMI) P6 = (Slope x Q) The boldface regressions are selected as the best models (see text). All regressions are significant at least with P < 0.05. The regression coefficient was calculated by forcing the regression to pass through the origin. It is only shown when the intercept was not significant in the regressions, which means that for the intercept P > 0.1 (P > 0.05 for the values in italics).

Per definition, there is an excess of water in the wet climate types, while the dry climates are water limited with respect to the annual totals. It is interesting to note from the comparison of the plots that in the temperate climate types, greatest Four values fall together with a tendency towards a water limitation potential in the soils. The opposite is the case for the tropical climate types, which is mainly true for the tropical wet climate type. Here, much of the annual Four is encountered when the soils should be at field capacity, making this parameter less meaningful in these climates. This could explain why Four is found in the regressions to be especially important in the temperate wet climate type.

An important open question is here, however, whether the parameter Four is at least not also partly correlated to precipitation intensity. Also this factor may naturally influence mechanical erosion rates in a drainage basin. One may speculate that an uneven precipitation distribution over the year (e.g. the abundance of rain periods) may fall together with a high frequency of thunder-storm events, and thus also with great precipitation intensities. But this may be only the case in certain regions.

118

Additional seasonal parameters reflecting precipitation intensity are needed to better understand why Four has been retained in the regression models.

-200

-100

0

100

200


-200

-100

0

100

200

Temperate Dry

Tropical Dry

-200

-100

0

100

200

-200

-100

0

100

200

-200

-100

0

100

200

Temperate Wet

Tundra & Taiga

Tropical Wet

MPPT - MPE (mm) Std. Dev. (mm)

Water Excess Water Limitation

-100

0

100

-100

0

100

-100

0

100

Temperate Wet

Tundra & Taiga

Tropical Wet

-100

0

100


-100

0

100

Temperate Dry

Tropical Dry

Four (mm)

Std. Dev. (mm) Fig. 63 Comparison of the potential of the soils to be water limited - as determined by the difference of monthly precipitation and monthly potential evapotranspiration (MPPT - MPE) - with the distribution of Four in different climate types. The values are calculated on the basis of all continentalgrid points of the northern Hemisphere that fall into the climate types. For further explanation, seetext.

For the group of the dry river basins, the statistics in Table 14 were performed in a different way. As mentioned above, great amounts of the total runoff in these basins can originate from small wet parts of the basins, which should also considerably influence the total sediment fluxes in these basin. I calculated therefore for the wet basin parts a theoretical sediment flux according to the best

119

regression models found in the wet river groups (boldface regressions), and subtracted the values from the observed total sediment flux. The resulting FTSS values were then used for regression with the basin characteristics, which were this time calculated on the basis of the dry basin grid points only. With this proceeding, P4 is found to be best correlated with sediment fluxes. Note that the regression coefficients in nearly all regressions are much greater than for the corresponding regressions in the wet

3.5 1.9 1.4--

1.2 1.8 5.9 6.4 5.2

0.8 0.8 1.8 6.2 7.9 8.8

0.5 1.3 1.9 3.7 5.1 11.8 10.0 --

0.5

0.7 1.7 2.5 4.6 6.4 11.9 11.8

0.7 1.1 2.1 3.7 4.9 4.9 --

0.5

0.25

1.0

2.0

4.0

8.0

16.0

ABT

APPTAPETR dry wet

Fig. 64 Mean biomass density (kg/m2) in the hexagons of the Holdridge Triangle (ice-covered and cultivated grid elements were omitted in the calculations). Among the triangle units, only the annual potential evapotranspiration ratio is depicted. For a description of the triangle, see chapter II.

climate types, which is in agreement with the general picture in Figure 62. This means that erodibility in dry climates is much greater than in wet climates. One explanation for this could be that water limitation in dry climates leads to a vegetation cover that represents a much less efficient protection of the soils against mechanical erosion. Soil protection by vegetation is probably a threshold phenomenon rather than a continuous effect. There may exist a minimal vegetation cover density, below which mechanical erosion rates rapidly increase. Note that biomass density is often strongly increasing from dry to wet conditions (Fig. 64). However, since I include modelled and observed FTSS values in the regressions, the here presented results have naturally to be taken with caution. More data especially of river basins that are dry over the complete basin area (monoclimatic) are needed to confirm such a trend.

Taking the most significant regressions in Table 14 (boldface regressions), the global sediment flux to the oceans can be determined on the basis of the corresponding data sets used in this study. This yields the following values (see also later in the text Table 19): tundra and taiga, 0.78 Gt/ yr (A = 27.10 x 106 km2), temperate wet, 3.10 Gt/yr (A = 16.91 x 106 km2), tropical wet, 5.86 Gt/yr (A = 24.90 x 106 km2), and dry climates, 5.08 Gt/yr (A = 37.34 x 106 km2). I refer to the corresponding regressions in the following as equation 27.1, 27.2, 27.3, and 27.4, respectively. The corresponding xy-plots are shown in Figure 65. In the value extrapolated with the model for the tundra and taiga climate, also the ice-free polar climate type (A = 3.89 x 106 km2) was included, and the value for the dry climates comprise also the desert climate type (A = 5.94 x 106 km2). Although these two latter climate types cover considerable parts of the continents, they contribute only very few to the total sediment flux (0.05 Gt/yr for the ice-free polar climate, and 0.03 Gt/yr for the desert climate). For the tropical wet climate, I selected the FTSS - P4 model compared to the FTSS - P5 model as the better model because of a slightly greater correlation coefficient for this regression. Note that taking the FTSS - P5 model instead would change the extrapolated sediment flux to a value that is only about 7% lower than the value calculated with the FTSS - P4 model.

120

3000

0 50 100 150 200 250

0

50

100

150

200

250

Tundra & Taiga(Slope x Q x LithMI)

y=x

0 100 200 300 400 500 600 700

0

100

200

300

400

500

600

700

Temperate Wet (Slope x Four)

y=x

0 500 1000 1500 2000 2500 3000

0

500

1000

1500

2000

2500

3000

Tropical Wet(Four x Slope x Q x LithMI)

y=x

0 100 200 300 400 500 600 700

0

100

200

300

400

500

600

700

Dry Climates(Four x Slope x Q x LithMI)

y=x

Fig. 65 Plot of predicted (y-axis) versus observed (x-axis) sediment yields (t km-2 yr-1) according to the best correlated parameter products in Table 14.

4.4.3. Influence of Lithology

Summing the sediment fluxes for the wet climate types results in a value of only 9.74 Gt/yr, which is considerably lower than the value that was calculated with equation 27 for all wet continental grid points. This is mainly due to the fact that LithMI was included into the regressions both for the tundra and taiga climate and for the tropical wet climate. Taking for these two climates the P3 - FTSS regressions instead, which have also still great correlation coefficients (Table 14), the corresponding sediment flux would increase to a value of 14.96 Gt/yr (tundra and taiga, 2.16 Gt/yr, temperate wet, 3.10 Gt/yr, tropical wet, 9.70 Gt/yr). Lithologies that are less resistant to mechanical erosion are over represented in the corresponding groups of river basins, especially in the case of the tropical wet rivers. Here, the mean LithMI value of all basins is 18.5, compared to 13.3 that is the global average of all continental grid points (exoreic and ice-free). In the other groups, the average LithMI values are closer to the global average (tundra and taiga, 14.0, temperate wet, 12.9, dry climates, 12.0). For this reason it is an important question whether the regressions would change with a set of tropical wet rivers which have more variable lithologies. Probably too much weight is given to LithMI in equation 27.3. This is important because it could lead to a considerable underestimation of the total sediment delivery to the oceans extrapolated in this study.

In this context one has also to mention that mechanical erosion in river basins consists of several processes which underlie not necessarily the same controls. Commonly one distinguishes hill slope erosion (or sheet erosion) which concerns all surfaces in the basins from more local processes such as gully erosion or river bed erosion (or channel erosion). Since the parameters I used were calculated by averaging the characteristics of all grid points in the basins, the correlation of FTSS and the here tested parameters and parameter products should mainly reflect the process of hill slope erosion. The influence of lithology on mechanical erosion rates, however, may be great with respect to channel erosion, but less important with respect to hill slope erosion, not at least because the outcropping

121

lithologies are normally covered by the soils. It is a well known fact that lithology can affect the hydrological network of a river system. Rivers often tend to follow the more erodible lithologies, which implies that highly erodible lithologies may especially increase channel erosion. Unfortunately, there is little known about the general contributions of channel erosion to total sediment fluxes, at least at the global scale. Kattan et al [1987] estimated channel erosion for the Senegal River to be at least 20% of the total river transport, and Etchanchu and Probst [1986] found for the Girou River (a tributary of the Garonne River in France) a value of 30%.

One may assume that the relative importance of channel erosion with respect to total erosion in a drainage basin could be also related to a great seasonality of the river discharge patterns because this can increase the transport capacity of a river. It is probably the coincidence of an erodible lithology with a strong seasonal variability of the river discharge patterns that enhances channel erosion. In this sight it is not very surprising that including lithology increases the correlation coefficient in Table 14 especially for the tundra and taiga rivers, where snow melting provokes considerable peak discharges of runoff. Nevertheless, such considerations are speculative and can neither be confirmed nor contested on the basis of the available data in this study. Further investigations should be designed to investigate the role of lithology for sediment fluxes. It would be interesting, for example, to determine the densities of the drainage networks in the basins from topographic maps, and to compare them with the hydroclimatic and lithological data sets.

4.4.4. Influence of General Basin Characteristics

Under general basin characteristics, all characteristics and factors that cannot be broken down to the grid point scale are understood here. In order to test whether such factors could influence sediment yields, I repeated the regression statistics with the parameter products (P2, P3, P4, P5) of Table 14, but this time I additionally grouped the rivers according to the general basin characteristics which are discussed in this section. If it is true that the above presented regression models represent the major controls for sediment yields, a further grouping of the river basins may help to identify additional controls. This proceeding underlies the assumption that different regression coefficients should appear for the individual groups if the tested basin characteristic significantly influence FTSS. Such differences have been found, for example, by comparing the regression coefficients between the rivers of the dry and the wet climate groups.

Because of the low significance of the average basin parameters for the rivers of the dry climate group, only the rivers of the wet climate groups are taken here. This should be of no importance because only geomorphological, geotectonical, and anthropogenic characteristics are used to distinguish the river groups. These characteristics are quite independent from climate. For grouping of the river basins, groups that comprise about the same number of rivers were formed, and the selection of the limiting values was therefore somewhat arbitrary. Because we have seen that the major controlling factors are not necessarily the same for the different climate types, also subgroups within the individual river climate groups were created when this was possible.

4.4.4.1. Basin Area and Hypsometry

The most prominent general basin characteristic is basin size itself (A). In many studies, basin area was found to be negatively correlated with FTSS (e.g., Milliman and Syvitski [1992], Probst and Amiotte-Suchet [1992], Probst [1992]). Figure 61 confirms the inverse relationship of A with FTSS at the global scale, but the correlation is only weak (as this was also noted by Summerfield and Hulton [1994]). The effect of drainage basin area on sediment yields has been explained in the literature in several ways. Among the most common arguments, it was mentioned that small basins generally exhibit steeper slopes and steeper stream gradients than large river basins (e.g., Wilson [1973]). The latter often have great areas with low slopes and low stream gradients especially in their downstream parts. Related to this, small basins tend to have limited flood plain development, so that most of the

122

material eroded from basin is just flushed out of the system, and little is deposited within the basin. Finally, Jansson [1982] noted also that the probability of an intense storm covering the whole of a river basins decreases with increasing basin size. For this reason, the maximum flood discharge per unit area should be inversely related to the size of the drainage area, and the capacity of the water flow to evacuate material which has been deposited in a channel after local thunderstorms should be smaller for large basins than for small basins.

Basin area can therefore be regarded to sum up several factors which may influence sediment yields. Many of these factors are considered as individual parameters in this study, as it is the case, for example, for Slope. Neglecting these factors, it is mainly the remainder correlation of basin size and the sediment storage capacity of a drainage basin that is needed to be investigated more in detail. The sediment storage capacity of a drainage basin is normally characterized with the sediment delivery ratio (SDR), which is the ratio of the amount of sediments that actually reaches the basin mouth over the amount of sediments being mobilized within the entire basin. In the long-term, SDR must be approximately 1, otherwise the river basin would become progressively choked with sediment (Summerfield [1991]). Only in very large basins, it may be less than 1 because sediment may progressively accumulate in the basin as the basin floor subsides under the load of overlying sediment. In the short-term, however, there may be a transfer of material within a drainage basin but little export of material from it. Debris removed far upstream can be stored in the river bed, in lakes, and on the floodplain before reaching the outlet. Studies investigating sediment delivery ratios in drainage basins found that from a few percent up to more than 100 percent of gross erosion (which is the total of all debris detached from the rocks and soils) can be exported from river basins (see, e.g., Walling [1983]). This is indicating that sediment retention in basins can be quite variable. However, when looking at these values one has to keep in mind that the determination of the sediment delivery ratio for a given basin needs, beside the measurement of sediment discharge at the river mouth, an estimate of gross erosion in the basin. Such an estimate is not easy to get because it requires to extrapolate local measurements over the entire basin area.

One can suppose that the sediment storage capacity of a drainage basin is greater when the basin is characterized by a relative high-standing and steep headwater region together with a relative low and flat downstream region compared to the case when the mountains and lowlands are more homogeneously distributed in the basin. The former case is often typical for great basins. But better as by basin size itself should this difference be reflected by the hypsometric curve of the basin. The hypsometric curve shows the percentage of cumulated surface above a given elevation level. Figure 66a depicts the hypsometric curves for all river basins of Table 12. In a first approximation, one may assume that the more the hypsometric curve is concave, the greater should be the storage capacity of a basin (as suggested, for example, by Pinet and Souriau [1988]). In order to derive numerical values for this effect, I determined for all basins a so-called hypsometry factor (HF), which is defined as:

HF = (ATri - ACurve) / ATri (28)

ATri is the area of the triangle formed by the points (0/0), (0/100), (100/100) in the plots of Figure 66a (that is the case when elevation linearly decreases with increasing area). ACurve is the area below the hypsometric curve. It was approximately determined by calculating the corresponding elevation for area increments of 5%.

The HF values of all basins are listed in Table 15. Great values close to 1 are typical for basins such as the Amazon Basin, which has his headwaters in the high Andes mountains, and passes then through large plains further downstream. Low values close to 0 characterize river basins where the morphology declines more or less continuously from upstream to downstream. This is the case, for example, for the Yana River. There can also negative values occur. This is normally found when large parts of elevated plateau regions are drained by the rivers. One example for this is the Zambesi River. As supposed above, large river basins tend to have great HF values (the correlation coefficient

123

between A and HF for all rivers is r = 0.50). Note also that for very small basins, the HF values are less reliable because the 0.5° x 0.5° longitude/latitude grid point resolution of the elevation data may be too coarse for the calculation of the hypsometric curves in these basins. For the statistics with all basins, a HF value of 0.35 was selected to separate two river groups (Table 16). This value corresponds, for example, to the hypsometric curve of the Liao He River (LHe in Fig. 66a). Within the climate river groups, the limiting value has been varied between HF = 0.30 and HF = 0.45 in order to obtain about equal numbers of rivers in each subgroup.

Table 15. Orogeny and hypsometry of the river basins of Table 12.

______________________________________________________________________________________________________________ Young Orogeny Hypso- Young Orogeny Hypso- Young Orogeny Hypso-

River (% of Total) metry River (% of Total) metry River (% of Total) metry _____________ Factor _____________ Factor _____________ Factor Area Runoff (1) Area Runoff (1) Area Runoff (1) ______________________________________________________________________________________________________________

Amazon 19 21 0.78 Yukon 100 100 0.49 Liao He 0 0 0.34 Zaire 0 0 0.43 Huanghe 14 20 0.30 Rufiji 0 0 0.01

Mississippi 10 6 0.59 Danube 73 80 0.61 Rio Negro (Arg.) 69 97 0.18 Ob 0 0 0.82 Orange 0 0 0.03 Hungho 0 0 0.02

Paraná 9 4 0.77 Colorado 84 97 0.14 Rhine 100 100 0.54 Yenisei 0 0 0.50 Columbia 94 84 0.01 Brazos 0 0 0.21

Lena 24 26 0.32 Kolyma 73 75 0.26 Loire 0 0 0.51 Amur 23 33 0.44 Sao Francisco 0 0 -0.14 Rhône 37 25 0.39 Nile 0 0 0.43 Si Kiang 46 52 0.40 Tana 0 0 0.34

Changjiang 51 48 0.44 Irrawaddy 100 100 0.66 Garonne 25 33 0.56 Ganges/Brahm. 56 79 0.56 Don 0 0 0.00 Po 67 72 0.32

Mackenzie 21 34 0.37 Senegal 0 0 0.48 Gambia 0 0 0.50 Niger 0 0 0.55 Indagirka 69 66 0.25 Fly 69 86 0.68

Zambesi 0 0 -0.18 Limpopo 0 0 0.02 Susitna 100 100 0.07 Murray 1 5 0.59 North Dvina 0 0 -0.15 Purari 100 100 -0.05

St. Lawrence 0 0 0.38 Godavari 0 0 0.16 Tiber 100 100 -0.28 Orinoco 25 12 0.74 Magdalena 100 100 0.18 Rioni 100 100 0.08

Tigris/Euphrates 39 88 0.57 Fraser 88 86 -0.11 Severn 0 0 0.27 Indus 68 98 0.43 Yana 100 100 0.05 Waikato 100 100 0.01

Mekong 81 80 0.69 Mahandi 0 0 0.08 Ems 0 0 0.04 ______________________________________________________________________________________________________________

(1) see equation 28

Table 16 shows that the regression coefficients are not much different in the corresponding groups, as far as the regressions are concerned for which in both groups great correlation coefficients are found. There are no indications that a concave hypsometry is accompanied by enhanced sedimentation in the basins, and thus to a lower sediment delivery out of the basins. On the contrary, the regression coefficients in the groups with low HF values tend to be even slightly greater than in the groups with great HF values. This does, of course, not mean that there is no sedimentation in the basins. It implies only that hypsometry is an insufficient criterion to detect sediment storage in the basins. The abundance of internal reservoirs, especially of lakes, should be more important in this context. It is probably not by chance that the lowest annual sediment concentration among the rivers of Table 12 is observed for the St. Lawrence River, which has an important part of its basin area covered by lakes. The presence of lakes is not necessarily related to hypsometry. Lakes can be found both in mountainous regions as well as in lowlands. A causal relationship between hypsometry and the occurrence of floodplains is more plausible, but here one has to keep in mind that an efficient storage of sediments on the floodplain needs a periodical over-banking of the river. This effect is also related to seasonality of the river discharge patterns. The Nile, for example, is well known for his sediment deposits in its lower course before the construction of the Aswan Dam, but the HF value of 0.43 for this river is only moderate.

124

0

40

80

0

40

80

0

40

80

0

40

80

0

40

80

0

40

80

0

40

80

0

40

80

0

40

80

0

40

80

0 40 80 0 40 80 0 40 80 0 40 80 0 40 80 0 40 80

427 m 296 m 563 m

Tib Sev

123 m

1861 m

829 m 913 m2042 m1790 mTan Po Gam Fly Sus

Wai Ems

Gar1924 m

1940 m

Pur

Rhi2374 m

1913 m

RNe

1900 m

Hun

1322 m

Brz Loi1193 m

Rho2177 m

Rio1844 m

1546 m

Yan

2023 m

Fra

2933 m

Mag

1877 m

Ruf

723 m

MahLHe

1550 m

NDi

180 m 947 m

God

1677 m

Lim

1963 m

Ink

241 m

Don Sen 813 m

3378 m

Cdo

2507 m

Col

1506 m

Kol

1090 m

SFr SKi 2139 m

Irr4166 m

2507 m

OrgDan

2640 m

Hua

5470 m

Yuk

2841 m

Mek

5510 m

Ind

5690 m

1700 m

Zam TiEu 2641 m

Ori2676 m

Sla 727 m

Nig1531 m

Mur1133 m

GaBra6133 m

Yts5742 m

Nil2946 m

Amo1891 m

Len1729 m

Mac 2013 m

Par4688 m

Ob2882 m

Mis 3269 m

Zai2404 m

Amz4792 m

Ien3000 m

Fig. 66a Hypsometric curves of the river basins of Table 12. The x-axis shows cumulative basin area (as percentage of total basin area) and the y-axis shows basin elevation (as percentage of the maximal elevation, which is given as numerical value in the plot). For abbreviations of the river names, seeTable 12.

125

Amz2.48

0

40

80 Zai2.12

Mis 2.85

Ob 1.98

Par 2.54

Ien 4.90

Len4.91

0

40

80 Amo 4.01

Nil 3.31

Yts 7.57

GaBra 7.75

Mac 5.52

Nig1.74

0

40

80 Zam 2.58

Mur 1.93

SLa 2.33

Ori 3.14

TiEu 5.34

Ind8.46

0

40

80 Mek 4.91

Yuk 8.41

Hua 6.85

Dan 7.16

Org 2.77

Cdo9.84

0

40

80 Col 12.59

Kol 6.85

SFr 2.57

SKi 4.61

Irr 8.46

Don0.84

0

40

80 Sen 1.23

Ink 7.22

Lim 2.85

NDi 0.81

God 2.48

Mag14.48

0

40

80 Fra 14.64

Yan 7.39

LHe 4.00

Ruf 5.52

Mah 2.85

RNe 4.43

0

40

80 Rhi 7.86

Hun 7.53

Brz 1.46

Loi 3.82

Rho 14.00

Tan2.20

0

40

80 Gar 6.94

Po 10.91

Gam 1.92

Fly 4.09

Sus 10.88

1 3 5 642

Pur13.04

0

40

80 Rio17.72

Tib 14.56

Sev 3.35

Wai 5.28

Ems 2.24

1 3 5 6421 3 5 6421 3 5 6421 3 5 6421 3 5 642

Fig. 66b Histograms of the slope distribution of the grid points in the river basins of Table 12. The following classes were created: 1, 0-3°; 2, 3-6°; 3, 6-9°; 4, 9-12°; 5, 12-15°; 6, > 15°. The numerical values in the plots depict mean basin slope. White colour corresponds to river basins related toorogenies older than 250 Mio yrs, dark grey colour to basins related to orogenies which are youngerthan 250 Mio yrs, and light grey colour to basins with an intermediate position. For further explanation, see text.

126

Table 16. Regression of sediment yield versus several parameter products. The rivers are groupedaccording to basin hypsometry. Only river basins of the wet climate types are selected.

____________________________________________________________________________________________________________

River Subgroup Number Correlation Coefficient r Regression Coefficient m (f(x) = m x) _______________________ of _____________________________________ _____________________________________ Climate Hypsometry Rivers P2 P3 P4 P5 P2 P3 P4 P5

[5] HF < 0.30 5 0.96 0.99 0.99 0.99 16.43 0.042 0.00254 0.125 [5] HF > 0.30 5 -- -- 0.85 0.89 -- -- 0.00264 0.128

[7] HF < 0.35 10 0.86 0.75 0.85 0.87 16.51 0.019 0.00109 0.104 [7] HF > 0.35 9 0.78 0.75 -- 0.61 16.55 0.024 -- 0.185

[8] HF < 0.45 5 0.85 0.99 0.97 0.97 31.65 0.022 0.00079 0.198 [8] HF > 0.45 7 -- 0.72 0.95 0.96 -- 0.017 0.00086 0.279

all wet HF < 0.35 19 0.87 0.98 0.96 0.93 28.00 0.022 -- 0.175 all wet HF > 0.35 22 0.69 0.81 0.92 0.92 18.62 0.017 0.00088 0.260 ____________________________________________________________________________________________________________

P2 = (Four x Slope) P3 = (Four x Slope x Q) P4 = (Four x Slope x Q x LithMI) P5 = (Slope x Q xLithMI) HF is a factor characterising the hypsometric curve (see text). Climate types: [5] tundra & taiga, [7] temperate wet, [8] tropical wet All regressions are significant at least with P < 0.05 (P < 0.1 for the values in italics). The regression coefficient was calculated by forcingthe regression to pass through the origin. It is only shown when the intercept was not significant in the regressions, which means that for the intercept P > 0.1 (P > 0.05 for the values in italics).

4.4.4.2. Orogeny

Another general basin characteristic that has been proposed to influence river sediment yields is orogeny (e.g., Pinet and Souriau [1988]). Young orogenies are often characterized by enhanced uplift rates because tectonical activity has not yet ceased. Related to this, the degree of fracturing of the outcropping rocks should be greater, facilitating the process of mechanical erosion. In old orogenies, the rocks are often highly consolidated. An investigation of the influence of tectonical activity on river sediment fluxes requires independent estimates of local uplift rates, which are, of course, not available for a global scale study such as this one. However, as mentioned above, Pinet and Souriau [1988] found mechanical erosion rates (river sediment yields and basin internal sedimentation) to be about 7 times greater in river basins related to orogenies younger than 250 Mio years compared to basins related to orogenies older than 250 Mio years. Such marked differences should be detectable here. I grouped therefore the river basins of Table 12 according to the orogeny type to which they belong (old, young) using the simplified map of Burchfiel [1983] that is shown in Figure 60. In the case of large rivers, the basins often stretch over both types, and for this reason I calculated for each basin the percentage of the total basin area that falls into the young orogeny type, as well as the percentage of the total basin runoff that originates from this area (Table 15). In many basins, much of the runoff originates from the headwater regions (which often represent in mixed basins the young orogeny type), and runoff distribution may be more significant here than only the area distribution. For a simple distinction, I considered river basins with more than 50% of total runoff originating from the young orogeny type (Qy) to represent the young orogeny type, and basins with Qy < 30% to represent basins of the old orogeny type. Basins in between should be in an intermediate position, but these river basins are not very numerous and I classified them into the old orogeny type. The results of the performed statistics when such a grouping was applied are shown in Table 17. In the subgroups, however, the old to intermediate orogeny types are overrepresented in the temperate wet climate, while the young orogeny type is overrepresented in the tropical wet climate. For this reason I set in these climates the limits to form the subgroups to Qy = 30% and Qy = 80%, respectively.

127

Table 17. Regression of sediment yield versus several parameter products. The rivers are groupedaccording to orogeny. Only river basins of the wet climate types are selected.

____________________________________________________________________________________________________________

River Subgroup Number Correlation Coefficient r Regression Coefficient m (f(x) = m x) ________________________ of ______________________________________ ______________________________________Climate Qrogeny Rivers P2 P3 P4 P5 P2 P3 P4 P5

[5] Qy > 50% 5 0.99 0.99 0.99 0.99 -- 0.040 0.00253 0.125 [5] Qy < 50% 5 -- -- -- 0.85 -- -- -- 0.161

[7] Qy > 30% 10 0.94 0.80 0.86 0.87 -- 0.019 0.00112 0.106 [7] Qy < 30% 9 0.79 0.69 -- -- 20.27 0.025 -- --

[8] Qy > 80% 6 -- 0.89 0.95 0.95 -- 0.021 0.00081 0.210 [8] Qy < 80% 6 0.95 0.96 0.97 0.98 16.69 0.014 0.00073 0.236

all wet Qy > 50% 20 0.78 0.93 0.96 0.91 25.0 0.0200 0.00089 0.188 all wet Qy < 50% 21 0.79 0.67 0.55 0.66 16.59 0.0187 0.00125 0.195 ____________________________________________________________________________________________________________

P2 = (Four x Slope) P3 = (Four x Slope x Q) P4 = (Four x Slope x Q x LithMI) P5 = (Slope x Q xLithMI) Qy is the percentage of total runoff that is formed in the basin part younger than 250 Mio yrs (see text). Climate types: [5] tundra & taiga, [7] temperate wet, [8] tropical wet All regressions are significant at least with P < 0.05 (P < 0.1 for the values in italics). The regression coefficient was calculated by forcingthe regression to pass through the origin. It is only shown when the intercept was not significant in the regressions, which means that for the intercept P > 0.1 (P > 0.05 for the values in italics).

Also for the grouping of the basins according to orogeny, the regression coefficients are not much different in the corresponding groups. Again, the river groups representing the old orogeny type have even slightly greater regression coefficients compared to the group representing young orogeny type, contrary to what should be expected. One has also to mention here that hypsometry and orogeny are not really independent from each other. The river basins belonging to the young orogeny type are often characterized by steep morphologies and relative straight hypsometric curves (see Fig. 66a and Fig. 66b). Note that of the 19 rivers with low HF values, 13 belong also to the young orogeny group. The fact that according to the statistics orogeny does not seem to influence FTSS does not mean that sediment yields are on average not much different in the two groups. Figure 66b shows that the river basins belonging to the young orogeny type have on average much greater slopes than the basins belonging to the old orogeny type. This should considerably influence the sediment delivery from these basins according to the empirical regression models established in this study.

4.4.4.3. Land Use

Finally, I considered also the percentage of cultivated area (CultA) as a general basin characteristic that could influence sediment yields. There has been much said about the possible effect of land use on river sediment fluxes, but in general it is difficult to quantify this effect at the global scale. Some authors argued that the delivery of river sediments to the oceans could have doubled in the last 200 yrs due to man's land use practices (e.g., Judson and Ritter [1964], Likens et al. [1981], Berner [1989]). In Table 18, only two groups are distinguished with respect to the criterion of CultA: basins which are affected by land use (CultA > 15%), and basins which are less affected by land use (CultA < 15%). It was not possible to form subgroups within the climate river groups. Basins with great CultA are mainly found in the temperate wet climate type, while the tropical wet basins are much less affected by land use, and there is nearly no land use in the basins of the tundra and taiga climate.

128

Table 18. Regression of sediment yield versus several parameter products. The rivers are grouped according to the percentage of cultivated area in the basins. Only river basins of the wet climatetypes are selected.

____________________________________________________________________________________________________________

Correlation Coefficient r Regression Coefficient m (f(x) = m x) Number ______________________________________ ______________________________________

River Group of Rivers P2 P3 P4 P5 P2 P3 P4 P5

CultA > 15% 19 0.88 0.86 0.81 0.73 15.71 0.0142 -- 0.141 CultA < 15% 22 0.86 0.97 0.97 0.95 30.56 0.0225 0.00829 0.202

____________________________________________________________________________________________________________

P2 = (Four x Slope) P3 = (Four x Slope x Q) P4 = (Four x Slope x Q x LithMI) P5 = (Slope x Q x LithMI) CultA is the percentage of cultivated area in the basins. All regressions are significant at least with P < 0.05. The regression coefficient was calculated by forcing the regression to pass throughthe origin. It is only shown when the intercept was not significant in the regressions, which means that for the intercept P > 0.1 (P > 0.05 for the values in italics).

The comparison of the regression coefficients in Table 18 do not indicate greater sediment yields in basins that are affected by land use compared to basins that are less affected by land use. The regression coefficients in the less-affected basins are even greater. Because of the uneven distribution of the different climates in the two groups, this result has naturally to be taken with more caution. In fact, the regressions in Table 18 mainly reflect a comparison of the basins of the tropical wet and the tundra and taiga climates on the one hand with the basins of the temperate wet climate on the other hand. But we have seen that in the latter climate type the parameter Four seem to be important than in the two other climate types. Moreover, bear in mind that the here presented statistics are made on the basis of wet climate rivers only. As it was discussed above, sediment yields in the dry climate rivers could be much more controlled by the density of the vegetation cover. For this reason, a perturbation of natural vegetation types through agricultural practices may essentially affect erosion rates in the dry climates. Finally, one has also to mention that the calculation of CultA comprises all types of land use, but certain practices could naturally have stronger effects on mechanical erosion than others.

4.5. A Global Map of River Sediment Yields from the Continents

The results of the previous sections indicated that present-day river sediment fluxes are mainly controlled by a combination of hydroclimatic, morphological, and lithological factors such as Q, Slope, Four, and LithMI. These results are in good agreement with the results of Phillips [1990] who found that slope gradient, runoff, and precipitation factors together should account for most of the global variation in soil erosion rates. It should be noted that the here presented models have a similar form as the Universal Soil Loss Equation (USLE) (e.g., Wischmeier et al. [1958]). Also the USLE includes rainfall intensity and slope as principal controlling factors, but additional parameters are needed for the USLE, including soil erodibility as a function of soil properties such as soil texture, and the density and structure of the vegetation cover. The USLE was originally designed for local scale assessments of soil loss by rainfall from agricultural land, but the regression models presented in this study indicate that similar approaches could be applied in order to predict the variability of sediment yields also at global and regional scales.

Figure 67a shows the result when equations 27.1, 27.2, 27.3, and 27.4 are applied to the overall continental area on the basis of the corresponding data sets. The so-created global sediment yield map depicts a regional variability which is in good agreement with field data. Greatest values are found for the regions along young orogenic belts, such as the Himalayas, the Andes, or the Alps, as well as for

129

the humid regions in the South of Asia and in Oceania. Note that the great sediment yields along the young orogenic belts can have a twofold cause: on the one hand these regions are characterized by steep slopes, and on the other hand drainage intensity can be great because the mountain chains form barriers for the atmospheric water circulation, leading to high precipitation. Low sediment yields are typical for the northernmost parts of the continents (north of 50° N), and naturally also for deserts. Global sediment yield maps have been proposed, for example, by Walling and Webb [1983], by Jansson [1988], or by Probst [1992]. These maps assigned FTSS values to the different regions of the continents on the basis of field data, and their resolutions are thus limited to the basin or sub-basin scale. Where no data were available, the values were normally assigned by subjective interpolation. Globally there is a good agreement between these maps and the here presented map in Figure 67a. This is especially the case for the map of Walling and Webb [1983], which is more detailed than, for example, the map of Probst [1992]). Jansson [1988] did not interpolate the values on her map between the basins she studied, but she noted that her map also agrees well with the one of Walling and Webb [1983].

Nevertheless, the outstanding sediment yields for the Huanghe and the Tana rivers manifest that the here presented modelling approach cannot account for all local particularities which are documented by field data. Another problem is the lower reliability of the regression models for the dry climate rivers compared to the wet climate rivers (see above). I created therefore a second sediment yield map by using the modelled values of Figure 67a as an interpolation matrix to distribute the observed sediment fluxes for the river basins of Table 12 over the continents. This means that within the borders of the river basins I corrected all grid point values according to:

FTSS (corrected) = FTSS (modelled) x (29) FTSS (basin average, observed) / FTSS (basin average, modelled)

For the grid points which do not fall into one of the basins, the values were obtained by a triangular interpolation between the next basin values. Also the interpolation was coupled to the modelled values according to:

FTSS (corrected) = FTSS (corrected, interpolated) x (30) FTSS (modelled) / FTSS (modelled, interpolated)

The so created map is shown in Figure 67b. I is not much different from the map in Figure 67a, but since the map respects perfectly the sediment yields of Table 12, now the great values that are found for the Huanghe and the Tana rivers are shown as well. I consider therefore this map to be the best representation of the global distribution of river sediment yields which can be made on the basis of the data used in this study. Locally, the values resulting from the applied proceeding could naturally change if additional basins would have been included into the interpolation procedure, but bear in mind that the here included river basins cover together more than 50% of the continental area that is drained to the oceans. Additional river basins could lead to some local changes but they should not affect much the regional and global patterns.

One has also to mention that both maps in Figure 67a and in Figure 67b tend to smooth local variability. One of the shortcomings of the applied modelling is that it cannot account for sediment storage in the basins. When there is storage in one part of the basin, the more must be eroded in another part of the basin to yield the average sediment flux observed at the river mouth. Another smoothing effect could be related to the limited grid point resolution of the data sets, which probably cuts off extremes by averaging the values. An idea of the possible extent of smoothing can be obtained by looking more in detail at the Amazon Basin, where sediment yields have also been studied at the sub-basin level. Gibbs [1967] and Meade et al. [1985] reported that at least 80% of the sediments transported by the Amazon River originate from the Andes. In both maps of Figure 67a and of Figure 67b, only about 60% of the total sediment load of the Amazon comes from regions with elevations > 500 m. These regions cover about 13% of the total basin area and represent mainly the Andes.

130

Table 19. Fluxes of water and of sediments from the continents to the oceans.

______________________________________________________________________________________________________________

Sediment Yield Sediment Flux __________________________ __________________________ Drainage Area a) Runoff b) Figure 67a Figure 67b Figure 67a Figure 67b (106 km2) (km3/yr) (t km-2 yr-1) (t km-2 yr-1) (Gt/yr) (Gt/yr) ______________________________________________________________________________________________________________

Polar, without ice 3.892 762 13 8 0.049 0.032 Tundra & Taiga 23.232 6930 32 28 0.733 0.646 Temperate Dry 9.635 729 115 249 1.111 2.403 Temperate Wet 16.918 7753 183 195 3.100 3.300 Tropical Dry 21.790 3101 181 207 3.943 4.520 Tropical Wet 24.919 22403 235 204 5.858 5.090

Desert 5.940 66 5 7 0.027 0.044

Total 106.326 41744 139 151 14.822 16.035

Africa 18.288 4120 33 53 0.610 0.973 Europe 9.564 3079 103 88 0.989 0.841

North America 23.020 7142 92 136 2.122 3.138 South America 17.732 11150 161 166 2.857 2.940

Asia 32.518 15318 247 244 8.032 7.930 Australia 4.476 773 46 46 0.207 0.205 Antarctis 0.728 162 7 10 0.005 0.007

Total 106.326 41744 139 151 14.822 16.035

Arctic Ocean 16.982 3239 18 14 0.313 0.235 North Atlantic 27.300 13484 109 132 2.984 3.600 South Atlantic 16.959 5074 35 31 0.599 0.523

Pacific 21.025 13532 293 352 6.165 7.407 Indian Ocean 16.594 5166 243 214 4.029 3.556 Mediterranean 6.739 1087 108 105 0.725 0.708

below 60° South 0.728 162 7 9 0.005 0.007

Total 106.326 41744 139 151 14.822 16.035 ______________________________________________________________________________________________________________

a) without endoreic and glaciated regions b) for the determination of the runoff values, see chapter III

Table 19 regionalizes the sediment fluxes from the continents to the oceans together with the corresponding water fluxes (which are taken from chapter III) with respect to major climates, different continents, and different ocean basins. Both the results from the pure modelling approach (Fig. 67a) and the results from the combined modelling/ interpolation approach (Fig. 67b) are listed. The comparison of the two approaches is interesting because it can help to identify the regions where the modelling deviates at the most from observations. Note, however, that also the interpolation approach is not without problems, since it has a tendency to extend local phenomenons to larger scales, which might not be true in all cases. The most striking difference between the corrected (Fig. 67b) and the non-corrected map (Fig. 67a) is the fact that in the corrected map the sediment contributions from the dry climate types increase considerably, while the contribution of the tropical wet climate is even decreasing. The tropical dry climate is now nearly of the same importance for the global sediment flux than the tropical wet climate, although the runoff contribution from the tropical dry climate is only less than 15% of the runoff contribution from the wet tropical climate. A part of the increased sediment contribution from the dry climates is naturally related to the fact that the Huanghe and the Tana rivers are included in the corrected map (both rivers belong to dry climate types), but the values also increased in other dry climate regions (e.g., in North America). This underlines the importance of additional investigations on the controls of river sediment fluxes especially in the climates.

Taking the values of the corrected map as the best estimates, global sediment delivery to the oceans is about 16 Gt/yr. The specific sediment export from Asia, the continent with the greatest average sediment yield, is about 5 times greater than from Africa or from Australia, the continents with the lowest average sediment yields. With respect to the different ocean basins, one can state that the Pacific and the Indian Ocean together receive about 68% of the global river sediment load

131

F

ig. 6

7a S

edim

ent y

ield

s fro

m th

e co

ntin

ents

as p

redi

cted

with

the

empi

rica

l rel

atio

nshi

ps fo

und

in th

is st

udy.

133

Fig

. 67b

As

Fig.

67a

, but

this

tim

e th

e m

odel

ling

was

cou

pled

with

an

inte

rpol

atio

n of

the

obse

rved

sed

imen

t yie

lds

for

the

60 r

iver

bas

ins

inve

stig

ated

in

this

stud

y (s

ee te

xt).

135

Milliman and Meade [1983] estimated that up to 70% of the world sediment transport may originate from southern Asia and the larger islands of Oceania. The map in Figure 67b agrees with the dominant role of these regions for the global budget, but since the mentioned 68% correspond to the complete Pacific and Indian Ocean basins, it is indicated that the importance of these regions is probably not as great as estimated by Milliman and Meade [1983]. It is also interesting to note from the values in Table 19 that the average sediment yields for the temperate wet, the temperate dry, the tropical wet, and the tropical dry climate types are quite similar (varying from about 200 to 250 t km-2 yr-1). Only for the other climate types, the values are considerably lower. This again indicates that climate alone is obviously not very meaningful in order to predict the variability of river sediment yields at the global scale.

4.6. Conclusions

The loads of suspended sediment transported by the rivers of the world are highly variable. This variability is not only found at spatial but also at temporal scales, making it difficult to determine reliable long-term averages for individual rivers. High-frequent sampling surveys over long periods are needed to establish average fluxes. In many cases the values found in the literature are rather order-of-magnitude estimates than precise values. This has to be kept in mind when using literature data compilations for empirical modelling approaches.

Up to now, several attempts have been made to relate the variability of river sediment yields to the variability of the environmental parameters characterizing the corresponding river basins. Some of these attempts were briefly reviewed on the previous pages. Some studies considered climatic parameters such as precipitation to be the dominant controlling factors, while others proposed morphological parameters such as basin elevation to be the most important. Although generalizations have to be made with caution, we have seen from the literature review that pure climatic models rather tend to overestimate the global sediment delivery to the oceans, while pure morphological models rather tend to underestimate the global sediment flux. Consequently, the data used in this study show that it is the combination of different factors which yields the most significant regression models. Sediment yields are best correlated by forming the products of a number of hydroclimatic, geomorphological, and lithological factors, that is Q, Slope, Four, and LithMI. The best correlated parameter combination varies when the rivers are grouped according to their average climatic situation, but it is always a combination of the above given parameters that yields greatest correlation coefficients. When these models are applied to the total continental area, they produce a map of the regional variability of river sediment yields that is in good agreement with field data. For the total sediment flux to the oceans, a value of 14.8 Gt/yr is calculated with the pure modelling approach, and a value of 16.0 Gt/yr is found when the modelling is combined with an interpolation of the observed sediment yields for the 60 river basins considered in this study.

The fact that the most significant regression model can vary for the different climate groups does not necessarily mean that the processes of mechanical erosion and the subsequent transport of the eroded materials out of the basins are different in these climates. One has to keep in mind that the average parameters that were determined and used in this study may only be indirectly related to the real processes that are responsible for the observed sediment fluxes. A good example for this can be the parameter Four, which is found to be of greater importance in the temperate wet climate than, for example, in the tropical wet climate. In the temperate wet climate, strong precipitation occurs mainly when the soils tend to be water limited, while in the tropical wet climate, maximum Four values rather coincide with soils that are at field capacity. The same Four value in the temperate wet climate can thus have a greater erosion potential than in the tropical wet climate. A strong seasonality of precipitation should mainly increase sediment yields when it leads to an additional overflow of the soils compared to a climatic situation that has the same annual characteristics, but a more uniform precipitation distribution over the year.

136

Moreover, an overflow of the soils when they have been dry before may also facilitate the detachment of particles from the soil complex. The latter effect can have a great importance for mechanical erosion rates and river sediment fluxes in dry climates. Although it is more difficult to establish relationships for the dry climate rivers because of the great environmental heterogeneity that is often typical for these basins, it is found that with the same controlling factors, the regression coefficients are greater in the dry climates than in the wet climates. This is indicating a much greater susceptibility of dry climate regions to mechanical erosion, which is in good agreement with observations. Another reason for a greater erobibility of the soils in these regions may be a less efficient soil protection by vegetation. Soil protection by vegetation is probably a threshold phenomenon. There may exist a certain vegetation cover density, below which erosion rapidly increases. Also other processes influencing mechanical erosion in dry climates may be threshold phenomenons, making it difficult to identify these processes in a study like this one.

Naturally the here-presented regression models cannot cover all aspects of the complex relationships controlling mechanical erosion rates and sediment fluxes in river basins. One shortcoming of the applied approach is the fact that it cannot account for sedimentation processes taking place in river basins related to basin subsidence or sediment trapping in internal reservoirs such as lakes. However, we have seen that a simple coupling of basin sedimentation to general basin characteristics such as basin area or basin hypsometry is not possible. This has been often proposed in the literature. Another shortcoming of the applied modelling is that it does not distinguish between the different types of erosion contributing to river sediment fluxes. There is little known about the general contributions of channel and gully erosion to total sediment fluxes, at least at global scales. Because the determined parameter averages refer to all grid points in the basins, it is possible that the approach in this work gives to much weight to sheet erosion.

Further studies should especially focus on river sediment fluxes in arid climates. In this study, dry climate rivers that are monoclimatic over most of their basin area are underrepresented, and additional data are needed to confirm the detected trends. This is important because it can considerably influence the global and regional budgets. Another point for further research activities could be the determination of additional parameters that better describe the seasonal patterns in the river basins. As soon as reliable water budget models will be available that can predict the variability of the water content in the soils over the year, it would be interesting, for example, to quantify the amount of strong precipitation only when the soils are water limited. This may be more meaningful than the here tested parameter Four. Finally also the role of lithology on river sediment fluxes should be further investigated with additional data. We have seen that omitting LithMI from the regressions leads also to quite high correlation coefficients, but the corresponding regression models predict considerably greater sediment fluxes at global and regional scales.

137

CHAPTER V

PREDICTING THE OCEANIC INPUT OF ORGANIC CARBON BY CONTINENTAL EROSION: FROM THE

BASIN SCALE TO THE GLOBAL SCALE

5.1. Introduction

The erosion of organic carbon from land to sea via rivers represents an important element in the global carbon cycle (Kempe [1979a], Degens et al. [1984]). Earlier estimates on the total amount of organic carbon transported by world rivers varied within one order of magnitude (for a review see Schlesinger and Mellack [1981]), but investigations during the last decade have improved our knowledge on fluvial carbon fluxes substantially (Degens [1982], Degens et al. [1983], [1985], [1987], [1988], [1991a], Kempe [1982], [1984], Kempe et al. [1993], Lewis and Saunders [1989], Richey et al. [1990]). Many of these studies are presented in chapter I. More recent estimates on the total amount of organic carbon transported by rivers range from 0.38 to 0.45 GtC/yr (gigatons of carbon per year), with about 0.20 to 0.22 GtC/yr being discharged as dissolved organic carbon (DOC; Meybeck [1982], [1993a], Spitzy and Leenheer [1991]), and 0.18 to 0.23 GtC/yr being discharged as particulate organic carbon (POC; Meybeck [1982], [1993a], Ittekkot [1988]). However, even if the global figures are more precise today, we still know little about the factors that control carbon fluxes. There exist a few studies that associated the variability of the fluxes with major climate types (Meybeck [1982], [1988], Thurman [1985]), but these studies are based exclusively on the selection of a few rivers under the assumption that each represents one climatic type, which is probably not true and may lead to errors.

After having established the major controls for the river fluxes of water and of sediments in the previous chapters (III and IV), it is now tested whether it is also possible to relate the variability of mean annual fluxes of dissolved and of particulate organic carbon to the environmental variability of the corresponding river basins. Data from numerous rivers that have been studied for their annual DOC and POC loads were collected from the literature, and the empirical relations existing between the carbon fluxes and various hydroclimatic, biological and geomorphological patterns characterizing the river basins are investigated. The main purpose is to determine the best possible regression models for the river carbon fluxes at the global scale in order to predict the specific fluxes for each site over the continents on the basis of the available controlling parameters. This allows not only the refinement of the global budget but also the coupling of erosion to biosphere and/ or ocean models in the scope of global change research. There is an increasing need for geochemical investigations to quantify river carbon fluxes world-wide (IGBP [1993], [1995]). It has been shown by Sarmiento and Sundquist [1992] that the global riverine carbon flux, even if it is small compared with the bulk fluxes between the atmosphere/ biosphere and atmosphere/ ocean interfaces, cannot be neglected if one wants to understand the fate of the anthropogenic released CO2.

In a further step, the here established models will then be combined with an empirical modelling for the global fluxes of inorganic carbon (Amiotte-Suchet [1995], Amiotte-Suchet and Probst [1993a, b], [1995]) in order to evaluate the role of continental erosion within the global carbon cycle. This will be the subject of the last chapter in this study, chapter VI.

138

The principal results that are presented in this chapter have been published in Ludwig et al. [1996]. Note that with respect to this publication, in the following all regressions and model outputs have been recalculated on the basis of the data sets used in this study (see chapter II). This concerns the global data sets for drainage intensity and for the organic carbon content in the soils, which are different compared to those of Ludwig et al. [1996]. Also this time the global sediment yield map presented in chapter VI was used, which is slightly different from the one presented in Ludwig et al. [1996].


5.2.1. River Fluxes of DOC and POC

For 32 of the rivers considered in this study, mean annual DOC and POC fluxes (FDOC and FPOC, respectively) could be determined. All fluxes were transferred into specific values by division with the corresponding basin areas. The resulting values are listed and referenced in Table 20. Note that they refer to the basin areas calculated in this study (see chapter II). For each river I determined the mean annual discharge weighted DOC and POC concentrations on the basis of the carbon fluxes and discharge values corresponding to the investigation periods. These average values were then multiplied with the long-term literature discharge estimates of chapter III to obtain the mean average fluxes.

As far as the river case studies are concerned which have been discussed in chapter I, the discharge-weighted annual mean concentrations were calculated according to:

n n Conc = ∑ (Conci x Qi) / ∑ (Qi) (31) i = 1 i = 1

Conc is the discharge-weighted annual mean DOC (or POC) concentration, Conci the instantaneous DOC (or POC) concentration, Qi the instantaneous discharge value, and n the number of samples analysed within the investigation period. Note that only data were taken into account that fall into one or several complete seasonal cycles. Most of the rivers were sampled in about monthly time intervals, and the resulting discharge-weighted means should be comparable. When there exist gaps of one month in the sampling series, I filled them by linear interpolation of the concentrations and/or the discharge values between the two closest data points.

For POC, one can also obtain an average flux by coupling the organic carbon content (POC%) in the total suspended solids to the long-term mean annual sediment flux (see chapter IV). In the here presented data, only the POC fluxes for the Ganges/ Brahmaputra, Indus, Chiangjiang and Huanghe rivers were calculated with this method, because no other data are available for these rivers. For the other rivers, I preferred an extrapolation with discharge, but I added in Table 20 also the discharge-weighted annual mean POC% values if both the POC and the TSS concentrations have been measured simultaneously. The POC% values correspond thus only to the observation period of the field studies. TSS-extrapolated POC fluxes can be obtained by a simple multiplication of the POC% values with the average sediment fluxes given in chapter IV. The sediment method yields normally greater values than the discharge method. A striking example is the Amazon River, where a POC% derived FPOC value yields a fourfold larger value than the discharge derived FPOC value (12.36 t km-2 yr-1 in comparison to 3.02 t km-2 yr-1). The tendency that the sediment method often yields greater values may be at least partly related to the fact that the long-term sediment fluxes refer to natural conditions (see chapter IV), whereas the field studies were carried out in recent years when certain rivers had considerably lower sediment loads due to damming of the rivers (e.g. the Nile or the Orange rivers). We will see in the

139

following that periods with relative low sediment loads may have greater mean annual POC% values than periods with relative high sediment loads.

Table 20. Mean annual fluxes of DOC and POC for some world rivers.

______________________________________________________________________________________________________________

River Abbreviation Dq FDOC, FPOC, POC%, DOC, POC, Source t km-2 yr-1 t km-2 yr-1 % mg/l mg/l

______________________________________________________________________________________________________________ Amazon Amz 3 4.760 3.020 6.1 4.5 2.8 Richey et al. [1990]

Zaire Zai 2 2.705 0.746 7.0 7.3 2.0 NKounkou and Probst [1987] Mississippi Mis 2 1.327 0.323 - 8.8 2.1 Leenheer [1982]

Ob Ob 1 1.215 0.118 - 9.1 0.9 Romankevitch and Artemyev [1985] Paraná Par 3 1.636 0.319 3.6 8.7 1.7 Depetris and Cascante [1985] Nile Nil 3 0.142 0.185 - 3.0 3.9 Abu el Ella [1993]

Changjiang Yts 1 6.314 6.820 2.5 12.4 13.4 Gan Wei-Bin et al. [1983] a Ganges/ Brahmaputra GaBra 1 2.851 6.726 - 3.9 9.1 Safiullah et al. [1987] b

Mackenzie Mac 3 0.824 0.841 3.0 4.9 5.0 Telang et al. [1991] Niger Nig 3 0.482 0.336 3.3 3.7 2.6 Martins and Probst [1991]

St. Lawrence SLa 3 1.514 0.303 6.0 3.8 0.8 Telang et al. [1991] Orinoco Ori 3 4.705 1.554 1.8 4.4 1.5 Lewis and Saunders [1989]

Indus Ind 3 3.788 2.320 0.5 14.4 8.8 Arain [1987] b Yukon Yuk 2 1.031 0.331 0.3 4.1 1.3 Telang et al. [1991]

Huanghe Hua 2 0.448 13.665 0.7 6.3 190.6 Zhang et al. [1992] Orange Org 3 0.038 0.016 2.2 2.5 1.1 Hart [1987]

Columbia Col 3 0.594 0.073 - 2.1 0.3 Dahm et al. [1981] Don Don 1 0.579 0.242 - 8.8 3.7 Romankevitch and Artemyev [1985]

Senegal Sen 3 - 0.102 - - 1.5 Degens et al. [1991b] Northern Dvina NDi 1 4.335 0.191 - 13.6 0.6 Romankevitch and Artemyev [1985]

Rhine Rhi 2 2.463 1.400 5.6 5.3 3.0 Eisma et al. [1982] Brazos Brz 3 0.202 0.233 0.8 3.7 4.3 Malcom and Durum [1976] Loire Loi 3 1.387 0.737 7.9 5.6 3.0 Meybeck et al. [1988] Rhône Rho 3 0.934 0.464 - 1.7 0.8 Kempe et al. [1991]

Garonne Gar 3 1.000 1.122 3.6 2.8 3.1 Etcheber, in preparation Po Po 1 2.176 0.926 - 3.1 1.3 Pettine et al. [1985]

Gambia Gam 3 0.317 0.142 2.4 2.4 1.1 Lesack et al. [1984] Rioni Rio 1 1.015 1.875 - 1.1 1.9 Romankevitch and Artemyev [1985] Tiber Tib 1 1.848 0.638 - 4.0 1.4 Pettine et al. [1985]

Severn Sev 3 2.557 - - 5.3 - Mantoura and Woodward [1983] Waikato Wai 3 5.865 1.450 10.1 5.5 1.4 ARA unpublished c

Ems Ems 1 3.284 1.051 7.0 8.1 2.6 Cadée [1987] ______________________________________________________________________________________________________________

All concentrations are discharge-weighted annual mean values Dq is an index for the data quality (see text): 3, high; 2, medium; 1, low a FPOC value from Milliman et al [1984] b FPOC value from Subramanian and Ittekkot [1991] c Auckland Regional Authority, New Zealand

One of the main constraints for a study such as this one are the differences in the quality of the river data. For a certain number of rivers, DOC and POC fluxes can be calculated on the basis of intensive field studies carried out during one or several years with short sampling time intervals. For others, the fluxes are only determined on the basis of few measurements, which possibly do not sufficiently reflect seasonal variability. Another problem are the different analytical techniques applied for the measurement of DOC and POC concentrations. For these reasons, I assigned to the rivers listed in Table 20 an index for the quality of the river data. A low data quality index can either ndicate less extensive field studies or that the measurements were done with a different analytical technique in comparison to the technique applied in most of the other studies. The latter is the case, for example, for the rivers of the former Soviet Union. The data quality index reflects the comparability of the data with regard to the bulk of the other studies and it denotes nothing about the quality of the studies itself.

Furthermore, keep in mind that in some cases the record periods coincided with unusual dry or wet years. Also this makes an extrapolation to average conditions difficult. I recall here from chapter I

140

Table 21. Average environmental characteristics for some world rivers. _____________________________________________________________________________________________________________

AT APPT Four VegC NPP ForR SoilC Elev Slope SoilH LithMI CultA PopD River (°C) (mm) (mm) (kg/m2) (kg/m2) (%) (kg/m3) (m) (radian) (cm) (%) (h/km2)_____________________________________________________________________________________________________________

Amazon 24.1 2228 235 11.7 0.669 68 11.2 455 0.0434 226 22.4 0.9 2 Zaire 23.6 1541 173 8.4 0.603 59 10.7 775 0.0370 310 9.6 3.5 15

Mississippi 11.0 895 86 3.1 0.487 19 12.3 646 0.0497 214 16.1 38.4 27 Ob -0.4 524 57 4.4 0.406 30 29.2 305 0.0345 216 26.0 12.6 7

Paraná 21.2 1223 135 3.3 0.424 19 11.3 503 0.0443 228 19.8 15.1 14 Yenisei -4.6 555 72 7.6 0.377 55 15.4 769 0.0855 93 9.6 2.7 2

Lena -8.5 445 59 6.0 0.318 50 13.6 608 0.0857 49 7.7 1.6 1 Amur -1.0 591 97 6.9 0.472 51 23.6 571 0.0699 137 11.6 7.5 29 Nile 24.2 816 111 2.8 0.454 23 10.8 892 0.0577 221 5.2 17.8 28

Changjiang 14.5 1164 143 5.3 0.457 31 12.8 1527 0.1321 196 12.3 25.5 205 Ganges/Brahm. 21.8 1516 269 3.7 0.477 22 12.1 1406 0.1353 153 20.8 37.9 193

Mackenzie -3.5 463 46 8.2 0.375 60 34.9 634 0.0964 184 8.5 0.2 1 Niger 27.1 1079 180 3.1 0.467 30 6.1 377 0.0303 294 11.9 9.5 43

Zambesi 21.3 930 163 5.4 0.522 45 7.4 994 0.0450 249 14.4 12.0 21 Murray 17.8 466 42 4.0 0.444 31 8.4 272 0.0337 161 15.8 20.5 6

St. Lawrence 5.5 1058 92 5.8 0.519 43 14.7 274 0.0407 210 4.6 24.1 25 Orinoco 25.6 2124 225 8.2 0.592 51 12.4 347 0.0549 228 23.8 2.1 13

Tigris/Euphrates 20.3 287 42 1.6 0.332 5 3.6 625 0.0933 200 15.9 23.2 46 Indus 19.8 593 98 2.2 0.331 11 5.3 1671 0.1476 92 20.9 19.6 121

Mekong 23.2 2225 322 8.1 0.609 55 11.1 854 0.0858 210 13.5 16.2 59 Yukon -4.2 502 55 4.2 0.251 35 19.7 748 0.1468 219 12.5 0.0 <1

Huanghe 8.7 501 78 2.0 0.367 8 11.0 1839 0.1196 179 22.5 29.8 157 Danube 9.4 889 81 4.1 0.577 28 15.5 514 0.1250 154 21.7 56.0 125 Orange 18.0 427 54 1.4 0.318 6 3.8 1256 0.0484 160 13.2 5.3 9

Colorado 13.3 429 44 4.2 0.322 24 4.3 1505 0.1718 176 11.2 2.1 6 Columbia 7.5 796 79 8.5 0.442 43 8.1 1336 0.2197 232 4.9 10.1 16 Kolyma -10.2 415 46 3.7 0.246 28 10.4 590 0.1195 24 12.0 0.0 <1

Sao Francisco 23.1 1132 156 4.0 0.468 32 6.5 627 0.0449 261 6.0 5.7 45 Si Kiang 20.7 1595 201 2.8 0.463 11 10.4 598 0.0805 198 9.6 30.4 199

Irrawaddy 23.1 1936 298 8.5 0.631 56 10.9 745 0.1477 242 22.9 15.3 39 Don 6.8 583 53 1.2 0.500 2 11.6 138 0.0147 234 22.0 73.2 37

Senegal 28.3 674 147 2.7 0.466 29 3.2 223 0.0216 133 14.3 10.5 6 Indagirka -14.5 342 44 3.6 0.241 27 10.7 738 0.1261 37 14.8 0.0 <1 Limpopo 20.2 738 101 3.2 0.434 30 6.5 795 0.0498 189 10.2 14.0 11

Northern Dvina 0.6 738 73 8.1 0.466 68 24.9 127 0.0141 342 19.3 0.5 1 Godavari 26.6 1149 216 3.8 0.568 30 11.6 399 0.0433 154 4.8 44.2 260

Magdalena 23.4 2225 219 5.3 0.548 38 11.0 1224 0.2527 194 16.5 7.5 21 Fraser 5.5 953 90 14.2 0.585 81 9.2 1173 0.2555 178 7.4 2.4 1 Yana -14.6 310 48 3.5 0.235 25 10.1 746 0.1290 30 11.6 0.0 <1

Mahandi 26.5 1563 295 4.7 0.590 29 8.3 318 0.0497 147 7.0 50.0 412 Liao He 6.3 635 111 3.6 0.503 23 17.3 488 0.0698 166 16.3 45.5 203 Rufiji 21.8 1061 168 4.2 0.496 35 9.3 864 0.0963 248 10.0 12.3 52

Rio Negro (Arg.) 12.5 370 40 1.7 0.264 9 3.9 697 0.0773 178 9.3 4.2 14 Hungho 21.1 1843 263 7.7 0.605 56 10.7 897 0.1315 306 11.0 13.3 133 Rhine 8.1 1215 105 5.3 0.539 33 17.8 579 0.1372 189 18.3 44.6 190 Brazos 18.1 795 74 2.5 0.471 15 12.3 438 0.0255 256 18.2 24.7 24 Loire 10.9 969 83 3.3 0.569 23 15.0 306 0.0667 161 8.7 50.8 135 Rhône 10.2 1131 99 3.7 0.513 30 14.4 805 0.2443 127 8.8 33.2 103 Tana 24.5 525 77 4.7 0.473 41 6.1 403 0.0383 316 1.1 11.1 31

Garonne 11.5 1079 93 4.7 0.540 31 13.7 483 0.1211 139 5.8 36.0 71 Po 11.8 1049 95 3.5 0.492 24 14.5 501 0.1903 161 21.2 53.1 167

Gambia 26.6 1067 214 2.3 0.501 22 5.5 223 0.0336 172 15.5 23.7 9 Fly 26.2 3455 327 12.6 0.512 60 11.6 198 0.0714 234 32.0 0.0 9

Susitna -0.5 711 76 4.1 0.226 29 17.4 790 0.1899 186 18.3 0.0 1 Purari 22.5 3402 294 15.1 0.706 75 18.4 1176 0.2276 177 20.4 0.0 22 Tiber 13.1 1153 108 4.4 0.533 39 17.7 437 0.2542 111 9.5 0.0 62 Rioni 8.3 1410 121 2.0 0.486 9 10.0 1421 0.3093 23 13.2 28.3 26

Severn 9.4 981 84 1.0 0.600 0 16.7 153 0.0584 165 9.3 100.0 389 Waikato 11.7 1837 157 7.4 0.590 59 15.9 433 0.0922 140 10.0 0.0 3

Ems 8.6 1043 89 1.6 0.580 6 27.3 113 0.0391 213 24.1 80.0 169 _____________________________________________________________________________________________________________ AT, mean annual temperature; APPT, mean annual precipitation total; Four, modified Fournier-index; VegC, average biomass density; NPP, net primary production; ForR, mean forest ratio; SoilC, average organic carbon content in the soils; Elev, mean modal basin elevation; Slope, average basin slope; SoilH, average soil depth, LithMI, index for the erodibility of the abundant basin lithology with respect to mechanical erosion (range: 1 to 40); CultA, percentage of the cultivated area in the basins; PopD, mean population density in the basins. For details on the parameters, see chapter II.

141

that, for example, the relative high average DOC concentration which is found for the Paraná River should be related to the exceptional flooding that occurred during the investigation period of this river (Depetris and Kempe [1993]). An example where the here determined average DOC concentration may be too low with respect to the real value may be the Mackenzie River, because in this study the high-water period was underrepresented. However, there are no objective means to correct such effects, and one has to suppose that they average out over the number of rivers which are considered in this study.

5.2.2. Environmental Data Sets and Empirical Modelling

Table 21 shows some of the climatic, biological, geomorphological, lithological and anthropogenic characteristics that were calculated for the 60 river basins investigated in this study. All data sets that were used to determine the average river basin characteristics are described in chapter II, where also details on the applied statistical procedures and other technics are given. The runoff and sediment yield data sets are described in chapter III and in chapter IV, respectively. All climatic distinctions follow the classification established in chapter II.

5.3. Factors Controlling DOC Fluxes

Figures 68a and 68b show that the best single variable model to predict the fluxes of dissolved organic carbon is a model based upon discharge (r = 0.67 for all rivers). Other variables showing a high correlation with FDOC are APPT and VegC, but this does not necessarily indicate a causal relationship because multicollinearity exists especially between the hydroclimatic and biological parameters (see also chapter V). The linear multiple correlation statistics reveal that the model performance is significantly improved if one introduces Slope and SoilC as additional parameters. On the basis of a set of 29 rivers, the following equation is the best model to estimate DOC fluxes:

FDOC = 0.0044 Q - 8.49 Slope + 0.0581 SoilC (32)

n = 29, r = 0.91, P < 0.0001 for Q, Slope, and P < 0.01 for SoilC

The units of the parameters are as follows: FDOC = t km-2 yr-1; Q = mm; Slope = radian; and SoilC = kg/m3; r is the correlation coefficient, P is the significance level, and n is the number of rivers considered in the equation. The two rivers of Table 20 that are not taken into consideration in equation 32 are the Indus and the Changjiang rivers. This is discussed below. Results are similar if the regressions are calculated by omitting the rivers with lower data quality. If one takes only the group of 17 rivers with the highest data quality index, the best model remains still a model based on Q, Slope, and SoilC, and it shows very similar regression coefficients to those of equation 32. The correlation coefficient rises up to a value of r = 0.98, while the influence of Q in the model increases (see also Fig. 68b). Figure 69 shows the corresponding plot of predicted versus measured DOC fluxes for the 29 river considered in equation 32.

The results of the regression analyses indicate that drainage intensity, the steepness of morphology, and the amount of organic carbon in the soils are the main factors which control DOC fluxes globally. I did not find any influence of basin area on DOC fluxes. The model may thus be applicable also to regional and local scales. DOC fluxes become greater with increasing drainage intensities, flatter morphologies, and larger carbon reservoirs in the soils. This suggests that soils are globally the major contributors to riverine DOC, which is in good agreement with most of the studies on this topic (e.g., mentioned by Spitzy and Leenheer [1991], Degens et al. [1991b], Kempe and Depetris [1992]), as well as with the findings of chapter I. However, the importance of basin morphology as potential controlling factor for organic carbon fluxes has only recently been recognised

142

elsewhere (Rasmussen et al. [1989], Eckhardt and Moore [1990], Clair et al. [1994]). For example, Clair et al. [1994] reported for a group of small watersheds in Canada that the export of organic carbon is inversely correlated with basin slope (one has to mention that they do not distinguish between DOC and POC fluxes in their models, but since the investigated rivers are poor in POC, their values represent mostly DOC).

F DOC

F POC

-

AT

Q APPT

+0.52

+0.85

+0.57

+0.75

N P P

S oilC

V e gC

- 0.57 P opDC u ltA

+0.75

E le v

S lop e

S o ilH

+0.53

(a)

+0.60+0.67-

- 0.74

-+0.64

- 0.64

-+0.56

Fig. 68a Correlation between fluxes of dissolved organic carbon (FDOC, upper boxes and bold lines), fluxes of particulate organic carbon (FPOC, lower boxes and bold lines), and the environmental patterns characterizing the river basins (fine boxes and fine lines) for all rivers of Table 20. Only correlation coefficients < - 0.5 and > + 0.5 are depicted. See Table 21 for parameter abbreviations.

Fig. 68b As Fig. 68a, but this timeonly calculated on the basis of therivers of Table 20 with the highestdata quality index.

F DOC

F POC

+0.59

AT

Q APPT

+0.62

+0.90

+0.73

+0.80

N P P

S oilC

V e gC

- 0.51 P opDC u ltA

+0.93

E le v

S lop e

S o ilH

+0.59

(b)

+0.68

- 0.67

+0.75+0.88+0.71

+0.52-

+0.51+0.56

- 0.76

An explanation for the great importance of Slope on DOC fluxes could be that a steep basin morphology results in a higher share of surface runoff with respect to total runoff (Probst and Sigha [1989]). Because of the restricted contact with the soils, these waters should be less concentrated in DOC (Kempe [1979b]). On the other hand, the water in a basin with a flat morphology may have a longer average residence time than the water in a basin with a steep morphology. Leaching processes can therefore be more intense, which should enrich the soil and runoff waters in DOC. Related to morphology is also a more likely occurrence of wetlands in flatter basins. Rasmussen et al. [1989] and Eckhardt and Moore [1990] proposed that this may be important in controlling organic carbon fluxes because peats and wetlands have organic rich soils. This is in good agreement with field data for large rivers of the tundra and taiga climate, which display generally elevated DOC concentrations (Romankevitch and Artemyev [1985]). The potential for peat formation is high in these regions (Lottes and Ziegler [1994]). It is also in agreement with the fact that the waters draining wetlands in small catchments are usually found to be very rich in DOC (Moore and Jackson [1989] - see also chapter I).

143

0 1 2 3 4 5 6 7

0

1

2

3

4

5

6

7

Tropical Wet

Temperate Wet

Tropical Dry

Tundra & Taiga

Temperate Dry

PredictedDOC Flux

Observed DOC Flux

Fig. 69 Comparison ofobserved DOC fluxes (t km-2

yr-1) with the DOC fluxespredicted in this study for the29 rivers of Table 20 (see text).The rivers are groupedaccording to their averageclimatic situation (see chapterII).

In Table 20, the Rioni and the Rhône rivers in Europe display the lowest average DOC concentrations (1.1 and 1.7 mg/l, respectively). They have low to moderate values of SoilC, but feature, together with the Tiber, the steepest basin morphologies of all rivers (Table 21). The only slightly higher concentrations of the Gambia and the Orange rivers (2.4 and 2.5 mg/l, respectively) can be related to their low SoilC values, while the high concentration of the Zaire River of 7.3 mg/l is an example for the combination of a low relief together with a moderate soil carbon pool. Nevertheless, it is important to point out that discharge is the major factor that controls the fluxes of dissolved organic carbon on a global scale, as it has been previously pointed out by Esser and Kohlmaier [1991]. This is not surprising because Q is globally more variable than DOC concentration. The value for the average DOC concentration in Table 20 lies between 1.1 and 14.4 mg/l, with more than 50% of the values lying in the range of 3-8 mg/l. Q is scattering over a much wider range, that is from 15 mm/yr (Orange) to 1070 mm/yr (Orinoco and Waikato - see chapter III).

As mentioned above, the Indus and the Changjiang rivers do not fit with the regression model of equation 32. Both rivers have high average DOC concentrations, but their SoilC values are low to intermediate, and their basin morphologies are steep. It is remarkable that both river basins are characterized by a great climatic variability (see chapter II). Note that more than 40% of the Indus watershed is desert. This is certainly responsible for the low average SoilC value found for the Indus, but on the other hand this region should have practically no importance for its DOC export. In the parts of the basin from which discharge is mainly derived, SoilC values can be expected to be much greater. Furthermore, the high DOC concentration in the Indus can also be related to the extreme seasonal variability of the discharge patterns in this river. I recall here from chapter I that among all discussed rivers the Indus shows the strongest flushing effect, that is a rapid increase of DOC concentrations on the rising limb of the hydrograph. This may also play an important role for the average annual DOC export.

There is, however, also another important reason that can explain why the Indus does not fit in the regression: It was shown in chapter I that the Indus looses probably up to 50% of its water from upstream to downstream, and it is likely that this exerts also a concentration effect on the dissolved solids in the river water. The mean annual DOC concentration in the Indus may thus be enhanced by a factor of two through evaporation. Such an effect that can naturally not be taken into account in this study.

144

One can also mention here that the high concentrations of both the Indus and the Changjiang rivers can be partly the result of a strong anthropogenic input. This may be especially true for the Changjiang River, around which the population density is very great (Table 21). Nevertheless, it is likely that there is also a methodological problem being responsible for the high average DOC concentration found for the Changjiang River. Note that most of the here presented rivers were analysed for DOC at the University of Hamburg in the framework of the SCOPE program (see chapter I). But DOC analyses for the Changjiang River were done at Tanjin University (Carbon Cycle Research Unit [1982], Gan Wei-Bin et al. [1983]). At the same time also samples of the Huanghe River were analysed at Tanjin University. It is remarkable that the analyses for the Huanghe yielded mean DOC concentrations as twice as high as the mean concentrations found in a later study by Zhang et al. [1992], who analysed the samples at the University of Hamburg (I use their results for the Huanghe in this study). Either the Huanghe reduced its DOC flux with about 50% within 5 years, or the analytical techniques do not yield the same values.

5.4. Factors Controlling POC Fluxes

For particulate organic carbon, the variability of the specific fluxes is greater than for dissolved organic carbon. No clear correlation with one or several of the climatic, biological, geomorphological, or lithological parameters can be detected for the set of all river basins in Table 20. Figure 68a suggests that there is a close relationship of FPOC with Elev, but this effect is caused by only a few rivers in the data set. These rivers are the Ganges/ Brahmaputra, Indus, Changjiang, and Huanghe rivers, which are all situated in the south and south-east of Asia. Taking again only the group of the 17 rivers with the highest data quality index, the results are completely different (Fig. 68b). Here, the best correlation for FPOC occurs with Q, and the correlation coefficient with Elev is almost zero. However, it would be misleading to predict global POC fluxes by coupling them to discharge only, as Esser and Kohlmaier [1991] proposed. We have seen in chapter IV that much the world's sediment transport to the oceans occurs from southern Asia and the larger islands of Oceania. One risks to bias the results if one excludes the rivers from this region from the regressions.

The here-presented data suggest that among all potential controlling factors tested in this study, it is the total sediment flux (FTSS) that shows the most significant relationship with FPOC on a global scale. Figure 70 shows the nature of this relationship. The relation of FPOC and FTSS becomes the most evident if one looks only at simultaneously measured fluxes without extrapolating them to long-term averages because extrapolation problems can arise for the determination of reliable average FPOC and FTSS that correspond to each other (if there exists a causal relationship between both parameters, it is problematic to linearly extrapolate independent observations for both parameters to long-term averages, e.g. by discharge). This reduces to some extent the number of data because not for all given POC fluxes in Table 20 it is not known to which FTSS values the observations correspond. In Figure 70 only the 19 rivers are shown for which both FPOC and FTSS have been measured simultaneously. In spite of the large scatter, one can see that the organic carbon content in the suspended solids clearly decreases with increasing TSS concentrations (cTSS) following a non-linear relationship. This is the case both for the individual measurements and for the annual means, suggesting that this relationship is similar for seasonal and spatial scales. However, we have seen in chapter I that seasonally the POC% - TSS relationship may be also influenced by resuspension of bedload particles during peak discharges. The following equation, which is the best mathematical fit to describe the relationship in Figure 70, is therefore only based on the annual means:

POC% = - 0.160 log(cTSS 3) + 2.83 log(cTSS 2) - 13.6 log(cTSS) + 20.3 (33)

n = 19, r = 0.83, P < 0.001

145

The unit of cTSS is mg/l. Other authors have found similar relationships between POC% and cTSS, both for seasonal (Probst [1992]) and for mean annual values (Meybeck [1982], Ittekkot [1988]). The equation 33 has a minimum for a TSS concentration of 2250 mg/l, corresponding to a POC% value of 0.5. This is in good agreement with the observation that POC% in TSS rarely falls below this level in rivers (Meybeck [1993a,b]).

1 10 100 1000 10000

0

5

10

15

20

25

30

35

40

a

bcd

f g

h jk lm op qrs

e

ni

a Waikatob Loirec Emsd Zairee Amazonf St. Lawrenceg Rhineh Garonnei Nigerj Mackenzie

k Paranal Chiangjiangm Gambian Orangeo Orinocop Brazosq Huangher Induss Yukon

Fig. 70 Plot of POC% (% TSS, y-axis) versus TSS concentration (mg/l, x-axis) for the 19 rivers of Table 20(see text). The circles represent the annual means of the shown rivers, while the dots represent individual measurements from the rivers: a, f, h, i, j, k, n, p, q, r.

There are two different processes that can account for the observed POC% - cTSS relationship. First, the decreasing POC% values with increasing TSS concentrations may reflect the variable contribution of the autochthonous carbon produced by riverine phytoplankton in the different rivers. Great sediment loads restrict the in-situ production in river water because of a reduced availability of light. This would mean that elevated POC% values result as an offset of the riverine carbon production to the carbon mobilized from the soil and vegetation pools. The fluxes of the allochthonous carbon are then probably more or less linearly coupled to the intensity of the mechanical erosion in the river basins (this is, of course, not true to the extent to which there occurs sediment storage in the basin). Second, the inverse POC% - cTSS relationship may result from an increasing dilution of riverine POC with mineral matter in highly turbid rivers caused by differences in the processes controlling the mechanical erosion in the drainage basins. In the basins of highly turbid rivers, gully and landslide erosion may be more active than linear erosion, leading to a greater share of mineral matter in TSS. If the first process is dominant globally, then POC export from the terrestrial biosphere would be lower than the input to the oceans. If the second process is dominant, then riverine POC should originate mostly from the terrestrial pools and the oceanic input should equal the terrestrial output, not considering, of course, possible losses of POC by oxidation during river transport and/ or in estuaries.

In the here presented, data the average POC% values vary between 0.3% and 10.1%. Values above 1.5% are only observed in rivers with TSS concentrations lower than 300 mg/l. The only exception is the Changjiang River. This river was sampled in its outer estuary, and the calculated

146

fluxes are probably influenced by the primary production in the estuary itself (Milliman et al. [1984]). Cauwet and Mackenzie [1993] have shown in a later study on the Changjiang River that primary production in the estuary leads to considerably greater POC% values in the suspended matter. The greatest POC% value in the data is observed for the Waikato River in New Zealand (10.1%). Other rivers with elevated values are the Loire River (7.9%), the Zaire River (7.0%), and the Ems River (7.0%). It is possible that these rivers have a considerable autochthonous share in their POC fluxes. On the basis of chlorophyll measurements, Meybeck et al. [1988] found for the Loire an autochthonous POC contribution of about 60%, and Kempe [1982] showed that in summer low pCO2 values (partial pressure of CO2) indicate strong autochthonous production. Also in chapter I, we have seen for the Waikato River that an enhancement of organic matter in the river waters in summer is possible (although this was mainly found for DOC).

However, most of the rivers have relative low POC% values between 0.5% and 5.0%. When looking in chapter I at the seasonal patterns of riverine POC, few indications for considerable contributions of autochthonous POC were found, mainly because in many cases much of the annual FPOC occurred during peak discharges, that is under hydrological conditions which do not allow the development of in-situ production in the river. Compositional characterization of particulate organic matter in rivers is scarce, but the few existing investigations confirm the dominance of allochthonous material in riverine POC. For example, Hedges et al. [1986] showed for the Amazon River that fine POC which forms the bulk fraction in the total POC (Richey et al. [1990]) derives primarily from soils, while coarse POC is composed of tree leaf debris and wood which is typical for litter. Lewis and Saunders [1989] determined for the Orinoco an autochthonous contribution of only 2% of the total POC flux (see chapter I). These are also indications that on a global scale autochthonous POC apparently plays a minor role only.

It is a striking feature that there is a similarity between the shape of the curve in Figure 70 and the typical profiles of the vertical organic carbon distribution in soils. Generally, total carbon density rapidly decreases with increasing depth, especially at high temperature (Desjardins et al. [1991]). Figure 70 may thus basically reflect the variability of mechanical erosion for different seasons and different regions. Low mechanical erosion intensity may mainly erode the uppermost soil horizon, and the POC% values in the river sediments should be characterized by elevated values. More intense mechanical erosion may also cut into deeper horizons, leading to lower POC% values in the mobilized material. Seasonally, the decreasing POC% values with increasing TSS concentrations can be also explained by the remobilization of mineral matter from the river bed during the rising hydrograph. In Table 21, SoilC values vary between 3.2 kg/m3 for the Senegal Basin and 34.9 kg/m3 for the Mackenzie Basin. Assuming that the soil has a mean density of 1.6 g/cm3 (Carvalho [1988]), the resulting percentage of organic carbon lies between 0.2% and 2.2%. These figures do not take into account the carbon stored in litter on top of the soil profile. Nevertheless, they are very close to the POC% values in river suspensions, in as much if one takes into account that erosion mainly occurs at the surface soil horizons where the organic carbon density is much greater and the litter pool acts as an additional source for POC. Cumulative carbon storage down to 20 cm can account for 50% of the total soil carbon (Zinke et al. [1986]). Moreover, we have seen in chapter I that most of the investigated rivers have C/N ratios close to the C/N averages found in soils (with maximum values that indicate litter contribution). Also this suggests that the observed POC% - cTSS relationship in Figure 70 principally results from the dilution of POC with mineral matter in turbid rivers, and that the bulk of riverine POC is derived from soil erosion.

There remains some uncertainty about the amount of fossil carbon in riverine POC. Meybeck [1993a, b] pointed out that the POC% value of 0.5 which seems to be a lower limit for organic carbon in river sediments is close to the organic carbon content commonly found in shales. Therefore he concluded that an important part of the particulate organic matter in rivers may be of fossil origin, and he estimated this flux globally to be about 0.08 GtC/yr (Meybeck [1993b]). This consideration underlies the assumption that there is no oxidation of rock organic matter before it is eroded. Net oxidation of rock organic matter has been estimated to range from 0.03 to 0.26 GtC/yr (Kramer [1994]). Selecting an intermediate value of 0.1 GtC/yr (Sarmiento and Sundquist [1992]), one finds

147

that net oxidation is at least in the same range as Meybeck's estimated fluvial transport of fossil POC. Since the organic matter oxidation is strongly related to rock granulation during soil formation (Kramer [1994]), one can assume that it should normally precede the erosion of the material. I consider therefore fossil POC to be of very small importance in riverine POC.

5.5. Extrapolation to the Global Scale

In order to establish global and regional budgets for the export of organic carbon to the oceans, the above discussed equations 32 and 33 were applied to the total continental area on the basis of the corresponding data sets (see chapters II, III, and IV). For POC, the average cTSS value was calculated for each grid element by combining the Q and FTSS data sets. Then the POC% values resulting from equation 33 were multiplied with the corresponding FTSS values (note that I used the sediment yield data set that was created by the combination of the modelling and interpolation method in chapter IV). According to the minimum in equation 33, all POC% values were limited to a minimum value of 0.5%. When calculating DOC fluxes, negative values which can result from equation 32 were set to zero. This occurred, however, only for a few grid elements and did not influence the global results much. The resulting maps for the specific DOC and POC fluxes on the continents are shown in Figure 71 and Figure 72, respectively. In Table 22, the modelled carbon fluxes are further detailed for the different continents, the different ocean basins, and for the major climate zones according to the climatic classification throughout applied in this study. The corresponding runoff and TSS fluxes are listed in Table 22 as well. Note that, as the case for FTSS (see chapter IV), also the FPOC values represent natural values, and do not reflect artificial reservoir retention. Present-day FPOC may be considerably lower for certain rivers because of the damming of these rivers.

With the here presented models, it is found that 0.363 Gt of organic carbon are transported to the oceans every year. About 57% of this carbon enters the oceans in dissolved form, and about 43% in particulate form. These figures are very close to previous estimates. For example, Meybeck [1993a] estimated annual transports of 0.198 GtC for DOC and of 0.170 GtC for POC, which is nearly identical with the global values in Table 22. On a regional scale, however, large differences occur with respect to Meybeck's figures. He estimated that about 65% of the global DOC flux originates from the wet tropics (Meybeck [1982], [1988], [1993b]), while this is only about 50% with the here presented values. Meybeck obtained his estimates by selecting one river as fully representative of one climatic type. The average concentrations for the selected rivers were then multiplied by an estimated runoff value attributed to each climate. As shown in chapter II, this is problematic because big river systems rarely fall exclusively in one climate zone. Moreover, the reported concentrations for rivers from similar climates often vary considerably, and selecting only one river is always subjective. For example, Meybeck [1988] considered an average DOC concentration of 8 mg/l to be representative for the wet tropics. This is exclusively based upon the elevated value found for the Zaire River. The values of the Amazon and the Orinoco rivers are both below 4.5 mg/l (Tab. 20).

It is interesting to compare the average concentrations that Meybeck attributed to different climates with the mean concentrations resulting from the water and organic carbon fluxes calculated in this study. His classification is not identical with the one I applied, but both classifications are comparable in the distinction of the major climates represented on Earth. Meybeck [1988] assumed low DOC concentrations for the semiarid (1.0 mg/l) and dry tropical regions (3.0 mg/l), intermediate to elevated DOC concentrations for the temperate (4.0 mg/l) and the tundra and taiga (6.0 mg/l) zones, and high DOC concentrations for the wet tropics (8.0 mg/l). His assignment is in general agreement with the here presented values both for the temperate and for the tundra and taiga climates. For the other climates, however, this is not the case. Going in Table 22 from dry to wet climates, DOC concentrations tend to decrease, and for the wet tropics one of the lowest concentrations of all climates is found. This is not a surprising feature. Neglecting the morphological effects on DOC fluxes, which can be supposed to average out over the major climates, the variation of average DOC concentrations

148

is principally a function of the variation of drainage intensity together with the variation of the soil carbon pool. Since on a global scale the latter varies less than drainage intensity does, it is mainly drainage intensity that determines not only specific DOC fluxes but also average DOC concentrations.

Table 22. Estimated fluxes of water, sediments, DOC, and POC to the oceans.

______________________________________________________________________________________________________________ Area Q FTSS FDOC FPOC DOC POC DOC/POC (103 km2) (109 m3/yr) (Gt/yr) (TgC/yr) (TgC/yr) (mg/l) (mg/l)

______________________________________________________________________________________________________________ Polar 3892 762 0.03 3.12 0.99 4.1 1.3 3.2 Tundra and Taiga 23232 6930 0.65 45.88 15.02 6.6 2.2 3.1 Temperate Dry 9635 729 2.40 2.90 12.51 4.0 17.2 0.2 Temperate Wet 16918 7753 3.30 35.37 32.69 4.6 4.2 1.1 Tropical Dry 21790 3101 4.52 15.96 27.82 5.2 9.0 0.6 Tropical Wet 24919 22403 5.09 101.72 68.51 4.5 3.1 1.5 Desert 5940 66 0.04 0.29 0.34 4.3 5.1 0.9 Total 106326 41744 16.03 205.24 157.88 4.9 3.8 1.3

Africa 18288 4120 0.97 19.97 10.72 4.9 2.6 1.9 Europe 9564 3079 0.84 16.60 10.52 5.4 3.4 1.6 North America 23020 7142 3.14 39.37 27.59 5.5 3.9 1.4 South America 17732 11150 2.94 51.83 34.31 4.7 3.1 1.5 Asia 32518 15318 7.93 73.00 71.55 4.8 4.7 1.0 Australia 4476 773 0.21 3.86 3.03 5.0 3.9 1.3 Antarctis 728 162 0.01 0.61 0.16 3.8 1.0 3.8 Total 106326 41744 16.03 205.24 157.88 4.9 3.8 1.3

Arctic Ocean 16982 3239 0.23 24.90 6.21 7.7 1.9 4.0 North Atlantic 27300 13484 3.60 71.02 39.33 5.3 2.9 1.8 South Atlantic 16959 5074 0.52 25.74 9.43 5.1 1.9 2.7 Pacific 21025 13532 7.41 57.01 68.58 4.2 5.1 0.8 Indian Ocean 16594 5166 3.56 21.35 28.10 4.2 5.4 0.8 Mediterranean 6739 1087 0.71 4.61 6.05 4.2 5.6 0.8 below 60° South 728 162 0.01 0.61 0.16 3.8 1.0 3.8 Total 106326 41744 16.03 205.24 157.88 4.9 3.8 1.3

______________________________________________________________________________________________________________ Calculated without the endoreic regions of the continents and the regions that are under permanent ice cover TgC = tera grams of carbon (1012 gC) The fluxes of water and of sediments are taken from chapter III and chapter IV, respectively

The climate type where the amount of organic carbon stored in the soils has the greatest effects on the average DOC concentration is the tundra and taiga climate type. On average, SoilC values are generally elevated in this type compared to the other climates types (Fig. 73). This has also been pointed out in many other studies on this parameter at the global scale (e.g., Post et al. [1982], [1985], Zinke et al. [1986], Post [1993]). Elevated SoilC is leading in Table 22 to the highest average DOC concentration for the tundra and taiga climate type among all climates. At the same time, fluxes of POC are relative low in this climate, and about 75% of total organic carbon (TOC) is eroded in the dissolved form. It is interesting to compare the values of Table 22 with the results of Clair et al. [1994], who investigated TOC fluxes on the basis of data for small to medium size Canadian rivers. These rivers represent a climate type that can be compared with the tundra and taiga climate type of this study. Clair et al. [1994] found that TOC fluxes were positively correlated with precipitation and negatively correlated with basin slope. With their model they extrapolated a median organic carbon export rate (DOC + POC) of 2.7 t km-2 yr-1 for the boreal climate. The here presented DOC and POC models yield a figure of 2.6 t km-2 yr-1 for the tundra and taiga climate.

For POC concentrations, the trend is similar than for DOC concentrations, but the general increase of POC concentrations in dry climate types is more pronounced than for DOC. This can be explained by the relative high mechanical erosion rates observed in these zones. For all dry climate types, the DOC/POC ratio is clearly below 1, while the global average is 1.3. Keep in mind that all

149

F

ig. 7

1 S

peci

fic D

OC

flux

es o

n th

e co

ntin

ents

as p

redi

cted

with

the

empi

rica

l rel

atio

nshi

ps fo

und

in th

is st

udy

(end

orei

c ba

sins

and

gla

ciat

ed re

gion

s ar

e om

itted

).

151

Fig

. 72

Spe

cific

PO

C fl

uxes

on

the

cont

inen

ts a

s pre

dict

ed w

ith th

e em

piri

cal r

elat

ions

hips

foun

d in

this

stud

y (e

ndor

eic

basi

ns a

nd g

laci

ated

regi

ons

are

omitt

ed).

153

concentration values given in Table 22 are calculated as the sum of the fluxes of all grid elements falling into one of the climatic types, divided by the corresponding runoff of these grid elements. It would be misleading to associate the values to individual rivers belonging to one of the climates. As discussed above, both for DOC and POC fluxes other factors such as the morphology of the river basins play an important role. Another important point is that especially for rivers basins belonging to dry climates, the climatic variability can be great and an average climate type is not always very meaningful to characterize the basin (see chapter II).

1.6 12.3 17.5--

4.7 4.6 22.9 30.1 12.5

2.1 3.6 5.4 14.8 7.9 14.4

2.2 2.6 4.4 5.0 5.8 7.5 -- 1.2 1.9 3.7 4.6 6.2 8.1 42.9

0.7 1.6 1.8 4.8 5.0 11.3 17.6 --

7.6

1.2ice

no ice

Standard Deviation

9.8 12.8 14.1--

6.5 9.5 20.5 28.8 13.3

1.9 3.0 6.9 16.2 13.3 13.7

2.5 2.3 4.3 8.2 11.1 11.3 --1.5 2.2 3.6 7.9 10.6 13.0 41.7

1.3 1.6 2.3 6.1 8.8 13.3 14.5 --

6.6

0.1ice

no ice

SoilC

(kg/m ) 3 (kg/m ) 3

ABT ABT

APETR APPT APETR APPT

Polar

Subpolar

Boreal

Cool TemperateWarm Temperate

TropicalSubtropical

Fig. 73 Distribution of the average organic carbon content in the soils (SoilC) in the Holdridge Triangle (data from USDA-SCS - see chapter II). The values in the hexagons represent the mean of all 0.5° x 0.5° longitude/latitude continental grid elements that fall into these hexagons (except grid elements which are cultivated). For a description of the Holdridge Triangle, see chapter II.

Finally, one should mention that globally the major part DOC is discharged to the Atlantic (49%, incl. Mediterranean), while the bulk of POC is discharged into the Pacific and Indian oceans (61%). This discrepancy reflects the great sediment delivery from the southern and south-eastern part of the Asian continent (see chapter IV). South America has the highest specific erosion rate of organic carbon. The rate for this continent is about three times greater than the rates for Australia and for Africa, which are the continents with the lowest erosion rates of organic carbon.

5.6. Conclusions

While most of the previous studies on the fluvial organic matter transport mainly tried to explain the seasonal variability of the observed fluxes in relation to the river hydrograph, it is shown in this chapter that it is also possible to relate the variability of mean annual DOC and POC fluxes with the environmental variability of the corresponding river basins. For DOC, a multiple regression model including drainage intensity, basin slope, and the amount of carbon stored in the soils is the best model to predict the fluxes globally. DOC fluxes become greater with increasing drainage intensities, flatter morphologies, and larger carbon reservoirs in the soils. This supports the general assumption made in most studies on this topic that the soils are globally the major contributors to riverine DOC. However, it is only recently that scientists became aware that basin morphology may also play an important role in controlling dissolved organic carbon fluxes. When the here determined regression model is applied to the total continental area on the basis of the corresponding data sets, the total amount of dissolved

154

organic carbon that is discharged to the oceans is calculated to be about 0.205 GtC/yr. This compares well with previous literature estimates.

For POC, sediment fluxes are the dominant controlling parameter, and no relationships with additional environmental factors can be established. Considering only the data when POC and sediment fluxes have been measured simultaneously, it can be shown that POC fluxes generally increase with increasing sediment fluxes, but the percentage of organic carbon in the total suspended solids clearly decreases with increasing sediment concentrations. This POC% - cTSS relationship was mathematically fitted, allowing thus to calculate the global POC fluxes to the oceans as a function of sediment yields and of drainage intensities. Applying the relationship to all continental grid points yields a total value of about 0.160 GtC/yr, which is also in good agreement with other literature estimates. Nevertheless, one has to mention here that this value may be less sure than the value for DOC. On the one hand, the strong link between FPOC and FTSS makes the global FPOC estimate very dependent on the approach that is applied to estimate global river sediment yields. The problems to predict this parameter that have been pointed out in chapter IV are therefore also valid for the prediction of POC fluxes. On the other hand, it is also possible that the dominance of sediment yield as controlling factor may mask the existence of relationships with other parameters, such as, for example, SoilC, VegC, and litterfall, which may also influence the global POC fluxes.

The here-presented modelling approaches underlie the assumption that the organic carbon fluxes which are measured at the river mouths can be explained as the sum of the fluxes taking place on the grid point levels. This implies that processes such as respiration, in-situ production, and/or sedimentation that can occur during the fluvial transport of organic matter are of small importance for the global fluxes. This is probably not true. For further studies, it is therefore important to investigate whether the findings of this study can be confirmed when looking at organic carbon fluxes in small catchments. Note, however, that the few studies on DOC fluxes in small catchments that have been discussed in chapter I are generally in good agreement with the results of this chapter. I recall here that going from forested catchments (Moore [1989]) over catchments with peaty soils (Grieve [1991]) to wetland catchments (Moore and Jackson [1989]), the reported average DOC concentrations increased considerably. It can be supposed that also the average SoilC in these catchments considerably increase in the same sense. At the same time, we have seen that in the Strengbach catchment it is probably also the steep basin morphology that leads to the quite low average DOC concentration encountered in this brook. The waters quickly percolates through the soils on the steep slopes, and leaching is mainly restricted to a small zone where the soil waters accumulate.

155

CHAPTER VI

ATMOSPHERIC CO2 CONSUMPTION AND RIVER CARBON INPUTS TO THE OCEANS BY

CONTINENTAL EROSION: PRESENT-DAY FLUXES AND IMPLICATIONS FOR THE LAST GLACIAL

MAXIMUM

6.1. Introduction

Continental erosion is a sink for atmospheric carbon. Atmospheric CO2 is consumed both by organic matter formation and chemical rock weathering, and subsequently transferred as dissolved organic carbon (DOC), particulate organic carbon (POC), and dissolved inorganic carbon to the oceans by rivers. The latter occurs mainly in the form of bicarbonate ions (HCO3

-). Concerning organic carbon, the CO2 consumption is governed by the photosynthesis reaction, which can be written as following:

CO2 + H2O → CH2O + O2 (34)

All carbon in the organic matter in rivers is of atmospheric origin. Only the part of fossil POC that is mobilized with the erosion of organic rich sedimentary rocks may be distinguished here from the bulk POC because of its much older age (but we have seen in the previous chapter that this probably negligible in the global POC budget).

Also for riverine HCO3- ions resulting from the weathering of silicate rocks, all carbon comes

from the atmosphere (mainly via soil CO2), as it can be seen, for example, in the following equation for the hydrolysis of albite:

2 NaAlSi3O8 + 2 CO2 + 11 H2O → Al2Si2O5(OH)4 + 2 HCO3

- + 2 Na+ + 4 H4SiO4 (35)

For HCO3- ions resulting from the weathering of carbonate rocks, however, only half of the carbon

originates from atmospheric/ soil CO2, while the other half comes from the carbonate mineral. This can be shown, for example, in the following equation for the calcite dissolution:

CaCO3 + CO2 + H2O → Ca2+ + 2 HCO3- (36)

The purpose of this chapter is to evaluate the role of these carbon fluxes within the global carbon cycle. Rising atmospheric CO2 concentrations related to industrial fossil fuel combustion and deforestation of the tropical rainforest are expected to lead to significant global climatic changes during the forthcoming decades. After 30 years of measurements in the atmosphere and in the oceans,

156

the global atmospheric CO2 budget is still surprisingly uncertain (Tans et al. [1990]). Previous studies estimated the present-day amount of atmospheric carbon that is consumed by continental erosion (FCO2) to be about 0.7 to 0.8 GtC/yr (gigatons of carbon per year), with 0.26 to 0.30 GtC/yr being discharged to the oceans as HCO3

- ions (Berner et al. [1983], Meybeck [1987], Probst [1992], Amiotte-Suchet [1995]), 0.20 to 0.22 GtC/yr as DOC (see chapter V), and 0.18 to 0.23 GtC/yr as POC (see chapter V). This may account for about one third of the estimated net oceanic carbon uptake for today (Sarmiento and Sundquist [1992]), and is hence not negligible if one wants to understand the dynamics of the present-day global carbon cycle.

Modelling the fate of the anthropogenic released CO2 revealed an inconsistency of the observed latitudinal CO2 gradient in the atmosphere with modelled transport fields (Tans et al. [1990]). Up to now, it was difficult to evaluate the role of the fluvial transport pathway in this context because of the lack of sufficient river data world-wide. In the following, the modelling of the river fluxes of organic carbon presented in the previous chapter is coupled with a modelling of the river fluxes of inorganic carbon. The latter has been recently developed at the CGS (GEM-CO2: Global Erosion Model for the consumption of CO2 by rock weathering - Amiotte-Suchet [1995], Amiotte-Suchet and Probst [1993a, b], [1995]). A detailed picture of the total consumption of atmospheric CO2 by continental erosion can thus be drawn. Furthermore, the transport of this carbon is followed by coupling the organic and inorganic carbon erosion with a global river routing scheme. The routing scheme also allows a prediction of the local inputs of the different carbon species into the oceanic system on a grid point scale. The fate of the river carbon once it entered the oceans is discussed, and a generalized steady state budget including the return of the river carbon from the oceans to the atmosphere by carbonate precipitation and organic matter respiration (which liberate CO2) is proposed.

Finally, it is also attempted to estimate the global river carbon fluxes for others than present-day conditions. Many authors pointed out the important role of continental erosion in the long-term or geochemical carbon cycle of the Earth (e.g., Garrels and Mackenzie [1971], Walker et al. [1981], Berner et al. [1983], Tardy et al. [1989], Berner [1991]) with respect to climate change during geological times. The overall empirical relationships established and used in this study are applied to a scenario for the last glacial maximum (LGM, 18000 yrs b.p.) defined by a general circulation model (GCM) run, and the resulting river carbon fluxes are determined. Not at least for this purpose, also a validation of the GEM-CO2 model output with measured HCO3

- fluxes for a set of major world rivers is done, and possible climatic effects on the consumption of atmospheric CO2 by rock weathering are discussed.

Note that many of the presented results have been published in Ludwig et al.(a), in press, and in Ludwig et al. (b), in press. With respect to these publications, the global and regional POC fluxes were in this chapter calculated with the global sediment yield map presented in chapter VI, which is slightly different from the one used in the above mentioned publications.


6.2.1. Modelling of Inorganic Carbon Fluxes

The fluxes of atmospheric CO2 consumed by rock weathering (FCO2-RW) are mainly a function of drainage intensity (Q) and of the rock type that is drained by the surface waters. This has been shown by Amiotte-Suchet and Probst [1993a], [1993b], [1995] and by Amiotte-Suchet [1995]. There exists a linear relationship between FCO2-RW and Q for each rock type. The relationships were established using data published by Meybeck [1986] concerning runoff and HCO3

- concentrations in 232 monolithologic watersheds in France. The watersheds were grouped into six lithological classes representative for the major rock types outcropping on the continents, and the empirical relationships between FCO2-RW and Q were determined for each of the six classes. According to equations 35 and 36,

157

FCO2-RW is considered to be equal to the HCO3- fluxes in waters draining silica rocks, and to be equal

to half of the HCO3- fluxes in waters draining carbonate rocks. Together, these relationships form the

Global Erosion Model for atmospheric CO2 consumption - GEM-CO2 (Amiotte-Suchet and Probst [1995]). The global distribution of FCO2-RW calculated by Amiotte-Suchet and Probst [1995] is available via ftp (ftp cdiac.esd.ornl.gov/pub/db1012) in a spatial resolution of 1° x 1° longitude/ latitude at the Carbon Dioxide Information Analysis Center CDIAC (Oak Ridge, USA).

For a given drainage intensity, FCO2-RW varies considerably for different rock types. FCO2-RW is 17 times greater for carbonate rocks, the rock type with the greatest specific CO2 consumption, than for plutonic and metamorphic rocks, the rock type with the lowest specific CO2 consumption. Going in the order of a decreasing specific CO2 consumption, the six rock types can be classified as follows: carbonate rocks, shales, basalts and gabbros, acid volcanic rocks, sands and sandstones, and plutonic and metamorphic rocks (Fig. 74). Bluth and Kump [1994] investigated the CO2 consumption by rock weathering based on data from different US rivers and found similar relationships.

1

4

3

2

65

Q (mm)

F CO2-RW

(t km yr ) -2 -1

Fig. 74 Relationships of the specific CO2consumption by rock weathering (FCO2-RW) and drainage intensity for six different rock types: (1) carbonate rocks, (2) shales, (3) basalts and gabbros, (4) acid volcanic rocks, (5) sands and sandstones, and (6) plutonic and metamorphic rocks (from Amiotte-Suchet [1995]).

6.2.2. Environmental Data Sets and Empirical Modelling

All data sets that were used to calculate the river carbon fluxes are described in chapter II, together with details on the applied statistical procedures and other technics. The runoff and sediment yield data sets are described in chapter III and in chapter IV, respectively. Note that the values calculated with GEM-CO2 that are given in this study were calculated with the runoff data set of the UNESCO runoff map, and they can therefore be somewhat different compared to the values given by Amiotte-Suchet and Probst [1995] (who used the water budget model of Willmott et al. [1985]). The models to calculate the river fluxes of DOC and of POC are presented in chapter V. Analogous to GEM-CO2, I will call these models in the following the Global Erosion Model for organic carbon (GEM-Corg). The flux of atmospheric CO2 consumed by the erosion of organic matter is thus FCO2-OM. All climatic distinctions follow the classification established in chapter II. Keep in mind that the

158

-150

-10

0

-50

0

50

1

00 1

50

-150

-10

0 -

50

0

50

10

0 1

50

-90

-60

-300306090

a

4°x5

° La

t/Lon

g

01

23

45

67

8

012345678

01

23

45

67

8

012345678

c

4°x5

° La

t/Lon

g

2°x2

.5°

Lat/L

ong

b

Fig

. 75a

Glo

bal 4

° x 5

° lat

itude

/long

itude

rive

r rou

ting

file

of M

iller

et a

l. [1

994]

. Sm

all r

ecta

ngle

s are

end

orei

c gr

id p

oint

s. 75

b P

lot o

f the

bas

in a

reas

(106

km2 )

res

ultin

g fr

om th

e gl

obal

2°

x 2.

5° la

titud

e/lo

ngitu

de r

iver

rou

ting

file

of M

iller

et a

l. [1

994]

ver

sus

the

basi

n ar

eas

calc

ulat

ed in

cha

pter

II fo

r 38

of t

heri

ver

basi

ns c

onsi

dere

d in

this

stu

dy (o

mitt

ing

the

smal

ler

basi

ns).

75c

As 7

5b,

but t

his

time

calc

ulat

ed w

ith th

e 4°

x 5

° la

titud

e/lo

ngitu

de r

iver

rou

ting

file

ofM

iller

et a

l. [1

994]

.

159

river carbon budgets given in this chapter refer only to the ice-free parts and endoreic parts of the continents.

6.2.3. River Routing Files

Calculating the inputs of river fluxes to the oceans on a grid point scale requires to know the paths connecting the grid points near a river's source to the grid points that are successively closer to its mouth for all grid points of the continents. Such river routing files have been created by Miller et al. [1994] in order to improve the hydrological module of the NASA/ GISS general circulation model (Hansen et al. [1983]). The river routing files are based on world maps (Korzoun et al. [1977], The Times of London [1967]) and they exist in the spatial resolutions of a 2° x 2.5° latitude/ longitude and of a 4° x 5° latitude/ longitude grid point resolution. James Miller and collaborators kindly made these files available to me. Figure 75a shows the routing file in the 4° x 5° resolution.

Because of the coarser resolution, the total continental area defined by the river routing files is lower than the one used in this study (which is based on the 0.5° x 0.5° latitude/ longitude grid point vegetation map of Olson [1983] - see chapter II). In the routing files, the coastal morphologies are much simplified and smaller islands cannot be represented. Note that whenever these routing files were used to bring the river carbon fluxes to the oceans, I calculated first the carbon fluxes on the continents in the finer 0.5° x 0.5° resolution with GEM-CO2 and GEM-Corg (respecting the limits of the endoreic part of the continents shown in chapter II), and superimposed then these data on the routing files. Doing that way makes it sure that no carbon is 'lost' through the change in resolution. For the reason of presentation facilities, only the ocean inputs calculated in the 4° x 5° resolution are discussed in this chapter, but calculations were also made in the finer 2° x 2.5° latitude/ longitude resolution. The coarser routing file is still a reasonable representation of the major hydrological pathways on Earth, although it has naturally a greater tendency to overestimate real basin area of rivers compared to the finer routing file (Fig. 75b and 75c).

6.2.4. Determination of the LGM Boundary Conditions

For the prediction of the Earth's climate prevailing during the last glacial maximum, the monthly temperature and precipitation climatologies calculated by the ECHAM2 general circulation model were used in this study. They exist both for a LGM simulation run (lgm) and for the corresponding present-day control run (contr). More details on the ECHAM2 model are given in chapter III. Esser and Lautenschlager [1994], for example, used the same model in order to estimate the change of carbon in the terrestrial biosphere from LGM to present with a global carbon cycle model. Since these ECHAM2 climatologies only exist in the 5.625° x 5.625° longitude/ latitude resolution of the model (so-called T21 resolution), the following algorithm was applied to derive 0.5° x 0.5° grid point climatologies from the model files:

ValM i = ValO i x ValM j / (ValO i) j (37)

ValM are either the monthly temperature (Kelvin) or precipitation (mm) values of the model, ValO are the corresponding values of the temperature and precipitation data sets used in this study (see chapter II), i and j are indices for the grid points in the 0.5° x 0.5° (i) and in the T21 (j) resolution, respectively, and (ValO i) j are the area-weighted averages of all i that fall into j. As this is a common practice also in other studies (e.g., Esser and Lautenschlager [1994], Gibbs and Kump [1994]), the final LGM climatologies were then obtained by subtracting the ECHAM2-derived anomalies from the present-day distributions according to:

ValO(lgm) i = ValO(present-day) i - (ValM(contr) i - ValM(lgm) i) (38)

160

For precipitation, it could happen that equation 38 yields negative values. In these cases, the grid point values were set to 0. Esser and Lautenschlager [1994] derived the LGM precipitation data set they used in their study by applying the relative changes between the LGM and control run of the ECHAM2 model to the actual precipitation distribution, and not the absolute changes. This method avoids negative values, but it holds the risk to alter the global precipitation change predicted by the model with respect to the absolute values. None of the methods is without shortcomings. Note that the here-applied method respects the absolute amount of precipitation change predicted by ECHAM2, except for the grid elements which are set to 0. Cutting off negative values overestimates to some extent the global LGM precipitation.

When applied to the total continental area for present-day (about 152 x 106 km2), the so-created LGM climatologies show a global cooling of 6.7 °C compared to the present-day climatologies (from 9.6 to 2.9 °C), and a reduction of global precipitation with 103 mm (from 817 to 714 mm). Without setting negative LGM precipitation values to 0 in equation 38, the global precipitation value for LGM would only be reduced with about 1.5%. This confirms the applicability of the above presented method. The global LGM temperature and precipitation data sets were then also used to derive global data sets for ABT and Four according to the methods described in chapter II. These data sets are needed to run the empirical models for runoff established in chapter III.

Both the changes of the continental area during LGM related to the sea level fall as well as the appropriate LGM ice-coverage of the continents were taken from the reconstructions of Peltier [1994]. These reconstructions exist as 1° x 1° longitude/ latitude global data files in 1000 yr intervals since LGM. They are available via ftp from the World Center for Paleoclimatology/ National Geophysical Data Center, Boulder, USA. A comparison of the LGM ice-sheet extensions of Peltier [1994] with those of the ECHAM2 model shows a good agreement between both predictions, although some differences occur. For example, the glaciation of the southern tip of South America emerging in the data of Peltier [1994] is not included in the ECHAM2 model. Generally, the model predicts a smaller LGM ice-sheet extension, as well as a smaller total continental area for LGM compared to Peltier's reconstructions. The latter is naturally also related to the coarse grid point resolution of the ECHAM2 model.

I assumed furthermore that the extensions of the endoreic parts of the continents were the same during LGM than for present-day.

6.3. River Carbon Fluxes in the Present-Day Global Carbon Cycle

6.3.1. Spatial Variability of the Atmospheric CO2 Consumption by Continental Erosion

Table 23 lists the carbon fluxes calculated with GEM-CO2 and with GEM-Corg for present-day. The fluxes are regionalized with respect to the major climates, different continents, and different ocean basins. Figure 76 and Figure 77 depict the global distributions of FCO2-RW and FCO2 (FCO2-RW + FCO2-OM), respectively. It is found that continental erosion represents a sink for atmospheric CO2 of 0.593 GtC/yr. In this value, 0.205 GtC/yr can be attributed to DOC, 0.158 GtC/yr to POC, and 0.230 GtC/yr to HCO3

- coming from the atmosphere. HCO3- that originates from carbonate dissolution (FCARB)

makes an additional flux of 0.090 GtC/yr. These values agree well with other literature estimates (see above), confirming the applicability of the modelling approaches presented in this study. Only the calculated POC and HCO3

- fluxes are rather at the lower side of the literature estimates. As mentioned in the previous chapter, the value for FPOC is strongly dependent on the approach that is applied to predict global sediment fluxes. Ludwig et al. [1996] and Ludwig et al. (a), in press calculated slightly

161

F

ig. 7

6 F

CO

2-RW

on

the

cont

inen

ts a

s pre

dict

ed w

ith G

EM-C

O2 (

afte

r Am

iotte

-Suc

het [

1995

]). E

ndor

eic

basi

ns a

nd g

laci

ated

regi

ons a

re o

mitt

ed.

163

F

ig. 7

7 F

CO

2 on

the

cont

inen

ts a

s pre

dict

ed w

ith G

EM-C

O2 a

nd G

EM-C

org.

Endo

reic

bas

ins a

nd g

laci

ated

regi

ons a

re o

mitt

ed.

165

greater FPOC values (0.173 and 0.187 GtC/yr, respectively). In these studies, sediment yields were not coupled to lithology, which yields a greater value for the total sediment flux to the oceans than determined in chapter IV (for details on this problem, see chapter IV). The possibility of an underestimation of real HCO3

- fluxes with GEM-CO2 is discussed below.

Table 23. Regional budgets for the atmospheric CO2 consumption by continental erosion and for river carbon fluxes to the oceans.

______________________________________________________________________________________________________________

Area FDOC FPOC FCO2-RW FCO2 FCARB FCO2-RW / (103 km2) (TgC/yr) (TgC/yr) (TgC/yr) (TgC/yr) (TgC/yr) FCO2-OM

______________________________________________________________________________________________________________ Polar [2] 3892 3.1 1.0 3.4 7.5 1.4 0.8 Tundra and Taiga [5] 23232 45.9 15.0 33.0 93.9 8.6 0.5 Temperate Dry [4] 9635 2.9 12.5 4.3 19.7 2.0 0.3 Temperate Wet [7] 16918 35.4 32.7 47.7 115.8 24.6 0.7 Tropical Dry [6] 21790 16.0 27.8 14.9 58.7 5.4 0.3 Tropical Wet [8] 24919 101.7 68.5 126.3 296.5 47.7 0.7 Desert [3] 5940 0.3 0.3 0.4 1.0 0.2 0.7 Total 106326 205.2 157.9 230.0 593.1 90.1 0.6

Africa 18288 20.0 10.7 11.5 42.2 4.9 0.4 Europe 9564 16.6 10.5 18.5 45.6 7.3 0.7 North America 23020 39.4 27.6 40.5 107.5 17.2 0.6 South America 17732 51.8 34.3 52.8 138.9 8.0 0.6 Asia 32518 73.0 71.6 104.3 248.9 52.7 0.7 Australia 4476 3.9 3.0 2.2 9.1 0.1 0.3 Antarctis 728 0.6 0.2 0.1 0.9 0.1 0.1 Total 106326 205.2 157.9 230.0 593.1 90.1 0.6

Arctic Ocean 16982 24.9 6.2 20.6 51.7 6.3 0.7 North Atlantic 27300 71.0 39.3 71.9 182.2 22.6 0.6 South Atlantic 16959 25.7 9.4 15.6 50.7 4.9 0.4 Pacific 21025 57.0 68.6 79.6 205.2 31.9 0.6 Indian Ocean 16594 21.4 28.1 32.9 82.4 19.6 0.7 Mediterranean 6739 4.6 6.1 9.4 20.1 4.6 0.9 below 60° South 728 0.6 0.2 0.1 0.9 0.1 0.1 Total 106326 205.2 157.9 230.0 593.1 90.1 0.6

______________________________________________________________________________________________________________ Calculated without endoreic regions and regions that are under permanent ice cover TgC = tera grams of carbon (1012 gC)

It can be seen in Table 23 that for all carbon forms, the tropical wet climate is the most important climate type with respect to the global fluxes. This points out the important role of drainage intensity for the consumption of atmospheric CO2 by continental erosion. In the wet tropics occur 54.9% of total FCO2-RW, 43.4% of total FPOC, and 49.6% of total FDOC. I recall here from chapter III that the corresponding runoff accounts for 50% of global runoff. The relative importance of FCO2-RW with respect to FCO2 increases in regions where carbonate and shale outcrops are abundant. The fact that the tropical wet climate contributes more than 50% to the global value of FCO2-RW is related to the great abundance of shales in the Amazon basin and in Oceania. Carbonate outcrops are especially important between the latitudes 20° and 40° N (Amiotte-Suchet [1995]). In the climate budgets in Table 23, this is reflected in the values for the temperate wet climate type, where the inorganic carbon fluxes are enhanced compared to the organic carbon fluxes. Nevertheless, the influence of lithology on FCO2-RW is naturally more evident at continental and regional scales. Note, for example, that for Africa the specific FCO2-RW value is about five times smaller than for South America (but the specific drainage intensity is only three times smaller). In Africa, plutonic and metamorphic rocks as well as sandstones are abundant over large areas. Because these rock types consume small amounts of atmospheric CO2, this continent has low specific FCO2-RW values. By far the greatest specific FCO2-RW values are observed in the south and south-east of Asia, where large carbonate outcrops coincide with great drainage intensities. This makes Asia to the continent with the highest specific FCO2-RW value. It can be

166

calculated that about 23% of global FCO2-RW takes place in the region of Asia between 75° to 135° E and 10° to 40° N, although this region covers only about 9% of the total exoreic continental area.

At the same time, the south and south-east of Asia is of similar importance for the fluxes of particulate organic carbon because of the great sediment delivery from this part of the world (see chapter III). About 25% of global FPOC is discharged to the oceans from the above specified region. This underlines the important role of this region with respect to the CO2 consumption by continental erosion as far as the fluxes of HCO3

- and of POC are concerned. For FDOC, only about 13% of the global flux comes from the same region.

-80 -60 -40 -20 0 20 40 60 80

0

20

40

60

80

FCO2-RW

FPOC

FDOC

Latitude N

Fig. 78 Holospheric distribution of FCO2 (TgC/yr, y-axis) as calculated with GEM-CO2 and GEM-Corg. Note that the carbon fluxes in the regions under permanent ice cover and in the endoreic parts of the continents are not included.

One has finally to notice that by far the major part of the atmospheric CO2 consumption by continental erosion takes place in the Northern Hemisphere. According to Figure 78, this is about 0.420 GtC/yr, being thus 70% of global FCO2. The pattern of high levels of CO2 consumption coincides with a net sink of atmospheric CO2 in the Northern Hemisphere postulated by Tans et al. [1989], [1990]. The authors did not consider river carbon fluxes in their budgets. The here calculated values can account for about 12 to 21% of the sink postulated by the authors. Again, this points out that continental erosion cannot be overlooked if one wants to understand the distribution of atmospheric CO2 in the present-day global carbon cycle.

6.3.2. Determining Local Inputs of Alkalinity, DOC, and POC to the Oceans

On the previous pages, the global river carbon fluxes have been presented on a grid point scale showing the origin of the fluxes on the continents. In the following, it is proposed to predict the local inputs of this carbon into the oceans through a coupling of the GEM-CO2 and GEM-Corg fluxes with the river routing file described above. The so-derived spatial distributions of river carbon discharges to the oceans may be included in other modelling studies in order to better understand the lateral transport of this carbon at the global scale, and eventually also to model the oceanic feedback to the river carbon input. Moreover, river carbon represents also an important contribution of nutrients to the oceans maintaining the biological productivity in the coastal waters. A detailed determination of the local inputs of river carbon to the oceans may also help to better understand the variability of metabolism in the coastal zones at the global scale.

Figure 79 shows the spatial distribution of river inputs of alkalinity (FCO2-RW and FCARB), dissolved organic carbon, and particulate organic carbon to the world's oceans in a 4° x 5° latitude/

Fig. 79 Global river inputs of alkalinity (FHCO3), dissolved organic carbon (FDOC), and particulate organic carbon (FPOC) to the oceans in a 4° x 5° latitude/longitude resolution.

169

longitude grid point resolution. In Table 24, the carbon fluxes are regionalized with respect to the different ocean basins for latitudinal bands of 4° each. Globally, 43% of the total river carbon enters the oceans in the Atlantic Ocean, 34% in the Pacific Ocean, 15% in the Indian Ocean, and 8% in the Arctic Ocean. Note that about 28% is discharged from the 10 greatest world river basins only, and more than one third (37%) of the carbon input into the Atlantic Ocean occurs in the 0-4° N latitude band, where the mouth of the Amazon River is situated.

Table 24. River fluxes of carbon (TgC/yr) to the world's oceans.

______________________________________________________________________________________________________________ Latitude Arctic Ocean Atlantic Ocean (1) Indian Ocean Pacific Ocean

Band _________________________ _________________________ _________________________ _________________________ HCO3

- DOC POC HCO3- DOC POC HCO3

- DOC POC HCO3- DOC POC

______________________________________________________________________________________________________________ 84 - 88 N 0.10 0.01 0.01 -- -- -- -- -- -- -- -- -- 80 - 84 N 0.12 0.05 0.03 0.07 0.01 0.01 -- -- -- -- -- -- 76 - 80 N 5.34 1.45 0.48 0.06 0.01 0.02 -- -- -- -- -- -- 72 - 76 N 12.72 14.25 2.72 0.63 0.10 0.05 -- -- -- -- -- -- 68 - 72 N 8.05 7.77 3.06 0.04 0.10 0.02 -- -- -- -- -- -- 64 - 68 N 1.31 1.53 0.28 1.08 1.37 0.63 -- -- -- -- -- -- 60 - 64 N -- -- -- 0.84 3.64 0.59 -- -- -- 2.81 1.96 1.45 56 - 60 N -- -- -- 5.02 8.98 1.78 -- -- -- 4.20 3.69 3.83 52 - 56 N -- -- -- 4.27 5.07 0.60 -- -- -- 2.75 5.05 3.08 48 - 52 N -- -- -- 9.99 4.38 1.15 -- -- -- 1.98 1.46 1.26 44 - 48 N -- -- -- 5.36 2.56 1.48 -- -- -- 1.56 0.92 0.76 40 - 44 N -- -- -- 10.30 2.34 3.07 -- -- -- 2.00 0.96 1.16 36 - 40 N -- -- -- 3.63 1.09 1.50 -- -- -- 1.09 1.27 6.46 32 - 36 N -- -- -- 1.49 1.40 1.11 -- -- -- 1.29 1.51 2.08 28 - 32 N -- -- -- 8.67 5.46 4.07 -- -- -- 27.94 4.12 7.76 24 - 28 N -- -- -- 0.66 0.52 0.35 2.25 0.13 1.83 3.52 0.92 0.98 20 - 24 N -- -- -- 4.57 0.78 2.21 1.56 0.31 1.07 11.35 2.17 2.92 16 - 20 N -- -- -- 1.32 0.44 0.42 11.01 7.72 10.22 2.77 1.62 1.81 12 - 16 N -- -- -- 2.72 2.75 4.18 16.38 4.87 7.92 2.42 0.68 4.44 08 - 12 N -- -- -- 4.95 6.85 6.36 4.13 1.30 1.35 7.58 3.76 3.97 04 - 08 N -- -- -- 1.00 2.48 1.16 2.57 0.61 0.47 3.96 2.60 2.68 00 - 04 N -- -- -- 48.36 36.20 20.22 0.43 0.82 0.76 11.35 6.10 5.78 00 - 04 S -- -- -- 2.92 1.47 0.84 0.41 0.20 0.78 7.19 5.43 3.61 04 - 08 S -- -- -- 7.09 6.81 1.53 4.93 1.83 1.29 2.64 1.14 1.52 08 - 12 S -- -- -- 0.75 0.52 0.18 0.53 0.32 0.41 10.77 2.84 3.69 12 - 16 S -- -- -- 0.05 0.13 0.02 1.43 1.96 1.38 0.20 0.08 0.19 16 - 20 S -- -- -- 0.76 0.42 0.17 1.14 0.65 0.44 0.04 0.20 0.21 20 - 24 S -- -- -- 0.08 0.27 0.14 0.60 0.76 0.74 0.07 0.22 0.23 24 - 28 S -- -- -- 0.05 0.17 0.08 0.22 0.20 0.35 0.09 0.13 0.10 28 - 32 S -- -- -- 0.50 0.71 0.83 0.02 0.04 0.07 0.01 0.01 0.02 32 - 36 S -- -- -- 3.76 4.60 1.44 0.01 0.02 0.09 0.16 0.19 0.32 36 - 40 S -- -- -- 0.04 0.09 0.02 0.20 0.47 0.44 0.79 1.09 0.99 40 - 44 S -- -- -- 0.12 0.16 0.06 0.03 0.07 0.03 0.87 1.52 1.18 44 - 48 S -- -- -- 0.01 0.01 0.01 -- -- -- 0.68 2.12 1.72 48 - 52 S -- -- -- 0.32 0.97 0.72 -- -- -- 0.07 0.35 0.07 52 - 56 S -- -- -- 0.13 0.19 0.14 -- -- -- 0.16 0.42 0.26

Total 27.55 25.05 6.56 131.63 103.03 57.15 47.86 22.27 29.64 112.30 54.53 64.55 ______________________________________________________________________________________________________________

(1) Including the Mediterranean Sea and the Black Sea

The general pattern is similar for the three carbon species because for all of them, the fluxes are strongly coupled to the spatial distribution of the freshwater inputs into the oceans. Differences are found, however, at local scales. Mainly the ratio of inorganic (HCO3

-) to organic (DOC + POC) carbon (RCinorg/Corg) can vary considerably. The variability of the fluxes of organic carbon is more closely coupled to the variability of drainage intensity than this is the case for the fluxes of alkalinity. HCO3

- fluxes are also strongly dependent on the outcropping rock types on the continents. Consequently, RCinorg/Corg follows more or less the distribution of lithology. Especially large carbonate outcrops in river basins lead to great RCinorg/Corg values in the ocean grid elements to which the rivers discharge. Because of carbonate rich watersheds of certain rivers discharging to these zones, RCinorg/Corg values of 1.5 to more than 2 can be found, for example, in the Atlantic Ocean in the

170

latitude bands of 48-52° N (e.g., the St. Lawrence River) and of 40-44° N (some European rivers as the Rhône River), or in the Pacific Ocean in the 28-32° N (e.g., the Changjiang River) and in the 20-24° N (e.g., the Hungho River) latitude bands. On the continents, carbonate outcrops are mainly abundant in the Northern Hemisphere, such as in the south-east of Asia, in western Europe, or in the eastern United States (see, for example, Amiotte-Suchet [1995]). For this reason, RCinorg/Corg is greater in the parts of the oceans north of the equator than south of it. Greatest average values are found in the North Pacific (RCinorg/Corg = 0.99) and in the northern Indian Ocean (RCinorg/Corg = 0.97). In the South Atlantic and in the southern Indian Ocean, the ratios of inorganic to organic river carbon inputs are low (RCinorg/Corg = 0.73 and 0.76, respectively).

With respect to the two organic carbon forms, DOC and POC, one can note that inputs of DOC are dominant in the Atlantic Ocean. The DOC to POC ratio, RDOC/POC, is on average 1.7 in the North Atlantic, and 2.7 in the South Atlantic. Particulate organic carbon inputs are more abundant in the Pacific and in the Indian Ocean. In the northern Indian Ocean, the share of POC in the organic matter fluxes is greatest (RDOC/POC = 0.7 on average) because of the great sediment fluxes that are characteristic for the rivers draining the South and Southeast of the Himalayan region (e.g., the Ganges/ Brahmaputra, Indus, and Irrawaddy rivers). By far the greatest average RDOC/POC value is found for the Arctic Ocean. Mechanical erosion rates are very small in the boreal climate zones (chapter IV) and the soils are generally rich in organic carbon (chapter V). For these reasons, nearly 80% of the total organic carbon input occurs here in the dissolved form (RDOC/POC = 3.8).

6.3.3. Fate of The River Carbon in the Oceans

An important question for the evaluation of the role of the fluvial carbon within the global carbon cycle is the fate of this carbon once it entered the oceanic system. Processes that withdraw carbon from the oceans are the sedimentation of organic matter and of carbonates, as well as the respiration of organic matter in the water column. Carbonate precipitation and organic matter respiration liberate CO2 to the atmosphere, whereas the carbon that is incorporated in the sediments becomes part of the lithosphere.

Supposing a steady state, i.e. assuming that the river input is balanced by the withdrawal of carbon in the oceans, Smith and Hollibaugh [1993] concluded from a review of literature data that approximately one third of the total organic carbon input by river is lost to the lithosphere, while two thirds return to the atmosphere by respiration. They estimated the total amount of the riverine organic carbon input to be 0.408 GtC/yr, which is close to the value of about 0.365 GtC/yr found in this study. According to their findings, about equal amounts of the fluvial organic carbon becomes involved in the coastal organic matter cycling (mainly in the estuaries) and in the offshore ocean organic matter cycling, but the ratio of burial to respiration in the coastal zone should be about 6 to 4, while this is only 1 to 9 in the open oceans. No distinction is made in their budgets between the fate of POC and DOC because much of the POC in the water column and in the upper sediment layers may be converted to DOC. When applied to the here presented values, this means that about 0.248 GtC/yr of the fluvial organic carbon input should return to the atmosphere after being oxidized in the oceans. 30% of this carbon is respired in the coastal waters, i.e. mainly in the grid elements in Figure 79 to which the river carbon is discharged. The remaining 70% is subjected to a slow oxidation in the open ocean waters.

For inorganic carbon, a steady state implies that carbonate sedimentation in the oceans balances the river HCO3

- fluxes. As an effect of carbonate precipitation, half of the HCO3 carbon is fixed in the sediments, while the other half is released to the atmosphere. With the global values of Table 23, this makes 0.160 GtC/yr for each flux. Regionally, carbonate sedimentation in the oceans is limited to the regions where sea floor lies above the calcite compensation depth (CCD). The present-day globally averaged CCD lies at a depth of 4.5 km (Delaney and Boyle [1988]). Assuming a steady state between the river input and the oceanic output fluxes, the carbon return from the oceans to the atmosphere can thus be estimated to be 0.408 GtC/yr. The remaining 0.187 GtC/yr should enter the atmosphere on the

171

continents via volcanism (about 0.070 GtC/yr, in agreement with Berner et al. [1983] and Williams et al. [1992]) and the oxidation of fossil organic matter in sedimentary rocks (about 0.117 GtC/yr, in agreement with Kramer [1994] and Sarmiento and Sundquist [1992]). Figure 80 summarizes the here discussed steady state budget for river carbon in the oceanic system.

a

b

c

de

gf

h

i

j

k

l

90

595

685

365

160 117

75

173

160 70

117

320

Atmosphere Oceans Lithosphere Fig. 80 Steady state budget for river carbon (adapted from Smith and Hollibaugh [1993]: a, river carbon input; b, atmospheric CO2 consumed by continental erosion; c, lithospheric carbon consumed by continental erosion; d, inorganic ocean carbon; e, organic carbon (DOC + POC) in the oceans; f, carbonate sedimentation; g, organic matter sedimentation; h, organic matter respiration in the coastal zone; i, organic matter respiration in the open ocean; j, CO2 degassing during carbonate precipitation; k, oxidation of fossil organic matter in sedimentary rocks; l, volcanism. All fluxes are in 1012 gC/yr.

However, a great uncertainty in the here discussed budget is the variability of the metabolism in the coastal zones. Although Smith and Hollibaugh [1993] concluded that the coastal ocean is globally net heterotrophic, i.e. respiration exceeds primary production, they showed also that this is not the case for all sites for which data can be found. They presented indications that the heteroprophic character of an estuary increases with greater primary production rates. Estuaries with low productivity rather tend to be net autotrophic, i.e. carbon production exceeds carbon consumption. More data are needed to confirm these trends. Looking at the very uneven distribution of the river carbon inputs in Figure 79, it becomes evident that a strong variability of the coastal metabolism could have a great influence on the fate of the organic river carbon once it is discharged to the oceans. Still little is known about the cycling and fate of riverine organic matter in coastal waters (Ludwig [1991], Ludwig and Spitzy [1992], Spitzy et al. [1990]).

Moreover, there are no means to verify whether the oceans are actually in a steady state with respect to the river carbon inputs, or not. This may be especially questionable for the inputs of organic

172

carbon. Likens et. al. [1981] and Berner [1989] supposed that delivery and burial of organic carbon in the coastal oceans may have doubled in the last 200 years, largely due to deforestation and agricultural practices. It cannot be expected that oceanic response is quick enough to compensate such rapid changes. On the other hand, river damming may have counteracted to such a trend, and it is difficult to estimate the sum of all human impacts on the river fluxes of POC and of sediments (see chapter IV). Kempe [1995] argued that large areas of the world's coastal zones may have lost their natural heterotrophic status and became autotrophic because of anthropogenic eutrophication, thereby causing either an increase in marine organic carbon burial or an increase in the oceanic DOC and dissolved inorganic carbon pools.

6.3.4. Influence of Climate on the Consumption of Atmospheric CO2 by Rock

Weathering

The applicability of the GEM-CO2 relationships to the scale of large river basins was tested for the Congo and the Amazon rivers under tropical wet climate, and for the Garonne River in France under temperate wet climate. The comparison of the results with estimates derived from field measurements revealed a good agreement between both methods (Amiotte-Suchet and Probst [1993a], [1993b], [1995]). Nevertheless, it is not excluded that additional factors which are not included in the model may not also influence the consumption of atmospheric CO2 by rock weathering. Especially the influence of climatic factors, such as temperature, have been proposed by numerous authors (e.g., Garrels and Mackenzie [1971], Holland [1978], Berner et al. [1983], Tardy [1986], Meybeck [1986], [1987], Probst [1992], Velbel [1993]).

A climatic influence is indicated if one compares HCO3- fluxes from the literature for various

major world rivers with the fluxes calculated with GEM-CO2 for these rivers (Fig. 81a). The literature values are listed and referenced in Table 25. It is interesting to note that the literature values compare well with the model values only in the tropical wet climate (black circles), while the model seems to underestimate the fluxes in the other climate types (white signs). Because of the climatic heterogeneity of many of the river basins (see chapter II), the following method was applied to test if there exists a systematic deviation of the GEM-CO2 values from the literature values with respect to climate. I calculated for all basins mean specific HCO3

- fluxes with GEM-CO2 for all climatic subunits that are represented in the basins. Note that all basin grid points that fall into the same class are taken as one subunit, which does not necessarily mean that this subunit is one geographically connected region. The average specific flux for an entire river basin (Fmodel) is then the sum of the specific values (Fi), multiplied by the percentage that the units occupy in the basin (ai), divided by 100:

Fmodel = (a1F1 + a2F2 + a3F3 + ...+ aiFi) / 100 (39)

The unit of the specific fluxes is t km-2 yr-1, and the indices (i) refer to the climate type according to the key in Table 23. A multiple regression between the observed fluxes (Fobserved) of all basins and the area weighted model fluxes of all climatic subunits (aiFi) in these basins can then help to identify the importance of each climate type with respect to the deviation between the observed and the modelled fluxes. It leads to the following regression equation:

Fobserved = (6.55 a4F4 + 2.91 a5F5 + 1.30 a7F7 + 0.85 a8F8) / 100 (40)

n = 31, r = 0.95, P < 0.001 for all aiFi

P is the significance level, r the correlation coefficient, and n the number of rivers considered in the regression. In equation 40, I omitted the Danube, Mahandi, Godavari and Magdalena rivers from the rivers in Table 25. Including these rivers, equation 40 would have a significant positive intercept. It is interesting to note that the first three rivers have the greatest percentages of cultivated area in their basins (around 50 %) compared to all other rivers in Table 25 (see Tab. 21 in chapter V). It is therefore

173

0 5 10 15 20 25 30

0

5

10

15

20

25

30

Tropical Wet

Temperate Wet

Tropical Dry

Tundra & Taiga

Temperate Dry

(a)

Fig. 81a Comparison of observed bicarbonate fluxes (t km-2 yr-1, x-axis) with the values calculated with GEM-CO2 (t km-2 yr-1, y-axis) for the river basins of Table 25. The rivers are classified according to their average climatic situation (see chapter II).

Fig. 81b As Fig. 81a, but this time the GEM-CO2 output was correctedaccording to the method discussed inthe text.

0 5 10 15 20 25 30

0

5

10

15

20

25

30

(b)

Tropical Wet

Temperate Wet

Tropical Dry

Tundra & Taiga

Temperate Dry

not excluded that these rivers have elevated fluxes because of the use of fertilizers in their basins. Fertilizers can increase natural HCO3

- fluxes without consuming atmospheric CO2 by dissolving carbonate minerals (Etchanchu and Probst [1988], Amiotte-Suchet [1995], Semhi [1996]). It is not known why also the Magdalena River does not fit here, but since this river has a low data quality index in Table 25, I did not further follow this question.

The regression coefficients in equation 40 indicate that GEM-CO2 considerably underestimates the HCO3

- fluxes in the temperate dry (i = 4) and the tundra and taiga (i = 5) climates (with 555% and 191%, respectively). In the temperate wet climate (i = 7), the underestimation is less (30%), whereas in the tropical wet climate (i = 8) the fluxes seem to be slightly overestimated (15%) by the model. The

174

tropical dry climate was not significant in the regression (it tends, however, rather to a regression coefficient between those of a5F5 and of a7F7). Also the other climate types which are partly represented in some of the basins were not found to be significant, which can be explained with the small areas they occupy in the basins (ice-free polar climate), and their very low contributions to the total fluxes (desert climate).

Table 25. Bicarbonate fluxes of major world rivers (from Amiotte-Suchet [1995]).

________________________________________________________________________________________________

River Basin Area HCO3- HCO3

- Dq Source (106 km2) (t km-2 yr-1) (mg/l)

________________________________________________________________________________________________ Amazon 5.903 5.544 5.2 2 Probst et al. [1994b]

Zaire 3.704 0.984 2.6 3 Probst et al. [1992] Mississippi 3.246 3.708 24.6 3 USDS [annual]

Ob 3.109 2.832 21.2 1 Meybeck [1987] Paraná 2.868 1.428 7.6 2 Kempe [1982] Yenisei 2.567 3.348 14.7 2 Gitelson et al. [1988]

Lena 2.465 2.844 13.2 1 Livingstone [1963] Amur 1.926 1.488 8.4 1 Meybeck [1987] Nile 1.874 1.464 30.5 2 Kempe [1983]

Changjiang 1.822 11.052 21.7 3 Gan Wei- Bin et al. [1983] Ganges/Brahmaputra 1.656 13.392 18.2 3 Meybeck [1979]

Mackenzie 1.615 3.492 20.9 1 Reeder et al. [1972] Niger 1.540 0.864 6.7 2 Martins [1983]

Zambesi 1.413 0.348 4.9 2 Meybeck [1987] Murray 1.131 0.192 18.1 3 Herczeg et al. [1993]

St. Lawrence 1.114 7.200 17.8 2 Cossa and Tremblay [1983] Orinoco 1.026 2.628 2.5 2 Paolini et al. [1987]

Indus 0.912 7.848 29.8 2 Arain [1987] Mekong 0.864 6.312 11.6 3 Meybeck [1979] Yukon 0.843 5.604 22.5 3 USDS [annual]

Huanghe 0.823 2.820 39.3 2 Gan Wei- Bin et al. [1983] Danube 0.773 9.612 37.1 2 Meybeck [1979] Orange 0.716 0.324 21.1 3 Meybeck [1979]

Colorado 0.708 0.804 28.5 3 USDS [annual] Columbia 0.664 4.488 16.0 3 USDS [annual] Si Kiang 0.464 16.932 26.2 1 Qunying et al. [1987] Limpopo 0.344 0.444 28.7 1 Livingstone [1963]

Northern Dvina 0.329 9.576 30.1 1 Livingstone [1963] Godavari 0.311 5.688 17.6 1 Biksham and Subramanian [1988]

Magdalena 0.285 8.436 10.0 1 Meybeck [1979] Fraser 0.248 4.680 11.8 1 Meybeck [1987] Yana 0.243 0.852 6.3 1 Livingstone [1963]

Mahandi 0.190 5.796 11.7 3 Subramanian [1979] Rio Negro (Arg.) 0.175 2.268 13.7 1 Depetris [1980]

Hungho 0.159 11.700 15.6 1 Ming-Hui et al. [1982] ________________________________________________________________________________________________

Dq is an index for data quality: 1, poor; 2, sufficient; 3, good

At first sight, one may be surprised because the empirical relationships forming GEM-CO2 were established in the temperate wet climate, but when applied to the scale of large basins, they fit best in the tropical wet climate. This may be explained with the fact that the basic relationships of the model uniquely reflect the situation in small watersheds with a mean size of about 8 km2 (Meybeck [1986]). At the scale of a large river basin, these watersheds represent more the headwater regions, where groundwater contributions to the river are normally small with regard to the lower course of the river. The relationships should therefore mainly reflect the composition of typical surface or subsurface waters. Note that the deviation of the GEM-CO2 fluxes with respect to the observed fluxes generally becomes greater for dryer and colder climates, and the regression coefficients in equation 40 decrease in the order of the increasing specific drainage intensities for each climate type (see chapter III). This is indicating that the disagreement of the fluxes in Figure 81a may be at least partly related to the residence time of the water in the basins, and thus to the reaction time for the water-rock-interactions. Given the same water storage capacity in a drainage basin, the average residence time of the waters in the tropical wet climate is naturally the shortest of all climate types because of the great water flux in

175

the tropical wet climate, and the chemical composition of the river waters probably reflects here closest the composition of typical surface waters with only short contact with the outcropping lithologies.

However, there may also be other effects related to climate that can influence river bicarbonate fluxes. For example, in the tropical climate, deep lateritic formations which are depleted in alterable minerals are typical soil formations. This may generally reduce chemical erosion, and thus also HCO3

- fluxes. An important question is this context is the role of the soils for weathering processes. Because it can be supposed that much of the CO2 involved in the weathering reactions (e.g., equations 35 and 36) may originate from the soils via the biological respiration of organic matter, the CO2 consumption by rock weathering should also be enhanced by elevated pCO2 values (partial pressure of CO2) in the soils. Very little is known about the spatial and temporal variability of this parameter at the global scale. It is possible that in certain regions, pCO2 values in the soils may be especially elevated in winter when the soils are covered by a snow cover, preventing the CO2 to degas to the atmosphere. Also this could explain why in equation 40 great regression coefficients are found for the colder climate types.

Table 26. Effects of the climatic correction factors determined in the text on the budgets for thebicarbonate fluxes. _________________________________________________________________________________________________________

corrected corrected corrected budget (1) (%) budget (1) (%) budget (1) (%)

_________________________________________________________________________________________________________

Polar 0.0 Africa -1.6 Arctic Ocean 176.0 Tundra and Taiga 191.0 Europe 90.2 North Atlantic 20.5 Temperate Dry 555.0 North America 68.0 South Atlantic 5.2 Temperate Wet 30.0 South America -5.1 Pacific 21.6 Tropical Dry 0.0 Asia 32.9 Indian Ocean 28.9 Tropical Wet -15.0 Australia 28.1 Mediterranean 74.6 Desert 0.0 Antarctis -0.7 below 60° South 1.2 Total 34.2 Total 34.2 Total 34.2

_________________________________________________________________________________________________________(1) with respect to the values given in Table 23

The here discussed approach can naturally only give indications for the influence of climatic factors on river HCO3

- fluxes, but it denotes nothing about any processes being responsible for it. In Figure 81b, the GEM-CO2 outputs of Figure 81a have been corrected by a simple multiplication of the grid point fluxes with the regression coefficients determined in equation 40, depending on the climate

-80 -60 -40 -20 0 20 40 60 80

0

20

40

60

80

FCO2-RW

FPOC

FDOC

Latitude N

Fig. 82 Holospheric distribution of FCO2 (TgC/yr, y-axis), as calculated with GEM-CO2 and GEM-Corg. Compared with Figure 78, the GEM-CO2 outputs have been corrected according to the climatic correction factors determined in the text.

176

type to which the grid elements belong. For the grid elements in the climates that were not found to be significant in the regression, the fluxes remained unchanged. The corrected fluxes compare now better with the observed ones. I consider therefore this method as a way to quantify possible effects of other controlling factors than drainage intensity and lithology (which are driving the GEM-CO2 model) on river HCO3

- fluxes at global and regional scales. Table 26 shows the consequences of this correction on the inorganic carbon budgets of Table 23. Note that the total amount of CO2 consumed by rock weathering (FCO2-RW) would increase to more than 0.310 GtC/yr. The consumption of atmospheric CO2 by rock weathering north of 30° N would become almost twice as great as without correction (see Fig. 78 and Fig. 82).

6.4. Implications for the Atmospheric CO2 Consumption by Continental Erosion During the Last Glacial Maximum

It has been proposed that increased carbon fluxes by continental erosion may have at least partly contributed to the low atmospheric CO2 concentrations during the last glacial maximum (e.g., Munhoven and François [1994], [1996], Gibbs and Kump [1994]) which are documented in the Vostok ice core record (Barnola [1987]). In the geochemical carbon cycle, the atmospheric CO2 consumption resulting from carbonate weathering is normally balanced in relative short time scales by carbonate sedimentation in the oceans, where all CO2 is released back to the ocean/ atmosphere system (equation 36, from the right to the left). This is not the case for all CO2 consumed by silicate weathering or by organic matter erosion. Here, a part of this carbon is lost to the lithosphere by carbonate and organic matter sedimentation, and only returns to the ocean/ atmosphere system via metamorphism/ volcanism and by the slow oxidation of old sediment carbon (kerogen) in sedimentary rocks (see also Fig. 80). Because the latter processes can vary considerably over geological time scales, this may result in large perturbations of atmospheric CO2.

6.4.1. Application of GEM-CO2 and GEM Corg to LGM Conditions

Table 27 compares the distribution of continental area and of major climate types at present-day with the situation during the last glacial maximum, as reconstructed in this study (see above). In order to investigate whether the total amount of atmospheric CO2 consumed by continental erosion was different from today, the GEM-CO2 and GEM Corg models were applied to the reconstructed LGM conditions. Figure 83 summarizes the parameter requirements of both models. Because global data sets for drainage intensity and for sediment yields are needed to run the models, also the empirical relationships that have been established in the chapters III and IV were applied to the corresponding LGM climatologies to create these data sets for LGM. Also here, the final data sets were created by subtracting the modelled anomalies between present-day and LGM from the global Q and FTSS data sets for present-day (equation 38). Again, this could lead to negative values for some grid elements, which were then set to 0. Permitting negative values would reduce the global LGM runoff only with about 1%, whereas for FTSS, the global sediment flux to the oceans would be reduced with nearly 6%. The final LGM data sets overestimate thus the global averages with these percentages compared to the purely modelled data sets.

For all empirical models, it was assumed that the morphological characteristics of the continents during LGM were the same as for today. One problem is that the morphological and lithological characteristics for the grid elements which became land area during LGM as a result of the lower sea level are not known. Consequently, the GEM-CO2 and GEM Corg models cannot be applied to these grid elements, and it is thus not possible to calculate one definitive value for FCO2 during LGM. Nevertheless, the concerned grid points represent only a minor part of the overall continental area, and I used therefore the following method to estimate the global budgets. For each carbon species, the

177

specific fluxes for each climate type, for each continent, and for each of the continental drainage areas corresponding to the different ocean basins were calculated on the basis of the grid points only that belong to the exoreic and ice-free continental area both for present-day and for LGM (about 87 x 106 km2). These specific fluxes were then applied to the real areas the units occupied during LGM (Table 27), and the global budgets were established within each category. The results are shown in the Tables 28a and 28b. As best estimate for the total FCO2 value, the arithmetic mean of the global values established for each of the three categories can then be selected.

Table 27. Comparison of the distribution of continental area and of major climate types atpresent-day with the situation during the last glacial maximum, as reconstructed in this study(see text).

_______________________________________________________________________________________________________ Present-Day Area Last Glacial Maximum Area ________________________________________ ________________________________________ total only exoreic exoreic, no ice total only exoreic exoreic, no ice (106 km2) (106 km2) (106 km2) (106 km2) (106 km2) (106 km2)

_______________________________________________________________________________________________________ Polar (ice covered) 15.26 15.26 - 42.55 42.52 - Polar (no ice) 3.89 3.89 3.89 6.33 6.12 6.12 Tundra and Taiga 24.09 23.21 23.21 20.74 17.97 17.97 Temperate Dry 16.18 9.63 9.63 15.88 10.04 10.04 Temperate Wet 18.56 16.92 16.92 12.32 11.15 11.15 Tropical Dry 25.53 21.79 21.79 23.61 20.77 20.77 Tropical Wet 25.07 24.93 24.93 26.08 25.99 25.99 Desert 23.88 5.96 5.96 25.35 7.51 7.51 Total 152.47 121.59 106.33 172.86 142.06 99.54

Africa 33.00 18.29 18.29 33.88 19.18 19.17 Europe 12.70 9.66 9.56 16.19 13.16 7.16 North America 25.40 25.04 23.02 29.61 29.26 12.07 South America 18.16 17.73 17.73 19.62 19.20 18.65 Asia 41.03 32.54 32.52 48.34 39.87 36.71 Australia 8.33 4.48 4.48 9.66 5.82 5.77 Antarctis 13.85 13.85 0.73 15.56 15.57 - Total 152.47 121.59 106.33 172.86 142.06 99.54

Arctic Ocean 17.55 17.55 16.98 22.42 22.43 11.90 North Atlantic 38.86 28.77 27.30 43.31 33.24 19.37 South Atlantic 18.55 16.96 16.96 19.58 17.99 17.86 Pacific 29.98 21.12 21.02 35.24 26.39 24.00 Indian Ocean 23.90 16.59 16.59 26.63 19.34 19.33 Mediterranean 9.77 6.74 6.74 10.13 7.10 7.09 below 60° South 13.85 13.85 0.73 15.56 15.57 - Total 152.47 121.59 106.33 172.86 142.06 99.54

_______________________________________________________________________________________________________

On the whole, FCO2 at LGM totals 0.513 GtC/yr. From this value, 0.169 GtC/yr can be attributed to FDOC, 0.141 GtC/yr to FPOC, and 0.203 GtC/yr to FCO2-RW. When the climatic correction factors determined in the previous section were taken into account, the total amount of FCO2 would increase to 0.594 GtC/yr. These values all range between 85% and 90% of the present-day values, indicating that the consumption of atmospheric CO2 during LGM was lower than today.

The most efficient way to increase river carbon fluxes at the global scale would be an increase of the continental area together with an increase of runoff (Berner et al. [1983], Tardy et al. [1989], Berner [1991]). Assuming that weathering is negligible under ice sheets (this point is also discussed below), the effective erodible continental area did not change much during LGM compared to present-day because the loss of area caused by the extensions of the ice sheets was more or less compensated by a greater exposure of shelf area due to the sea level fall (Fairbanks [1989]). In the here-presented

178

LGM scenario, the effective erodible area even slightly decreases. At the same time, the total amount of water running off the ice-free regions of the continents was considerably lower during LGM compared to present-day (with about 14%). Note that despite the increase of the tropical wet climate area during LGM, the total amount of runoff originating from this climate type decreases. Also the areas of deserts and of the temperate dry climate become greater for LGM, while the areas of the temperate wet and the tundra and taiga climates decrease notably (Tab. 27 and 28b). This is in agreement with observations. Reconstruction of vegetation distribution based on palynological, pedological, and sedimentological evidence globally indicate a greater aridity during LGM than today (Adams et al. [1990], Adams and Faure [1996]). Starkel [1988] concluded from palaeohydrological evidences that water discharge by rivers was generally lower during LGM than today.

GEM-CO2 GEM-Corg

etc. . . .FCO2 (granites) = fn3(Q)

FTSS(climate) =

FPOC = fn(Q, FTSS)FDOC = fn(Q, Slope, SoilC)

FTSS

FPOCFDOCFCO2-RW

FCO2

FCO2 = FCO2-RW + FDOC + FPOC

Q Lithol. SoilC Slope Four Climate

AT APPT

FCO2-RW =

a = fn(climate)

Global Data Sets

Small Monolithologic

Watersheds

Major World

River Basins

FCO2 (shales) = fn2(Q)FCO2 (carb.) = fn1(Q)

Global Data Sets

a * FCO2(rock type)

fn(Q, Four, Slope, LithMI)

Fig. 83 Flow diagram of the relationships forming the models GEM-CO2 and GEM Corg in order to predict the consumption of atmospheric CO2 by continental erosion. Abbreviations: AT, mean annualtemperature; APPT, mean annual precipitation total; Lithol., Lithology; Q, drainage intensity, Slope,average grid point slope; SoilC, average organic carbon content in the soils; Four, modifiedFournier-index characterizing the precipitation distribution over the year; LithMI, index for theerodibility of the abundant basin lithology with regard to mechanical erosion; FCO2-RW, consumption of atmospheric CO2 by rock weathering; FTSS, sediment flux; FDOC, flux of dissolvedorganic carbon; FPOC, flux of particulate organic carbon; FCO2, consumption of atmospheric CO2by continental erosion

179

Table 28a. Estimated fluxes of water, sediments, DOC, POC, and atmospheric CO2 consumed by rock weathering to the oceans during the last glacial maximum. For further explanations, see text.

______________________________________________________________________________________________________________ Area (1) Q FTSS FDOC FPOC FCO2-RW FCO2-RWc (2) (106 km2) (km3) (Tg/yr) (TgC/yr) (TgC/yr) (TgC/yr) (TgC/yr)

______________________________________________________________________________________________________________ Polar (no ice) 6.12 1831 160 10.5 4.2 15.0 15.0 Tundra and Taiga 17.97 3975 976 23.8 13.6 22.8 52.6 Temperate Dry 10.04 1005 1243 3.4 7.6 8.0 51.2 Temperate Wet 11.15 5899 2854 21.9 26.7 36.0 49.1 Tropical Dry 20.77 2754 2786 14.7 17.8 17.6 17.6 Tropical Wet 25.99 19957 6276 92.8 65.2 101.7 90.1 Desert 7.51 231 1045 1.4 4.1 1.5 1.5 Total 99.54 35652 15340 168.5 139.2 202.7 277.1

Africa 19.17 3963 681 19.0 9.5 14.6 17.4 Europe 7.16 1908 976 9.0 8.5 15.9 31.6 North America 12.07 4146 1650 18.2 15.8 26.0 45.1 South America 18.65 10299 2812 48.5 30.1 51.2 52.6 Asia 36.71 14375 9301 68.7 74.0 90.6 134.3 Australia 5.77 757 274 4.0 3.2 2.2 4.4 Antarctis - - - - - - - Total 99.54 35450 15694 167.4 141.1 200.5 285.3

Arctic Ocean 11.90 1509 181 12.7 3.8 8.0 14.7 North Atlantic 19.37 11045 2925 53.1 32.8 64.4 78.8 South Atlantic 17.86 4666 661 24.1 10.4 17.1 19.2 Pacific 24.00 11480 7056 49.4 58.5 71.3 106.0 Indian Ocean 19.33 6389 4385 25.8 32.6 35.7 51.1 Mediterranean 7.09 985 775 4.4 5.6 8.9 19.9 below 60° South - - - - - - - Total 99.54 36074 15984 169.5 143.6 205.4 289.6

______________________________________________________________________________________________________________

Best estimate 35725 15673 168.5 141.3 202.9 284.0

______________________________________________________________________________________________________________ (1) Only exoreic and without ice cover. (2) Corrected according to the climatic correction factors determined in the text.

Gibbs and Kump [1994] also investigated the changes of bicarbonate fluxes on glacial/ interglacial time scales in a similar study like this one. They found that river HCO3

- fluxes during LGM may have been with about 20% greater than today because of a greater exposure of carbonate rocks on the shelves. In their study, global runoff from the ice-free regions of the continents nearly remained unchanged. In the here presented work, the individual grid point lithologies of the shelves that became land area during LGM were not taken into account in the modelling. Assuming that the continental shelves consist in large parts of carbonates, one should expected that the here presented values for FCO2-RW are somewhat too low. However, it is unlikely that this effect should have been strong enough to compensate the reduction of runoff during LGM. The fact that Gibbs and Kump [1994] found no significant changes in global runoff between present-day and LGM may be related to the method they applied. They used the precipitation minus evaporation fields predicted by a GCM model to derive runoff. The resulting fields they obtained could be highly negative over large areas on the continents, which makes physically no sense and may lead to errors. This underlines again the importance to improve hydrology in GCM models (see also chapter III).

Finally, one has also to mention that the FDOC values given in Table 28 may be too great because they were calculated with the SoilC data set for present-day. There are indications that the total amount of carbon stored in the soils may have been considerably lower during LGM than today. For example, Adams et al. [1990] estimated that the global soil carbon pool during LGM was reduced with about 50% compared to the present-day value. Also the here-reconstructed LGM conditions (Table 27) imply to some extent a reduction of the global SoilC pool because of the increase of deserts

180

and the considerable reduction of the tundra and taiga climate. We have seen in the previous chapter that the latter climate type has on average quite elevated SoilC values. Note that according to GEM-Corg, a reduction in the global amount of soil carbon as great as proposed by Adams et al. [1990] would reduce the global FDOC value for LGM with about 15%.

Table 28b. Changes in percent of the fluxes in Table 28a compared to the present-day fluxes.

______________________________________________________________________________________________________________ Area (1) Q FTSS FDOC FPOC FCO2-RW FCO2-RWc (2)

______________________________________________________________________________________________________________ Polar (no ice) 57.2 140.3 433.3 238.7 320.0 341.2 341.2 Tundra and Taiga -22.6 -42.6 50.2 -48.1 -9.3 -30.9 -45.2 Temperate Dry 4.2 37.9 -48.2 17.2 -39.2 86.0 81.8 Temperate Wet -34.1 -23.9 -13.5 -38.1 -18.3 -24.5 -20.8 Tropical Dry -4.7 -11.2 -38.4 -8.1 -36.0 18.1 18.1 Tropical Wet 4.3 -10.9 23.3 -8.8 -4.8 -19.5 -16.1 Desert 26.4 250.0 2512.5 366.7 1266.7 275.0 275.0 Total -6.4 -14.6 -4.3 -17.9 -11.8 -11.9 -10.2

Africa 4.8 -3.8 -29.8 -5.0 -11.2 27.0 53.8 Europe -25.1 -38.0 16.2 -45.8 -19.0 -14.1 -10.2 North America -47.6 -41.9 -47.5 -53.8 -42.8 -35.8 -33.7 South America 5.2 -7.6 -4.4 -6.4 -12.2 -3.0 5.0 Asia 12.9 -6.2 17.3 -5.9 3.4 -13.1 -3.1 Australia 28.9 -2.1 30.5 2.6 6.7 0.0 56.1 Antarctis -100.0 -100.0 -100.0 -100.0 -100.0 -100.0 -100.0 Total -6.4 -15.1 -2.1 -18.4 -10.6 -12.8 -7.6

Arctic Ocean -29.9 -53.4 -21.3 -49.0 -38.7 -61.2 -74.1 North Atlantic -29.0 -18.1 -18.8 -25.2 -16.5 -10.4 -9.0 South Atlantic 5.3 -8.0 27.1 -6.2 10.6 9.6 17.0 Pacific 14.1 -15.2 -4.8 -13.3 -14.7 -10.4 9.5 Indian Ocean 16.5 23.7 23.2 20.6 16.0 8.5 20.5 Mediterranean 5.2 -9.4 9.2 -4.3 -8.2 -5.3 21.2 below 60° South -100.0 -100.0 -100.0 -100.0 -100.0 -100.0 -100.0 Total -6.4 -13.6 -0.3 -17.4 -9.1 -10.7 -6.2

______________________________________________________________________________________________________________

Best estimate -6.4 -14.4 -2.2 -17.9 -10.5 -11.8 -8.0

______________________________________________________________________________________________________________ (1) Only exoreic and without ice cover. (2) Corrected according to the climatic correction factors determined in the text.

6.4.2. Other Factors Related to Continental Erosion

The application of the GEM-CO2 and GEM Corg models to LGM climate conditions implies that continental erosion did not contribute to the low atmospheric CO2 concentrations during the last glacial maximum. FCO2 is found to be lower during LGM than for present-day. One has naturally to mention here that the determined FCO2-RW values (Tab. 28) may be somewhat too low because carbonate outcrops could have been increased during LGM due to the sea level fall. But this is probably of minor importance for the atmospheric CO2 budget. In the geochemical carbon cycle, it is mainly the chemical erosion of silicate rocks which can represent a net consumption of atmospheric CO2 (see above).

An open question is, however, whether the above made assumption that no chemical weathering takes place underneath ice sheets is correct or not. When analysing melt waters from a Swiss glacier, Sharp et al. [1995] observed higher CO2 consumption rates underneath the ice-cover than in a non-glaciated catchment. If high consumption rates hold also for extended parts of the LGM ice sheets, the continental area subjected to chemical weathering was considerably greater for LGM than today. Chemical weathering could have occurred at least at the ice sheet margins, where melt waters were in

181

contact both with atmospheric CO2 and large amounts of fine grained glacial debris. This is important because it may have especially affected silicate erosion. Shales, which show among silicate rocks the greatest specific CO2 consumption, are widely outcropping in the northern latitudes north of 40° N (Amiotte-Suchet [1995]).

One has also to mention here that it is not only the amount of carbon discharged by rivers but also the oceanic response to this carbon input that determines the role of continental erosion in the glacial/ interglacial carbon cycle. If the results of Smith and Hollibaugh [1993] on the ratios of burial to respiration of riverine organic matter in the coastal zone and in the open ocean (see above) can be extrapolated to LGM, one may conclude that due to the reduced shelf area, greater amounts of riverine organic matter may have reached the open ocean, leading to lower sedimentation and greater respiration rates in the ocean compared with today. This means that even if the river inputs were lower during LGM, the amount of carbon that returned to the atmosphere may not have changed or it was even greater than for present-day. Both a reduced CO2 consumption by continental erosion and relative greater respiration rates in the ocean could have acted as a negative feedback to stabilize the Earth's climate and atmospheric composition against the documented reduction of the atmospheric CO2 concentrations during LGM.

In this context, also the study of Franzén [1994] is interesting. He proposed that peat deposits of the late Eem-interglacial may have been as great as 250-550 GtC in the regions that were later affected by the ice sheets during LGM. Because actually there is no evidence for the existence of large peat deposits at LGM, he supposed that nearly all of the Eem deposits have been destroyed and washed out by melt waters during the retreat of the ice masses at the end of LGM. This should have led to a considerable peak in organic carbon erosion, and a great amount of this carbon may have ended up in the oceans. Slow oceanic respiration of this carbon could have accounted for at least a part of the increase of the atmospheric and terrestrial carbon pools during Holocene (note that according to Smith and Hollibaugh [1993] the turnover time of organic carbon in the ocean may be in the range of 4000 to 8000 years). In effect, this increase creates a very large 'missing source' of carbon, and the carbon cycle on the glacial/ interglacial time scale is still far away to be truly understood (Adams and Faure [1996]).

6.5. Conclusions

The last chapter of this thesis evaluated the role of continental erosion within the global carbon cycle. An important element was here the coupling of the river fluxes of organic carbon with the river fluxes of inorganic carbon, and a global carbon erosion model on a grid point scale was proposed. This model combines the previous studies on the controls of river HCO3

- fluxes carried out by Amiotte-Suchet and Probst [1993a], [1993b], [1995], and Amiotte-Suchet [1995] (forming together the GEM-CO2 model) with the controls for river DOC and POC fluxes that have been identified in this study (forming together the GEM-Corg model).

To summarize these models, it has to be noted that HCO3- fluxes are mainly related to drainage

intensity together with the rock type that is drained by the surface waters. According to the stoecheometry of the weathering reactions, all carbon in the bicarbonate ions resulting from the weathering of silicate rocks originates from atmospheric/ soil CO2. For bicarbonate ions resulting from the weathering of carbonate rocks, only half of the carbon stems from the atmosphere, while the other half comes from the dissolving carbonate mineral. For DOC, fluxes are related to drainage intensity, basin slope, and the amount of organic carbon stored in the soils. POC fluxes are calculated as a function drainage intensity and of sediment yields, which can be best correlated by forming the products of hydroclimatic and geomorphological factors, that is drainage intensity, basin slope, an index characterizing rock hardness, and an index characterizing rainfall variability over the year. All carbon in the organic matter in rivers can be considered to be of atmospheric origin.

182

However, a comparison of bicarbonate fluxes calculated with GEM-CO2 with literature values for a set of major world rivers revealed that the literature values compare well with the model values in the tropical climates, while the model seems to underestimate the fluxes in the temperate and in the tundra and taiga climate types. For this reason a statistical approach was developed in this study in order to estimate the model deviation from observations with respect to climate. Even if this approach denotes nothing about the processes being responsible for this deviation, it may be considered as a way to quantify possible effects of other controlling factors than drainage intensity and lithology on river HCO3

- fluxes at global and regional scales.

When the GEM-CO2 and GEM-Corg models were applied to the total continental area on the basis of the corresponding data sets, the total amount of atmospheric CO2 which is consumed by continental erosion was calculated to be about 0.595 to 0.670 GtC/yr, depending on whether one includes the correction of the GEM-CO2 model or not. This compares well with previous literature estimates. An additional flux of 0.090 (0.120) GtC/yr comes from the dissolution of carbonate minerals (figures in brackets are the corrected ones). In the total FCO2 value, 34.6% (30.5%) can be attributed to DOC, 26.6% (23.5%) to POC, and 38.8% (46.0%) to rock weathering. Moreover, the transport of this carbon is followed by coupling the modelling with a global river routing scheme, leading to a prediction of the local inputs of the different carbon forms into the oceanic system. Both the spatial distribution of the carbon fluxes at the atmosphere/ continent as well as at the continent/ ocean interfaces could thus be drawn on a grid point scale. These distributions may be included in further modelling studies in order to better understand the lateral transports carbon in the global carbon cycle. Note that the pattern of high levels of atmospheric CO2 consumption by continental erosion coincides with a net sink of atmospheric CO2 in the Northern Hemisphere postulated by Tans et al. [1989], [1990], who did not consider river carbon fluxes in their budgets. The here calculated values show that FCO2 north of the equator may be as great as 0.5 GtC/yr (including the corrected values for GEM-CO2), which can account for about 15 to 25% of the sink postulated by the authors. This points out that continental erosion cannot be overlooked if one wants to understand the fate of the anthropogenically released CO2 in the present-day global carbon cycle.

Finally, the GEM-CO2 and GEM Corg models were applied to a LGM scenario in order to investigate whether the total amount of atmospheric CO2 consumed by continental erosion may have been different from today. The climatic boundary conditions for LGM were reconstructed by the use of a GCM simulation run, together with the empirical modelling for continental runoff that has been presented in chapter III. It is found that FCO2 during LGM was about 10 to 15% lower than today, mainly because global runoff was also reduced with about the same percentage. The reliability of these figures is naturally strongly dependent on the reliability of the GCM simulations, which are still at an early stage of development. Nevertheless, the predicted trend that the LGM world was drier than today is also confirmed by palynological, pedological, and sedimentological evidences, suggesting that continental erosion did not contribute to the low atmospheric CO2 concentrations prevailing during LGM.

An open question is, however, whether the assumption that no chemical weathering takes place underneath ice sheets is correct or not. If this is not the case, the here presented global budgets for the CO2 consumption by rock weathering would be too low, especially for the LGM budgets. More work should be devoted to study this aspect. Another important point for further investigations is to better understand the role of pCO2 in soils for weathering processes. Since it can be supposed that much of the CO2 involved in the weathering of minerals may originate from soil organic matter via biological respiration, the controls of this parameter may also have a great influence on the CO2 consumption by rock weathering. This may at least partly explain why the GEM-CO2 model seems to deviate from observations in certain climates. Unfortunately, very little is known about the spatial and temporal variability of pCO2 in soils at the global scale.

183

GENERAL CONCLUSIONS

English Version

On the previous pages, the riverine transport of organic matter has been investigated at very different scales. The spatial scale ranged from a small watershed in the French Vosges mountains of about 80 ha to the overall continental area, whereas the temporal scale varied from a single storm event during about 36 hours to a last glacial maximum scenario at 18000 yrs b.p.

Before the modelling tools were developed, which allow an extrapolation of the available data on river fluxes of organic carbon to the global scale, a detailed review of many river case studies carried out up to present was done in the first chapter of this study. Although this review essentially focused on the seasonal and interannual variability of the carbon transport, it already revealed a strong coupling between the fluxes of organic carbon and the fluxes of water and of sediments which is generally found in rivers. It became quickly evident that without an understanding of the major controls of these two parameters, all estimates for the amount of organic carbon transported by world rivers can only be as precise as the available data for the water and sediment fluxes, and it will be difficult to extrapolate these estimates to others than present-day conditions. For this reason, two chapters in this thesis were dedicated to investigate the controlling factors of continental runoff and river sediment yields at the global scale (chapter III and IV, respectively). At the same time, these chapters represent an important link between this study and many other scientific questions related to the erosion and fluvial transport of materials from the continents to the oceans.

Also for several other reasons, the review of the river case studies is an important element of this study. First, it allows to become familiar with the problems related to the determination of mean annual river fluxes, which emerge not at least because of the great interannual and long-term variations that can be found for river discharges. These problems have to be kept in mind when looking at the empirical models proposed in this study. Moreover, some particularities of certain case studies were revealed in this review, which were a useful information for the following chapters. A good example for this is the Indus River study. It was shown that the high mean annual concentrations of dissolved organic carbon (DOC) found for this river may be at least partly related to a concentration of the waters by evaporation from upstream to downstream. Such an effect cannot be accounted for in the regression analyses applied in this study, which can explain the outstanding behaviour of this river with respect to the proposed model for DOC fluxes. But probably the most important lesson that may be retained from the review of the river case studies is that in nearly all cases it was indicated that the allochthonous part in the total organic matter fluxes clearly dominates the autochthonous part. Even if it is naturally difficult to generalize because detailed biogeochemical analyses which are needed to identify the different sources of organic matter in rivers were not done in the investigated case studies, it could be shown that seasonally by far most of the organic matter is normally mobilized and washed out of the basins at the beginning and during the high-water periods, which are not favourable for the development of in-situ production in the river waters. Also in the few cases studies where additional measurements from different reaches and tributaries of the rivers are available (e.g., Niger River), it was found that the organic matter transport in the river may be rather conservative. This also implies that processes such as in-situ production, respiration, and/or sedimentation which can occur during the transport of organic matter in rivers are of minor importance for the global fluxes.

Consequently, when the available data for various world rivers on their mean annual DOC fluxes are investigated on the background of the numerous environmental parameters that could be determined in this study for the corresponding river basins, it was found that a multiple regression model including drainage intensity, slope, and the amount of carbon stored in the soils is the best

184

model to predict fluxes of dissolved organic carbon globally (chapter V). DOC fluxes become greater with increasing drainage intensities, flatter morphologies, and larger carbon reservoirs in the soils. Also this supports the general assumption that the bulk of riverine DOC is of allochthonous nature, and mainly originates from the carbon pool in the soils. However, the importance of basin slope as potential controlling factor for dissolved organic carbon fluxes has probably been overlooked up to now. Only in a few recent studies scientists became aware that morphology may also influence DOC fluxes.

For the fluxes of particulate organic carbon (POC) sediment fluxes are the dominant controlling parameter, and no relationships with additional environmental factors could be established. It was shown that POC fluxes generally increase with increasing sediment fluxes, but the percentage of organic carbon in the total suspended solids clearly decreases with increasing sediment concentrations. This relationship was mathematically fitted, allowing thus to calculate POC fluxes to the oceans as a function of sediment yield (sediment fluxes divided by drainage area) and of drainage intensity (chapter V).

Sediment yields can be best correlated by forming the products of hydroclimatic, geomorphological, and lithological factors, that is drainage intensity, basin slope, an index characterizing rock hardness, and an index characterizing rainfall variability over the year (chapter IV). The best correlated parameter combination varies to some extent when the rivers are grouped according to their average climatic situation, but it is always a combination of the above mentioned parameters that yields greatest correlation coefficients. In dry climates, however, regression coefficients are greater than in wet climates, indicating that erodibility is much greater in dry climates. When the regression models are applied to the total continental area, they produce a map of the regional variability of river sediment yields that is in good agreement with field data. This is important because the modelling approaches applied in this study are generally based upon the assumption that the fluxes which are measured at the river mouths can be explained as the sum of the fluxes taking place on the grid point levels. Because this proceeding cannot account for sedimentation processes taking place in river basins related to basin subsidence or sediment trapping in internal reservoirs such as lakes, this is certainly more problematic in the case of sediment yields, than, for example, in the case of dissolved organic carbon fluxes. The good agreement between modelled and observed variability of river sediment yields at the regional scale is therefore an argument that the factors controlling the erosion processes clearly dominate over the factors controlling the sedimentation processes for river sediment fluxes. For the total sediment flux to the oceans, a value of 14.8 Gt/yr is calculated with the pure modelling approach, and a value of 16.0 Gt/yr when the modelling is combined with an interpolation of the observed sediment yields for the 60 river basins considered in this study. These values compare well with other literature estimates.

When the here determined regression models are applied to the total ice-free and exoreic area of the continents (about 106 x 106 km2), the global amount of total organic carbon (TOC) discharged to the oceans is calculated to be 0.365 GtC/yr. The corresponding sediment flux is 16 Gt/yr, and the corresponding water flux is 41750 km3/yr. In the global TOC value, 0.205 Gt/yr can be attributed to DOC, and 0.160 GtC/yr to POC. These values agree well with previous literature estimates. Consequently, the overall empirical relationships established in this study for the fluxes of organic carbon (forming together the model GEM-Corg - Global Erosion Model for organic carbon) were coupled with an empirical modelling of the global fluxes of inorganic carbon (GEM-CO2 - Global Erosion Model for the consumption of atmospheric CO2 by rock weathering) which has been developed by Amiotte-Suchet [1995], and a global carbon erosion model on a grid point scale could thus be proposed (chapter VI). The global flux of inorganic carbon originating from atmospheric CO2 consumption by rock weathering totals 0.230 GtC/yr, and inorganic carbon that is mobilized by the dissolution of carbonate minerals makes an additional flux of 0.090 GtC/yr. All three major carbon fluxes represent a permanent transfer of carbon from the atmosphere to the oceans, a detailed picture of the consumption of atmospheric CO2 by continental erosion could thus be drawn. For all carbon forms, global maps showing the spatial distributions of the river carbon fluxes on the continents were shown, and detailed budgets for the carbon fluxes were proposed with respect to the different

185

continents, ocean basins, and major climate types. In addition, the transport of this carbon has been followed by coupling the modelling with a global river routing scheme, allowing also a prediction of the local inputs of the different carbon forms into the oceanic system on a grid point scale.

The overall amount of atmospheric CO2 consumed by continental erosion was calculated to be about 0.6 GtC/yr. This can account for about one third of the estimated net oceanic carbon uptake for present-day (Sarmiento and Sundquist [1992]) and is therefore not negligible in the global carbon cycle. The calculated spatial distribution of the river carbon fluxes can be used in further modelling studies in order to better understand the transport of carbon in the present-day carbon cycle. Up to now, it was difficult to consider the river fluxes in this context because of the lack of sufficient data world-wide. With the GEM-CO2 model developed in the framework of previous studies carried out at the CGS (Amiotte-Suchet and Probst [1993a,b], [1995], Amiotte-Suchet [1995]), already a prediction tool for the river fluxes of inorganic carbon was available. With the here presented GEM-Corg model, a prediction tool for organic carbon fluxes is added, and it is now possible to respond entirely to the data need for global river carbon fluxes. Actually, the data sets for the local inputs of river carbon to the world's oceans calculated in this study are included into test runs of two different 3-D ocean circulation models in order to improve the capacities of the models to predict the oceanic carbon cycle (in collaboration with James Orr, LMCE, and Richard Murnane, Princeton University).

Finally, it was also tested to apply the GEM-CO2 and GEM-Corg to others than present-day conditions. Here it becomes evident why the detailed investigation of the principal controlling factors for continental runoff which was done in chapter III is of fundamental importance for the study of river carbon fluxes. Both in the GEM-CO2 and GEM-Corg models, drainage intensity is the dominant controlling factor for the fluxes of inorganic and of organic carbon, but actually available global hydrological models seem to be still far away from a reliable reproduction of the observed runoff patterns at the global scale. This makes it difficult to apply carbon erosion models to scenarios of global change or for paleoclimatic studies.

Based on multiple correlation statistics, it was found that precipitation and potential evapotranspiration, which is here uniquely calculated as a function of temperature, are the principal controlling factors for continental runoff. The data in this work confirm the common feature found in many hydrological studies that mean annual runoff ratio (runoff divided by precipitation) generally increases as more as precipitation exceeds potential evapotranspiration. It is important to point out, however, that morphology also represents a non-negligible controlling factor. Given the same temperature and precipitation values, greater slope increases runoff ratio, while greater elevation decreases it. The effect of slope may be related to the fact that a uneven morphology locally favours water saturation and hence an overflow of the soils. The effect of elevation may directly influence potential evapotranspiration because atmospheric pressure decreases with increasing elevation, which may enhance evapotranspiration. Finally, it is also the seasonal precipitation distribution that influences runoff ratio. An uneven precipitation distribution may increase the possibility that soils achieve saturation on short-term, and more water is lost to runoff compared to a uniform precipitation distribution.

The empirical modelling for continental runoff together with the GEM-CO2 and GEM-Corg models was then applied to global temperature and precipitation climatologies for the last glacial maximum (LGM) derived from general circulation model (GCM) runs (chapter VI). It was found that global runoff from the ice-free exoreic regions of the continents was reduced with about 14% during LGM compared to present-day. At the same time, the river carbon fluxes were also reduced with about 10% to 15%, depending on the carbon species. This is indicating that the consumption of atmospheric CO2 during LGM was lower than today, even if it has to be mentioned that some uncertainty remains especially for the value for the atmospheric CO2 consumption by rock weathering. The lithologies of the grid points on the emerging shelves during LGM are unknown, and it is also unknown whether the made assumption that no chemical weathering takes place underneath ice sheets is correct or not.

186

The results of this study allow the development of a number of perspectives for further studies. Although it has been shown that the applied statistics are a useful tool in order to identify the general controls for the river fluxes of organic carbon, of water, and of sediments, it is evident that the proposed simple regression models cannot account for the physical and chemical processes which are responsible for the observed fluxes. Further investigations may therefore focus on the possibility to include the controlling factors identified in this study into process-based models which may better describe the concerned fluvial erosion and transport processes.

A good example for this can be the water fluxes. It seems that global water budget models will remain also in the next future the most frequently applied models for the determination of continental scale runoff, not at least because they form the hydrological modules in GCMs. Water budget models normally ignore the variable morphology of the Earth, but it has been shown in this study that morphological factors, especially basin slope, have a non-negligible role influence on runoff. For this reason, it would be interesting to test, for example, whether the inclusion of Slope can improve the prediction abilities of water budget models. Clearly, there are still considerable efforts to do on the field of global hydrology. This is important because not only erosion models, but also many other modelling approaches at the global scale (e.g., vegetation models) can be strongly dependent on reliable water budgets.

Another important point is the need of parameters which take seasonal effects into account. In this study, the parameter Four, which characterizes the variability of precipitation over the year, was retained in many regression models. But when investigating the major controls for river sediment yields, it was found that this parameter is probably only indirectly correlated with the erosion processes. A strong seasonality of precipitation may be only important when it leads to a more frequent overflow of the soils compared to a situation that is characterized by the same mean annual climatic parameters, but with a more uniform distribution of precipitation over the year. Especially seasonal data on soil moisture are needed in this respect (here again the above mentioned need of better global hydrological models may be claimed). This information could probably help to considerably improve the regression models for the prediction of sediment yields. Another important open question in this context is the role of precipitation intensity for river sediment yields, i.e. the frequency and intensity of storm events. At present, little is know about this factor at the global scale, but global data sets on this parameter may be developed in the next future, which could then also be used for a study such as this one.

The here proposed global map for sediment yields is certainly a good example where the results of this study can find many applications for other scientific problems related to continental erosion. By coupling it to global lithological or pedological maps, the oceanic input and global cycling of many other elements and substances may be investigated. For the extrapolation to other than present-day conditions, however, more should be known about the role of catastrophic events for the global fluxes. There is some evidence from the geological record that such events may have had a considerable influence on global sediment fluxes during geological times.

As far as the organic matter fluxes are concerned, it is now important to investigate whether the findings of this study can be confirmed when looking at the situation in small catchments. This study mainly extrapolated the data which have been acquired at the scale of major river basins to smaller scales, i.e. the grid point scale. It would be interesting to test whether the results are similar when the opposite proceeding is applied, i.e. when the data acquired at the scale of small catchments would be applied to larger scales. For example, this could help to answer the question whether the assumption made in this study that processes such as in-situ production or organic matter respiration during the river transport are globally of small importance for the organic carbon fluxes is true or not. Unfortunately, insufficient data on small catchments are currently available for such an approach. Moreover, even if data exist, it is often difficult use them in this respect because additional information of the environmental characteristics of the investigated catchments is missing. Further studies should therefore try to respond to this data need.

187

Finally, one has to mention that looking at small catchments may also facilitate the task to better discern the physical and chemical processes which are basically responsible the mobilization of organic matter on the level of the vegetation and soil carbon pools. This concerns also the fluxes of inorganic carbon. It is probably at the soil level that to some extent the fluxes of organic and of inorganic carbon are linked together. On the one hand, the soil organic matter may supply the bulk of the organic carbon that is found in river and stream waters. On the other hand it is probably the respiration of the same organic matter which supplies most of the CO2 which is then involved in the weathering reactions.

Version Française

Au cours de ce travail, les transports fluviaux de matière organique ont été étudiés à des échelles différentes. L'échelle d'espace s'étend d'un petit versant de 80 ha dans les Vosges jusqu'à l'ensemble des surfaces continentales et l'étude dans le temps va d'une crue d'environ 36 heures jusqu'à un scénario pour le dernier maximum glaciaire (18000 ans b.p.).

Avant de développer la modélisation empirique qui nous a permis d'extrapoler à l'échelle globale les données disponibles sur les transports fluviaux de carbone organique, nous avons présenté une revue détaillée de plusieurs cas d'études réalisés jusqu'à présent. Cette revue qui porte essentiellement sur la variabilité saisonnière et interannuelle de transports fluviaux de carbone a confirmé l'étroite relation qui existe entre les flux de carbone organique et les flux d'eaux et de sédiments. Il est ainsi rapidement apparu qu'il était essentiel de comprendre les processus qui contrôlent ces deux paramètres pour extrapoler les données disponibles à des périodes climatiquement différentes de la période actuelle. C'est pour cette raison que deux chapitres de cette thèse ont été dédiés à l'étude des facteurs de contrôle du drainage continentale et des transports de sédiments à l'échelle globale (chapitre III and IV, respectivement). Ces deux chapitres devraient permettre aussi de faire le lien avec d'autres problèmes scientifiques liés à l'érosion continentale et aux transports fluviaux vers les océans.

Cette revue de différents cas d'études est aussi un élément important de ce travail pour plusieurs autres raisons. Tout d'abord, elle permet de se familiariser avec les problèmes liés à la détermination des flux moyens annuels dans les fleuves qui sont dus, entre autre, aux variations interannuelles des débits des fleuves. Ces problèmes doivent être gardés en tête quand la modélisation empirique est abordée dans les différents chapitres de cette étude. De plus, cette revue de cas d'études a permis de mettre en évidence certaines particularités qui peuvent être utiles pour la compréhension des chapitres qui suivent cette revue. A ce titre, l'étude du bassin versant de l'Indus est un bon exemple. Nous avons pu montrer que les fortes concentrations moyennes annuelles en carbone organique dissous (COD) pouvaient être en parti attribuées pour ce fleuve à une forte concentration des eaux par évaporation en allant de l'amont vers l'aval du bassin. Les processus de ce genre ne peuvent pas être pris en compte dans les régressions qui ont été appliquées dans cette étude; ce qui explique que ce fleuve n'a pas été pris en compte dans la modélisation des flux de carbone organique dissous. Cependant, la leçon la plus importante que nous puissions probablement tirer de cette revue est que dans la plupart des cas d'études la contribution du carbone organique d'origine allochtone domine celle d'origine autochtone. Même s'il est difficile ici de généraliser car des analyses biogéochimiques plus détaillées sont nécessaires pour identifier les différentes sources de matière organique dans les fleuves, nous devons souligner que la matière organique est normalement mobilisée et exportée saisonnièrement en grande partie au début et pendant les périodes des hautes eaux, périodes qui ne sont pas favorables au développement de phytoplancton dans les fleuves. Ces cas d'études nous ont aussi montrés, notamment quand des mesures étaient disponibles pour les différents affluents (par exemple le Niger), que le transport de matière organique dans un fleuve pouvait être relativement conservatif. Ce qui signifie que les processus comme la production de matière organique, la respiration ou la sédimentation de ces matières dans le fleuve, sont relativement mineurs par rapport aux flux totaux exportés vers les océans.

188

L'aspect conservatif du comportement du transport de la matière organique dissoute à l'échelle globale a été également confirmé par la recherche des facteurs qui contrôlent les flux moyens annuels (chapitre V). En effet, les données de COD pour de nombreux grands fleuves du monde ont été étudiées en relation avec les différents paramètres environnementaux qui ont pu être déterminées dans cette étude pour les différents bassins versants correspondants. Nous avons ainsi pu calculé un modèle de régression de type régression multiple entre les flux de carbone organique dissous et les différents paramètres environnementaux ; le meilleur modèle à l'échelle globale est celui qui prends en compte l'intensité de drainage, les pentes et le stock de carbone dans le sol. Ainsi les flux de COD augmentent avec l'intensité de drainage, avec les pentes les plus faibles et des stocks de carbone organique les plus importants dans les sols. Ce résultat confirme en quelque sorte l'hypothèse précédente selon laquelle le transport fluvial de COD est principalement d'origine allochtone et fourni par le carbone des sols. Cependant, l'importance des pentes qui apparaît comme un facteur potentiel de contrôle du flux de carbone organique dissous a probablement été sous-estimée jusqu'à maintenant et ce n'est que très récemment que quelques études ont montré l'influence de la morphologie du relief sur les flux de COD.

En ce qui concerne le flux de carbone organique particulaire (COP), le facteur de contrôle dominant est le flux de sédiment ; aucune autre relation avec les différents facteurs environnementaux n'a pu être déterminée. Nous avons pu montré ainsi que les flux de COP augmentent généralement avec les flux de sédiments, mais le pourcentage de carbone organique dans les matières en suspension diminue de manière évidente quand la concentration en sédiments augmente. L'ajustement d'un modèle mathématique à cette relation nous a permis de calculer les flux de COP aux océans en fonction des transports spécifiques de sédiments et de l'intensité de drainage (chapitre V).

Les flux des sédiments on pu être modélisés par un produit de facteurs hydroclimatiques, géomorphologiques et lithologiques, facteurs représentés dans cette étude par l'intensité de drainage, un index caractérisant la variabilité saisonnière des précipitations, la pente des bassins et un index caractérisant la dureté des roches (chapitre IV). Le produit corrélant le mieux ces différents paramètres varie quelque peu si les fleuves sont regroupés en fonction de leurs caractéristiques climatiques moyennes, mais c'est toujours la combinaison des différents facteurs mentionnés ci-dessus qui donne les meilleurs coefficients de corrélation. Pour les climats secs, cependant, les coefficients de régression sont plus élevés que pour les climats humides, indiquant ainsi que l'érodibilité des bassins est plus élevée sous climat sec. Si ces modèles de régression sont appliqués à l'ensemble des surfaces continentales, une carte globale des flux spécifiques de sédiments est obtenue, avec une variabilité régionale en bon accord avec les données de terrain. Ce résultat est important car l'approche qui a été utilisée dans cette étude est fondée sur l'hypothèse selon laquelle les flux mesurés à l'exutoire d'un bassin peuvent s'exprimer comme la somme des flux provenant des différents mailles résultant du découpage de ce bassin. Dans la mesure où cette procédure ne prends pas en compte le processus de sédimentation qui se développe dans les bassins versants à cause de la subsidence ou du piégeage des sédiments dans des réservoirs internes comme les lacs, cette approche est certainement plus problématique quand elle s'applique aux flux des sédiments que, par exemple, aux flux de carbone organique dissous. La bonne correspondance qui existe entre la variabilité modélisée à l'échelle régionale et celle observée est cependant un bon argument montrant que les transports fluviaux des sédiments sont plutôt déterminés par les facteurs qui contrôlent les processus d'érosion que par ceux qui contrôlent les processus de sédimentation. Le flux total de sédiments vers les océans est estimé à 14,8 Gt/an par cette approche de modélisation, et à 16,0 Gt/an si cette modélisation est combinée avec une interpolation des données de la littérature pour les 60 bassins qui ont été considérés dans cette étude. Ces valeurs sont en bon accord avec d'autres estimations trouvées dans la littérature.

Quand les modèles de régression déterminés dans cette étude sont appliqués à l'ensemble des surfaces continentales exoreiques et non-couvertes par la glace (environ 106 x 106 km2), la quantité totale de carbone organique exportée vers les océans est estimée à 0,365 GtC/an, dont 0,205 GtC/an sous forme de COD et 0,160 GtC/an sous forme de COP. Le flux de sédiment correspondant est de 16 Gt/an et le flux d'eau de 41750 km3/an. Ces valeurs sont aussi en bon accord avec les précédentes estimations trouvées dans la littérature. Par conséquence, l'ensemble des relations empiriques

189

déterminées dans cette étude pour les flux de carbone organique (qui forment le modèle GEM-Corg - Global Erosion Model pour le carbone organique) ont été couplées à une modélisation empirique des flux de carbone inorganique (GEM-CO2 - Global Erosion Model pour la consommation de CO2 atmosphérique par l'altération des roches) développée récemment par Amiotte-Suchet [1995], permettant ainsi de proposer un modèle d'érosion de carbone à l'échelle globale (chapitre VI). Le flux global de carbone atmosphérique consommé par l'altération chimique des roches est ainsi calculé à 0,230 GtC/an, et le flux de carbone inorganique issu de la dissolution des carbonates est de 0,090 GtC/an. Comme les trois principaux flux de carbone représentent un transfert permanent de carbone de l'atmosphère / sols vers les océans, nous pouvons fournir une cartographie détaillée de la consommation de CO2 atmosphérique par l'érosion des continents. Pour chaque forme de carbone, des cartes globales montrant la distribution spatiale des transports fluviaux de carbone sur les continents sont proposées ainsi que des bilans détaillés de ces flux de carbone pour les différents continents, les principaux climats et pour les différents océans. De plus, ces transports de carbone ont été couplés à un schéma de circulation des eaux continentales, permettant ainsi d'estimer les apports régionaux des différentes formes de carbone vers les océans avec une résolution suffisamment fine.

L'ensemble des flux de CO2 atmosphérique consommés par l'érosion continentale est estimé à 0,6 GtC/an. Ce flux représente environ un tiers du flux net des échanges estimé à l'heure actuelle entre l'océan et l'atmosphère (Sarmiento et Sundquist [1992]) et il n'est donc pas négligeable dans le cycle global du carbone. Les distributions spatiales calculées dans cette étude concernant les flux de carbone transportés par les fleuves peuvent être utilisées dans d'autres modélisations pour mieux comprendre les transports latéraux de carbone dans le cycle global actuel de carbone. En effet, il était difficile jusqu'à présent de prendre en compte les transports des fleuves dans cette modélisation parce qu'il n'y avait pas suffisamment de données à l'échelle globale. Avec le modèle GEM-CO2 développé dans des études antérieures au CGS (Amiotte-Suchet et Probst [1993a,b], [1995], Amiotte-Suchet [1995]), un outil de prédiction des flux de carbone inorganique transportés par les fleuves vers les océans existait déjà. Avec le modèle GEM-Corg présenté dans cette étude, un outil de prédiction des flux de carbone organique est ajouté et il devient maintenant possible de répondre entièrement au besoin concernant l'ensemble des flux de carbone transportés par les fleuves. Actuellement, les données produites dans cette étude concernant la régionalisation des apports aux océans sont interprétées dans des modèles de circulation océanique en 3D pour améliorer les capacités de ces modèles à prédire le cycle du carbone océanique (en collaboration avec James Orr, LMCE, and Richard Murnane, Princeton University).

Enfin nous avons aussi essayé d'appliquer les modèles GEM-CO2 et GEM-Corg à d'autres conditions climatiques que l'actuel. Il devient alors évident que l'étude détaillée que nous avons réalisée sur les facteurs de contrôle du drainage continentale (chapitre III) est fondamentale pour simuler les transports fluviaux de carbone. Dans les deux modèles, GEM-CO2 et GEM-Corg, l'intensité du drainage est le principal facteur de contrôle des flux de carbone inorganique et organique. Cependant, les modèles hydrologiques actuellement disponibles qui ont été développés à l'échelle globale sont encore loin de produire des valeurs de drainage continentale en accord avec les données observées à l'échelle globale sur les grands bassins fluviaux du monde. Ce qui rend difficile leur utilisation dans nos modèles d'érosion quand des scénarios de changements globaux ou des reconstitutions paléoclimatiques sont envisagés.

En appliquant des procédures de régressions multiples, nous avons montré dans cette étude que les précipitations et l'évapotranspiration potentielle - qui est ici uniquement calculée en fonction de la température - sont les facteurs principaux de contrôle du drainage à l'échelle globale. Les données utilisées et les résultats obtenus dans cette étude permettent de confirmer un résultat déjà obtenu dans de nombreuses études hydrologiques, à savoir que le coefficient d'écoulement (le rapport entre l'écoulement fluvial et les précipitations) augmente d'autant plus que les précipitations dépassent l'évapotranspiration potentielle. Cependant, il est important de souligner que la morphologie représente également un facteur de contrôle non-négligeable pour le drainage. Dans les mêmes conditions de température et de précipitations, de fortes pentes augmentent le coefficient d'écoulement, tandis que des élévations plus fortes le diminuent. L'effet des pentes peut être expliquer par le fait qu'une morphologie plus pentue favorise localement la saturation en eaux des sols. L'effet de l'élévation

190

devrait influencer directement l'évapotranspiration potentielle, car la pression atmosphérique diminue quand l'élévation augmente. Enfin la distribution saisonnière des précipitations influence aussi le coefficient d'écoulement. Des périodes de fortes précipitations dans l'année augmentent la probabilité que les sols soient rapidement saturés et qu'une quantité d'eau plus importante s'écoule, par rapport à une meilleure répartition des précipitations tout au long de l'année.

Les résultats obtenus dans cette thèse permettent d'envisager un certain nombre de perspectives pour des études futures. Bien qu'il ait été montré que les méthodes statistiques sont un outil très précieux pour identifier les facteurs qui contrôlent de manière générale les flux de carbone organiques, d'eaux et de sédiments, il est évident que les modèles de régression proposés dans cette étude ne peuvent rendre compte des processus physico-chimiques responsables des flux observés. Les études à venir devraient ainsi s'efforcer d'inclure les facteurs de contrôle qui ont été identifiés au cours de cette étude dans des modèles phénoménologiques qui pourraient donner une meilleure description des processus d'érosion et de transport par les fleuves.

Un bon exemple pourrait être constitué par les flux d'eaux. Il semble que les modèles de bilans hydrologiques globaux resteront encore dans le futur les modèles les plus couramment appliqués pour déterminer les débits à l'échelle continentale, notamment parce qu'ils constituent les modules hydrologiques dans les modèles GCM (General Circulation Models). Les modèles de bilans hydrologiques ignorent généralement la morphologie variable de la Terre. Il a été montré dans cette étude que les facteurs morphologiques, en particulier la pente, jouent un rôle non négligeable au niveau des débits. Il serait intéressant de tester par exemple si la prise en compte de ce paramètre améliorerait les capacités de prévision des modèles de bilans hydrologiques. Il apparaît que des efforts considérables restent à faire dans le domaine de l'hydrologie globale. Ceci est important car non seulement les modèles d'érosion, mais aussi d'autres approches de modélisation à l'échelle globale (par exemple, les modèles de végétation) exigent des bilans hydriques fiables.

Une autre nécessité est d'inclure dans les modèles des paramètres qui rendent compte des effets saisonniers. Dans cette étude, le paramètre Four, caractérisant la variabilité des précipitations au cours d'une année, a été retenu dans de nombreux modèles de régression. Toutefois, en considérant les principaux facteurs de contrôle pour les flux spécifiques de sédiments, on peut penser que ce paramètre n'est probablement qu'indirectement lié aux processus d'érosion. Un fort caractère saisonnier des précipitations peut être important seulement quand cela conduit à une inondation plus fréquente des sols par rapport à une situation caractérisée par les mêmes paramètres climatiques annuels moyens, mais par une distribution plus uniforme des précipitations au cours de l'année. Des données de taux d'humidité dans les sols sont requises pour traiter ce problème. A nouveau, le besoin de meilleurs modèles hydrologiques globaux apparaît. Une telle information pourrait également être utile pour améliorer les modèles de régression pour la modélisation des flux de sédiments. Un autre aspect important à ce niveau est le rôle joué par l'intensité des précipitations pour les flux de sédiments dans les rivières, c'est-à-dire le rôle joué par la fréquence et l'intensité des événements orageux. A l'heure actuelle, on sait peu de choses à propos de ce facteur à l'échelle globale, mais des séries de données globales peuvent être développées pour ce paramètre dans le futur, et pourraient alors être utilisées dans le cadre d'une étude telle que celle-ci.

La carte globale proposée ici pour les flux de sédiments constitue certainement un bon exemple où les résultats de cette étude pourraient trouver de nombreuses applications pour d'autres problèmes scientifiques liés à l'érosion des continents. En l'associant à des cartes lithologiques ou pédologiques globales, on peut étudier les apports aux océans et le cycle global de beaucoup d'autres éléments ou substances. Pour une extrapolation à des conditions autres que les conditions actuelles, il faudrait en savoir plus sur le rôle des événements catastrophiques pour les flux de sédiments à l'échelle globale. Il y a quelques indications dans l'enregistrement géologique que de tels événements puissent avoir influencé considérablement les flux de sédiments globaux au cours de temps géologiques.

En ce qui concerne les flux de matière organique, il est important de vérifier si les résultats obtenus dans ce travail peuvent être confirmés par l'étude de petits bassins versants. Nous avons ici

191

surtout extrapolé des données acquises sur les grands bassins fluviaux à des plus petites échelles, c'est-à-dire à l'échelle de la maille. Il serait intéressant de vérifier si nous obtenons des résultats semblables en appliquant la démarche inverse, c'est-à-dire en appliquant à plus grande échelle des données acquises sur de petits bassins versants. Cela permettrait par exemple de vérifier l'hypothèse faite dans cette étude selon laquelle des processus tels que la production in-situ ou la respiration de la matière organique ont qu'une importance limitée pour les flux de carbone organique globaux. Malheureusement, nous ne disposons actuellement sur des petits bassins versants que des données insuffisantes pour une telle approche. De plus, même si ces données existent parfois, leur utilisation est rendue difficile par le manque d'information au niveau des caractéristiques environnementales des sites étudiés. Il est donc nécessaire d'initier des études pour répondre à toutes ces questions.

Finalement, il faut souligner que l'étude de petits bassins versants peut aussi aider à discerner les processus physico-chimiques qui sont à l'origine de la mobilisation de la matière organique au niveau de la végétation et des sols. Ceci s'applique également aux flux de carbone inorganique. Il est probable que les flux de carbone organique et inorganique soient liés dans une certaine mesure au niveau des sols. D'une part, la matière organique du sol fournit la majeure partie du carbone organique qui se retrouve dans les rivières. D'autre part, il est probable que la respiration de cette même matière organique fournisse la majeure partie du CO2 impliqué dans les réactions d'altération.

193

REFERENCES

Abramopoulos, F., Rosenzweig, C. and Choudhury, B., 1988. Improved ground hydrology calculations for global climate models (GCMs): Soil water movement and evapotranspiration. J. Climate, 1, 921-941.

Abu el Ella, E.M., 1993. Preliminary studies on the geochemistry of the Nile river basin, Egypt. In: S. Kempe, D. Eisma and E.T. Degens (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 6. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 74, Hamburg, pp. 115-134.

Adams, J.M. and Faure, H., 1996. Changes in moisture balance between glacial and interglacial conditions: influence on carbon cycle processes. In: J. Branson, A.G. Brown and K.J. Gregory (Editors), Global Continental Changes: the Context of Palaeohydrology. Geol. Soc. Spec. Publ. 115, pp. 27-42.

Adams, J.M., Faure, H., Laure-Denard, L. McGlade, J.M. and Woodward, F.I., 1990. Increases in terrestrial carbon storage from the last glacial maximum to the present. Nature, 348, 711-714.

Ahamer, G., Spitzer, J., Weiss, C.O. and Fankhauser, G., 1992. Der Einfluß einer verstärkten energetischen Biomassennutzung auf die CO2-Konzentration in der Atmosphäre. Abschlussbericht. Institut für Energieforschung, Joanneum Research, Graz.

Ahnert, F., 1970. Functional relationships between denudation, relief and uplift in large mid-latitude drainage basins. Amer. J. Sci., 268, 243-263.

Amiotte-Suchet, P., 1995. Cycle du carbone, érosion chimique des continents et transferts vers les océans. Sci. Géol. Mém., Strasbourg, 97, 156 pp.

Amiotte-Suchet, P. and Probst, J.L., 1993a. Modeling of atmospheric CO2 consumption by chemical weathering of rocks: Application to the Garonne, Congo and Amazon basins. Chemical Geology, 107, 205-210.

Amiotte-Suchet, P. and Probst, J.L., 1993b. Flux de CO2 consommé par altération chimique continentale: influence du drainage et de la lithologie. C. R. Acad. Sci. Paris, 317, 615-622.

Amiotte-Suchet, P. and Probst, J.L., 1995. A global model for present day atmospheric/ soil CO2 consumption by chemical erosion of continental rocks (GEM-CO2 ). Tellus, 47B, 273-280.

Arain, R., 1985. Carbon and mineral transport of Indus River 1982-1983. In: E.T. Degens, S. Kempe and R. Herrera (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 3. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 58, Hamburg, pp. 487-494.

Arain, R., 1987. Persisting trends in carbon and mineral transport monitoring of the Indus River. In: E.T. Degens, S. Kempe and Gan Wei-Bin (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 4. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 64, Hamburg, pp. 417-421.

Arain, R., 1988. Carbon transport by Indus River - An overview after five years (1981-1985) of monitoring of river discharge to the Arabian Sea. In: E.T. Degens, S. Kempe and A.S. Naidu (Editors), Transport of Carbon and Minerals in Major World Rivers, Lakes and Estuaries, Pt. 5. Mitt. Geol.-Paleont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 66, Hamburg, pp. 93-97.

Arain, R., and Khuhawar, M.Y., 1982. Carbon transport in the Indus River: preliminary results. In: E.T. Degens (Editor), Transport of Carbon and Minerals in Major World Rivers, Pt. 1. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 52, Hamburg, pp. 449-456.

Bagrov, N.A., 1953. On the average long term evaporation from the land surface (in Russian). Meteorol. Gidrol., 10, 20-25.

194

Barber, L.B., Leenheer, J.A., Pereira, W.E., Noyes, T.I., Brown, G.K., Tabor, C.F. and Writer, J.H., 1995. Organic contaminants of the Mississippi River from municipal and industrial wastewater. In: R.H. Meade (Editor) Contaminants in the Mississippi River, 1987-92. U.S. Geological Survey Circular 1133, U.S. Geological Survey, Denver, pp.115-135.

Barnola J.M., Raynaud D., Korotkevitch Y.S. and Lorius C., 1987. Vostok ice core provides 160 000 year record of atmospheric CO2. Nature, 329, 408-414.

Baumgartner, A. and Reichel, E., 1975. The World Water Balance. Elsevier, New York, 177 pp. Berner, R.A., 1989. Biogeochemical cycles of carbon and sulfur and their effect on atmospheric

oxigen over Phanerozoic time. Pal. Pal. Pal., 75, 97-122. Berner, R.A., 1991. A model for atmospheric CO2 over Phanerozoic time. Amer. J. Sci., 291, 339-

376. Berner, R.A., Lasaga, A.C. and Garrels, R.M., 1983. The carbonate-silicate geochemical cycle and

its effect on atmospheric carbon dioxide over the past 100 millions years. Amer. J. Sci., 283, 641-683.

Biksham, G. and Subramanian, V., 1988. Nature of solute transport in the Godavari basin, India. J. Hydrol., 103, 375-392.

Bird, M.I., Fyfe, W.S., Pinheiro-Dick, D. and Chivas, A.R., 1992. Carbon isotope indicators of catchment vegetation in the Brazillian Amazon. Global Biogeochem. Cycles, 6, 293-306.

Bluth G.J.S. and Kump L.R, 1994. Lithologic and climatologic controls of river chemistry. Geochim. Cosmochim. Acta, 58, 2341-2359.

Boeglin, J. and Probst, J.L, 1996. Transports fluviaux de matières dissoutes et particulaires sur un bassin versant en région tropicale: le bassin amont du Niger au cours de la période 1990-1993. Sci. Géol. Mém., Strasbourg, in press.

Bolin, B., Degens, E.T., Duvigneaud, P. and Kempe, S., 1979. The global biogeochemical carbon cycle. In: B. Bolin, E.T. Degens, S. Kempe and P. Ketner (Editors), The Global Carbon Cycle, SCOPE Rep. 13. John Wiley, New York, pp. 1-56.

Bryers, G.G. 1985. Water quality of lake Taupo and the Waikato River. In: E.T. Degens, S. Kempe and R. Herrera (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 3. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 58, Hamburg, pp. 525-537.

Budyko, M.I., 1950. Klimaticheskie faktorii vneshnego fisiko-geographicheskego protzessa (Climatic factors of external physico-geographical processes). Trudy Glavnaia Geofizicheskaia Observatoria, 19, 25-40.

Budyko, M.I., 1974. Climatic factors of geographical zonality in climate and Life. Academic, San Diego, Calif., pp. 317-340.

Budyko, M.I. and Zubenok, L.I., 1961. Ob opredelenii ispareniya s poverkhnosti sushi (On the determination of evaporation from the land surface). Izv. Akad. Nauk SSSR Ser. Geogr., 6, 3-17.

Burchfiel, C.B., 1983. The continental crust. Sci. Am., 249, 86-98. Cadée, G.C., 1987. Organic carbon in the Ems River and estuary: A comparison of summer and

winter conditions. In: E.T. Degens, S. Kempe and Gan Wei-Bin (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 4. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 64, Hamburg, pp. 359-374.

Carbon Cycle Research Unit, 1982. The amount of carbon transported to the sea by the Yangtze and Huanghe rivers (People's Republic of China) during the half year July-December, 1981. In: E.T. Degens (Editor), Transport of Carbon and Minerals in Major World Rivers, Pt. 1. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 52, Hamburg, pp. 437-448.

Carvalho, N.D., 1988. Sediment yield in the Velhas River basin (Minais Gerais, Brazil). In: M.P. Bordas and D. Walling (Editors), Sediment Budgets. Proceedings on the Porto Alegre Symposium. IAHS Publ., 174, IAHS Press, Wallingford, pp. 369-375.

Cauwet, G. and Mackenzie, F.T., 1993. Carbon inputs and distribution in estuaries of turbid rivers: The Yang Tse and Yellow rivers (China). Mar. Chem., 43, 235-246.

195

Charania, S.H., 1988. A strategy for organizing a sediment data collection network based on the available hydrological records for a catchment in Kenia. In: M.P. Bordas and D. Walling (Editors) Sediment Budgets. Proceedings on the Porto Alegre Symposium. IAHS Publ., 174, IAHS Press, Wallingford, pp. 181-195.

Clair, T.A., Pollock, T.L. and Ehrman, J.M., 1994. Exports of carbon and nitrogen from river basins in Canada's Atlantic Provinces. Global Biogeochem. Cycles, 8, 441-450.

Claussen, M., Lohman, U., Roeckner, E. and Schulzweida, U., 1994. A global data set of land-surface parameters. MPI report 135, Max-Planck-Institut für Meteorologie, Hamburg, 23 pp.

CORINE, 1992. Soil Erosion Risk and Important Land Resources in the Southern Regions of the European Community. Report EUR 13233, Office for Official Publications of the European Communities, Brussels, 99 pp.

Cossa, D. and Tremblay, G., 1983. Major ions composition of the St. Lawrence River: Seasonal variability and fluxes. In: E.T. Degens, S. Kempe et H. Soliman (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 2. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 55, Hamburg, pp. 253-259.

Dahm, C.N., Gregory, S.V. and Park, P.K., 1981. Organic carbon transport in the Columbia River. Estuarine Coastal Shelf Sci., 13, 645-658.

Degens, E.T., 1982. Transport of Carbon and Minerals in Major World Rivers, Pt. 1. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 52, Hamburg, 766 pp.

Degens, E.T. and Ittekkot, V., 1985. Particulate organic carbon - an overview. In: E.T. Degens, S. Kempe and R. Herrera (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 3. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 58, Hamburg, pp. 7-27.

Degens, E.T., Kempe, S. and Soliman, S., 1983. Transport of Carbon and Minerals in Major World Rivers, Pt. 2. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 55, Hamburg, 535 pp.

Degens, E.T., Kempe, S. and Spitzy, A., 1984. Carbon dioxide: A biogeochemical portrait. In C.O. Hutzinger (Editor), The Handbook of Environmental Chemistry, vol. 1. Springer-Verlag, Berlin, pp. 127-215.

Degens, E.T., Kempe, S. and Herrera, R., 1985. Transport of Carbon and Minerals in Major World Rivers, Pt. 3. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 58, Hamburg, 645 pp.

Degens, E.T., Kempe, S. and Gan Wei-Bin, 1987. Transport of Carbon and Minerals in Major World Rivers, Pt. 4. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 64, Hamburg, 512 pp.

Degens, E.T., Kempe, S. and Naidu, A.S., 1988. Transport of Carbon and Minerals in Major World Rivers, Pt. 5. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 66, Hamburg, 422 pp.

Degens, E.T., Kempe, S. and Richey, J.E., 1991a. Biogeochemistry of Major World Rivers. SCOPE Rep. 42. John Wiley, New York, pp. 356.

Degens, E.T., Kempe, S. and Richey, J.E., 1991b. Summary: Biogeochemistry of major world rivers. In: E.T. Degens, S. Kempe and J.E. Richey (Editors), Biogeochemistry of Major World Rivers. SCOPE Rep. 42. John Wiley, New York, pp. 323-347.

Delaney, M.L. and Boyle, E.A., 1988. Tertiary paleoceanic chemical variability: Unintended consequences of simple geochemical models. Palaeoceanography, 3, 137-156.

Depetris, P.J., 1980. Hydrochemical aspects of the Negro River, Patagonia, Argentina. Earth Surf. Processes, 5, 181-186.

Depetris, P.J. and Cascante, E.A., 1985. Carbon transport in the Paraná River. In: E.T. Degens, S. Kempe and R. Herrera (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 3. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 58, Hamburg, pp. 299-304.

Depetris, P.J. and Cascante, E.A., 1987. Hydrochemical transport of selected heavy metals in the Paraná River (Argentina). In: E.T. Degens, S. Kempe and Gan Wei-Bin (Editors), Transport

196

of Carbon and Minerals in Major World Rivers, Pt. 4. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 64, Hamburg, pp. 339-347.

Depetris, P.J. and Kempe, S., 1990. The impact of the EL Niño 1982 event on the Paraná River, its discharge and carbon transport. Pal. Pal. Pal., 89, 239-244.

Depetris, P.J. and Kempe, S., 1993. Carbon dynamics and sources in the Paraná River. Limnol. Oceanogr., 38, 382-395.

Depetris, P.J. and Lenardón, M.L., 1982. Particulate and dissolved Phases in the Paraná River. In: E.T. Degens (Editor), Transport of Carbon and Minerals in Major World Rivers, Pt. 1. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 52, Hamburg, pp. 385-395.

Depetris, P.J. and Lenardón, M.L., 1983. A second report on: Particulate and dissolved Phases in the Paraná River. In: E.T. Degens, S. Kempe and H. Soliman (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 2. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 55, Hamburg, pp. 167-181.

Depetris, P.J. and Paolini, J., 1991. Biogeochemical aspects of South American rivers: The Paraná and the Orinoco. In: E.T. Degens, S. Kempe and J.E. Richey (Editors), Biogeochemistry of Major World Rivers. SCOPE Rep. 42. John Wiley, New York, pp. 105-125.

Desjardins, T., Volkoff, B., Andreux, F. and Cerri, C., 1991. Distribution du carbone total et de l'isotope 13C dans les sols ferrallitiques du Brésil. Sci. Sol, 29, 175-187.

Dooge, J.C.I., 1992. Hydrological models and climate change. J. Geophys. Res., 97, D3, 2677-2686.

Douglas, I., 1967. Man, vegetation, and the sediment yield of rivers. Nature, 215, 925-928. Dubus, P., 1989. Paramétrisation des grands bassins fluviaux du monde. Modélisation de l'érosion

mécanique et de l'érosion chimique des continents. Mém. de DEA, Université Louis Pasteur, Strasbourg, 30 pp.

Dümenil, L., Isele, K., Liebscher, H.J.,Schröder, U., Schumacher, M., and Wilke, K., 1993. Discharge data from 50 selected rivers for GCM validation. MPI report 100, Max-Planck-Institut für Meteorologie, Hamburg, 61 pp.

Eckhardt, B.W. and Moore, T.R., 1990. Controls on dissolved organic carbon concentrations in streams, southern Québec. Can. J. Fish. Aquat. Sci., 47, 1537-1544.

Eisma, D., Cadée, G.C. and Laane, R., 1982. Supply of suspended matter and particulate and dissolved organic carbon from the Rhine to the coastal North Sea. In: E.T. Degens (Editor), Transport of Carbon and Minerals in Major World Rivers, Pt. 1. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 52, Hamburg, pp. 483-506.

El Gh'mari, A., 1995. Etude de la dynamique d'altération d'un granite soumis à des précipitations atmosphériques acides. Thèse de Doctorat de l'Université Louis Pasteur, Strasbourg.

Emanuel, W.R., Shugart, H.H. and Stevenson, M.P., 1985. Climatic change and the broad-scale distribution of terrestrial ecosystem complexes. Climatic Change, 7, 29-43.

Esser, G., 1991. Osnabrück Biosphere Model: Structure, construction, results. In: G. Esser and D. Overdieck (Editors), Modern Ecology. Elsevier, New York, pp. 679-709.

Esser, G. and Kohlmaier, G.H., 1991. Modeling terrestrial sources of nitrogen, phosphorus, sulphur and organic carbon to rivers. In: E.T. Degens, S. Kempe and J.E. Richey (Editors), Biogeochemistry of Major World Rivers. SCOPE Rep. 42. John Wiley, New York, pp. 297-322.

Esser, G. and Lautenschlager, M., 1994. Estimating the change of carbon in the terrestrial biosphere from 18000 b.p. to present using a carbon cycle model. Envionm. Poll., 83, 45-53.

Eswaran, H., Van Den Berg, E. and Reich, P., 1993. Organic carbon in soils of the world. Soil Sci. Soc. Am. J., 57, 192-194.

Etchanchu, D. et Probst, J.L., 1986. Erosion et transport de matières en suspension dans un bassin versant en région agricole. Méthode de mesure du ruissellement superficiel, de sa charge et des deux composantes du transport solide dans un cours d'eau. C. R. Acad. Sci. Paris, 302, 1063-1068.

Etchanchu, D. and Probst, J.L., 1988. Evolution of the chemical composition of the Garonne river water during the period 1971-1984. Hydrol. Sci. J., 33, 243-256.

197

Etcheber, H. 1986. Biogeochimie de la matière organique en milieu estuarien: comportement, bilan, proprietés. Cas de la Gironde. Mém. Inst. Géol. Bas. Aquitaine, Univ. Bordeaux, 19, 379 pp.

Ewel, J.J. and Whitmore, J.L., 1973. The ecological life zones of Puerto Rico and the U.S. Virgin Islands. Forest Service Research Paper ITF-18. Inst. of Tropical Forestry, USDA, Rio Piedro, Puerto Rico.

Fairbanks, R.G., 1989. A 17000 year glacio-eustatic sea level record: influence of glacial melting rates on the Younger Dryas event and deep-ocean circulation. Nature, 342, 637-642.

FAO (Food and Agriculture Organization/ U.N. Educational, Scientific, and Cultural Organization), 1971, 1975, 1976, 1978, 1979, 1981. Soil map of the World. Vol. I to Vol. X. UNESCO Press, Paris.

FNOC (Fleet Numerical Oceanography Center), 1992. Global elevation, terrain, and surface characteristics. Digital raster on a 10 minute geographic (lat/long) 1080x2160 grid. In: NOAA/ National Geopysical Data Center (Editor), Global Ecosystems Database, Version 1.0 Disc (CD-ROM). National Geopysical Data Center, Boulder, Colorado.

Fournier, F., 1960. Climat et érosion. Presses Universitaires de France, Paris, 201 pp. Frakes, L.A., Probst, J.L. and Ludwig, W., 1994. Latitudinal distribution of paleotemperature on

land and sea from early Cretaceous to middle Miocene. C. R. Acad. Sci. Paris, 318, 1209-1218.

Franzén, L., 1994. Are wetlands the key to the ice-age cycle enigma? Ambio, 23, 300-308. Gan Wei-Bin, Chen Hui-Min and Han Yun-Fang, 1983. Carbon transport by the Yangtse (at

Nanjing) and Huanghe (at Jinan) rivers, Peoples Republic of China. In: E.T. Degens, S. Kempe and S. Soliman (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 2. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 55, Hamburg, pp. 459-470.

Garrels, R.M. and Mackenzie, F.T., 1971. Evolution of sedimentary rocks. W. W. Norton and Co. Inc., New York, 397 pp.

Gibbs, M.T. and Kump, L.R., 1994. Global chemical erosion during the last glacial maximum and the present: Sensivity to changes in lithology and hydrology. Palaeoceanography, 9, 529-543.

Gibbs, R.J., 1967. The geochemistry of the Amazon river system, the factors that control the salinity and composition and concentration of suspended solids. Geol. Soc. Am. Bull., 78, 1203-1232.

Gitelson, I.I., Abrosov, N.S. and Gladyshev, M.I., 1988. The main hydrological and hydrobiological characteristics of the Yenisei River. In: E.T. Degens, S. Kempe and A.S. Naidu (Editors), Transport of Carbon and Minerals in Major World Rivers, Lakes and Estuaries, Pt. 5. Mitt. Geol.-Paleont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 66, Hamburg, pp. 43-46.

Global Runoff Data Center Koblenz, 1991. Flow rates of selected world rivers: A database (DS 552.0). National Center for Atmospheric Research (NCAR), Boulder, Colorado.

Grieve, I.C., 1991. Dissolved organic carbon trends in small streams, land use effects and models of temporal variation. In: Sediment and Stream Water Quality in a Changing Environment: Trends and Explanation. Proceedings on the Vienna Symposium 1991. IAHS Publ., 203, IAHS Press, Wallingford, pp. 201-208.

Hansen, J., Russell, G., Rind, D., Stone, P., Lacis, A., Lebedeff, S. Ruedy, R. and Travis, L., 1983. Efficient three-dimensional global models for climate studies: Models I and II. Mon. Wea. Rev., 111, 609-662.

Hart, R.C., 1982. The Orange River: Preliminary Results. In: E.T. Degens (Editor), Transport of Carbon and Minerals in Major World Rivers, Pt. 1. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 52, Hamburg, pp. 435.

Hart, R.C., 1983. The Orange River: Supplementary Results. In: E.T. Degens, S. Kempe and H. Soliman (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 2. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 55, Hamburg, pp. 451-457.

Hart, R.C., 1985. Aspects of the hydrogeochemistry of the upper Orange River. In: E.T. Degens, S. Kempe and R. Herrera (Editors), Transport of Carbon and Minerals in Major World Rivers,

198

Pt. 3. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 58, Hamburg, pp. 435-442.

Hart, R.C., 1987. Carbon transport in the upper Orange River. In: E.T. Degens, S. Kempe and Gan Wei-Bin (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 4. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 64, Hamburg, pp. 509-512.

Hedges, J.I., Clark, W.A., Quay, P.D., Richey, J.E., Devol, A.H. and Santos, U.d.M., 1986. Composition and fluxes of particulate organic material in the Amazon River. Limnol. Oceanogr., 31, 717-738.

Hedges, J.I., Clark, W.A. and Cowie, G.L., 1988. Organic matter sources to the water column and surficial sediments of a marine bay. Limnol. Oceanogr., 33, 1116-1136.

Henderson-Sellers, A., 1993. Continental vegetation as a dynamic component of a global climate model: A preliminary assessment. Climatic Change, 23, 337-377.

Herczeg, A.L., Simpson, H.J. and Mazor, E., 1993. Transport of soluble salts in a large semiarid basin: river Murray, Australia. J. Hydrol., 144, 59-84.

Holdridge, L.R., 1947. Determination of world plant formations from simple climatic data. Science, 105, 367-368.

Holdridge, L.R., 1959. Simple method for determining potential evapotranspiration from temperate data. Science, 130, 572.

Holdridge, L.R., 1964. Life Zone Ecology. Tropical Science Center, San José, Costa Rica. Holeman, J.N., 1968. The sediment yield of major rivers of the world. Water Resour. Res., 4, 737-

747. Holland, D.H., 1978. The chemistry of atmosphere and oceans. John Wiley, New York, 351 pp. Hulme, M., 1991. An intercomparison of model and observed global precipitation climatologies.

Geophys. Res. Let., 18, 1715-1718. International Geosphere Biosphere Program (IGBP), 1993. Science Plan. In: P.M. Holligan and H.

de Bois (Editors), Land-Ocean Interactions in the Coastal Zone. IGBP Rep. No. 25, Stockholm, 50 pp.

International Geosphere Biosphere Program (IGBP), 1995. Implementation Plan. In: J.C. Pernetta and J.D. Milliman (Editors), Land-Ocean Interactions in the Coastal Zone. IGBP Rep. No. 33, Stockholm, 215 pp.

Ittekkot, V., 1988. Global trends in the nature of organic matter in river suspensions. Nature, 332, 436-438.

Jansen, J.M.L. and Painter, R.B., 1974. Predicting sediment yield from climate and topography. J. Hydrol., 21, 371-380.

Jansson, M.B., 1982. Land erosion by water in different climates. UNGI Rapp. 57, Uppsala University, 151 pp.

Jansson, M.B., 1988. A global survey of sediment yield. Geografiska Annaler, 70, A, 81-98. Judson, S. and Ritter, D.F., 1964. Rates of regional denudation in the United States. J. Geophys.

Res., 69, 3395-3401. Kalle, K., 1966. The problem of gelbstoff in the sea. In: Barnes (Editor), Oceanographic Marine

Biological Annual Review, No. 4. Allen and Unwin, London, pp. 91-104. Kattan, Z., Gac, J.Y. and Probst, J.L., 1987. Suspended sediment load and mechanical erosion in

the Senegal Basin - Estimation of the surface runoff concentration and relative contributions of channel and slope erosion. J. Hydrol., 92, 59-76.

Kayser, N., Probst, J.L., Cadet, D. et Tardy, Y., 1990. Propagation des ondes de sécheresse et d'humidité à travers le monde. C. R. Acad. Sci. Paris, 310, 757-763.

Kempe, S., 1979a. Carbon in the rock cycle. In: E.T. Degens, S. Kempe and P. Ketner (Editors), The Global Carbon Cycle. SCOPE Rep. 13, John Wiley, New York, pp. 343-377.

Kempe, S., 1979b. Carbon in the freshwater cycle. In: E.T. Degens, S. Kempe and P. Ketner (Editors), The Global Carbon Cycle. SCOPE Rep. 13, John Wiley, New York, pp. 317-342.

Kempe, S., 1982. Long-term records of CO2 pressure fluctuations in fresh waters. In: E. T. Degens (Editor), Transport of carbon and minerals in major world rivers, Pt. 1. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderdb. 52, Hamburg, pp. 91-332.

199

Kempe, S., 1983. Impact of Aswan high dam on water chemistry of the Nil. In: E. T. Degens, S. Kempe et H. Soliman (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 2. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 55, Hamburg, pp. 401-423.

Kempe, S., 1984. Sinks of the anthropogenically enhanced carbon cycle in surface fresh waters. J. Geophys. Res., 89, D3, 4657-4676.

Kempe, S., 1995. Coastal seas: a net source or sink of atmospheric carbon dioxide? LOICZ reports and studies No. 1, Stockholm, 27 pp.

Kempe, S. and Depetris, P.J., 1990. Labile carbon in the Paraná River basin. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, 69, pp. 157-165.

Kempe, S. and Depetris, P.J., 1992. Factors controlling the concentration of particulate carbohydrates and amino acids in the Paraná River. Hydrobiologia, 242, 175-183.

Kempe, S., Pettine, M. and Cauwet, G., 1991. Biogeochemistry of European rivers. In: E.T. Degens, S. Kempe and J.E. Richey (Editors), Biogeochemistry of Major World Rivers. SCOPE Rep. 42. John Wiley, New York, pp. 169-211.

Kempe, S., Eisma, D. and Degens, E.T., 1993. Transport of Carbon and Minerals in Major World Rivers, Pt. 6. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 74, Hamburg, 319 pp.

Kern, J.S., 1994. Spatial patterns of soil organic carbon in the contiguous United States. Soil Sci. Soc. Am. J., 58, 439-455.

Köppen, W., 1936. Das geographische System der Klimate. In: W. Köppen and R. Geiger (Editors), Handbuch der Klimatologie, Band 1, Teil C. pp. 1-44.

Korzoun, V.I., Sokolov, A.A., Budyko, M.I., Voskresensky, G.P., Kalinin, A.A., Konoplyantsev, E.S., Korotkevich, E.S. and Lvovich, M.I., 1977. Atlas of World Water Balance. UNESCO Press, Paris, 36 pp.

Kramer, J.R., 1994. Old sediment carbon in global budgets. In: M.D.A. Rounsevell and P.J. Loveland (Editors), Soil Responses to Climate Change, NATO ASI Series, 123. Springer Verlag, Berlin, pp. 169-183.

Kuhl, S. and Miller, J.R., 1992. Seasonal river runoff calculated from a global atmospheric model. Water Resour. Res., 28, 2029-2039.

Ladouche, B., Idir, S., Viville, D., Probst, A., Loubet, M., Probst, J.L., Ambroise, B. et Barriac, T., 1995. Approche géochimique de la décomposition de l'hydrogramme de crue. Rapport d'avancement du programme DBT II, 51 pp.

Langbein, W.B. and Schumm, S.A., 1958. Yield of sediment in relation to mean annual precipitation. American Geophysical Union Transactions, 39, 1076-1084.

Latron, J., 1990. Charactérisation géomorphologique et hydrologique du bassin versant du Strengbach (Aubure). Mémoire de maitrise de l'Université Louis Pasteur, Strasbourg, 96 pp.

Lautenschlager, M., 1991. Simulations of ice-age atmosphere - January and July means. Geol. Rdsch., 80, 513-534.

Lautenschlager, M. and Herterich, K., 1990. Atmospheric response to ice age conditions: Climatology near the Earth's surface. J. Geophys. Res., 95, 22547-22557.

Leavesley, G.H., 1994. Modeling the effects of climate change on water resources - a review. Climatic Change, 28, 159-177.

Leemans, R. and Cramer, W.P., 1992. IIASA database for mean monthly values of temperature, precipitation, and cloudiness on a global terrestrial grid: Digital raster on a 30 minute geographic (lat/long) 360x720 grid. In: NOAA/ National Geopysical Data Center (Editor), Global Ecosystems Database, Version 1.0 Disc (CD-ROM). National Geopysical Data Center, Boulder, Colorado.

Leenheer, J.A., 1982. United States Geological Survey data information service. In: E.T. Degens (Editor), Transport of Carbon and Minerals in Major World Rivers, Pt. 1. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 52, Hamburg, pp. 355-356.

Leenheer, J.A., 1991. Organic substance structures that facilitate contaminant transport and transformations in aquatic sediments. In: R.A. Baker (Editor), Organic Substances and Sediments in Water, vol. I, Humics and Soils. Lewis Publ. Chelsea, pp. 3-21.

200

Legates, D.R., 1989. A high-resolution climatology of gage-corrected global precipitation. In: B. Sevruk (Editor), Precipitation Measurement. Proceedings of the WMO/IAHS/ETH International Workshop on Precipitation Measurement, St. Moritz, Switzerland. Zürich Swiss Federal Institute of Technology, pp. 519-526.

Legates, D.R. and Willmott, C.J., 1990. Mean seasonal and spatial variability in gauge-corrected global precipitation. Int. J. Climatol., 10, 111-127.

Legates, D.R. and Willmott, C.J., 1992. Monthly average surface air temperature and precipitation: Digital raster on a 30 minute geographic (lat/long) 360x720 grid. In: NOAA/ National Geopysical Data Center (Editor), Global Ecosystems Database, Version 1.0 Disc (CD-ROM). National Geopysical Data Center, Boulder, Colorado.

Lesack, L.R., Hecky, R.E. and Melack, J.M., 1984. Transport of carbon, nitrogen, phosphorus, and major solutes in the Gambia River, West Africa. Limnol. Oceanogr., 29, 816-830.

Lewis, W.M. and Saunders III, J.F., 1989. Concentration and transport of dissolved and suspended substances in the Orinoco River. Biogeochemistry, 7, 203-240.

Lewis, W.M., Saunders III, J.F., Weibezahn and Levine, S.N., 1986. Organic carbon in the Caura River, Venezuela. Limnol. Oceanogr., 31, 653-656.

Lewis, W.M., Hamilton, S., Jones, S. and Runnels, D. 1987. Major element chemistry, weathering and element yields for the Caura River drainage. Biogeochemistry, 4, 159-181.

Likens, G.E., Mackenzie, F.T., Richey, J.E., Sedel, J.R. and Turekian, K.K., 1981. Flux of Organic Carbon by Rivers to the Oceans. Rep. CONF-8009140, U.S. Dept. of Energy, 397 pp.

Livingstone, D.A., 1963. Chemical composition of rivers and lakes. Data of geochemistry. U. S. Geol. Survey Prof. Paper, 440 G, 1-64.

Lottes, A.L. and Ziegler, A.M., 1994. World peat occurence and the seasonality of climate and vegetation. Pal. Pal. Pal., 106, 23-37.

Ludwig, W., 1991. Untersuchung gelöster gebundener und gelöster freier Aminosäuren im westlichen Mittelmeer (Golf von Lion) mittels HPLC. Diplomarbeit der Universität Hamburg, Hamburg, 71 pp.

Ludwig, W. and Probst, J.L., 1996. A global modelling for the climatic, morphological, and lithological control of river sediment discharges to the oceans. In: D.E. Walling and B.W. Webb (Editors) Erosion and Sediment Yield: Global and Regional Perspectives. Proceedings on the Exeter Symposium. IAHS Publ. no. 236, IAHS Press, Wallingford, 21-28.

Ludwig, W. and Spitzy, A., 1992. Spatial and seasonal variations in the distribution of dissolved amino acids in the Gulf of Lions. In: J.-M. Martin and H. Barth (Editors), Water Pollution Research Reports No. 28 of the Commission of the European Communities. pp. 219-235.

Ludwig, W., Amiotte-Suchet, P., Munhoven, G. and Probst, J.L.. Atmospheric CO2 consumption by continental erosion: Present-day controls and implications for the last glacial maximum. Global and Planetary Change, in press (a).

Ludwig, W., Amiotte-Suchet, P. and Probst, J.L. River discharges of carbon to the world's oceans: determining local inputs (4° x 5° grid resolution) of alkalinity and of dissolved and particulate organic carbon. C. R. Acad. Sci. Paris, in press (b).

Ludwig, W., Probst, J.L. and Kempe, S., 1996. Predicting the oceanic input of organic carbon by continental erosion. Global Biogeochem. Cycles, 10, 23-41.

Malcom, R.L. and Durum, W.H., 1976. Organic carbon and nitrogen concentrations and annual organic carbon load of six selected rivers of the United States. U.S. Geological Survey Water-Supply Paper, 1817F, 1-21.

Mallows, C.P., 1973. Some comments on CP. Technometrics, 15, 661-675. Mantoura, R.F.C. and Woodward, E.M.S., 1983. Conservative behaviour of riverine dissolved

organic carbon in the Severn Estuary: Chemical and geochemical implications. Geochim. Cosmochim. Acta, 47, 1293-1309.

Marengo, J.A., Miller, J.R., Russell, G.L., Rosenzweig, C.E. and Abramopoulos, F., 1994. Calculations of river-runoff in the GISS GCM: Modeling the hydrology of the Amazon River. Climate Dynamics, 10, 349-361.

Mariotti, A., Gadel, F., Giresse, P. and Kinga-Mouzeo, 1991. Carbon isotope composition and geochemistry of particulate organic matter in the Congo River (Central Africa): Application

201

to the study of Quaternary sediments off the mouth of the river. Chemical Geology, 86, 345-357.

Martins, O., 1982. Geochemistry of the Niger River. In: E.T. Degens (Editor), Transport of Carbon and Minerals in Major World Rivers, vol. 1. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 52, Hamburg, pp. 397-418.

Martins, O., 1983. Transport of carbon in the Niger river. In: E. T. Degens, S. Kempe et H. Soliman (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 2. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderdb. 55, Hamburg, pp. 435-449.

Martins, O. and Probst, J.L., 1991. Biogeochemistry of major African rivers: Carbon and mineral transport. In: E.T. Degens, S. Kempe and J.E. Richey (Editors), Biogeochemistry of Major World Rivers. SCOPE Rep. 42. John Wiley, New York, pp. 127-155.

McDowell, W.H. and Wood, T., 1984. Podzolization: Soil processes control dissolved organic carbon concentrations in stream water. Soil Science, 137, 23-32.

Meade, R.H. and Parker, R.S., 1985. Sediments in rivers of the United States. U.S. Geol. Surv. Water Suppl. Pap., 2275, 49-60.

Meade, R.H., Nordin, C.F., Perez Hernandez, D., Mejia, A.B. and Perez Godoy, J.M., 1983. Sediment and water discharge in Rio Orinoco, Venezuela and Colombia. In: Proceedings on the II. Int. Symp. on River Sedimentation. Water Res. and Electric Power Press, Bejing, China, pp. 1134-1144.

Meade, R.H., Dunne, T., Richey, J.E., Santos, U.d.M. and Salati, E., 1985. Storage and remobilization of suspended sediment in the lower Amazon River of Brazil. Science, 228, 488-490.

Meybeck, M., 1979. Concentrations des eaux fluviales en éléments majeurs et apports en solution aux océans. Rev. Géol. Dyn. Géogr. Phys., 21, 215-246.

Meybeck, M., 1982. Carbon, nitrogen and phosphorus transport by world rivers. Amer. J. Sci., 282, 401-450.

Meybeck, M., 1984. Les fleuves et le cycle géochimique des éléments. Thèse Sciences, Univ. Paris VI, 558 pp.

Meybeck, M., 1986. Composition chimique des ruisseaux non pollués de France. Sci. Géol. Bull., 39, 3-77.

Meybeck, M., 1987. Global chemical weathering of surficial rocks estimated from river dissolved loads. Amer. J. Sci., 287, 401-428.

Meybeck, M., 1988. How to establish and use world budgets of river material. In: A. Lerman and M. Meybeck (Editors), Physical and Chemical Weathering in Geochemical Cycles. Kluwer, Norwell, pp. 247-272.

Meybeck, M., 1993a. C, N, P, and S in rivers: From sources to global inputs. In: R. Wollast, F.T. Mackenzie and L. Chou (Editors), Interactions of C, N, P and S. Biogeochemical Cycles and Global Change. Springer Verlag, Berlin, pp. 163-193.

Meybeck, M., 1993b. Riverine transport of atmospheric carbon: Sources, global typology, and budget. Water, Air and Soil Pollution, 70, 443-463.

Meybeck, M., Cauwet, G., Dessery, S., Somville, M., Gouleau, D. and Billen, G., 1988. Nutrients (organic C, P, N, Si) in the eutrophic river Loire and its estuary. Estuarine Coastal Shelf Sci., 27, 595-624.

Michaelis, W. and Ittekkot, V., 1982. Biogeochemistry of rivers: Field and analytical techniques. In: E.T. Degens (Editor), Transport of Carbon and Minerals in Major World Rivers, Pt. 1. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 52, Hamburg, pp. 69-89.

Miller, J.R., Russell, G.L. and Caliri, G., 1994. Continental-scale river flow in climate models. J. Climate, 7, 914-928.

Milliman, J.D. and Meade, R.H., 1983. World-wide delivery of river sediment to the oceans. J. Geol., 91, 1-21.

Milliman, J.D. and Syvitski, J.P.M., 1992. Geomorphic/ tectonic control of sediment discharge to the ocean: The importance of small mountainous rivers. J. Geol., 100, 525-544.

202

Milliman, J.D., Xie Qinhun, and Yang Zuocheng, 1984. Transfer of particulate organic carbon and nitrogen from the Yangtze River to the ocean. Am. J. Sci., 284, 824-834.

Milliman, J., Qin Yun-Shan, Ren Mei-E, and Saito, Y., 1987. Man's influence on the erosion and transport of sediment by Asian rivers: The Yellow River (Huanghe) example. J. Geol., 95, 751-762.

Milliman, J.D., Rutkowski, C. and Meybeck, M., 1995. River discharge to the sea. A global river index (GLORI). LOICZ Core Project Office, Texel, the Netherlands, 125 pp.

Milliman, J.D., Snow, J., Jaeger, J. and Nittrouer, C., 1996. Catastrophic discharge of fluvial sediment to the ocean: evidence of Jøkulhlaups events in the Alsek Sea Valley, southeast Alaska (USA). In: D.E. Walling and B.W. Webb (Editors) Erosion and Sediment Yield: Global and Regional Perspectives. Proceedings on the Exeter Symposium. IAHS Publ. no. 236, IAHS Press, Wallingford, 367-379.

Ming-Hui, H., Stallard, R.F. and Edmond, J.M., 1982. Major ion chemistry of some large Chinese rivers. Nature, 298, 550-553.

Mintz, Y. and Serafini, Y.V., 1992. A global monthly climatology of soil moisture and water balance. Climate Dynamics, 8, 13-27.

Moore, J.G. and Mark, R.K., 1986. World slope map. EOS Trans. AGU, 67, 1353-1356. Moore, T.R., 1989. Dynamics of dissolved organic carbon in forested and disturbed catchments,

Westland, New Zealand. 1. Maimai. Water Resour. Res., 25, 1321-1330. Moore, T.R. and Jackson, R.J., 1989. Dynamics of dissolved organic carbon in forested and

disturbed catchments, Westland, New Zealand. 2. Larry River. Water Resour. Res., 25, 1331-1339.

Munhoven, G. and François, L.M., 1994. Glacial-interglacial changes in continental weathering: Possible implications for atmospheric CO2. In: R. Zahn et al. (Editors), Carbon Cycling in the Glacial Ocean: Constraints on the Ocean's Role in Global Change. Springer Verlag, Berlin, pp. 39-58.

Munhoven, G. and Probst, J.L., 1995. Influence of Continental Erosion Processes on the Glacial/ Interglacial Evolution of Atmospheric Carbon Dioxide. Final Scientific Report, EC contract ERBCHBIC 94 1053, Strasbourg, 16 pp.

Munhoven, G. and François, L.M., 1996. Glacial-interglacial variability of atmospheric CO2 due to changing continental silicate rock weathering. J. Geophys. Res., 101, D16, 21423-21437.

Németh, A., Paolini, J. and Herrera, R., 1982. Carbon transport in the Orinoco River: Preliminary Results. In: E.T. Degens (Editor), Transport of Carbon and Minerals in Major World Rivers, Pt. 1. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 52, Hamburg, pp. 357-364.

NKounkou, R., 1989. Hydrogéodynamique actuelle du Congo et de l'Amazone Cycle global de l'eau et bilan de l'érosion au cours des temps phanérozoïques (derniers 600 millions d'années). Thèse de doctorat, Université Louis Pasteur, Strasbourg, 183 pp.

NKounkou, R.R. and Probst, J.L., 1987. Hydrology and geochemistry of the Congo river system. In: E.T. Degens, S. Kempe and Gan Wei-Bin (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 4. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 64, Hamburg, pp. 483-508.

Ohmori, H., 1983. Erosion rates and their relation to vegetation from the viewpoint of world-wide distribution. Bull. Dept. Geogr., Univ. Tokyo, 15, 77-91.

Ol'dekop, E.M., 1911. Ob isporenzi s poverk n osti rechnykh basseinov (On evaporation of from the surface of river basins). Tr. Meteorol. Observ. Iur'evskogo Univ. Tartu., 4.

Olivry, J.C., Bricquet, J.P., Thiebaux, J.P. and Sigha, N., 1988. Transport de matière sur les grands fleuves des régions intertropicales: les premiers résultats des mesures de flux particulaires sur le bassin du fleuve Congo. In: M.P. Bordas and D. Walling (Editors), Sediment Budgets. Proceedings on the Porto Alegre Symposium. IAHS Publ., 174, IAHS Press, Wallingford, pp. 509-521.

Olson, J.S., Watts, J.A. and Allison, L.J., 1983. Carbon in live vegetation of major world ecosystems. Report ORNL-5862, Oak Ridge Laboratory, Oak Ridge, Tennessee, 164 pp.

203

Olson, J.S., Watts, J.A. and Allison, L.J. 1985. Major world ecosystem complexes ranked by carbon in live vegetation: A database, Rep. NDP-017. Carbon Dioxide Information Analysis Center, Oak Ridge Nat. Laboratory, Oak Ridge, Tennessee, 19 pp.

Orange D., Feizoure C., Wesselink, A. and Callede, J., 1995. Variabilites hydrologiques de l'Oubangui a Bangui au cours du XXme siecle. In: Proceedings des Journees Scientifiques de FRIEND-AOC, Cotonou, December 1995. UNESCO Press, Paris, 20 p.

Orange D., Feizoure C., Laraque, A. and Wesselink, A., 1996. Hydroclimatological changes on the central African continent: Evidence from the Oubangui River. XXI General Assembly of EGS - The Hague/ May 1996, 7 p.

Paolini, J., Herrera, R. and Németh, A., 1983. Hydrochemistry of the Orinoco and Caroní rivers. In: E.T. Degens, S. Kempe and H. Soliman (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 2. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 55, Hamburg, pp. 223-236.

Paolini, J., Hevia, R. and Herrera, R., 1987. Transport of carbon and minerals in the Orinoco and Caroni rivers during the years 1983-1984. In: E.T. Degens, S. Kempe and Gan Wei-Bin (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 4. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 64, Hamburg, pp. 325-338.

Peltier, W.R., 1994. Ice age palaeotopography. Science, 265, 195-201. Penman, H.L., 1948. Natural evaporation from open water, bare soil and grass. Proc. Roy. Soc.,

London, A193, 120-145. Pettine, M., La Noce, T., Pagnotta, R. and Puddu, A., 1985. Organic and trophic load of major

Italian rivers. In: E.T. Degens, S. Kempe and R. Herrera (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 3. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 58, Hamburg, pp. 417-429.

Phillips, J.D., 1990. Relative importance of factors influencing fluviatil soil loss at a global scale. Amer. J. Sci., 290, 547-568.

Pike, J.G., 1964. The estimation of annual runoff from meteorological data in a tropical climate. J. Hydrol., 2, 116-123.

Pinet, P. and Souriau, M., 1988. Continental erosion and large scale relief. Tectonics, 7, 563-582. Pocklington, R., 1982. Carbon transport in major world rivers: The St. Lawrence, Canada. In: E.T.

Degens (Editor), Transport of Carbon and Minerals in Major World Rivers, Pt. 1. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 52, Hamburg, pp. 347-353.

Pocklington, R., 1985. The contribution of organic matter by the St. Lawrence River to the Gulf of St. Lawrence, 1981-1983. In: E.T. Degens, S. Kempe and R. Herrera (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 3. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 58, Hamburg, pp. 323-329.

Pocklington, R. and Tan, F., 1983. Organic carbon transport in the St. Lawrence River. In: E.T. Degens, S. Kempe and H. Soliman (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 2. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 55, Hamburg, pp. 243-251.

Pocklington, R. and Tan, F., 1987. Seasonal and annual variations in the organic matter contributed by the St. Lawrence River to the Gulf of St. Lawrence. Geochim. Cosmochim. Acta, 51, 2579-2586.

Post, W.M., 1993. Organic carbon in soil and the global carbon cycle. In: M. Heimann (Editor), The Global Carbon Cycle. NATO ASI Ser. I., vol. 15. Springer-Verlag, New York, 599 pp.

Post, W.M., Emanuel, W.R., Zinke, P.J. and Stangenberger, A.G., 1982. Soil carbon pools and world life zones. Nature, 298, 156-159.

Post, W.M., Pastor, P., Zinke, P.J. and Stangenberger, A.G., 1985. Global patterns of soil nitrogen storage. Nature, 317, 613-616.

Prentice, K.C., 1990. Bioclimatic distribution of vegetation for general circulation model studies. J. Geophys. Res., 95, D8, 11811-11830.

Probst, A., Fritz B., Ambroise B., and Viville D., 1987. Forest influence on the surface water chemistry of granitic basins receiving acid precipitation in the Vosges massif, France.

204

Proceedings on the Vancouver Symposium, 9-22 August 1987. IAHS Publ., 167, IAHS Press, Wallingford, pp. 109-120.

Probst, A., Dambrine E., Viville D. and Fritz B., 1990. Influence of acid atmospheric inputs on surface water chemistry and mineral fluxes in a declining spruce stand within a small granitic catchment (Vosges massif, France). J. Hydrol., 116, 101-124.

Probst, A., Viville D., Fritz B., Ambroise B. and Dambrine E., 1992. Hydrochemical budgets of a small granitic catchment: The Strengbach catchment case study (Vosges massif, France). Water, Air and Soil Pollution, 62, 337-347.

Probst, J.L., 1983. Hydrologie du bassin de la Garonne, modèle de mélanges, bilans de l'érosion, exportation des phosphates et des nitrates. Thèse 3ème cycle, Université Paul Sabatier, Toulouse, 148 pp.

Probst, J.L., 1992. Géochimie et hydrologie de l'érosion continentale. Mécanismes, bilan global actuel et fluctuations au cours des 500 derniers millions d'années. Sci. Géol. Mém., Strasbourg, 94, 161 pp.

Probst, J.L. et Tardy, Y., 1985. Fluctuations hydroclimatiques du Bassin d'Aquitaine au cours des 70 dernières années. Rev. Géol. Dyn. Géogr. Phys., 26, 59-76.

Probst, J.L. et Bazerbachi, A., 1986. Transports en solution et en suspension par la Garonne supérieure. Sci. Géol. Mém., Strasbourg, 39, 79-98.

Probst J.L. and Tardy Y., 1987. Long range streamflow and world continental runoff fluctuations since the beginning of this century. J. Hydrol., 94, 289-311.

Probst J.L. and Tardy Y., 1989. Global runoff fluctuations during the last 80 years in relation to world temperature change. Amer. J. Sci., 289, 267-285.

Probst, J.L. et Sigha, N., 1989. Estimation de l'écoulement superficiel et de sa charge en suspension sur quelques grands bassins fluviaux du monde. C. R. Acad. Sci. Paris, 309, 357-363.

Probst, J.L. and Amiotte-Suchet, P., 1992. Fluvial suspended sediment transport and mechanical erosion in the Maghreb (North Africa). Hydrol. Sci. J., 37, 621-637.

Probst, J.L., Nkounkou, R.R., Krempp, G., Bricquet, J.P., Thiébaux, J.P. and Olivry J.C., 1992. Dissoloved major elements exported by the Congo and the Ubangi rivers during the period 1987-1989. J. Hydrol., 135, 237-257.

Probst, J.L., Amiotte-Suchet, P. and Ludwig, W., 1994a. Continental erosion and river transports of carbon to oceans. Trends in Hydrology, 1, 453-468.

Probst J.L., Mortatti J. and Tardy, Y., 1994b. Carbon river fluxes and weathering CO2 consumption in the Congo and Amazon river basins. Applied Geochem., 9, 1-13.

Qunying, Z., Feng, L., Xun, L. and Minghui, H., 1987. Major ion chemistry and fluxes of dissolved solids with rivers in southern coastal China. In: E.T. Degens, S. Kempe and Gan Wei-Bin (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 4. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 64, Hamburg, pp. 243-249.

Rasmussen, J.B., Godbout, L. and Schallenberg, M., 1989. The humic content of lake water and its relationship to watershed and lake morphology. Limnol. Oceanogr., 34, 1336-1343.

Reeder, S.W., Hitchon, B. and Levinson, A.A., 1972. Hydrogeochemistry of the surface waters of the Mackenzie River drainage basin, Canada. Factors controlling inorganic composition. Geochim. Cosmochim. Acta, 36, 825-865.

Richey, J.E., Hedges, J.I., Devol, A.H. and Quay, P.D., 1990. Biogeochemistry of carbon in the Amazon River. Limnol. Oceanogr., 35, 352-371.

Romankevitch, E.A. and Artemyev, V.E., 1985. Input of organic carbon into seas and oceans bordering the territory of the Soviet Union. In: E.T. Degens, S. Kempe and R. Herrera (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 3. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 58, Hamburg, pp. 459-469.

Russell, G.L. and Miller, J.R., 1990. Global river runoff calculated from a global atmospheric general circulation model. J. Hydrol., 117, 241-254.

Safiullah, S., Mofizuddin, M., Iqbal-Ali, S.M. and Enamul-Kabir, S., 1987. Biogeochemical cycles of carbon in the rivers of Bangladesh. In: E.T. Degens, S. Kempe and Gan Wei-Bin (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 4. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 64, Hamburg, pp. 435-442.

205

Sarmiento, J.L., and Sundquist, E.T., 1992. Revised budget for the oceanic uptake of anthropogenic carbon dioxide. Nature, 356, 589-593.

SAS Institute Inc., 1986. SAS system for regression, 1986 edition. SAS Institute Inc., Cary, N.C., 164 pp.

Saunders III, J.F. and Lewis, W.M., 1988. Transport of phosphorus, nitrogen, and carbon by the Apure River, Venezuela. Biogeochemistry, 5, 323-342.

Sawer, J.O. and Lindsey, A.A., 1963. The Holdridge bioclimatic formations on the Eastern and Central United States. Indiana Acad. Sci. J., 73, 105-112.

Schlesinger, W.H. and Melack, J.M., 1981. Transport of organic carbon in the world's rivers. Tellus, 31, 172-187.

Schreiber, P., 1904. Über die Beziehung zwischen dem Niederschlag und der Wasserführung der Flüsse in Mitteleuropa. Z. Meteorol., 21, 441-452.

Schumm, S.A. and Hadley, R.F., 1961. Progress in the application of landform analysis in studies of semi-arid erosion. Geol. Surv. Circ. no. 437, Washington.

Semhi, K., 1996. Erosion et transferts de matières sur le bassin versant de la Garonne. Influence de la sécheresse. Thèse de Doctorat de l'Université Louis Pasteur, Strasbourg, 203 pp.

Sharp, M., Tranter, M., Brown, G.H. and Skidmore, M, 1995. Rates of chemical denudation and CO2 drawdown in a glacier-covered alpine catchment. Geology, 23, 61-64.

Smith, S.V. and Hollibaugh, G.T., 1992. Coastal metabolism and the oceanic organic carbon balance. Rev. Geophys., 31, 75-89.

Spitzy, A. 1988. Dissolved organic matter in groundwaters from different climates. In: E.T. Degens, S. Kempe and J.E. Richey (Editors), Biogeochemistry of Major World Rivers. SCOPE Rep. 42. Mitt. Geol.-Paleont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 66, Hamburg, pp. 377-413.

Spitzy, A. and Leenheer, J., 1991. Dissolved organic carbon in rivers. In: E.T. Degens, S. Kempe and J.E. Richey (Editors), Biogeochemistry of Major World Rivers, SCOPE Report 42. John Wiley, New York, pp. 213-232.

Spitzy, A., Ludwig, W. and Wobig, M., 1990. Survey of amino acid nitrogen in water, sediment and aerosol from the Gulf of Lions. In: J.-M. Martin and H. Barth (Editors), Water Pollution Research Reports No 20 of the Commission of the European Communities. pp. 187-207.

Starkel, L., 1988. Global paleohydrology. Bull. Pol. Acad. Sci, 36, 71-89. Subramanian, V., 1979. Chemical and suspended sediment characteristics of rivers of India. J.

Hydrol., 44, 37-55. Subramanian, V. and Ittekkot, V., 1991. Carbon transport by the Himalayan rivers. In: E.T.

Degens, S. Kempe and J.E. Richey (Editors), Biogeochemistry of Major World Rivers. SCOPE Rep. 42. John Wiley, New York, pp. 157-168.

Summerfield, M.A., 1991. Rates of uplift and denundation. In: M.A. Summerfield (Editor), Global Geomorphology. Longman Sci. Publ., Singnapore, pp. 371-402.

Summerfield, M.A. and Hulton, N.J., 1994. Natural controls of fluvial denudation rates in major world drainage basins. J. Geophys. Res., 99, B7, 13871-13883.

Tan, F.C., 1987. Discharge and carbon isotope composition of particulate organic carbon from the St. Lawrence River, Canada. In: E.T. Degens, S. Kempe and Gan Wei-Bin (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 4. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 64, Hamburg, pp. 301-310.

Tans, P.P., Conway T.J. and Nakazawa T., 1989. Latitudinal distribution of the sources and sinks of atmospheric carbon dioxide derived from surface observations and atmospheric transport model. J. Geophys. Res. 94 (D4), 5151-5172.

Tans, P.P., Fung, I.Y. and Nakazawa, T., 1990. Observational constraints on the global atmospheric CO2 budgets. Science, 247, 1431-1438.

Tardy, Y., 1986. Le cycle de l'eau; climats, paléoclimats et géochimie globale. Edition Masson, Paris, 338 pp.

Tardy, Y., N'Kounkou, R.R. and Probst, J.L., 1989. The global water cycle and continental erosion during phanerozoic time. Am. J. Sci., 289, 455-483.

206

Tardy, Y., Moratti, J. Ribeiro, A., Victoria, R. et Probst, J.L., 1994. Fluctuations de la pluviosité, de l'écoulement et de la température sur le bassin de l'Amazone et oscillations du climat global au cours du siècle écoulé. C. R. Acad. Sci. Paris, 318, 955-960.

Tardy, Y., Furlan, J., Moratti, J. et Probst, J.L., 1995. Distribution globale des oscillations du climat au cours du siècle éculé. Fluctuations de débit de cinquante grands fleuves du Monde. C. R. Acad. Sci. Paris, 320, 945-952.

Telang, S.A., 1985. Transport of carbon and minerals in the Mackenzie River. In: E.T. Degens, S. Kempe and R. Herrera (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 3. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 58, Hamburg, pp. 337-344.

Telang, S.A., Korchinski, M. and Hodgson, G.W., 1982. Abundances and transport of ions, nitrogen, and carbon in the Mackenzie River. In: E.T. Degens (Editor), Transport of Carbon and Minerals in Major World Rivers, Pt. 1. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 52, Hamburg, pp. 333-346.

Telang, S.A., Scott R.J.. and Hodgson, G.W., 1983. Concentrations and transport of carbon and minerals in the Mackenzie River. In: E.T. Degens, S. Kempe and H. Soliman (Editors), Transport of Carbon and Minerals in Major World Rivers, Pt. 2. Mitt. Geol.-Paläont. Inst. Univ. Hamburg, SCOPE/UNEP Sonderbd. 55, Hamburg, pp. 261-266.

Telang, S.A., Pocklington, R., Nadu, A.S., Romankevitch, E.A., Gitelson, I.I. and Gladyshev, M.I., 1991. Carbon and mineral transport in major North American, Russian Arctic and Siberian rivers: The St. Lawrence, the Mackenzie, the Yukon, the Arctic Alaskan rivers, the Arctic basin rivers in the Soviet Union and the Yenisei. In: E.T. Degens, S. Kempe and J.E. Richey (Editors), Biogeochemistry of Major World Rivers, SCOPE Rep. 42. John Wiley, New York, pp. 75-104.

The Times of London, 1967. Times Atlas of the World. John Bartholomew and Son Ltd. Thornthwaite, C.W., 1948. An approach toward a rational classification of climate. Geogr. Rev.,

38, 85-94. Thurman, E.M., 1985. Organic Geochemistry of Natural Waters. Martinus Nijhoff, Dortrecht, 497

pp. Tosi, J., 1983. Provisional life zone map of Brazil at 1:5000000 scale. Tropical Science Center, San

José, Costa Rica. Turc, L., 1954. Le bilan d'eau des sols. Relation entre les precipitations, l'evaporation et

l'écoulement. Ann. Agron., 5, 491-569. U.S. Geological Survey. Quality of surface waters of the United States. Geol. Surv. Water Supply

Paper. Van Blarcum, S.C., Miller, J.R. and Russell, G.L., 1995. High latitude river runoff in a doubled

CO2 Climate. Climatic Change, 30, 7-26. Vebel, M.A., 1993. Temperature dependence of silicate weathering in nature. How strong a

negative feedback on long-term accumulation of atmospheric CO2 and global greenhouse warming? Geology, 21, 1050-1062.

Veyssy E., Colas C., Etcheber H., Maneux E., Probst J.L.. Flux organiques à l'entrée d'un systeme estuarien: cas de la Garonne. Mem. Sci. Géol., Strasbourg, in press.

Viville, D., Ambroise B., Probst, A., Fritz B., Dambrine, E., Gelhaye, D. et Deloze, C., 1988. Le bassin du Strengbach à Aubure (Haut-Rhin, France) pour l'étude du dépérissement forestier dans les Vosges (Programme DEFPORA). I. Equipement climatique, hydrologique, hydrochimique. In: Mathy (Editor), Proceedings of the Int. Symposium Air Pollution and Ecosystems, Grenoble, 18-22 May 1987. Reidel Publ. Co., Bruxelles, pp. 829-834.

Vörösmarty, C.J., Moore III, B., Grace, A.L., Gildea, M.P., Melillo, J.M., Peterson, B.J., Rastetter, E.B. and Steudler, P.A., 1989. Continental scale models of water balance and fluvial transport: An application to South America. Global Biogeochem. Cycles, 3, 241-256.

Walker, J.C.G., Hays, P.B. and Kasting, J.F.A., 1981. A negative feedback mechanism for the long term stabilisation of Earth's surface temperature. J. Geophys. Res., 86, 9776-9782.

Walling, D.E., 1983. The sediment delivery problem. J. Hydrol., 65, 209-237.

207

Walling, D.E. and Webb, B.W., 1983. Patterns of sediment yield. In: K.J. Gregory (Editor), Bachground to Palaeohydrology, Whiley, New York. pp. 69-100.

Walling, D.E. and Webb, B.W., 1985. Estimating the discharge of contaminants to coastal waters by rivers: Some cautionary comments. Marine Poll. Bull., 16, 488-492.

Webb, R.S., Rosenzweig, C.E. and Levine, E.R., 1991. A global data set of soil particle size properties. NASA Technical Memorandum, # 4286. NASA, Washington D.C., 33 pp.

Webb, R.S., Rosenzweig, C.E. and Levine, E.R., 1992. A global data set of soil particle size properties: Digital raster on a 1 degree geographic (lat/long) 180x360 grid. In: NOAA/ National Geopysical Data Center (Editor), Global Ecosystems Database, Version 1.0 Disc (CD-ROM). National Geopysical Data Center, Boulder, Colorado.

Webb, R.S., Rosenzweig, C.E. and Levine, E.R., 1993. Specifying land surface characteristics in general circulation models: Soil profile data set and derived water holding capacities. Global Biogeochem. Cycles, 7, 97-108.

Williams S. N., Schaeffer S. J., Calvache M. L. and Lopez, D., 1992. Global carbon dioxide emission to the atmosphere by volcanoes. Geochim. Cosmochim. Acta, 56, 1765-1770.

Willmott, C.J. and Legates, D.R., 1991. Rising estimates of terrestrial and global precipitation. Climate Research, 1, 179-186.

Willmott, C.J. and Legates, D.R., 1993. A comparison of GCM-simulated and observed mean January and July surface air temperature. J. Climate, 6, 274-291.

Willmott, C.J., Rowe, C.M. and Mintz, Y., 1985. Climatology of the terrestrial seasonal water cycle. J. of Climatology, 5, 589-606.

Wilson, L., 1973. Variations in mean annual sediment yield as a function of of mean annual precipitation. Amer. J. Sci., 273, 335-349.

Wischmeier, W.H., Smith, D.D. and Uhland, R.E., 1958. Evaluation of factors in the soil loss equation. Agric. Eng., 39, 458-474.

Zhang, S., Gan Wei-Bin, and Ittekkot, V., 1992. Organic matter in large turbid rivers: The Huanghe and its estuary. Mar. Chem., 38, 53-68.

Zinke, P.J., Stangenberger, A.G., Post, W.M., Emanuel, W.R. and Olson, J.S., 1986. Worldwide organic soil carbon and nitrogen data, Rep. ORNL/CDIC-18. Oak Ridge Nat. Laboratory, Oak Ridge, Tennessee, 136 pp.

Zobler, L., 1986. A world soil file for global climate modeling. NASA Technical Memorandum, # 87802. NASA, Washington D.C..

A-I

LIST OF FIGURES

Fig. 1 - Generalized scheme of the present-day global carbon cycle according to various literature estimates. .................................................................................................................. 2

Fig. 2 - Continuum of dissolved and particulate organic carbon in natural waters............................ 3 Fig. 3 - Distribution of the eleven river case studies that are presented in the text............................ 10 Fig. 4 - Map of the sampling sites in the Strengbach catchment at Aubure....................................... 11 Fig. 5 - Evolution of DOC and TSS concentrations in relation to discharge at four sampling

sites in the Strengbach catchment during a storm event in May 1994..................................... 13 Fig. 6 - Fluxes of dissolved organic carbon at four sampling sites in the Strengbach

catchment during a storm event in May 1994.......................................................................... 14 Fig. 7a - Evolution of the concentrations of dissolved organic carbon in relation to discharge

in the Strengbach Brook at SS4. .............................................................................................. 16 Fig. 7b - Two details of Fig. 7a: Three successive high-water periods in winter 1993/94 and a

storm event in summer 1995.................................................................................................... 17 Fig. 8a Mean monthly discharge of the Garonne River from 1977 to 1992 at La Réole. .................. 20 Fig. 8b Comparison of the daily discharges at La Réole and at Portet during the investigation

period. ...................................................................................................................................... 20 Fig. 9a Evolution of DOC, POC, and TSS concentrations in relation to discharge in the

Garonne River from 1989 to 1992. The station is Portet in the upstream part of the basin. ........................................................................................................................................ 22

Fig. 9b Evolution of DOC, POC, and TSS concentrations in relation to discharge in the Garonne River from 1989 to 1992. The station is La Réole in the downstream part of the basin. .................................................................................................................................. 23

Fig. 10 - Drainage intensity in the Mackenzie Basin. ........................................................................ 25 Fig. 11 - Evolution of Q, DOC, TSS, POC, and C/N in the Mackenzie River from 1981 to

1983. ........................................................................................................................................ 26 Fig. 12 - Mean monthly Discharge of the Mackenzie River at Norman Wells since 1973................ 27 Fig. 13 - Evolution of Q, DOC, TSS, POC, C/N, and 13C in the St. Lawrence River from

1981 to 1985. ........................................................................................................................... 28 Fig. 14 - Mean monthly discharge of the St. Lawrence River at Cornwall and at Québec City. ....... 29 Fig. 15 - Comparison of the monthly water contribution from Cornwall in percent with the

mean monthly 13C isotope signature of suspended matter in the waters at Québec City. ....... 30 Fig. 16 - Mean monthly discharge of the Waikato River at Ngaruawahia in comparison with

discharge at Mercer during the sampling period...................................................................... 31 Fig. 17 - Evolution of Q, DOC, TSS, POC, and C/N in the Waikato River from November

1981 to October 1984. ............................................................................................................. 32 Fig. 18 - Evolution of Q, DOC, TSS, and POC in the Orinoco River from February 1981 to

July 1982. The sampling station was Ciudad Bolivar.............................................................. 35 Fig. 19 - Average seasonal patterns of Q, DOC, TSS, and POC in the Orinoco River at

Barrancas.................................................................................................................................. 35 Fig. 20 - Paraná's water heights measured in the city of Paraná from March 1981 to April

1984. ........................................................................................................................................ 36 Fig. 21 - Variation of Q, DOC, TSS, POC, and C/N in the Paraná River from March 1981 to

April 1984. ............................................................................................................................... 37 Fig. 22 - Drainage intensity in the Zaire Basin. ................................................................................. 39 Fig. 23 - Comparison of the mean hydrographs of the Ubangi River at Bangui and of the

Zaire River at Brazzaville. ....................................................................................................... 39

A-II

Fig. 24 - Evolution of DOC concentration in relation to discharge in the Ubangi from July 1992 to July 1995. ....................................................................................................................40

Fig. 25 - Drainage intensity in the Niger Basin. .................................................................................41 Fig. 26 - Evolution of Q, DOC, TSS, POC, and C/N in the Niger River at Lokoja (May 1980

to April 1981)...........................................................................................................................42 Fig. 27 - DOC and POC concentrations during 1980/81 in the Niger River at Jebba, at

Lokoja, and in the Kaduna River, one of the Niger's tributaries. .............................................43 Fig. 28 - Evolution of Q, DOC, and TSS in the Niger River at Bamako (December 1991 to

July 1993).................................................................................................................................44 Fig. 29 - Drainage intensity in the Orange Basin. ..............................................................................45 Fig. 30 - Runoff at the outlet of Le Roux dam in comparison with runoff at Aliwal North. .............45 Fig. 31 - Evolution of Q, DOC, TSS, POC, and C/N in the Orange River at Le Roux Dam

from 1982 to 1984....................................................................................................................46 Fig. 32 - Drainage intensity in the Indus Basin. .................................................................................47 Fig. 33 - Mean monthly discharge at four gauging stations in the Indus Basin. ................................47 Fig. 34 - Evolution of Q, DOC, TSS, POC, and C/N in the Indus River at Kotri Barrage from

1982 to 1984.............................................................................................................................48 Fig. 35a - World-wide distribution of the river basins investigated in this study. .............................55 Fig. 35b - Continental divides for the different oceans. .....................................................................56 Fig. 36 - Drainage intensity on the continents....................................................................................59 Fig. 37 - Global distribution of organic carbon in the soils................................................................59 Fig. 38 - Mean modal elevation of the continents. .............................................................................61 Fig. 39 - Mean surface slope on the continents. .................................................................................61 Fig. 40 - Textural classes for soil erodibility assessment. ..................................................................63 Fig. 41 - The Holdridge Life Zone Triangle (a) and the upon based climatic classification

applied in this study (b)............................................................................................................65 Fig. 42 - Ratio of ABT over AT versus AT as determined for 2° temperature slices. .........................66 Fig. 43 - Distribution of all 0.5° x 0.5° longitude/latitude grid elements of the Mackenzie

Basin and of the Indus Basin in the Holdridge Triangle. .........................................................66 Fig. 44 - Distribution of the climate types distinguished in this study. ..............................................69 Fig. 45 - Global distribution of Four on the continents. .....................................................................69 Fig. 46 - Distribution of DT and Four in the Holdridge Triangle.......................................................71 Fig. 47 - Global distribution of the gauging stations included in the WMO database . .....................75 Fig. 48a - Plot of the specific drainage intensities extracted from the UNESCO runoff map

versus literature estimates for the 60 river basins of Table 6. ..................................................77 Fig. 48b - Comparison of total runoff extracted from the UNESCO runoff map with literature

estimates for the 60 river basins of Table 6..............................................................................77 Fig. 49a - Holospheric runoff distribution as determined by Baumgartner and Reichel [1975]

and as determined in this study on the basis of the UNESCO runoff map. .............................81 Fig. 49b - Holospheric runoff distribution of the UNESCO runoff map, as it is generated on

land, and as it enters the oceans. ..............................................................................................81 Fig. 50 - Limiting conditions for annual actual evapotranspiration. ..................................................83 Fig. 51 - Plot of mean annual potential evapotranspiration versus mean annual temperature

for the 60 river basins of Table 6. ............................................................................................85 Fig. 52a - Comparison of the holospheric distribution of precipitation and runoff for four

different precipitation data sets. ...............................................................................................87 Fig. 52b - Standard deviation of the average runoff ratio in Fig. 52a. ...............................................87 Fig. 53a-h - Combinations of different estimations for precipitation and potential

evapotranspiration for the 60 river basins of Table 6...............................................................89 Fig. 54 - Runoff ratio as a function of APPT over APE according to Pike [1964]. ...........................90 Fig. 55a - Comparison of the mean runoff ratio predicted in this study with the mean runoff

ratio resulting from the UNESCO runoff and precipitation maps for the 60 river basins of Table 6. ................................................................................................................................95

Fig. 55b - Comparison of total runoff predicted in this study with total runoff extracted from the UNESCO runoff map for the 60 river basins of Table 6....................................................95

A-III

Fig. 56 - Comparison of the holospheric runoff distribution between the UNESCO runoff map (corrected) and the empirical approach developed in this study. ..................................... 98

Fig. 57 - Comparison of total runoff determined with the water budget model of Willmott et al. [1985] with total runoff extracted from the UNESCO runoff map for the 60 river basins of Table 6. ..................................................................................................................... 99

Fig. 58a - Holospheric distribution of precipitation and evapotranspiration resulting from the UNESCO precipitation and maps and of the ECHAM2 outputs. ............................................ 100

Fig. 58b - Comparison of the spatial runoff distribution between the UNESCO runoff map and the model output of the ECHAM2 General Circulation Model. ....................................... 101

Fig. 59 - Proposed relationships between denudation rate and mean annual precipitation................ 108 Fig. 60 - Regions of the continents belonging to orogenies younger than 250 Million years. .......... 111 Fig. 61 - Correlation between sediment yields and different environmental parameters for the

60 river basins of Table 12, as well as correlation between these environmental parameters. ............................................................................................................................... 114

Fig. 62 - Plot of mean annual sediment concentration versus mean annual drainage intensity for the river basins of Table 12. ............................................................................................... 116

Fig. 63 - Comparison of the potential of the soils to be water limited with the distribution of Four in different climate types. ................................................................................................ 118

Fig. 64 - Mean biomass density in the hexagons of the Holdridge Triangle. .................................... 119 Fig. 65 - Plot of predicted versus observed sediment yields according to the best correlated

parameter products in Table 14................................................................................................ 120 Fig. 66a - Hypsometric curves of the river basins of Table 12. ......................................................... 124 Fig. 66b - Histograms of the slope distribution of the grid points in the river basins of Table

12. ............................................................................................................................................ 125 Fig. 67a - Sediment yields from the continents as predicted with the empirical relationships

found in this study.................................................................................................................... 131 Fig. 67b - As Fig. Erreur ! Signet non défini.a, but this time the modelling was coupled

with an interpolation of the observed sediment yields for the 60 river basins investigated in this study.......................................................................................................... 133

Fig 68a - Correlation between fluxes of dissolved organic carbon fluxes of particulate organic carbon and the environmental patterns characterizing the river basins for all rivers of Table 20. .................................................................................................................... 142

Fig 68b - As Fig. 68a, but this time only calculated on the basis of the rivers of Table 20 with the highest data quality index. ................................................................................................. 142

Fig. 69 - Comparison of observed DOC fluxes with the DOC fluxes predicted in this study. .......... 143 Fig. 70 - Plot of POC% versus TSS concentration for 19 world rivers. ............................................ 145 Fig. 71 - Specific DOC fluxes on the continents as predicted with the empirical relationships

found in this study.................................................................................................................... 149 Fig. 72 - Specific POC fluxes on the continents as predicted with the empirical relationships

found in this study.................................................................................................................... 151 Fig. 73 - Distribution of the average organic carbon content in the soils in the Holdridge

Triangle. ................................................................................................................................... 153 Fig. 74 - Relationships of the specific CO2 consumption by rock weathering and drainage

intensity for six different rock types. ....................................................................................... 157 Fig. 75a - Global 4° x 5° latitude/longitude river routing file of Miller et al. [1994]. ....................... 158 Fig. 75b - Plot of the basin areas resulting from the global 2° x 2.5° latitude/longitude river

routing file of Miller et al. [1994] versus the basin areas calculated in chapter II for 38 of the river basins considered in this study. ............................................................................. 158

Fig. 75c - As 75b, but this time calculated with the 4° x 5° latitude/longitude river routing file of Miller et al. [1994]. ....................................................................................................... 158

Fig. 76 - FCO2-RW on the continents as predicted with GEM-CO2. .................................................... 161 Fig. 77 - FCO2 on the continents as predicted with GEM-CO2 and GEM-Corg. ................................. 163 Fig. 78 - Holospheric distribution of FCO2 as calculated with GEM-CO2 and GEM-Corg. ................ 166

A-IV

Fig. 79 - Global river inputs of alkalinity (FHCO3), dissolved organic carbon (FDOC), and particulate organic carbon (FPOC) to the oceans in a 4° x 5° latitude/longitude resolution..................................................................................................................................167

Fig. 80 - Steady state budget for river carbon. ...................................................................................171 Fig. 81a - Comparison of observed bicarbonate fluxes with the values calculated with GEM-

CO2 ..........................................................................................................................................173 Fig. 81b - As Fig. 81a, but this time the GEM-CO2 output was corrected according to the

method discussed in the text.....................................................................................................173 Fig. 82 - Holospheric distribution of FCO2 , as calculated with GEM-CO2 and GEM-Corg.

With respect to Figure 78, the GEM-CO2 outputs have been corrected according to the climatic correction factors determined in the text. ...................................................................175

Fig. 83 - Flow diagram of the relationships forming the models GEM-CO2 and GEM Corg in order to predict the consumption of atmospheric CO2 by continental erosion. .......................178

A-V

LIST OF TABLES

Table 1. Monthly export of dissolved organic carbon from the Strengbach catchment (October 1993 to September 1995).......................................................................................... 18

Table 2. Mean annual runoff of the Niger River from upstream to downstream. ............................. 42 Table 3. Summary of the river case studies presented in the text. .................................................... 50 Table 4. Digitized river basins and their basin areas......................................................................... 56 Table 5. Abundance of the different climate types in the investigated river basins.......................... 67 Table 6. Comparison of drainage intensities extracted from the UNESCO runoff map with

literature estimates. .................................................................................................................. 76 Table 7. Estimated fluxes of river water to the oceans...................................................................... 80 Table 8. Correlation coefficients between RR and certain parameters, calculated on the basis

of the mean river basin values.................................................................................................. 91 Table 9. Comparison of the distribution of wet and dry basin parts in the river basins of

Table 6 with the amount of total runoff originating from these basin parts. ........................... 92 Table 10. Correlation coefficients between RR and certain parameters, calculated on the

basis of the mean grid point values.......................................................................................... 93 Table 11. Multiple correlation models for RR in different climate types. ........................................ 94 Table 12. Sediment yields for 60 world rivers. ................................................................................. 106 Table 13. Summary of the different models for sediment yield prediction discussed in the

text. .......................................................................................................................................... 113 Table 14. Regression of sediment yield versus several parameters and parameter products.

The rivers are grouped according to their average climatic situation. ..................................... 117 Table 15. Orogeny and hypsometry of the river basins of Table 12. ................................................ 123 Table 16. Regression of sediment yield versus several parameter products. The rivers are

grouped according to basin hypsometry. ................................................................................. 126 Table 17. Regression of sediment yield versus several parameter products. The rivers are

grouped according to orogeny. ................................................................................................ 127 Table 18. Regression of sediment yield versus several parameter products. The rivers are

grouped according to the percentage of cultivated area in the basins...................................... 128 Table 19. Fluxes of water and of sediments from the continents to the oceans. ............................... 130 Table 20. Mean annual fluxes of DOC and POC for some world rivers........................................... 139 Table 21. Average environmental characteristics for some world rivers. ......................................... 140 Table 22. Estimated fluxes of water, sediments, DOC, and POC to the oceans. .............................. 148 Table 23. Regional budgets for the atmospheric CO2 consumption by continental erosion

and for river carbon fluxes to the oceans. ................................................................................ 165 Table 24. River fluxes of carbon to the world's oceans..................................................................... 169 Table 25. Bicarbonate fluxes of major world rivers.......................................................................... 174 Table 26. Effects of the climatic correction factors determined in the text on the budgets for

the bicarbonate fluxes. ............................................................................................................. 175 Table 27. Comparison of the distribution of continental area and of major climate types at

present-day with the situation during the last glacial maximum.............................................. 177 Table 28a. Estimated fluxes of water, sediments, DOC, POC, and atmospheric CO2

consumed by rock weathering to the oceans LGM.................................................................. 179 Table 28b. Changes in percent of the fluxes in Table 28a compared to the present-day

fluxes........................................................................................................................................ 180

Résumé

L'objectif scientifique de cette étude est de quantifier à l'échelle globale les flux de carbone organique apportés chaque année aux océans, afin d'évaluer le rôle de ces flux dans le cycle global du carbone. Les flux de carbone organique concernent, de manière à peu près équivalente, le carbone organique dissous (COD) et le carbone organique particulaire (COP). Pour ces deux formes de carbone, les facteurs de contrôle sont identifiés et des modèles empiriques sont établi pour extrapoler les flux à l'échelle globale et régionale. Ces flux étant fortement associés aux flux d'eau (COD, COP) et de sédiments (COP), une telle investigation n'est possible qu'avec un examen détaillé des facteurs principaux de contrôle de ces deux paramètres clés à l'échelle globale. La plupart des relations établies dans cette étude peuvent être appliquées à d'autres problèmes scientifiques concernant l'érosion et les transports fluviaux de matière des continents vers les océans.

L'approche de cette étude est basée sur un ensemble de 60 grands fleuves du monde. Les caractéristiques hydroclimatiques, biologiques, géomorphologiques et lithologiques des bassins versants de ces fleuves sont extraites d'un grand nombre de banques de donnés globales à partir des contours digitalisés de ces bassins. Ensuite, ces caractéristiques sont utilisées pour des analyses statistiques avec les données de la littérature sur les flux de carbone, d'eau et des sédiments dans ces fleuves.

En ce qui concerne le COD, un modèle linéaire de régression multiple incluant l'intensité de drainage, les pentes et la quantité de carbone stockée dans les sols constitue le meilleur modèle pour prévoir les flux à l'échelle globale. Les flux de COD deviennent plus importants avec des intensités de drainage croissantes, des morphologies plus aplaties et de des stocks de carbone plus importants dans les sols. En ce qui concerne le COP, les flux de sédiments représentent le principal facteur de contrôle. Les flux de COP augmentent généralement avec les flux de sédiments, mais le pourcentage de carbone organique dans ces sédiments en suspension diminue nettement quand les concentrations en sédiments augmentent. Cette relation a pu être modélisée dans cette étude, nous permettant ainsi de calculer le flux total de COP vers les océans en fonction du flux de sédiments et de l'intensité du drainage.

Les flux spécifiques de sédiments sont corrélés avec le produit des facteurs hydroclimatiques, géomorphologiques et lithologiques, c'est-à-dire l'intensité du drainage, les pentes, un index pour caractériser la dureté des roches et un index pour caractériser la variabilité des précipitations au cours de l'année. Les précipitations et l'évapotranspiration potentielle - qui est ici uniquement calculée en fonction de la température - sont les facteurs principaux de contrôle du drainage. En moyenne annuelle, le coefficient d'écoulement (le rapport entre l'écoulement fluvial et les précipitations) augmente d'autant plus que les précipitations dépassent l'évapotranspiration potentielle. Cependant, la morphologie représente également un facteur de contrôle non négligeable pour le drainage. Dans les mêmes conditions de température et de précipitations, de fortes pentes augmentent le coefficient d'écoulement, tandis que des élévations plus fortes le diminuent. Enfin une forte saisonnalité des précipitations augmente aussi le coefficient d'écoulement.

Lorsqu'on applique le modèle de régression pour le COD à la surface totale des continents à partir de l'ensemble des données existantes, la quantité totale de carbone organique dissous exporté vers les océans est estimée à environ 0,21 gigatonnes de carbone par an (GtC/an). Pour le COP, cette quantité est d'environ 0,16 GtC/an. Les flux correspondant d'eaux et de sédiments sont respectivement de 41750 km3/an et 16 Gt/an. Ces valeurs sont proches d'estimations précédemment publiées. Par conséquent, pour tous ces transports fluviaux, des cartes globales représentant la distribution spatiale des flux spécifiques sur les continents sont ainsi présentées, et des bilans détaillés sont proposés pour les différents continents, bassins océaniques et type de climats. De plus, la modélisation des transports fluviaux de carbone organique est couplée à une modélisation de carbone inorganique développée par Amiotte-Suchet [1995]. Comme les transports fluviaux de carbone organique et inorganique représentent un transfert permanent de carbone atmosphérique des continents vers les océans, cette étude propose une régionalisation de la consommation de CO2 atmosphérique sur les différents continents. Ainsi, le flux total de carbone atmosphérique utilisé par l'érosion continentale et transféré annuellement vers l'ensemble des océans peut être estimé à 0,60 Gt/an, dont 38% sous forme de carbone inorganique dissous, 35% sous forme de COD et 27% sous forme de COP.

Finalement, il est tenté d'estimer les flux de carbone liés à l'érosion continentale pour des conditions climatiques autres que celles d'aujourd'hui. L'ensemble des relations empiriques présentées dans cette étude sont enfin appliquées à un scénario climatique pour le dernier maximum glaciaire (18000 ans B.P.) et les implications pour les transports fluviaux de carbone sont discutés.

Abstract

The purpose of this study is to develop a global and regional quantification of the amount of organic carbon that is discharged to the oceans every year by rivers in order to evaluate the importance of this transport pathway within the global carbon cycle. This concerns in about equal parts the river fluxes of dissolved organic carbon (DOC) and of particulate organic carbon (POC). For both carbon forms, the major controlling factors at the global scale are determined, and empirical regression models are established which allow an extrapolation of the fluxes to regional and global scales. Because the organic carbon fluxes are strongly coupled to the fluxes of water (DOC, POC) and of sediments (POC), such an investigation is not possible without a similar examination of the major controls of these two key parameters at the global scale.

The approach considered in this study is based on a set of 60 major world rivers. The hydroclimatic, biological, geomorphological, and lithological characteristics of the drainage basins are extracted from a large number of environmental data sets using the digitized basin contours. These characteristics are then used for statistical analyses together with literature data for these rivers on the observed fluxes of organic carbon, water, and sediments.

For DOC, it is found that a multiple regression model including drainage intensity, basin slope, and the amount of carbon stored in the soils is the best model to predict fluxes on a global scale. They become greater with increasing drainage intensities, flatter morphologies, and larger carbon reservoirs in the soils. For POC, sediment fluxes are the dominant controlling parameter. It can be shown that POC fluxes generally increase with increasing sediment fluxes, but the percentage of organic carbon in the total suspended solids clearly decreases with increasing sediment concentrations. This relationship was mathematically fitted, allowing to calculate POC fluxes to the oceans as a function of sediment yields (sediment fluxes divided by basin area) and of drainage intensity.

Sediment yields can be best correlated by forming the products of hydroclimatic, geomorphological, and lithological factors, that is drainage intensity, basin slope, an index characterizing rock hardness, and an index characterizing rainfall variability over the year. The best correlated parameter combination varies to some extent when the rivers are grouped according to their average climatic situation, but it is always a combination of the above mentioned parameters that yields the greatest correlation coefficients. Precipitation and potential evapotranspiration, which is here uniquely calculated as a function of temperature, are the principal elements controlling runoff. Mean annual runoff ratio (runoff divided by precipitation) generally increases as precipitation exceeds potential evapotranspiration. Morphology, however, also represents a non-negligible controlling factor. Given the same temperature and precipitation values, a greater slope increases runoff ratio, while greater elevation decreases it. Finally, runoff ratio is also increased by a highly seasonal precipitation distribution over the year.

When the determined regression model for DOC is applied to the total continental area on the basis of the corresponding data sets, the total amount of dissolved organic carbon discharged to the oceans is calculated to be about 0.21 gigatons of carbon per year (GtC/yr). For POC, it amounts 0.16 GtC/yr. The corresponding fluxes of water and of sediments are 41750 km3/yr and 16 Gt/yr, respectively. These values compare well with previous literature estimates. Consequently, for all of these river fluxes, global maps showing the distribution of the specific fluxes on the continents are shown, and detailed budgets are given with respect to the different continents, ocean basins, and major climate types. Moreover, the modelling of the river fluxes of organic carbon is coupled with a modelling of the river fluxes of inorganic carbon that has been developed by Amiotte-Suchet [1995]. Because both inorganic and organic carbon fluxes account for a permanent transfer of carbon from the atmosphere to the oceans, a detailed picture of the consumption of atmospheric CO2 by continental erosion can thus be drawn. It is calculated that a total amount of 0.60 GtC of carbon is withdrawn from the atmosphere every year by continental erosion, with about 38% that can be attributed to dissolved inorganic carbon, 35% to DOC, and 27% to POC.

Finally, an attempt is made to estimate the carbon fluxes for other than present-day conditions. The overall empirical relationships established and used in this work are then applied to a last glacial maximum scenario (18000 yrs b.p.) derived from a general circulation model run, and the consequences for the river carbon fluxes are discussed.

THESE - univ-perp.fr

Documents

Transcript of THESE - univ-perp.fr