Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

36
The mathematics of desertification: searching for early warning signals Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA) OR The dynamics of vegetation patterns

description

The mathematics of desertification: searching for early warning signals. Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA). OR. The dynamics of vegetation patterns. TexPoint fonts used in EMF. - PowerPoint PPT Presentation

Transcript of Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

Page 1: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

The mathematics of desertification: searching for early warning signals

Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

ORThe dynamics of vegetation patterns

Page 2: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

• Vegetation patterns & early warning signals.

• Pattern formation in reaction-diffusion models.

• The Busse balloon.

• The dynamics of modulated waves.

• Bifurcations with slowly varying parameters.

• Desertification scenarios.

• Claims/conjectures/comments/observations.

Structure of the talk

An intermezzo: the Hopf dance.

Page 3: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

Desertification: • a fast (‘catastrophic’) process in which a patterned state suddenly collapses. • caused by a slow change in the ‘environment’.

A(n almost) irreversible process: the desert is ‘very stable’.

Page 4: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

Early warning signals• The appearance of vegetation patterns is the first (very) early warning signal.• Patterned states are quite robust: they are stable under a large range of varying circumstances.• Nevertheless, they may suddenly collapse when a ‘threshold’ is crossed.

Tiger bushes/ Brousse tigrée

Questions• Can we predict the collapse?• Can we measure how close a system is to threshold?• Can we invent tools by which this ‘distance to threshold’ can be measured?

Page 5: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

Crucial difference: this (type of) system(s) is not spatially extended, the models only involve temporal dynamics.

The most simple example: The ‘catastrophic collapse’ induced by a saddle node bifurcation.

_y= ¯ +y2

A very simple early warning signal: the transition is preceded by ‘a slowing down’ behavior.

ODEs ↔ PDEs

[M. Scheffer, J. Bascompte, W.A. Brock, V. Brovkin, S.R. Carpenter, V. Dakos, H. Held, E.H. van Nes, M. Rietkerk, G. Sugihara (2009), Early-warning signals for critical transitions, Reviews, Nature 461(3) 53-59]

Page 6: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

YES

There are many PDE models in the ecological literature, all of reaction-diffusion type.

Central components:

• Biomass• Water• Groundwater

Translation into mathematical terms

Can one model vegetation dynamics?

[Meron, Klausmeier, Rietkerk, ...]

Page 7: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

The Klausmeier & Gray-Scott (GS) models

See the talk of Sjors van der Stelt this afternoon

Page 8: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

The dynamics of patterns in the GS equation

[J. Pearson (1993), Complex patterns in a simple system]

Page 9: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

‘stripes’ ‘labyrinths’

‘spots’

There is a (very) comparable richness in types of vegetation patterns ...

Page 10: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

It’s tempting to play a game of finding corresponding structures in the GS system and in actual vegetated areas ...

• Can we use our insights in GS to find ‘early warning signals’?

• Has this question perhaps already been considered (in disguise)?

More scientifically:

Page 11: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

The GS equation perhaps is one of the most well-studied reaction-diffusion equations of the last decades.

[Pearson, Swinney et al. 1994]

chemical reaction

numerical simulation

It’s mostly famous’ for exhibiting ‘self-replication dynamics’

Page 12: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

water

Let’s first consider one-dimensional patterns

In one space dimension:

Vegetation patterns correspond to singular spatially periodic solutions of the reaction-diffusion model.

biomass

Page 13: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

First question:

For which parameter values can spatially periodic patterns be observed?

bifu

rcat

ion

para

met

er R

wave number k

Region in (wavenumber, parameter)-space in which stable spatially periodic patterns exist

onset

[Busse, 1978] (convection)

The Busse balloon

Page 14: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

[D, Rademacher & van der Stelt, ’11]

Turing

‘Fall of patterns’ at k=0

A Busse balloon for the GS model

↔ yearly rainfall

Page 15: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

Periodic patterns near k=0: singular localized pulses (of vegetation pattern kind)

Coexisting stable patterns (for the same parameter values)

Page 16: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

• Near onset/the Turing bifurcation: ‘full analytical control’ through Ginzburg-Landau theory.

• A complete classification of the generic character of the boundary of the Busse balloon [Rademacher & Scheel, ’07].

• Near the ‘fall of patterns’: existence and stability of singular patterns [Doelman, Gardner & Kaper, ’01; van der Ploeg & D, ’05].

What do we know analytically?

See the talk of Sjors van der Stelt this afternoon

Note: Collapse takes place as ‘the system’ crosses through the boundary of the Busse balloon.

Page 17: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

A spin-off: the Hopf dance, a novel fine-structure!

A ‘dance’ of intertwining Hopf bifurcations.

The homoclinic (k=0) ‘oasis’ pattern is the last to destabilize

(Ni’s conjecture)

AN INTERMEZZO

Page 18: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

Two types of Hopf bifurcations?

in phase

out of phase

Why only these two?

Page 19: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

Spectral analysis

¤i(s)

¤j (s)

Note: § 1 endpoints correspond to H§ 1 Hopf bifurcations.

Page 20: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

+1

-1

» 1=k

Im(¸)

¤h(s)

The long wavelength limit ( k ~ 0)

Page 21: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

A novel general insight in the ‘fall of patterns’

In a general class of reaction-diffusion models:

• The homoclinic ‘oasis’ pattern is the last pattern to become unstable (↔Ni’s conjecture).

• The Hopf dance: near the destabilization of the homoclinic pattern, the Busse balloon has a ‘fine structure’ of two intertwining curves of Hopf bifurcations.

[D, Rademacher & van der Stelt, ’11]

END INTERMEZZO

Page 22: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

There is no uniquely determined ‘attractor’ (as in ODEs).

The dynamics within a (continuous) family of stable

patterns.

Back to desertification/the catastrophic collapse of patterns

What is still missing?

I. Which pattern does nature choose?

,

Page 23: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

The dynamics of modulated ‘waves’

[van Harten, ’95; D, Sandstede, Scheel & Schneider, ’09]

Page 24: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

Various ‘phase equations’ can be obtained:

• The Burgers equation.

• The Korteweg-de Vries equation.

• The Ginzburg-Landau equation.

• The Kuramoto-Sivashinsky equation.

• …

Near a side band type boundary of the Busse balloon,

,

Page 25: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

What is still missing?

II. The dynamics induced by slowly varying parameters.

Page 26: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

Answer: It may (for instance) significantly delay the bifurcation.

from [Baer, Erneux, Rinzel, ’89]

¹x = ¹x("t)

unstablestable

O(1) delay!

Hopf bifurcation curve

This type of issues has not (at all!) been studied in PDEs

Page 27: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

Desertification & early warning signals

Various different scenarios

A typical Busseballoon,destabilization possibleby:² thesidebandmechanism,² a saddlenodebifurcation,² a Hopf bifurcation,² co-dimension 2 corners.

à A=A("t)

Turing

Page 28: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

(i) – the Hollywood scenario: A decreases (relatively) fast

• The vegetation pattern cannot adapt (although the destabilization may be delayed).• The vegetation patterns collapse into the desert state with the same wave number as given by the Turing bifurcation.

Unlikely, has not been not observed

Page 29: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

(ii-iv) – Sliding along the boundary of the Busse balloon

A decreases so slow that the ‘internal dynamics’ within the family of spatially periodic patterns can adapt.

By decreasing its (average) wave number k the pattern slides back into the Busse balloon.

By this sliding mechanism, k decreases to realistic values

Page 30: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

Claims and/or conjectures

•The adaptation process is governed by the local character of the (boundary of the) Busse balloon.

(side band: a Kuramoto-Sivashinsky equation with non-constant coefficients (?))

•The ‘flexibility of the internal dynamics’ can be determined by a phase equation analysis.

And thus:

• There will be various destabilization mechanisms.• These can be recognized by the character of the internal dynamics.• This provides explicit early warning signals.

Page 31: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

Some final comments/observations

• Numerical simulations will play a crucial role in the analysis.• A priori, significant aspects of the analysis will have to be formal.

- Phase equations: about 30 years between first derivations & rigorous proofs.- Delayed bifurcations in ODEs: proper tool (‘geometric blow-up’) only developed 10-15 years after the first papers appeared in the literature.

• This is all only in one spatial dimension!- There’s not even a proper Ginzburg-Landau theory in 2D. - Similar conceptional ‘challenges’ for phase equations in 2D. - …

• However: 1-D insight will lead to 2D conjectures …

Page 32: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)
Page 33: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

THE BELLY DANCE

The spectral branch is only to leading order a straight line/an interval.

In general it will be (slightly) bent.

This may yield small regions of ‘internal Hopf destabilizations’ and the corners in the boundary of the BB will disappear ↔ the orientation of the belly.

belly

Page 34: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

A more typical Busse balloon?

Or more generic (?): sometimes a co-dimension 2 intersection, sometimes an `internal Hopf bridge’?

Page 35: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

This is however not the case. In the class of considered model system, a BELLY DANCE takes place.

The belly always points away from the Im-axis near the ‘corner’ at which +1 and -1 cross at the same time.

Page 36: Arjen Doelman (UL), Max Rietkerk (UU) Jens Rademacher (CWI) , Sjors van der Stelt (UvA)

WHY??

The theory includes in essence ‘all explicit models in the literature’ (!?). HOWEVER, if one looks carefully it’s clear these models are in fact very special.

→ Back to the drawing board …