Presets
Playback
Rainfall and time
Plants
Slope
Camera
Layers
Quality
Model check
The science
What tiger bush is. On gentle semi-arid slopes (Niger, Mexico, Australia) vegetation can organise itself into stripes that run along the contours, separated by bare crusted soil. The stripes slowly march uphill. The classic explanation is the Klausmeier (1999) model: bare ground sheds rainwater downhill, and the next plant band downslope captures it. Plants get water, but also trap it, so bands feed themselves and starve the strip just uphill of them.
∂w/∂t = a − w − w n2 + v ∂w/∂x ∂n/∂t = w n2 − m n + ∂2n/∂x2Here w is soil water, n is plant biomass, a is rainfall, m is plant mortality (plus any grazing you add) and v is the runoff speed (it grows with slope steepness). The term w n2 is plant uptake: the square means that denser plants infiltrate water disproportionately better. All quantities are dimensionless, as in the original paper; the domain here is 240 units long with periodic ends and uphill is +x.
Uniform state and threshold. Setting the derivatives to zero gives uniform vegetation n± = (a ± √(a2 − 4m2)) / 2m, which exists only if a ≥ 2m. The upper branch is the realistic one. For a steep enough slope this uniform state is unstable to perturbations of a particular wavelength and breaks up into bands; the growth rate of a Fourier mode of wavenumber k comes from the eigenvalues of
[[ −1 − n2 + i v k , −2 w n ], [ n2 , 2 w n − m − k2 ]]Bands survive drought. Because bands harvest water from the bare strips, they persist at rainfall well below 2m, where uniform cover is impossible. The fourth graph shows this: ramp the rainfall down and biomass falls gradually, then the whole pattern collapses to bare soil at a low value of a. Ramping rainfall back up starting from bare soil with a few seeds shows that recovery needs much wetter conditions than collapse did (hysteresis), which is why overgrazed or drought-hit drylands are hard to restore.
Numerical checks run on this page. (1) The well-mixed ODE is integrated and compared with the analytic uniform state n+. (2) A fresh PDE run from a slightly noisy uniform state is allowed to develop and the wavelength of its strongest Fourier mode is compared with the maximum of the linear-stability growth rate. (3) Global budgets on the live state: the change in mean biomass and mean water over one time unit must equal the integrated sources and sinks, because diffusion and runoff move material but create none.
Numerics. Explicit diffusion and growth with step 0.004, an implicit decay for water (stable for dense plants) and runoff by an exact semi-Lagrangian shift. 480 cells.
Honest simplifications. A one-dimensional, dimensionless model: no rainfall pulses, soil layers, seed banks, tree roots, or terrain roughness, and the contour bands in the 3D view are the one-dimensional solution repeated along the contour with a wiggle added for realism and random shrub placement. Real band wavelengths (50-200 m) and migration speeds (0.2-1 m per year) depend on units you can only get by calibrating the dimensionless numbers. The 3D shrubs are a visual device: each is shown or hidden by comparing the local model biomass with a fixed random number.