Spatial interactions influence the global distribution of savanna and forest
Tropical savanna and forest are home to more than one-fifth of global human population, supporting their livelihoods, but they sustain increasingly human impacts that constitute a substantial biodiversity risk. Predicting how these two biomes will respond to ongoing global change is of high priority.
Recent global analyses, supported by theoretical work, have suggested that savannas persist where climate can support forest because fire limits tree cover, which in turn promotes fire. This hypothesis constitutes a well-accepted framework explaining the two biomes coexistence in the tropics. Yet, questions remain, especially regarding the spatial distribution of forest and savanna in areas of potential bistability: savanna tends to occur near savanna, and forest near forest defining a sharp boundary between the two. What, then, explain this spatial patterning?
Using global tree cover (Landsat data), we quantified the degree of the spatial aggregation in savanna and forest in Africa and South-America; we estimated the transition rates from one biome to the other decadally. Whereas previous work has modeled stochastic transitions depending on constant rate of changes, we hypothesized that these rates depend both on environment and nearest neighbors. We then used these estimates to computationally model the temporal evolution of biome extent and distribution.
We found that spatial aggregation is very strong within each biome, and that the geometry of spatial aggregation depended on the geography of the continent studied. Estimates of transition rates in the modern context showed that the probability of a switch from savanna to forest, depended on rainfall but not on nearest neighbors, but that the probability of a switch from forest to savanna depended on both rainfall and nearest neighbors.
Simulations produced using these modern estimates could not exactly mimic the spatial pattern of forest and savanna currently observed in the tropics. We suggest two possibilities, (1), that transition rates have changed and modern estimates do not reflect the dynamic that produced the current distribution of biomes, and (2) that spatial interactions are not strong enough to define forest/savanna boundaries and the spatial pattern observed is a legacy of past climate.