The effect of fire on tree–grass coexistence in savannas: a simulation study

The savanna biome has the greatest burned area globally. Whereas the global distribution of most biomes can be predicted successfully from climatic variables, this is not so for savannas. Attempts to dynamically model the distribution of savannas, including a realistically varying tree : grass ratio are fraught with difficulties. In a simulation study using the dynamic vegetation model LPJ-GUESS we investigate the effect of fire on the tree : grass ratios as well as the biome distribution on the African continent. We performed simulations at three spatial scales: locally, at four sites inside Kruger National Park (South Africa); regionally, along a precipitation gradient; and for the African continent. We evaluated the model using results of a fire experiment and found that the model underestimates the effect of fire on tree cover slightly. On a regional scale, high frequencies were able to prevent trees from outcompeting grasses in mesic regions between ~700 and 900 mm mean annual precipitation. Across the African continent, incorporation of fire improved notably the simulated distribution of the savanna biome. Our model results confirm the role of fire in determining savanna distributions, a notion that has been challenged by competing theories of tree–grass coexistence.