Fire feedbacks with vegetation in a cellular automaton model of savannas

Background/Question/Methods Disturbance models have demonstrated that positive feedbacks between vegetation and fire can stabilize ecosystems in a savanna state with trees and grasses coexisting. Models that incorporate positive fire-vegetation feedbacks on savanna systems do not usually consider space, despite the importance of fire spread as an inherently spatial process. We use a cellular automaton model to investigate the potential for a spatially explicit, local feedback between vegetation and fire probability to stabilize the landscape in a savanna state. Our model allows for a cell to be in a grass, savanna tree, or forest tree state. Trees can be in either a juvenile or adult age class with tree mortality in fire dependent on age class. Our model explicitly considers dispersal and fire spread through the landscape. We specifically allow for the probability of fire spreading to a cell to be a function of the local neighborhood, with the probability maximized where grasses interface with savanna trees. This behavior is based on observations from empirical studies in savannas of the southeastern U.S. where fire temperature and spread are highest where flammable needles from adjacent savanna pines mix with grass fuels. Results/Conclusions We show that this spatially localized feedback stabilizes the ecosystem in a savanna state with a spatially overdispersed pattern of savanna trees that is consistent with observations from southeastern pine savannas. The absence of this feedback results in an abrupt transition from grassland to forest both spatially within the landscape and globally as fire becomes less frequent in the landscape: no intermediate savanna state is reached.