Savannah-forest dynamics: encroachment speed, model inference and spatial simulations

Forest encroachment over savannahs has been recurrently reported in the tropics over the last few decades, especially in northern tropical Africa. However, process-based, spatially explicit modelling of the phenomenon is still trailing broad-scale empirical observations. In this paper, we use remotely sensed diachronic data from Central Cameroon to calibrate a simple reaction-diffusion model, embodying dynamical interactions between grass and woody biomasses in the savannah biome. Landsat satellite image series over the Mpem and Djim National Park (MDNP) (Cameroon) witnessed a dramatic extension of forest over the last five decades, and our estimates of forest front speeds based on randomly sampled transects indeed yield higher values (5-7 m yr-1) than in the existing literature (0.5-2 m yr-1). We use model simulations to provide the first estimates of woody biomass dispersal coefficients. Since the study region did not provide examples of savannah progression, estimates of grass dispersal proved inconsistent and we reverted to literature-based historical data to reach rough estimates. We demonstrate that broad-scale remote sensing data allows for calibrating simple reaction-diffusion models of vegetation dynamics in the savannah biome. Such calibrated models become a general baseline of expected changes and a valuable tool to understand how spatial environmental factors may locally modulate the overall dynamics.