Widespread forest-savanna coexistence but limited bistability at a landscape scale in Central Africa

Tropical forest and savanna frequently coexist under the same climatic conditions, which has led to the hypothesis that they could represent alternative ecosystem states, stabilized by internal feedbacks. An implication of this hypothesis is that forest and savanna may be bistable and exhibit tipping behavior in response to changing conditions. However, we pose that the local presence of forest and savanna within coexistence landscapes is not sufficient evidence that these are alternative stable states at larger ecosystem scales. Therefore, we explore forest-savanna coexistence and bistability at landscape scale in Central Africa. Using remote sensing data on tree cover, we classify 0.1 degrees x 0.1 degrees (approx. 10 x 10 km) landscapes as homogeneous forest, homogeneous savanna, or coexistence, and analyze the roles of climate, topography and soil sand content in driving their distributions. We find that local coexistence of forest and savanna within landscapes is common and occurs for the whole range of mean annual precipitation in our study area. At low precipitation, however, coexistence increases with topographic roughness and is therefore likely driven by local redistribution of resources rather than internal feedbacks. Coexistence within intermediate and high precipitation landscapes remains unexplained by the studied variables, and may be caused either by heterogeneity in unmeasured drivers or by feedback-driven bistability. At landscape scale, the precipitation ranges for which homogeneous forest and savanna occur have only limited overlap, and this overlap can be largely explained by other external drivers, such as seasonality, soil sand content, and topography. This lack of evidence that homogeneous forest and savanna in Central Africa are alternative ecosystem states at this landscape scale means that transitions between them may be mostly local, resulting in coexistence states. Therefore, we conclude that the likelihood of large-scale tipping between homogeneous forest and savanna ecosystems may be lower than previously thought.