Continental scale investigations of tree cover in Africa have suggested that forest and savanna act as alternative stable states, with observed bimodality between 1300 and 2000 mm of rainfall. This bimodality in tree cover is taken as evidence for the existence of bistability in the system as it is nearly impossible to detect bistability directly. However, human activities, such as deforestation, agriculture, pastoralism can also result in bimodal distributions in tree cover. In this study, we used data from sub-Saharan Africa to determine the extent of bimodality arising due to intrinsic dynamics and due to human activities. We used data from the Human footprint map to delimit sub-Saharan Africa into 'human-impacted' (high human footprint index scores) and 'natural' (low human footprint index scores) regions. For each region, we obtained a cumulative environmental predictor that includes mean annual rainfall, rainfall seasonality and soil variables. The extent of bimodality for each region was determined from a plot of vegetation cover (using enhanced vegetation index) as a function of the cumulative environmental predictor. Finally, using spatial datasets of fire frequency and domestic herbivore densities, we determined how human activities may impact existing feedback loops that maintain forests and savannas.
Results/Conclusions
The first principal component of a PCA explained >95% of the variance and was used as the environmental predictor; it correlated strongly with mean annual rainfall. We find that the range of environmental values for which we observe bimodality is larger for the human-impacted region than the natural region. This suggests that some of the observed continental-scale bimodality is contributed by human activities. Unlike previous results from South America, our results show that many regions in Africa do show patterns consistent with intrinsic dynamics, although the extent is smaller than previously thought. We find that fire frequencies and stocking densities of domestic herbivores were different in the human-impacted and the natural regions. Our results show that forest loss, fire frequencies and herbivore densities influence the switches from high-to-low and low-to-high cover in distinct ways, contingent on human prioritisation of land use. To conclude, we find that human-impacted regions do show a larger range of environmental variables over which bimodality in vegetation cover is observed, and that changes in fire frequencies and herbivore densities maybe driving this observed pattern.