Greve, Michelle6; Lykke, Anne Mette7; Overgaard, Anne Blach8; Svenning, J.-C.8
1 Ecoinformatics & Biodiversity, Faculty of Science, Aarhus University, Aarhus University2 Department of Terrestrial Ecology, National Environmental Research Institute, Aarhus University, Aarhus University3 Department of Bioscience - Soil Fauna Ecology and Ecotoxicology, Department of Bioscience, Science and Technology, Aarhus University4 Department of Bioscience - Plant and Insect Ecology, Department of Bioscience, Science and Technology, Aarhus University5 Department of Bioscience - Ecoinformatics and Biodiversity, Department of Bioscience, Science and Technology, Aarhus University6 Department of Bioscience - Soil Fauna Ecology and Ecotoxicology, Department of Bioscience, Science and Technology, Aarhus University7 Department of Bioscience - Plant and Insect Ecology, Department of Bioscience, Science and Technology, Aarhus University8 Department of Bioscience - Ecoinformatics and Biodiversity, Department of Bioscience, Science and Technology, Aarhus University
Aim To assess the influence of natural environmental factors and historic and current anthropogenic processes as determinants of vegetation distributions at a continental scale. Location Africa. Methods Boosted regression trees (BRTs) were used to model the distribution of African vegetation types, represented by remote-sensing-based land-cover (LC) types, as a function of environmental factors. The contribution of each predictor variable to the best models and the accuracy of all models were assessed. Subsequently, to test for anthropogenic vegetation transformation, the relationship between the number of BRT false presences per grid cell and human impact was evaluated using hurdle models. Finally, the relative contributions of environmental, current and historic anthropogenic factors on vegetation distribution were assessed using regression-based variation partitioning. Results Deserts and evergreen forests were best predicted by environmental variables, though most other LC classes were also relatively well predicted by the environment. Annual precipitation emerged as the most important determinant of all LC classes. At low rainfall levels, LC classes with increasing woody cover replaced each other as rainfall increased, while LC class rainfall optima overlapped at high rainfall levels. With some exceptions, anthropogenic factors had a relatively small influence on the distribution of most LC classes. However, anthropogenic factors did have an influence on the inaccuracies in BRT models, and these models provided an indication of which LC classes have been most reduced by transformation. Main conclusions Here we show, for the first time, how environmental and anthropogenic factors influence vegetation distribution across Africa. LC classes at rainfall extremes are best predicted by the environment. In addition, we corroborate, also for the first time, the much-stated claim that rainfall is the most important variable for the distribution of African vegetation for all African vegetation types. Finally, we indicate how anthropogenic drivers affect LC distributions.