Species specific modelling
When representative data (spatial and taxonomic) is available species-specific models are always the best approach. Using the same point data outlined above with the same environmental variables (but at resolution of 1km (0.008 degrees) rather than 10km (0.08 degrees)). Species with at least 3 points were modelled (this is too low for individual species models, but can still be useful in an approach where only patterns are being examined, though output accuracy can then be assessed by comparing to the IUCN ranges as whilst ranges within IUCN are artificial the overall patterns in terms of geographic regions should be broadly comparable). As noted variables should represent the conditions likely to delimit species ranges, reflect ecophysiological thresholds, and other habitat constraints, including habitat structure. Variable correlation is accounted for within Maxent, and species-specific variable selection can also be made using various R packages (i.e. ENMEval).
Once again, we ran models using the default parameters and used 3 replicates, and the average reclassified to a binary map (0:1) using the 10 percentile cloglog threshold. As models will note all environmentally suitable habitat regardless of biogeographic constraints, to reflect biogeography we then used the IUCN redlist to note species endemic to Madagascar, in Madagascar and continental Africa, and those limited to the African continent. It should be noted that a number of Madagascan endemic species lacked sufficient data to model, thus richness in Madagascar may be under-estimated, especially given the small ranges of some bats, but all models had an AUC exceeding 0.9, additional indicies such as Boyce index or AIC can be used to give independent measures of model performance and accuracy. These were then masked in batches using masks of each of those three regions. The raster calculator was then used to batch sum groups of 30 species within each of these three groups and numbered sequentially. We then combined all those species restricted to continental Africa using the raster calculator, before using the mosaic to new raster tool to combine the three types of biogeographic map to refine ranges and map modelled richness.
All maps are provided to show how they note richness patterns across the African continent.