Can Species Richness Patterns Be Interpolated From a Limited Number of Well-Known Areas? Mapping Diversity Using GLM and Kriging

The limitations of biodiversity data are commonly overcome by modelling the geographic distribution of species and community characteristics. Here we evaluate two Assemblage-level Modelling (ALM) techniques, General Linear Models (GLM) and kriging, assessing their ability to predict scarab dung beetle richness in the Iberian Peninsula using two different strategies. Calibration Errors (ability to interpolate values within the conditions where the model was built) were assessed by means of a leave-one-out jackknife. Validation Errors (ability to provide partial extrapolations to different environmental conditions within the same geographic domain) were calculated by comparing model projections with an independent dataset. Although the forecasts within the calibration dataset were very good for GLM and extremely good for kriging, both techniques provided surprisingly poor extrapolations. We discuss why such poor performance may be related to non-stationarity in the factors driving diversity patterns, and how ALM may be improved to account for it.