Distribution modeling of species at a coarse scale: How are we doing?

The distribution and abundance of species are a major subject of research in ecology and conservation biology. Much advance has been made in distribution modeling in the last decades owing to modern technology (computing power and remote sensing) and innovate statistical techniques in particular at coarse scales and over large extents. We shed critical light on the success of coarse scale models based on environmental conditions and how that success was judged. First, we compared the performance of environment-based models to pure spatial interpolation (prediction through spatial position and/or neighboring values) on data from the North American Breeding Bird Survey. Second, we investigated the explanatory power of environmental variables on artificial ranges, to find out whether spatial coincidence could be partly responsible for the success of environmental models. Third, we evaluated environmental models by splitting ranges into eastern and western halves for training and test data. All approaches indicated that models based on environmental predictors were much weaker than presumed so far.