Habitat models based on climate have been used in a variety of scenarios to make predictions about the ranges of species under novel conditions (such as hindcasting through past climate change or invasive species on new continents). The consenus is that these analyses: 1) suggest that habitat models work fairly well, and 2) that climate clearly must be the dominant mechanism driving species ranges. However, most of these assessments have used non-quantitative (often visual) measures of how well the new range is predicted and have failed to address issues like statistical non-independence due to spatial autocorrelation. We present a quantitative assessment of how successful habitat models are at predicting ranges. We also explore how strongly habitat models implicate climate as a causal mechanism by comparing the predictive power of climate to the predictive power of other factors. Both analyses are done using the North American Breeding Bird Survey, allowing us to address these questions across approximatlely 100 species.
We find that the predictive success of habitat models depends heavily on what exactly we try to predict (presence/absence vs. abundance) and on whether we allow spatial autocorrelation to 'assist" the predictions. We also find that the exclusive importance of climate needs to be questioned. We explore the implications of these results for applied situations and draw a distinction between interpolation vs. extrapolation. We also explore methods that bring about improvement.