Implications of Spatial Autocorrelation and Dispersal for the Modeling of Species Distributions

Modeling the geographical distributions of wildlife species is important for ecology and conservation biology. Spatial autocorrelation in species distributions poses a problem for distribution modeling because it invalidates the assumption of independence among sample locations. I explored the prevalence and causes of spatial autocorrelation in data from the Breeding Bird Survey, covering the conterminous United States, using Regression Trees, Conditional Autoregressive Regressions (CAR), and the partitioning of variance. I also constructed a simulation model to investigate dispersal as a process contributing to spatial autocorrelation, and attempted to verify the connection between dispersal and spatial autocorrelation in species\u27 distributions in empirical data, using three indirect indices of dispersal. All 108 bird species modeled showed strong spatial autocorrelation, which was significantly better modeled with CAR models than with traditional regression-based distribution models. Not all autocorrelation could be explained by spatial autocorrelation in the underlying environmental factors, suggesting another process at work, which I hypothesized to be dispersal. In the simulation model, dispersal produced additional autocorrelation in the distribution of population abundances. The effect of dispersal on autocorrelation was modulated by the potential population growth rate, with low growth rates leading to a stronger effect. The effect of dispersal on population sizes was different between populations at the periphery and core of a range. Due to their relative isolation, peripheral populations received fewer immigrants than populations at the core, causing lower population sizes. Dispersal could therefore be an explanation for range structures independent of environmental conditions. The verification of dispersal as a partial cause of autocorrelation failed. The most plausible cause was the indirectness of the indices used to represent dispersal. Distribution modelers should generally include space explicitly in their models, especially for species with low potential population growth rates. Dispersal has a strong potential to shape species distributions and requires more explicit consideration in distribution models and conservation plans. To reach this goal, direct research on dispersal distances and strength is urgently needed. Disruptions in natural dispersal patterns through removal of habitat isolates populations and thus may harm species beyond the effects of only direct habitat removal