It is appropriate to analyze count data in ecology, such as species richness, using Poisson models. However, there has been limited work on incorporating spatial autocorrelation in Poisson models. The objective of this study was to use three spatial Poisson modeling techniques to investigate
the relationships between bird species richness and patterns of climate and land cover diversity in New York State. The three spatial Poisson models included auto-Poisson (AP), generalized linear mixed Poisson (GLMP), and geographically weighted Poisson regression (GWPR). The results from
the three models were compared with a global nonspatial Poisson (GP) model. Moran's I correlograms and local estimates of Moran's I were used to evaluate the global and local spatial autocorrelations of model residuals and spatially assess model performance. We found that the
spatial Poisson models produced better model predictions for bird species richness, significantly reduced spatial autocorrelation in model residuals, and generated more desirable spatial distributions for model residuals than the GP model. Overall, we found that the GWPR models were more effective
in reducing spatial autocorrelation of model residuals and incorporating spatial heterogeneity at different spatial scales than the AP and GLMP models. We conclude that the analysis of count data (e.g., species richness) can be effectively modeled using spatial Poisson models but that the
coefficients of environmental predictors may shift as a result of which method is used.