The threats posed by global climatic change and the spread of invasive species has increased focus on quantifying environmental factors that influence the geographic distributions of species and how these factors change across space and time. In practice, the issue is one of detecting ecological boundaries and determining factors associated with them. There are two major challenges in this regard. First, the data required for inference are often lacking or, when available, are often spatially homogenized, i.e., summarized over geographical or political regions such states, countries, biomes, etc. Data of this sort are often termed areal data. Second, data arising from neighboring regions are often more highly correlated than distant neighbors. Such spatial structure is often of ecological interest, but must be accounted for in order to obtain valid inference. Here we describe the application of Bayesian areal wombling, an established method in public health and epidemiology, to investigate and quantify environmental factors that influence the spread of an invasive species, the hemlock woolly adelgid ('HWA', Adelges tsugae Annand). Despite being a pest of great concern, the best data available for describing the spread of HWA are the year of first reported infestation for counties in the eastern Untied States. We ask: How do temperature, human population density, and host (hemlock) abundance influence spread rates and how does this influence vary across the current range of HWA?
Results/Conclusions
High probability boundaries (i.e., likely barriers to spread) were concentrated in regions of low hemlock abundance and low winter temperatures. However, it was difficult to separate the effects of hemlock abundance and temperature in northeastern United States, as these variables exhibit strong correlation in this region. Low probability boundaries were common in the southeastern Untied States, where spread of HWA has been rapid and where local hemlock abundance is relatively high and winters are relatively warm. There was no discernible influence of human population density on spread rates. These findings generally are consistent with the theorized influence of host abundance and temperature on population and dispersal dynamics of HWA. More broadly, our results demonstrate the potential utility of Bayesian areal wombling in identifying and quantifying local landscape influences on the geographic distribution of species. Additionally, this approach may facilitate rigorous analysis of existing but underutilized biogeographical datasets, the outcomes of which may inform mathematical models for the description and prediction of invasive spread and range shifts under climate change.
