Broad-scale patterns of avian biodiversity in response to habitat heterogeneity in a semi-arid landscape

The rapid decline in biodiversity makes urgent the need to understand the distribution of species over broad spatial extents. Traditionally-used classified imagery-based approaches have limited usefulness for this because they may overlook important within-habitat components in highly heterogeneous ecosystems. The main objective of my dissertation was to develop remote sensing and statistical approaches, informed by ecological theory, for mapping and understanding patterns of avian biodiversity in a semi-arid ecosystem. The study area was the McGregor Range of Fort Bliss Army Reserve in the northern Chihuahuan Desert.;In the first three chapters I tested different remote sensing approaches for understanding the ecological factors that influence bird species richness and guild abundance. I used image texture measures as proxies for habitat heterogeneity and the Normalized Difference Vegetation Index as a proxy for habitat productivity for modeling species richness. I subsequently used spectral mixture analysis to calculate proportions of discrete habitat components within each 30 m pixel of a given study plot. My results emphasize that habitat heterogeneity is a main determinant of bird species richness and the abundance of some guilds in that ecosystem.;My fourth chapter addressed the ecological factors that affect the occurrence and fitness of the Loggerhead Shrike (Lanius ludovicianus). While I found significant statistical relationships between bird occurrence and habitat variables such as NDVI texture, I found no significant relationship between the habitat variables measured and measures of fitness. These results suggest a greater need for understanding what limits individual bird fitness in that ecosystem.;My fifth chapter stems from my M.S. in biometry, and focused on testing the usefulness of Bayesian Model Averaging for building predictive models in ecology. I found that the choice of model prior influences the accuracy of the predictions and that the prior associated with AIC model averaging does not necessarily lead to improved predictions.;Finally, I implemented the tools developed in my dissertation to produce maps of species abundance and occurrence across the study area. Using measures of image texture and elevation variables is a cost-effective and easy-to-implement alternative to the traditional use of classified imagery for wildlife habitat mapping.