Evaluation and application of predictive habitat modeling in ecology.

摘要

My dissertation research is an important contribution to the growing field of predictive habitat modeling in ecology. I investigate innovative approaches for evaluating the performance of different predictive habitat models and applying these methods to large scale ecological phenomena. Several predictive habitat models currently exist. It has been the focus of much research to determine which the best model(s) is. However, much of this research is undermined by biased data sets. To resolve this issue, I tested model performance with simulated data that is not prone to the usual biases of real data sets. In general, my results support the findings of previous studies in that models that accurately predicted species distributions with real occurrence data also showed superior performance using simulated occurrence data. Using the conclusions from the model evaluation analysis as a basis, I applied these methods to two independent research questions. I first identified certain variables that best predicted the occurrence of chronic wasting disease (CWD) in Nebraska. Chronic wasting disease is a newly emerging infectious disease found only in members of the deer family (Family Cervidae). Analysis of several different combinations of spatial, temporal, and environmental variables showed that the chance of recording a positive CWD case was greater the more time spent sampling and when that sampling was conducted in western Nebraska. For the second question, I predicted range expansion among six North American mammals and ascertaining what role environmental variables have in predicting those expansions. I used two predictive habitat models combined with climate, land cover, and elevation variables to predict distributions. I predicted range expansions accurately for two of the six species, suggesting that other factors influenced the distributions of the remaining species. My results demonstrate the applicability of predictive habitat modeling in ecology and provide insights into novel methods of evaluating model performance.