The use of species distribution models to inform amphibian conservation in western Canada

Species distribution models (SDMs) have become a widely used tool for understanding species distributions and informing conservation actions. However, user decisions made during model development can impact model predictions and performance. In this thesis, I address the sensitivity of model predictions and performance to several modeling decisions. I first reviewed modeling decisions used in studies that validated SDMs with independent data and found that the impacts of some decisions (i.e., choice of geographic study extent) have yet to be fully tested using independent data. I then used independent data to test the impacts of study extent on SDMs for six amphibian species at the edge of their respective ranges in western Canada. I found that model predictions were highly sensitive to the study extent used, and that many models failed to accurately predict independent occurrence regardless of the study extent used. Following from this result, I explored the joint impacts of choice of study extent, the type of environmental data used, and whether sampling bias was accounted for in SDMs developed to inform conservation translocations of long-toed salamanders. I found that these decisions impacted the prioritization of potential release sites, and that models developed using random background points and local study extents tended to perform best. In this study I also demonstrated an approach for incorporating future climatic predictions into the selection of potential release sites and found that tradeoffs exist between developing accurate models and those that can make future predictions without extrapolation beyond the conditions with which models were generated. Overall, my thesis contributes several validated SDMs for use in amphibian conservation in western Canada while simultaneously adding to our understanding of how key modeling decisions can impact SDM predictions, performance, and downstream conservation decisions.