Evaluating performance of regional wildlife habitat models: Implications to resource planning

Wildlife management and policy decisions are increasingly being made at the regional, national and international level. Much of the motivation for extending the geographic context of resource planning comes from U.S. federal laws that require resource managing agencies to examine the potential impacts from land resource use on wildlife resources. Several regional approaches have demonstrated the feasibility of using extant resource inventories in developing wildlife habitat models. While testing for feasibility is a logical first step in the development of large-scale resource response models, resource planners need to know if these methods provide information that is useful in the decision-making process. This study has taken the next step of quantifying model performance as a means of characterizing the uncertainty in model predictions. Discriminant function analysis was used to relate the composition of land use and land cover within counties of the south-eastern United States to abundance and presence/absence classes of white-tailed deer, wild turkey and red-cockaded woodpecker. We evaluated model performance by examining: (1) classification accuracy of the model as a whole; and (2) the error and precision of the classification rules comprising the models. Although all models were found to perform significantly better than a random model, the performance criteria indicated that significant improvements in the models could be realized. Sources of model uncertainty are reviewed, and suggestions for how to control for that uncertainty are discussed as a future research need.