PARENT SESSION
Oral Session # 99: Late-Breaking Newsworthy Presentations.
Presiding: E Preisser and G Larocque
Friday, August 8. 8:30 AM to 12:00 PM, SITCC Meeting Room 205.
A landscape-scale approach for predicting songbird occurrence: An evaluative criterion for selecting models for wildlife management.
Wynne, J Judson1, 2, Block, William1, Sisk, Thomas 2, 1 USDA Forest Service, Rocky Mountain Research Station, Flagstaff, AZ2 Center of Environmental Science and Education, Northern Arizona University, Flagstaff, Arizona
ABSTRACT- Wildlife-habitat relationship models are employed routinely in guiding land management decisions. Understanding and identifying potential sources of error is imperative to providing managers with the highest quality models. We developed an evaluative criterion to 1) identify potential sources of error in model datasets, and 2) select the best habitat models. Using a competing models framework, we modeled habitat using classification tree and logistic regression models for eight songbird species on the Pinaleños Mountains, southeastern Arizona. A three-year dataset (1993-1995) of bird survey points, habitat information derived from literature, and landscape-scale variables were used to develop models. Models were verified using a one-year dataset (2002) of bird survey points. GIS information were considered of the highest quality with the best elevation model (for deriving elevation, slope and aspect), vegetation land cover (overall accuracy = 71.2%), and maps of springs and streams used. Sample sizes in the model-building dataset were considered small (≤ 30 samples for presence and absence) and verification data were collected during the 2002 drought. Although none of the species′ models attained 80% accuracy, most yielded overall accuracy values better than chance and were comparable to other studies using similar habitat variables. Low predictive success of these models was probably due to a combination of inappropriate study design, small sample size, environmental stochasticity in the verification dataset, and lack of fine-scale GIS information. Although use of these models in guiding management decisions is considered limited, the criterion developed provides a systematic framework for evaluating data quality for modeling wildlife-habitat relationships. We recommend using a method, which includes elements identified here for evaluating model datasets for the potential errors and to potentially reduce error propagation. This approach will ultimately provide land managers with higher quality wildlife-habitat relationship models.
Key words: model error, habitat modeling, error propagation, resource management