Using Landscape-Level Data to Predict Presence of Hammond's Flycatcher, Dusky Flycatcher, and Gray Flycatcher in Dry-Pine Forests of North-Central Washington

Abstract

I develop a model to predict the presence of three species of flycatcher; Hammond's Flycatcher (Empidonax hammondii), Dusky Flycatcher (Empidonax oberholseri), and Gray Flycatcher (Empidonax wrightii) using landscape-level data, statistical software packages, and ArcGIS software that were readily available via the internet. Point-count data used in the study were collected as part of a United States Department of Agriculture (USDA) Forest Service Birds and Burning study in northcentral Washington. The geospatial data for this study included three 30-m resolution Landsat Thematic Mapper (LTM) raster files and a 30-m resolution Digital Elevation Model (DEM) raster file covering the study area. Model development was achieved using logistic regressions and habitat selection calculations. Arc GIS raster calculator was used to create a predictive raster layer for each target species representing those habitats selected in the modeling process. Predictive raster layers were compared to point-count stations where,presence/absence for each species was known and percent concordance was recorded. The Hammond's Flycatcher model had an 81.0% concordance with point-count stations where the species was present. The Dusky Flycatcher model accurately predicted the species presence 78.0% of the time, and the model for Gray Flycatcher achieved 30.0% concordance. Predictive models were compared to randomly generated points to test model performance. The mean percent concordance between the Hammond's Flycatcher model and random sites was 66.9% (SD= 12.2), 22.3% (SD= 8.80) for Dusky Flycatcher, and 21.7% (SD= 8.70) for Gray Flycatcher. Results oft-tests suggest that model performance was significantly better at predicting species presence than random sites. The analysis procedures presented in this study differed from other methods in their relative simplicity, yet achieved results similar to other predictive models with an average model concordance of 63% for all three models.