K-fold cross-validation was used to determine the predictive ability of logistic regression estimated resource selection function (RSF) models. Models were evaluated and selected based on their general, spatial, and temporal predictive ability (3-way RPI or 3-way RSF Plot Index). This method was used to evaluate which remotely sensed and GIS-based predictor variables, acting as proxies for structural habitat characteristics, were effective for modelling habitat selection of eleven grassland bird species.; Five years of bird point count data from an area of native prairie were used. The 3-way RPI method is dependent on the assignment of output to arbitrary suitability classes. Methods to partially ameliorate the threshold dependency created by this class assignment were developed. The use of random, temporal, and spatial partitions of the data to evaluate general, temporal, and spatial model robustness was demonstrated to be superior to standard methods of general testing. |