Modelling butterfly species richness using mesoscale environmental variables: Model construction and validation for mountain ranges in the Great Basin of western North America

If species richness can be modelled as a function of easily quantified environmental variables, the scientific foundation for land-use planning will be strengthened. We used Poisson regression to develop a predictive model of species richness of resident butterflies in the central Great Basin of western North America. Species inventory data and values for 14 environmental variables from 49 locations (canyon segments) in the Toquima Range (Nevada, USA) were used to build the model. We also included squares of the environmental variables to accommodate potential non-linear relationships. Species richness of butterflies was a significant function of elevation and local topographic heterogeneity, with the selected model explaining 57% of the total deviance of species richness. Predictive variables can be derived efficiently from GIS-based data for areas in which species inventories have not yet been conducted. Therefore, in addition to evaluating the ability of the model to explain observed variation in species richness, we generated and tested predictions of species richness for ‘new’ locations that had not been used to build the model. Predictions were effective for five new segments also located in the Toquima Range, but not for 22 new segments in the nearby Shoshone Range. We discuss issues related to generalizability and data quality in model development.