Characterizing uncertainty in species distribution models derived from interpolated weather station data

Species distribution models (SDMs) are used to generate hypotheses regarding the potential distributions of species under different environmental conditions, such as forecasts of species range shifts in response to climate change and predictions of invasive species range expansions. However, an accurate description of species’ geographic ranges as a function of the environment requires that species observations and climatic variables are measured at the same spatial and temporal resolution, which is usually not the case. Weather station data are interpolated and these resulting continuous data layers are incorporated into SDMs, often without any uncertainty assessment. Here we quantify the effects of three unrelated but complementary aspects of uncertainty in weather station interpolations on SDM performance using MaxEnt. We examine the influence of topographic heterogeneity, interannual variability, and distance to station on the overand under-prediction of modeled North American bird distributions. Our species observations are derived from presence-absence information for 20 bird species with wellknown distributions. These three metrics of uncertainty in interpolated weather station data have varying contributions to overand under-prediction errors in SDMs. Topographic heterogeneity had the highest contribution to omission errors; the lowest contribution to commission errors was from Euclidean distance to station. The results confirm the importance of establishing an appropriate relational basis in time and space between species and climatic layers, providing key operational criteria for selection of species observations fed into SDMs. Our findings highlight the importance of identifying weather stations locations used in interpolated products, which will allow a characterization of some aspects of uncertainty and identification of regions where users need to be particularly careful when making a decision based on a