The importance of snow in species distribution models of arctic vegetation

Abstract

Snow cover is characteristic of high-latitude and -altitude ecosystems where snowpack properties regulate many ecological patterns and processes. Nevertheless, snow information is only rarely used as a predictor in species distribution models (SDMs). Methodological difficulties have been limiting both the quality and quantity of available snow information in SDMs. Here, we test whether incorporating remotely sensed snow information in baseline SDMs (using five climate-topography-soil variables) improves the accuracy of species occurrence and community level predictions. We use vegetation data recorded in 1200 study sites spanning a wide range of environmental conditions characteristic of mountain systems at high-latitudes. The data consist of 273 species from three ecologically different and evolutionarily distant taxonomical groups: vascular plants, mosses, and lichens. The inclusion of the snow persistence variable significantly improved the predictive performance of the distribution and community level predictions. The improvements were constant, irrespective of the evaluation metric used or the taxonomic group in question. Snow was the most influential predictor for 36% of the species and had, on average, the second highest variable importance scores of all the environmental variables considered. Consequently, models incorporating snow data produced markedly more refined distribution maps than simpler models. Snow information should not be neglected in the construction of species distribution models where ecosystems characterized by seasonal snow cover are concerned.