Aim. To quantify whether species distribution models (SDMs) can reliably forecast species distributions under observed climate change. In particular, to test whether the predictive ability of SDMs depends on species traits or the inclusion of land cover and soil type, and whether distributional changes at expanding range margins can be predicted accurately. Location. Finland. Methods. Using 10-km resolution butterfly atlas data from two periods, 1992–1999 (t1) and 2002–2009 (t2), with a significant between-period temperature increase, we modelled the effects of climatic warming on butterfly distributions with boosted regression trees (BRTs) and generalized additive models (GAMs). We evaluated model performance by using the split-sample approach with data from t1 ("non-independent validation"), and then compared model projections based on data from t1 with species’ observed distributions in t2 ("independent validation"). We compared climate-only SDMs to SDMs including land cover, soil type, or both. Finally, we related model performance to species traits and compared observed and predicted distributional shifts at northern range margins. Results. SDMs showed fair to good model fits when modelling butterfly distributions under climate change. Model performance was lower with independent compared to non-independent validation and improved when land cover and soil type variables were included, compared to climate-only models. SDMs performed less well for highly mobile species and for species with long flight seasons and large ranges. When forecasting changes at northern range margins, correlations between observed and predicted range shifts were predominantly low. Main conclusions. SDMs accurately describe current distributions of most species, yet their performance varies with species traits and the inclusion of land cover and soil type variables. Moreover, their ability to predict range shifts under climate change is limited, especially at the expanding edge. More tests with independent validations are needed to fully understand the predictive potential of SDMs across taxa and biomes.
Global Ecology and Biogeography, 2013, Vol 22, Issue 12