Predicting range shifts of northern bird species: Influence of modelling technique and topography

Abstract

Using eight modelling techniques, bird atlas data of Finland and northern Norway, and data on ‘current’ climate (1971–1990) and projected future climate (2051–80; climate scenarios HadCM3 A2 and B1), we forecasted current and future distributions of 28 land bird species in northern Europe. Specifically, we investigated (i) the impact of inclusion of topographical information on the pure bioclimatic envelope models (‘climate -only’ models), (ii) the accuracy and spatial differences of the predictions from the different models, and (iii) the co-occurrence of hotspots in species numbers and proportion of protected areas, both currently and in the future. Modelling accuracy was measured as the cross-validation area under the curve (AUC) of the receiver operating characteristic plot. Generalized additive models (GAM) and generalized boosting method (GBM) gave parallel projections with high predictive accuracy for the species distributions and their hotspots, but random forests (RF) and artificial neural networks (ANN) also showed good model performance. Inclusion of topographical variables showed an overall tendency to increase the accuracy of the climate-only models, and this increase was statistically significant in GAM, GLM (generalized linear models) and RF. According to both climate-only and climate-topography models, coincidence of grid cells harboring species hotspots and larger protected areas appears to decline by 2051–80. This calls for further evaluation of the adequacy of the reserve network in northern latitudes, where species are probably particularly susceptible to the effects of climate changes.