Should topographic metrics be considered when predicting species density of birds on a large geographical scale? A case of Random Forest approach

Abstract

Species Distribution Modelling (SDM) is a group of statistical tools that describe species distribution in environmental gradients in order to create their predictive distribution. However, due to the complexity of factors that influence the occurrence or density of species these methods’ effectiveness is still debatable. That is why we decided to explore how topographic metrics, such as altitude, slope, roughness and aspect, would affect the density of farmland (Icterine warbler) and forest (Eurasian golden oriole) bird species. We generated two sets of SDMs for each of the two bird species: One set of models contained topographic metrics as a predictor variable, and the other did not. Out-of-back procedures in the Random Forest approach and evaluation models based on independent dataset scores showed that omitting topographic metrics as predictors resulted in a substantial reduction of model performance for both lowland and upland bird species. Further analysis of predictive maps revealed that neglecting topographic metrics resulted in large over-predictions of species’ densities in regions where these species were rare. Importantly, our results also support the notion that detailed topographic metrics can be considered as a surrogate for elusive climatic factors and the habitat’s condition. Hence, the study emphasises that the process of selecting predictor variables, especially topographic metrics, is one of key elements in developing Species Distribution Models for birds, even for those species which are not directly dependant on the topographic metrics.