Can we predict the distribution of heathland butterflies with heathland bird data?

Abstract

Abstract National or regional conservation strategies are usually based on available species distribution maps. However, very few taxonomic groups achieve a full coverage of the focal region. Distribution data of well-mapped taxonomic groups could help predict the distribution of less well-mapped groups and thus fill gaps in distribution maps. Here, we predict the distribution of five heathland butterflies in Flanders (north Belgium) using typical heathland bird distribution data as predictor variables. We compare predictions with those using only biotope or a combination of both biotope and bird data as variables. In addition, we test the transferability of 'bird', biotope and combined models to the Netherlands, an ecologically similar region. Transferability was tested in three separate sandy regions in the Netherlands at different distances from the region in which the models were built. For each of the five heathland butterflies, we applied logistic regressions on ten random model sets and tested the models on ten random evaluation sets within Flanders. We used the area under the curve (AUC) of the receiver-operating characteristics (ROC) plots to estimate model accuracy. Overall, bird models performed significantly better than biotope models but were not significantly different from the combined models in Flanders. In the Netherlands, the transferred biotope and the combined models performed better than the transferred 'bird models'. We conclude that on a local scale, birds can, to some extent, serve as proxies for biotope quality, but that biotope models are more robust when transferred to another region.