Combining habitat suitability with simulated movements for the predictive modelling of beaver distribution in Flanders

Abstract

Abstract In Flanders, beavers reappeared since 2000 and numbers have expanded rapidly over the last decade, challenging impact management. To help identify areas with increased vulnerability to beaver damage we developed a species distribution model (SDM) in which we combined a MaxEnt-based approach to estimate habitat suitability and potential future spread with simulated dispersions using the SiMRiv R-package. We constructed our model in 2020 based on 2019 distribution data of beavers in Flanders and used it to predict population distribution probability up to 2022. For validation of this combined model output, we compared the output per km² grid cell to the observed distribution of beavers in Flanders in 2020–2022. Our model predicted possible beaver occupancy in 4.361 out of 14,343 km² grid cells in Flanders. In 51.4% of these grid cells predicted probability of occupancy was accidental, in 29.3% it was low, in 13.0% medium and in 6.4% high. In 2020–2022 beavers were observed in 908 km² grid cells in Flanders. Comparison with the model output showed that the model performed well overall (AUC: 0.825–0.864); although the negative predictive power was much higher than positive predictive power since the model tends to overestimate actual occupancy over this three-year period. However, the accuracy rises with the projected level of occupancy probability. Although the overestimation fits the intended risk management purposes of the model, expanding it beyond the borders of Flanders and strengthening the backbone of the model with a more detailed population model could further improve the output and help improve the positive predictive power.