Modelling bat distributions and diversity in a mountain landscape using focal predictors in ensemble of small models

Abstract

Aim: Bats are important components of mammalian biodiversity and strong bioindicators, but their fine-scale distributions often remain less known than other taxa (e.g., plants, birds). Yet as highly mobile species with multiple needs in the landscape, bats impose serious modelling challenges, such as advanced use of neighbourhood analyses. The aims of this study were to test the use of a designed sampling of bats for biodiversity and conservation assessments, and to find appropriate modelling solutions for providing nature practitioners with reliable potential bat distribution maps in a mountain area of high conservation interest. Location:The western Swiss Alps of Vaud. Methods:We conducted a one-year field survey combining passive acoustic recordings supplemented by mist net catching to collect data on bats. These data were then used to create univariate models with focal land use/cover variables using different focal window sizes to detect the optimal species-specific scale of influence for each variable. The large number of selected variables was then used to create ensembles of small models at a 100 m × 100 m resolution, and the resulting habitat suitability maps were transformed into species distribution maps for practitioners. Results:We were able to collect data to model 14 different bat species representing 66% of the Swiss bat diversity, including four red list species. In general, the most important variables were Euclidean distance to road or water, temperature and slope, but there was large variation among species both for the variable importance and for the optimal focal window size. Main conclusion: Our study greatly increased the knowledge of bats in this regionand showed that many of the red list species are nowadays disappearing from the intensively used lowland plains and restricted to the remaining forests along the slopes. Additionally, we highlighted the importance of selecting the variable scale on a species-specific basis accounting for their mobility and range sizes. K E Y W O R D S bioindicators, Continuous Boyce Index, habitat suitability model, multi-scale model, presenceonly model, rare species, Switzerland | 771 SCHERRER Et al.

Current and future climatic regions favourable for a globally introduced wild carnivore, the raccoon Procyon lotor

“…Model calibration. Nine different modelling techniques were calibrated and evaluated 126–129 : Artificial Neural Networks – ANN 130 (Ripley, 1996); Classification Tree Analysis - CTA 131 , Flexible Discriminant Analyses – FDA 132 ; Generalised Additive Models – GAM 133 ; Generalised Boosted Models – GBM 134 ; Generalised Linear Models – GLM 135 ; Multivariate Adaptative Regression Splines – MARS 136 ; MAXimum ENThropy – MAXENT 1 ; Random Forests – RF 137 . For each set of presence/pseudo-absence, model calibration was realized with 70% of all data.…”

Combining ensemble models and connectivity analyses to predict wolf expected dispersal routes through a lowland corridor

“…To integrate the output of the performed distribution models and develop the ensemble predictions, we used a weighted-averaging approach through which each single distribution model was weighted according to its predictive accuracy as independently assessed [43]. The performance of the ensemble models was evaluated using the AUC of the ROC [44,45,46,47,48]. Based on the results obtained from the models run for the two study areas, we developed two ensemble predictive maps for the whole landscape.…”

First assessment of habitat suitability and connectivity for the golden jackal in north-eastern Italy

“…To develop the ensemble prediction by integrating the output of the single distribution models, we used a weighted-averaging approach through which the distribution models were weighted according to their predictive accuracy (Thuiller et al 2009;Rodrıguez-Soto et al 2011). The performance of the ensemble model was evaluated using the AUC of the ROC (Fernandes et al 2019;Mohammadi et al 2019;Scherrer et al 2019).…”