Fig. 3

Comparison of model performance metrics for Dermacentor reticulatus using three modelling algorithms [generalised additive models (GAM), maximum entropy (MaxEnt) and random forests (RF)], with four explanatory variable sets (bioclimatic variables, WorldClim, TerraClimate and MODIS satellite-derived variables) and with eight training extents (100 km–700 km buffering extents around occurrence data increasing in increments of 100 km and the European extent). The area under the receiver-operating characteristic curve (AUC), true skill statistic (TSS), omission rate (OR), Miller’s calibration slope (MCS) and Continuous Boyce Index (CBI) were derived from the mean values from cross-validation folds. The uncertainty index represents the range in predictions between folds and the area suitable shows the percentage of Europe predicted as suitable for the species using binary maps made using a threshold optimised for TSS. Note that the y-axes are different in each panel and the data points are jittered around the x-axis to allow for better visualisation