Fig. 2: Summary of modelling variable importance, performance and estimated suitability.

A Variable importance across model runs. Points are the median and bars the 95 percentile. B Partial plots of estimated suitability versus the top five most important variables, presented as aggregates across continental Portugal, Spain and Italy. Horizontal dashed lines at suitability 0.5 show the threshold for model decision on suitable versus unsuitable. C On the left, model performance as measured by accuracy for each model across model runs. On the right, model performance as measured by true positive rate (TPR) and true negative rate (TNR) for each model across model runs. Models are represented by symbols as indicated in the legend to the right. Points are the median and bars the 95 percentile measured across data points (including all months). D Maps presenting the distribution of WNV occurrences used to train the models, and estimated suitability in September and median across time. E,F,G Temporal dynamics of estimated suitability for each of the three countries used for training, together with the proportion of case reports per month within each country (orange). Only months with reported occurrences are visually presented (orange), the other months have zero occurrences. For (C,E–G), the shaded bars per month present the 25, 50, 75 and 95 percentiles (from darker to lighter), and points and lines present medians.