Extended Data Fig. 1.

Framework for model generation. Step one involves the generation of an artificial simulated threat likelihood surface. In step 2 real range-maps are overlaid this surface and for each range, a category of either “affected” or “not affected” is assigned based on the values of the threat surface. Several different approaches based on median, mean, max was tested. In step three the ranges with the binary threat categorization are modelled using a series of probabilistic models. In step 4 the models from step 3 are correlated with the original simulated threat surface to identify the modelling approach that best re-produced the simulated surface. Once the best model is identified we progressed to validate it against two independent datasets.