A Comparative Multi-Model Analysis for Predicting Wesselsbron Virus Suitability in sub-Saharan Africa

Wesselsbron virus (WSLV) is an understudied mosquito-borne zoonotic pathogen with important veterinary and public health implications in sub-Saharan Africa: it causes abortions and congenital malformations in livestock and can infect humans, resulting in febrile illness. To better characterize its ecological suitability and guide surveillance, we developed a comparative species distribution modelling framework using multiple machine-learning algorithms and ensemble approaches. Using occurrence records and a broad set of climatic, environmental, and anthropogenic predictors, we evaluated the performance of Random Forest, XGBoost, Neural Network, and ensemble models, including Rank Averaging and Stacking. The best-performing models achieved high discriminatory ability, with ensemble approaches providing the most robust and generalizable predictions. Variable importance analyses indicated that WSLV suitability was strongly associated with climatic variability, precipitation-related conditions, altitude, human population density, and livestock density. Spatial predictions identified major areas of suitability across southern, eastern, and western Africa, with uncertainty concentrated in transitional ecological zones where further field data are required. These findings are broadly consistent with previous Maxent-based modelling work, which also highlighted high ecological suitability in equatorial and southern Africa, while suggesting that model choice may influence the spatial extent of predicted risk. Overall, this study demonstrates that WSLV distribution is shaped by interacting climatic and anthropogenic drivers and that ensemble machine-learning frameworks can improve risk estimation for neglected arboviruses with sparse and heterogeneous data. The resulting maps provide a useful basis for One Health surveillance, targeted field validation, and preparedness planning in high-risk livestock-producing regions.