Bhosekar, U.; Ventura, P. C.; Hill, M. D.; Kummer, A. G.; Mhade, S.; Chitturi, J.; Vasquez, C.; Mutebi, J.-P.; Townsend, J.; Litvinova, M.; Wilke, A. B. B.; Ajelli, M.

Conventional mosquito surveillance typically relies on contemporaneous data, making it challenging to anticipate future vector surges. To support proactive vector management, this study evaluates a multi-model forecasting framework designed to generate probabilistic 1- to 4-week-ahead forecasts of Aedes aegypti relative abundance per trap night. The framework was validated using multi-year surveillance data across four US jurisdictions spanning varied environments (from subtropical to temperate and arid). We found that an ensemble approach aggregating statistical and machine learning models generally achieved the best performance across all locations and forecast horizons. Relative forecast performance improved as the forecast horizon extended from 1 to 4 weeks ahead. The most challenging data to forecast were primarily restricted to low mosquito activity periods or atypical population peaks with unusual timing or magnitude. While full integration into routine vector management workflows represents a long-term process requiring operational adaptation, this work advances forecasting research and establishes a baseline for translating these approaches into real-time applications for public health authorities, with downstream effects in mitigating the risks of mosquito-borne diseases.