Forecasting Culex pipiens/restuans abundance in Ontario, Canada using surveillance and weather data

Abstract Reliable short-term forecasting of vector abundance is critical for guiding localized mosquito control, yet week-ahead prediction at the Public Health Unit (PHU) scale remains challenging. We developed and evaluated a week-ahead forecasting framework for Culex pipiens/restuans abundance using 22 years of routine surveillance and meteorological data from 34 PHUs across Ontario, Canada. Based on a subset of predictors selected through feature-selection techniques, models integrating recent mosquito abundance and lagged weather variables were evaluated with multiple regression-based approaches, including linear, neural network, and random-forest regression, under random, temporal, and spatial validation schemes. Under cross-validation, top-performing models explained approximately 69% of the variance in weekly mosquito abundance. In independent testing on 2022 to 2023 data, the best-performing model achieved a 14.4% reduction in RMSE and an absolute increase in R2 of 0.124 relative to the persistence baseline. Across validation schemes, predictive performance was driven primarily by recent mosquito abundance history, whereas recent weather conditions provided additional gains, with contributions attenuating at longer lags. Increasing predictor complexity yielded diminishing returns for week-ahead forecasting. Consistent performance across validation schemes indicates model robustness. Overall, these results demonstrate a parsimonious approach for routine short-term forecasting at the PHU scale, with potential applications in surveillance, resource allocation, and targeted mosquito control.