Abstract
BACKGROUND: Seasonal surges in malaria place significant strain on health systems, resulting in frequent shortages of diagnostic tests and antimalarial drugs. Reliable forecasts of outbreaks can help health authorities plan resource allocation more efficiently. This study aimed to develop and validate a predictive model of malaria outbreaks in a low-transmission setting in southern Zambia using remote sensing data on temperature and rainfall. METHODS: Weekly health facility malaria case data were obtained over fifteen years for a sub-district catchment area in Choma District, Zambia. Remotely sensed rainfall estimates were derived from Climate Hazards Group InfraRed Precipitation with Station data. Land surface temperature was obtained from the MODerate-resolution Imaging Spectroradiometer on NASA's Terra satellite. Lagged correlation analysis guided selection of interval lags of temperature and rainfall with the highest predictive value. A negative binomial regression model was then trained on 2010-2016 data to predict total cases in a malaria season, with validation from 2017 to 2024. An incremental training approach was used to investigate how adding new data affected model performance. RESULTS: The final predictive model identified mean nighttime temperature during November-January and mean daily rainfall in December as optimal interval lags for forecasting seasonal malaria incidence. Specifically, malaria cases increased significantly with higher nighttime temperatures (β = 2.1, 95% CI: 0.7-3.5, P = 0.003) and greater precipitation (β = 23.5, 95% CI: -0.7 to 47.8, P = 0.057). A negative interaction term (β = -1.2, 95% CI: -2.5 to 0.01, P = 0.052) indicated that the highest malaria burden occurred during periods characterized by higher temperatures and lower precipitation. The model reproduced the 2020 malaria outbreak within 4% of observed case numbers (2367 observed vs. 2467 mean predicted cases, 95% prediction interval: 1105-3828) based on weather conditions available four months before the seasonal peak. While model predictions were reliable in most years, overestimation occurred in two transmission seasons, suggesting important influences from unmeasured ecological or programmatic factors in these periods. CONCLUSIONS: This simple, weather-driven forecasting approach accurately predicted malaria outbreaks up to four months ahead in a low-transmission setting. Such predictions could inform targeted stock management and preemptive resource mobilization, thereby reducing the risk of commodity shortages during outbreak periods.