Leptospirosis is a major environmental health concern in Thailand, where transmission is strongly influenced by hydrometeorological conditions such as rainfall, surface water, and temperature. This study developed machine-learning models for one-month-ahead leptospirosis forecasting using lagged environmental predictors across five endemic provinces from 2017 to 2023. Monthly case data from the Department of Disease Control were integrated with meteorological variables from the Thai Meteorological Department and Sentinel-2-derived NDVI and MNDWI. Spearman correlation analysis revealed delayed environmental effects, with optimal lags of 2-3 months for MNDWI (ρ = 0.48), rainfall (ρ = 0.41), and temperature (ρ = 0.36), while humidity showed a shorter one-month lag (ρ = 0.43). Among XGBoost, Gradient Boosting (GBT), and Random Forest, GBT achieved the best performance (MAE = 6.83, R2 = 0.617). Feature-importance analysis indicated that recent incidence history dominated predictions (2-month rolling mean = 55.6%), with rainfall at 3-month lag (5.1%) and current rainfall (4.5%) being the strongest environmental predictors. Forecast skill varied across provinces, with highest accuracy in Si Sa Ket (R2 = 0.681) and weakest in Songkhla (R2 = 0.056), highlighting spatial heterogeneity in environmental and socio-behavioral transmission drivers. These findings highlight the utility of environmental monitoring for early-warning systems and demonstrate the feasibility of climate-informed forecasting for leptospirosis in tropical endemic settings.
Kokkaew et al. (Mon,) studied this question.