Use of time-series forecasting of dengue incidence based on meteorological predictors in Western Nepal

Dengue is an emerging public health concern in Nepal, with rapidly increasing incidence and recurrent outbreaks, particularly in Gandaki Province. Climatic factors play a crucial role in dengue transmission dynamics, yet high-resolution time-series analyses integrating both statistical and machine-learning approaches remain limited in this setting. We conducted a time-series study using weekly dengue incidence data from January 2020 to December 2025 in Gandaki Province, Nepal. Meteorological variables (temperature, precipitation, humidity, and wind speed) were obtained from the ERA5-Land dataset. Seasonal-trend decomposition, correlation, cross-correlation, and negative binomial regression analyses were performed to assess climatic associations with lag effects. Forecasting models including Seasonal Autoregressive Integrated Moving Average (SARIMA), SARIMA with exogenous variables (SARIMAX), and Extreme Gradient Boosting (XGBoost) were developed and compared using rolling-origin cross-validation based on MAE and RMSE. A total of 34,364 dengue cases were reported during the study period, with pronounced post-monsoon peaks in September–October. Seasonal decomposition showed that annual seasonality accounted for a substantial proportion of the temporal variability in dengue incidence. Temperature was the strongest and most consistent climatic predictor, remaining significantly associated with dengue incidence in negative binomial regression across contemporaneous and 1–2 month lag models (IRR: 1.145–1.187, all p < 0.05). Structural break analysis identified a significant shift in transmission dynamics around 2024–2025. Among the forecasting models, SARIMAX demonstrated superior out-of-sample predictive performance (MAE: 240.35; RMSE: 397.51) compared with SARIMA, while XGBoost achieved comparable average accuracy but showed greater variability and reduced stability during periods of structural change. Forecasts indicate continued seasonal outbreaks with moderate epidemic activity through 2027. Dengue transmission in Gandaki Province is strongly seasonal and climate-sensitive, with temperature emerging as the most robust climatic predictor after accounting for overdispersion. SARIMAX models incorporating meteorological variables provided the best out-of-sample forecasting performance and offer an interpretable framework for climate-informed dengue early warning systems. Integrating such forecasting approaches into routine surveillance may enhance preparedness and support timely public health interventions in Nepal.