Temporal Association of Climatic Factors with Dengue Incidence and Forecasting Using SARIMAX Model: A Time Series Study in Karnataka, India

Abstract

Background: Dengue are Aedes-transmitted arboviral diseases whose transmission dynamics are strongly modulated by climatic conditions. Karnataka in India, experiences recurrent, seasonally variable dengue but region-specific forecasting models that incorporate climatic predictors remain limited.

Objectives: (1) To describe the temporal distribution of monthly dengue cases and climatic variables in Karnataka, India; (2) To develop and evaluate a SARIMAX model incorporating climatic variables for forecasting monthly dengue cases in Karnataka, India

Methods: In this retrospective time-series study, monthly dengue case counts from September 2021 to December 2025 were obtained from the Commissionerate of Health & Family Welfare Services, Government of Karnataka and matched with monthly climatic data like minimum, average and maximum temperature, relative humidity, rainfall, wind speed, and solar radiation. Series were log-transformed and rendered stationary by first-order non-seasonal and seasonal differencing. Cross-correlation function (CCF) analysis identified biologically plausible lagged climatic predictors, which were incorporated as exogenous variables in candidate SARIMAX models compared using stationary R², RMSE, MAPE, normalized BIC, and the Ljung-Box test, in IBM SPSS Statistics 2025, with a significance level of p < 0.05.

Results: Monthly dengue cases averaged 856.5 (SD 1029.0; range 8-5738), with a marked outbreak in 2023. CCF analysis identified significant positive lagged associations for rainfall (lag 7), wind speed (lag 7), solar radiation (lag 6) and humidity (lag 3) and a significant negative association for minimum temperature (lag 2). SARIMAX(0,1,1)(0,1,1)₁₂ was selected as the best-fitting model (stationary R² = 0.218, RMSE = 3022.6, normalized BIC = 16.894, Ljung-Box Q = 12.98, df = 16, p = 0.675), although none of the individual exogenous predictors reached conventional statistical significance (all p > 0.05), with minimum temperature at lag 2 showing the strongest, near-significant association (p = 0.098). Twelve-month forecasts indicated low dengue activity from January-April 2026 followed by a sharp monsoon-associated rise, peaking in July-August 2026, with wide prediction intervals during the peak months.

Conclusion: A SARIMAX (0,1,1) (0,1,1)₁₂ model incorporating lagged climatic predictors adequately captured  the seasonal, monsoon-linked pattern of dengue transmission in Karnataka. Therefore, the model may be more useful for identifying periods of potentially increased dengue risk than for predicting the precise number of dengue cases in 2026, despite limited statistical power to confirm individual climatic effects. These findings support the potential value of climate-informed early-warning systems for dengue control in Karnataka, while highlighting the need for longer surveillance series to strengthen predictor level inference.