Development of a Hybrid Model for Rainfall Forecasting in Northeastern Thailand: Integration of Seasonal Autoregressive Integrated Moving Average, Support Vector Regression, And Variational Mode Decomposition

This study presents a hybrid model integrating Seasonal Autoregressive Integrated Moving Average (SARIMA), Ensemble Variational Mode Decomposition (EVMD), and Support Vector Regression (SVR) to improve monthly rainfall forecasting in Northeastern Thailand. The approach addresses the challenges posed by the non-stationary and nonlinear nature of rainfall data. SARIMA is first applied to extract linear components, while EVMD is used to decompose residuals into Intrinsic Mode Functions (IMFs). Each IMF and the remaining residuals are forecasted using SVR. A dataset comprising 496 months of rainfall records (January 1983 to April 2024) from 12 meteorological stations under the Thai Meteorological Department was used. Model performance was evaluated using five statistical metrics: RMSE, PRMSE, RPD, MAE, and R². The hybrid SARIMA-EVMD-SVR model consistently outperformed SARIMA and SVR standalone models, achieving R² values above 0.84 and RPD values greater than 2.5 in most stations. The hybrid model improved forecasting accuracy by up to 39.26% over SVR and 36.11% over SARIMA. The results highlight the model’s ability to effectively capture complex rainfall dynamics. Its adaptability offers potential for application in other time series forecasting tasks, contributing to enhanced decision-making in water resource planning and climate-related policy development.