Forecasting Rainfall in the Ping, Wang, Yom, and Nan River Basins of Thailand using Decomposition and Holt-Winters Methods Enhanced by GRG Nonlinear Optimization

Abstract

The Ping, Wang, Yom, and Nan rivers in Northern Thailand are vital for national sustainability, serving as key sources of energy and public utilities. This study utilizes secondary data from monitoring stations operated by the Upper Northern Region Irrigation Hydrology Center under the Royal Irrigation Department (RID), Thailand. The primary objective is to develop optimal parameter estimation methods for rainfall forecasting in the upper Ping, Wang, Yom, and Nan river basins. Time series forecasting techniques examined include decomposition with the Whale Optimization Algorithm (WOA-D), Holt-Winters smoothing enhanced by WOA (WOA-HW), and Generalized Reduced Gradient (GRG) nonlinear optimization methods (GRG-D and GRG-HW). These are compared with traditional decomposition and Holt-Winters models developed using Minitab (Minitab-D) and Excel (ForecastSheet-HW). The secondary objective is to forecast rainfall over a 24-month horizon to support trend analysis and improve water resource management. Seasonal ARIMA (SARIMA) is also employed for comparative analysis. Results show that traditional tools like Minitab-D and ForecastSheet-HW are accessible and effective in minimizing forecasting errors. Metaheuristic models such as WOA offer improved accuracy but require programming expertise. GRG solvers provide a practical balance, offering near-comparable accuracy without the need for coding. GRG-D and GRG-HW produced forecasts closely matching actual rainfall across all basins. GRG-HW and WOA-HW achieved the lowest sMAPE values for the Yom and Nan Rivers, GRG-D outperformed Minitab-D for the Wang River, and ForecastSheet-HW remained most effective for the Ping River.