Multi-step-ahead prediction of water levels using machine learning: A comparative analysis in the Vietnamese Mekong Delta

Abstract

This study evaluates the efficacy of five machine learning algorithms Support Vector Regression (SVR), Decision Tree (DT), Random Forest (RF), Light Gradient Boosting Machine Regressor (LGBM), and Linear Regression (LR) in predicting water levels in the Vietnamese Mekong Delta's tidal river system, a complex nonlinear hydrological phenomenon. Using daily maximum, minimum, and mean water level data from the Cao Lanh gauging station on the Tien River (2000-2020), models were developed to forecast water levels one, three, five, and seven days in advance. Performance was assessed using Nash-Sutcliffe Efficiency, coefficient of determination, Root Mean Square Error, and Mean Absolute Error. Results indicate that all models performed well, with SVR consistently outperforming others, followed by RF, DT, and LGBM. The study demonstrates the viability of machine learning in water level prediction using solely historical water level data, potentially enhancing flood warning systems, water resource management, and agricultural planning. These findings contribute to the growing knowledge of machine learning applications in hydrology and can inform sustainable water resource management strategies in delta regions.