Enhancing daily peak water level forecasts in tidal-dominated regions using machine learning and future tide integration

Tidal-dominated coastal urban areas, such as the Dong Nai-Sai Gon River Basin, face escalating flood risks from heavy rainfall, reservoir releases, and tides. Accurate daily peak water level (Z_max) forecasting is vital for flood and traffic management. Conventional harmonic analysis and standard machine learning (ML) methods often overlook future tidal data, limiting predictive accuracy. This study proposes a hybrid framework (M3) integrating 24 h forecasted tides, derived via Fourier analysis and a multilayer perceptron, with hydro-meteorological predictors using ensemble ML. Using data from four stations between 2019 and 2022, the optimal M3-cbr model achieved strong performance, yielding RMSE values from 0.0581 to 0.1103 m, KGE from 0.8499 to 0.9414, WI from 0.9509 to 0.9719, and |PBIAS| < 2.4325%. It improved accuracy by 9% to 43% over M2 and 45% to 86% over M1 during flood seasons. This cost-effective framework supports operational early warning and sustainable flood management.