Ensemble Machine Learning for Water Quality Index Prediction in the Johor River Basin: A Comparative Study with HydroMT-based Hydrodynamic Modelling

Timely and cost-effective assessment of surface water quality is essential for effective management of tropical river systems. Conventional Water Quality Index (WQI) methods, though widely adopted, require multiple laboratory tests and prolonged sampling, which present significant challenges in regions with limited data availability. This study presents an ensemble machine learning (ML) framework for predicting WQI in the Johor River, Peninsular Malaysia, providing a faster and more affordable alternative that enables reliable WQI prediction using fewer inputs and reduced costs. A decade-long monitoring dataset comprising 1,637 samples from 44 Department of Environment stations was utilised. The analysis employed a multistage workflow, including calculation of numerical WQI based on Malaysia's National Water Quality Standards, temporal trend analysis across upstream, midstream, and downstream sections, Pearson correlation analysis, and feature importance evaluation. Two optimised ensemble algorithms, Gradient Boosting Regression (GB) and Random Forest Regression (RF), were developed with hyperparameters tuned via randomised cross-validation. Temporal analysis identified magnesium, Escherichia coli, suspended solids, and dissolved solids as primary contributors to water quality decline. Feature importance analysis demonstrated that biochemical oxygen demand (BOD), chemical oxygen demand (COD), and dissolved oxygen percentage (DO%) alone provided strong predictive power, enabling an efficient three-parameter input set. Using only these variables, the RF and GB models achieved R² values of 0.86 and 0.85, respectively, with the GB ensemble accurately categorising water quality in over 96% of validation samples. The predictive performance of the ML framework was benchmarked against HydroMT, an open-source Python toolkit for process-based hydrodynamic modelling using global datasets. The ensemble ML approach demonstrated comparable predictive accuracy with significantly reduced cost and complexity. Future research will integrate the ML framework with HydroMT-driven hydrodynamic simulations under projected climate scenarios, establishing a consistent and efficient decision-support system for managing tropical river basins while accounting for climate variability.