Assessing the performance of HEC-HMS Model vs.machine learning algorithms in predicting river discharge

Abstract

Accurate prediction of river discharge is essential for effective water resources management and flood risk mitigation. This study presents a comparative analysis of the Hydrologic Engineering Center - Hydrologic Modeling System (HEC-HMS), a process-based hydrological model, and several data-driven machine learning algorithms, including Random Forest, Artificial Neural Networks, Long Short-Term Memory networks, and XGBoost, in streamflow estimation. The models were tested in two contrasting river basins in Sri Lanka: the Kelani River basin in the wet zone and the Malwathu Oya basin in the dry zone. Daily precipitation and discharge data were utilized for model calibration and validation. Model performance was evaluated using multiple statistical metrics, including Nash-Sutcliffe Efficiency, Root Mean Square Error, Coefficient of Determination, Percent Bias, and Kling-Gupta Efficiency. The study compared HECHMS and machine learning models for streamflow prediction in two Sri Lankan basins. In the Kelani Basin (wet zone), LSTM outperformed others (NSE = 0.89, KGE = 0.90), while HEC-HMS showed moderate accuracy (NSE = 0.74). In the Malwathu Oya Basin (dry zone), LSTM excelled (NSE = 0.80), but HEC-HMS struggled (NSE = 0.49, PBIAS = 15.82). ML models, especially LSTM, adapted better to diverse flow regimes, highlighting their potential to complement traditional hydrological models.