Comparative assessment of HEC-HMS_ann and HEC-HMS_XGboost model performance in simulating future streamflow in mountainous and low-lying catchments

Abstract

Reliable streamflow prediction is critical for effective water resource management, particularly in data-scarce regions where traditional hydrological models like HEC-HMS often underperform due to structural uncertainties and limited data. This study addresses this challenge by developing and comparing hybrid models that integrate HEC-HMS with machine learning techniques Artificial Neural Networks (ANN) and Extreme Gradient Boosting (XGBoost) to enhance streamflow simulation accuracy in contrasting terrains. The Manafwa (mountainous) and Sezibwa (low-lying) catchments in Uganda served as case studies, utilizing historical hydrological data (2010–2018) for calibration and validation. The HEC-HMS_ANN and HEC-HMS_XGBoost models were evaluated against standalone HECHMS, ANN, and XGBoost models using Nash-Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), and coefficient of determination (R²). Results demonstrated that hybrid models significantly outperformed individual approaches, with HEC-HMS_XGBoost achieving the highest accuracy (NSE: 0.9341 calibration, 0.9191 validation for Manafwa; NSE: 0.9513 calibration, 0.937 validation for Sezibwa). The XGBoost-based hybrid excelled in capturing both peak and low flows, while the ANN hybrid showed limitations in low-flow estimation. Topographical differences influenced performance, with the low-lying Sezibwa catchment exhibiting marginally better metrics than the mountainous Manafwa. This research underscores the superiority of hybrid models, particularly HEC-HMS_XGBoost, in overcoming data scarcity and hydrological complexity. The findings provide actionable insights for policymakers and water managers to adopt context-specific models, enhancing flood preparedness and infrastructure resilience in diverse catchments. The study advances hydrological modeling in data-deficient regions, aligning with global sustainable water management goals.