Modeling stage‐discharge and sediment‐discharge relationships in data‐scarce Himalayan River Basin Dhauliganga, Central Himalaya, using neural networks

Abstract

AbstractThis study focuses on the hydro‐sedimentological characterization and modeling of the Dhauliganga River in Uttarakhand, India. Field data collected from 2018–2020, including stage, velocity, and suspended sediment concentration (SSC), showed notable variations influenced by melting snow, glaciers, and precipitation. Challenges in accurately modeling rivers with a topography and sparse gauging stations were addressed using artificial neural networks (ANN). The calibrated models precisely predicted stage‐discharge and sediment‐discharge relationships, demonstrating the effectiveness of machine learning, particularly ANN‐based modeling, in such challenging terrains. The model's performance was assessed using coefficient of determination (R2), root mean square error (RMSE), and mean square error (MSE). During the calibration phase, the model exhibited notable performance with R2 values of 0.96 for discharge and 0.63 for SSC, accompanied by low RMSE values of 5.29 cu m s–1 for discharge and 0.61 g for SSC. Subsequently, in the prediction phase, the model maintained its robustness, achieving R2 values of 0.97 for discharge and 0.63 for SSC, along with RMSE values of 5.67 cu m s–1 for discharge and 0.68 g for SSC. The study also found a strong agreement between water flow estimates derived from traditional methods, ANN, and actual measurements. The suspended sediment load, influenced by both water flow and SSC, varied annually, potentially modifying aquatic habitats through sediment deposition, and altering aquatic communities. These findings offer crucial insights into the hydro‐sedimentological dynamics of the studied river, providing valuable applications for sustainable water‐resource management in challenging terrains and addressing environmental concerns related to sedimentation, water quality, and aquatic ecosystem.