Abstract

Sediment transport in Himalayan rivers is highly dynamic, driven by intense monsoon rainfall, steep topography, and fragile geology, posing challenges for water resource management and infrastructure sustainability. This study develops season-specific sediment rating curves (SRCs) for Station 120 in the Mahakali River Basin, Nepal, using machine learning (ML) models to improve sediment load estimation under varying hydrological conditions. Daily discharge and suspended sediment data from 2007 to 2014 were analyzed across four seasons Pre-Monsoon, Monsoon, Post-Monsoon, and Winter accounting for seasonal variability in sediment transport dynamics. Three ML models K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Random Forest (RF) were evaluated, and the best-performing model for each season was selected based on R² and Mean Absolute Percentage Error (MAPE). SVM outperformed others in Pre-Monsoon, Monsoon, and Winter seasons, while RF showed superior accuracy in Post- Monsoon. Power-law SRCs were derived from predicted sediment concentrations, yielding equations: S=4.28×Q1.16 (Monsoon), S=1.21×Q1.19 (Post-Monsoon), S=3.81×Q1.17 (Pre-Monsoon), and S=826.88×Q −0.71 (Winter). Despite improved accuracy, higher MAPE during the Monsoon season highlights the limitations of ML models in capturing extreme events. The findings support the need for advanced deep learning approaches, as suggested by prior studies, to better represent non-linear and time-dependent sediment processes. This research provides a robust, seasonally adaptive framework for sediment load estimation in data-scarce Himalayan basins, supporting improved sediment management.