Development of a sediment monitoring model for river systems

Abstract

This study presents a comprehensive framework for monitoring and managing sedimentation in River Malaba, a transboundary river shared by Uganda and Kenya within the Malaba subcatchment. The research integrated hydrological modeling, hydraulic simulation and machine

learning

to address challenges related to sediment yield estimation, sediment dynamics and real-time sediment monitoring. The Soil and Water Assessment Tool (SWAT) was used to simulate catchment hydrology and estimate sediment yield. Model calibration and validation were performed using SUFI-2, with a Nash-Sutcliffe Efficiency (NSE) of 0.72 and R² of 0.76 for calibration, and NSE of 0.68 and R² of 0.74 for validation. The simulation revealed an average annual sediment yield of 142,248.7 tons with spatial variability across sub-basins ranging from 1.72 to 86.93 tons/km², highlighting sediment hotspots in the catchment. Hydraulic modeling was carried out using HEC-RAS 6.4.1 to evaluate sediment transport and deposition patterns along a critical reach of the river within Kalait Sub-County. The model achieved an NSE of 0.70 and R² of 0.73 during calibration (1999–2008) and NSE of 0.69 and R² of 0.71 during validation (2009–2022). Key outputs included sediment routing characteristics, trap efficiency estimations and the identification of erosion-prone and deposition-prone reaches along the river corridor. To facilitate real-time sediment prediction, a soft sensor was developed using a Multi-Layer Perceptron (MLP) Artificial Neural Network. Rainfall, temperature and flow velocity were used as input parameters due to their influence on sediment concentration. The model was trained and optimized using grid search and achieved R² scores of 0.90 for the training set, 0.80 for validation, and 0.85 for the testing set, indicating strong generalization and predictive performance. A sediment concentration heatmap was also generated using geospatial coordinates and predicted concentrations. River segments were ranked and categorized into low, moderate and high sediment concentration zones. The heatmap visually identified critical locations that require dredging thus offering actionable insights for maintenance planning. This integrated approach offers a robust decision support tool for sediment monitoring and management in river basins.