Satellite-driven machine learning framework for monitoring river sediment discharge: integrating Sentinel-2 spectral data with hydrometric insights

초록

Accurate prediction of river sediment discharge is crucial for effective water resource management and flood mitigation, especially as direct measurements are challenging. This study integrates spectral bands from Sentinel-2 satellite imagery and total suspended matter (TSM) data, processed using the C2RCC processor, with ground-based hydrometric data to enhance machine learning models for sediment discharge prediction. The Karun River, Iran's most vital and water-rich river, was selected as the study area due to its significant role in regional water supply. In this study, we analyzed several machine learning techniques, including K-nearest neighbors, random forest, gradient boosting, multi-layer perceptron, and extreme gradient boosting. The findings reveal that these models significantly outperformed traditional sediment rating curves, with the Random Forest model achieving an R2 value of 0.92 and low error metrics (RMSE = 454 g/s, MAE = 286 g/s, and MBE = −22 g/s). By integrating TSM data, spectral bands, and hydrometric measurements, the study provides frequent sediment discharge updates every five days, reducing the need for in-situ measurements. A new SRC was developed by combining in-situ river discharge data with sediment discharge estimates from the RF model, improving prediction accuracy and reducing errors compared to traditional methods. This approach enhances SRCs and improves sediment transport analysis, providing more reliable tools for water resource planning and hydraulic structure design. © 2025 COSPAR

키워드