Integrated Satellite-Derived Bathymetry and Morphodynamic Assessment for Regulated River Monitoring Using Machine Learning and Sentinel-2 Data
Ahmed Nour-Eldeen, Rofyda Abdelrehem, Alban Kuriqi, Ismail Abd-Elaty, Hickmat Hossen
This study presents an integrated, data-driven framework for satellite-derived bathymetry and morphodynamic assessment in large, regulated rivers, providing a spatial database to support reach-scale hydromorphological monitoring and river management. Satellite-derived bathymetry (SDB) was developed using 24,768 in situ depth measurements and Sentinel-2 multispectral data to train Random Forest (RF) and Artificial Neural Network (ANN) models. Under turbid water conditions, the Random Forest model outperformed the Artificial Neural Network model in simulating the non-linear relationship between the water spectrum and water depth; the RF model achieved an R2 of 0.828 and an RMSE of 0.93 m, while the ANN model produced an R2 of 0.608 and an RMSE of 1.40 m. Depth-dependent errors were smallest at intermediate depths and larger in shallow and deep water. Morphometric parameters, including the Sinuosity Index (SI) and Braiding Index (BI), were calculated for 2017, 2019, and 2021 using the NDWI-based water mask to define channel boundaries. The reach exhibited moderate sinuosity (SI ≈ 1.16), and an increase in braiding was observed (BI ranging from 1.33 to 1.36). From 2017 to 2019, erosion (3.51 km2) exceeded deposition (1.25 km2). In contrast, the 2019–2021 period showed approximately equal areas of erosion and deposition (1.63 km2 each). The analysis is constrained by a single 2015 calibration survey, the optical penetration limit of Sentinel-2, and the reliance on three morphometric snapshots (2017, 2019, 2021), which may not capture short-term adjustments. The novelty of this study lies in integrating ML-based Sentinel-2 bathymetry with multi-temporal morphometric indicators to characterize the vertical and horizontal dynamics of regulated rivers jointly.