Machine learning-driven SSC retrieval in the high-turbidity lower Yellow River leveraging a novel red-edge and narrow NIR band synergy

Spatiotemporal dynamics of suspended sediment concentration (SSC) in the lower Yellow River, China (LYR), are critical for understanding channel evolution and optimizing reservoir operations but remain poorly quantified at high resolution. This study combined Sentinel-2 imagery and daily SSC data from seven hydrological stations to develop a retrieval model integrating machine learning feature selection with empirical regression. The proposed normalized difference index (B8A – B7)/(B8A + B7) achieved high retrieval accuracy ( R 2 = 0.82) and was applied to reconstruct SSC variations during 2019–2024. Daily SSC during this period ranged from approximately 0 to 30 kg/m 3 , with daily discharge varying between approximately 100 and 5200 m 3 /s, representing typical conditions of a highly turbid, flow-regulated river. The results reveal a distinct spatiotemporal regime: high-concentration events predominantly occur in summer, synchronized with flood seasons and reservoir discharges, while a clear erosion–deposition transition exists approximately 20–30 km downstream of Luokou station, a representative hydrological station located in the downstream channel of the LYR. This spatial pattern indicates the limited efficacy of current regulation strategies in maintaining full-stream scour. The study provides a novel monitoring framework for sediment transport in highly regulated rivers and highlights the need for adaptive management strategies to mitigate discontinuous scour and deposition.