Spatiotemporal dynamics and driving mechanisms of suspended sediment concentration in the Yellow River revealed by machine learning and multi-source remote sensing

Dynamic monitoring of suspended sediment concentration (SSC) in the Yellow River is crucial for understanding sediment variability. However, the lack of systematic high-resolution data has hindered long-term SSC dynamics analysis. This study develops an SSC remote sensing inversion model utilizing machine learning and multi-source remote sensing technologies for spatiotemporal monitoring across the main stream and major tributaries of the Yellow River. We inverted SSC changes from 1986 to 2019 in the main stream and five major tributaries in the middle reaches, while examining the driving factors behind SSC variability using structural equation modeling. Results show: (1) The XGBoost-based SSC inversion model performed well, with R 2 of 0.78, RMSE of 1685.74 mg/L, and MAE of 967 mg/L, outperforming other models such as linear regression and random forest. (2) SSC in the Yellow River declined significantly between 1986 and 2019, with an annual average decrease of 75.4 mg/L, totaling a 63.6% reduction. Spatially, 97.39% of monitored river sections showed a decline, with 50% of these passing significance tests ( p < 0.05). (3) Structural equation modeling identified key drivers of SSC changes, with climate and hydrological factors like precipitation as primary influences, along with socio-economic development, ecological restoration, and reservoir sedimentation. The study offers a practical framework for remote sensing-based SSC monitoring in rivers with varying turbidity levels, providing insights for water-sediment resource management and ecological conservation in the Yellow River Basin.