Measurement of suspended sediment concentration at the outlet of the Yellow River Canyon: Using Sentinel-2 images and machine learning

The suspended sediment concentration (SSC) is one of the most critical water quality parameters in rivers. Previous studies mainly focused on low-concentration sediment rivers. Besides, the traditional situ monitoring techniques and low-resolution imagery cannot meet the vastly spatial and temporal coverage. The Yellow River is famous for its high sediment content in the world. This research considers the outlet of the Yellow River Canyon as a case study. Based on the SSC data of Longmen Hydrological Station of the Yellow River provided by the Yellow River Conservancy Commission and the Sentinel-2 images from 2019 to 2020, the optimal band combination was selected to construct a random forest non-parametric regression prediction model, and the quantitative estimation of SSC of rivers with high sediment concentration was realized. The results showed that: (i) The determination coefficient (R 2 ) of the SSC estimation model derived from the random forest model is 0.94, satisfying the requirements of high-precision SSC estimation; (ii) The red band and near-infrared band are important predictors of suspended sediment concentration; (iii) Seasonal differences and spatial variation in sediment in YuMen Kou waters are evident. The study's findings demonstrate that the random forest regression model is superior to traditional modeling methods for predicting SSC in sediment-rich rivers.