Author(s): Pang Yizhe; Zhang Yongxian; Xue Yuan; Ma Zipu; Xing Xuanwei; Xu Mengzhen
Linked Author(s): Mengzhen Xu
Keywords: Lower Yellow River; Multispectral optical satellite; Water depth retrieval; Machine learning
Abstract: River depth is a crucial boundary for studying river geomorphological evolution and calculating river material flux. Traditionally, river depth is obtained through field measurements. It’s still challenging to use satellite-based methods to obtain and extract river depth. Current studies on water depth retrieval primarily focus on clear and water bodies with large areas, such as lakes, shallow seas, and wide rivers with high visibility. These studies often utilize semi-empirical methods and machine learning methods. However, retrieving water depth in sediment-rich rivers remains difficult. In this study, we utilized Sentinel-2 optical satellite imagery and in-situ data from 73 cross sections in the lower Yellow River to develop water retrieval models using machine learning and semi-empirical methods. We investigated the correlation between blue, green, and red band reflectance and water depth at each cross section. The results demonstrated that machine learning methods exhibit high adaptability and accuracy in water depth retrieval for multiple cross sections, even where river sediment concentration varies significantly. In contrast, semi-empirical methods do not perform as well under these conditions. The reflectance and water depth exhibited an exponential decay relationship at certain cross sections, although this correlation displayed significant spatial variability. This study enhances understanding of the correlation between reflectance in the blue, green, and red bands and water depth in the lower Yellow River, thereby providing valuable insights for future remote sensing-based river monitoring and management.
Year: 2024