Author(s): Siyoon Kwon; Il Won Seo

Linked Author(s): Siyoon Kwon, Il Won Seo

Keywords: Remote sensing; Suspended sediment; Hyperspectral image; OBRA; RFE

Abstract: Suspended sediment, one of the crucial physical factors occurring in natural rivers, changes the river morphology, and has an enormous influence on water quality management. However, the measurement data of suspended sediment is insufficient since the measurement of the Suspended Sediment Concentration (SSC) mainly depends on the direct measurement method using a conventional sampler, which requires a lot of cost and manpower. Thus, in this study, a hyperspectral image-based SSC estimation model was developed through a UAV-based hyperspectral measurement system to perform a remote measurement that overcomes the limitations of the existing point-based measurement. In the previous studies, most remote sensing-based empirical models for estimation of SSC were site-specific since the effect of the sediment properties such as grain size and types of minerals were not considered. Therefore, in order to develop a global SSC estimation model that takes into account these effects, an intrinsic hyperspectral dataset of SSC for each sediment characteristic was constructed through a laboratory experiment and a field-scale experiment. Besides, the optimal spectral bands for SSC estimation were selected according to the characteristics of suspended sediment by using Optimal Band Ratio Analysis (OBRA) and Recursive Feature Elimination (RFE), a feature selection algorithm based on machine learning. Finally, SSC measurement models were developed by training a Support Vector Regression (SVR) with the selected optimal spectral band. The performances of the developed SSC estimation models were assessed according to sediment properties at the Hwang river in South Korea. The results showed that consideration of sediment types and grain size in developing a remote sensed SSC estimation model from hyperspectral imagery would show better performance than previous models with more accurate estimations of SSC in rivers.

DOI: https://doi.org/10.3850/IAHR-39WC252171192022415

Year: 2022