Continuous Measurement of River Sediment Load Using Acoustic Parameters of H-ADCP

The Horizontal Acoustic Doppler Current Profiler (H-ADCP), widely used for real-time discharge monitoring, measures flow velocity through the Doppler shift of acoustic pulses. Furthermore, the attenuation and scattering of acoustic waves in water enable the estimation of suspended sediment concentration (SSC), extending its utility from hydrodynamic to sediment monitoring. In Korea, 66 gauging stations equipped with H-ADCPs are currently in operation. By applying sediment estimation techniques based on acoustic backscatters, these systems are expected to enable the simultaneous measurement of both discharge and sediment load. SSC estimation using H-ADCP is based on the linear relationship between SCB(sediment Corrected Backscatters) and individually sampled SSC. Recent studies have attempted to improve the accuracy of SSC estimation by including additional hydraulic and acoustic parameters. In particular, multiple regression models that include water level along with SCB, as well as the attenuation–backscatter ratio (ABR), which jointly accounts for both attenuation and scattering effects, have demonstrated enhanced predictive capability. This study analyzed data from 5 H-ADCP gauging stations to examine the relationships between sediment-related (sediment attenuation coefficient, SCB and ABR) and hydraulic variables (water level, velocity and discharge) using machine learning, aiming to improve accuracy of SSC estimation. The testbeds were equipped with Channel Master H-ADCPs operating at frequencies of 300, 600, and 1200 kHz, and SSC sampling was obtained during the 2024 flood seasons using the D-74 sampler. Using H-ADCPs data and individually measured SSC from the testbeds, both a simple linear regression between SCB and SSC, and a multiple regression including water level were developed and compared against measured SSC. In addition, SSC estimates derived from machine learning models that integrate both hydraulic and acoustic variables were also evaluated for comparison. Application to the testbeds showed that multiple regression including water level improved accuracy compared with simple SCB–SSC linear regression. Furthermore, when machine learning was applied with optimal variable model using diverse variables, the estimation achieved over 85% accuracy relative to individually measured values.

Keywords: SSC, SCB, ABR, water level, H-ADCP

Acknowledgements

This work was supported by Korea Environment Industry & Technology Institute (KEITI) through Research and development on the technology for securing the water resources stability in response to future change Program, funded by Ministry of Climate, Energy, Environment (MCEE) (RS-2024-00397970).