Resumen
The spatio-temporal dynamism of sediment discharge (Qs) in rivers is influenced by various natural and anthropogenic factors. Unfortunately, most rivers are only monitored at a limited number of stations or not gauged at all. Therefore, this study aims to provide a remote-sensing-based alternative for Qs monitoring. The at-a-station hydraulic geometry (AHG) power?law method was compared to the at-many-stations hydraulic geometry (AMHG) method; in addition, a novel AHG machine-learning (ML) method was introduced to estimate water discharge at three gauging stations in the Tisza (Szeged and Algyo) and Maros (Makó) Rivers in Hungary. The surface reflectance of Sentinel-2 images was correlated to in situ suspended sediment concentration (SSC) by support vector machine (SVM), random forest (RF), artificial neural network (ANN), and combined algorithms. The best performing water discharge and SSC models were employed to estimate the Qs. Our novel AHG ML method gave the best estimations of water discharge (Szeged: R2 = 0.87; Algyo: R2 = 0.75; Makó: R2 = 0.61). Furthermore, the RF (R2 = 0.9) and combined models (R2 = 0.82) showed the best SSC estimations for the Maros and Tisza Rivers. The highest Qs were detected during floods; however, there is usually a clockwise hysteresis between the SSC and water discharge, especially in the Tisza River.