Journal Article

Reconstruction of the Vertical Distribution of Suspended Sediment Using Support Vector Machines

Wen, Yuncheng, Yuan, Qiang, Wang, Yuke, Zhang, Fanyi, Lv, Jinyang, Yu, Zhe, Xia, Mingyan, Cheng, Zelin

2026

Overview

Accurately quantifying vertical sediment transport rates in large seaward rivers is vital for estimating basin-scale water and sediment fluxes and assessing riverbed evolution. Traditional multi-point velocity and suspended sediment concentration (SSC) measurements are costly and slow, hindering long-term online monitoring. Bidirectional flows in tidal reaches further exacerbate this challenge. We propose a physics-constrained support vector machine (SVM) inversion method to estimate vertical sediment transport rates from single-point measurements. Constrained by modified logarithmic velocity and Rouse suspended sediment concentration profiles, it quantitatively relates single-point hydraulic variables to key parameters governing vertical distributions. Lower Yangtze River tidal reach field data validate the hybrid model’s successful reconstruction of vertical distributions. It accurately captures transient sediment responses across maximum flood and ebb. Inverted transport rates match measurements closely (RMSE = 0.085, NSE = 0.969, PBIAS = 2.50%) and exhibit strong cross-site generalization. Sensitivity analysis identifies 0.4 times the water depth above the riverbed as the optimal single-point sensor position. Although currently validated only in the lower Yangtze River, this low-cost, reliable method supports local basin management, flood control, and disaster mitigation by enabling continuous sediment flux monitoring. However, applying it to other river or estuarine systems may require recalibration or retraining to adapt to different local conditions.

• Publisher: MDPI AG
• Subject: Accuracy / Algorithms / Artificial intelligence / Basins (Geology) / Boundary conditions / Brackishwater environment / Constraints / Disaster management / Emergency preparedness / Engineering / Estuaries / Flood control / Flood management / Flow velocity / Fluid mechanics / Hydraulics / Hydrology / Image processing / Machine learning / Maximum probable flood / Neural networks / Physics / physics-constrained learning / Position sensing / profile reconstruction / Reconstruction / River beds / Riverbeds / Rivers / Sediment / Sediment concentration / Sediment transport / Sediment, Suspended / Sediments / Sensitivity analysis / Soil erosion / support vector machine (SVM) / Support vector machines / suspended sediment vertical distribution / Suspended sediments / tidal river / Velocity / Vertical distribution / Water / Water depth / Watershed management