Author(s): Binbin Wang; Ruichen Xu; Qingqing Sun; Baolin Deng

Linked Author(s): Binbin Wang

Keywords: Freshwater mussel; Turbulence; Stream; Sediment; Lagrangian particle tracking; Machine learning

Abstract: Sediment spills from construction and engineering activities can rapidly introduce large sediment loads into streams and rivers, posing risks to freshwater mussels through burial and habitat degradation. To evaluate downstream zones of potential mussel exposure, we examined 49 stream reaches across Missouri where field measurements and gaging data were available. Empirical hydraulic relationships were established for three representative flow classes, which were then used to drive a process-based Lagrangian particle tracking model. The model simulated sediment transport and deposition under a wide range of hydraulic conditions (velocities: 0.124–2.191 m/s; depths: 1–12 m) and sediment sizes (0.01–20 mm in diameter), resulting in more than 40,000 simulation scenarios. Model outputs showed that particles exceeding critical diameters (0.041 –1.975 mm, depending on flow conditions) settled in downstream habitats, with median settling distances ranging from 0 to 40 m and variances up to 110 m². Broader distribution s of settling locations extended from tens to hundreds of meters downstream, with mode, median, and mean values generally falling within 10–200 m based on particle counts, and 1–100 m when weighted by sediment mass. To extend predictive capability, nine machine learning (ML) models were trained on the simulation data to predict probability distributions of sediment deposition. Three models—Random Forest, Extreme Gradient Boosting, and Gradient Boosting Trees— performed best (R² > 0.99) across tested conditions, with water depth emerging as the most influential predictor.

DOI: https://doi.org/10.64697/iahr.proc.ise2026.abs.106

Year: 2026