Exploring suspended sediment concentration determinants in mountain rivers using machine learning and structural equation modeling
Year: 2026
Type: article
Abstract: Study region The lower Jinsha River Basin, China, a mountainous basin with high sediment yield. Study focus Flood sediment transport in mountain rivers reflects interactions among hydrological and physiographic factors. This study combines flood-event identification, sediment rating curves, and hysteresis analysis to characterize temporal shifts in sediment transport, and then integrates data-driven models and pathway analysis to identify and quantify the main controls on daily suspended sediment concentration. New hydrological insights for the region Using 60,254 daily records from ten rivers (10²–10⁴ km²), we found lower sediment concentrations at comparable discharges in 2007–2020 than in 1970–1987, while clockwise discharge–sediment hysteresis persisted across the studied rivers. Among five data-driven models, the Artificial Neural Network (ANN) performed best overall, achieving NSE values of 0.743 and 0.714 in the two periods. SHAP analysis revealed interactions between current hydrological forcing and antecedent sediment conditions. PLS-SEM estimated positive direct effects of current discharge and antecedent sediment concentration on sediment concentration ( β = 0.794–1.030 and 0.680–0.695, respectively). Antecedent discharge showed a positive indirect effect through current discharge ( β = 0.719–0.952) and a negative direct effect ( β = −1.031 to −0.831). These findings indicate that daily sediment dynamics in the lower Jinsha River reflect both current discharge and antecedent sediment conditions, supporting their joint consideration in regional sediment monitoring and flood-related sediment assessment.
Authors: Yuning Tan, Ying Chen, Chongming Yu, Wenjie Li
Language: English
Subfield: Soil Science
Domain: Life Sciences
Open Access status: gold
Estimated APC paid: $2,600
Award: G4941854304
OpenAlex ID: W7226005657
