A cross-scale predictive framework for suspended sediment concentration in mountain rivers integrating multivariate empirical mode decomposition and machine learning
Year: 2026
Type: article
Abstract: Suspended sediment concentration (SSC) in mountain rivers is jointly influenced by hydrological forcing, environmental conditions, and human activities, and exhibits pronounced multiscale variability that complicates daily prediction. This study examined 16 Upper Yangtze mountain rivers spanning 102–105 km2 using daily SSC and hydro-environmental records from 2007–2020. A cross-river framework integrating Multivariate Empirical Mode Decomposition (MEMD), common temporal scales, and machine learning was developed to characterize multiscale SSC dynamics and improve prediction. SSC showed the greatest daily variability, with a median coefficient of variation of 3.88 and a range of 2.89–7.99. MEMD components were grouped into ten common scales, with SSC-related information concentrated mainly within C01–C07 (4.57–374.26 days). Across these scales, runoff and precipitation showed the most widespread associations with SSC, while all five dynamic environmental variables were significantly correlated with SSC across all rivers at C07. Reconstruction using C01–C07 improved prediction over raw inputs, particularly for nonlinear models: test NSE/KGE increased from 0.535/0.543 to 0.643/0.641 for random forest and from 0.586/0.579 to 0.654/0.641 for XGBoost. SHAP analysis identified runoff-related variables as the dominant shared predictive signal, contributing 50.8% and 48.0% in RF and XGB, respectively, whereas secondary controls from precipitation, topography, temperature, reservoir influence, and drainage characteristics varied among rivers. Overall, the framework identifies transferable temporal information across heterogeneous catchments and reveals a hierarchical SSC control structure combining common hydrological forcing with river-specific modifiers.
Source: Figshare
Authors: Yuning Tan, Ying Chen, Wenjie Li, Hongbo Du, Yu Wan
Language: English
Subfield: Environmental Engineering
Field: Environmental Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225752149
FWCI:
Cites:
Cited by:
Related to: