Scholarlink

<p>The study of sediment transport in rivers is critical for efficient river management, engineering, and environmental conservation. Neglecting this issue can lead to reservoir sedimentation and ecological degradation. This study focuses on estimating the suspended sediment load (SSL) of the Thoubal River, Manipur, India—a data-scarce and sediment-rich basin—using three machine learning (ML) algorithms: Decision Tree, Random Forest, and Extreme Gradient Boost. A total of 534 datasets, including discharge, suspended sediment concentration, flow area, velocity, flow depth, and flow width, recorded at nine stations between 2021 and 2025, were used as inputs. Beyond prediction, the novelty of this work lies in its comprehensive framework that combines model comparison, uncertainty analysis, and interpretability. First, the models were evaluated using statistical indices (coefficient of determination (R<sup>2</sup>), root mean square error (RMSE), overall index of model performance, and ratio of RMSE to standard deviation), where the Random Forest achieved the best performance (<i>R</i><sup>2</sup> = 0.98). Second, Monte Carlo Simulation was applied to verify prediction uncertainty, demonstrating the superior robustness of Random Forest. Finally, two feature analysis methods—feature importance and SHapley Additive exPlanations (SHAP)—were used to identify the relative contribution of hydrological variables, showing that discharge and sediment concentration are the dominant predictors. This integrative approach provides not only accurate predictions but also enhanced interpretability, which is critical for guiding monitoring strategies in data-limited regions. The findings highlight Random Forest as a reliable tool for SSL prediction and suggest its broader application in river basin management, sediment control, and disaster risk reduction.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predicting river system dynamics using machine learning: a multivariable approach to assess the interactions and impacts of hydrological variables on suspended sediment transport

  • 期刊论文

  • 发表日期：

    2025年12月22日

查看全文
  • Mukesh Kumar Yadav,
  • Ngangbam Romeji

摘要

The study of sediment transport in rivers is critical for efficient river management, engineering, and environmental conservation. Neglecting this issue can lead to reservoir sedimentation and ecological degradation. This study focuses on estimating the suspended sediment load (SSL) of the Thoubal River, Manipur, India—a data-scarce and sediment-rich basin—using three machine learning (ML) algorithms: Decision Tree, Random Forest, and Extreme Gradient Boost. A total of 534 datasets, including discharge, suspended sediment concentration, flow area, velocity, flow depth, and flow width, recorded at nine stations between 2021 and 2025, were used as inputs. Beyond prediction, the novelty of this work lies in its comprehensive framework that combines model comparison, uncertainty analysis, and interpretability. First, the models were evaluated using statistical indices (coefficient of determination (R2), root mean square error (RMSE), overall index of model performance, and ratio of RMSE to standard deviation), where the Random Forest achieved the best performance (R2 = 0.98). Second, Monte Carlo Simulation was applied to verify prediction uncertainty, demonstrating the superior robustness of Random Forest. Finally, two feature analysis methods—feature importance and SHapley Additive exPlanations (SHAP)—were used to identify the relative contribution of hydrological variables, showing that discharge and sediment concentration are the dominant predictors. This integrative approach provides not only accurate predictions but also enhanced interpretability, which is critical for guiding monitoring strategies in data-limited regions. The findings highlight Random Forest as a reliable tool for SSL prediction and suggest its broader application in river basin management, sediment control, and disaster risk reduction.