Scholarlink

<p>Soil erosion remains one of the most critical forms of land degradation globally and its assessment is essential for efforts aimed at preventing and mitigating land degradation. This study assessed soil loss and sediment yield in the upper uThukela Catchment, South Africa and sought to provide spatially explicit data to support erosion control and prevention strategies. A hybrid modelling framework was employed, integrating the Revised Universal Soil Loss Equation (RUSLE) with machine learning (ML) algorithms. Initially, RUSLE was applied to estimate annual soil loss across the catchment. These estimates subsequently served as inputs for predictive ML models to assess spatial patterns of soil loss, sediment transport, and the influence of twelve predictors. Elastic Net, Support Vector Machines (SVM), and Random Forest (RF) were evaluated using 10-fold cross-validation to determine the most effective predictive model. RF demonstrated the best predictive performance, explaining 56% of soil loss variance and producing the lowest prediction errors. The RF model estimated an average annual soil loss of 1.33 t/ha/year, with soil loss values ranging from 0 t/ha/year in low-lying areas to 55 t/ha/year on the footslopes of mountainous areas. Areas under communal rangeland management mostly experienced moderate to high erosion risk, contributing approximately 78% of the total annual soil loss (3.80&#xa0;million tonnes) as sediment delivered to water bodies. Such sedimentation poses a potential threat to water quality and storage capacity in the Woodstock and Spioenkop Dams. Among the predictors, the slope length and steepness (LS) factor was the most significant, with strong interactions with rainfall erosivity and the topographic wetness index. Overall, these findings highlight the importance of considering these topographic and climatic variables when assessing soil loss in similar environments.</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

Integrating an empirical erosion model and machine learning for assessing soil loss and sediment delivery dynamics in a sub-humid catchment

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月30日

查看全文
  • Kwanele Phinzi

摘要

Soil erosion remains one of the most critical forms of land degradation globally and its assessment is essential for efforts aimed at preventing and mitigating land degradation. This study assessed soil loss and sediment yield in the upper uThukela Catchment, South Africa and sought to provide spatially explicit data to support erosion control and prevention strategies. A hybrid modelling framework was employed, integrating the Revised Universal Soil Loss Equation (RUSLE) with machine learning (ML) algorithms. Initially, RUSLE was applied to estimate annual soil loss across the catchment. These estimates subsequently served as inputs for predictive ML models to assess spatial patterns of soil loss, sediment transport, and the influence of twelve predictors. Elastic Net, Support Vector Machines (SVM), and Random Forest (RF) were evaluated using 10-fold cross-validation to determine the most effective predictive model. RF demonstrated the best predictive performance, explaining 56% of soil loss variance and producing the lowest prediction errors. The RF model estimated an average annual soil loss of 1.33 t/ha/year, with soil loss values ranging from 0 t/ha/year in low-lying areas to 55 t/ha/year on the footslopes of mountainous areas. Areas under communal rangeland management mostly experienced moderate to high erosion risk, contributing approximately 78% of the total annual soil loss (3.80 million tonnes) as sediment delivered to water bodies. Such sedimentation poses a potential threat to water quality and storage capacity in the Woodstock and Spioenkop Dams. Among the predictors, the slope length and steepness (LS) factor was the most significant, with strong interactions with rainfall erosivity and the topographic wetness index. Overall, these findings highlight the importance of considering these topographic and climatic variables when assessing soil loss in similar environments.