Scholarlink

<p>Soil erosion and sediment yield pose significant threats to agricultural sustainability, water security, and ecosystem stability in subtropical landscapes. This study aimed to quantify and spatially predict soil erosion and sediment yield across Hunan Province, China, by integrating field measurements, the Universal Soil Loss Equation (USLE), and advanced digital soil mapping techniques. The central research question was which modeling approach provides the most accurate and reliable spatial predictions to guide soil and water conservation planning.&#xa0;A total of 467 composite surface soil samples (0–30&#xa0;cm) were collected during the dry season using a random sampling design. Laboratory analyses included soil texture, bulk density, soil organic matter content, aggregate stability, and infiltration rate. USLE factors were derived from regional rainfall data, soil properties, slope metrics from digital elevation models, NDVI, and land management practices. Three machine learning algorithms—Random Forest (RF), Artificial Neural Network (ANN), and Support Vector Machine (SVM)—were trained using 80% of the dataset and validated with the remaining 20%. Model performance was evaluated using R², RMSE, and MAE, and uncertainty was assessed using the prediction interval coverage probability (PICP).&#xa0;The RF model consistently outperformed ANN and SVM in predicting both soil erosion (training R² = 0.92; test R² = 0.86) and sediment yield (training R² = 0.89; test R² = 0.80). Spatial predictions revealed severe soil erosion (up to 19.8 t ha⁻¹ yr⁻¹) and sediment yield (up to 1,623 t ha⁻¹ yr⁻¹) in the southern and southeastern mountainous areas, while flatter central regions exhibited lower values. Elevation and slope were identified as the dominant controlling factors, and uncertainty was minimal in stable lowlands and highest in dynamic, high-relief landscapes.&#xa0;These findings demonstrate the robustness and effectiveness of the RF model for spatial environmental modeling and provide high-resolution maps of soil erosion and sediment yield that can support targeted soil and water conservation policies in subtropical China. Future studies should incorporate climate change scenarios and land-use dynamics to evaluate long-term impacts on erosion processes and management effectiveness.</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

Spatial Analysis of Soil Erosion and Sediment Yield in Hunan Province, China, Using Digital Soil Mapping Techniques

  • 期刊论文

  • 发表日期：

    2026年2月19日

查看全文
  • Shihao Zhang

摘要

Soil erosion and sediment yield pose significant threats to agricultural sustainability, water security, and ecosystem stability in subtropical landscapes. This study aimed to quantify and spatially predict soil erosion and sediment yield across Hunan Province, China, by integrating field measurements, the Universal Soil Loss Equation (USLE), and advanced digital soil mapping techniques. The central research question was which modeling approach provides the most accurate and reliable spatial predictions to guide soil and water conservation planning. A total of 467 composite surface soil samples (0–30 cm) were collected during the dry season using a random sampling design. Laboratory analyses included soil texture, bulk density, soil organic matter content, aggregate stability, and infiltration rate. USLE factors were derived from regional rainfall data, soil properties, slope metrics from digital elevation models, NDVI, and land management practices. Three machine learning algorithms—Random Forest (RF), Artificial Neural Network (ANN), and Support Vector Machine (SVM)—were trained using 80% of the dataset and validated with the remaining 20%. Model performance was evaluated using R², RMSE, and MAE, and uncertainty was assessed using the prediction interval coverage probability (PICP). The RF model consistently outperformed ANN and SVM in predicting both soil erosion (training R² = 0.92; test R² = 0.86) and sediment yield (training R² = 0.89; test R² = 0.80). Spatial predictions revealed severe soil erosion (up to 19.8 t ha⁻¹ yr⁻¹) and sediment yield (up to 1,623 t ha⁻¹ yr⁻¹) in the southern and southeastern mountainous areas, while flatter central regions exhibited lower values. Elevation and slope were identified as the dominant controlling factors, and uncertainty was minimal in stable lowlands and highest in dynamic, high-relief landscapes. These findings demonstrate the robustness and effectiveness of the RF model for spatial environmental modeling and provide high-resolution maps of soil erosion and sediment yield that can support targeted soil and water conservation policies in subtropical China. Future studies should incorporate climate change scenarios and land-use dynamics to evaluate long-term impacts on erosion processes and management effectiveness.