Scholarlink

<p>Three-dimensional (3D) modeling of soil texture components serves as an essential tool for the sustainable management of soil and agricultural resources. In this study, a 3D modeling framework was applied in Jiangsu Province, China, integrating the equal-area spline function with the Quantile Regression Forest (QRF) algorithm. Soil profile measurements of sand, silt, and clay contents were smoothed using the spline function across five standard depth intervals: 0–25, 25–50, 50–75, 75–100, and 100–125&#xa0;cm. These processed data, together with 16 environmental covariates, were employed as inputs to the QRF model to generate depth-specific predictive maps of soil texture components. The model exhibited strong predictive performance in the surface layer (0–25&#xa0;cm), with coefficients of determination (R2) of 0.78 for clay, 0.76 for sand, and 0.77 for silt. However, accuracy declined with depth, reaching R2 values of 0.55, 0.54, and 0.52 for clay, sand, and silt, respectively, at 100–125&#xa0;cm. Model error, expressed as the root mean square error (RMSE), increased with depth—from 8.1% for surface clay to 13.4% in the deepest layer. Variable importance analysis identified piezometric elevation, precipitation, the Grain Size Index (GSI), the spectral band ratio Band3/Band7, and topographic indices such as the Multiresolution Ridge Top Flatness (MrRTF) as the most influential predictors of soil texture. Furthermore, uncertainty evaluation using the Prediction Interval Coverage Probability (PICP) index revealed decreasing prediction reliability with greater depth. Based on these findings, future research should prioritize the use of higher-resolution elevation datasets, increase sampling density, and incorporate geological and land-use variables to enhance model accuracy and generalizability. This study presents a robust methodology for generating accurate, spatially consistent estimates of soil texture components across defined profile depths and provides a scientific foundation for advancing regional-scale 3D soil mapping.</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

Enhanced three-dimensional mapping of soil texture components using quantile regression forest and spline techniques in Jiangsu Province, China

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月9日

查看全文
  • Zhefeng Lin,
  • Jiangang Xu,
  • Muqiu Lu

摘要

Three-dimensional (3D) modeling of soil texture components serves as an essential tool for the sustainable management of soil and agricultural resources. In this study, a 3D modeling framework was applied in Jiangsu Province, China, integrating the equal-area spline function with the Quantile Regression Forest (QRF) algorithm. Soil profile measurements of sand, silt, and clay contents were smoothed using the spline function across five standard depth intervals: 0–25, 25–50, 50–75, 75–100, and 100–125 cm. These processed data, together with 16 environmental covariates, were employed as inputs to the QRF model to generate depth-specific predictive maps of soil texture components. The model exhibited strong predictive performance in the surface layer (0–25 cm), with coefficients of determination (R2) of 0.78 for clay, 0.76 for sand, and 0.77 for silt. However, accuracy declined with depth, reaching R2 values of 0.55, 0.54, and 0.52 for clay, sand, and silt, respectively, at 100–125 cm. Model error, expressed as the root mean square error (RMSE), increased with depth—from 8.1% for surface clay to 13.4% in the deepest layer. Variable importance analysis identified piezometric elevation, precipitation, the Grain Size Index (GSI), the spectral band ratio Band3/Band7, and topographic indices such as the Multiresolution Ridge Top Flatness (MrRTF) as the most influential predictors of soil texture. Furthermore, uncertainty evaluation using the Prediction Interval Coverage Probability (PICP) index revealed decreasing prediction reliability with greater depth. Based on these findings, future research should prioritize the use of higher-resolution elevation datasets, increase sampling density, and incorporate geological and land-use variables to enhance model accuracy and generalizability. This study presents a robust methodology for generating accurate, spatially consistent estimates of soil texture components across defined profile depths and provides a scientific foundation for advancing regional-scale 3D soil mapping.