Scholarlink

<p>Estimating forest soil organic carbon density (SOCD) is particularly challenging in dry tropical valleys (DTVs) due to their complex topography, highly variable soils, diverse vegetation types, and unique climatic conditions. In this respect, this study incorporates multi-source remote sensing data and multiple machine learning algorithms to improve the accuracy and reliability of forest SOCD estimation. This study integrates 47 variables, including optical remote sensing, environmental factors, and topographic factors, and employs the Boruta algorithm for variable selection. Eight machine learning models are constructed to estimate the forest SOCD in the DTVs of Yuanmou County, Yunnan Province, China. The results indicate that: (1) Regarding the remote sensing estimation of DTV forest SOCD, the soil factors are relatively critical, followed by the vegetation index factors, while the climatic factors have relatively little effect. (2) There are significant differences in the performances of the various models. The extreme gradient boosting algorithm achieves the highest estimation accuracy with an R<sup>2</sup> of 0.68 and a root mean square error (RMSE) of 1.29&#xa0;kg&#xa0;C&#xa0;m<sup>−2</sup>, followed by the multilayer perceptron model with an R<sup>2</sup> of 0.67 and an RMSE of 1.30&#xa0;kg&#xa0;C&#xa0;m<sup>−2</sup>. The Bayesian regularized neural network and multilayer perceptron have largely equal estimation performance, with the elastic net model having the lowest fitting accuracy with an R<sup>2</sup> of 0.32 and RMSE of 1.89&#xa0;kg&#xa0;C&#xa0;m<sup>−2</sup>. The most applicable machine learning models for DTV forest SOCD estimation are extreme gradient boosting, multilayer perceptron, and Bayesian regularized neural network, with extreme gradient boosting showing the best performance. This study provides valuable insights into selecting data sources and models for remote sensing-based SOCD estimation in DTVs.</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

Estimation of Forest Soil Organic Carbon Density in Dry Tropical Valleys Based on Multi-source Data and Machine Learning Algorithm

  • 期刊论文

  • 发表日期：

    2026年5月26日

查看全文
  • Xiongwei Xu,
  • Can Xu,
  • Liankai Zhang,
  • Jinjiang Yang,
  • Guiren Chen,
  • Canfeng Li,
  • Chen Zhang,
  • Fuyan Zou,
  • Min Yan,
  • Zhenhui Wang,
  • Xuefeng Peng

摘要

Estimating forest soil organic carbon density (SOCD) is particularly challenging in dry tropical valleys (DTVs) due to their complex topography, highly variable soils, diverse vegetation types, and unique climatic conditions. In this respect, this study incorporates multi-source remote sensing data and multiple machine learning algorithms to improve the accuracy and reliability of forest SOCD estimation. This study integrates 47 variables, including optical remote sensing, environmental factors, and topographic factors, and employs the Boruta algorithm for variable selection. Eight machine learning models are constructed to estimate the forest SOCD in the DTVs of Yuanmou County, Yunnan Province, China. The results indicate that: (1) Regarding the remote sensing estimation of DTV forest SOCD, the soil factors are relatively critical, followed by the vegetation index factors, while the climatic factors have relatively little effect. (2) There are significant differences in the performances of the various models. The extreme gradient boosting algorithm achieves the highest estimation accuracy with an R2 of 0.68 and a root mean square error (RMSE) of 1.29 kg C m−2, followed by the multilayer perceptron model with an R2 of 0.67 and an RMSE of 1.30 kg C m−2. The Bayesian regularized neural network and multilayer perceptron have largely equal estimation performance, with the elastic net model having the lowest fitting accuracy with an R2 of 0.32 and RMSE of 1.89 kg C m−2. The most applicable machine learning models for DTV forest SOCD estimation are extreme gradient boosting, multilayer perceptron, and Bayesian regularized neural network, with extreme gradient boosting showing the best performance. This study provides valuable insights into selecting data sources and models for remote sensing-based SOCD estimation in DTVs.