Scholarlink

To address the challenges of insufficient exploration of spatial correlation features among environmental variables and inadequate analysis of synergistic mechanisms in multi-source data within current digital soil mapping (DSM) research, this study pioneers a three-dimensional convolutional neural network (3D-CNN) architecture that innovatively captures volumetric spatial-environmental interactions to investigate the enhancement effect of multi-variable correlations on soil organic carbon (SOC) prediction in Shandong Province. Results indicate that the 3D-CNN model, leveraging its three-dimensional convolutional kernels to extract correlations among environmental variables, achieved superior prediction accuracy compared to Random Forest (RF) and two-dimensional convolutional neural network (2D-CNN) models, with improvements of 31.4% and 11.1%, respectively. Furthermore, the 3D-CNN demonstrated enhanced performance in error control. Spatial pattern analysis revealed a gradient distribution of SOC content in Shandong Province, characterized by lower values in the northwestern alluvial plains and coastal regions and higher values in the hilly areas of central and southern Shandong. By effectively uncovering correlations among multi-source environmental variables through deep learning techniques, this study confirms the unique advantages of the 3D-CNN model in analyzing multi-source environmental data and provides novel perspectives and methodologies for high-precision DSM.

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

Application of 3D-CNN and Multi-Source Environmental Variables in SOC Prediction for Cultivated Land in Shandong Province

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Erlong Xiao,
  • Daocheng Li,
  • Yingxin Xia,
  • Jun Yan,
  • Lixin Ning

摘要

To address the challenges of insufficient exploration of spatial correlation features among environmental variables and inadequate analysis of synergistic mechanisms in multi-source data within current digital soil mapping (DSM) research, this study pioneers a three-dimensional convolutional neural network (3D-CNN) architecture that innovatively captures volumetric spatial-environmental interactions to investigate the enhancement effect of multi-variable correlations on soil organic carbon (SOC) prediction in Shandong Province. Results indicate that the 3D-CNN model, leveraging its three-dimensional convolutional kernels to extract correlations among environmental variables, achieved superior prediction accuracy compared to Random Forest (RF) and two-dimensional convolutional neural network (2D-CNN) models, with improvements of 31.4% and 11.1%, respectively. Furthermore, the 3D-CNN demonstrated enhanced performance in error control. Spatial pattern analysis revealed a gradient distribution of SOC content in Shandong Province, characterized by lower values in the northwestern alluvial plains and coastal regions and higher values in the hilly areas of central and southern Shandong. By effectively uncovering correlations among multi-source environmental variables through deep learning techniques, this study confirms the unique advantages of the 3D-CNN model in analyzing multi-source environmental data and provides novel perspectives and methodologies for high-precision DSM.