Scholarlink

Abstract <p>Soil pore structure is a crucial indicator of soil quality, significantly influencing water movement, gas exchange, microbial activity, and plant root growth. This study statistically analyzes the seasonal changes in soil pore structure of Calcic Cambisol in Luancheng County, Shijiazhuang City, using computed tomography (CT) to reveal the variation patterns of soil pore space across different depths and seasons. Based on these analyses, the Lasso method was employed for variable selection, and support vector machine regression (SVR) and decision tree regression (DTR) models were used to predict soil pore percentage area across seasons. The results indicated that during spring, summer, and autumn, both models achieved high prediction accuracy, with coefficients of determination exceeding 0.89. The DTR model outperformed the SVR model in terms of accuracy. However, prediction accuracy declined in winter due to structural complexity caused by freeze-thaw cycles. SHapley Additive exPlanations analysis was conducted to improve model interpretability. Results indicated that the number of pores (Count) was the most influential feature in spring, likely driven by root activity and actinomycetes. In summer and autumn, grey scale density of the pore (IntDen), depth of the soil (Depth), and Count contributed most, reflecting vertical heterogeneity and water redistribution. In winter, perimeter of the pores (Perim) and IntDen were dominant, suggesting freeze-thaw–induced pore deformation. These findings demonstrate the seasonal variability of pore-forming mechanisms and highlight the potential of machine learning for improving soil structural assessment under dynamic environmental conditions.</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

Study on the Seasonal Change Pattern of Soil Pore Structure and Its Prediction Model

  • 期刊论文

  • 发表日期：

    2025年1月1日

查看全文
  • Tao Ban,
  • Fei Wang,
  • Xiaoming Guo,
  • Yifan Wang

摘要

Abstract

Soil pore structure is a crucial indicator of soil quality, significantly influencing water movement, gas exchange, microbial activity, and plant root growth. This study statistically analyzes the seasonal changes in soil pore structure of Calcic Cambisol in Luancheng County, Shijiazhuang City, using computed tomography (CT) to reveal the variation patterns of soil pore space across different depths and seasons. Based on these analyses, the Lasso method was employed for variable selection, and support vector machine regression (SVR) and decision tree regression (DTR) models were used to predict soil pore percentage area across seasons. The results indicated that during spring, summer, and autumn, both models achieved high prediction accuracy, with coefficients of determination exceeding 0.89. The DTR model outperformed the SVR model in terms of accuracy. However, prediction accuracy declined in winter due to structural complexity caused by freeze-thaw cycles. SHapley Additive exPlanations analysis was conducted to improve model interpretability. Results indicated that the number of pores (Count) was the most influential feature in spring, likely driven by root activity and actinomycetes. In summer and autumn, grey scale density of the pore (IntDen), depth of the soil (Depth), and Count contributed most, reflecting vertical heterogeneity and water redistribution. In winter, perimeter of the pores (Perim) and IntDen were dominant, suggesting freeze-thaw–induced pore deformation. These findings demonstrate the seasonal variability of pore-forming mechanisms and highlight the potential of machine learning for improving soil structural assessment under dynamic environmental conditions.