Journal of Integrative Agriculture ›› 2026, Vol. 25 ›› Issue (8): 3453-3468.DOI: 10.1016/j.jia.2026.01.016

• • 上一篇    下一篇

  

  • 收稿日期:2025-07-15 修回日期:2026-01-14 接受日期:2025-11-17 出版日期:2026-08-20 发布日期:2026-07-09

A zoning-based machine learning framework for accurate soil organic matter prediction across Mollisol and non-Mollisol regions

Xue Li1, 2, Bo Jiang2, Depiao Kong1, Deqiang Zang2, Ya Chen2, Changkun Wang3, Huanjun Liu1, Chong Luo1#   

  1. 1 State Key Laboratory of Black Soils Conservation and Utilization, Northeast Institute of Geography and Agroecology, Chinese Academy of Sciences, Changchun 130102, China

    2 School of Public Administration and Law, Northeast Agricultural University, Harbin 150030, China

    3 Nanjing Institute of Soil Science, Chinese Academy of Sciences, Nanjing 210008, China

  • Received:2025-07-15 Revised:2026-01-14 Accepted:2025-11-17 Online:2026-08-20 Published:2026-07-09
  • About author:Xue Li, Mobile: +86-18846763278, E-mail: S231202038@neau.edu.cn; #Correspondence Chong Luo, Mobile: +86-13304807096, E-mail: luochong@iga.ac.cn
  • Supported by:

    This study was supported by the National Natural Science Foundation of China (42401460) and the National Key R&D Program of China (2021YFD1500100).  

PDF

摘要：

土壤有机质（SOM）是土壤肥力和生态系统功能的核心指标。然而，在黑土（Mollisol）和非黑土共存的地区，由于地形异质性明显以及高维协变量冗余性高，高精度空间制图面临重大挑战。本研究提出了一种“遥感分区-特征选择优化-随机森林（RSZ-FSO-RF）”框架。通过整合2014-2023年Landsat-8多时相影像与地形、气候因素，并利用Google Earth Engine（GEE）平台，实现了对黑土和非黑土区域的高精度遥感分区（总体精度：92.13%，Kappa系数：0.70）。随后，在每个分区内建立了局部随机森林回归模型进行SOM预测，并使用递归特征消除（RFE）方法优化预测变量。结果表明，与基于FAO分区的建模方法相比，RSZ-FSO-RF框架显著提高了预测精度（R²=0.619，RMSE=6.849 g kg-1）。进一步的特征优化继续提升了模型表现（R²=0.627，RMSE=6.781 g kg-1）。值得注意的是，不同分区的最佳预测变量组合差异显著，非黑土地区的SOM空间变异性普遍高于黑土地区。通过有机结合遥感分区与特征选择，本框架有效减轻了协变量冗余性，同时考虑了局部异质性，显著提高了高分辨率SOM制图的准确性和稳定性。此外，本研究为在复杂地形条件下的土壤资源管理和可持续农业发展提供了科学依据和决策支持。

Abstract:

Soil organic matter (SOM) is a core indicator of soil fertility and ecosystem function.  However, in regions where Mollisol and non-Mollisol coexist, high-precision spatial mapping faces significant challenges due to pronounced terrain heterogeneity and redundancy in high-dimensional covariates.  This study proposes a “remote sensing zoning-feature selection optimization-random forest (RSZ-FSO-RF)” framework.  By integrating Landsat-8 multi-temporal imagery from 2014–2023 with topographic and climatic factors, and leveraging the Google Earth Engine (GEE) platform, it achieves high-precision remote sensing zoning of Mollisol and non-Mollisol areas (overall accuracy: 92.13%, Kappa coefficient: 0.70).  Subsequently, local Random Forest (RF) regression models were established within each zone for SOM prediction, with predictive variables optimized using recursive feature elimination (RFE).  Results demonstrate that compared to FAO-zone-based modeling, the RSZ-FSO-RF framework significantly enhances prediction accuracy (R2=0.619, RMSE=6.849 g kg–1).  And further feature optimization continued to enhance model performance (R2=0.627, RMSE=6.781 g kg–1).  Notably, optimal predictor combinations varied significantly across zones, with SOM spatial variability generally higher in non-Mollisol areas than in Mollisol regions.  By organically integrating remote sensing zoning with feature selection, this framework effectively mitigates covariate redundancy while accounting for local heterogeneity, significantly enhancing the accuracy and stability of high-resolution SOM mapping.  Furthermore, this study provides scientific basis and decision support for soil resource management and sustainable agricultural development under complex topographic conditions.

Key words: soil organic matter , remote sensing zoning , feature selection algorithm , random forest algorithm , prediction