Scholarlink

Background and Aims <p>High-resolution mapping of Soil organic matter (SOM) in black soil regions using remote sensing technology has been extensively studied. However, in areas with complex terrain, where rolling hills and gully zones coexist, the spatial heterogeneity of hydrothermal conditions often limits the effectiveness of traditional global models. To address this challenge, this study proposes a terrain-based local regression model to improve the accuracy of SOM spatial distribution prediction in complex terrains.</p> Methods <p>Using Shuguang Farm in northeastern China, which features Planosol, the study area was divided into partition I (upper slope) and partition II (lower slope and valley bottom). Bare soil imagery from Sentinel-2 (2019–2024, April and May) and 170 soil samples were used to develop both global and terrain-based local regression models using the Random Forest (RF) algorithm. And then model performance was then compared.</p> Results <p>The study found that: (1) May imagery provided the best window for SOM prediction, with higher accuracy than April; (2) the local regression model (R<sup>2</sup> = 0.653, RMSE = 0.489%) outperformed the global model (R<sup>2</sup> = 0.609, RMSE = 0.518%); (3) environmental covariates, especially climatic factors, were crucial in enhancing model performance, with terrain covariates being particularly important in partition I.</p> Conclusion <p>The terrain-based local regression model effectively improves SOM prediction accuracy in complex terrains, offering a valuable tool for precision agriculture and soil management.</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

Terrain-partitioned local regression for high-resolution soil organic matter mapping in complex soil landscapes

  • 期刊论文

  • 发表日期：

    2026年5月25日

查看全文
  • Xue Li,
  • Bo Jiang,
  • Deqiang Zang,
  • Depiao Kong,
  • Changkun Wang,
  • Ya Chen,
  • Huanjun Liu,
  • Chong Luo

摘要

Background and Aims

High-resolution mapping of Soil organic matter (SOM) in black soil regions using remote sensing technology has been extensively studied. However, in areas with complex terrain, where rolling hills and gully zones coexist, the spatial heterogeneity of hydrothermal conditions often limits the effectiveness of traditional global models. To address this challenge, this study proposes a terrain-based local regression model to improve the accuracy of SOM spatial distribution prediction in complex terrains.

Methods

Using Shuguang Farm in northeastern China, which features Planosol, the study area was divided into partition I (upper slope) and partition II (lower slope and valley bottom). Bare soil imagery from Sentinel-2 (2019–2024, April and May) and 170 soil samples were used to develop both global and terrain-based local regression models using the Random Forest (RF) algorithm. And then model performance was then compared.

Results

The study found that: (1) May imagery provided the best window for SOM prediction, with higher accuracy than April; (2) the local regression model (R2 = 0.653, RMSE = 0.489%) outperformed the global model (R2 = 0.609, RMSE = 0.518%); (3) environmental covariates, especially climatic factors, were crucial in enhancing model performance, with terrain covariates being particularly important in partition I.

Conclusion

The terrain-based local regression model effectively improves SOM prediction accuracy in complex terrains, offering a valuable tool for precision agriculture and soil management.