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

Soil organic matter is a key indicator of soil fertility, soil quality and ecosystem functioning. Accurate information on its spatial distribution is therefore important for soil resource management and sustainable agricultural production. Yet mapping soil organic matter across large and environmentally complex regions remains challenging, particularly where different soil types coexist and topography, climate and other environmental factors vary substantially.

To that end, a team of researchers from China proposed a “remote sensing zoning-feature selection optimization-random forest (RSZ-FSO-RF)” framework. Their findings are published in the Journal of Integrative Agriculture.

The researchers focused on Northeast China, one of the world’s major Mollisol regions. They collected 1,603 surface (0-20 cm) soil samples and integrated these measurements with Landsat-8 multi-temporal imagery from 2014–2023, with topographic and climatic factors, and leveraging the Google Earth Engine (GEE) platform.

Instead of relying solely on existing soil maps to define modelling zones, the researchers first used remote sensing and environmental covariates to distinguish Mollisol from non-Mollisol areas and then constructed separate Random Forest models within each zone.The remote-sensing zoning model achieved an overall classification accuracy of 92.13%, with a Kappa coefficient of 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).

“When remote-sensing bands, topographic and climatic variables were included, we found that models based on the new zoning scheme showed higher overall prediction accuracy than those based on conventional FAO zoning,” shares corresponding author Prof. Chong Luo from the Northeast Institute of Geography and Agroecology, Chinese Academy of Sciences.

The researchers then compared three feature-selection approaches—recursive feature elimination, Boruta and Lasso—and found that recursive feature elimination algorithm provided the most effective optimization. After feature selection, the overall model reached an R² of 0.627 and an RMSE of 6.781 g kg⁻¹.

“Soil organic matter mapping can be improved by first accounting for regional soil heterogeneity and then optimizing the variables used within each zone,” says Luo. “Remote-sensing zoning provides a more adaptive basis for regional modelling than applying the same relationships across an entire heterogeneous landscape.”

The resulting maps also revealed differences between the two zones. “Mean soil organic matter was slightly higher in Mollisol areas, while non-Mollisol areas showed greater spatial variability, suggesting that the factors controlling soil organic matter differ between regions,” adds Luo. “These differences highlight the importance of using zone-specific predictors when mapping soil properties over complex landscapes. The framework could provide useful support for more precise soil monitoring and land management.”

Contact the authors: Xue Li, E-mail: S231202038@neau.edu.cn; #Correspondence Chong Luo, E-mail: luochong@iga.ac.cn

The publisher KeAi was established by Elsevier and China Science Publishing & Media Ltd to unfold quality research globally. In 2013, our focus shifted to open access publishing. We now proudly publish more than 300 world-class, open access, English language journals, spanning all scientific disciplines. Many of these are titles we publish in partnership with prestigious societies and academic institutions, such as the National Natural Science Foundation of China (NSFC).