The Innovation Team of Smart Agriculture at the Institute of Agricultural Resources and Regional Planning (IARRP) of the Chinese Academy of Agricultural Sciences (CAAS), has made a series of significant advances in the remote sensing monitoring of heavy metal concentration distribution in farmland soils. 

The research findings were published in two leading journals: Ecological Indicators (IF = 7.0) with the article titled "Retrieval of chromium and mercury concentrations in agricultural soils: Using spectral information, environmental covariates, or a fusion of both?" and IEEE Transactions on Geoscience and Remote Sensing (IF = 7.5) with "A Spectral Hierarchical Machine Learning for Predicting Arsenic Concentration in Farmland Soil Using Sentinel-2 Imagery."

Soil health is a crucial foundation for maintaining the Earth's life systems, ensuring food security, and achieving ecological sustainability. With the rapid pace of urbanization and industrialization, heavy metals have increasingly accumulated in farmland soils, leading to soil quality degradation and significantly impacting crop production and human health. Therefore, accurately and efficiently obtaining spatial distribution information on heavy metal concentrations in large-scale farmland soils has become an urgent scientific issue.

Multispectral satellite remote sensing, with its high spatiotemporal resolution, wide coverage, and low cost, shows great potential for large-scale monitoring of heavy metals in farmland soils. However, due to the complexity of soil environments and insufficient spectral information, challenges remain in accurately retrieving heavy metal concentrations using multispectral remote sensing.

To address these challenges, the team proposed a machine learning modeling strategy that integrates spectral information and environmental variables to enhance the accuracy of heavy metal concentration retrieval. Results indicated that the fusion of spectral information and environmental variables using the Extreme Gradient Boosting (XGBoost) model yielded the best performance in retrieving chromium (Cr) and mercury (Hg) concentrations in farmland soils, significantly outperforming Support Vector Machine (SVM) and Back Propagation Neural Network (BPNN) models (Wang et al., Ecological Indicators, 2024). Additionally, in the hierarchical modeling strategy, the Cubist model based on soil organic matter (SOM) stratification achieved higher prediction accuracy for arsenic (As) concentration retrieval, surpassing the global Convolutional Neural Network (CNN) model (Wang et al., IEEE Transactions on Geoscience and Remote Sensing, 2025).

Thus, the modeling approach that combines spectral stratification with environmental variables effectively addresses the impact of soil environmental complexity on predictive performance, providing a new perspective for efficient monitoring and early warning of heavy metal pollution in farmland soils.

Figure 1: Technical Route for Soil Heavy Metal Retrieval Integrating Multispectral Satellite Data and Environmental Covariates

Figure 2: Comparison of Validation Accuracy for Arsenic Concentration Retrieval Based on Global and Hierarchical Models

Dr. Wang Li, a postdoctoral researcher from the Innovation Team of Smart Agriculture, is the first author, and Associate Researcher Zha Yan is the corresponding author. The research was supported by State Key Laboratory of Efficient Utilization of Arable Land in China, the National Key Research and Development Program of China, the National Natural Science Foundation of China, and the China Postdoctoral Science Foundation.

Citation and Original Article Links:

- Wang L, Zhou Y, Sun X, et al. Retrieval of chromium and mercury concentrations in agricultural soils: Using spectral information, environmental covariates, or a fusion of both? Ecological Indicators, 2024, 167: 112594. [Link](https://doi.org/10.1016/j.ecolind.2024.112594)

- Wang L, Zhou Y, Zhou Z, et al. A Spectral Hierarchical Machine Learning for Predicting Arsenic Concentration in Farmland Soil Using Sentinel-2 Imagery. IEEE Transactions on Geoscience and Remote Sensing, 2025, 63: 1-14. [Link](https://doi.org/10.1109/TGRS.2025.3532678)