Smarter soils: Machine learning reveals lead hotspots beneath our feet
image: Trivariate maps created using Pb as the optimal element, combined with two of Zn, soil organic C (SOC), and Fe, to show the spatial distribution relations between Pb and Zn, SOC, and Fe in the study area in the Moravian-Silesian region of Frydek Mistek Precinct in the Czech Republic.
Credit: Pedosphere
Lead (Pb) contamination in agricultural soils is an invisible threat that can infiltrate food systems and jeopardize public health. To address this risk, scientists developed a powerful new method that combines machine learning algorithms with spectral imaging and terrain data to map and predict soil Pb levels. Among several tested models, the extreme gradient boosting (EGB) algorithm, paired with specially preprocessed spectral data and topographic features, achieved the highest prediction accuracy. This integrated approach not only pinpoints contamination hotspots but also offers a fast, non-invasive tool for monitoring soil quality and informing land-use decisions.
Lead (Pb), a persistent and toxic element in soils, often originates from industrial activities, vehicular emissions, and agricultural inputs. Its presence in croplands can lead to serious health consequences, particularly for children, and poses challenges for food safety. Traditional soil analysis methods are costly and labor-intensive, making large-scale monitoring impractical. In contrast, remote sensing technologies, especially those using visible and infrared spectra, offer a rapid alternative—but spectral data alone can be noisy and unreliable without advanced processing. Due to these limitations, there is a growing need to develop precise, scalable, and data-driven approaches for detecting and managing Pb contamination in soils.
In a study (DOI: 10.1016/j.pedsph.2024.01.002) published on March 26, 2025, in Pedosphere, researchers from the Czech University of Life Sciences Prague, together with international collaborators, unveiled a novel soil contamination prediction framework. By fusing spectral data, topographic variables, and six advanced machine learning models, the team successfully predicted the distribution of Pb in farmland soils. The study not only enhances the accuracy of contamination mapping but also reveals the environmental factors most responsible for Pb dispersion.
The researchers collected 115 topsoil samples across agricultural fields in the Czech Republic, measuring Pb levels alongside iron, zinc, and soil organic carbon (SOC). Using a high-resolution spectrometer, they obtained VNIR-SWIR spectral data and combined this with six key terrain attributes such as slope, elevation, and drainage patterns. These datasets were fed into six machine learning algorithms, including artificial neural networks (ANN), support vector machines (SVM), and extreme gradient boosting (EGB). After extensive model testing, the EGB algorithm, combined with standard normal variate (SNV)–processed spectra and terrain features, delivered the most accurate predictions, with an R² of 0.75 and a low error margin. The study also employed trivariate mapping to visualize Pb’s spatial relationships with SOC and iron, revealing that elevation and slope were major drivers of Pb distribution. These insights underscore the power of combining environmental sensing with AI to unravel complex pollution dynamics in soil ecosystems.
“Our results show that artificial intelligence can revolutionize how we detect and manage soil pollution,” said Dr. Prince Chapman Agyeman, lead author of the study. “By integrating machine learning with spectral and terrain data, we’ve created a reliable, cost-effective system for predicting Pb contamination. This approach equips land managers and environmental agencies with the tools they need to act quickly and efficiently—before contamination becomes a crisis.”
This innovative approach paves the way for real-time, scalable soil monitoring systems. It can help farmers and policymakers identify contamination hotspots, prioritize remediation, and ensure food safety. The framework is adaptable for detecting other pollutants such as cadmium or arsenic and could be enhanced with additional data like land use, climate, or crop history. Future research may integrate deep learning, mid-infrared spectroscopy, or portable X-ray fluorescence (PXRF) to improve prediction depth and resolution. Ultimately, this fusion of AI and environmental sensing offers a promising path toward cleaner soils and safer agriculture in a changing world.
###
References
DOI
Original Source URL
https://doi.org/10.1016/j.pedsph.2024.01.002
Funding Information
This study was supported by an institutional Ph.D. grant (No. 21130/1312/3131) from the Faculty of Agrobiology, Food, and Natural Resources at the Czech University of Life Sciences Prague (CZU), Czech Republic.
About Pedosphere
Pedosphere is a peer-reviewed international journal established in 1991 and published bimonthly in English by Elsevier and Science Press. It is jointly sponsored by the Soil Science Society of China and the Institute of Soil Science, Chinese Academy of Sciences, in collaboration with five leading Chinese institutions in soil science. Under the editorship of Prof. Shen Ren-Fang, the journal publishes high-quality original research and reviews spanning the full spectrum of soil science, including environmental science, agriculture, ecology, bioscience, and geoscience. Topics of interest include soil physics, chemistry, biology, fertility, plant nutrition, conservation, and global change. All submissions undergo rigorous double-blind peer review by an international editorial board and expert panel. Pedosphere is indexed in major databases such as SCI Expanded, SCOPUS, BIOSIS, CAB Abstracts, and CNKI, making it a widely recognized platform for advancing soil science research globally.