Key Points
- Researchers demonstrated that combining diffuse reflectance spectroscopy (DRS) with satellite radar data and environmental variables improving soil organic carbon (SOC) prediction accuracy.
- The study validated the effectiveness of Sentinel-1 synthetic aperture radar (SAR), for consistent and scalable SOC monitoring in semi-arid and tropical regions.
- While the integrated approach using spectroscopy, radar, and environmental data enhanced model performance, it also introduced challenges in handling mixed data types.
A recent study examined how diffuse reflectance spectroscopy (DRS) can improve prediction of the soil organic carbon (SOC) levels. This study, which was published in the journal Geoderma, investigates the benefit of using radar-derived vegetation indices and environmental variables with soil spectroscopy to improve the accuracy of predicting soil organic carbon (SOC) levels (1). Led by Erli Pinto dos Santos and a team of researchers in Brazil, the study offers critical insights into improving soil health assessments, particularly in challenging semi-arid environments.
What is soil organic carbon?
Soil organic carbon (SOC) is one of the most vital components of soil health (1,2). It plays an important role in climate regulation, nutrient cycling, and agricultural productivity (1,2). Because SOC is important for farmers and agricultural scientists to evaluate to optimize crop yields, there is an ongoing push to develop sustainable methods that are effective at measuring SOC. To that end, the research team explored a novel method that can be a reliable, cost-effective alternative to typical methods for analyzing SOC.
What was the method the researchers tested in the study?
In this study, the method involved using DRS, specifically in the visible-near infrared (Vis-NIR) spectrum, through integration with satellite-based remote sensing and environmental data. As part of the experimental procedure, the researchers built a Regional Soil Spectral Library (R-SSL) using 127 soil profiles collected from semi-arid areas in northeastern Brazil (1). This library formed the foundation for testing four modeling scenarios using Cubist machine learning (ML) models, which is a type of rule-based regression method well-suited for handling both categorical and continuous data types (1).
Four scenarios were evaluated in their study. The first scenario was diffuse reflectance data from laboratory spectroscopy alone (SSL only) (1). The second was SSL combined with radar vegetation indices from Sentinel-1 satellite data (1). The third was SSL combined with environmental factors like soil type, land use, and elevation (1). And finally, the fourth scenario was SSL combined with both radar vegetation indices and environmental covariates (1).
What were the findings of the study?
The researchers found that including all covariates resulted in huge improvements. The RMSE dropped to 3.94 g kg⁻¹, which was a reduction of approximately 13% (1). Meanwhile, the R² increased to 0.71, marking a 14% boost in model performance (1). These improvements suggested that SAR-derived radar vegetation indices, when used alongside traditional soil spectroscopy, provide a fuller picture of the complex interactions influencing SOC (1).
An important finding of the study was the value in using the Sentinel-1 synthetic aperture radar (SAR) data. Because the radar was not affected by cloud cover, it was able to provide consistent global coverage (1). This makes it highly valuable for soil monitoring applications, especially in tropical or semi-arid regions where optical sensors are often limited by atmospheric conditions (1). By integrating Sentinel-1 radar data with vis-NIR soil spectra and a suite of environmental variables, the researchers were able to build a more robust and scalable framework for SOC estimation (1).
However, the study also highlights the complexity of integrating diverse data sources. Although the combination of categorical and continuous variables yielded better results, it introduced new requirements for model preprocessing. The researchers found that the models needed to be capable of managing discrete covariates or had to undergo transformations to maintain interpretability and accuracy (1).
What are the future directions of this research?
Soil monitoring technologies are going to continue to grow in importance. As the study highlighted, combining field spectroscopy with remote sensing can be used to build better models for soil health assessment (1).
Accurate SOC prediction is essential for evaluating carbon sequestration strategies, guiding sustainable agriculture, and contributing to international climate commitments. By validating the use of Sentinel-1 SAR data for soil carbon monitoring, the researchers demonstrate how the integration of radar and optical data with ML models opens up new possibilities for scalable, reliable soil testing in regions where traditional laboratory methods are impractical or cost-prohibitive (1).
References
- dos Santos, E. P.; Moreira, M. C.; Fernandes-Filho, E. I.; et al. Integrating Satellite Radar Vegetation Indices and Environmental Descriptors with Visible-infrared Soil Spectroscopy Improved Organic Carbon Prediction in Soils of Semi-arid Brazil. Geoderma 2025, 457, 117288. DOI: 10.1016/j.geoderma.2025.117288
- Beillouin, D.; Corbeels, M.; Demenois, J.; et al. A Global Meta-analysis of Soil Organic Carbon in the Anthropocene. Nat. Commun. 2023, 14, 3700. DOI: 10.1038/s41467-023-39338-z
