Near-Infrared Spectral Tool for Rapid Soil Property Assessment
A machine learning–based spectral method that uses near-infrared (NIR) measurements to rapidly estimate key soil chemical properties—including pH, organic matter, calcium, magnesium, sodium, and potassium—reducing the cost and time of conventional laboratory analysis. The tool is designed for farmers, agronomists, and researchers seeking to improve crop yield through precision agriculture and is being developed using AGROSAVIA's soil sample bank.
Overview
Accurate knowledge of soil properties is essential for effective crop management and economic performance in agricultural production. Conventional laboratory soil analysis is often too costly and time-consuming to support precision agriculture, limiting farmers' ability to implement optimal practices. This project proposes a spectral method for land property assessment based on near-infrared (NIR) measurements of soil samples. By leveraging AGROSAVIA's extensive soil sample bank—containing more than 5,000 samples with standard laboratory results—the team is training and validating machine learning models to predict key soil chemical properties rapidly and cost-effectively. The resulting tool is intended to serve any farmer, agronomy professional, or scientist aiming to improve crop yields.
Technical specifications
- Input: Visible and near-infrared (vis-NIR) spectral measurements of soil samples
- Output: Estimated values for six chemical properties: acidity (pH), organic matter (OM), calcium (Ca), magnesium (Mg), sodium (Na), and potassium (K)
- Machine learning approach: LASSO regression model, benchmarked against Partial Least Squares Regression (PLSR) and single-band chemometric techniques
- Validation dataset: Over 500 sugarcane soil samples from Colombia, with cross-validation correlations above 0.6 for pH, OM, Ca, and Mg
- Comparison results: LASSO outperforms traditional chemometric approaches used at AGROSAVIA for most properties, though Na and K remain challenging
- Future development: Expanded training using the full ~5,000-sample bank, with additional mathematical and computational tools explored for improved spectral modeling
Technology readiness level
The technology is currently at an early-to-mid stage of development. Initial validation on 500+ Colombian sugarcane soil samples has demonstrated proof of concept, with machine learning predictions achieving correlations above 0.6 for most target properties and outperforming traditional chemometric methods. The team is now seeking funding to scale up NIR measurements across nearly 5,000 pre-tested soil samples available at AGROSAVIA, which will enable broader model training, improved accuracy, and assessment of the tool's limitations before deployment to end users in the agricultural sector.
About Agrosavia
AGROSAVIA is Colombia’s public agricultural R&D corporation near Bogotá, operating a nationwide network of research centers and experimental farms. Co-located labs, greenhouses, pilot plots and living collections—including national germplasm banks—enable rapid prototyping and access to diverse genetic resources. An outreach and innovation network links companies and producer groups to multi-site trials and farmer engagement across varied agroecological zones. Research is backed by the Government of Colombia—principally the Ministry of Agriculture and Rural Development—with competitive national and international funding, and a technology transfer function to structure collaboration and IP.
delivered to your inbox.