Machine learning–enabled NIR soil sensing for rapid, low-cost nutrient analysis

Overview

This solution combines field-portable near-infrared (NIR) spectral sensors with machine learning algorithms to dramatically reduce the cost and time required for soil nutrient analysis. Current commercial soil testing is too expensive and slow for large datasets and rapid in-field sensing, limiting the ability of agronomists and farmers to implement precision nutrient management. By replacing conventional wet-chemistry analysis with spectroscopy plus predictive modeling, the technology aims to cut per-sample analysis costs by more than 90% and deliver results in near real time.

The value proposition is straightforward: faster, cheaper soil data enables more frequent and spatially dense sampling, which supports better fertilizer recommendations, improved crop yields, and reduced environmental impacts from over-application of nutrients. The approach is particularly relevant for broad-acre agriculture in regions such as western Canada, where detailed soil fertility information is needed to optimize nitrogen, phosphorus, and potassium inputs.

Technical specifications

• Uses FT-NIR spectroscopy (laboratory) and a field-compatible NIR probe to collect soil spectra directly in the field and on removed samples at the 0–15 cm depth
• Applies machine learning algorithms to spectral data to predict soil properties including pH, soil organic carbon (SOC), total nitrogen, total phosphorus, plant-available nitrogen, phosphorus, and potassium, and effective cation exchange capacity (eCEC)
• Builds on established NIR performance for directly measuring SOC, total N, clay content, and cation exchange capacity, while extending prediction to nutrients that are difficult to measure directly due to their low concentration in the soil matrix
• Validation plan includes collecting approximately 1,000 soil spectra from agricultural fields alongside agronomists, with parallel laboratory reference analysis on dried, ground samples
• Designed to integrate with existing agronomic calibration frameworks, such as the Brazilian model that links SOC levels to crop yield potential and nitrogen fertilizer recommendations

Technology readiness level

The technology is at an early-to-mid stage of development. The underlying NIR spectroscopy capability is already proven in laboratory settings, and a commercial soil testing laboratory in western Canada is currently collecting FT-NIR spectra of dried samples and using internal calibrations to estimate nutrient levels. The next phase of validation will test the system under real field conditions using a portable NIR probe, compare in-field versus laboratory spectral measurements, and benchmark machine learning predictions against standard soil chemistry results. Successful completion of this validation will establish whether the approach can deliver reliable, low-cost, in-field soil nutrient analysis suitable for adoption by agronomy service providers and precision agriculture programs.