Overline

Mid-Infrared Spectroscopy for Soil Health Assessment.

Developing rapid, cost-effective approaches to predict soil health indicators using spectroscopy and machine learning.

Funded by the Florida Cattle Enhancement Board

Project Overview

Traditional soil health testing can be costly, labor-intensive, and time-consuming. This project explores the use of mid-infrared diffuse reflectance Fourier transform spectroscopy (mid-DRIFTS) combined with machine learning to rapidly predict soil health indicators in Florida ranchlands.

1

Reduce the cost and time associated with soil health assessments.

2

Evaluate machine learning approaches.

3

Improve prediction accuracy through spectral partitioning.

Why It Matters?

Traditional laboratory analyses require substantial time and resources. Spectroscopy provides a pathway to generate soil health information faster and at lower cost, potentially increasing the adoption of soil health testing among producers and land managers.

Major Findings

Partitioning spectra into informative regions reduced noise and improved model performance across physical, chemical, and biological soil indicators.

Machine learning approaches (Neural networks and support vector machines) captured complex relationships within soil spectra more effectively than traditional PLS models, resulting in more accurate predictions of soil properties.

Clay, sand, and silt showed some of the largest improvements in model performance.

Indicators such as soil organic matter, active carbon, protein, and mineralizable carbon were successfully predicted using mid-DRIFTS, demonstrating its potential for rapid soil health assessment.

Project Outcomes

Improving Soil Health Predictions with Spectroscopy.

By combining mid-infrared spectroscopy with machine learning, this project demonstrates a rapid, scalable, and cost-effective approach for predicting soil health indicators in Florida’s sandy soils.

Prediction Error
Reduction in RMSE across soil indicators
Prediction Accuracy
Improvement in R²
Model Robustness
Increase in RPD values
Best Models
Best-performing machine learning approaches
Soil Use and Management

Related Publications

Celestin, F., Deiss, L., Champiny, Ryan E., Dubeux, Jose C.B., Maltais-Landry, G., Mylavarapu, R., and Lin, Y. (2026). Predicting soil health indicators in sandy soils using diffuse reflectance mid-infrared Fourier transform spectroscopy. Soil Use and Management, 42(1), e70184. https://doi.org/10.1111/sum.70184

Read Publication