Hyperspectral reflectance and machine learning to monitor legume feed quality

Authors: H.K., CHINMAYI - Oak Ridge Institute For Science And Education (ORISE); Flynn, Kyle; BAATH, GURJINDER - Texas Agrilife Research; LEE, TREY - Former ARS Employee; Gowda, Prasanna; Northup, Brian; Ashworth, Amanda

Submitted to: Agronomy Journal
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 3/20/2026
Publication Date: 4/22/2026
Citation: H.K., C., Flynn, K.C., Baath, G.S., Lee, T.O., Gowda, P.H., Northup, B.K., Ashworth, A.J. 2026. Hyperspectral reflectance and machine learning to monitor legume feed quality. Agronomy Journal. https://doi.org/10.1002/agj2.70386.
DOI: https://doi.org/10.1002/agj2.70386

Interpretive Summary: Hyperspectral remote sensing is a promising technology that can help farmers optimize crop production through detailed spectral information. In this study, we examined the quality of forage from three legume species: soybean, tepary bean, and mothbean. Using data collected on-site, we made predictions about fiber content, digestibility, and protein levels in the forage. We used different machine learning methods, like k-Nearest Neighbors, partial least squares regression, support vector machine, and random forest, to build predictive models for the data. We also explored how this technology could be applied to a future hyperspectral satellite by adjusting the wavelengths, we simulated the satellite's bands and investigated their potential. We compared models that considered the legume species as a variable with models that separated the data by each species. The results showed that hyperspectral remote sensing is a valuable tool, both on-site and through convolved satellite bands. The random forest and support vector machine models consistently outperformed other models in predicting the nutritional quality of legumes. The support vector machine model was particularly efficient. Generally, creating models for individual species yielded better results than models including all legume species together. These findings provide insights into the nutritional composition of legume forage, aiding various agricultural applications. Moreover, they contribute to the advancement of precision agriculture for producing high-quality forage crops.

Technical Abstract: Hyperspectral remote sensing is a promising technology that can provide precision agriculture the ability to capture detailed spectral information and actionable insights to enable farmers to optimize crop production. This study investigated the quality of forage from three legume species(soybean[Glycine max], tepary bean [Phaseolus acutifolius] and mothbean [Vigna aconitifolia]) using in-situ hyperspectral data and convolved wavebands to predict concentrations of neutral detergent fiber (NDF), acid detergent fiber (ADF), in vitro true digestibility (IVTD), and crude protein (CP) using different methods of machine learning (k-Nearest Neighbors [KNN], partial least squares regression [PLS], support vector machine [SVM], and random forest [RF]). Exploration of potential applications of a future hyperspectral satellite (i.e. CHIME) was conducted by convolving hyperspectral wavelengths to fit hyperspectral bands from the proposed satellite. Exploration of global models that included legume species as a variable, as well as an alternative exploration of splitting the data into individual sub-datasets by species were compared. Results suggest hyperspectral remote sensing can be a valuable tool for either in-situ or convolved satellite bands (i.e., CHIME). The RF and SVM models consistently outperformed other models, demonstrating superior predictive performance when evaluating legume nutritive quality, with SVM being less computationally demanding. Among model comparisons, building models for individual species generally outperformed models that incorporated legume species as a variable. Results provide insight into the nutritional composition of forage legume that will aid in their use in different agronomic applications. The presented findings also provide guidance and contributions to applying precision agriculture to producing forages.