Skip to main content
ARS Home » Plains Area » El Reno, Oklahoma » Grazinglands Research Laboratory » Forage and Livestock Production Research » Research » Publications at this Location » Publication #365127

Research Project: Integrated Agroecosystem Research to Enhance Forage and Food Production in the Southern Great Plains

Location: Forage and Livestock Production Research

Title: Predicting forage quality of warm-season legumes by Near Infrared Spectroscopy coupled with machine learning techniques

Author
item BAATH, GURJINDER - Oklahoma State University
item BAATH, HARPINDER - Oklahoma State University
item Gowda, Prasanna
item THOMAS, JOHNSON - Oklahoma State University
item Northup, Brian
item RAO, SRINIVAS - Retired ARS Employee
item SINGH, HARDEEP - Oklahoma State University

Submitted to: Sensors
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/4/2020
Publication Date: 2/6/2020
Citation: Baath, G.S., Baath, H.K., Gowda, P.H., Thomas, J.P., Northup, B.K., Rao, S., Singh, H. 2020. Predicting forage quality of warm-season legumes by Near Infrared Spectroscopy coupled with machine learning techniques. Sensors. 20(3):867. https://doi.org/10.3390/s20030867.
DOI: https://doi.org/10.3390/s20030867

Interpretive Summary: There is a growing interest in the cultivation of warm-season legumes for forage in the southern United States. Quantifying the forage quality of warm-season is crucial for research as well as advisory work. The near infrared spectroscopy (NIRS) is an alternative to the time-consuming and expensive traditional analytical techniques of determining forage quality. However, the effectiveness of NIRS in predicting forage quality of warm-season legumes has not been tested. Therefore, we evaluated the accuracy of NIRS in predicting forage quality of different warm-season legumes using three different calibration methods: partial least square (PLS), support vector machine (SVM) and gaussian processes (GP). Forage quality parameters tested include crude protein (CP), neutral detergent fiber (NDF), acid detergent fiber (ADF) and in vitro true digestibility (IVTD). Results showed that PLS performed best at calibrating each of the forage quality parameter in all species based and global models, while SVM consistently resulted in better prediction accuracy on cross-validation and external validation among the three tested methods. This study demonstrated that machine learning algorithms like SVM could make it possible to develop robust models using a small number of samples; thus, their use should be encouraged in other NIRS based applications.

Technical Abstract: Warm-season legumes have been receiving increased attention as forage resources in the southern United States and other parts of the world. However, the use of near infrared spectroscopy (NIRS) in predicting forage quality of warm-season has not been much explored. The objective of this study was to evaluate the performance of NIRS in predicting the forage quality of five warm-season legumes namely guar (Cyamopsis tetragonoloba), tepary bean (Phaseolus acutifolius), pigeon pea (Cajanus cajan) soybean (Glycine max), and mothbean (Vigna aconitifolia) using three different statistical methods: partial least square (PLS), support vector machine (SVM) and gaussian processes (GP). Additionally, the efficacy of global models to predict forage quality of warm-season legumes was investigated. A set of 70 forage samples was used in developing species-based models for crude protein (CP), neutral detergent fiber (NDF), acid detergent fiber (ADF) and in vitro true digestibility (IVTD) in guar and tepary bean, and CP and IVTD in pigeon pea and soybean forages. All species-based statistical models were tested through 10-fold cross-validation, and external validation using 20 samples of respective species. The global models for CP and IVTD of warm-season legumes were developed using a set of 150 random samples, including 30 samples for each of the five species. The global models were tested through 10-fold cross-validation, and external validation using five individual sets of 20 samples each for different legume species. Among methods, PLS consistently performed best (R2c = 0.94-0.98) for all forage quality parameters in both species-based and global models. The SVM gave better prediction accuracy for guar and soybean crops, and global models, and both SVM and PLS performed better for tepary bean and pigeon pea crops. The global modeling approach that developed single model for all five crops yielded sufficient accuracy (R2cv/R2v = 0.92-0.99) in predicting CP of five different legumes, but the predictions of in vitro true digestibility (IVTD) for different crops were highly varied (R2cv/R2v = 0.42-0.98). Machine learning algorithms like SVM could develop robust NIRS-based models with a relatively small number of samples, and thus needs further attention in different NIRS based applications.