Botanical Composition Prediction of Alfalfa–Grass Mixtures using NIRS: Developing a Robust Calibration

Abstract

Botanical composition of alfalfa (Medicago sativa L.)–grass fresh and ensiled mixtures is a key parameter for assessing forage and diet quality as well as for managing mixed stands. Previous attempts to validate near‐infrared reflectance spectroscopy (NIRS) equations for estimating botanical composition have had mixed results. This study was conducted to develop a robust NIRS method to estimate botanical composition of binary alfalfa–grass mixtures. Alfalfa–grass samples were collected across New York State over four growing seasons, hand separated, and a subset were ensiled separately. Dry samples were coarsely ground, mixed in known proportions, and reground for analysis by NIRS at Dairy One Forage Laboratory, Ithaca, NY. Samples were mixed to range from 0 to 100% alfalfa for NIRS calibration, with a total of 741 individual samples from 3 yr used for calibration of three NIRS instruments and samples from a fourth year used for validation. Grass composition was predicted with good precision and accuracy showing biases of 2.49 and standard errors of prediction (SEP) of 5.06, with R2 of 0.972, using the equation developed across multiple instruments. With selection of a robust set of calibration samples over many environments, NIRS can be used to determine the botanical composition of fresh‐dried or ensiled‐dried alfalfa–grass samples, and replicate scans from multiple instruments can be combined to develop a single calibration that will perform with equal efficiency across different instruments.