Estimation of Botanical Composition in Alfalfa/Ryegrass Mixtures by Near Infrared Spectroscopy

Abstract

AbstractDetermining the botanical composition of forage samples by near infrared reflectance spectroscopy (NIRS) can save time and expense. The purpose of this study was to test the efficacy of NIRS technology for predicting botanical composition of forage samples. Estimating the number of samples required for reliable calibration equations, determining if pooling calibration samples across locations and years (broad‐based) would improve accuracy and precision, and comparing NIRS errors with experimental and sampling errors were primary objectives. Samples were collected from four different alfalfa (Medicago sativa L.) and ryegrass (Lolium perenne L.) management trials. Calibration samples were hand‐separated into alfalfa, ryegrass, and weed components and each component was dried and weighed. Alfalfa and ryegrass components were then remixed, ground, and processed through a scanning monochromator. Calibration equations for percent alfalfa were chosen using statistical analysis, Random subsets of samples were selected for validation of calibration equations using R2, bias, and standard error of performance (SEP). Broad‐based calibration equations required as little as 200 samples (approximately 5%) to accurately predict botanical composition of alfalfa/ryegrass mixtures with R2 of 0.95, SEP of 6.5%, and bias of −0.3%. Calibration equations generated from each of the four studies separately only marginally improved SEP and slightly increased bias. When botanical composition was estimated on samples from locations not included in the calibration data set, large increases in SEP and bias were observed. Errors associated with NIRS estimations were often one‐half as large as the experimental and sampling errors.