Pasture Sampling Requirement for near Infrared Reflectance Spectroscopy Estimates of Botanical Composition

Abstract

Near infrared reflectance spectroscopy (NIRS) has been used predict species composition and plant components in hay and pasture samples. Two measures of error, bias and standard error of prediction (SEP), are important to consider with respect to accuracy of NIRS predictions of laboratory values of samples. Our objectives were to determine how sample size affects the precision and accuracy of the NIRS estimate of the mean of pasture samples. In Exp. I, four grazed pastures of pure and mixed alfalfa (Medicago sativa L.) and tall fescue (Festuca arundinacea Schreb.) were sampled over 12‐d grazing period in July 1987. Samples were separated into percentage of tall fescue, alfalfa leaf, and alfalfa stem and used for development of NIRS calibration equations. The SEP was used to calculate boundary limits, which should contain the NIRS estimates for 1, 5, 8, or 10 pasture samples. Boundary limits decreased from 11.0 to 2.3% units as sample number increased. The increase in accuracy was inversely related to the square root of the sample size. Experiment 2 was similar to Exp. I, except all samples gathered on the first day of grazing within one of the pasture mixtures were used as a verification set for the NIPS calibration equations. Pasture means and standard deviations were similar for laboratory and NIPS values. We conclude that the pasture sample size needed to obtain an acceptable confidence interval of the pasture mean for botanical composition is a sufficient sample size for accurate and precise NIPS prediction of that mean.