Near Infrared Reflectance Analysis with Single and Multiproduct Calibrations

Abstract

Single‐product calibrations for near infrared (NIR) spectrophotometers are normally developed for each type of material to achieve a high level of prediction accuracy. Developing multiproduct calibrations with thousands of samples was not practical with personal computers until recently. Advancements in software now make it possible to include thousands of samples into multiproduct calibrations. This study was conducted to evaluate the accuracy of a multiproduct calibration consisting of forage samples. Large single‐product files of hay and fresh forage, haylage, and small grain silage spectra were available for this study. Every sixth sample from each file was reserved to test calibration accuracy. Single‐product calibrations were developed for each product and a multiproduct calibration was developed for all four products. There was very little difference between the accuracy of the multiproduct calibration containing all samples and the single‐product calibrations. In a second study, a multiproduct calibration developed from 756 haylage and small grain silage samples was used to predict the quality of 100 hay samples. The prediction error was found to be unsatisfactory, so 50, 100, 166, and 500 hay samples were added to the multiproduct calibration file to see if the hay prediction errors could be improved. The expanded multiproduct calibration was more accurate than a custom single‐product hay calibration when <150 hay samples were available to develop a custom calibration or expand the multiproduct calibration. This suggests that if only a few samples are available, it may be more cost‐effective to expand a similar product library and recalibrate than to gather a large number of samples for a new calibration.