This paper deals with the problem of how to utilize a large calibration set with 10 different analytes in order to make the best predictions possible on a routine basis. Ten different strategies of using the data set were studied with the use of numbers of principal components ranging from 4 to 12. We found positive effects of scatter correction for most of the analytes. On average, the local regression methods were superior to the others. The optimum number of samples for local regression seems to be between 50 and 100. The largest reduction in root mean square error of prediction (RMSEP), in comparison to results for the traditional method, was found on scatter-corrected spectra and a proposed local calibration with 50 calibration samples. The gain in RMSEP for neutral detergent fiber (NDF), acid detergent fiber (ADF), and crude fiber was about 25% and for protein and <i>in vitro</i> digestible dry matter digestibility (IVDMD) about 10%, compared to results for the traditional universal calibration method.
You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an OSA member, or as an authorized user of your institution.
Contact your librarian or system administrator
Login to access OSA Member Subscription