Comparison between single and mixed-species NIRS databases’ accuracy of dairy fiber feed value detection

Abstract

Abstract Near Infrared Reflectance Spectroscopy (NIRS) accuracy is affected by its database. However, our previously developed database for dairy cattle dietary fiber feed (DFF) showed low accuracy for complex organic substance detection due to mixed-species used in the database. This paper aimed to compare single and mixed-species NIRS database accuracy in predicting DFF nutrient contents. In the mixed database, five feeds from three areas of dairy cattle farming were sampled. In the single database, thirty Napier grass from 30 areas were collected. Samples were analyzed chemo-metrically for their nutrient contents. Spectra of each sample were collected three times (two spectra for calibration and a spectrum for validation) using FT-NIRS Spectrometer Solid Cell. Calibration and validation models were carried out using the Partial Least Squares (PLS) regression. For external validation, seven independent Napier grass samples were tested. The result showed that the single species NIRS database developed using Napier grass was less accurate than mixed-species DFF due to huge nutrient content variations between varieties of Napier grass. It is concluded that database accuracy developed from mixed dietary fiber feed were more accurate in comparison to single species and suggested to used combination of mixed and single database for more accurate DFF prediction.