Prediction error and repeatability of near infrared reflectance spectroscopy applied to faeces samples in order to predict voluntary intake and digestibility of forages by ruminants

The study examined the prediction error of near infrared reflectance spectroscopy of faeces (FNIRS) in estimations of in vivo organic matter digestibility (OMD) and dry and organic matter voluntary intake (DMVI and OMVI) of forages by ruminants in established calibration datasets from our laboratories. It also examined the repeatability of the FNIRS measurements of these parameters and the proportion of grass and clover in ingested forage. The prediction error of NIRS calibrations depends on the accuracy and precision of reference data. In this study, the variability in reference data for OMD, DMVI and OMVI was less than 10% of the mean. Correction for the error in the reference method almost halved the standard error of prediction (SEP) for OMD to 0.0155. For DMVI and OMVI, the corrected SEP was 8-9 g/kg BW0.75, similar to the apparent SEP. These results suggested that the FNIRS calibrations were precise enough to predict OMD adequately, but probably not DMVI or OMVI. The repeatability of FNIRS spectra and predictions assessed by repeated measurements of the same sample was satisfactory for all tested parameters. For OMD, DMVI, OMVI and the grass and clover proportion of the ingested forage, prediction repeatability was lower than, or similar to, the standard error of cross-validation (SECV) of the FNIRS calibration. The study showed that FNIRS could be a reliable method for predicting the OMD of ruminants, but the prediction of voluntary intake with acceptable error was less satisfactory because of uncertainties in the FNIRS calibration models. (C) 2015 Elsevier B.V. All rights reserved.