Predicting the quality of browse-containing diets fed to sheep using faecal near-infrared reflectance spectroscopy

The potential of data collected from past feeding trials to derive faecal near-infrared reflectance spectroscopy (fNIRS) calibrations for predicting the attributes of browse-containing sheep diets was examined. Reference data and faecal near-infrared spectrum pairs (n = 240) originated from five feeding trials involving 40 diets consisting of varying levels of fresh browse and oaten chaff. The fNIRS calibrations were developed to predict crude protein (CP), total phenolics (TP), total tannin (TT) and phosphorus (P) contents, protein precipitation capacity of tannin (PPC), in vivo digestibility of dry matter (DMD), organic matter (OMD) and crude protein (CPD) and in vitro OMD (IVOMD), metabolisable energy (ME) and short chain fatty acid production (eSCFA) in the diet. The precision of calibrations was evaluated by the coefficient of determination (R(c)(2)) and standard error (SEC) of calibration. The predictive ability of calibrations was evaluated by standard error of cross-validation (SECV), standard error of prediction (SEP), slope of the validation regression and the ratio of the standard deviation of the reference data to the SECV (RPD). For all fNIRS calibrations, R(c)(2) was >0.80 and SEC was close to the respective SECV. Slope of the validation regressions did not deviate from 1 for chemical attributes but deviated from 1 for functional attributes (except eSCFA). The RPD of DMD and OMD was <3, whereas the ratio was >3 for CP, TP, TT, PPC, P, CPD, IVOMD, ME and eSCFA calibrations. Data derived from the past feeding trials could be used to derive robust fNIRS calibrations to predict chemical attributes (CP, TP, TT, PPC, P) of browse-containing sheep diets. Although, fNIRS calibrations predicting dietary in vitro functional properties (digestibility and ME) were superior to those predicting in vivo functional properties, both were not so robust. Statistics of fNIRS calibrations derived using reference data originating from in vitro methods needs to be carefully interpreted.