Faecal-NIRS for predicting digestibility and intake in cattle: efficacy of two calibration strategy
Résumé
Diet digestibility and intake are animal traits difficult to measure in practice. Thus, several indirect methods including faecal visible/near infrared (VIS/NIR) spectroscopy, have been developed for predicting these traits. The development of this technology requires great number of samples, which are not always obtained in standardized conditions among experiments. Consequently, the lack of standardization could be associated to a loss of precision of VIS/NIR models. In order to test this hypothesis, we used 88 VIS/NIR spectra of faeces sampled on cattle beef and dairy cows from 3 different experiments (De la Torre et al., 2019; Herremans et al., 2019; Fanchone et al., 2013). Experiments included measurements of the individual organic matter digestibility (OMD) and daily dry matter intake (DI) of animals. Two calibration strategies were compared for predicting these measurements: (1) specific experiment calibration and (2) a global calibration procedure. Models were assessed by standard error of cross-validation (SECV) and coefficient of determination of cross-validation (R2CV). The SECV was separated into bias and SECV corrected by bias SECV(C). Bias and SECV(C) for both calibration strategies were compared using the procedure of Fearn (1996). Data sets gave value ranges of 0.60-0.78 for OMD and 53-181 g/kg body weight (BW)0.75 for DI. For OMD, the SECV and R2CV were 0.02 and 0.86 for both specific experiment calibration and global calibration procedure. For DI, the SECV and R2CV were 6.90 g/kg BW0.75 and 0.96 respectively, for the specific experiment calibration and 7.62 g/kg BW0.75 and 0.95, respectively, for the global calibration procedure. For both, OMD and DI, no significant differences between calibration strategies were found for bias and SECV(C). From the results of this preliminary study, we conclude that mixing data from experiments conducted in very different conditions does not degrade the precision of the obtained VIS/NIRS models to predict OMD and DI in cattle.