Prediction of test-day body weight from dairy cow characteristics and milk spectra

The knowledge of individual body weight (BW) is a management key in terms of feed efficiency and to assess the environmental footprint of dairy production. From 6 farms, BW were measured on 735 Holstein cows. Daily milk samples were collected on these weighed cows and analysed by mid-infrared spectrometry. The stage and number of lactation were also collated. A spectral cleaning was conducted by calculating GH distances from 17 principal components. Spectra with a GH greater than 3 were discarded. The final dataset contained 720 records. Predicting equations were based on Partial Least Squares regressions. Cross-validation coefficient of determination (R2cv) and root mean square error (RMSEPcv) of the equation including only spectral data were of 0.19 and 65 kg. Then, days in milk, month of test and lactation stage were added. The obtained R2cv and RMSEPcv increased (0.43 and 54 kg). The part of the information derived from the spectral data was equal to 6%. By adding the daily milk yield, the BW prediction was slightly improved and showed a R2cv of 0.45 and a RMSEPcv of 53 kg. The use of Legendre Polynomials to regress the spectral data following the day in milk did not improve the predictability. By deleting samples showing a squared residual higher than its mean + 3 times of its standard deviation, the final equation included 668 samples (93% of the initial set) and had a R2cv of 0.58 and RMSEPcv of 42 kg. A herd cross-validation was then performed to assess the robustness of the developed equation. RMSEPv ranged from 40 to 58 kg. This preliminary study showed the potentiality to predict an indicator of body weight. As this prediction uses easy to record explicative variables and if a larger validation confirmed the obtained results, this prediction equation could be used to develop large scale study about feed efficiency. Moreover, this method allows to consider the past information if spectral data are available.