Contribution of milk mid-infrared spectrum to improve the accuracy of test-day body weight predicted from stage, lactation number, month of test and milk yield

A regular and repeated recording of body weight (BW) is useful information for herd management. BW can bepredicted regularly from animal characteristics such as age, lactation number, or lactation stage. Those traits areunfortunately animal unspecific. Adding animal specific information, which can be easily obtained on a largescale, to the BW prediction would be of utmost importance. There are good scientific reasons to suspect linksbetween BW and animal specific characteristics, available in a repeated fashion, such as milk yield and milkcomposition. This study aimed to demonstrate the feasibility of predicting test-day BW from stage, lactationnumber, month of test, milk yield and mid-infrared spectra, representing milk composition. Five models weretested initially from 721 BW records collected in 6 herds: day in milk + number of lactation (equation 1a);equation1a + milk yield (equation 1b); only spectral data (equation 1c); equation 1c + equation 1a (equation2); equation 2 + milk yield (equation 3). Then 3 other equations included the same explicative variables, exceptthat the spectral data were regressed using second order Legendre Polynomials (PL) to take into account changesof spectral data within lactation. Equation 1a and 1b were built using linear regressions and equation 1c until 3were built using partial least square regressions. These 3 last equations had a higher number of factors. Adding ofMIR data in the equation increased of 7% the values of cross-validation R² (R²cv). Potential BW outliers werediscarded using a residual analysis based on equation 3. From 662 records, the following statistical parameterswere obtained: the calibration coefficient of determination (R²c) = 0.65, R²cv = 0.61, calibration root meansquared error of prediction (RMSEP)=38 kg, and RMSEPcv=40 kg. Low variation of R²c and RMSEPc valuesobtained from the herd validation confirmed the herd independence of predictions. However, large variabilitywas observed for RMSEPv (37 to 64 kg) suggesting the need to increase the dataset in order to improve therobustness of the equation. By applying the equations on a large spectral database, it was confirmed that theaddition of MIR data allows to better model the BW evolution within lactation. Based on these preliminaryresults, and if a larger validation confirms thesefindings, this approach could be used to develop equations thatare better able to assess BW throughout lactation(s), BW being an important element for management andselection tools.