Dairy cattle

Genetic evaluation of milk production using robotic milking data

---

Abstract

Robotic milking systems (RMS) have the potential to deliver accurate genetic evaluations by leveraging high-resolution daily milk yield data. The goal of this study was to develop the prototype for a genetic evaluation of milk production using RMS data. The dataset comprised 26,261,671 daily milk yield records from 30,603 lactating daughters of 4,018 sires, collected between 2008 and 2025 from 170 farms in Italy (6.7M), the UK (6.6M), France (6.5M), Canada (3.9M), the US (1.2M), the Netherlands (0.8M), Brazil (0.4M), and Norway (0.1M). All farms used Lely milking robots. Three modeling approaches were evaluated. The first method applied a farm-level multiple linear regression adjusting for lactation number, year, days in milk (DIM), and day of the year to impute missing daily yields. DIM and day effects were decomposed into orthogonal polynomials to capture non-linear associations. Observed and imputed yields were summed to obtain total 305-day milk production per cow, and predictions from all farms were analyzed using a sire model with farm and age at first calving as fixed effects and sire as random effect. The second method treated raw daily yields as the trait without imputing missing records, using an extended test day model grouping milking dates by Herd-Year-Season (HYS). Lactation number, DIM, and HYS were fixed effects, with DIM decomposed into 14 orthogonal polynomials. Estimated breeding values (EBV) were expressed in kg/day and scaled to 305-day lactations for comparison. The third method used the same trait as the second method but employed a random regression model (RRM), allowing sire EBV to regress across p orthogonal polynomials of DIM. In the first method, regression models for imputation achieved an average adjusted R2 of 0.71. A total of 81,624 estimated 305-day milk productions from 151 farms passed quality control. Based on 11,293 sires, EBV accuracies averaged 0.41 ± 0.36 and 0.56 ± 0.27 for methods 1 and 2, respectively (mean ± SD). The EBV estimates from both methods were equivalent when accuracies exceeded 0.88. RRM generated sire-specific polynomial curves describing genetic ability across DIM. Heritability was higher beyond 150 DIM (0.38 vs 0.51). Also, sires with similar average EBV from the first two methods often showed distinctive curves of genetic ability across lactation. Overall, this study suggests that daily milking yields from RMS have potential to improve the accuracy of genetic evaluations, though accounting for serial correlation among consecutive records remains a necessary refinement. The RRM revealed that genetic selection could target specific phases of the lactation curve, offering opportunities beyond what single-trait 305-day evaluations can capture.

Keywords: 2026

How to Cite:

Chasco, J., Nani, J. & Peà±agaricano, F., (2026) “Genetic evaluation of milk production using robotic milking data”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2284033. doi: https://doi.org/10.31274/wcgalp.23546

Rights: 1