Productive life span and resilience rank can be predicted from on-farm first-parity sensor time series but not using a common equation across farms.

A dairy cow's lifetime resilience and her ability to recalve gain importance on dairy farms, as they affect all aspects of the sustainability of the dairy industry. Many modern farms today have milk meters and activity sensors that accurately measure yield and activity at a high frequency for monitoring purposes. We hypothesized that these same sensors can be used for precision phenotyping of complex traits such as lifetime resilience or productive life span. The objective of this study was to investigate whether lifetime resilience and productive life span of dairy cows can be predicted using sensor-derived proxies of first-parity sensor data. We used a data set from 27 Belgian and British dairy farms with an automated milking system containing at least 5 yr of successive measurements. All of these farms had milk meter data available, and 13 of these farms were also equipped with activity sensors. This subset was used to investigate the added value of activity meters to improve the model's prediction accuracy. To rank cows for lifetime resilience, a score was attributed to each cow based on her number of calvings, her 305-d milk yield, her age at first calving, her calving intervals, and the DIM at the moment of culling, taking her entire lifetime into account. Next, this lifetime resilience score was used to rank the cows within their herd, resulting in a lifetime resilience ranking. Based on this ranking, cows were classified in a low (last third), moderate (middle third), or high (first third) resilience category within farm. In total, 45 biologically sound sensor features were defined from the time series data, including measures of variability, lactation curve shape, milk yield perturbations, activity spikes indicating estrous events, and activity dynamics representing health events (e.g., drops in daily activity). These features, calculated on first-lactation data, were used to predict the lifetime resilience rank and, thus, to predict the classification within the herd (low, moderate, or high). Using a specific linear regression model progressively including features stepwise selected at farm level (cutoff P-value of 0.2), classification performances were between 35.9 and 70.0% (46.7 ± 8.0, mean ± SD) for milk yield features only, and between 46.7 and 84.0% (55.5 ± 12.1, mean ± SD) for lactation and activity features together. This is, respectively, 13.7 and 22.2% higher than what random classification would give. Moreover, using these individual farm models, only 3.5 and 2.3% of cows were classified high when they were actually low, or vice versa, whereas respectively 91.8 and 94.1% of wrongly classified animals were predicted in an adjacent category. The sensor features retained in the prediction equation of the individual farms differed across farms, which demonstrates the variability in culling and management strategies across farms and within farms over time. This lack of a common model structure across farms suggests the need to consider local (and evidence-based) culling management rules when developing decision support tools for dairy farms. With this study we showed the potential of precision phenotyping of complex traits based on biologically meaningful features derived from readily available sensor data. We conclude that first-lactation milk and activity sensor data have the potential to predict cows' lifetime resilience rankings within farms but that consistency between farms is currently lacking.