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

摘要

A dairy cow’s lifetime resilience and her ability torecalve gain importance on dairy farms, as they affectall aspects of the sustainability of the dairy industry.Many modern farms today have milk meters and activitysensors that accurately measure yield and activityat a high frequency for monitoring purposes. Wehypothesized that these same sensors can be used forprecision phenotyping of complex traits such as lifetimeresilience or productive life span. The objective of thisstudy was to investigate whether lifetime resilience andproductive life span of dairy cows can be predicted usingsensor-derived proxies of first-parity sensor data.We used a data set from 27 Belgian and British dairyfarms with an automated milking system containingat least 5 yr of successive measurements. All of thesefarms had milk meter data available, and 13 of thesefarms were also equipped with activity sensors. Thissubset was used to investigate the added value of activitymeters to improve the model’s prediction accuracy.To rank cows for lifetime resilience, a score was attributedto each cow based on her number of calvings, her305-d milk yield, her age at first calving, her calvingintervals, and the DIM at the moment of culling, takingher entire lifetime into account. Next, this lifetimeresilience score was used to rank the cows within theirherd, resulting in a lifetime resilience ranking. Based onthis ranking, cows were classified in a low (last third),moderate (middle third), or high (first third) resiliencecategory within farm. In total, 45 biologically soundsensor features were defined from the time series data,including measures of variability, lactation curve shape,milk yield perturbations, activity spikes indicating estrousevents, and activity dynamics representing healthevents (e.g., drops in daily activity). These features,calculated on first-lactation data, were used to predictthe lifetime resilience rank and, thus, to predict theclassification within the herd (low, moderate, or high).Using a specific linear regression model progressivelyincluding features stepwise selected at farm level (cutoffP-value of 0.2), classification performances werebetween 35.9 and 70.0% (46.7 ± 8.0, mean ± SD) formilk yield features only, and between 46.7 and 84.0%(55.5 ± 12.1, mean ± SD) for lactation and activityfeatures together. This is, respectively, 13.7 and 22.2%higher than what random classification would give.Moreover, using these individual farm models, only 3.5and 2.3% of cows were classified high when they wereactually low, or vice versa, whereas respectively 91.8and 94.1% of wrongly classified animals were predictedin an adjacent category. The sensor features retainedin the prediction equation of the individual farms differedacross farms, which demonstrates the variabilityin culling and management strategies across farms andwithin farms over time. This lack of a common modelstructure across farms suggests the need to considerlocal (and evidence-based) culling management ruleswhen developing decision support tools for dairy farms.With this study we showed the potential of precisionphenotyping of complex traits based on biologicallymeaningful features derived from readily available sensordata. We conclude that first-lactation milk andactivity sensor data have the potential to predict cows’lifetime resilience rankings within farms but that consistencybetween farms is currently lacking.