Peri-estrus activity and rumination time and its application to estrus prediction: Evidence from dairy herds under organic grazing and low-input conventional production

Abstract

The objective of this research was to investigate the peri‑estrus changes in activity and rumination time and their use for estrus prediction using data collected by a collar-mounted accelerometer and rumination system (SCR Engineers Ltd., Netanya, Israel). Two experimental dairy herds (low-input conventional: CONV and organic: ORG) of the University of Minnesota West Central Research and Outreach Center, Morris, MN were monitored during the period from June, 2014 to August, 2017. The herds were seasonally bred, and the study comprised three summer breeding seasons (SUM: June to August) and four winter breeding seasons (WIN: December to February). In the winter, both herds were kept in an outwintering lot or a compost-bedded pack barn. In the summer the CONV herd was housed in a dry-lot, and the ORG herd on pasture. The breeding records of 1462 estrus events (849 CONV and 613 ORG) were used as estrus dates. When the day of estrus (day 0) was compared to the reference-period (day -3, -2, -1, +1, +2, +3, relative to the reference period), increased daily activity was observed for 96% of the estrus events (97% for CONV-SUM, 97% for CONV-WIN, 91% for ORG-SUM, and 95% for ORG-WIN), and a decreased daily rumination was observed for 82% of the estrus events (81% for CONV-SUM, 88% for CONV-WIN, 73% for ORG-SUM, and 86% for ORG-WIN). A logistic regression model was developed based on the deviation from a 7-day moving average, 7-day moving standard deviation, and the daily absolute maximum of 6-h-window cumulative change of the data. A successful threshold of alerting estrus (predicted probability > 0.70) was possible when sensitivity and specificity (%) of the model were 75 and 96 for CONV-SUM, 91 and 99 for CONV-WIN, 51 and 92 for ORG-SUM, and 80 and 99 for ORG-WIN. At a lower threshold (predicted probability > 0.30), the sensitivity and specificity (%) were 89 and 81 for CONV-SUM, 97 and 90 for CONV-WIN, 79 and 75 for ORG-SUM, and 90 and 90 for ORG-WIN. A range of trade-offs between sensitivity and specificity was characterized by receiver operating characteristic curves. Additionally, incorporating rumination time data to activity data alone did not improve the prediction performance substantially. In sum, this technology can help detect peri‑estrus changes and there is potential for optimizing prediction performance through improving prediction algorithms or refining alert thresholds.