Optimal replacement policies for dairy cows based on daily yield measurements

Discipline: lactation management; Keywords: reproduction, model, MDP, daily information, economic implications.

Management decisions including if and when to replace a cow will be facilitated if models can be developed that integrate all variables affecting these decisions. As modern-day computers can effectively handle large sets of data, even with daily inputs, a modelling exercise was done by L.R. Nielsen and co-workers, which was reported in the Journal of Dairy Science, Volume 93 of 2010, pages 79 to 92, with the title: Optimal replacement policies for dairy cows based on daily yield measurements.

The decision of when to replace a cow with a heifer is affected by many factors such as the cow’s current and future milk yield, illness, availability of replacement heifers, prices, reproduction and the goals of the producer. The problem is sequential in nature: at a specific time the decision of when to replace the cow or not is based on known information and expectations about the future. At the next decision stage, updated information is available and the decision choice is re-evaluated. Thus, a model that can integrate sequential data with all variables taken into account will be valuable to the dairy farmer, the shorter the intervals (say daily instead of monthly) the better.

The techniques used are based on so-called Markov decision processes (MDP), but thus far the length of the period is unacceptable for effective decision making. In this article Nielsen and co-workers have extended the MDP model to use daily information, such as daily milk yield measurements. These were incorporated by means of the so-called Bayesian prediction model that predicts performance of the cow based on these measurements, but modified every day as new information becomes available. The economic implications of every prediction accordingly come into play. The model was theoretically developed and then tested in specific herds against known lactation curves. The outcomes were highly satisfactory and the authors are confident that the basis has been laid that can be further improved on in future.

Bottom line: In the modern day dairy farm operation milk yield recording and management facilitating measurements and observations are very much part of the daily activities. To introduce a model with the capacity to integrate all relevant information to improve decisions on cow replacement will significantly improve the economic output of the operation. It is recommended that the assistance of an animal scientist and economist be obtained to incorporate such a model into the computer management decision system of the dairy farm.