Multivariate analyses of multilevel processes in dairy herds with automated milking systems
Professor Carsten Enevoldsen, Dept. of Large Animal Sciences (grant holder).
Main supervisor Ilka Klaas, Dept. of Large Animal Sciences.
PhD student .
Background and context
Automated Milking Systems (AMS or robots) have been implemented in about 900 Danish herds during the latest decade (1). They produce huge amounts of measurements. In addition, farmers are constantly offered new automatic measurement tools for various management purposes. The time-series data that can be available from these AMS herds are collected at several organizational levels (udder quarter, udder, cow, group of cows, herd, producer of milking system, and veterinary practice) with variable time intervals. Feed-back mechanisms across organizational levels are common. Decision support tools for efficient utilization of these complex data are still very immature. Some producers of measurement devices have applied huge R&D efforts (2) but the efforts seem to be restricted to a rather narrow aspect of the production process and, obviously, centered on that particular product/measurement device. The measurement devices usually focus much on supporting health management (e.g. early detection of disease). Consequently, the local veterinarian will be deeply involved in using the output from the devices. Numerous products are on the market and the output for decision support is often poorly documented (3). Consequently, the veterinary consultant is often very frustrated about how to use the system and how to weigh the information from the AMS with other information from other organizational levels. Basically, the consultant needs a framework where she/he is able to assess the usefulness of existing and new measurement tools in a specific context. The milk production system in the AMS context can be characterized as a complex multilevel dynamic system with numerous feed-back mechanisms and constraints. There are several types of outcome variables (interval, ordinal, dichotomous, counts, time-to-event). The organizational structure is not strictly hierarchical. Very advanced statistical tools have been developed and applied for many years in the cattle industry; in particular mixed models for animal breeding. Multi-process Kalman filters and hierarchical Markov processes are examples of approaches in herd management science (for monitoring and replacement, respectively). Latent class analysis is used for evaluation of diagnostic tests. However, we need a joint framework that handles multiple organizational levels and utilizes the large amount of data available from the AMS. The concepts and tools from chemometric analysis appear to offer promising solutions to our problems.
The research question
How to assess the usefulness (added value) of new and existing automated data measurements at multiple organizational levels in the dairy herd?
The scientific support and expected profile of the candidate
The scientific platform of the project will be established by bringing together the following competencies: The Danish academic training of specialized cattle veterinarians and Ph.D.s in the field, which is delivered by the Herd Health Management field (Cattle Health) at LIFE and represented by Professor Carsten Enevoldsen and Associate Professor Ilka Klaas. Diagnostic test evaluation by means of Latent Class Analysis (LCA) is represented by Associate Professor Nils Toft. Herd Management Science is represented by Professor Anders Ringgaard Kristensen. We expect CHANCE-competencies will bring new solutions to these fields that will provide better decision making in complex systems in a major food-producing industry.