Feed intake prediction models
|
Francisco Maroto-Molina, Spain, co-ordinator
g02mamof@uco.es There is a contact form below Partner countries: Estonia Italy The Netherlands Denmark Switzerland Open to all COST countries |
Once you have read about the project, if you would like to join click on the button:
|
Members of the group have access to the Discussion Blog once they are logged in:
|
description
The voluntary feed intake or dry matter intake of dairy cows is very important because adequate nutrient intake facilitates milk production, but also is essential to maintain cow health. In addition, a reduced dry matter intake often announces upcoming health disorders in the days before other clinical signs and thereby facilitates early detection. Moreover, feed intake data facilitates the nutritionally and economically accurate formulation of rations and, together with milk yield, can be used to estimate feed efficiency of an individual cow. In recent years, several authors have published models for the prediction of feed intake of dairy cows. Most of them used performance data (body weight, milk yield, etc) as dependent variables. Some of them are widely used, e.g. the NRC equation, but they are not applicable to an individual cow for online decisionmaking. Low-cost feeding-behavior sensors are expected to be available for commercial use on dairy farms in the near future. It will allow improving current prediction models by including feeding time, feeding rate and other feeding behavior data. The aim of this project is to explore the opportunities and challenges of developing feed intake models for the individual dairy cow that include feeding behavior data. To this end, we will use a meta-analysis approach, i.e. we will merge and analyze data from different sources and originally recorded with different purposes. There are several research facilities in Europe that allow the automatic recording of feed intake and feeding behavior data. Most of them use the Roughage Intake Control (RIC) system (Hokofarm Group). These facilities are continuously recording feed intake data, which are normally used in nutritional studies (comparison of diets, effects of particle size, etc). Besides, these studies usually record other data, e.g. body weight, milk production, parity, health status, etc. All this information, put together, can be used to build robust models for the prediction of feed intake and, as stated before, this is the main goal of this project. It should be noted that our approach is in line with the recommendations of the Directive 2003/98/CE on the reuse of public sector information. The group plans to meet twice: for a planning meeting and for a meeting discussing results, but the members will chat on Skype in between. In addition, one or more members of the group may apply for an STSM.
relationship to deliverables
|
D3: Feeding behavior assessment. DairyCare shall deliver one or more novel technologies for automated monitoring of feeding behavior. We will not work in the development of a novel technology itself, but we will work in some important issues of novel technologies, mainly feasibility and utilities. Among others, we will explore the relevance of recording feeding behavior data and using them to predict feed intake. We will address the next questions: Are feeding behavior data useful for predicting feed intake? Can additional data improve the predictability or accuracy of feed intake prediction models? Which data?
D5: Decision support models. DairyCare shall deliver a conceptualized model for integration of sensor data into a decision support system for dairy farmers and this is the core of our project, as we will work in prediction models using different data (different sensors). Moreover, the main goal of the prediction of individual feed intake is to make management decisions for the individual dairy cow. These decisions can be related to different fields: early detection of health problems, daily ration adjustment, e.g. by computerized concentrate feeders, identification of efficient cows, which is also related to the environmental footprint of cows, etc. Some important issues will be addressed: What is the precision required for each decision? And the accuracy? What is the time resolution we need? Which variables can we include in the prediction for each use of the predicted data (in order to be mathematically and statistically correct)? |
|
cross-disciplinarity
|
WG1: welfare biomarkers. Different health disorders cause specific effects, but especially in case of ketosis and other metabolic disorders, a decrease in feed intake often appears several days before detection by the farm staff. Lameness has also been reported to have an important effect on feed intake and feeding behavior of the individual cow.
WG2: activity. Most of the members of the group belong to this WG and the activities to be carried out in this project are closely related to the key task “behavioral monitoring based on milking records and feed station record” and to the key target “feeding behavior”. WG3: system-level welfare management. This project will integrate data and develop decision support systems. The team is quite multidisciplinary. Three of the partners (50%) define themselves as animal scientists, so they have a deep knowledge of the science behind feeding and intake. Besides, they are accustomed to handle feed intake data coming from RIC systems, as their institutions have research facilities including these systems. One of these three is a specialist in energy balance during the transition period (late dry period and early lactation) which has been reported as a period during which the prediction of feed intake is especially difficult, and models have to be adjusted. One partner is an ethologist, which is of major importance when working with behavior data. This person will share with the team her knowledge about factors affecting feeding behavior, methodologies for calculating behavior bouts, e.g. meals, etc. The other two members of the team are focused in data science, which includes prediction models, although they have a background in animal science. One of them has experience in outlier detection and management, which is important because feeding behavior data automatically recorded use to be dirty (nonfeeding visits to the trough, recording errors, etc). The other person is a specialist in managing data coming from different type of sensors. |
|
stsm resources
|
This project does not need too many resources, in terms of research facilities, laboratories, etc because we are using existing historical data. These data come from several research facilities across Europe, which is a guarantee of variability and robustness of the results, and include experiments with a very different number of cows and days. There are experiments with few cows (up to ten) during few days (some weeks) and experiments including hundreds of cows during the whole lactation. When necessary, specific data-sharing agreements and IP rights will be established. Anyway, the main resource of this project is the expertise of the different researchers involved in it.
|
|
contact the co-ordinator
|
You may use this Contact Form to send an enquiry directly to Francisco Maroto-Molina
|
|
