Update on the ClearFarm platform development

bringing the data from the farm to the screen

After nearly two years of trials on farms in six European countries, collected data is now being thoroughly analysed to feed the algorithm behind the ClearFarm platform.

To find out more about this experience, we have talked to the leaders of the two work packages focusing on the validation in the pig and the dairy cattle value chain: Eddie Bokkers, associate professor at the Animal Production Systems group of Wageningen University & Research (WUR), and Elisabetta Canali, associate professor at the Department of Veterinary Medicine of the University of Milano.

ClearFarm: How would you describe the ClearFarm data collection experience? Have you had any setbacks? How did you overcome them?

Eddie Bokkers: I don’t think there were unusual setbacks. Animal experiments and data collection on farms always give some issues. I think the main setback we had was COVID-19. We had some slowdown on deliveries of some technical devices and some arrangements, but at the end, I think we’ve done pretty well.

Elisabetta Canali: The data collection was a positive experience, both in terms of the contribution of all the researchers and of the collaboration of the people in the dairy cattle farms involved with in the trials. The pandemic was the major setback and it meant only a slight delay.

CF: What kind of data is still being collected?

EB: In Denmark data is analysed of different trials with weaners and fattening pigs. In Spain, they are still collecting sensor data and working on analysing data of the first trials. In the Netherlands, we have now finalised two trials with fattening pigs, the trial with sows is still running. We are also analysing the data to see how we can use it in a proper way.

EC: We are collecting information about the behaviour of the cows from the sensors. And we’re also collecting data on animal-based indicators for each one of them. We are collecting the same type of data that during the pilots, to refine the algorithm and make sure that it works properly.

CF: What are the main differences between the data collecting in pig farms and dairy cattle farms? Are the differences related to the indicators, to the technology, to the analysis…?

EB: I think the main difference is that in dairy farming, there are many more sensors and these sensors are focused on measurements of individual animals. So far, in the pig production sector data collection is more focused on group level, or even farm level. So, we have less choice in sensors for data collection. But we do have some very interesting options. For example, feeding station data, which generated information about feeding patterns at the individual level. In Denmark and the Netherlands, we are collecting data of 3D cameras in fattening pigs, to estimate body weights. So, we have individual data especially related to body weight and to feed intake and feeding patterns in general, that, for example, can be used to offer real time information for the farmer.

EC: The main differences are related not to the type of data, but to the technology used to collect it. We use different kinds of sensors that can be placed on each cow, like collars. But in the case of pigs, due to their anatomy, this is not possible, so the researchers must apply other kind of observation techniques and technologies.

CF: Is all the generated data being used to develop the algorithm or does it need to pass a preselection? What kind of criteria has been applied to do so?

EB: It depends on the sensor. We have some sensors in Spain from which we only receive the output, and we don’t know how the algorithm behind it works. We can connect the output directly to the ClearFarm platform. But then we also have the feeding station data, which is raw data, and we have to build an algorithm ourselves first to provide useful information for the farmer and the platform.

There are many ways to look at these patterns: are they individual differences? Are they consistent over time? Is there a difference between males and females? between breeds? There are many options to look at, and that makes it also interesting for us to have a proper analysis, to understand the value of the different data.

EC: We are currently evaluating precisely what data will be used to develop the algorithm. Which, in any case, must be significant to evaluate the welfare of the dairy cows. This is the most important value of the indicators.

CF: How is the data being integrated to develop the algorithm? Some data was automatically generated and other was produced from observation or onsite health checks, for instance. Could you briefly describe the process from the farm to the screen?

EB: For play behaviour, for example, we collected a lot of data manually by video analysis. Similarly, we used video analysis for behavioural observations in the Netherlands. Both in the Netherlands and Spain, we conducted clinical observations manually. We use this data to validate the sensor data and whether it is useful for a platform. At the end, we only want to use sensor data for the platform. The manual collected data functions as background information to see the value of the sensor data for animal welfare assessment.

EC: We have different sources for the data and different types of data. For instance: for each cow we have information about their behaviours (eating rumination, lying) from sensors and we collected manually health and productive data. We also used some animal-based indicators to assess the welfare of the cows. We are preparing a huge file that contains all these data and it will be the starting point of our analysis to develop the algorithm. The most difficult thing is to put together and aggregate all the different indicators and data.