Evaluating dairy cow welfare with sensors

At the moment I am, together others in the Organizing Committee, working on the last details for the First International Precision Dairy Farming Conference, that we are going to organize from 21-23 June in Leeuwarden, the Netherlands.

Worldwide, there is quite some attention for precision technology. One of the interesting aspects of precision technology is that their use can move beyond the farm yard. The information gathered by some of these sensor system can be very useful along the value chain. An example of such use is animal welfare. Our consumers more and more demand not only a fresh and healthy dairy product, but also a dairy product that is produced by cows taken well care of. Large food companies start bringing programs in place where improved welfare is more and more demanded. Some years ago I talked to people of FrieslandCampina and they indicated that these large customers demand prove for the product specifications. How do you prove the welfare of cows?  Of course, a large majority of farmers care for their animals, but we all know that some farmers are not taking very good care of their animals. How do you know as dairy industry. Well, maybe sensor technology can be used for that.

For instance, an important indicator of dairy cow welfare is the time they are lying. Cows need lying time. Proper amounts of lying time are a sign that there is no overstocking (in terms of cows per cubicle) and that the comfort of the cubicles is good etc. 3D accelerometers (those are the sensors we all have in our mobile phones and that detect in what direction you move your phone) are used and successfully marketed to detect oestrus. However, these sensors can also be used to evaluate the lying time of cattle. If a leg is horizontal, the cow is probably lying, but not always. She might be scratching for instance. Akke Kok, a PhD student of whom I am one of the supervisors, is using 3D accelerometer sensors to evaluate lying time of dairy cattle and she wanted to validate whether the lying bouts as recorded by the IceQube sensor were correct. Moreover, she wanted to determine a threshold to define whether a horizontal leg means lying or not. That work has recently been published in the Journal of Dairy Science.

Now to evaluate the accuracy of sensor systems, normally you have to create a golden standard: sit down in the barn and observe all cows while writing down when cows are lying or not. Then you compare the sensor readings with this gold standard. Akke took an interesting different approach. She equipped both hind legs of 28 lactating dairy cows with an IceQube sensor for a period of 6 days and used the two sensors as each other’s validation. Classification of lying bout records as true (actual LB) or false (recorded while standing) was based on three assumptions. First, all standing records were assumed to occur whilst standing. Second, false lying bout records due to short leg movements could not occur in both hind legs simultaneously. Third, true lying bouts only occurred if the records of the paired sensors coincided. Based on maximum accuracy, a minimum duration of lying bout records of 33 seconds was determined, with an accuracy of 0.992, a sensitivity of 0.993 and a specificity of 0.977. Applying the threshold hardly affected estimates of daily lying time, but improved estimates of frequency and mean duration of LBs for individual cows.

All in all, an interesting application to use sensors as their own gold standard. Now let’s see how and when dairy processors become interested in these types of data to guarantee welfare of animals throughout the value chain.