Hide-And-Seek in The Norwegian Forest

What happens when you want to watch something, but cannot, due to difficult conditions (shy animals, rocky or steep areas, far away from roads …)? You use tools! In Norway, farmers have a limited number of fields they can use to produce food, both for humans and for livestock. In the summer, fields must be used to produce food for the long winter. Farmers do this by sending their animals out into the forest during the summer growing season (Figure 1). The cows get some sun and fresh air, eat grass and other plants, and their calves grow big and strong!

However, the forest can contain very dense vegetation and rough terrain, and the cows are alone—no shepherd or guard dogs are with them. After not seeing humans for months, the cows can also become shy, making it difficult to approach them. Other difficulties these cows face can include hot and cold weather, diseases, accidents, and even carnivores! Very few researchers have studied how cows act in hard-to-reach, remote areas [1]. For farmers, it is very important to know if negative things happen to their cows that can affect the cows’ health. Happy cows are the goal for everyone! Scientists and engineers have worked together to build tools to help both researchers and farmers watch the animals from back home [2, 3].

Different Tools Used to See the Invisible

The cows we studied were equipped with collars (Figure 2) that contain small devices that allow us to watch them without actually being with them in the forest. The collars collect information using sensors and send it back to researchers in the form of numbers. One sensor is a GPS device, just like the one in your phone or in your parent’s cars. GPS can be very useful for locating cows when they are far away in the forest, and it can tell us where the cows go every day and night. Another sensor is called a movement sensor (or accelerometer). It records how the cow is moving its head and neck, whether it is moving side-to-side, forward/backwards, or up and down. Your phone uses a similar technology to count your steps or to automatically switch the way your screen is turned when you turn your phone.

The movement sensors can detect even the tiniest movements up to 10 times per second. Each time, the collar stores three numbers, one for each direction the body is moving (forward/backwards, side-to-side, and up/down). The movement data comes out of the sensor as long lists of numbers for each direction. Finally, the collar contains a radio that communicates to a satellite in space, which can send all of these numbers back to us in the lab (or to the farmer in his house) so we can analyze the data.

What Do These Numbers Mean?

Cows have many different body movements, also called behaviors, which include walking, eating, and sleeping. Every behavior has its own unique combination of the three numbers the collars records, and this is called a movement signature. We must figure out what the list of numbers returned by the collar means! But how do we even do this? In movies, solving a secret code requires a “key” that connects the secret signs to actual letters. In the same way, solving movement sensor data also requires a key that connects different combinations of numbers to specific behaviors. Our quest was to create this key for the forest cow movements. The first step of this quest was going into the forest to film the cows (Figure 1)!

After equipping our herds with our special collars, we snuck into the forest and filmed some of our cows, catching them doing many behaviors in many different situations. We wanted to film many situations: different cows, different breeds of cows, different times of the day, and different types of forests. Capturing this variety is important because cows can move in unique ways, and we wanted to understand how differences in cow movements would show up in the data from the collars. So, we gathered 41 h of videos and carefully watched all of it, writing down what the cows were doing (laying, standing, walking, feeding, fighting, running, etc.). We labeled a total of 42 behaviors [2]. What was the next stop on our quest? Matching our behaviors with movement data. We could do this because both our video and our collar data were linked to the date and time when the numbers were recorded.

This brought us to the final step of our quest to find the movement signatures: finding the combination of numbers linked to each behavior. We needed some help for this, as we had millions of rows of numbers and behaviors, and it would have taken us forever using just our own brains! So, we taught a computer how to do the job of creating the “key” to solve the mystery of the collar data (Figure 3). This process is called machine learning. Just like you learn from doing things over and over, computers can also learn from lots of examples. They can figure out patterns and make decisions on their own. It is like a computer becoming really good at a chess game by playing it a lot—but with all kinds of stuff, like pictures, numbers, and words! Our job was to give the computer many examples of behaviors and the numbers that are connected to those behaviors, and then to test whether the computer could recognize special patterns that are repeated. Once this was done, we taught our program to run through all of our data, to assign a movement signature to a behavior. We could also test the computer by checking whether it could recognize what the cows were doing if we gave it movement data it had never seen before.

What Were the Cows Doing in the Forest?

What we ultimately wanted was to be able to use the numbers we got from the collars to say, for example: “These cows were busy eating from 9 a.m. until 11 a.m., then they rested for a bit, and then…”. We were absolutely able to do that! The computer could tell us whether the cows were walking, resting, or feeding. This had been done before, but only on cows that stayed in the safe fields around their farms, so we expected that simple behaviors like these would be recognized. But, to our surprise, the computer could recognize up to 20 behaviors (Figure 2) [2]! Some of those included social behaviors, such as “our cows are bumping their heads against each other”, “our cows are licking each other”, “our mamma cows are feeding their calves”. Recognizing these behaviors was new, and very exciting for our future projects. This kind of data can also help farmers take better care of their animals, keeping them happy while they are in the forest.

Importance of Spying on Cows

Cows are important for meat, milk, leather, soap, and many more products that humans use every day. Because the wellbeing of animals in our care is very important, we must make sure we are not causing cows unnecessary stress by releasing them into nature, even for a limited time. With the tools we created, we can identify what forest cows are doing without ever disturbing them, all with the help of a collar. In addition to making sure that the cows seem happy and are acting normally, the data from collars can help us study how the cows “chat” and hang out with each other, how different herds communicate, or how mothers and calves behave. It could also give us more information about how cows respond to carnivores in the wild, or how various conditions (diseases, insects, weather) impact their energy levels. On top of this, we might be able to identify a cow that is sick or injured, which would help the farmer take better care of his entire herd.

Glossary

Livestock: ↑ Animals like cows and sheep that farmers raise for things like milk and meat.

Carnivores: ↑ Animals that eat other animals to stay alive (for example: lions, tigers, wolves).

Remote Area: ↑ A far away and isolated place that is not easy to reach.

Sensors: ↑ Just like our senses help us understand the world, sensors help machines understand what is going on. For example, GPS sensors reveal locations, and movement sensors (accelerometers) reveal different movement.

Behavior: ↑ How someone or something behaves. Just like how you might smile when you are happy or run when you are excited, animals have behaviors that can tell us about them.

Movement Signature: ↑ A movement signature is like a picture that shows how something moves in three different directions. It helps us figure out and understand how things move by looking at numbers.

Machine Learning: ↑ Just like you learn from doing things over and over, computers learn from lots of examples. They can figure out patterns and make decisions on their own. (see here).

Acknowledgments

This research was part of the CarniForeGraze project which was funded by the Norwegian Research Council (project number 302674).

Conflict of Interest

Author SW was employed by Ökologische Schutzstation Steinhuder Meer e.V.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.