Every Step Cow-nts: Big, Bovine Data

There are not many creatures I've ran into that I'm not a fan of (from a distance, at least.) When it comes to cows, they grew on me pretty quickly. I went to college in the rural, Central Washington town of Ellensburg. Outside of the University, the next largest industry is farming, predominately focusing on Timothy Hay (most of which ends up in the bellies of Japanese horses.) Scattered throughout the amber fields of grain are herds of cattle, and a few horses and sheep in between. After graduating, and to begin preparing for environmental work with the Peace Corps, I decided to take up an offer to work for a friend of mine, Joann, who had a small herd of grass-fed Hereford cows.

Hereford cattle on ARS' Fort Keogh Livestock and Range Research Labratory near Miles City, Montana. Image Credit: USDA, photo by Keith Weller
https://www.flickr.com/photos/usdagov/

There's a lot of things I learned to love about these animals: their emotional intelligence, the unique personalities they develop, and above all, their stubborn determination to do whatever it is they want at all times. I was constantly amazed how every mother knew the exact sound of her calves above the balling of all the others. Watching the little ones sneak under the movable fencing to get to the greener pastures always made me laugh. Oh and I learned that the best tool for reaching that hard to reach itchy spot is the mirrors and bumpers of a Ford truck, preferably with humans inside trying to take a break from the sun. I only spent a few months working with Joann and the Herefords, but it changed my outlook on cows, food, and our relationship with the environment for years to come.

Some members of Joann's herd of Hereford cattle waiting to move to greener pastures. Photo taken August 2016.

Joann had spent a lot of time with her cows. She had worked with cattle since she could walk, and had been working with Herefords in Ellensburg since 2006. Her knowledge, intuition, and skills were exceptional. She had tricks to make everything a bit easier, on us and the cows, and it almost seemed like she knew what they were thinking at any given time. Even with her years of experience and expertise, a time would come when Joann was mostly at the whims of nature: breeding and birthing. Cows only undergo estrus (their period of fertility) for relatively short periods, typically 12-18 hours. The intervals between estrus is irregular, and difficult to predict with conventional methods. Much to the frustration of cattle handlers, this estrus period typically occurs at night. All in all, this makes a recipe for a stressful, high-stakes guessing game. I'm sure Joann spent a lot of sleepless nights trying to come up with ways to make this process easier and more effective. Little did she know, a group of researchers in Japan were working on the same problem.

Bovine Biometrics

The story began on the Japanese island of Hokkaido, with a group of dairy farmers who, like Joann, were losing sleep over how to improve breeding successes in their cattle. When the group approached the Japanese IT and computing company Fujitsu to help solve this problem, the researchers began trying an interesting approach. They began to collect physical, biometric, and other data from the cows, and see if any useful patterns could be identified. When the project was complete, a surprisingly simple solution was found: pedometers. Yes, pedometers.

As it turns out, when a female cow is undergoing estrus, her physical activity levels begin to increase. Cows, like most animals with legs, apparently express this by taking more steps. So Fujitsu figured out that by measuring a cows steps using a fancy pedometer hooked up to data-tracking software, they could accurately identify when a cow was beginning her estrus cycle. This alone is pretty cool, and takes out a massive chunk of the guessing on the part of farmers. No more checking in every night at 2 in the morning, hooray! Amazingly, test research from farms in Japan saw their fertilization success rate jump from 44% to closer to 99%. That's a hell of an increase, essentially ensuring that every cow inseminated will result in a pregnancy. But the insights didn't stop there.

Graphic demonstrating the GYUHO SaaS, Akisai Food & Agricultural Cloud Service from Fujitsu
https://journal.jp.fujitsu.com/en/2016/04/08/01/

Along with accurately predicting the onset of estrus, the data being collected by researchers allowed for even more detailed predictions to be made. They claimed that inseminating the cow at different points of estrus (i.e. in the first four hours, last four hours, etc.) had a measurable impact on the gender ratio of the offspring. Maintaining a proper gender ratio in a herd is important, and can save a lot of time and resources when done right. Other options for affecting calf gender usually involves procuring 'sexed sperm', or sperm which has been sorted based on the presence/absence of X and Y chromosomes (this involves a complicated process known as Flow Cytometry, where cells are stained with fluorescent dye, then 'read' with a laser or other light source, and sorted.)

Furthermore, they claim to be able to predict at least 10 bovine diseases and disorders days or weeks earlier than standard diagnostic methods, purely based on data collected from their pedometers. Cows, like many people I know, change their habits when things aren't quite right physically. Some conditions, like Heartwater, Cerebellar Hypoplasia, Blackleg, and others have a direct and observable impact on how the animal moves. Lethargy and reduced physical activity are an early warning sign for a wide range of conditions, and pedometers are a relatively cheap and effective way to monitor individual animals (see this study in the Canadian Veterinary Journal.) They aren't perfect: after all, many issues can cause lowered activity including a foot injury, skidding on wet floors, etc. However, even non-pathogenic issues are helpful to be aware of, and early detection of any problem is almost always beneficial.

Some examples of where current data monitoring is being used in cows.
https://www.ft.com/content/2db7e742-7204-11e7-93ff-99f383b09ff9

As of today, sensors and monitors can collect, analyze, and report data from almost any system within the cow. Many farmers now begin their day by checking their phone for notifications on the health of their entire herd. But what does this all mean for the future, and specifically for conservation efforts? Let's take a look at what it takes to make a system like this work.

Big Data, Cloud Computing, and Machine Learning

The first thing one needs in order to analyze massive amounts of data, is someplace to keep it all. While I can certainly keep a hundred or so high-def movies on my laptop, I would need A LOT more space to hold the entire Netflix library. Luckily, I don't have to store it all myself! Instead, Netflix hosts a larger network of servers located in some other place, and allows each user to access the movie or show they want. This is the wonder known as the Cloud. It's not actually in the sky, as it's actually servers just like everything else. It just allows you to access these servers from essentially anywhere with an internet connection, and on any internet-compatible device. While it may not sound that mindblowing, it has allowed for an incredible revolution in data storage. I no longer need to build a server farm myself to store insane amounts of data, I can simply rent the server space from Microsoft or Amazon. And it's becoming actually affordable.

Big Data and Conservation; Image Credit: Smithsonian's National Zoo
https://www.flickr.com/photos/nationalzoo/

So, I start collecting as much data as I can scoop up from pedometers on my herd of cows. Let's say I want everything from how many steps, to how big each step was, to which direction the step was in relation to the sun at that exact time, and beyond. How do I make this meaningful? I'm not a mathematician, so data tables and spreadsheets tend to turn to gibberish for me, and fast. Even a numbers enthusiast will have a hard time finding patterns and useful information in a pile of data this size. The solution is a fascinating field known as 'Machine Learning'. To paint with a broad brush, machine learning programs focus on teaching a piece of software to look for, recognize, and make decisions based on patterns and other observed phenomena within a data set. A simple machine learning program may be trained to analyze photographs and look for a particular object or set of objects (think: a Google captcha asking you to select every photo with a stoplight in it; you're actually helping an AI program learn.) Initially, the program may think every black rectangle is a stoplight, but after analyzing more and more photos, the software will begin to grow in accuracy. The more data you throw at them, the faster they learn and improve upon themselves. This is the heart of the software analyzing the movement of each cow. The algorithms began by learning what normal activity and movement looks like, and then was able to differentiate outlying data that suggests something may be amiss. This field is so fascinating and volatile, I will have to come back in another post just to talk about machine learning.

So I have the ability to gather massive amounts of data, I have space to store it all thanks to the cloud, and now I have a piece of software that will dig through the mountains of data for me, while picking out any relevant information, patterns, and interesting tidbits it can find. Say one of my cows gets sick, I can tell that to the software, which could then analyze the previous 10 days of data for example. Any anomalies found could help the software to pay attention when a similar outlier is seen at a later point. Voila! We now have a machine-learning, cloud-based network which keeps tabs on our herd while we eat, sleep, relax, or whatever else it is farmers do when they're not working. The best part? It (theoretically) gets better as it works!

Now back to Conservation

So we can make raising cattle more efficient. That's great for those of us who love milk or a good steak, but the real question is what else can we do with this tech? I believe the answer to that question is: oh so much! Access to cloud-based servers has already opened the floodgates for a huge range of programs and activities that were not previously possible (check out this article for a breakdown of the past 10 years in cloud advances.) One of the coolest things about this technology is that it allows a series of devices to work together on the same task, all connected via the cloud. My laptop might not be beefy enough to analyze terabytes of data, but I now have access to hundreds if not thousands of CPUs capable of teaming up, turning a huge process into a series of smaller ones. With the continued spread of open-source software (generally free software which can be accessed, used, and improved upon by anyone,) literally anyone with internet access can work on some pretty incredible computational tasks.

Example of an animal detection program for wildlife camera traps. Image credit University of Washington.
https://github.com/microsoft/CameraTraps

GitHub, a community of programmers, coders, and general software enthusiasts, has a massive repository of open-source programs geared for a wide range of uses. Searching GitHub for the keywords Conservation Technology already brings up some pretty cool stuff (check out this program for tracking abalone conservation, for example!) If drones are what you're into, there's some interesting stuff coming out allowing for image recognition, self-navigation, and so on which will like radically change the use of drones in conservation and ecology efforts.

Aerial photographs, taken by drone, of orangutan nests in a SE Asian rainforest. Drone software can be taught to ID these anomalies, allowing for an autonomous nest-finding drone. Image Credit: ConservationDrones.org

Machine learning and AI also have a lot to bring to the conservation table. I already mentioned image recognition, which is being used everywhere from tracking the migration of sea turtles to identifying animals captured on wildlife cameras. Even more amazing, machine learning is being applied to the soundscape as well. Acoustic monitoring has been used for years to monitor species such as whales, but advances in technology are stretching the limits. For example, this neural network from MIT can identify speech and music as well as a human. It takes a bit longer to learn than a human does, but as the article points out, it won't complain about being forced to listen to audio files all day! Imagine if, instead of relying on a series of static wildlife cameras limited by their capture area, we had a series of microphones which could pick up any sound for tens of meters around. Day or night, rain or shine, sounds could be recorded and fed into analytical software which could learn to pick out a loris's cry, or a frogs croak, and ignore irrelevant background noise. This is no longer farfetched. I imagine acoustic monitoring will soon become the dominant form of data acquisition for identifying and tracking species as well as monitoring population densities and movements. Check out Wildlife Acoustics for an idea of what's available, and to access a free copy of acoustic analysis software (Kaleidoscope Lite.)

Screenshot from Kaleidoscope, Acoustic Monitoring Analysis Software from Wildlife Acoustics.
https://www.wildlifeacoustics.com/products/kaleidoscope

I am already blown away by the advancements which have been made in the past 15 years, and I do not doubt for a moment that the future will bring things I could never have imagined. I'm very excited to see how we can utilize these breakthroughs in a way that benefits us all, and contributes to a more healthy and just world. It's up to us to guide our progress to this end, and we need all the great minds we can muster!