A group of data scientists walk into a farmer-owned cooperative. No, that’s not the start of a joke. Rather, it was the start of a crash course in agriculture for five students at the University of Minnesota (U of M) Carlson School of Management.

In early 2017, as part of the Carlson Analytics Lab program, a diverse set of companies challenged Masters of Science in Business Analytics students to use real-world, oftentimes complex and messy, data to answer big questions and create actionable recommendations. Land O’Lakes, Inc. brought a farm-to-fork question to the table—how can we increase the confidence of farmers in our Answer Plot^® program to help them sustainably grow more food?

“Myself and a team at Land O’Lakes have been working with the U of M on a big data proof of concept,” says Teddy Bekele, vice president, Ag Technology, at WinField^® United. “Essentially, we gave the students a subset of Answer Plot^®, climate and soil data. Their goal was to develop algorithms leveraging big data techniques. The output was to see if they could create a correlation between a given farmer’s field and our test plots.”

According to Teddy, this correlation could create greater confidence in the program. Simply put, if we show farmers research data from the Answer Plot^® locations that most resemble their fields, it gives them confidence that their results will match ours. Today, most farmers turn to the research plots closest to home, which may or may not be similar.

Five students signed up to answer our question, partnering with WinField United R&D, Land O’Lakes IT and Carlson faculty along the way. They analyzed data, developed business recommendations and learned a whole lot about agriculture.

Growing data with the Answer Plot^® program

WinField United’s nearly 200 Answer Plot^® locations make us unique in the industry. Think of these as living laboratories spread out across the country. At these plots, we plant and test seed, crop protection and plant nutrition technologies, collecting more than six million points of data each year.

Within hours of an Answer Plot^® harvest, the raw field data makes its way from the combine to a central database. And it's not just during harvest that this database fills. Throughout the year, the plots are managed by 18 Answer Plot^® research teams. They are continually gathering data and insights to help enhance our agronomic expertise and provide credible facts for farmers using similar conditions.

Providing even more value, Answer Plot^® events held at different points throughout the growing season serve as a “show and tell" experience for our retailers and farmers. Whether it's digging and analyzing roots, looking for insects or diseases, or standing back and noticing an overall different appearance between two treatments, the demos provide an outdoor classroom opportunity to gain growers' and retailers' trust with WinField United's products and insights. And these events have proved popular. We regularly host more than 10,000 visitors at more than 600 Answer Plot^® events each year.

But if you’re a farmer talking with an agronomist or attending an event, a question you might find yourself asking about the Answer Plot^® program is, “How can I trust this is right for my farm?”

Considering differences in climate, soil types, native pests and countless other factors, it’s a fair question that prompted our question for the students.

Meet the students

When Jeff Macaluso, Aayush Agrawal, Bill Eerdmans, Jialing Zhang and Shawn Spence walked into Land O’Lakes for the first time, no one had what you’d call a traditional ag background. But they had just the right mix of experiences and skills. Several had previous analytical work experience, in a variety of industries. There was even a former Air Force weather forecaster on the team.

Two themes, though, were consistent through the group. 1) Their obvious interest in data and business analytics, and 2) the desire to work on something that would advance the idea of feeding human progress.

“I’m from Austin, Texas, and have no ag experience. I’d never been on a farm,” says Jeff. “But this sounded like a really interesting project that would have a tangible impact. I like the idea of making the environment better by improving efficiency.”

“It’s a challenge—having accessible food in the future,” says Bill. “This was a chance to work on a project that could have a real societal impact.”

With this greater purpose in mind, the team set out to use data to answer our question on improving the Answer Plot^® program.

Analyzing the Answer Plot^® model

Answer Plots^® are currently based on geography, with farmers attending events at the plot closest to their farms. The idea is what works at the plot should work for the nearby farms.

But here’s the catch—conditions between farms and plots are often very different regardless of proximity. Even within distances of a few miles, there are different soil types that come into play, terrain changes and even varying climate conditions. To get the students started, a team at Land O’Lakes, including Ag Tech Spatial Data Specialist Ryan Miller and IT Mobile and Emerging Technology Developer Brandon Budnicki, gathered open source climate data and soil maps for the entire United States, along with data from nearly 200 research plots.

The open source data was spread across thousands of files, mostly formatted in large blocks of unorganized text. Not very searchable. So, the students formatted and cleaned the data to make it accessible for open source analytics tools.

The students met with WinField United agronomists, including Joel Wipperfurth and John Kinnard, and reviewed academic publications to better understand the data. In fact, every week throughout the semester, the students came the Land O’Lakes headquarters building. This gave them the opportunity to build the model together and receive feedback in real-time from experts.

After talking to our agronomists, the team decided to focus on climate, topographic and soil data. They also decided to focus their study on data available for 300,000 farms across Minnesota and Wisconsin.

They created a conditioned-based similarity metric that could determine the most similar (in terms of climate, topography and soil type) Answer Plot^® for an individual field. When they ran the metric on all the fields in their data set, they discovered a field might be most similar to a research plot hundreds of miles away than the plot just up the road.

By comparing a field’s similarity to a plot, instead of geographic location, they could improve overall program representation and effectiveness.

Here’s a hypothetical example. A farmer with loamy soil in Wisconsin might not be represented by the local plot that has more sand in the soil, but would be represented by data from a loamy plot in Minnesota. The data collected by the students makes finding the most similar plot possible in a quantifiable way.

Where to go from here

The students’ work is opening doors to new business opportunities and new ways to make a difference.

After the semester wrapped, the students presented their proof of concept to WinField United R&D. Their findings and recommendations were also handed off to an internal Land O’Lakes IT team that will take the data forward and see how we can use it to improve our programs and ultimately help farmers.

This partnership is one step in our journey to help farmers adjust to weather, biodiversity and any other challenges in their way, and still produce efficient crops and a sustainable food supply.