Global agribusiness company Syngenta is at the forefront of using data analytics techniques to feed the world and help farmers overcome challenges. One way it encourages innovation to tackle these ambitious goals is with the Syngenta Crop Challenge, a global open innovation competition that invites participants to improve farm productivity. For the 2018 version of the contest, solvers were challenged to apply their analytics skills to develop models that predict how well corn hybrids will perform in untested locations.
This year’s winners hail from The International Center for Tropical Agriculture (CIAT) in Colombia, a not-for-profit dedicated to reducing poverty and hunger while protecting natural resources in developing countries. The team consisted of almost a dozen members and was headed by Dr. Daniel Jimenez an agronomist, data scientist and pioneer in the use of artificial intelligence in agriculture. They worked solidly for months on their solution, beating off stiff competition to take the USD $5,000 first prize. This is how they did it.
CIAT had been aware of previous crop challenges but hadn’t taken part due to funding issues. The organization’s work is project based and taking unpaid time away from these is difficult. However, Dr. Jimenez decided the time was right to enter the 2018 competition and persuaded his colleagues to take part.
“Well, it was more like an instruction to the team. I said ‘look this is very interesting I think we all have the capability to do this. We know how to use artificial intelligence and data mining, we have a statistician who’s learned a lot of techniques. I think we can do well. We have the team right now so let's do it.’”
With that spirit of confidence and can-do attitude, the team entered the international competition. The research and development organization has been working on agriculture and analytics for almost ten years and has received global recognition for their endeavors from such institutions as the United Nations and the World Bank Group, collecting a ‘Momentum for Change’ Climate Solutions award. However, participation was something of a learning curve.
CIAT hadn’t previously explored its analytics capabilities in a worldwide contest against university and private sector teams. Nor had it used data from the developed world as the data it usually works with is from the developing world.
To create models to predict how corn hybrids will perform in untested locations participants were given datasets with genetic information and yields from trials of experimental maize hybrids, as well as information on the soil and weather conditions of a growing region in the United States.
“The contest had a strong genetics focus which is something we hadn't done before and so the interpretation of the results was challenging because of this genetic component. Also, we tried some tools we hadn't previously used like TensorFlow which is the Google platform for artificial intelligence.”
The team worked intensely for several months including throughout the Christmas period. There were meetings nearly every day and lots of technical discussions about techniques and approaches. Among the many hurdles they faced were how to handle a huge dataset with missing values and generating a weather forecast for 2017 because there was no data on conditions of the coming season.
The end result of a feverish bout of activity was a solution that can speed up maize hybrid breeding using machine learning. It was one of the best that Syngenta received and consequently, CIAT was invited to Baltimore along with four other teams to present their submission to the Crop Challenge prize committee.
Preparations for this important day were undertaken in earnest and it was here that Dr. Jimenez experienced some of his proudest moments as the team’s leader.
“At CIAT we have a team that's a nice combination of agronomists and geeks. Some of the team are like how I used to be years ago. They come from working-class families and went to public schools and some couldn’t speak English. Up until three years ago one of them couldn’t speak it at all well.
“CIAT has enabled members of my team and myself to build and strengthen our capacities to not only speak English but to interact with peers and thought leaders at international events.
“Then when I saw one of our team members speaking to Syngenta and IBM in English I was like, as a leader I’m done, I’m very happy. You see it's not necessarily about the prize but how the team is growing with you.”
Nonetheless waiting for the announcement of the winner was a tense, nail-biting time for CIAT. Submissions were evaluated according to a range of criteria including how well predicted performance aligns with observed hybrid performance for 2017, the simplicity of the solution, clarity of explanation and novel ideas used to create the predictive model.
With their decisions made the judging panel announced the top three winners in reverse order.
“The third place winner was announced, a really good university team,” continues Dr. Jimenez. “Then second place went to a PhD candidate from Brazil, who is the closest thing I have ever seen to a data unicorn. He did a lot of things by himself.
“So, then I said to myself we didn't make it because I thought the team from the University of São Paulo was going to win. But then they said the first place is for CIAT and I just couldn’t believe it.”
CIAT’s prize-winning solution can be applied to any crop anywhere in the world as long as the genetic information of the crop is known as well as the weather and soil conditions of the area where the crops are to be planted.
Participation had been a worthwhile endeavor not just for developing a brilliant solution and the satisfaction of winning but also for the knowledge and experience gleaned along the way.
“We are going to integrate the things that we learned in this Syngenta contest in our research and it has the potential to benefit hundreds of thousands of farmers. This isn’t blue skies research it's research that has applications in the real world."