Iowa farmers face a ‘data deluge.’ Could artificial intelligence help?

Imagine every stalk of corn in an Iowa field as a single data point. Add them all up; then account for yield-affecting variables like soil type, row width, irrigation practices, seasonal pests, weather patterns and more.

Modern-day farmers have a way to track those variables, thanks in part to ongoing advancements in precision agriculture. But put them all together, and you get a “data deluge” that can leave farmers feeling adrift in a sea of numbers and percentage points.

To help Iowa farmers navigate the waters, some agricultural organizations and researchers are turning to artificial intelligence.

“Over the past decade, we’ve seen a dovetailing of various technologies that can essentially transform agriculture. … You have drones, you have satellite data. You have cheap sensors,” said Baskar Ganapathysubramanian, a professor of engineering at Iowa State University. “But at the end of the day, farmers … want actionable insight. (They’re saying) ‘Don’t just give me the data. Give me insight into what I should do with this data.’”

Speaking Thursday on a panel at The Gazette’s Iowa Ideas conference, Ganapathysubramanian said artificial intelligence can provide some of that insight to help farmers tackle topics like profitability, yield outcomes, plant genomics and more.

Ganapathysubramanian was joined on the panel by ISU graduate research assistant Karlene Negus and Matthew Carroll, analytics and insight lead for the Iowa Soybean Association, to discuss the growing overlap of agriculture and artificial intelligence.

Panel: AI has varied uses across industry

Panelists emphasized that the uses for artificial intelligence in the agriculture industry are numerous and likely to evolve over time as AI technologies improve and awareness grows.

Negus is studying the intersection of AI and plant breeding/genetics and is working to create AI models that more effectively leverage large-scale genetic data to predict important agronomic traits in corn such as yield, plant height and flowering time.

For her, AI opens the door to more efficient plant breeding through predictive modeling to see how a plant would react in a diverse range of scenarios and how altering that plant through selective breeding could change those outcomes.

“If we think about the plant breeding cycle, … it’s going to take eight to 10 or even more years for the plants that are developed (now) to finally arrive in fields,” she said. “AI models for plant breeding specifically present the opportunity to shorten that cycle.”

In his role as director of the ISU-based AI Institute for Resilient Agriculture, Ganapathysubramanian has also leveraged AI to help farmers identify insects found in the field through a new app, InsectNet.

The app was developed over two years as part of a cross-disciplinary collaboration between agronomists, computer engineers, statisticians, data scientists and artificial intelligence specialists.

“We collected as much data as we could … and we trained these large-scale models on super computers over several months,” Ganapathysubramanian said. The result “is one more expert in the (toolbox) of the farmer that they can have on their phones.”

InsectNet can identify and classify more than 2,500 insect species using an AI model trained on more than 12 million insect images. The app has an accuracy rate around 96 percent, and when an insect cannot be identified, it gives users an “uncertain” response rather than using a best guess.

Using the app, Ganapathysubramanian said farmers can submit photos of an insect found on their own farm and quickly get an identification that lets them know if it’s a pest that needs to be taken care of or an insect that is harmless or even beneficial to have around.

Limitations, questions remain around in-field integration

One thing that all AI tools have in common is that they require massive amounts of data to build and train. This can pose a challenge to developers when the data needed for a specific tool are either scarce or difficult to access.

“Obviously, we’re kind of in the big data age, and lots and lots of people are working to generate more data sets that have lots of data in them,” Negus said. “But I think there’s kind of a bottleneck” at this point in time.

Individual farmers might be wary to share their on-farm data because of the relative lack of regulatory frameworks around the collection and sharing of such data or out of concerns related to privacy, security and/or licensing.

Private companies might have their own data collected from the producers they work with, although they’re unlikely to share that information with the broader public over concerns related to proprietary information.

Panelists offered a few potential solutions to those issues such as federated learning, a machine learning method that offers increased privacy protections by not requiring data be uploaded to a centralized server.

Speakers also stressed the importance of trust building and partnerships, saying that farmers are more likely to share data with an organization they have a good working relationship with and who they trust to use it appropriately.

The same is also true of getting farmers to implement AI-powered tools. In his time with the Iowa Soybean Association, Carroll said he’s seen a trend where some farmers are quicker to adopt new technologies than others. The same is true of AI.

But by leveraging long-standing relationships between growers and the Iowa Soybean Association, Carroll said more farmers are willing to take a chance on something new and to report back on their findings with suggestions for what could be improved.

“I’d say we’ve got a lot of folks who are excited and some folks who are more ‘Wait and see’ but that’s not unexpected with any technology adoption,” Carroll said. “It’ll always come back to building those trusted relationships.”

Comments: grace.nieland@thegazette.com