We rely more and more on artificial intelligence and powerful computers to accumulate and process information. Can artificial intelligence help us grow a better soybean crop?
In this Soybean School episode, Bernard Tobin is joined by Shawn Conley, University of Wisconsin-Madison soybean extension specialist, to talk about his research in artificial intelligence derived management systems for soybeans.
Conley is performing trials with growers this year, that will compare how those growers manage their soybeans to management recommendations from artificial intelligence.
Recommendations are based off of data amassed over the last eight years, says Conley, including variety testing data and agronomic trial data from all over the U.S. “Can we use machine learning tools to be able to predict on a field-by-field basis what farmers should be doing on a specific field,” says Conley. More information on how the trials are set up can be found here.
In the video, Conley outlines how machine learning tools will be used to predict the practices that can be used to maximize yield or to maximize profitability. (Story continues below video)
“That’s what we’re just trying to do in this first year, is see if we can ‘break the tool’ or to see if it actually works,” says Conley. “Hopefully over time we’ll keep building it and building it and making it into a place where farmers can upload their own data, for free, run the model in the background, and try it on their own farm.”
The artificial intelligence is quite formidable, and as Conley explains, the model can run a million different iterations and has the capacity to assess multiple interactions between things like soil type, fertility, seeding rate, row spacing, etc.. In-season management decisions (spray timing) can also be plugged into the model and is part of the objective list for Conley’s project this year.
Looking at the site-specific management systems that the model generates, Conley admits that he’s seen some weird thing so far, but because it’s an experiment he’s willing to test them out for accuracy. “All models are inherently flawed, you just have to keep working your way through it to figure out how to fix it to keep the errors to a minimum,” he says.
Other models and databases are being built for corn production too, adds Conley.