Advancing Precision Crop Yield Prediction With Data Analytics

Something went wrong. Please try again later...
Yield Image

We’re getting really good at predicting crop yields, and we’re about to get a lot better. That’s good for farmers—and for the planet.

It’s the dead of winter—time to think about what to plant this spring. The decisions you make now will affect all kinds of things. Not just your profitability come harvest time, but how much you’ll pay for crop insurance, any futures contracts you enter into, and the cost of the seeds and other inputs you’ll need.

Since 1980, farmers around the world have been turning to the World Agricultural Supply and Demand Estimates prepared by the U.S. Department of Agriculture (USDA) for help in making these decisions. Every month, the USDA releases supply-and-demand forecasts, an exhaustive analysis compiled from farmer surveys and historical weather patterns, for major crops like corn and soybeans.

Now, however, a number of other players have entered the game, bringing a new level of expertise and computing power to the problem. Their efforts aren’t just making life easier for growers. Their ultimate goal: to make agriculture safer and more sustainable far into the future.

Going forward

The USDA is surprisingly good at predicting the future. Two years ago, University of Nebraska researchers compared these USDA crop reports going back decades with actual end-of-season production. The result: the agency’s estimates were consistently within a statistically acceptable margin of ultimate real-world yields. For instance, the university study found that in each of the 20 months of estimates for a given crop year (each USDA crop report starts before and ends after the growing season), the USDA, on average, never overestimated U.S. corn output or underestimated Chinese corn yield by more than 4 percent.

That’s a performance that the USDA can be proud of. But recently, it has been outpaced by a slew of computer-savvy academics and start-ups that have helped the business of projecting yields take a giant leap forward. Companies like Descartes Labs, Indigo Agriculture, Farmers Edge Precision Consulting, and a handful of universities (primarily land-grant schools) have adopted a more sophisticated approach to predicting crop yields. And although sales to farmers have been slow, these tools are introducing growers to sophisticated modeling software that gauges water, pesticide, fertilization, and sowing requirements plant by plant to generate optimum yields, a category of products that most experts expect will be commonplace in the next five years or so.

Yield Image
Yield Image
“Before long, better crop yield predictions will be viewed as a primitive achievement.”
—Kaiyu Guan, University of Illinois

A leader among this crowd is a team at the University of Illinois at Urbana-Champaign whose track record is noteworthy. Between 2010 and 2016, the USDA yield report for June during each growing season was off, on average, by 17.66 bushels per acre, while the Illinois program missed the mark by only 12.75 bushels per acre. Comparing August reports, the USDA misestimated by an average of 5.63 bushels per acre, while the Illinois team whittled that figure down to 4.37 bushels per acre.

Yielding results

That may sound like a small difference, but farmers don’t see it that way. Growers like Trevor Scherman, who produces wheat, canola, and lentils in Saskatchewan, Canada, say that having access to more-precise readings of anticipated yields in their region and beyond helps determine how much to plant, and thus better align their farming strategy with likely prices and profitability potential. “I can make decisions about where to put resources, informed more by expected supply and demand and much less by guesswork,” says Scherman. “Do I grow 30 acres of crop this year or only 20, and what are the likely results?”

But it’s not only farmers that are already benefiting from yield estimate programs. Yield predictors are used by grain and consumer goods companies to plot raw material purchases and logistics schedules. They also influence premiums for crop insurance, the size of agriculture-related bank loans, and cost volatility for inputs like fertilizer and seeds. And they provide crucial data points for federal, state, and local governments in economic and farm subsidy planning. Simply put, more-accurate forecasts mean fairer and more-efficient markets. “This information allows me to have conversations about business with neighbors, bankers, and lawyers based on facts,” says Scherman.

Yield Image
Yield Image

Yield prediction also influence premiums for crop insurance, the size of agriculture-related bank loans, and cost volatility for inputs like fertilizer and seeds.

Inside the crystal ball

The primary distinction between the USDA’s estimates and the yield forecasts produced by the new coterie of competitors that claim to offer more-accurate options lies in the complexity and capacity of the software and the intricacy of the data. Just as the U.S. Department of Labor surveys businesses to collect employment data, the USDA’s gets its crop yield information mostly from surveys of farmers—providing assessments of the health and maturity of their crops over the course of the growing season—and the impact of these factors on production expectations. In more recent years, the USDA has bolstered the accuracy of its calculations with historical climate information used as predictors for future patterns.

By contrast, the alternative organizations are tapping into much more complicated databases and more advanced crop analytics to improve on the USDA approach. The University of Illinois team, for example, is amassing spatial and temporal satellite images of every individual field in the U.S., and eventually around the world. These pictures from space are blended with separate tranches of data to compensate for missing information resulting from, for instance, cloud cover. The results are high-resolution images that provide penetrating views of crop growth activity, including plant health and stress, sensitivity to heat, and yield variability.

More recently, the program’s experts have been capturing data on solar-induced florescence, providing plant-by-plant information on photosynthesis. The Illinois group runs this brew of byzantine data through algorithms and applications on Blue Waters, the university’s resident supercomputer, taking advantage of the machine’s learning and intelligence capabilities to generate ever more detailed and timely crop yield forecasts. The computer even suggests the types of questions researchers should be asking as they sift through all of the new information at their disposal.

Kaiyu Guan, a professor at the University of Illinois and a principal investigator on the crop monitoring project, says that in the next couple of months, his team will debut perhaps the most advanced predictive yield model yet. “We will map out daily yield analyses for every individual field with staple crops in the U.S.,” Guan says.

Yield Image
Yield Image
The program’s experts are capturing data on solar-induced florescence, which provides information on plant-by-plant photosynthesis.

Going forward

Similar innovative approaches to developing crop yield programs are underway at private-sector companies and other universities. Indeed, innovations in agricultural predictive modeling are developing so fast these days that researchers are already looking beyond predicting crop yields to solving much more complex and intractable farming problems using similar approaches to data collection and analysis.

Current satellite images are extremely detailed but cannot see below the crop canopy and the aboveground biomass. As advances in artificial intelligence and machine learning generate data patterns that consistently predict specific levels of crop growth and plant maturation, this information could be compared with underground water, nitrogen and carbon availability and readings of root health and pest infestations to produce maps of underground conditions. This, in turn, would provide farmers with a suite of actionable intelligence to drive precision farming strategies that could enhance food security and minimize resource and chemicals usage at a time when these issues are becoming critical priorities for the survival of the planet.

“Before long, better crop yield predictions will be viewed as a primitive achievement,” says Guan, who notes that the University of Illinois plans to give the output of its research away to farmers for free—in a usefully simplified form, of course. “What we can learn from it when we can see what is beneath the surface is much more valuable, impacting not just crop yield but profits and environmental protection too.”

Which means that a sustainable future for farming may eventually be traced to breakthroughs that enable us to better see the future today.