Farm Progress is part of the Informa Markets Division of Informa PLC

This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.

What satellites can't see

harryfn/Thinkstock Aerial view of small fields
USDA says satellite images can't be relied upon to predict annual corn, wheat or soybean harvests.

by Jeff Wilson

For decades, government satellites have been taking detailed photographs of crops around the world that are now being tapped by traders like Cargill Inc. to gain an edge in global grain markets.

But the U.S. Department of Agriculture -- the benchmark in forecasting domestic crops -- says the images by themselves still can’t be relied upon to predict annual corn, wheat or soybean harvests. Instead, the government’s main source of information remains farmer surveys and random field samples.

“Satellites are not advanced enough to differentiate crop acres yet, so there is a loss of precision,” said Seth Meyer, the chairman of the World Outlook Board, the USDA agency responsible for world crop forecasts. “This technology is going to get better, but right now it’s just one tool in our forecasting toolbox.” 

Getting accurate assessments of major U.S. crops valued at more than $100 billion last year is a recurring challenge for traders, consumers and farmers. Crop conditions can change with the weather over the long growing season, so any early forecasts may be far off the mark when harvest rolls around. 

Some scientists expected satellite images to eventually make the job easier. The U.S. has been taking pictures from space since the 1970s to track everything from the weather to troop movements. But it wasn’t until the last few years that advances in digital technology and computing power made those billions of images more useful in crop forecasting.

Making Switch

Statistics Canada, the government agency that produces the country’s monthly crop forecasts, already is using the technology in key crop assessments. In 2016, StatsCan switched to using only satellite- and weather-based models for the monthly production report published in September, saving about C$150,000 ($95,000) in farm and field surveys for that month. Other reports during the year still rely on the surveys.

“There’s been a lot research put into this model to verify its robustness,” said Gordon Reichert, head of remote sensing analysis at Statistics Canada. “Feedback from the grain companies in Canada has been favorable.”

Satellites scan thousands of square miles of agricultural land and record daily changes in areas as small as two dining tables, mostly by analyzing how green the fields are from planting to harvest. Machine-learning algorithms then match those characteristics with historical data and production results to make forecasts.

Computer Models 

In 2008, the U.S. began offering its satellite data and three decades of history for free, and the European Union made all its space-based observations public in 2013. Companies like Descartes Labs Inc., TellusLabs Inc. and Planalytics Inc. used the information to develop computer models to make all kinds of predictions, including for grain production. Other companies even sent their own monitoring devices into orbit and market the data to farmers looking to catch nutrient deficiencies or diseases soon enough to be treated.

It’s an effective early-warning tool, said Bruno Basso, an environmental scientist at Michigan State University and co-founder of CiBO Technologies in Cambridge, Massachusetts. 

For example, the world was caught off-guard in 1972 when harsh weather led to a big drop in wheat harvests in the former Soviet Union. The country kept the damage secret and filled the shortage with 440 million bushels of U.S. wheat, or 28% of what American farmers produced that season using subsidies provided by the U.S. government. Prices tripled, and the Soviet purchases were dubbed the “great grain robbery” by traders.

How Green? 

The key to today’s satellites determining crop health is sensors that measure different light bands reflected or absorbed by plants. At their healthiest, the leaves tend to soak up more red and blue light and reflect more green, which indicates good growth potential. That real-time information is combined with other variables including weather, temperature and soil conditions, and then compared with historical data and crop results for the area.

Traders are beefing up their satellite capability. Minneapolis-based Cargill invested in Descartes Labs in August. Last year, Descartes got a $1 million grant from the U.S. Defense Advance Research Projects Agency to develop ways of using satellite imagery to predict crop performance and avert food insecurity in the Middle East and North Africa.

But the USDA is in no rush to rely more heavily on satellite imagery. For one thing, while images from space can identify corn or soybean fields in big growing regions like the Midwest, they are less reliable for areas with diverse crops or smaller fields. Still, the agency cites assessments of vegetation health from satellite imagery in its monthly reports. 

Based on two decades of data through 2016, the USDA’s track record remains good -- especially for the August report that tracks a key point in the growth cycle through the end of July. There’s a two-in-three chance that the government estimate that month won’t deviate by more than 3.5% from the final tally in December, long after the harvest, according to Scott Irwin, an economist at the University of Illinois and part owner of crop forecaster YieldCast. 

“It’s going to be a long time before there is a system from the big data forecasters that is as accurate as USDA,” Irwin said. Companies selling analysis of satellite imagery “will have to prove they can beat USDA in real time, not based on historical simulations of model performance,” he said. 

To contact the reporter on this story: Jeff Wilson in Chicago at [email protected]

To contact the editors responsible for this story: Simon Casey at [email protected]

Steve Stroth 

© 2018 Bloomberg L.P

Hide comments


  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.