Minding Ag's Business

Crop Predictions at the Speed of Data

Katie Micik Dehlinger
By  Katie Micik Dehlinger , Farm Business Editor
Just as the cost of computers declined significantly following improvements to its microchips and processors, the falling cost of collecting and storing large quantities of data foretells dramatic upticks in it use. (DTN file photo)

Data is everywhere, and sometimes there's simply so much of it that's it's hard to make heads or tails of it.

When DTN partnered with Gro Intelligence in 2018 to launch a Digital Yield Tour, we matched Gro's corn and soybean yield models with boots on the ground commentary to paint a top-down picture of how the crop was progressing, instead of the bottom-up method employed by other crop tours.

The derecho that tore across Iowa, Illinois and parts of Indiana posed a challenge this year that wasn't easy to assess, for our tour or the USDA. It took USDA a full month before it included the windstorm's impact in its analysis, but it started showing up in Gro's yield models shortly after the event.

Why do I bring this all up now? James Heneghan, Gro's vice president of agribusiness, recently appeared on our Field Posts podcast (https://fieldposts.buzzsprout.com/…) to talk about the derecho, but also about the future of data and quickly it's becoming more accurate and useful.

County yields in Iowa declined markedly after the storm. Jasper County, for instance, was hit by some of the strongest winds in the derecho, but it also had the highest county-wide average yield when we conducted the tour at 209.8 bushels per acre. Two weeks after the storm, Gro's models shaved off 9 bpa, and of this writing, the county yield projection is at 193.5 bpa.

The derecho also lowered Gro's national yield estimate, Heneghan said in the podcast, with the corn yield coming in just over 180 bpa, a shade over USDA's September yield estimate of 178.5 bpa. USDA will release its updated forecast on Friday. It won't release county-level data until next year.

"With the technology that's out there and with the data that's out there, you can look at this daily, weekly," he said, adding that USDA's monthly releases are no longer the best, most timely estimates of the crop. "It's not so much that the old process is done poorly. In fact, it's done quite well. But with new technology you can do it even better and look at it more frequently than you used to."

But can you trust it?

This is where Heneghan starts talking about error rates. During our August crop tour, he said the target is a 5% error rate, meaning the yield estimate is within a range of 5% above or below UDSA's final yield projection in January. By mid-November, when Gro's national yield models cut off, the goal is to be within 1% of USDA's final estimate.

Heneghan said the ultimate goal is to get the error rate as low as possible as early in the season as possible, and a rapidly evolving data landscape could help in that effort.

Gro's models, like many that are out there, utilize machine learning, which help make the models more accurate over time. But from Heneghan's perspective at a company that lives, breathes and sleeps data, better data yields better results.

"Data sets are coming in more frequently at higher resolution, and the costs to store that data and process that data is getting cheaper and cheaper," he told Field Posts host Sarah Mock. For example, the national differential vegetative index, or NDVI, is a big driver of their yield models, and it is going from an eight-day data series to running every day. Heneghan said Gro just added a second source of soil moisture data to is models, allowing it to compare and contrast.

He points to the rapid evolution of weather forecasts as a prime example of how data is changing the way we understand the atmosphere. "Some of the forecasts going even a month out are getting very, very accurate. If you think about even 10 years ago, you wouldn't go out a week," he said.

"As the meteorological models get better, adding data around what the forecast tells you about the progression of crops, and add that to our models, that's stuff we're excited about."

If you want to learn more about weather models and their increasing accuracy, make sure to tune into the all-virtual DTN Ag Summit from Dec. 7-9. We'll have a breakout session on exactly that topic. You can register here: www.dtnagsummit.com.

Katie Dehlinger can be reached at Katie.dehlinger@dtn.com

Follow her on Twitter @KatieD_DTN


To comment, please Log In or Join our Community .