USDA to launch AI tool to check crop production data after errors
By Tom Polansek
CHICAGO, Nov 6 (Reuters) - The US Department of Agriculture will use an artificial intelligence tool it built to check data in next week's monthly crop production report, agency officials said, after a series of data missteps eroded farmers' and traders' trust in its forecasts.
After the USDA discovered errors in its corn data from September, officials determined the AI tool would have flagged the problem and decided to use it on the crop production report due on November 10, USDA Under Secretary Scott Hutchins said in an interview.
The tool's debut tests whether the agency can restore confidence in data that often moves markets.
"I'm very bullish on the fact that we'll be able to utilize these tools in the future to help us avoid these kinds of really, truly unfortunate errors," Hutchins said on Thursday.
Last month, the USDA corrected its September corn production and yield data. The correction raised the yield estimate by about 3% in Nebraska, the third-biggest corn-producing state, and lowered it by about 2% in Kansas.
The agency, long viewed as the gold standard of global agricultural market information, previously faced criticism after it made sharp upward revisions to estimates for last year's corn acreage. Over the summer, it reported inaccurate beef export sales data.
The USDA previously said it was launching a pilot project that would rely more heavily on technology to improve crop acreage and yield estimates as fewer farmers respond to its surveys.
COMPUTER FILE ERROR
Lance Honig, chair of the agricultural statistics board for USDA's National Agricultural Statistics Service, told Reuters that a computer file incorrectly combined data on irrigated and non-irrigated corn acres gathered by field surveyors, causing the errors in September's data. Kansas and Nebraska were affected because they have higher percentages of irrigated fields, which often have higher yields, he said.
"This error that occurred in September is literally a very isolated occurrence that, quite frankly, with the proper validation, we should have caught," Honig said. "There was no issue with the data that was gathered at the field level."
The errors angered farmers who are struggling to make money as they pay soaring prices for diesel, fertilizer and other inputs. Last month, the National Corn Growers Association, which represents 35,000 farmers, said the September errors were unacceptable and had further hurt farmers' confidence in the USDA's ability to produce accurate reports.
"You just don't know what the USDA is going to do anymore," said Jim Gerlach, president of brokerage A/C Trading in Indiana, on Friday. "Their reports have become so erratic."
Questions around the credibility of federal data increased after President Donald Trump's administration sharply downsized the government last year. Thousands of employees at the USDA have left, and experts have worried the cuts hobbled the agency's ability to produce accurate and timely data.
Hutchins said the September errors were not related to staffing losses. The new AI tool will add to, not replace, the agency's data validation, he said.
"The tool is basically going to be used as a check to make sure that there are no anomalies that we were not able to see or identify in the normal process," Hutchins said.
This week, market analysts questioned the accuracy of another monthly USDA report on soy processing which showed a sharp divide between how many soybeans were crushed and how much soymeal was produced in Iowa. The USDA said it was reviewing the report as part of a normal process and would make any necessary revisions next month.