USDA turns to AI to check crop estimates after data errors

The U.S. Department of Agriculture will introduce an internally developed artificial intelligence tool to review its November 10 crop production report, officials said, following data mistakes that have weakened confidence among farmers and traders.

USDA Under Secretary Scott Hutchins said officials concluded the technology would have identified errors in September's corn estimates. He said it would serve as an extra safeguard alongside existing checks, rather than replacing those procedures.

The agency's corrections lifted Nebraska's estimated corn yield by about 3% and cut the Kansas estimate by about 2%. Lance Honig, who chairs the agricultural statistics board at USDA's National Agricultural Statistics Service, said a computer file had wrongly combined survey information for irrigated corn acreage with information for non-irrigated acreage.

Honig said the states were affected because irrigation, which often produces higher yields, is more prevalent there. He said the field survey data were not faulty and adequate validation should have identified the error.

The National Corn Growers Association, representing 35,000 farmers, called the mistakes unacceptable and said they further damaged confidence in USDA reporting. Earlier criticism followed sharp upward revisions to corn acreage estimates and incorrect beef export sales figures.

Thousands of USDA employees have departed after government downsizing, prompting experts to question whether the agency could maintain timely, accurate reporting. Hutchins said staffing reductions were not responsible for September's mistakes.

Separately, analysts challenged figures in a monthly soy-processing report showing a large discrepancy between Iowa's soybean crush and soymeal production. USDA said a routine review was underway, with any necessary corrections to follow the next month.