Yield forecasts are produced using an ensemble of machine learning-based models that translate weather conditions, soil conditions, USDA crop condition reports, as well as various other data and metadata into forecasted crop yields at the county level. These county-level yield forecasts are then disaggregated across all fields hosting that same crop within the county using both historical and current-season satellite-based indicators of crop development and health. Once the USDA Risk Management Agency (RMA) releases ‘final’ yields for a season for each crop and county combination, those yields are similarly disaggregated to produce final field-level estimates of crop yields.