U.S. counties require at least 40 reported millet acres and 40 classified acres in two of 2023–2025. Grain concentrations additionally use native crop cells from all three years. Canada uses three annual native inventories plus documented production geography.

Recurring 0.15° rotation neighborhoods anchor 0.03° crop cells with at least 2 acres and 2% occupied area. A bounded 1.5 km morphological closing preserves separated concentrations; no counties or convex hulls fill the map. Other U.S. uses have recurring county evidence but 2025 detailed spatial support. Species remain unresolved where the classifier cannot distinguish them.

Inclusion rules, source years and exclusions · Mapped versus reported acreage. Manitoba proso reported acreage exceeds classified support; intended uses, classification omission and source coverage prevent forcing the totals to match. Quebec pearl/forage evidence is separate from the Prairie reference. Isolated Canadian pixels outside corroborated regions are withheld.

UNL / CSU / SDSU / Wyoming / USDA-ARS proso guide supplies the Great Plains planting reference and cultivar GDD thresholds. Planting ±14 days and additional delays represent scenarios. Applying this model elsewhere is an explicit transfer assumption; it does not identify species or observed planting.

Open-Meteo modeled daily weather supplies map surfaces and stage exposure. In CO, NE and SD grain regions, NASA POWER supplies rain, temperature, stages and modeled soil wetness from the native 0.5° × 0.625° grid; each crop cell stays in its own source grid cell. Other North American regions retain their separately labeled Open-Meteo samples. Missing grid cells are unavailable, never filled from a different stratum. Reference ET is not supplied by this NASA package, so the affected rain-minus-ET values are unavailable; stage, rain and heat remain usable. No station observations are substituted into this modeled surface. Weather inputs are checked against downloaded native rasters. Map values use the assigned crop-support cell and remain clipped to growing regions or the native pixel mask. Fine crop shapes do not increase weather resolution.

Observed IEM station reports are a separate optional overlay with actual UTC timestamps, minimum 60% daily coverage and no imputation. Hourly reports are not daily rainfall. Station coverage and provenance.

NASA POWER supplies the same matched historical and seasonal weather for the corrected U.S. map, stage comparison and coarse-yield hindcast. Clipping does not give gridded weather field-scale accuracy.

USDA annual summaries supply dated proso grain outcomes. Historical ranges and experimental weather-based estimates are distinguished. Canada and other U.S. states retain geography and seasonal scenarios; numerical yield is withheld where local outcomes and calibration are absent. Historical classified acres are weights, not a 2026 acreage census or production total.