Large Universe Models in agriculture

What it ingests

Satellite and drone imagery, soil moisture and nutrient sensors, weather forecasts and accumulated degree days, pest and disease reports from surrounding regions, input prices, and commodity forward curves.

The belief that matters

Expected yield and its distribution per field, revised through the season rather than estimated at planting and discovered at harvest.

Decisions are irreversible and time-boxed

Irrigation, fertiliser application, fungicide timing and harvest date are each decisions with narrow windows. Being right a week late is identical to being wrong.

A maintained belief that widens on a dry forecast and tightens after rainfall gives these decisions an explicit uncertainty rather than a farmer's recollection of a similar year.

Regional evidence

Pest and disease pressure propagates geographically. Reports from neighbouring regions are direct evidence about local risk, and they arrive continuously through extension services and grower networks — a stream that maps naturally onto belief revision.

Limits

Agronomic models are well developed and locally calibrated. The contribution is integration across imagery, sensors, regional reports and markets, not replacement of the agronomy.