A proposal for every field, thanks to Fields of the World
Where there is no agricultural register, FairGrow can propose your fields from Fields of the World: an open world map of field boundaries, drawn by an AI model on Sentinel-2 imagery from 2024 and 2025. A proposal, not a decision: you decide which field is yours.
The project behind the boundaries
We did not draw this map. A group of researchers did, for the whole world, and then decided that anyone may use it. FairGrow is one of the things that decision makes possible.
Fields of the World, an open project for field boundaries
Fields of the World Global is produced by the Taylor Geospatial Institute and the Microsoft AI for Good Research Lab. The model is called PRUE, a segmentation model that estimates for every ten-meter pixel whether it belongs to a field, a field edge or something else. Those probabilities are turned into polygons: about 3.2 billion fields in 219 countries.
The model was trained on the Fields of the World benchmark: more than 70,000 samples from 24 countries on four continents, where people drew the field boundaries by hand. The map itself is open data on Source Cooperative, per country, as GeoParquet.
FairGrow only fetches what it needs: the squares of about five kilometers around the locations of our users. All that goes to the source is the outline of such a square, no name, no company and no exact location.
A candidate, not the truth
A field boundary drawn by a model at ten meters is a good start. This is how we treat it, so you know what you are looking at.
The register comes first
Where an official register exists, such as the BRP in the Netherlands, it leads. There we show Fields of the World as a check layer, for example to spot a field that has been split since the last declaration.
You choose
A boundary from Fields of the World is always a candidate. FairGrow never adopts it as your field unasked; that only happens when you pick it or edit it.
Only where it is needed
We do not download the whole world, only squares around the places where FairGrow has users. That way our storage grows with our users, not with the planet.
A copy of our own
The dataset is still an alpha release and may change. For every country where our users are, we therefore keep a verified copy on our own storage, with a checksum per file.
Honest numbers
We compute area and perimeter ourselves, in meters, in the projection that fits your location. The values from the source are kept alongside.
Your data stays home
Your fields, your sensors and your measurements are on our own servers in a Dutch data center, ISO 27001 certified. Fields of the World only supplies public boundaries; nothing of yours goes back.
What to keep in mind
Ten meters per pixel is sharp for a satellite, but coarse for a grower. These are the limits we measured ourselves or that the makers point out.
Small fields
Below roughly half a hectare it becomes unreliable. We keep polygons from 500 m² and flag everything under 2,000 m² and under 5,000 m². In Armenia and Tajikistan a third of the polygons are under those 500 m².
Strips, orchards and greenhouses
Narrow strips merge, tree rows can show up as separate fields, and to a satellite a greenhouse is a roof, not a field. Drawing or a register works better there.
What confidence means
Each polygon carries a value from 0 to 100. It tells how reliable the model is in a square of 500 by 500 meters, not how accurate your boundary is. The makers recommend 69 as the threshold. If the value is missing, the square is outside the coverage; that is not the same as a low score.
A snapshot
The boundaries come from imagery of 2024 and 2025. A field you merged or split this year is not yet visible to the model.
The license, in one sentence
Fields of the World Global is licensed under Creative Commons Attribution 4.0: free to use, also commercially, as long as you say where it comes from and that you changed it.
We changed the polygons: geometry repaired where it was invalid, holes and slivers under 500 m² left out, and a simplified version made for the map. We keep the original shape alongside.
Publications
Robinson, C., Muhawenayo, G., Khanal, S., Fang, Z. et al. (2026). The first global agricultural field boundary map at 10m resolution. arXiv:2605.11055
Muhawenayo, G. et al. (2026). PRUE: A Practical Recipe for Field Boundary Segmentation at Scale. arXiv:2603.27101
Kerner, H. et al. (2025). Fields of the World: A Machine Learning Benchmark Dataset for Global Agricultural Field Boundary Segmentation. AAAI-2025. arXiv:2409.16252
Corley, I. et al. (2026). Fields of The World: A Field Guide for Extracting Agricultural Field Boundaries. arXiv:2602.08131
Fields of the World is one of several sources in FairGrow. The satellite imagery itself comes from Copernicus, and in the Netherlands field and crop come from the Basisregistratie Gewaspercelen of RVO, via PDOK.
See which fields lie around your location
Create your account and add your location. FairGrow fetches the field boundaries nearby, so that soon you can point out your own fields instead of drawing them.
Create your account → About Copernicus →