The Global Fields of The World - 2nd Edition release provides updated global-scale estimates of agricultural fields for 2017-2025. The dataset is run on the Sentinel-2 Level 3 Quarterly Cloudless Mosaics from CDSE rehosted by Taylor Geospatial on Source Cooperative here. The data product contains raw prediction COGs, vectorized GeoParquet, and PMTiles for visualization.
This product is sponsored and developed by Taylor Geospatial in partnership with the FTW organization and its partners.
The 2018, 2022, 2023, 2024 and 2025 collections use the updated processing documented in their year collections. Other year links retain historical published collections; they do not imply completion of the updated recipe for those years. Consult each collection for its provenance, footprint and asset inventory.
Fields of the World global field-boundary predictions, organized by year as GeoParquet vectors and probability COGs. Both derive from the TGE Labs Sentinel-2 quarterly cloudless mosaics, and both are part of Fields of the World.
These historical illustrations predate the updated 2018/2022/2023/2024/2025 processing; they are not validation of the updated outputs.
Sorriso, Brazil, 2025 Q3. Large fields, same imagery in both panels.
Normandy, France, 2025 Q3. Irregular fields and hedgerows, same imagery in both panels.
Matopiba, Brazil. New rectangular field blocks.
West Bahia, Brazil. Rectangular fields replaced by center pivots.
Imagery is Copernicus Sentinel data 2017–2025 from the TGE Labs quarterly mosaics, outlines are FTW predictions, and these are selected examples, not an accuracy assessment.
https://data.source.coop/ftw/global-data-2e/.AGENTS.md beside this file is the guide for agents. Read Limitations before drawing conclusions from any of these numbers.
Data license: CC-BY-4.0
The vector product is one GeoParquet file per UTM zone per year, at vector/{year}/zone=NN/utm{NN}.parquet, with a PMTiles archive per year for maps.
The vector README has a runnable query.
The raster product is one COG per tile at 2.5 m, at raster/{year}/…/{tile}.tif, with field (band 1) and boundary (band 2) probabilities.
All years share one tile grid, so a tile can be compared across years.
Find tiles in the raster index.
The raster README has the details.
The vectors are polygons derived from the same probabilities as the rasters, so the two are not independent measurements.
For 2018, 2022, 2023, 2024 and 2025, the observed model is ftw-q-unet-efb3.onnx, SHA256 d14285f91870a79833aac3529650f385afec605ed415acdf2227a02a0a88d036. The four Sentinel-2 quarterly mosaics supply the inputs; each raster item records its exact quarter IDs and model provenance.
For 2018, 2022, 2023, 2024 and 2025, extraction uses a minimum mean field probability of 0.10, fills enclosed holes strictly smaller than 100 m², and uses a minimum area of 900 m². Independent crop filtering evaluates polygons with finite area from 900 through 5,000,000 m² against the union of native cropland observations from 2017, 2020 and 2024, removing polygons with less than 5% overlap. Unknown or incomplete coverage, nonfinite areas and polygons outside that size interval are retained by this filtering stage. Downstream vector cleanup is documented separately in each vector collection; its 20 m² hole-cleanup description is distinct from extraction hole filling.
These are model predictions, not a survey. A field here is a remote-sensing field unit, a connected component of predicted field-interior pixels, and not a cadastral or legal parcel. This is not a land-tenure product, and one legal parcel may map to many polygons or to none. Counts, areas and perimeters are predicted quantities that carry the model's errors. See Fields of the World for the project and its definitions.
score column is the model's mean field probability inside the parcel (0 to 100), not a calibrated probability, so a score of 80 is not an 80% chance that the parcel is real.score over a hard threshold there.id carries no meaning across years, so compare years in the vectors with a spatial join.
The rasters share one grid and compare per pixel.Vector GeoParquet is WGS 84 lon/lat (EPSG:4326) in every zone file, so ST_Area on geometry returns square degrees, so read metrics:area (m²) instead.
Each COG is in its own tile's UTM zone, so a mosaic across zones needs a warp.
PMTiles are Web Mercator (EPSG:3857).
See the raster README for a short rasterio example that mosaics tiles for a lon/lat box.
catalog/ in fieldsoftheworld/ftw-global-data-catalog is this catalog, and the same repository holds the processing pipeline.
Report problems in the issue tracker.
| 2023 | Updated | 14,590 | 217,506,501 |
| 2024 | Updated | 14,593 | 204,609,812 |
| 2025 | Updated | 14,590 | 225,519,173 |