South American Field Boundaries

Agricultural field boundaries produced by Trazo models for key geographies in South America for the 2023-2024 planting season. Post-processed.

Contents

root

trazo2-fields

Details

Visibility: Public
Owner: WRI Data Lab
Created: 22 Jun 2026
Last Updated: 17 Jul 2026

Analytics

28 days

Downloads

11,698

Data served

909.5 GB

Countries

28-day downloads

9 Jul 2026

5 Aug 2026

Analytics

README

Trazo2 Field Boundaries — Brazil & Paraguay (2023–2024)

Agricultural field boundaries for five Brazilian states (Goiás, Pará, Piauí, São Paulo, Tocantins) and Paraguay (Chaco and non-Chaco), delineated from Sentinel-2 imagery by the Trazo2 model for the 2023–2024 planting season. Trazo2 is the best evaluated and characterized model in the suite. Files are named {model}{country}{region}_{year}; read them all with the glob trazo2-fields/*.parquet. Full-attribute GeoPackages for desktop GIS are in geopackages/.

Spatial Coverage

Bounding Box: [-62.99413046374593, -28.106997121038603, -41.282401930088895, 1.34858515435252]

Temporal Coverage

Start: 2023-07-01T00:00:00Z
End: 2024-06-30T23:59:59Z

Schema

Column	Type	Description
Id	int32
mbmode24	double
mbcov_area_2024	double
mbvalid_area_2024	double
geometry	binary

Files

File	Size	Checksum
./trazo2_paraguay_non_chaco_2024.parquet	-	-
./llms.txt	-	-
./README.md	-	-
./trazo2_brazil_goias_2024.parquet	-	-
./trazo2_brazil_para_2024.parquet	-	-
./trazo2_brazil_piaui_2024.parquet	-	-
./trazo2_brazil_sao_paulo_2024.parquet	-	-
./trazo2_brazil_tocantins_2024.parquet	-	-
./trazo2_paraguay_chaco_2024.parquet	-	-
./geopackages/trazo2_paraguay_non_chaco_2024.gpkg	-	-
./geopackages/trazo2_brazil_goias_2024.gpkg	-	-
./geopackages/trazo2_brazil_para_2024.gpkg

Quick Start

STAC Metadata

root: ../catalog.json
parent: ../catalog.json

Source

https://huggingface.co/worldresourcesinstitute/Trazo2

Processing Notes

The best evaluated and characterized model in the suite — a good choice where documented performance matters. Covers five Brazilian states (Goiás, Pará, Piauí, São Paulo, Tocantins) and Paraguay (Chaco and non-Chaco).

U-Net with EfficientNet-B3 backbone over Sentinel-2 chips (256×256 px, RGB + NIR, two seasonal windows). Trained on the full global FTW dataset combined with all 17 South American ecoregions and partner data, with rebalanced class weights (0.75 boundary / 0.05 background / 0.20 interior).

Test set	Pixel IoU	Pixel precision	Pixel recall
Mato Grosso (2,462 fields)	76.27%	87.19%	85.90%
SA Soy (1,400 fields)	76.05%	91.01%	82.22%
Chiquitania (3,961 fields)	66.61%	85.56%	75.04%

Trazo2 model card

Bulk access. Files are partitioned by country; read them all with the glob trazo2-fields/*/*.parquet.

Each file carries the mbmode24 column: the mode of MapBiomas' 2024 agriculture classification (see Table A1 of the technical note).

Citation

Grupp et al. (2026). Field Boundaries of South America. World Resources Institute & Arizona State University.

Attribution

Trazo / World Resources Institute

License

CC-BY-4.0

Contact

Tristan Grupp — World Resources Institute tristan.grupp@wri.org

Known Issues

Trazo2 is the most thoroughly evaluated model, but for most general use cases trazo3-fields is recommended for its broader coverage and stronger Mato Grosso performance.

Object-level metrics on the model card (object precision 34.83%, object recall 28.37% on SA Soy) use a punishing 50% IoU object-matching threshold, which systematically depresses them. Treat the pixel metrics as the headline figure.

Trazo2 exposes the same attribute columns as Trazo1 — the models themselves are distinct, with different training data and class weighting. Trazo3 adds Hansen forest-loss columns.

Strongest for annual crops (soy, corn, other row crops); not designed for agroforestry or other matrixed agricultural contexts. Performance degrades in underrepresented regions or cropping systems. Boundaries are model-derived and have not been field-verified.

Generated by Portolan from STAC metadata and .portolan/metadata.yaml