Trazo Field Boundary Annotations (fiboa)

Field boundary annotations from the Trazo project (WRI & Arizona State University), converted to fiboa GeoParquet.

Trazo extends the Fields of The World (FTW) baseline with 60,386 manually annotated field boundary polygons across 17 South American ecoregions — the Chaco, Chiquitania, Araucaria, Amazon, Cerrado, Caatinga, Pampas and others — created to adapt field boundary segmentation models to South American agricultural landscapes.

Relationship to FTW (no duplication)

This dataset contains only the new Trazo annotations. The upstream Trazo repository intentionally excludes FTW data ("intended to be used in conjunction with the FTW benchmark dataset"), and we verified this spatially: FTW's only South American subset (LEM, West Bahia, Brazil — 1,854 fields) has zero polygon intersections with this dataset. You can safely combine this dataset with the FTW benchmark without deduplication.

Contents

• wri_trazo.parquet — fiboa GeoParquet (fiboa 0.3.0, GeoParquet 1.1.0), 60,386 features, EPSG:4326
• stac/collection.json — STAC Collection metadata

Columns

Regions prefixed test (testchiquitania, testmatogrosso, testsoy) were truly randomly sampled and are the statistically rigorous choice for model evaluation. Other regions were selected for poor baseline model performance, active-learning query strategies (matogrosso_brazil_*), or partner data availability (maparaguay — Mundo Agropecuario), and are best used for training/validation.

Lineage

• Source: source.coop/wri-data-lab/trazoannotations (per-ecoregion GeoPackages)
• Normalization: reprojected to EPSG:4326, geometries made valid, multipart polygons exploded, Z dimension dropped; per-feature year filled from the source filename where the column was absent (test folders)
• Converted with fiboa-cli 0.21.0 using the wri_trazo converter

License and Citation

Creative Commons Attribution 4.0 International (CC-BY-4.0). Please cite:

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

When used together with the FTW benchmark, also cite:

Kerner et al. (2025) — Fields of The World: A Machine Learning Benchmark Dataset for Global Agricultural Field Boundary Segmentation. AAAI 2025, 39(27), 28151–28159. DOI: 10.1609/aaai.v39i27.35034

Trazo was produced by WRI and Arizona State University with support from the Land and Carbon Lab, funded through a Walmart Foundation grant.