Overview
Trazo is a deep learning training dataset and suite of field boundary segmentation models developed to support sustainable supply chain monitoring across South America. Built on the Fields of The World (FTW) baseline dataset and model architecture, Trazo extends FTW to the varied and challenging agricultural landscapes of South America. Read about Trazo.
Trazo provides a large-scale dataset of over 10.9 million agricultural field boundaries across South America, generated by the Trazo deep learning models and refined through post-processing. The performance of each model and its use case are described in the technical note and associated model checkpoint documentation.
This dataset represents one of the most comprehensive open field boundary datasets specialized to and available for South American agricultural regions, with strong coverage of major soy-producing areas. These fields were generated for the 2023–2024 planting season. This dataset offers a complement to Global Fields of the World.
Field boundaries were generated from Sentinel-2 imagery using the Trazo model suite, based on a U-Net architecture trained on locally annotated training data spanning 17 soy-producing ecoregions. All boundaries underwent post-processing to improve accuracy and reduce false positives. Each dataset is presented under a directory named after the model which produced it. Each model's inference result directory has a geopackage folder and a parquet folder. We present every dataset as a geopackage and parquet. Each fields boundary dataset has an "mbmode" column. This is the mode value of Mapbiomas's classification of agriculture in 2024, see table A1.
We reccomend Trazo3 for most use cases. Trazo2 is the best evaluated and characterized model. Trazo1 has a greater sensitivity for soft boundaries and might be most useful in large scale row cropping systems where these soft boundaries — a boundary defined by two abutting field interiors rather than a hard boundary like a road — are subtle.
Example Outputs (Trazo3)
Coverage
The dataset contains 10,935,129 field boundaries across South America.
Regional Coverage with Model and Temporal Windows
These are the field counts of various states produced by different models. These counts are examples from a mix of models. For most use cases, use Trazo3 fields.
Intended Use & Relevance
Field boundaries are the foundational unit of supply chain traceability. By linking a food product to a specific plot of land, supply chain actors can determine whether that land was subject to deforestation or ecosystem conversion.
Field plots assessed as deforestation-free after the end of 2020 is a core requirement of the EU Deforestation Regulation (EUDR), which applies to soy, palm oil, cattle, cocoa, coffee, rubber, and timber placed on the EU market.
This dataset is intended to support:
- Supply chain risk assessment for soy and other crops (EUDR use cases)
- Deforestation and land conversion monitoring at the field level
- Geospatial research on agricultural land use in South America
- Benchmarking and evaluation of field boundary detection methods
Data Generation & Quality
Boundaries were generated using the Trazo model suite, trained on manually annotated data spanning 17 soy-producing ecoregions.
Processing steps include:
- Deep learning inference using Trazo models (Trazo1, Trazo2, Trazo SA, Trazo3)
- Post-processing to remove artifacts and improve geometry
- Regional expert review to assess coverage and quality
For full technical details on:
- Model architecture
- Training data
- Performance benchmarks
Refer to:
- The Technical Note
- Trazo Training Data
- Trazo Model Checkpoints
Acknowledgments
This dataset was produced in collaboration with WRI and Arizona State University, with support from the Land and Carbon Lab. This work was funded through a Walmart Foundation grant.
Citation, License and Contact
Creative Commons Attribution 4.0 International (CC BY 4.0)
Free to use, share, and adapt for any purpose, including commercial applications such as supply chain traceability, provided that Trazo / WRI is properly cited.
Grupp et al. (2026) — Field Boundaries of South America. World Resources Institute & Arizona State University. Technical Note
Corresponding author: Tristan Grupp - tristan.grupp@wri.org
Kerner et al. (2025) — Fields of The World: A Machine Learning Benchmark Dataset For Global Agricultural Field Boundary Segmentation AAAI 2025, vol. 39(27), pp. 28151–28159. DOI: 10.1609/aaai.v39i27.35034
Corresponding author: Hannah Kerner - hkerner@asu.edu