Trazo1 Field Boundaries — Early Adopters Program (2025)
Superseded — not the recommended data. For current work use trazo3-fields. Field boundaries from the Trazo1 model, shared through the Early Adopters Program in late 2025. Nine regions across Argentina, Brazil and Paraguay — the only EAP collection that includes Paraná. brazil_rio_grande_do_sul is the "grassland left in" variant.
Spatial Coverage
- Bounding Box: [-67.68489242903367, -35.321528124461906, -41.28218977809995, 1.4457373426083768]
Temporal Coverage
- Start: 2025-06-17T00:00:00Z
- End: 2025-10-29T23:59:59Z
Schema
Files
Quick Start
STAC Metadata
- root:
../../catalog.json - parent:
../../catalog.json
Source
https://www.wri.org/research/field-boundaries-south-america
Processing Notes
Not the recommended data. For current work use the main collections —
trazo3-fieldsfor most use cases. This directory holds the earlier boundaries described in the technical note.
These are the field boundaries reported in the Trazo technical note, representing the state of the work in late 2025. EAP stands for Early Adopters Program: we shared these datasets with end users and collected structured feedback on them. They are published here for three reasons — the technical note documents them, they carry the user feedback summarised below, and together they show how the data product evolved.
Organised by model, mirroring the main catalog. Model characteristics are documented in the technical note and in each model's Source Cooperative folder; use Trazo3 for most cases and select another model only when its specific characteristics suit your work.
All files are {country}_{region}_{year}.parquet, reprojected to EPSG:4326 from their
original UTM zones, and share the schema det_method, id, area, perimeter, geometry
(area in hectares, perimeter in metres).
Trazo on the ground
What follows is end-user feedback gathered through the Early Adopters Program.
Barriers users face
Users described the obstacles they meet in their work on sustainable agriculture:
"Existing forest cover loss remote sensing products are coarse. Products focused on agriculture are often paywalled." — Environmental NGO
"We have very little up-to-date information regarding the agricultural areas of the territories in which we work." — Research Organization
"The main challenge in achieving these goals is that the development of methodologies to achieve them first requires extensive data generation and gathering. One of the key inputs consists of an updated vector layer of individual agricultural fields." — Private bank investing in sustainable agriculture
"Insufficient access to modern agricultural technology and data systems restricts farmers' productivity and resilience." — Farmer advocacy cooperative
The need for a "solid MRV for impact monitoring and clear claim framework for commercialisation of impact." — Grain trader
"Complex spatial analysis and lack of traceability." — Environmental sustainability consultant
Scaling beyond manual mapping
Users described how manual methods limit their reach:
"We have a database of boundaries collected through a variety of means but need to scale to more areas." — Agricultural analytics company
"Staff time for smallholder parcel mapping is considerable. Recently we mapped roughly 700 ha of rubber plantations. This required 3 teams of 5 people each, working over 1 month. This creates well-defined plot boundaries but is a significant commitment per project." — Carbon capture company
"Field boundary data either comes from the local land manager or hired consultants." — NGO
"For regional analyses, we use Varda's Global Field ID product. However, we have identified several inconsistencies and data gaps in the regions where we operate. We have considered developing our own model for field boundary delineation, but we currently face time and team constraints." — Service provider
Automated boundaries from a single model version give a consistent output for comparative analysis and fill these data gaps.
Feedback on the Trazo data
Users assessed the datasets in this directory directly:
"The data strongly aligns with our expectations. The precision, spatial coherence, and compatibility with national land-use datasets make it highly useful for monitoring, planning, and agricultural intelligence. It supports ongoing workflows related to production estimation, land-use classification, and farm-level analysis. The delineation accuracy is remarkably high, especially in the Chaco region, where the boundaries aligned very well with our land-use planning layers. The outputs were more consistent and detailed than anticipated for such a heterogeneous landscape." — Geotechnical Specialist, Mundo Agropecuario
"The spatial patterns and visual cues in the data were helpful because they provide a basis for identifying field boundaries, which could become even more useful as the model improves." — GIS Analyst, agricultural service provider
"Las parcelas quedan muy bien delineadas a partir de cierto tamaño … Los datos ajustan mejor a sistemas agrícolas o sistemas ganaderos en pasturas puras sin árboles. Pero tienen problema para detectar los contornos de las parcelas silvopastoriles o silvícolas puros (cultivos de árboles)." — Investigator, national agricultural technical institute
"Most farms were identified so the layer would be very useful for aggregated agricultural area measurements." — Senior Data Scientist, private sector
"The dataset was generally very clear. In a few highly fragmented areas, small-holder plot boundaries were slightly harder to interpret, but this is expected and understandable given the inherent landscape complexity." — Geotechnical Specialist, Mundo Agropecuario
"The model's data performance was generally in line with what I expected. More specifically, I expected a high-quality result but anticipated that complex areas (for example, fields that differ from the average in the Chaco region with different textures, diverse geometries, color, etc.) would be challenging, which was largely the case. Our organization is interested in accurate and up-to-date data on agricultural and livestock plots, both for monitoring deforestation and for tracking productivity, degradation status, and the implementation of sustainable practices in these areas. The data format aligns perfectly with what we typically use and prefer, and the spatial resolution is high.
These initial datasets demonstrated the quality that can be achieved with this methodology and align well with the needs mentioned, with the only exception being temporal coverage, as we require the most up-to-date data possible. However, this is not a major concern since the datasets that will be published next year will enable us to generate this type of information as frequently as needed.
It is also worth noting that we are highly interested in differentiating between crop types, and the segmentation models and training samples that will be shared next year will allow us to customize the model to better meet this need." — GIS Specialist, major bank investing in sustainable agriculture
Source: Trazo on the ground
Citation
Grupp et al. (2026). Field Boundaries of South America. World Resources Institute & Arizona State University.
Attribution
Trazo / World Resources Institute
License
Contact
Tristan Grupp — World Resources Institute tristan.grupp@wri.org
Known Issues
Superseded data. These boundaries represent the state of the work in late 2025 and
have since been improved. Prefer the main collections; trazo3-fields is recommended
for most use cases.
Narrower attributes. EAP files carry only det_method, id, area, perimeter. They do
not include the MapBiomas class (mbmode24) or the Hansen forest-loss columns available
in the released trazo3-fields.
Counts differ from the released data for the same region, because these come from different model runs. That is expected, not an error — Trazo1 and Trazo2 fragment fields more finely than Trazo3, so they report more boundaries over the same ground.
Boundaries are model-derived and have not been field-verified. As with all Trazo data, they are not intended to be used alone to support deforestation- or conversion-free claims.
Generated by Portolan from STAC metadata and .portolan/metadata.yaml