Agricultural field boundaries produced by Trazo models for key geographies in South America for the 2023-2024 planting season. Post-processed.
# Trazo1 Field Boundaries — Brazil (2023–2024)
> Agricultural field boundaries for Mato Grosso do Sul and Paraná, Brazil, delineated from
> Sentinel-2 imagery by the Trazo1 model. WRI / Arizona State University, CC-BY-4.0.
Part of the [Trazo catalog](../catalog.json). Trazo1 is the **most boundary-sensitive** model in
the suite, because of its heavy boundary weighting: class weights 0.88 boundary / 0.04 background
/ 0.08 interior (LR 0.001, 100 epochs), against the rebalanced 0.75/0.05/0.20 used by Trazo2. That
makes it well suited to soft boundaries — those defined by two abutting field interiors rather
than a hard feature such as a road — in large, uniform row-crop systems.
It posts the highest SA Soy pixel IoU of the suite (77.09%, versus 47.69% for the Fields of The
World baseline) but is markedly weaker outside soy (66.57% Mato Grosso, 57.76% Chiquitania),
having been trained on only 559 South American chips from 10 of the 17 soy ecoregions.
[Model card](https://huggingface.co/worldresourcesinstitute/Trazo1).
Outside soy-dominated row cropping, prefer [trazo3-fields](../trazo3-fields/).
## Files (2)
- `brazil_mato_grosso_do_sul_2024.parquet`
- `brazil_parana_2024.parquet`
```sql
SELECT count(*) FROM read_parquet('trazo1-fields/*.parquet');
```
## Schema
| Column | Meaning |
|---|---|
| `geometry` | MultiPolygon field boundary, EPSG:4326 |
| `Id` | auto-generated feature id; no external meaning |
| `mbmode24` | MapBiomas 2024 agriculture class, area-weighted mode (Table A1 of the technical note) |
| `mbcov_area_2024`, `mbvalid_area_2024` | QA: MapBiomas raster coverage / valid area (m²) |
Trazo1 exposes the same attribute columns as Trazo2. Trazo3 additionally carries Hansen
forest-loss columns.
## Caveats
Boundaries are model-derived and have not been field-verified. Object-level metrics on the model
card use a punishing 50% IoU matching threshold that systematically depresses them — treat pixel
metrics as the headline. Optimized for annual crops (soy, corn); pasture and tree-crop performance
is untested, and agroforestry may be undetectable at Sentinel-2 resolution. **Not intended to be
used alone** to support deforestation- or conversion-free claims.