Agent Guide: MapBiomas Annual Land Cover and Land Use, South America

Read this before you query the catalog. Every number here is measured from the STAC objects in this catalog or from a listing of the bucket. Nothing is quoted from memory.

Machine entry point: https://data.source.coop/tristangruppwri/mapbiomas/catalog.json Human entry point: https://source.coop/tristangruppwri/mapbiomas Source repository: https://github.com/wri/rural-land

What This Is

Annual land cover and land use maps from the MapBiomas Project for ten South American countries, 2019-2024, as Cloud-Optimized GeoTIFFs at 30 m. One collection per country, one item per year. Pixel values are the raw national MapBiomas class codes - nothing was reclassified. Because each MapBiomas national initiative maintains its own legend, codes are not comparable across borders as published; the README documents a three-tier harmonisation (exact level-1 collapse, a pan-MapBiomas legend, and a corrected crosswalk onto MapBiomas Brazil Collection 10) and ships it as a machine-readable harmonization.csv.

The catalog is a mirror. MapBiomas produces the maps and licenses them. Source Cooperative serves the files. See providers on any collection.

Shape

11 collections, 59 items. One item per country-year. One Cloud-Optimized GeoTIFF per item, at 30 m in EPSG:4326, uint8, nodata 0.

The One Thing That Breaks Analyses

The same integer means different things in different countries. Code 4 is Savanna Formation in Brazil and Open forests in Argentina. Code 50 is Herbaceous Sandbank Vegetation in Brazil and Xerophytic grassland in Venezuela. Code 63 is Shrub and herbaceous mosaic in Argentina and Steppe in Chile.

Never compare raw codes across collections. Bolivia and Chile also change MapBiomas collection part way through their own year range, so their years are not comparable to each other either.

Three crosswalks are published, one exact and two lossy, in https://data.source.coop/tristangruppwri/mapbiomas/provenance/harmonization.csv. The catalog README explains which tier to use. Apply one before any cross-country statistic.

Reading a Pixel

Every COG embeds the official MapBiomas colour table for its country, and one CLASS_<code> TIFF tag per legend entry, so the file is self-describing. The same legend is in the STAC as classification:classes on the data asset's band, with color_hint carrying the official colour. Use that rather than a palette of your own.

The bucket answers range requests, so a windowed read fetches only the tiles it needs. It does not need credentials.

Before You Query

Check tools/availability.py in the source repository, or the Availability section of the catalog README. Some declared rasters are not in the bucket yet and return HTTP 404.

Conformance

Portolan profile https://schemas.portolan-sdi.org/portolan/v0.1.2/schema.json. Validated with rashid 0.1.7. Known deviations are listed in docs/conformance.md in the source repository, each with the reason it stands.