// the take
geobr is a Python/R package that downloads official Brazilian spatial datasets — municipalities, states, census tracts, schools, indigenous lands, health facilities, and more — as ready-to-use geopandas or sf objects with consistent projections. It's aimed at researchers, data scientists, and GIS people who'd otherwise be stitching together shapefiles from a dozen different government agencies by hand.
Dozens of datasets spanning multiple decades (some back to 1872) with a consistent CRS (SIRGAS2000) and a uniform function signature across all readers, so switching between geography types is low-friction. The Python side moved to a GeoParquet pipeline with a DuckDB backend, letting you run SQL — including spatial joins like ST_Within — directly against snapshots without loading full datasets into memory. Missing views auto-download on first use, so the package itself stays small. Both R (CRAN) and Python (PyPI) packages are maintained in parallel with their own CI and codecov.
Every first use of a dataset hits IPEA's servers at runtime with no documented offline mirror or fallback, so any pipeline built on this has a hard external dependency with no local cache guarantee beyond a session. Geometries default to a simplified 1:250,000 scale for fast rendering, which is a problem if someone uses the output for anything needing precise boundaries and doesn't catch that it's simplified. The available-years table has real gaps — some functions are marked dev-only and read_comparable_areas is flagged as temporarily suspended — so the API surface isn't fully stable across the catalog. Output quality is entirely downstream of IBGE/INEP/FUNAI/DataSUS source data with no visible versioning contract, so upstream schema or boundary changes can silently change what you get back.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →