---

name: geoparquet description: "Use the GeoParquet specification and bundled inspection helpers to author, validate, distribute, or safely review geospatial Parquet data, including GeoParquet 2.0 vector metadata, external writer choices, and the alpha Parquet Raster proposal." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

GeoParquet

Use this skill when a task involves GeoParquet, Parquet GEOMETRY/GEOGRAPHY logical types, WKB geometry columns, the Parquet geo metadata key, CRS/PROJJSON, spatial row-group statistics, GeoParquet distribution, STAC publication, or the alpha Parquet Raster proposal.

This repository is a specification and example suite, not an installable Python writer package. Do not tell a user to install geoparquet as if it were a library. For local inspection helpers, install the dependencies appropriate to the task, then run a neutral import check:

python -m pip install "pyarrow>=22" "jsonschema>=4" "geoarrow-pyarrow>=0.2"
python -c "import pyarrow, jsonschema; print(pyarrow.__version__)"

The helper scripts are self-contained and offline by default. They do not fetch PROJJSON schemas, resolve authority registries, install external writers, or contact out-of-database raster URIs. Keep the exact writer/reader versions and raw inspection output with any conformance or publication decision.

Route by task

• Validate or inspect a vector file: read sub-skills/validate-geoparquet/SKILL.md. Use it for geo JSON, native logical types, WKB layout, CRS, geometry types, bbox, and row-group geospatial statistics. It distinguishes a full 2.0 file from metadata-only or native-type-only Parquet.
• Plan a distribution or choose a writer: read sub-skills/distribute-geoparquet/SKILL.md. Use it for GDAL/OGR, DuckDB, gpio/geoparquet-io, Sedona, compression, ordering, row groups, spatial partitioning, STAC, and old-reader tradeoffs.
• Review raster-in-Parquet designs: read sub-skills/parquet-raster/SKILL.md. Treat it as an alpha proposal only; do not transfer vector conformance rules to raster columns.

Shared operating rules

1. Establish the claimed GeoParquet version, file or dataset scope, writer and reader versions, expected CRS, and compatibility requirements before giving a pass/fail result.
2. Inspect the Parquet footer, not only the filename or a successful generic Parquet read. GeoParquet 2.0 requires native root-level, non-repeated BYTE_ARRAY geometry columns annotated GEOMETRY or GEOGRAPHY, plus a JSON geo metadata block.
3. Validate metadata locally, then reconcile metadata with physical schema, WKB representation, native CRS, geometry-type statistics, and row-group extents. Mark unresolved PROJJSON or CRS-equivalence checks as not verified, never as passed.
4. Choose GeoParquet 2.0 for native geometry types and geospatial statistics when the reader matrix supports them. Choose 1.1 with bbox covering when older readers require it, and label the artifact with its actual version.
5. Before publication, plan compression, spatial ordering, row groups, partitioning, and STAC separately from logical conformance. A valid bbox or native statistic does not make unsorted data fast.
6. Never fetch an out-db raster URI during review. Preserve alpha-proposal ambiguities and external dependencies in the result.

Read references/troubleshooting.md for cross-cutting recovery guidance, and read references/repo-provenance.md before deciding whether this graph is stale relative to a repository checkout.

Scope limits

This graph does not implement a general GeoParquet writer, run production-scale conversion or benchmarks, install external tools, resolve every CRS authority, or certify the alpha raster proposal. It does include safe bundled vector metadata/footer helpers; their ownership and usage are described by the validation sub-skill.