GeoParquet Validation Skill

Guide users through GeoParquet workflows with a gpio-first approach: inspect, validate, optimize, and distribute GeoParquet files following current best practices.

What this skill is for

Use this skill when the user is working with GeoParquet files and needs:

• metadata inspection
• validation and auto-fix workflows
• conversion and optimization with gpio
• partitioning and publishing
• DuckDB support for heavier SQL transforms

This is not a general "anything about GeoParquet" skill. It is centered on the gpio toolchain.

Tools

gpio (geoparquet-io) - Preferred

Always prefer gpio for GeoParquet operations. It applies important best practices by default.

Installation:

pipx install --pre geoparquet-io
pip install --pre geoparquet-io
uv pip install --pre geoparquet-io

If gpio is missing, guide the user through installation before proceeding.

DuckDB - For Advanced Operations

Use DuckDB for complex SQL, joins, aggregations, or geometry operations.

pip install "duckdb>=1.5"

When using DuckDB, apply GeoParquet best practices manually:

• ORDER BY ST_Hilbert(geometry)
• COMPRESSION ZSTD with COMPRESSION_LEVEL 15
• ROW_GROUP_SIZE 100000
• validate output with gpio check all

Workflow

1. Inspect

gpio inspect <file>
gpio inspect stats <file>

Report row count, geometry type, CRS, columns, and file size.

gpio inspect and gpio check accept only (Geo)Parquet files. For non-Parquet sources (shapefile, GPKG, GeoJSON, FGB), skip pre-inspection — convert first, then inspect the output. Do not spend calls pre-inspecting a non-Parquet source unless the conversion fails; if you truly need source metadata first, use pyogrio.read_info("<src>").

2. Convert or optimize

gpio convert geoparquet <input> <output>   # input: any OGR-readable format (shp, gpkg, geojson, fgb, csv)
gpio convert geoparquet <input> <output> --compression-level 15

convert applies best practices by default (Hilbert sort, bbox covering column, ZSTD) and validates its own output — a clean convert rarely needs --fix afterwards.

3. Validate

gpio check all <file>
gpio check all <file> --fix --output <fixed>

check all passes when Spec Validation reports every check with a checkmark. Lines marked as informational (the "GeoParquet 2.0 is available" pointer in particular) are not failures — stay on 1.1.0, the widest-compatibility version, unless the user explicitly asks for 2.0. convert and check all already print file size and row-group stats; one gpio inspect on the final output is enough for reporting.

4. Scale based on size

• Small: single file, Hilbert sorted, bbox column
• Medium: single file, covering metadata, compression level 15
• Large: partition with kdtree/admin/quadkey and generate STAC

5. Publish

gpio publish stac <input> <output.json>
gpio publish upload <file> s3://bucket/path/

Quick Reference

# Inspect (Parquet input only)
gpio inspect <file>
gpio inspect stats <file>

# Convert (input: any OGR-readable format — shp, gpkg, geojson, fgb, csv)
gpio convert geoparquet <input> <output>
gpio convert geoparquet <input> <output> --compression-level 15

# Validate
gpio check all <file>
gpio check all <file> --fix --output <fixed>

# Extract
gpio extract <input> <output> --bbox "minx,miny,maxx,maxy"
gpio extract <input> <output> --where "column > value"

# Partition and publish
gpio partition kdtree <input> <output_dir> --max-rows-per-file 500000
gpio publish stac <input> <output.json>

References

• references/gpio-commands.md
• references/distribution-best-practices.md
• references/tool-comparison.md