Zoe/GS-QA2-source

GS-QA2 Source Data

Reference geospatial data underlying GS-QA2, a benchmark for question answering over raster–vector data. This repository contains the raw vector and raster inputs needed to rebuild the reference PostGIS database, execute the benchmark's ground-truth SQL queries, and run the baselines — use it together with the QA pairs in Zoe/GS-QA2 and the code in github.com/ZhuochengShang/QARV.

Contents

Path	Contents	Format
osm/osm_extract/{lakes,parks,pois,postal_codes,roads}/	Raw OSM extracts of the United States (~53 GB, 3,152 files) — the exact input to the benchmark's ingestion scripts	GeoJSON
dem/tiles/*.tif	1,182 ASTER GDEM v3 1°×1° GeoTIFF tiles (1 arc-second ≈ 30 m, EPSG:4326) covering the contiguous United States (~18 GB)	GeoTIFF
dem/tile_list.txt	Listing of all DEM tile filenames	text
dem/needed_dem_tiles.txt	The tile set required by the benchmark (usable to re-download from NASA Earthdata instead of fetching dem/tiles/)	text

Rebuilding the reference database

The benchmark's ground-truth SQL expects PostGIS tables pois, roads, parks, lakes, regions and a raster table public.dem_us. Follow GS-QA/ingestion/ in the QARV repository:

1. Set DATA_ROOT to the downloaded osm/ directory (so the layers are at $DATA_ROOT/osm_extract/{lakes,parks,pois,postal_codes,roads}) and load the vector tables with ingest_osm_postgis.sh (which uses the *_processor.py loaders and schema files from GS-QA/generator/).
2. Set DEM_ROOT to the downloaded dem/tiles/ directory and load with ingest_dem_postgis.sh. The script runs raster2pgsql with 256×256 tiling — the 1,182 source GeoTIFFs become the 265,950 in-database raster tiles reported in the paper — and builds a GiST spatial index.

All geometries and rasters use EPSG:4326 (WGS 84).

Provenance

• The OSM extracts were produced with osmx from Geofabrik United States extracts, by the authors of the original GS-QA benchmark (also mirrored at this Google Drive folder).
• The DEM tiles are ASTER GDEM Version 3 (ASTGTM v003) granules obtained from NASA Earthdata / LP DAAC; dem/needed_dem_tiles.txt identifies the granules so they can equally be re-downloaded from the source.

Licensing and attribution

• OSM-derived data (osm/): © OpenStreetMap contributors, licensed under the Open Database License (ODbL) 1.0. Any derived database must comply with ODbL share-alike terms.
• DEM tiles (dem/): ASTER GDEM is a product of METI and NASA. ASTER GDEM v3 data are freely available and redistributable; retain this attribution. See the ASTGTM v003 documentation.

Citation

@inproceedings{shang2026gsqa2,
  title     = {GS-QA2: A Benchmark for Question Answering over Raster--Vector Data},
  author    = {Shang, Zhuocheng and Elmahallawy, Shahd and Al Nazi, Zabir and Hristidis, Vagelis and Eldawy, Ahmed},
  year      = {2026},
  note      = {Benchmark and code: https://github.com/ZhuochengShang/QARV}
}

@article{saeedan2026gsqa,
  title   = {GS-QA: A Benchmark for Geospatial Question Answering},
  author  = {Saeedan, Majid and Shihab Rashid, Muhammad and Eldawy, Ahmed and Hristidis, Vagelis},
  journal = {arXiv preprint arXiv:2605.22811},
  year    = {2026}
}

Contact

Zhuocheng Shang — zshan011@ucr.edu — University of California, Riverside