---
license: odbl
task_categories:
  - other
language:
  - en
tags:
  - geospatial
  - openstreetmap
  - osm
  - polygons
  - landuse
  - landcover
  - remote-sensing
  - foundation-model
size_categories:
  - 1M<n<10M
---

# osm-polygon-selection dataset

A curated set of OpenStreetMap polygons across 33
European countries, classified by size bin (small/medium/large,
area in [0.1, 100] km²) and tagged by continent (Natural Earth
admin0 lookup).

## Geographic distribution

![Polygon distribution across Europe](map_preview.png)

The map above shows a stratified sample of 1,200 polygons from the
dataset (color-coded by country, sized by area). It gives a rough
visual sense of which European regions are well-covered and which
are sparse.

## Files

- `all_europe.parquet` — all 1,067,098 polygons in a single file
- `<country>.parquet` — per-country files
- `map_preview.png` — geographic distribution preview (this README)
- `metadata.yaml` — HuggingFace dataset viewer metadata
- `manifest.json` — machine-readable build manifest
- `README.md` — this file

## Resources

To understand where this dataset comes from and how the filters
were designed, see these resources:

- **Blog post** (long-form context): [OSM data analysis for landuse](https://noeflandre.com/posts/osm-data-analysis) on noeflandre.com. Explains the osm-stats clustering that produced the whitelist, including the arithmetic correction (the blog originally claimed 326 union base keys; the correct value is 236).
- **GitHub repo** (code + documentation): [NoeFlandre/osm-polygon-selection](https://github.com/NoeFlandre/osm-polygon-selection). The pipeline source code, whitelist decisions, and dataset-state documentation.
- **Related repo** (whitelist source data): [NoeFlandre/osm-stats](https://github.com/NoeFlandre/osm-stats). The TF-IDF and embeddings analyses on OSM tags that fed into the 22,075-tag whitelist.
- **Project repo docs** (in-repo documentation): `docs/whitelist_decisions.md` (Tier A / Tier B inclusion rules) and `docs/dataset_state.md` (per-country coverage breakdown and known issues).

## Schema

| column | type | description |
|--------|------|-------------|
| osm_id | int64 | OSM object id |
| osm_type | string | "way" or "relation" |
| centroid_lon | float64 | polygon centroid longitude (WGS84) |
| centroid_lat | float64 | polygon centroid latitude (WGS84) |
| area_km2 | float64 | polygon area in km² (Web Mercator, accurate at mid-latitudes) |
| tags | list(string) | OSM `key=value` tags |
| continent | string | Natural Earth admin0 lookup of the centroid |
| size_bin | string | "small" (0.1-1), "medium" (1-10), or "large" (10-100) km² |
| country | string | ISO-style country name |
| extract_status | string | "clean" (extract process ran to completion) or "killed" (extract was interrupted before completion) |
| pbf_date | string | date of the source PBF file (from mtime) |

## Provenance

- Pipeline version: v0.1.0
- Git SHA: 6b7ce5e
- Built: 2026-06-27T14:21:35.187579
- Source: Geofabrik regional extracts (`https://download.geofabrik.de/`)
- Whitelist: 22,075 OSM `key=value` tags from osm-stats (see
  `docs/whitelist_decisions.md` in the project repo)

## Filter chain

Each polygon in this dataset has passed three filters:

1. **Size filter (Stage 0)**: area in [0.1, 100] km².
   Polygons smaller than 0.1 km² or larger than 100 km² are dropped.
2. **Whitelist filter (Stage 2)**: at least one OSM tag in the
   22,075-tag whitelist. The whitelist is derived from a clustering
   of OSM tags across both `tfidf` and `embeddings` analyses.
3. **Classify (Stage 3)**: continent assigned via Natural Earth
   admin0 shapefile, size_bin assigned by area.

## Known issues

Some countries in this dataset (those with `extract_status = "killed"`)
had their Stage 0 extract process interrupted before the entire PBF
was yielded. The yielded polygons that made it to disk are valid and
complete, but a small tail of polygons (typically 5-10% of the
country's total) is missing. See `docs/dataset_state.md` in the repo
for the full list. To complete them, re-run `scripts/run_country.sh
<country>` — the `.seen_ids` WAL preserves the work already done.

## Per-country summary
- albania: 8,423 polygons (killed)
- andorra: 776 polygons (clean)
- austria: 133,711 polygons (clean)
- azores: 2,640 polygons (clean)
- belarus: 32,828 polygons (killed)
- belgium: 108,572 polygons (killed)
- bosnia-herzegovina: 19,252 polygons (killed)
- bulgaria: 21,405 polygons (killed)
- croatia: 13,038 polygons (killed)
- cyprus: 4,846 polygons (clean)
- czech-republic: 269,806 polygons (killed)
- denmark: 175,795 polygons (clean)
- estonia: 28,134 polygons (killed)
- faroe-islands: 1,278 polygons (clean)
- guernsey-jersey: 670 polygons (clean)
- hungary: 6,715 polygons (killed)
- iceland: 27,749 polygons (killed)
- isle-of-man: 2,648 polygons (clean)
- kosovo: 109 polygons (killed)
- latvia: 18,679 polygons (killed)
- liechtenstein: 565 polygons (clean)
- lithuania: 13,363 polygons (killed)
- luxembourg: 11,460 polygons (clean)
- malta: 620 polygons (clean)
- moldova: 15,434 polygons (killed)
- monaco: 2 polygons (clean)
- montenegro: 11,785 polygons (clean)
- portugal: 66,287 polygons (clean)
- romania: 694 polygons (killed)
- serbia: 6,138 polygons (killed)
- slovakia: 1,887 polygons (killed)
- slovenia: 633 polygons (killed)
- switzerland: 61,156 polygons (clean)