sports-facilities-africa-osm

Technical repository for an open, reproducible pipeline that turns OpenStreetMap into a dataset of sports facilities across the African continent, then maps and analyzes it.

📄 Preprint: Construire les données manquantes à l'échelle continentale : une méthode reproductible à partir d'OpenStreetMap — Benoît Prieur (2026), Zenodo. https://doi.org/10.5281/zenodo.20738614

At continental scale, few domains have an open, homogeneous and comparable dataset. There is none for African sports infrastructure, so this repo builds one from OSM (imperfect, non-representative, but open and re-runnable) and publishes the analysis as a static site.

• 🌍 Site: https://thepriben.github.io/sports-facilities-africa-osm-2026/
• 📄 Preprint: https://doi.org/10.5281/zenodo.20738614
• 🗓️ Data snapshot: Geofabrik Africa extract; see SNAPSHOT_DATE in scripts/common.py

Dataset at a glance (snapshot 2026-06-15): 123,936 geolocated sports sites · 194 distinct sport tags · 61,679 football pitches. Figures are regenerated by the pipeline (see data/sport_counts.csv and docs/data/sport_counts.json).

---

Pipeline

Geofabrik africa-latest.osm.pbf
        │  osmium tags-filter (leisure=pitch | leisure=stadium | building=stadium)
        ▼
   filtered.osm.pbf
        │  osmium export -f geojsonseq  →  shapely (geometry → WKT)
        ▼
data/sports_facilities.csv            (geometry WKT, sport)   ← raw, kept intact
        │  explode ";" + normalize tags (scripts/common.py)
        ├─► data/sport_counts.csv      (sport, count)
        ├─► docs/data/sport_counts.json (top N + totals, for the site)
        └─► docs/maps/*.html           (Folium maps)

The first stage (download + osmium) is encapsulated in scripts/refresh_from_geofabrik.py. Because the continental extract is ~8 GB, the script downloads it to a temp folder, filters it, and deletes the .pbf afterwards (use --keep-pbf to keep it).

Repository layout

.
├── data/
│   ├── sports_facilities.csv   # raw extract: columns (geometry WKT, sport)
│   └── sport_counts.csv        # cleaned per-sport counts (generated)
├── scripts/
│   ├── common.py               # paths, SNAPSHOT_DATE, sport normalization, extract_point()
│   ├── refresh_from_geofabrik.py  # OSM → data/sports_facilities.csv
│   ├── build_counts.py         # counts CSV + docs/data/sport_counts.json
│   └── make_map.py             # Folium maps: heatmap | sport | british
├── docs/                       # GitHub Pages site (served from /docs)
│   ├── index.html
│   ├── assets/{css,js,img}/
│   ├── data/sport_counts.json
│   ├── maps/*.html
│   └── .nojekyll
├── preprint/                   # French article (source .md + figures); published on Zenodo (DOI above)
├── images/                     # static figures
├── requirements.txt
└── LICENSE

Data schema

data/sports_facilities.csv contains one row per OSM feature carrying a sport=* tag:

column	description
geometry	WKT (POINT, LINESTRING, POLYGON, …) in WGS84
sport	raw OSM sport value (may be multi-valued: a;b)

Multi-valued tags are split on ;; values are lowercased, trimmed, and a small alias map (SPORT_ALIASES in scripts/common.py) fixes frequent typos before counting.

Reproduce

python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt              # pandas, folium, shapely
# requires the `osmium` CLI (osmium-tool) on PATH

# 1. Rebuild the dataset from OpenStreetMap (downloads ~8 GB, then discards it)
python scripts/refresh_from_geofabrik.py

# 2. Counts + JSON for the site
python scripts/build_counts.py

# 3. Maps (written to docs/maps/)
python scripts/make_map.py heatmap
python scripts/make_map.py sport --sport netball --color "#1d3557" --name netball
python scripts/make_map.py british

Refresh for another year

refresh_from_geofabrik.py always pulls africa-latest. To pin a specific date or to build a time series, pass --url with a dated Geofabrik extract, bump SNAPSHOT_DATE in scripts/common.py, re-run steps 2–3, and commit the regenerated data/ and docs/. Comparing snapshots is the whole point: it turns a single measurement into a trend.

Deployment

GitHub Pages, source = main branch, /docs folder. The site is fully static (vanilla HTML/CSS/JS + Chart.js via CDN + embedded Folium maps); no build step.

License

• Code: MIT, see LICENSE.
• Data: derived from OpenStreetMap, © OpenStreetMap contributors, under the Open Database License (ODbL).