Darrkzero/nigeria-police-coverage

# Nigeria Police Station Coverage A pipeline that builds a geocoded, deduplicated dataset of Nigerian police stations from public sources, joins it against LGA (Local Government Area) boundaries, and reports station coverage per LGA — for all 36 states + FCT. Started as a one-state experiment (Rivers State) to figure out what data was even available. It wasn't much: OpenStreetMap alone turns up a handful of stations per state, GRID3's police-station layer only covers ~21 of the country's 37 states, and neither is close to a complete picture. The pipeline that emerged combines both with manually-sourced Google Maps exports, resolves duplicates and cross-state contamination automatically, and scales the same process to every state. ## Sources | Source | What it gives | Coverage | |---|---|---| | OpenStreetMap (via Overpass API) | `amenity=police` points | Sparse everywhere, ~1–30 per state | | GRID3 NGA Police Stations feature service | Purpose-built police POI layer | ~21 of 37 states only | | Google Maps (manual export per state) | Precise, named, building-level pins | The main source of volume — user-collected | ## Pipeline ``` pipeline/ state_config.py path resolution + canonical state name list fetch_osm.py Overpass query for a state fetch_grid3.py GRID3 feature service query for a state parse_google_export.py cleans a Google Maps export: extracts precise coordinates from place URLs, filters out landmark-reference rows ("opposite the police station"), rejects points outside the state's actual boundary (catches cross-state stragglers that show up in Maps exports near borders) scrape_nigeriagalleria.py best-effort scrape of a secondary directory (heuristic — always review output) geocode_manual_stations.py Nominatim geocoding for manually-sourced lists merge_sources.py merges all available sources, dedupes by proximity and matched name, drops low-confidence geocodes compute_coverage.py spatial …