OSM H3 Metrics Pipeline

This repository contains an OSM → PostGIS + H3 → pmtiles pipeline that:

• Downloads OpenStreetMap data for a bounded region (Utah first).
• Imports it into PostGIS using osm2pgsql flex + Lua.
• Computes metrics on H3 hexagons (initially "age of POIs", e.g. shop=*).
• Generates vector tiles as pmtiles using tippecanoe.
• Serves tiles via a simple static server or object storage + CDN.

The design is planned to be region-parameterised: once Utah works, you should be able to deploy the same stack other regions by changing a few environment variables and it would "just work". Not there yet.

The deployment target assumed here is:

• Hetzner Cloud VM (e.g. 4 vCPU, 8 GB RAM, 160+ GB SSD).
• Docker Compose as the main "infra as code" artifact.
• Dokploy on the same host to orchestrate the Compose app, manage TLS, and run scheduled jobs.

Dokploy is completely optional - you can totally orchestrate this with just docker-compose, but I'm currently in "Self Hosted PaaS" experimentation mode and testing the limitations of those systems.

architecture

1. OSM Source

   • Geofabrik extract (*-latest.osm.pbf). For state / province / country
   • Could eventually be planet PBF + replication diffs for global/large-region coverage, but that would require some rethinking of bits of this setup.

2. OSM → PostGIS

   • Import with osm2pgsql flex using a custom Lua style:
     • Right now I want to demonstrate POI age (initially just shop=* and perhaps a few amenity=*).
     • Preserve timestamps (and other extra attributes) so we can compute feature age.
   • Store in PostGIS as osm_poi_shop (or similar).
   • this would be an extensible design - drop in another lua file + another metrics declaration, and we'd have another QA layer.

3. Metrics + H3

   • Use h3 + h3_postgis extensions to: generate H3 cells at res 6,7,8 (for example) and compute relevant aggrregate metrics like:
     • POI counts
     • Average age
     • P90 age
     • Counts of "older than 1 year", "younger than 30 days", etc.

4. Tiles

   • Export H3 cells with metrics as GeoJSON/NDJSON.
   • Run tippecanoe to generate one or more pmtiles files.
   • Drop into Cloudflare R2

5. Orchestration

   • A pipeline container runs the above steps.
   • Scheduled daily:
     • Via cron on the host, or
     • Via Dokploy’s "Schedule Jobs" against the pipeline service.

The of course a simple map viewer! Can be MapLibre with minimal UX. Needs a legend and an about page.

There we go.