Flatmap

Flatmap is a tool that generates Mapbox Vector Tiles from geographic data sources like OpenStreetMap. Flatmap aims to be fast and memory-efficient so that you can build a map of the world in a few hours on a single machine without any external tools or database.

Vector tiles contain raw point, line, and polygon geometries that clients like MapLibre can use to render custom maps in the browser, native apps, or on a server. Flatmap packages tiles into an MBTiles (sqlite) file that can be served using tools like TileServer GL or even queried directly from the browser. See awesome-vector-tiles for more projects that work with data in this format.

Flatmap is named after the "flatmap" operation that it performs: map input elements to rendered tile features, flatten them into a big list, and sort by tile ID to group into tiles. The output is also a "flat map" where zoom level 0 contains the entire planet in a 256x256 px tile, and each zoom level splits parent tiles into 4 quadrants, revealing more detail.

See ARCHITECTURE.md for more details on how Flatmap works or this blog post for more of the backstory.

Demo

See the live demo of vector tiles created by Flatmap.

Style © OpenMapTiles · Data © OpenStreetMap contributors

Usage

To generate a map of an area using the basemap profile, you will need:

• Java 16+ or Docker
• at least 1GB of free disk space plus 5-10x the size of the .osm.pbf file
• at least 1.5x as much free RAM as the input .osm.pbf file size

To build the map:

Using Java, download flatmap.jar from the latest release and run it:

wget https://github.com/onthegomap/flatmap/releases/latest/download/flatmap.jar
java -Xmx1g -jar flatmap.jar --download --area=monaco

Or using Docker:

docker run -e JAVA_TOOL_OPTIONS="-Xmx1g" -v "$(pwd)/data":/data ghcr.io/onthegomap/flatmap:latest --download --area=monaco

⚠️ This starts off by downloading about 1GB of data sources required by the basemap profile including ~750MB for ocean polygons and ~240MB for Natural Earth Data.

To download smaller extracts just for Monaco:

Java:

java -Xmx1g -jar flatmap.jar --download --area=monaco \
  --water-polygons-url=https://github.com/onthegomap/flatmap/raw/main/flatmap-core/src/test/resources/water-polygons-split-3857.zip \
  --natural-earth-url=https://github.com/onthegomap/flatmap/raw/main/flatmap-core/src/test/resources/natural_earth_vector.sqlite.zip

Docker:

docker run -e JAVA_TOOL_OPTIONS="-Xmx1g" -v "$(pwd)/data":/data ghcr.io/onthegomap/flatmap:latest --download --area=monaco \
  --water-polygons-url=https://github.com/onthegomap/flatmap/raw/main/flatmap-core/src/test/resources/water-polygons-split-3857.zip \
  --natural-earth-url=https://github.com/onthegomap/flatmap/raw/main/flatmap-core/src/test/resources/natural_earth_vector.sqlite.zip

You will need the full data sources to run anywhere besides Monaco.

To view tiles locally:

Using Node.js:

npm install -g tileserver-gl-light
tileserver-gl-light --mbtiles data/output.mbtiles

Or using Docker:

docker run --rm -it -v "$(pwd)/data":/data -p 8080:8080 maptiler/tileserver-gl -p 8080

Then open http://localhost:8080 to view tiles.

Some common arguments:

• --download downloads input sources automatically and --only-download exits after downloading
• --area=monaco downloads a .osm.pbf extract from Geofabrik
• --osm-path=path/to/file.osm.pbf points Flatmap at an existing OSM extract on disk
• -Xmx1g controls how much RAM to give the JVM (recommended: 0.5x the input .osm.pbf file size to leave room for memory-mapped files)
• --force overwrites the output file
• --help shows all of the options and exits

Generating a Map of the World

See PLANET.md.

Creating a Custom Map

See the flatmap-examples project.

Benchmarks

Some example runtimes (excluding downloading resources):

Input	Profile	Machine	Time	mbtiles size	Logs
s3://osm-pds/2021/planet-211011.osm.pbf (65GB)	Basemap	DO 16cpu 128GB	3h9m cpu:42h1m avg:13.3	99GB	logs, VisualVM Profile
Daylight Distribution v1.6 with ML buildings and admin boundaries (67GB)	Basemap	DO 16cpu 128GB	3h13m cpu:43h40m avg:13.5	101GB	logs
s3://osm-pds/2021/planet-211011.osm.pbf (65GB)	Basemap (without z13 building merge)	Linode 50cpu 128GB	1h9m cpu:24h36m avg:21.2	97GB	logs, VisualVM Profile
s3://osm-pds/2021/planet-211011.osm.pbf (65GB)	Basemap (without z13 building merge)	c5ad.16xlarge (64cpu/128GB)	59m cpu:27h6m avg:27.4	97GB	logs

Alternatives

Some other tools that generate vector tiles from OpenStreetMap data:

• OpenMapTiles is the reference implementation of the OpenMapTiles schema that the basemap profile is based on. It uses an intermediate postgres database and operates in two modes:
  1. Import data into database (~1 day) then serve vector tiles directly from the database. Tile serving is slower and requires bigger machines, but lets you easily incorporate realtime updates
  2. Import data into database (~1 day) then prerender every tile for the planet into an mbtiles file which takes over 100 days or a cluster of machines, but then tiles can be served faster on smaller machines
• Tilemaker uses a similar approach to Flatmap (no intermediate database), is more mature, and has a convenient lua API for building custom profiles without recompiling the tool, but takes about a day to generate a map of the world

Some companies that generate and host tiles for you:

• Mapbox - data from the pioneer of vector tile technologies
• Maptiler - data from the creator of OpenMapTiles schema
• Stadia Maps - what onthegomap.com uses in production

If you want to host tiles yourself but have someone else generate them for you, those companies also offer plans to download regularly-updated tilesets.

Features

• Supports Natural Earth, OpenStreetMap .osm.pbf, and Esri Shapefiles data sources
• Java-based Profile API to customize how source elements map to vector tile features, and post-process generated tiles using JTS geometry utilities
• Merge nearby lines or polygons with the same tags before emitting vector tiles
• Automatically fixes self-intersecting polygons
• Built-in basemap profile based on OpenMapTiles v3.12.2
• Optionally download additional name translations for elements from Wikidata
• Export real-time stats to a prometheus push gateway using --pushgateway=http://user:password@ip argument (and a grafana dashboard for viewing)
• Automatically downloads region extracts from Geofabrik using geofabrik:australia shortcut as a source URL
• Unit-test profiles to verify mapping logic, or integration-test to verify the actual contents of a generated mbtiles file (example)

Limitations

• It is harder to join and group data than when using database. Flatmap automatically groups features into tiles, so you can easily post-process nearby features in the same tile before emitting, but if you want to group or join across features in different tiles, then you must explicitly store data when processing a feature to use with later features or store features and defer processing until an input source is finished (boundary layer example)
• Flatmap only does full imports from .osm.pbf snapshots, there is no way to incorporate real-time updates.

Roadmap

• Enough flatmap-core functionality to support basemap profile based on OpenMapTiles
• Basemap profile based on OpenMapTiles v3.12.2
  • Port all layers
  • Download name translations from wikidata
  • Merge buildings at z13
  • adm0_l/adm0_r boundary labels
  • Abbreviate road names to improve visibility
  • Poi layer agg_stop tag
• Get flatmap-core into Maven Central
• Remove geotools dependency for reading shapefiles (not in Maven Central)
• Remove graphhopper dependency for reading OSM files, and use LocationsOnWays to skip node location storage when present and reduce memory requirement by 70%.
• "Sparse mode" to only store node and relation data for elements used by a profile
• Support zoom levels higher than 14
• Handle nodes and relations in relations (only ways handled now)
• Lake centerline support in flatmap-core
• Improve line merging to combine nearby parallel roads
• Basemap schema improvements for onthegomap.com
• Accept other kinds of data sources
• Extract reusable utilities for complex schemas from flatmap-basemap to flatmap-core
• Other schemas

Contributing

Pull requests are welcome! See CONTRIBUTING.md for details.

Support

Have a question or want to share something you've built? Start a GitHub discussion.

Found a bug or have a feature request? Open a GitHub issue to report.

This is a side project, so support is limited. If you have the time and ability, feel free to open a pull request to fix issues or implement new features.

Acknowledgement

Flatmap is made possible by these awesome open source projects:

• OpenMapTiles for the schema and reference implementation that the basemap profile is based on
• Graphhopper for utilities to process OpenStreetMap data in Java
• JTS Topology Suite for working with vector geometries
• Geotools for shapefile processing
• SQLite JDBC Driver for reading Natural Earth data and writing MBTiles files
• MessagePack for compact binary encoding of intermediate map features
• geojson-vt for the basis of the stripe clipping algorithm that flatmap uses to slice geometries into tiles
• java-vector-tile for the basis of the vector tile encoder
• imposm3 for the basis of OSM multipolygon processing and tag parsing utilities

See NOTICE.md for a full list and license details.

Author

Flatmap was created by Michael Barry for future use generating custom basemaps or overlays for On The Go Map.

License and Attribution

Flatmap source code is licensed under the Apache 2.0 License, so it can be used and modified in commercial or other open source projects according to the license guidelines.

Maps built using flatmap do not require any special attribution, but the data or schema used might. Any maps generated from OpenStreetMap data must visibly credit OpenStreetMap contributors. Any map generated with the profile based on OpenMapTiles or a derivative must visibly credit OpenMapTiles as well.