osm2rdf

osm2rdf is a tool for converting the complete OpenStreetMap (OSM) data to RDF Turtle (TTL).

We consider the following prefixes being used in our code and documentation:

PREFIX ogc: <http://www.opengis.net/rdf#>
PREFIX geo: <http://www.opengis.net/ont/geosparql#>
PREFIX osm2rdf: <https://osm2rdf.cs.uni-freiburg.de/rdf#>

The produced triples include:

1. One triple <o> <key> <value> . for each key-value pair (called "tag" in OSM) of each object.
2. Two triples <o> geo:hasGeometry <geom_o> . <geom_o> geo:asWKT <wkt> . for the shape of each object (using WKT to represent the shape).
3. One triple <o1> ogc:sfContains <o2> . for each pair of objects, where <o1> geometrically contains <o2>.
4. One triple <o1> ogc:sfIntersects <o2> . for each pair of objects, where <o1> geometrically intersects <o2>.
5. Triples for various other kinds of information (see osm2rdf --help for the many options of the tool).

For the complete OSM data, the tool takes around 15 hours on standard hardware and produces around 40 billion triples with a size of around 200 GB for the compressed TTL file.

Accompanying services and materials

Weekly updated downloads for the complete data as well as for each individual country are provided on https://osm2rdf.cs.uni-freiburg.de .

A SPARQL endpoint (based on the QLever SPARQL engine) for the complete OSM data is available under https://qlever.cs.uni-freiburg.de/osm-planet . The UI provides many example queries (click on the "Examples" button).

The basic method behind osm2rdf is described in this SIGSPATIAL'21 paper.

The algorithm behind computing the predicates geo:sfContains and geo:sfIntersects will be published soon. For the complete OSM data, there are around 2 billion non-trivial geometric objects (this count excludes points that are merely base points of more complex shapes). It is very challenging to compute geometric relations between such a large number of objects in finite time.

Getting started

Some features require boost version 1.78 or newer. To enable these features boost has to be updated manually on most systems as distributions do ship older versions.

Docker

The provided Makefile contains targets which build a Docker image, prepare input and output directories and downloads data. To run osm2rdf on germany it is enough to call:

make docker-de

This creates the directories input, output, and scratch, downloads the data and runs osm2rdf with default parameters.

Alternatively use the provided Dockerfile to compile and run osm2rdf:

$ mkdir input
$ mkdir output
$ mkdir scratch

$ docker build -t osm2rdf .

$ wget -O ./input/germany-latest.osm.pbf https://download.geofabrik.de/europe/germany-latest.osm.pbf
$ docker run --rm -v `pwd`/input/:/input/ -v `pwd`/output/:/output/ -v `pwd`/scratch/:/scratch/ -it osm2rdf /input/germany-latest.osm.pbf -o /output/germany-latest.osm.ttl -t /scratch/

Without Docker

On Ubuntu 20.04 the following packages are required to build osm2rdf:

clang clang-tidy g++ libboost-dev libboost-serialization-dev libexpat1-dev cmake libbz2-dev zlib1g-dev libomp-dev

clang is optional, but without it clang-tidy has unrelated error messages.

Clone and build osm2rdf:

$ git clone https://github.com/ad-freiburg/osm2rdf.git
$ cd osm2rdf
$ mkdir -p build && cd build && cmake -DCMAKE_BUILD_TYPE=Debug .. && make -j4 && cd ..

Prepare input, output and scratch folders and download some data and convert the data to ttl:

$ mkdir input
$ mkdir output
$ mkdir scratch
$ wget -O ./input/freiburg-regbez-latest.osm.pbf https://download.geofabrik.de/europe/germany/baden-wuerttemberg/freiburg-regbez-latest.osm.pbf
$ ./build/osm2rdf ./input/freiburg-regbez-latest.osm.pbf -o ./output/freiburg-regbez-latest.osm.ttl

More .pbf files can be found on the Geofabrik Download Server