Pyrosm – Python’s Rapid OSM Parser

Pyrosm is a Python library for reading OpenStreetMap data from Protocolbuffer Binary Format -files (*.osm.pbf) into Geopandas GeoDataFrames. Pyrosm makes it easy to extract various datasets from OpenStreetMap pbf-dumps including e.g. road networks, buildings, Points of Interest (POI), landuse and natural elements. Also fully customized queries are supported which makes it possible to parse the data from OSM with more specific filters.

Pyrosm is easy to use and it provides a somewhat similar user interface as OSMnx. The main difference between pyrosm and OSMnx is that OSMnx reads the data over internet using OverPass API, whereas pyrosm reads the data from local OSM data dumps that can be downloaded e.g. from GeoFabrik's website. The library has been developed by keeping performance in mind, hence, it is mainly written in Cython (Python with C-like performance) which makes it fast to parse OpenStreetMap data from PBF files. With its out-of-core ("streaming") engine, the default reader, it is possible to read whole-country (or even continent) extracts quickly without running out of memory. Pyrosm decodes the PBF data with Google's Protocol Buffers library (using its fast upb C backend). Protocol Buffers is a commonly used and efficient method to serialize and compress structured data which is also used by OpenStreetMap contributors to distribute the OSM data in PBF format (Protocolbuffer Binary Format).

Backend change. Since v0.8.0, the backend used to parse the protocol-buffer messages is Google's Protobuf (its fast C upb backend) instead of the previously used Pyrobuf. The switch was made for maintainability and installation reliability: Pyrobuf is no longer maintained and its source build fails with modern setuptools, which broke pip install pyrosm, whereas Google's Protobuf is actively maintained and ships prebuilt wheels and conda-forge packages for every Python version pyrosm supports. The change does not slow down parsing — see the backend benchmark. v0.7.0 was the last release that used Pyrobuf.

Out-of-core ("streaming") engine. Pyrosm reads with its out-of-core engine by default: OSM(filepath) decodes PBF files in a single streaming pass with bounded memory, decodes larger files in parallel on threads (workers="auto", the default), and caches each layer's result automatically. It reads whole-country and even whole-continent extracts (e.g. all of South America) quickly on modest machines without running out of memory. OSM(filepath, engine="in_memory") selects the in-memory reader, which parses the whole file once and serves every read of the reader from it; history reads always use it.

Documentation is available at https://pyrosm.readthedocs.io.

Current features

• download PBF data easily from any location in the world
• find and download the right extract for a bounding box or a place name (NEW in v0.9.0)
• download the smallest single extract that contains an area, comparing Geofabrik, BBBike and Movisda extracts (NEW in v0.14.0)
• read street networks (separately for driving, cycling, walking and all-combined)
• read buildings from PBF
• read Points of Interest (POI) from PBF
• read landuse from PBF
• read "natural" from PBF
• read boundaries from PBF (such as administrative borders)
• read any other data from PBF by using a custom user-defined filter, including regular-expression values and Overpass-style filters (NEW in v0.11.0)
• read large PBF extracts (country level, even some continents) with bounded memory using the out-of-core ("streaming") engine, the default reader, with parallel decoding and automatic result caching
• filter data based on bounding box
• control which OSM tags are parsed into columns
• crop a PBF to a smaller area and write modified OSM data back to PBF (NEW in v0.9.0), or export only selected layers (NEW in v0.11.0)
• merge overlapping PBF extracts, such as two country extracts around a border town, into one file (NEW in v0.14.0)
• edit the geometry of an OSM network — move nodes, reshape ways, delete elements — and write it back to PBF (NEW in v0.12.0)
• export networks as a directed graph to igraph, networkx and pandarm, optionally with topological simplification (NEW in v0.11.0)

Install

Pyrosm is distributed via PyPI and conda-forge.

Install pyrosm with pip:

$ pip install pyrosm

pip installs prebuilt wheels of pyrosm and its dependencies, so nothing is compiled. Wheels are published for CPython 3.11–3.15 on glibc-based Linux (x86_64 and aarch64), macOS (Intel and Apple silicon) and Windows on x86_64.

You can also install pyrosm from conda-forge with mamba (or its standalone variant micromamba), a fast drop-in replacement for conda. If you don't have it yet, download and install mamba via Miniforge from the conda-forge download page — it ships mamba preconfigured with the conda-forge channel. Then install pyrosm with:

$ mamba install -c conda-forge pyrosm

or, with micromamba:

$ micromamba install -c conda-forge pyrosm

(the same command works with conda if you have it).

Troubleshooting

If pip starts building pyrosm from source, no wheel matches your platform or Python version, and the build needs a C compiler. On musl-based Linux such as Alpine, and on Windows on ARM, GeoPandas' dependency pyogrio has no wheels either, and building it needs the GDAL library; in Docker, a Debian-based image such as python:3.12-slim gets wheels for everything. On the platforms conda-forge supports, which do not include musl-based Linux, installing from conda-forge avoids building from source.

When should I use Pyrosm?

Pyrosm can of course be used whenever you need to parse data from OSM into geopandas GeoDataFrames. However, pyrosm is better suited for situations where you want to fetch data for whole city or larger regions (even whole country).

If you are interested to fetch OSM data for smaller areas such as neighborhoods, or search data around a specific location/address, we recommend using OSMnx which is more flexible in terms of specifying the area of interest. That being said, it is also possible to extract neighborhood level information with pyrosm and filter data based on a bounding box (see docs).

How to use?

Using pyrosm is straightforward. See docs for instructions how to use the library.

Get in touch + contributions

If you find a bug from the tool, have question, or would like to suggest a new feature to it, you can make a new issue here.

We warmly welcome contributions to pyrosm to make it better. If you are interested in contributing to the library, please check the contribution guidelines.

Development

You can install a local development version of the tool by 1) creating an environment with the necessary packages using mamba/micromamba and 2) building pyrosm from source:

1. create an environment for one of the supported Python versions (3.11–3.15) by:

   • e.g. Python 3.14 (you might want to modify the env-name which is test by default): $ mamba env create -f ci/314-conda.yaml (or $ micromamba create -f ci/314-conda.yaml)
   • environment files for other versions are available under ci/ (e.g. ci/312-conda.yaml)

2. build pyrosm development version from master (activate the environment first):

   • pip install -e . --no-build-isolation
   • (--no-build-isolation builds the Cython extensions against the build dependencies provided by the environment, i.e. Cython, instead of refetching and recompiling it in an isolated build environment)

You can run tests with pytest by executing:

$ pytest . -v

Citing pyrosm

If you use pyrosm in your research or other work, please cite it. Pyrosm is archived on Zenodo, which provides a citable DOI (10.5281/zenodo.3755057, always resolving to the latest version). An example citation for the pyrosm version 0.13.1 is as follows:

Tenkanen, H. (2026). pyrosm: A Python library for reading and writing OpenStreetMap PBF data with GeoDataFrames. (v0.13.1) Zenodo. https://doi.org/10.5281/zenodo.3755057

A BibTeX entry and version-specific DOIs are available in the documentation and on the Zenodo record.

License and copyright

Pyrosm is licensed under MIT (see license).

The OSM data is downloaded from two sources:

Data © Geofabrik GmbH, BBBike, Movisda and OpenStreetMap Contributors

All data from the OpenStreetMap is licensed under the OpenStreetMap License.