Basic Usage

The following chapter gives a practical introduction on how to use Pyosmium. It is assumed that you already have a basic knowledge about the OSM data model.

For a more detailed introduction into the design of the osmium library, the reader is referred to the osmium documentation.

Reading OSM Data

Using Handler Classes

OSM file parsing by osmium is built around the concept of handlers. A handler is a class with a set of callback functions. Each function processes exactly one type of object as it is read from the file.

Let’s start with a very simple handler that counts the nodes in the input file:

import osmium

class CounterHandler(osmium.SimpleHandler):
    def __init__(self):
        self.num_nodes = 0

    def node(self, n):
        self.num_nodes += 1

A handler first of all needs to inherit from one of the handler classes. At the moment this is always osmium.SimpleHandler. Then it needs to implement functions for each object type it wants to process. In our case it is exactly one function node(). All other potential callbacks can be safely ignored.

Now the handler needs to be applied to an OSM file. The easiest way to accomplish that is to call the apply_file() convenience function, which in its simplest form only requires the file name as a parameter. The main routine of the node counting application therefore looks like this:

if __name__ == '__main__':

    h = CounterHandler()


    print("Number of nodes: %d" % h.num_nodes)

That already finishes our node counting program.

Inspecting the OSM objects

Counting nodes is actually boring because it completely ignores the content of the nodes. So let’s change the handler to only count hotels (normally tagged with tourism=hotel). They may be tagged as nodes, ways or relations, so handler functions for all three types need to be implemented:

import osmium

class HotelCounterHandler(osmium.SimpleHandler):
    def __init__(self):
        self.num_nodes = 0

    def count_hotel(self, tags):
        if tags['tourism'] == 'hotel':
            self.num_nodes += 1

    def node(self, n):

    def way(self, w):

    def relation(self, r):

A reference to the object is always given as the only parameter to the handler functions. The objects have some common methods and attributes, listed in osmium.osm.OSMObject, and some that are specific to each type. As all objects have tags, it is possible to reuse the same implementation for all types. The main function remains the same.

It is important to remember that the object references that are handed to the handler are only temporary. They will become invalid as soon as the function returns. Handler functions must copy any data that should be kept for later use into their own data structures. This also includes attributes like tag lists.

Handling Geometries

Because of the way that OSM data is structured, osmium needs to internally cache node geometries, when the handler wants to process the geometries of ways and areas. The apply_file() method cannot deduce by itself if this cache is needed. Therefore locations need to be explicitly enabled by setting the locations parameter to True:

h.apply_file("test.osm.pbf", locations=True, idx='sparse_mem_array')

The third parameter idx is optional and states what kind of cache osmium is supposed to use. The default sparse_mem_array is a good choice for small to medium size extracts of OSM data. If you plan to process the whole planet file, dense_mmap_array is better suited. If you want the cache to be persistent across invocations, you can use dense_file_array giving an additional file location for the cache like that:

h.apply_file("test.osm.pbf", locations=True, idx='sparse_file_array,example.nodecache')

where example.nodecache is the name of the cache file.

Interfacing with Shapely

Pyosmium is a library for processing OSM files and therefore offers almost no functionality for processing geometries further. There are other libraries for that purpose. To interface with these libraries you can simply convert the osmium geometries into WKB or WKT format and import the result. The following example uses the libgeos wrapper Shapely to compute the total way length:

import osmium
import shapely.wkb as wkblib

# A global factory that creates WKB from a osmium geometry
wkbfab = osmium.geom.WKBFactory()

class WayLenHandler(osmium.SimpleHandler):
    def __init__(self):
        osmium.SimpleHandler.__init__(self) = 0

    def way(self, w):
        wkb = wkbfab.create_linestring(w)
        line = wkblib.loads(wkb, hex=True)
        # Length is computed in WGS84 projection, which is practically meaningless.
        # Lets pretend we didn't notice, it is an example after all. += line.length

if __name__ == '__main__':
    h = WayLenHandler()
    h.apply_file("test.osm.pbf", locations=True)
    print("Total length: %f" %

Writing OSM Data

osmium.SimpleWriter is the main class that takes care of writing out OSM data to a file. The file name must be given when the writer is constructed. Its suffix determines the format of the data. For example:

writer = osmium.SimpleWriter('nodes.osm.bz2')

opens a new writer for a packed OSM XML file. Objects can be written by using one of the writers add_* functions.

A simple handler, that only writes out all the nodes from the input file into our new nodes.osm.bz2 file would look like this:

import osmium

class NodeWriter(osmium.SimpleHandler):
    def __init__(self, writer):
        self.writer = writer

    def node(self, n):

This example shows that an unmodified object can be written out directly to the writer. Normally, however, you want to modify some data. The native osmium OSM types are immutable and cannot be changed directly. Therefore you have to create a copy that can be changed. The node, way and relation objects offer a convenient replace() function to achieve exactly that. The function makes a copy and at the same time replaces all attributes where new values are given as parameters to the function.

Let’s say you want to remove all the user names from your nodes before saving them to the new file (maybe to save some space), then the node() handler callback above needs to be changed like this:

class NodeWriter(osmium.SimpleHandler):

    def node(self, n):

replace() creates a new instance of an osmium.osm.mutable object. These classes are python implementations of the native object types in osmium.osm. They have exactly the same attributes but they are mutable.

A writer is able to process the mutable datatypes just like the native osmium types. In fact, a writer is able to process any python object. It just expects suitably named attributes and will simply assume sensible default values for attributes that are missing.


It is important to understand that replace() only makes a shallow copy of the object. Tag, node and member lists are still native osmium objects. Normally this is what you want because the writer is much faster writing these native objects than pythonized copies. However, it means that you cannot use replace() to create a copy of the object that can be kept after the handler callback has finished.