How complete are addresses in OpenStreetMap in Belgium? After several status these previous years (see September 2022, in May 2023, and in May 2024, November 24), I finally adapt the process to analyse addresses completeness for all Belgium!
The aim of this small article is to assess the completeness of addresses in Belgium for OSM, compared to official data. Basically we count the address points in OSM and in official data by municipalities. We don’t check the address quality here, we know there can be errors in OSM data, and even in official data.
How I proceeded
Using QGIS, I gathered OSM addresses and official ones. OSM data is imported in a PostgreSQL database using osm2pgsql
based on a pbf file for Belgium from the 24th of April. Official data is coming from BEST address csv files, that gathers Belgian addresses for the 3 regions of the country.
Addresses in OSM can be mapped on buildings (most of the cases), hence on a way, or on a node. I consider only objects with the “addr:housenumber” tag. First, I made a single layer of addresses points from OSM data by merging the centroids of buildings with this tag and the points with these tags, by excluding the places where these two layers overlaps (typically points with an address inside a building where there is already an address). The method is far from perfect: I don’t consider relations, I miss some way objects with addresses but without the building tag (typically some schools), and probably a lot of other edge cases due to the “joyeux bordel” of the OSM data model.
Official data considers several address points for each building unit, or box number. It means that there are often several points at the same place, especially in cities, accounting for each box number. Since the mapping of box number is often not so used in OSM, I merged all points at the same place (given a buffer of 1 meter) as a first step. This reduces the number of official address from 7,004,321 to 4,454,017.