OpenBuildingMap

In my Geofabrik work, I often process OSM data for clients in one way or the other. One thing we do is sell a standardised shape file export where we select the most common OSM features and sort them into a couple of thematic layers. We used to export the buildings as well – it wasn’t a lot of extra effort and some clients had use for it. But soon we’ll have to stop doing that, and instead provide the buildings only if someone explicitly asks for them.

The reason is that the amount of building data in OSM is exploding, mainly due to building imports.

Building imports are relatively easy to do – you don’t usually have large objects where multipolygons have to be constructed, and they are not usually interconnected with other, already existing data. Buildings make a map look nice; everybody wants to have them shown. There seems to be a wide-ranging consensus that mappers should not be bothered with something as mind-numbing as drawing buildings, so an import is the only solution. (I’ll come back to that later.)

I’ll take France as an example because they have imported a lot of buildings and because they’re next door from where I live. The France dataset has 210 million nodes and 30 million ways in OSM. Of these, 144 million nodes and 27 million ways are used for buildings; that means that if you download French data from OSM, 80% of it will be buildings, and 20% will be “other stuff”, like streets, cycle routes, or POIs. This has knock-on effects on everything else of course. Simplifying only slightly, this means that of all the time the OSM severs spend on rendering the map of France, 80% is spent on drawing buildings. It means that if you have to wait one minute for your editor to download an area in France you want to edit, 50 seconds are spent assembling and transferring buildings. It means that if you carefully selected a small area for editing because otherwise your editor gives you an error message saying “the area you selected for download contains too much data”, you could work on an area six times as large if it weren’t for the buildings.

I’m not saying we should get rid of buildings. But I want to draw attention to the fact that they are, for the most part, expensive eye candy. We want our data to be the basis for beautiful maps, so eye candy must be permitted, maybe even welcomed – but it comes at a price, and it is transforming OSM. Currently we don’t differentiate thematic “layers” in OSM. The planet file, the API “map” call, the export tab, the data layer, and Geofabrik’s download service – in all these places you get “everything”, and if you want layers of any kind, you have to sort things yourself. But this will not make sense or be possible for much longer; too many use cases do not require buildings, and it is a waste of resources to force-feed 80% building data to everyone who is only interested in the other 20%.

(If you’re now saying that buildings are more than eye candy because they can carry addresses and therefore they are useful for navigation and finding addresses, then yes, you’re absolutely right in theory. Of the 27 million buildings in France, only 0.16 million actually have a house number though. So I revise my figures; not 27 million eye candy buildings, only 26.84 million.)

Of course this issue isn’t new; unless you use a specialized, filtering download service like XAPI or Overpass, you always get more than you’re interested in. But it has never been as massive as it is becoming with building imports. I’m saying imports because imports allow people to produce much more data in a shorter time than traditional mapping. In many places, buildings are meticulously traced by hand from aerial imagery, and in those areas the volume of building data rises too – but it doesn’t explode. France has an active community of about 1,000 mappers; for them to contribute 27 million buildings, even if they’re industrious, would take several years.

Where changes are gradual and happen over a longer time, the ecosystem has time to adapt, and such adaption will come naturally. We haven’t yet adapted to the new world of OpenBuildingMap. I don’t like imports; I think that in areas where we don’t have mappers who care enough to trace buildings from an aerial image, we should be honest about it and not paint our map with some government’s building data. But I fear that this battle is lost, and so we’ll have to find ways to cope with the influx of such data.