In a “discussion” on this changeset (between @9_tab , @habi and me), @habi made me aware that there seem to be (at least) two ways to map a building in Switzerland:
Map a building outline from what we see on the ground
EGID has another definition of what is a building, and there are some cases where a building seems to be a single piece if we look at it from the ground with multiple entrances and addresses. However, it has multiple EGID numbers and is registered as multiple buildings.
It happens in the “discussed” changeset, but I’ve tried to find a better example with a stronger online source (note that SCHL has a shield on each building with some of the information that you can find on their website):
The owner website mentions 3 buildings and 6 entreances. RegBL has 6 addresses and 6 EGID numbers so 6 buildings. I guess that through the eyes of an expert, it’s obvious that there is a wall in the middle of the building, but not for me, especially as the hipped roof is shared between the two.
I have two issues with this way of mapping:
I’m not sure how to tag the roof:shape=hipped. Should it be a building:part=roof? Should I use a relation or something else?
There is a difference between the spoken language (we’ll talk about “one building”) and the official way. It could be an issue when we count the number of buildings for example per owner.
Isn’t there another way to reunite those two approaches? One of my ideas was to map the building with a single outline and use building:part inside with Key:ref:EGID — OpenStreetMap Wiki (overpass query for usage) (btw shouldn’t it be ref:CH:EGID instead?) and the addresses. But according to Tag:building=terrace - OpenStreetMap Wiki this kind of usage seems deprecated (note that the French version of this page hasn’t been updated).
@habi was saying that it’s a “local preferences”, but shouldn’t we agree on a way to map building for the whole country instead?
I couldn’t find any discussion or elements that happened in the past; I probably have missed something.
building separation as defined by the RegBl is not necessarily directly observable and might be hard to nail down without detailed plans of a specific building (often this is done based on roof shapes in OSM, but that can naturally be totally wrong).
building:parts are actually used for modelling the 3d hull of buildings and need not to have anything, and very often have nothing at all, to do with a building as defined both in the RegBl and OSM.
the RegBl just gives us ids, no geometry, so while that can hint at multiple OSM buildings without the geometry it isn’t really that helpful.
and then there’s the whole issue of building foot print vs. roof outline mapping this could be partially be addresses by using 3d-mapping (I’ve suggested that the last time we had the topic), but that just makes things more complicated if we are just trying to 2d map buildings*.
Simon
* if it is any consolation swisstopo has a mix of AV/Cadastre building footprints and roof outlines in it’s 3d buildings dataset. so we are not the only people grappling with these issues.
I suspect this is simply due to old imagery being not particularly well orthorectified. We had an awful lot of buildings traced essentially as soon as the 1st generation of Bing imagery became available.
This has been a longstanding issue with I subsumed under “local mapping conventions”.
As mentioned in the linked changeset this leads to changesets from different users like:
where several building outlines mapped/imported from cadastral data were merged to one outline, decreasing the mapping finesse. I absolutely see the argument that these buildings look like one single block from the outside, “but” have several EGIDs.
but are mapped as single, touching ways. And are each inhabited by a single group of people, usually a family. I actually grew up in one of those singular building outlines in a larger block of 10 of those…
I think that mapping buildings as such if data on their outline in the cadastral survey is available is perfectly suitable, and mapping them as single, encompassing outline is also suitable. But it’s not suitable to merge precisely mapped single buildings into one encompassing outline if they’ve been mapped in detail already.
and then there’s the whole issue of building foot print vs. roof outline mapping this could be partially be addresses by using 3d-mapping (I’ve suggested that the last time we had the topic), but that just makes things more complicated if we are just trying to 2d map buildings*.
Yes, now I can think of those buildings like farm where the roof is actually much wider than the building. I guess building:part=roofis appropriate in those cases.
Thanks @habi for the details. I think I kind of get what you said by local as by curiosity I’ve checked the “land registry” (Katasterplan / Cadastre) map for the example given above and it’s seems that the canton has a different opinion from swisstopo (one building with an id, which is linked to only of the two EGID numbers). It sounds like a mess like you said @SimonPoole.
