[Tagging] Overlapping lakes

[Stefan Tauner]

On Fri, 12 Feb 2021 13:50:14 +0000
Richard Fairhurst <richard at systemed.net> wrote:

> In Finland, Puruvesi (a large lake) is part of Saimaa (a really large lake).
> 
> Wikipedia says "Puruvesi is… part of the Saimaa lake system": https://en.wikipedia.org/wiki/Puruvesi
> 
> Currently we have a Puruvesi multipolygon with 194 member ways:
> https://www.openstreetmap.org/relation/899111
> 
> but we also have a Saimaa multipolygon with 7223 member ways:
> https://www.openstreetmap.org/relation/7379046
> (warning - opening this will boggle your browser)
> 
> All the member ways of Puruvesi are repeated in Saimaa.
> 
> Duplicating the geometries like this seems wrong to me. In "One feature, one OSM element" (https://wiki.openstreetmap.org/wiki/One_feature,_one_OSM_element) we say "A feature that consists of several smaller features is usually best described using a relation".
> 
> How should this be mapped?

I am not that familiar with the Finnish lakes but my impression from
maps as well as my visit in the Helsinki/Tampere area is that they are
very interconnected and sometimes borders are indistinguishable. This
seems to be the case here as well. Saimaa itself is defined as *a* lake
too in Wikipedia. If that's what the Finns think as well then I am not
so sure if there is a need to change something from that abstract
standpoint. I don't think this a comparable to the Great Lakes which do
not overlap in this sense. Nobody thinks of them as *a* lake.

However, having a relation with that many members easily breaks and
that has a bigger impact than if a smaller relation breaks. They are
also harder to work with (opening Saimaa in the browser does take a
while but even editors often struggle with this kind of objects).

> The Great Lakes seems to provide a good precedent:
> https://www.openstreetmap.org/relation/1124369
> 
> The Great Lakes is currently represented as a 'master' relation of natural=water, type=site, water=lake, which contains all the individual lake multipolygons. The tagging appears to be a recent change from type=group (see https://help.openstreetmap.org/questions/72769/is-there-a-relationsite-variant-to-use-for-natural-objects).

Master relations are not really special apart from the fact that they
indicate some kind of hierarchy, which is sometimes very useful (IMHO
they are essential in large route relations but there are other
opinions), but technically all relations can contain other relations.
While this is also true for MPs in a technical sense, the semantics are
not defined, the wiki does not document this as valid, and there are
relatively few relations with relations as members (2331 out of >25M
~0.009%).

Maybe somebody with more insights into that latter option will give a
more educated opinion on that MP issue. I think it's better to be
avoided. However, that leaves us with no viable alternative to the
current one IMHO because Saimaa is neither a site nor a group of other
(types of) objects.

tl;dr: *shrug*

-- 
Kind regards/Mit freundlichen Grüßen, Stefan Tauner