[Talk-ca] Multipolygon problems

Sat Jul 1 07:52:48 UTC 2017

On Fri, Jun 30, 2017 at 11:47:36PM +0200, Frank Steggink wrote:
> On 30-06-2017 21:21, Jochen Topf wrote:
> > On Fri, Jun 30, 2017 at 08:16:40PM +0200, Frank Steggink wrote:
> > > Maybe I'm not understanding it, but in the OSM inspector [1] I just see one
> > > case of old style multipolygon, in Manitoba. Last week, when you posted your
> > > original message, I just saw one case in New Brunswick. IIRC, it was a park,
> > > not even from the Canvec import.
> > The types of problems I am talking about don't show up in the OSM
> > inspector. This is not old-style multipolygons (where tags are on the
> > outer ways and not on the relation), but multipolygons where the tags
> > are on the relation AND on the ways.
> Ah, ok, now I understand. Since there was a lot of discussion about old
> style multipolygon tagging, and since this type of problem hasn't been added
> to OSM inspector,  this wasn't immediately obvious.
> > > In the OSM inspector other errors can be seen, but the most prevalent one is
> > > "Touching rings". Maybe indeed a case of suboptimal mapping, but nothing
> > > which seems urgent to me.
> > > 
> > > Here is an example of a forest multipolygon, imported by me
> > > (canvec_fsteggink). It is still version 1, but it has tags on the relation,
> > > not on the rings (except for the quarries): [2]
> > > This is from Canvec v7.0. IIRC, we started at v6.0, and the last version I
> > > know of is v10.0. Maybe v6.0 had wrong tagging, but I'm not seeing any such
> > > cases in the OSM inspector.
> > > 
> > > So, I'd like to ask you to give a couple of examples where data imported
> > > from Canvec is clearly wrong with regard to old style multipolygon tagging.
> > Here are all cases in Canada (not only those from the imports):
> > https://tmp.jochentopf.com/954226a3acab882d28d8500ddef8203d/same-tags-ca.pbf
> > 
> > Here is one example where you can clearly see the problem:
> > http://www.openstreetmap.org/relation/541821
> How difficult would it be to add this to OSM inspector? Not everybody has
> Postgres running, and is able to use osm2pgsql. Yes, there is documentation,
> but it requires some technical skills. Also, it would be very convenient to
> have this updated daily.

It is not that difficult to add to the OSM Inspector and if I have the
time I'll work on that together with the Geofabrik people.

> > > When we have clear examples, then it might be easier to come up with a plan
> > > how to fix it. But so far, I see absolutely no reason why Canada stands out
> > > in a negative way. Yes, we all acknowledge that Canvec data is suboptimal,
> > > but as others already have pointed out, mapping everything by hand in
> > > especially remote areas is nearly impossible.
> > Canada stands out in a negative way, because
> > a) there are so many problems. Nearly a third of the cases worldwide are in
> >     Canada and
> > b) most of these problems are probably caused by one little program, the
> >     program used to convert/import the CanVec data.
> As you might have noticed, later imports, like the example I provided, don't
> have that issue anymore. I'm mentioning this to express that not _all_
> Canvec data is at fault! Only the first couple of versions. However, for
> some reason this was never noticed up until a point that collaborative
> action was done to have it fixed. Probably because the rendering pipeline of
> the slippy map was accepting this kind of tagging up until recently.

Okay, that is a big relief already. At least we are not making this
problem worse by new imports that might happen in the future.

> > Mapping Canada "by hand" might be difficult because it is such a huge
> > country and there aren't that many mappers. But the same arguments goes
> > for why you have to be extra careful importing data. If you break
> > something, there are not enough people to fix it manually. And, yes,
> > errors do happen. And if we find them, we fix them and move on. But
> > errors from imports can be so huge there aren't enough people there to
> > fix them manually.
> The world is so huge that there aren't enough people to create and maintain
> a global world map. However, OSM exists. Fixing errors can also be
> crowdsourced. Martijn van Exel is really doing a great job with MapRoulette,
> for instance. Although fixing errors (cleaning up the mess left behind by
> others) is not nearly as rewarding as mapping, it might be easier to do,
> especially since there is no need for a lot of creativity when fixing the
> same kind of errors.

You might have seen that I spent a lot of time in the last months to
create more than 60 Maproulette challenges for all sorts of different
multipolygon problems in different communities. And the community worked
tirelessly on all these problems. (http://area.jochentopf.com/fixed.html)

If the Canadian community steps up and is willing to do this work
manually, I'd he happy to provide such Maproulette challenges. I have
challenges running at this moment for this exact problem for other parts
of the world (http://area.jochentopf.com/fixing.html). But I wanted to
give the Canadian community the chance for some input first, because of
the unique situation here.

> Personally, I think that, although things were far from perfect, they were
> done with the best intentions and with the support of the majority of the
> Canadian OSM community. We have to deal with this situation now. A much more
> cooperative tone would have been very welcome, especially since you would
> like to see us coming off our lazy butts and fix our mess.
> 
> There is really nothing to gain by threatening to contact the DWG in order
> to have those imports removed. They already exist for about 7 years! And if
> the Canadian community at large wouldn't have welcomed it, this would have
> come to the surface way sooner.  So, why has this suddenly become such a
> huge problem because of the way how the slippy map is rendered?

Well, I tried writing a nice mail informing you all about the problem. A
week later, when nobody had even acknowledged the problem, I wrote the
next mail. And that did exactly what it was supposed to do: It got you
all off your butts and discuss this problem. Now to the next step:

> We can much better focus on getting the job done, than criticizing each
> other. If 150 people are fixing 100 multipolygons each, this is doable! We
> could do it with the help of OSM inspector, and eventually a MapRoulette
> task.

Thank you. That is the first time somebody from the Canadian community
is actually addressing the issue at hand and proposing a way forward.
Lets see whether other people in the community see it the same way and
you can come to a solution together.

I can help by providing Maproulette challenges and OSMI views and
downloads with lists of affected relations etc. But I can't decide how
you want to address this problem. The Canadian community has to do this.

> p.s. Are you still wearing your t-shirt with Lake Manicouagan on it, based
> on OSM data? I hope it doesn't contain wrong tagging or imported data. ;)

Unfortunately the lake has faded a lot on that t-shirt so I don't wear
it much any more. I should make a new one! :-)

Jochen
-- 
Jochen Topf  jochen at remote.org  https://www.jochentopf.com/  +49-351-31778688