[Tagging] Using multipolygons to map bays in Alaska

[Daniel Koć]

W dniu 16.11.2018 o 14:31, Christoph Hormann pisze:

> Quoting from: 


Thanks, that was really helpful answer.

The problem is that I have asked you how to draw verifiable node, not
"what is a good practice for drawing a node using verifiable
operations". Let's look closer:


> in the middle between the coasts of the strait at the 
> narrowest point"

The hidden prerequisite here is to know which part of the coast belongs
to the strait and which is not. But when you know it, it's also easy to
define equally (also not fully) verifiable area:

"take the coast that belongs to the strait and link the ends with lines
without crossing them"


> approximately at 
> the middle of the bay with equal distance to the coast enclosing the 
> bay on all sides."


When you claim that you don't know which part of the coastline belongs
to the bay, how do you measure a middle of that?


> "If you want to formulate a formal mathematical rule for where the node 
> for a bay is best placed: Place it so the variance of the distance of 
> the node to the bay's shores is minimized.  Most existing nodes comply 
> with this rule remarkably well."


Again - distance "to the bay's shores", so first you know which parts of
the shore belong to the bay (you don't take the whole continent for
sure) and then do some strict operation with it (but chosen by some
convention).

So we claim that we don't know the bay borders, but we derive a node
from them. When somebody will claim that the coastline around the bay is
different (longer, shorter, has different starting and ending nodes
etc.), the node would be placed somewhere else. And while you both can
agree on the algorithm, the input is different, so most probably the
output will be different too. There can be only verification of applying
the same procedure, but not the object (or a node) itself.

And it can't be. The only fully verifiable water area is the one fully
surrounded by the land (excluding problems with intermittent waters).
Then drawing the coastline is fully accurate, because you might choose
different ways to draw a node for it and only one way to draw the area.
Any water area that is even partly open (to the sea, river, lake etc.)
is not fully (100%) verifiable. But the node is also not verifiable,
because you need to start with some assumptions about where it starts or
ends for a given object to make some formal operations on it.

My preference is to show the borders explicitly, so the user will choose
what to do with it (measure the area, show the borders, put a node this
or the other way...) and has a mostly realistic picture of the object.
Making node is hiding the borders, size and shape as an implicit choice
on behalf of all the users.

I don't say that nodes should be never used for bays or straits. I just
claim that they are less usable than areas, because they loose more
data, while being also never fully verifiable. Both are just conventions
to describe some place on the water, because there are no clear borders
on the water.

That is also why the coastline is never fully verifiable, but I would be
against solving it by using a node instead of a conventional straight
line - visible for example on La Plata estuary up to z7:

https://www.openstreetmap.org/query?lat=-36.29153&lon=-56.77843#map=7/-35.782/-55.684


-- 
"Excuse me, I have some growing up to do" [P. Gabriel]