[Tagging] Using multipolygons to map bays in Alaska

Kevin Kenny kevin.b.kenny at gmail.com
Thu Nov 15 15:03:49 UTC 2018

```On Thu, Nov 15, 2018 at 6:11 AM Christoph Hormann <osm at imagico.de> wrote:

> Mapping bays with polygons is always non-verifiable to a large extent.
> Mapping bays with polygons as you describe it above is always
> completely non-verifiable and amounts to pure (low quality) label
> painting which should not be done and should not be incentivized by
> maps with a mapper feedback goal.

I'm afraid that I'm not following this argument very well. What about a bay
is 'completely non-verifiable?'

The existence of the bay itself?  It's there. I can see it. The locals have
a name for it.

The fact that the bay comprises an area? Even common speech refers to
something being "in" or "out of" a bay - implying that the thing is an
area. "My ancestors earned their living gathering clams and oysters in
Jamaica Bay." "He sailed out from San Francisco Bay in search of
adventure." "Mariners entering the Bay of Fundy from the ocean must beware
of tidal rips." That sort of speech does not sound to me like the language
that describes a point. The thing is an area.

The fact that the boundary of the bay is in places indefinite?  Let's look
at the three sorts of ways surrounding the bay.

1. Coastline that the consensus of locals agree is the shoreline of the bay.
2. Coastline near the mouth of the bay, where there may be a certain degree
of 'fuzziness' about whether a particular square centimetre of water is or
is not part of the bay.
3. The entirely indefinite line across the water that encompasses the bay's
mouth and serves to close the polygon that defines the bay's shape.

I saee no problem at all with borrowing the ways that define the coastline
for (1.).  it's been drawn once, why draw it again?  If it's been drawn
imperfectly, that's an entirely different problem - unless you argue that
nothing should ever be mapped until its survey has reached whatever your
standard of perfection is. I really don't want to get into the situation
where we say to mappers, "you really can't map A until and unless you've
fixed B." That's a sure way to drive away a mapper who has valuable

(2.) and (3.) are exactly the same problem that we see at the mouth of a
river, and here, as with rivers, the common understanding of the locals has
to come into play. I can recall an earlier thread in which I tried to apply
the 'objective' standard that someone proposed for river mouths to the
Hudson River in New York - and came up with the (to me) ludicrous
conclusion that the coastline, and the Atlantic Ocean, come up over 200  km
from New York City, and that the river has no independent existence until
that point and should not be named as a river on the map. Yes, it is
estuarine. It flows both ways, even in the long stretch where the water is
fresh. If your argument relates to the indefiniteness of the boundary, you
would be arguing that because a perfectly objective location for the river
mouth cannot be located, then the surface of the river cannot be an area.

I know from experience that some mappers advance that argument. I
personally find it ridiculous. Water features are subject to the same sort
of uncertainty as land features - the boundaries among seas, for instance,
are not well defined. But surely nobody would argue that a sailor embarking
in Venice or Trieste is sailing the Ionian Sea just because the boundaries
between it, the Ægean Sea, the Tyrrhenian Sea and the Adriatic Sea are all
indefinite. The Gulf of Bothnia and the Gulf of Finland both exist. Oulu is
one one and Tallinn is on the other, and they don't lose their existence as
areas just because there is no boundary on the water's surface.

If you argue, "well, rivers are different, you can model them with linear
features", I challenge you in the Hudson River example to come up with an
appropriate mapping for the Tappan Zee, a body of tidal water of
intermediate salinity, measuring about 5 km by 30, through which the Hudson
runs.

The same thing happens with land borders. Saint Lawrence and Franklin
Counties in New York are less than 100% defined because some kilometres of
their border are indefinite. The indefinite part is in an uninhabited tract
of swamp land that belongs to the state. Absolutely nobody cares where the
exact line is. All agree that surveying it could be delayed - in this case,
it has been delayed for a couple of centuries - until and unless there's
some dispute that gives a reason to fix it. But all the inhabitants of
those counties know what county they're in - because the border is
well-defined in the inhabited places where anyone cares about it. It used
to be that the boundaries between Saudi Arabia, Yemen, Oman and the
Emirates were similarly indefinite, simply because nobody cared about where
exactly an arbitrary line was in the middle of the great desert of ar-Rub'
al-Khali.  It's the same situation with the mouth of bays and rivers -
strike an arbitrary line, and improve it later if there is ever a reason.

You go on to argue that information about the shape of a bay would have no
value beyond what a point feature would give.  I agree with you only if we
confine our attention to the current renderers. There are more
sophisticated rendering algorithms out there (for example, the one that
Barrault describes in
that place labels on lines or curves that are fitted to the shape of area
features. I anticipate a counterargument that a mapper could draw a line
for the label placement - but that, much more than attempting to outline an
indefinite area, is 'label painting' of the worst kind. If you read the
Barrault paper, you'll see that it includes techniques for choosing the
line in such a way as to avoid interfering labels, something that a mapper
cannot do. In addition, the mapper cannot possibly anticipate all the
possible projections and data layouts in which the map might appear, and
which might inform the label placement. For instance, consider a printed
map on a rectangular sheet of paper. (Yes, I still use such things. I never
find that one of them has a dead battery.) Let's say that the area of the
paper happens to place a portion of the bay in one corner of the map. A
point label cannot determine the label placement - it's likely off the
sheet entirely. Only a treatment of the bay as an area would allow a
renderer to get that situation right. (I do that on paper maps all the
time. ST_PointInPolygon(ST_Intersect(theFeature, theBoundingBox)) in
PostGIS gives a reasonable first approximation for the placement of such a
label.)

So I agree: the outline of a bay is and will always be inaccurate. It
suffers from Mandelbrot's 'coastline problem' on the coastline and has an
entirely arbitrary line across the mouth. There are plenty of other map
features - up to and including national borders - that are equally
indefinite.

That still doesn't keep them from being map features. As with anything
else, the indefiniteness can be improved if there is ever a reason to do
so. It matters very much that the area exists and that its definite borders
are identified. It matters very little that one portion of the border is
not definite. "The barge is lying at anchor in Jamaica Bay" is almost
always a verifiable concept, because nobody anchors in Rockaway Inlet, the
mouth of the bay. Said barge is veriifiably *not* anchored in nearby
Bannister Bay, Brosewere Bay or Hewlett Bay. The fact that I can't tell
exactly where Bannister Bay ends and Reynolds Channel begins does not mean
that I don't have the information to answer the question, "Where is Seville
Central's transfer barge #3?"

The whole argument strikes me as expecting perfection from the map of an
imperfect world. And the dispute is between those mappers who say, "this
thing is good enough for now, I have more important projects than fixing
it," against those who say, "this thing isn't 100% well-defined and
correct, we can't possibly consider anything else until it's either
perfected or expunged - with expunction being the only legitimate fate of
objects that are imperfect in the field as well as on the map. The first
approach leaves messes that have to be cleaned up. The second makes no
progress at all.

Incidentally, this discussion also gives one possible approach for mapping
unincorporated communities. If you ask Hank what community his farm is in,
he'll give you an answer. Generally speaking, his neighbours will agree. If
you ask Mary on the next farm over what community she lives in, she too
will give you an answer, and again, generally, the neighbours will agree.
If the two answers are different, there's a boundary between the two farms.
At least near where I live, the borders of most named unincorporated
communities are understood by the locals at that level of detail, which is
why I've not messed with admin_level=8 locally except for a case or two
where I can verify that TIGER got it wrong. (How do I verify that? One
example: The border of the community follows a riverbank, the fence of a
military reservation, and the boundary line of a large parcel of parkland.
The locals understand that those are the bounds. They are observable in the
field.) Right now in my part of the world, the admin_level=8 data from
TIGER is "at least plausible", not worth sorting out right now unless I'm
in there fixing something else. There's no shortage of things to be fixing!

We have an imperfect map, and an imperfect world, and we work on tidying
both. We always leave it imperfect, but perhaps can at least leave it
better than we found it. What else can we do?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/tagging/attachments/20181115/6d87b7ba/attachment.html>
```