[Tagging] coastline v. water

Tue Nov 24 17:32:36 UTC 2020

On Tue, Nov 24, 2020 at 9:23 AM Christoph Hormann <osm at imagico.de> wrote:

> The problem we have here is that of a widening gap between the goals and
> aspirations of the mapper community - which naturally grow as OSM grows in
> ambitions - and the abilities and engagement in the non-mapping part of the
> community to develop and satisfy similar ambitions in cartographic quality
> without outsourcing the hard part of that work to the mappers.  Too many
> people have followed the illusion for too long that the large corporate OSM
> data users will provide the necessary support in that field while it turns
> out (non-surprisingly in my eyes) that they have neither an interest in
> above average cartographic quality nor in substantially sharing methods and
> competency in the little work they do in that domain.
>

(Brief summary: 1. Many area features are indefinite only at margins that
do not have a significant deleterious effect on statistics when analyzed or
on the understanding of the map when rendered. 2. Topology still matters -
for analyzing or rendering them. 3. Algorithm development needs data to
chew on. Blocking the data while waiting for the algorithms is a 'deadly
embrace.' 4. Mappers are continuing to enter the data for approximate
regions because they understand 1.-3. above. 5. Which argument are you
willing to give up in order not to argue against all progress in this
domain?)

In my earlier message, I was speaking not as a mapper looking to enter
data, nor as a map user looking for a pretty rendering - although I wear
both those hats from time to time - but as a newly-retired applied
mathematician (A.B., mathematics, Dartmouth College, MS in electrical
engineering, Arizona State University, PhD, computer science, University of
Illinois, about forty years of experience with GE, Northrop, Honeywell, and
others), with a reasonable background in computational geometry, thinking
of what challenges I ought to tackle next.

Given that one of my principal avocations is hiking, my chief rendering
interest is not with an endlessly-panning map, as useful as that is; it is
with paper maps where labeling must conform with the neatline.  For those
maps, simply placing a point for 'label painting' near the center of an
indefinite feature is not sufficient. Instead, a first step has to be
calculating the intersection of the area feature with the region of
interest, leading to one of the results: (a) the area is totally within the
region; (b) the area is totally outside the region and may be discarded;
(c) the area intersects the region partially and one or more regions of
intersection must have labels placed individually. The 'one or more' arises
from the fact that a non-convex area feature or a non-convex region of
interest (a rectangle, for instance, with a corner cut out for placement of
a legend) may yield more than one polygon of intersection.

You have on several occasions advanced the argument that the central label
should be enough for this and made a contention that I don't understand
about projecting from the central label of a bay onto the shoreline to
reconstruct the area. With that contention came the implication that the
topological information about an indefinite area would not be needed, if
only the renderers and data analysts worked hard enough. Unless you can
provide me with literature citations to what you have in mind, I'm afraid
that I'll have to dismiss your claim as hand-waving. As far as I can tell,
there is no known way to achieve the result that mappers want - or at least
I want - without the detailed geometry of the partially indefinite area. If
you can provide such citations, I'm eager to follow up with you!

I should digress into the phrase, 'partially indefinite,' that I've already
been using.  For the contentious areas such as the Red Sea, the indefinite
portion about which the controversy arises is typically small, and
typically of a nature where a rough approximation is acceptable to all
users. There is no controversy arising from the shoreline of the Red Sea
except for a trivial amount of border.  Very few claim that the Red Sea
exists only as a social construct. Scientists discuss its hydrology and
ecology in contradistinction to that of the region of the Indian Ocean to
which it connects.  Mariners speak of Port Sudan, Jeddah, Sharm al-Sheikh,
or Eilat as Red Sea ports (Eilat may be further specialised as being a port
on the Gulf of Aqaba, a smaller area that is similarly well-defined; the
relation is one of hierarchy rather than exclusion. (Moving somewhat to the
northwest, I've seen papers on hydrology that have tabulated observations
in rows labeled 'Ionian Sea', 'Ægean Sea', 'Tyrrhenian Sea', 'Adriatic
Sea', and so on. There is _some_ shared understanding that those words have
meanings.)

'Partially indefinite' extends to other features such as peninsulæ (a
mirror image of bays - the indefinite boundary is one of land rather than
water); straits (indefinite water margins at both ends); isthmi (indefinite
land margins at both ends); and even rivers and lakes (indefinite at river
mouths, lake inlets and outlets; confluences where tributaries enter, and
divergences where distributaries leave). It even extends to political
boundaries.  There are a number of townships and even counties in my part
of the world where the political boundaries are well known, surveyed, and
monumented in the inhabited regions, but have never been successfully
surveyed; knowing where the line is in the woods and swamps is simply not
worth the cost of mounting a survey. (The general attitude has been, "if it
ever becomes an issue, we'll sort it out.")
https://caltopo.com/map.html#ll=43.91623,-74.49297&z=14&b=t indicates one
example, where - on an official topo map from the US Geologic Survey - the
cartographer has indicated the political boundary as 'INDEFINITE BOUNDARY'
and even indicated an error of closure at the northern corner of the
township. This is NOT a bad splice between two map sheets; note that
streams, lakes, contour lines, are continuous.  The discontinuity of the
grid lines arises from the fact that one map sheet shows coordinates in
NAD83 (nearly the same as WGS84) while the other is NAD27. The error of
closure arises from projecting the boundaries of incomplete surveys that
yielded inconsistent results. Still, if you follow the indefinite line
around, you eventually come to where it transects the hamlet of Oxbow,
https://caltopo.com/map.html#ll=43.43812,-74.49185&z=14&b=t, east of
Rudeston and southeast of Piseco, where the town line is indeed signed on
the highway (source: personal observation). The inhabitants of Oxbow no
doubt care very much about which township collects their property taxes and
maintains their roads; whose justice of the peace adjudicates their minor
disputes, officiates at their marriages, and so on; whose town clerk
maintains their vital statistics.  The reductio-ad-absurdum of your
argument would say that the township of Arietta must not be mapped as other
than a point feature because large portions of its boundary are
indeterminate. The townsfolk would disagree, in light of the fact that such
a mapping loses the information of where the town line is, in the places
where that's known perfectly well.

Returning to the main point, I don't see how to approach the problem of
rendering partially indefinite areas without the geometry of what is known
precisely, and some approximation to the indefinite margins so that the
areas present a closed, non-self-intersecting topology.  Without that
information _somewhere_, there's no place to start from! I would also
appreciate some sort of deprecatory tag (`boundary=indefinite` on the way,
perhaps, or a role of `outer:indefinite` or `inner:indefinite` on the way
within the relation) to indicate where a particular way is arbitrarily
added in order to give a partially-indefinite region a complete topology. A
renderer is free to refrain from rendering indefinite margins, or to treat
them differently in the rendering (like the narrower dashed line and the
callout, 'INDEFINITE BOUNDARY' in the government topo above). Statistical
analysis is free to make the boundary 'fuzzy' and explore how moving it
will affect results. If you find even this approach unacceptable, your
argument presents us with an insoluble 'chicken and egg' problem.  Without
the data, there is no foundation on which to develop and explore how
rendering and statistical algorithms, both known and yet to be developed,
perform against the real world. Without the algorithms already in place,
you argue, the data ought not to be provided.

In my opinion, the solution of 'outer:indefinite' and 'inner:indefinite' on
the role makes more sense, since an object that is indefinite with respect
to one object may be perfectly definite with respect to another. Whether a
particular bit of coastline on the Bar al-Mandab Strait is or is not
fronting on the Red Sea does not make it any less the shoreline of Yemen or
Djibouti.

You have also, in the past, argued passionately against 'foreign keys'
being present in OSM (for instance, labeling waters in the US with their
'reach codes', which are digital identifiers identifying stream and river
segments that are used in diverse government systems in the US) in order to
connect it with non-OSM databases. This argument also forecloses on the
idea that perhaps mappers should record the topology of these areas
externally to OSM , while renderers or analysts should somehow fuse the
multiple sources in their rendering or analysis.

This combination of constraints feels as if you are arguing against the
existence of the Red Sea or the Town of Arietta for some other reason, and
using whatever argument will block it in the moment.  We can't enter the
data regarding indefinite features in OSM. We can't connect to the data
regarding indefinite features outside OSM. We can't develop the algorithms
without real data to prove them against. We can't have the actual data in
the database until the algorithms are developed to consume the data.  What,
pray tell, may we do? All I hear from you is arguments against this and
that. They are diverse and, taken together, are comprehensive enough, that
in the absence of a proposed path forward among them, it is tempting to
dismiss them as simple obstructionism.

That's what mappers are doing. They're mapping indefinite area features
because they see the potential benefits of doing so, even if the
algorithmics have yet to catch up with them.  Until and unless you offer a
better alternative, they will continue to work around your objections.

Note that I am not arguing in favor of anything that I have done on the
map. In the vanishingly rare cases where I have adjusted a indefinite
margin of an object, I've provided detailed discussion of why I believe
that the line is the "best possible from the available information," citing
sources.  https://www.openstreetmap.org/user/ke9tv/diary/391486 and
https://www.openstreetmap.org/user/ke9tv/diary/42951 are examples).  I've
refrained from entering any new indefinite objects at Frederik's request,
which at the time appeared to be made in his official capacity as DWG
member or as OSM board member. (Whether the request actually reflected an
official position or was instead made in his personal capacity remains
unclear to me.) While I've argued about the definition of 'coastline' here,
I've not moved the coastline in the estuarine environment other than to
correct obvious errors in the boundary between land and water. The argument
(which I've heard in the past, and don't recall from whom) that I'm
advancing this argument to justify in retrospect damage that I've already
done to the map does not hold water.

So, how do we move forward? Dismissing the Red Sea as a mere social
construct is unlikely to achieve consensus. Moreover, social constructs are
part of what we map; we live in the human world as well as the physical
one. Objects have names; regions have political boundaries; amenities draw
tourists; historic sites enjoy protection; and we map all those things. The
human world is a world of ambiguity. We can, and should, try to make the
map as definite as possible. We must not discard the human view in doing so.
-- 
73 de ke9tv/2, Kevin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/tagging/attachments/20201124/5145745b/attachment-0001.htm>