[Tagging] Peak-based mountain range proof of concept

Wed Dec 30 00:38:20 UTC 2020

On Tue, Dec 29, 2020 at 10:01 AM Brian M. Sperlongano <zelonewolf at gmail.com>
wrote:

> I went with what I could objectively verify with the help of local
> knowledge.  The Green Mountains are not the only mountain range in Vermont,
> there is also the Taconic range which runs through Vermont and over the
> line into New York.  It's west of US7 in southern Vermont, and there are a
> few mountains in the "Green Mountain National Forest" that are actually
> part of the Taconic range.  Part of the discussion was "should foothills be
> included?" - and the answer was that "Vermonters wouldn't consider the
> foothills to be part of the mountains".  So, I specifically attempted to
> exclude them.  I excluded anything with "Hill" in the name which eliminated
> both the foothills and many (but not all) of the sub-peaks.
>
> The initial data set was formed by pulling down a rectangle around VT with
> natural=peak objects, then filtering out for elevation so I could pick out
> the parallel spines that make the shape of the mountain range that matched
> the rough shape of where I understood the range to be.  (I've done some
> hiking in the GM, but certainly not enough to declare any sort of local
> knowledge).  From that starting point, I consulted actual local knowledge
> to figure out what to add or subtract from that set.
>

 I decided to run with Brian's idea to see how far I could take it, taking
into account some reasonably reilable, public-domain, but non-OSM
information (plus some local knowledge).  Here's a detailed analysis of how
an indefinite 'Catskill Mountains' polygon might be produced - and some
comments on the challenge of identifying a 'spine' as a linear feature.

Warning: This is quite technical. You're welcome to skim and look at the
pictures. I suggest having the Flickr set open in a separate window, so you
can hit the 'forward' control to go through them.

I began the analysis by asking the wrong question.  I picked Slide
Mountain, the highest peak in the range, and asked, 'what's the lowest
contour that encircles Slide but no higher peak?'  That' I thought, would
give the ultimate outer border of the range - anything outside that contour
would be either in the flatlands/valleys, or would "look to" a different
range. It was surely an outer limit: it wound up spanning a comma-shaped
area from the eastern Finger Lakes to the Capital Region of New York, down
almost to Harrisburg, Pennsylvania, encompassing not only the Catskills,
but also the Poconos, a lot of the rolling hills of central New York, and a
sizable chunk of the Appalachian ridge.
https://www.flickr.com/photos/ke9tv/50777004162/in/album-72157717614241653/
Clearly, I needed a tighter boundary than that!  (Although this shows just
how prominent Slide Mountain is: the nearest higher peak along its
prominence line is in West Virginia, and the nearest more prominent peak
along the line is all the way in North Carolina.  The deep valley of the
Mohawk River cuts it off from the nearer higher summits to the north and
east.)

Instead, I asked the question, 'what is the highest-elevation contour that
still encloses the 35 summits of the Catskill high peaks.
https://en.wikipedia.org/wiki/Catskill_High_Peaks   The resulting polygon
was encouraging:
https://www.flickr.com/photos/ke9tv/50777003997/in/album-72157717614241653/.
The lowest point needed to make a single connected polygon with the 35
peaks under consideration was a pass on New York State Route 42 separating
the drainages of West Kill - Schoharie Creek - Mohawk River - Hudson River
and Bushnellsville Brook - Esopus Creek - Hudson River.
https://www.flickr.com/photos/ke9tv/50777003802/in/album-72157717614241653/
Raising the contour by another ten metres divides the polygon there. (I
actually took a shortcut: I asked "what's the HIGHEST enclosing contour for
Hunter Mountain (the second most prominent peak in the range) that excludes
Slide Mountain or any higher peak?" and found that thresholding at that
elevation produced a polygon that enclosed all 35 high peaks.)

Along most of its border, it appeared to be nearly coterminuous with the
Catskill Park, and a quick glance (informed by intimate local knowledge)
made me say, "yeah, that's all mountainous"... and prompted me to look for
further gaffes.  Here's where subjectivity starts to come in a little bit -
I'm applying local knowledge about what the inhabitants would say "are the
Catskills" or "are not the Catskills". Fortunately, none of my calls would
be at all controversial locally.

https://www.flickr.com/photos/ke9tv/50777003787/in/album-72157717614241653/
- This area is accounted by the locals, including the local geologists, as
belonging to the Helderbergs rather than the Catskills.  The demarcation is
pretty clear - the narrow neck at the pass on County Route 3.  Given this
local knowledge, I cut this area off manually - without changing the
elevation threshold elsewhere.

https://www.flickr.com/photos/ke9tv/50776134543/in/album-72157717614241653/
- The elevation threshold isolated Bald Mountain, Tower Mountain and some
unnamed heights of land in the Town of Jewett.  Retaining the same
elevation threshold, I added the missing polygons.

https://www.flickr.com/photos/ke9tv/50776892081/in/album-72157717614241653/
- The elevation theshold isolated Mount Tremper, Carl Mountain, Mount
Tobias and Mount Guardian, on a ridge between Mount Tremper village and
Woodstock.  Again, I kept the same elevation threshold and added the
missing polygons.

https://www.flickr.com/photos/ke9tv/50776892026/in/album-72157717614241653/
- Ticetonyk Mountain is below the Catskill Escarpment, but is counted among
the Catskills.  It would have been isolated even using the criteria of the
failed attempt with Slide Mountain. I retained the elevation theshold and
added the missing area.

https://www.flickr.com/photos/ke9tv/50777003512/in/album-72157717614241653/
- Ashokan High Point, and the rest of the peaks on its ridge, are cut off
by the pass between the Rondout and Ashokan valleys. Again, they're very
close to the boundary of the original region. I retained the elevation
theshold and added the missing areas.

https://www.flickr.com/photos/ke9tv/50776891851/in/album-72157717614241653/
- Plattekiill Mountain and Mount Utsayantha are accounted among the
'Catskill Hundred Highest', and are again separated by only a short
distance (and the East Branch of the Delaware). The same drill - retain the
elevation threshold and add the missing areas.

https://www.flickr.com/photos/ke9tv/50776134113/in/album-72157717614241653/
- The good burghers of Delhi and Andes consider the hills to their
southwest to be among the Catskills, and who am I to argue?  Again, the
elevation threshold can be kept and gives a clean polygon.

https://www.flickr.com/photos/ke9tv/50776133938/in/album-72157717614241653/
- The mountains are really petering out into rolling hills here, and hardly
anyone would argue with fixing the East Branch of the Delaware as the
ultimate western limit of the range. I excluded this region.

That pretty much got me around the whole border of the region, and left me
with a reasonably tidy, and objectively mountainous, area shown in
https://www.flickr.com/photos/ke9tv/50777002877/in/album-72157717614241653/.

The chief purpose of having such an area would be shaped label placement.
Rendering its boundaries would never be appropriate. But having the shape
would be important, particularly in rendering paper maps whose area of
interest partially overlaps with the region.  For instance, a map showing
only the extreme eastern portion might well want to have a curved label,
'CATSKILL MOUNTAINS', following the curve of the ridge above the eastern
escarpment.    I know that this sort of sophisticated label placement is
beyond our current capabilities, but we have a chicken-and-egg problem that
it's impossible to do better cartography without the data to support it.

In many cases, we'd want to simplify (more accurately, generalize) the area
for rendering at a smaller scale.  For stand-alone areas like this one, one
of my favourite methods for generalization is morphological closure.  I
include a few images indicating how this can work.  For the closure, you
have to pick a length scale.  A workable scale is often about 1-2% of the
largest dimension of the area feature; that suffices to retain the shape
but to fill in gaps and smooth out rough edges. We get this dimension by
calculating the smallest enclosing circle.
https://www.flickr.com/photos/ke9tv/50777002732/in/album-72157717614241653/
- here I found that the radius of the circle was 51.7 km.

I therefore buffered the area by 2.07 km - 2% of the diameter, to arrive at
https://www.flickr.com/photos/ke9tv/50776133413/in/album-72157717614241653/
- which is overly large, but fills in all the gaps.  The 'morphological
closure' operation is then accomplished by buffering by _negative_ 2.07 km,
removing the excess around the perimeter but leaving gaps and indentations
filled.  This should be a very nice simplified enclosing polygon for
applications that need one.
https://www.flickr.com/photos/ke9tv/50776890916/in/album-72157717614241653/

I also did a rough ridge line analysis at
https://www.flickr.com/photos/ke9tv/50776308698/in/album-72157717614241653/
.  The most significant ridge - the only one separating the basins of
first-order rivers - is the Catskill Divide, with the basin of the Delaware
to the west and that of the Hudson to the east.  It's hard for me to see a
good way to choose a 'spine' for the range with the ridges running as they
do in every direction.  If I were asked to choose the next most significant
ridges, they would be the one separating the East Branch and West Branch of
the Delaware, running southwest from the Catskill Divide at top center and
ending at the border of the mountains near the Cannonsville Reservoir; and
the one separating the Mohawk River basin from those of other Hudson River
tributaries. It starts from the Catskill Divide near the west border of the
Hunter-West Kill Wilderness, runs generally east almost to the eastern
escarpment, then turns abruptly north and follows the ridge of the
escarpment, continuing north out of the region into the Helderbergs
beyond.  But then I'd feel to add the ridge of the Pepacton range to the
west (south of the reservoir near the center of the image), so as not to
lose all linear labeling in that area. I think that the 'label the ridges'
idea has been defeated by the complex topography (and equally complex
topology).

Fortunately, I think that maintaining a fixed elevation threshold for a
given range gives us some hope of having manageable (multi)polygons with a
modicum of objectivity.  I'd even be willing to make the sacrifice of,
'that lobe to the north is really part of the Helderbergs, but OSM rules
insist it has to be grouped with the Catskills' as a less-than-ideal
solution if it will get the verifiability police off my back.

-- 
73 de ke9tv/2, Kevin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/tagging/attachments/20201229/d74c3f21/attachment-0001.htm>