[Tagging] Feature Proposal – RFC – natural=peninsula (Was: Feature Proposal – RFC – place=peninsula)

Kevin Kenny kevin.b.kenny at gmail.com
Thu Jan 10 22:52:11 UTC 2019


On Thu, Jan 10, 2019 at 4:47 PM Daniel Koć <daniel at koć.pl> wrote:
> BTW: if we want to use proportion of length to area, we claim that we
> know this area somehow.

If the feature is represented as a polygon, we do.

If we have confidence that at least some shoreline is unambiguously
part of the peninsula's shoreline, take the smallest shoreline in
which we have confidence, strike an arbitrary line between its
endpoints to complete a polygon, use that polygon to compute the area.
Since this is a rough guideline, it doesn't need to be mathematically
precise. Whether you estimate that the boundary of Cape Cod is the
canal, or a line from approximately Wareham to Plymouth, or a line
running approximately NE from Buttermilk Bay, you'll still find that
the length of the land border is many times less than either the
length of the waterfront or the square root of the area. Similarly,
with Nova Scotia, the answer won't change whether you draw the border
from Truro to Trenton or across the narrowest part of the Chignecto
Isthmus, or at the provincial border.

For almost any tag, in almost any situation, we defer to local mappers
when we come close to the borders of "what sort of object is this?"  I
would not protest if a Michigander were to map the Upper and Lower
Peninsulas, setting their boundaries as the state line, because that's
the local understanding.  If the Iberian Peninsula were to be mapped,
I'd defer to the Europeans where to put the arbitrary land border. It
seems that cutting across the narrowest part would be wrongheaded - it
would exclude much of Catalonia, including Barcelona. But I'd let the
mappers of Andorra decide whether the arbitrary line should include
them.

For the use cases I have in mind (using the shape of a
partially-indefinite area to guide label shaping as well as
placement), the precise arbitrary line will make no significant
difference to the rendered result.  The worst case will be that the
terminal A in a curved IBERIA might be a little closer to Zaragoza or
a little closer to Toulouse.  If the 'G' in 'Gulf of Bothnia' (or the
'P' in 'Pohjanlahti) is a little bit closer to or farther from Åland,
that will not matter a bit - the label will still render on a curved
light, roughly south to north, from there to somewhere offshore of
Oulu. But rendering it well does require information about the shape
of the object.

In short, there are effective uses to be made of the data, and
effective guidelines that can be set for tagging them, even if there's
less than perfect precision in the guidelines, or the shape depends on
some indefinite boundary.  To discard that information on the basis of
'verifiability' is 'all or nothing' thinking. Surely the Gulf of
Finland verifiably exists. Surely Helsinki, Tallinn and Saint
Petersburg are all verifiably on its shores.  The fact that there is
no precise line to separate it from the Baltic Sea should not prevent
an application from being able to benefit from the incomplete
information that is knowable about its shape.

It's a delicate philosophical point, and it's one that we stumble over
in this list regularly: when the existence of an object is verifiable,
but there is only partial information known (or knowable, as in the
case of virtually any waterbody or most area landforms) about its
geometry or other attributes, is that object 'verifiable' for the
purposes of OSM?  We have decided that we can map rivers - despite
there being an impossible-to-define line that separates the river from
the lake or sea into which it empties. (And we haggle endlessly over
the rules for drawing that line.) We've decided that mapping bays is
mostly unacceptable - because of the arbitrariness of the mouth of the
bay but in practice more because larger bays produce data that
overwhelms databases and data consumers (hence the revert of the Gulf
of Bothnia relation) - and similarly, we reject seas. We mostly accept
political divisions with indefinite borders (but struggle over
disputed ones). Now we struggle with peninsulas.
The stated reasons are all the same - that a portion of the geometry
is not verifiable, therefore even the verifiable part of the geometry
must not be mapped.

I'm no philosopher. I simply have a wish to render certain features
based on the incomplete data that are knowable and verifiable. Some of
the philosophers tell me that my understanding of the world is wrong -
that such features aren't 'real'. I *think* I see them.



More information about the Tagging mailing list