[Tagging] Lyft and nameless sectioning in OSM

Wed Oct 12 21:18:57 UTC 2022

> You seem to imply this is trivial, so feel free to build a prototype of
this to see how accurate it is (note that you will need to judge the
automated method against manual mapping, not the other way around). But
most areas are probably not mapped in enough detail yet for this to work.
For example, in the Houston area you shared most buildings are only tagged
with building=yes or not even mapped at all, how will you be able to
automatically infer the landuse from that? But someone can manually
identify and map a residential area even if they don't want to map all the
houses there yet (https://www.openstreetmap.org/relation/7916453).

> Even if you have an area mapped in high detail, I don't expect this to be
that easy. In this case landuse will also generally have been accurately
mapped a long time ago.

I want to highlight this point from Enno.

I'd say if you think it's going to work, build it and show the community
examples of where it works well and where it doesn't. Discussing the
hypothetical makes us all revert to our own assumptions rather than looking
at a real comparison. I'm personally skeptical that a model will work very
well, because right now, land use tagging appears to fill in for when we're
missing more detail in the DB already, which suggests to me that to take
the more detailed data and try to infer the general case is something the
dataset isn't ready for except, maybe, in some areas.

I'd also say, as someone who writes spatial analysis models for a living,
computer models are not objective. We imbue them with our assumptions, we
set thresholds, we define data sources, and more. I'd suggest that the
model will be a subjective model that encodes how its author would
categorize land use. It will just do it at a different scale (and create a
different set of challenges for verification).

But again, I'm not saying don't build it - just expect serious scrutiny of
its outputs. I also don't think the existence of a hypothetical better
method is a reason for others to stop doing it their own way, assuming both
result in acceptable data - OSM is a project of people adding the edits
they're willing and able to edit. The methodology is only of concern if the
result is bad, and I shared my opinion on that above.

-Nick
http://nicksantos.com

On Wed, Oct 12, 2022 at 1:31 AM Enno Hermann <enno.hermann at gmail.com> wrote:

> You seem to imply this is trivial, so feel free to build a prototype of
> this to see how accurate it is (note that you will need to judge the
> automated method against manual mapping, not the other way around). But
> most areas are probably not mapped in enough detail yet for this to work.
> For example, in the Houston area you shared most buildings are only tagged
> with building=yes or not even mapped at all, how will you be able to
> automatically infer the landuse from that? But someone can manually
> identify and map a residential area even if they don't want to map all the
> houses there yet (https://www.openstreetmap.org/relation/7916453).
>
> Even if you have an area mapped in high detail, I don't expect this to be
> that easy. In this case landuse will also generally have been accurately
> mapped a long time ago.
>
> -Enno (eginhard)
>
> On Tue, Oct 11, 2022 at 10:55 PM Evan Carroll <me at evancarroll.com> wrote:
>
>> Thanks  Joseph Eisenberg! That's exactly what I'm looking for. Good
>> answer. So basically the primary use case of an **unnamed** residential,
>> commercial, industrial, and retail "Zones" is not to convey (additional)
>> information but to serve as a good-enough styling solution about what the
>> zone conveys? As I suspected, it is solving a spatial clustering problem
>> manually to achieve a good-enough different visual style. I guess you can
>> see here in the landcover.
>>
>>
>> https://github.com/gravitystorm/openstreetmap-carto/blob/cae2309efd4ee0338fcdf9f201e92f20b338426c/style/landcover.mss#L16
>>
>> The next question is can this be defined such that this can be automated?
>> It would seem to me like if we,
>>
>> 1. Take a bounding box.
>> 2. From that, subtract out the landuse polygons for named zones, LEAVING
>> ONLY "unnamed zones, and land not in a zone."
>> 3. From that, subtract out the landuse polygons for zones NOT of type
>> "common landuse key values - developed land", per the wiki
>> https://wiki.openstreetmap.org/wiki/Key:landuse LEAVING ONLY "unnamed
>> zones of developed land, and land not in a zone."
>> 4. Subtract out by key:highway lines LEAVING ONLY "unnamed zones of
>> developed land and land not in a zone that does NOT intersect a highway."
>> 5. Extract "unnamed zones of developed land and land not in a zone that
>> does NOT intersect a highway" into a zone set.
>> 6. Infer from the contents of the polygon what type of developed land the
>> zone is.
>>
>> This would allow us to be precise and objective and determine what kind
>> of "developed land" an unnamed zone was, as well as to provide a gauge of
>> the accuracy of the zone.
>>
>> If this was done, do you think this would satisfy all use cases of
>> unnamed landuse zones for developed land like commercial, retail,
>> residential, educational?
>>
>> (I said in the above "unnamed zone" for simplicity, and unnamed zone can
>> still have an operator, and in every case when I said unnamed zone, I meant
>> a zone without a name or an operator).
>>
>> --
>> In the event this email pertains to Real Estate, Texas law
>> <https://texreg.sos.state.tx.us/public/readtac$ext.TacPage?sl=R&app=9&p_dir=&p_rloc=&p_tloc=&p_ploc=&pg=1&p_tac=&ti=22&pt=23&ch=531&rl=20>
>> requires all license holders provide to prospective clients the following
>> forms: Information About Brokerage Services
>> <http://docs.evancarroll.com/realestate/iabs.pdf>, Consumer Protection
>> Notice
>> <https://www.trec.texas.gov/sites/default/files/pdf-forms/CN%201-2.pdf>.
>> My TREC license number is 610570
>> <https://www.trec.texas.gov/apps/license-holder-search/?detail_id=881377951>
>> and my sponsoring broker is NB Elite Realty LLC.
>>
>> --
>> Evan Carroll - me at evancarroll.com
>> System Lord of the Internets
>> web: http://www.evancarroll.com
>> ph: 281.901.0011 <+1-281-901-0011>
>>
>>
>> _______________________________________________
>> Tagging mailing list
>> Tagging at openstreetmap.org
>> https://lists.openstreetmap.org/listinfo/tagging
>>
> _______________________________________________
> Tagging mailing list
> Tagging at openstreetmap.org
> https://lists.openstreetmap.org/listinfo/tagging
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/tagging/attachments/20221012/abac1ed1/attachment-0001.htm>