[Talk-GB] Beta release: Address mapping tool

Sun Feb 13 11:50:32 UTC 2022

Rob, Pieter,

Thanks for this. It looks like a good step in the right direction, but
I've got a few observations so far:

- As you've already noted, the loading of the data can be a bit slow
sometimes. It'd be great if anything could be done to improve this at
all, as I expect that potential contributors may not have such a fast
computer / device and Internet connection as myself, therefore may see
slower performance and may get frustrated by this. I've tried with both
Firefox 97.0 and Chrome 98.0.4758.80 on Linux.

- Looking at some addresses near to me it seems to show fewer address
fields as populated compared with if I use JOSM to load the same data.
For example for 1 Manor Close [1][2] I see just addr:housenumber and
addr:street elements rendered on the example envelope, but OSM's data
does already have addr:city and addr:postcode, too. If this is because
the tool is focussed on just the street-level elements, then perhaps it
would be more reassuring to users to show that likes of addr:city,
addr:postcode etc. with their actual values in OSM, where known, in the
blue text instead of the placeholders such as [Town] and [Postal code]?

- In a nearby adjacent area such as Athelstan Road [3] with a number of
probable council properties the suggested addresses look somewhat
random. I realise that this is due to the effective incompleteness of
the input data, but wonder whether this may lead to some addresses
getting incorrectly input as a result of this (it wouldn't be
immediately obvious exactly how far along the road the marks shown are,
without switching to the aerial imagery view, so the user may
inadvertently think that they're looking at the adjacent house or two
further along the road)

- When in the view to add an address after having clicked the "Add this
address" text, such as [4], then the top map (immediately beneath the
"Testing - changes won't be saved" text, with the red North - South -
East - West arrows) doesn't show any tiles beneath it for the OSM or
satellite views. The minimap at the foot of the view does correctly
show the OSM tiles and the place mark atop it, though. I'm seeing this
just on Chrome, but not on Firefox.

- Without wanting to over-complicate the UI too much, could other
imagery provider's tiles also be shown? (Where data derivation for OSM
is permitted, of course). The Mapbox Satellite view shown is far too
blurry to be of any use in the areas that I've looked at, but I know
that the Bing imagery here is quite a bit better.

- I'm trusting that this text is simply for development purposes? It
seems too techy for targetting at end-users:
The new POI will have urpn_count=1&ref:GB:uprn=100060813240

Looking forward to seeing the next iterations,

Gregory

[1] https://www.openstreetmap.org/way/889279907
[2]
https://pietervdvn.github.io/mc/develop/uk_addresses.html?test=true&z=19&lat=51.2659&lon=1.053995&filter-to_handle=true&language=en#way/889279907
[3]
https://pietervdvn.github.io/mc/develop/uk_addresses.html?test=true&z=19&lat=51.26738&lon=1.057388&filter-to_handle=true&language=en
[4]
https://pietervdvn.github.io/mc/develop/uk_addresses.html?test=true&z=19&lat=51.26882&lon=1.056312&filter-to_handle=true&language=en#https://osm-uk-addresses.russss.dev/addresses/16/32960/21862.json/67

On Fri, 2022-02-11 at 22:39 +0000, Rob Nickerson wrote:
> Hi all,
> 
> As previously discussed OSM UK has been working on a new way of
> mapping addresses. We're at a stage where we are ready to share this
> and ask for feedback. My thanks to Russ and Pieter for the
> development work and the many people who have provided support and
> advice since we came up with this idea 12 months ago.
> 
> First a recap: The new tool makes use of openly licensed datasets in
> order to predict where an addressable point is. To do this we take OS
> generalised buildings and land ownership (Cadastral) parcels. The
> land parcels are used to split the generalised buildings (e.g split
> semi-detached houses into their two parts) and the centre of these
> split buildings become our predicted locations for addresses. These
> are then compared to the existing data in OpenStreetMap. We check for
> cases where the point falls within a building way in OSM, and when
> OSM already has an address within the Cadastral parcel. All of this
> is then presented in the user interface (we are using MapComplete as
> it works across desktop and mobile). Contributors can then add the
> address following a ground survey.
> 
> With that in mind here is the link to the demo tool (in test mode so
> no data is written to OSM):
> https://pietervdvn.github.io/mc/develop/uk_addresses.html?test=true&z=7&lat=52.92215&lon=-1.87866&filter-to_handle=true&language=en&welcome-control-toggle=true#welcome
> 
> A few initial observations:
>    1. As it is in test mode, you see some developer debug (e.g. the
> code names for each block in the questions popup). Please ignore this
> as it won't appear in the final version.
>    2. It can get slow if it tries to load too much data. If this is
> the case I find refreshing the page at a fully zoomed in position
> works well.
>    3. Currently when you try to import a point you get the option to
> move the location. As the data is coming from OS and the cadastral
> parcels this doesn't make sense. I'm hoping we can get rid of the
> screen that allows you to move the point.
>    4. A ref:GB:uprn tag is added if only one UPRN is found within a
> cadastral parcel.
>    5. A uprn_count tag is also added. I personally find this useful
> as it is a good indicator of how many addresses a point covers,
> therefore giving a good piece of info to support QA checks. It also
> can help interpret if something like house number "1-6" is six
> addresses or three (if just odd or even numbers). Keen to get
> feedback on this proposed tag.
>    6. There may be some points proposed that are not actually
> addressable locations. Currently you should just ignore these but
> another option is that we have a button to say "Not an address". We
> could then ask what it is and add a noaddress=yes and fixme= tag to
> OSM. Again feedback on this idea is welcomed.
>    7. Due to the approach we selected, it does not always offer an
> address point. This is particularly true in city centres and areas
> with social housing. It comes from the lack of Cadastral parcels. We
> previously discounted the raw UPRN data as our source of address
> locations (due to too much noise) so there is not much we can do
> about this. You can pre-add a partial addr:* point to OSM (e.g.
> addr:street) in another map editor and then add the rest of the
> address using this tool.
>    8. It is not yet possible to add a completely new point that is
> not in the pre-processed data. I am curious as to whether this
> functionality should be added or not given that the aim is to share
> this tool with people completely new to OSM.
> 
> Finally, for those of a more technical nature, the datasets I
> described in the recap section are available as tiled GeoJSONs and
> MVT tiles. You can then use these in other places should you wish.
> For example the MVT can be added to JOSM (Imagery -> Imagery
> Preferences... -> Click "+MVT" button -> paste in
> https://osm-uk-addresses.russss.dev/addresses/{z}/{x}/{y}.mvt and add
> a maxzoom=21). Being a MVT layer, you can right click the layer name
> in the Layers side-panel and click "Convert to OSM data" to access
> it. Don't upload the whole thing; instead carefully pick features
> that you want to use for a ground survey / manual inspection.
> 
> I look forward to hearing your feedback on this. The aim is to then
> get this in the hands of people completely new to mapping so that we
> can speed up the address mapping task.
> 
> Thank you
> Rob
> _______________________________________________
> Talk-GB mailing list
> Talk-GB at openstreetmap.org
> https://lists.openstreetmap.org/listinfo/talk-gb