[Talk-GB] Solar update Q3 2020

[SK53]

I'd meant to provide an update at the end of August, just after the peak of
a frenzy of adding solar power from the new higher quality Bing imagery. As
I didn't, now is a good time to summarise recent progress: September has
seen more modest gains.

Some headline figures:

   - We passed 250k OSM elements on 31st August and 30% completion
   (measured against FIT, which does not cover the totality of solar
   installations) a few days later
   - 2 additional LA areas joined the, anomalous, Isles of Scilly, in
   passing 100% of the FIT target: Torridge and Tower Hamlets.
   - 7 more LAs have reached >90% of FIT. Joining Nottingham and Plymouth,
   these are Canterbury, North Devon, Exeter, Torbay, Derby and Chesterfield
   - 30 LAs are now between 80 & 90% of the FIT numbers
   - in total around 100 LAs have over half of the FIT totals mapped,
   including 4 in Greater London
   - Devon, Derbyshire & Nottinghamshire have all LAs over 80%
   - Leicestershire, Lincolnshire and Warwickshire are around 60% complete
   - Much of the rest of the West Midlands is around 30%
   - Edinburgh was the first Scottish LA to pass 50% (thanks Donald)
   - Visibility of solar in Greater London has changed dramatically. Russ,
   Dan & Yvonne have been the most active mappers.

>From my own mapping, it's now clear that in most parts of the country the
imagery is probably good enough to find at least 90% of installations. This
raises the bar compared with previous imagery where for most LAs finding
much more than 50% of solar was too hard to be practical. It perhaps also
changes what our focus should be. The following paragraphs touch on some of
these.

The other significant aspect of better imagery is that it is now feasible
to provide module counts and orientations for the vast majority of panels.
At present only 8.5% of the panels already mapped have these tags (although
this represents 20k panels, valuable in its own right). I've therefore
moved to improving the percentages in the counties I focussed on recently
(Devon, Derbys and Notts). It's slightly tedious work, but I think it's now
important to avoid building up 'tagging debt' in this area. For LAs with a
lower percentage of mapping (upto 60%) it's often possible to combine
adding these tags with looking for unmapped solar power, which is rather
more engaging, as can be seen by totals for places like Gedling, Rushcliffe
& Derbyshire Dales. This latter type of activity can combine well with
general updating of an area, refinement of existing elements and correction
of errors. If half the new panels added as we advance to 300k have these
tags we'll double this percentage. I've moved to generally using direction
rather than generator:orientation because this is visualised in iD and I
can't see any ambiguity in the use of direction on solar elements (the
exception is where solar tags have been added to a building). This
<https://overpass-turbo.eu/s/YAF> overpass-turbo query can be used to
visualise the status of solar within any LA: green has both tags, purple is
a large installation with "lots" of modules, red lacks orientation & module
count, yellow & orange lack one or other.

Frederik mentioned a while back about solar elements in places where little
else is mapped. Solar is slightly unusual for OSM in that the utility of
the data in most cases needs fairly comprehensive coverage, whereas a
single address is useful if I want to route to it. Use of nodes for much of
the mapping does allow local mappers adding more detail to correct position
and so on, and there is a small, but steady, flow of edits of this type.

Adding buildings themselves comes up against the imagery alignment issues
which have been discussed here recently, and therefore, in general, needs
more care. In many places there were already substantial numbers of
buildings mapped, but a particular problem is that these can appear
misaligned with modern imagery (in some places as much as a 5 m
discrepancy). In the main topological relationships between solar panels
and supporting buildings have been maintained, but some cases this is not
so: notably extensions built since the building was originally mapped and
buildings mapped from OS StreetView. My particular bugbear here is a
multi-storey car park in Exeter where the individual parking spaces on the
roof have been mapped, but subsequently solar panels have been installed
above these spaces.

For buildings in general I think any mapping in relation to solar should
focus on larger buildings with big solar installations: farm & retail
sheds, schools & other educational facilities, hospitals, factories and so
forth. This <https://overpass-turbo.eu/s/YAG> overpass query shows how
solar on a particular amenity class can be found (don't use on too large an
area). Obviously landuse classes can be substituted
<https://overpass-turbo.eu/s/YAG>. In principle panels mapped as nodes with
a large number of modules can be found with overpass, but I haven't worked
out exactly how to do this, or to use the MapCSS eval to style them
differently. These data lend themselves to refinement (converting to areas,
adding tags, adding underlying buildings) via a MapRoulette approach.

In terms of coverage the general accretionary approach is working well NW
of the Foss, but Southern England and East Anglia are the two areas now
with low numbers.

Dan will also have views as to what is most useful from a data consumer
viewpoint.

Thanks to everyone who's contributed. There's still plenty of mileage in
this area: in particular, use of the data for training machine learning
algorithms to identify panels from imagery and then combining that with
regular mapping by people.

On to 300k!

Jerry
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/talk-gb/attachments/20201001/162bfebf/attachment.htm>