[Talk-GB] Rooftop Solar & UPRNs

[SK53]

When I wrote to this list at the end of June I little suspected that we'd
achieve 25% completion of solar panels by the end of July. Obviously,
access to greatly improved imagery has made a big difference.

The places mentioned by Dan : LAs around Exeter & the Midlands now have
good coverage. The Devon LAs are all over 60% and other LAs in the county
are also progressing. In the Midlands it's easier to list places with low
coverage: N. Staffordhsire, most of Shropshire, East Lincs and
Northamptonshire.

One of my concerns mentioned last time was missing installations in rural
areas. Even with 60% coverage it is noticeable that rural LSOAs are less
well populated. I've recently been experimenting to see if this can now be
addressed and have good results.

Initially I used areas where OSM has most buildings mapped (Derbyshire
Dales near Bakewell & in South Hams between Kingsbridge & Dartmouth). I
pull buildings down into JOSM with an Overpass query, and add existing
solar mapping. Buildings are selected and added to the to do list (a
plugin) & then I step through each building. This was effective, but the
bbox of individual LSOAs resulted in very large numbers of buildings
(~8000), which is really too many for a single task.

I then turned to UPRNs. It is relatively easy to filter UPRNs by LSOA
(e.g., in QGIS) and numbers are more manageable (say 1,500-2,000). Again
stepping through these resulted in finding virtually all the solar
installations expected from the FIT numbers (tested on East Devon 0009A -
Branscombe & Derbyshire Dales 0008C - Tissington & Parwich). However the
number of items is still too high even when divided into batches, and
requires quite some time to work through.

One problem is the sheer number of UPRNs which are not related to
buildings. Numerous minor tracks, possibly some footpaths, farm ponds,
mobile phone masts, old quarries etc. These may make up as much as 40% of
all UPRNs.

An obvious solution would be to use only UPRNs which pertain to buildings,
but I didnt have an OS Local building layer available and even then the
total number of search locations is still too high.

Instead I've used clustering of UPRNs which seems to give reasonable
results. A simple clustering based on distance yields around 100 clusters
which can be searched visually. The non-building UPRNs tend to move the
centroid away from groups of buildings, but not so far as to be unworkable.

I've used QGIS so I thought I'd document that approach in case anyone
fancies using it in there own area (obviously it can be used for things
other than solar):

* *Filter *UPRNs by LSOA. I use a clipping operation in QGIS. A shapefile
of LSOAs is available from the ONS site, but there is also a file
<https://geoportal.statistics.gov.uk/datasets/ons-uprn-directory-january-2019>
of UPRN=>Administrative Geographies which may enable this to be done on a
Unix command line.
* *Cluster*. Search for clustering on the Toolbox option of the Processing
Menu. A number of clustering techniques are available. The one I used is
DBSCAN. Open this can apply settings of minimum cluster size of 1 and
maximum distance of 0.0025 (approximating 250 m in WGS assuming 100 km /
degree). Run the tool and results appear as a new layer. This appears
identical to the original UPRNs, but each is now assigned a cluster id.
* *Group Clusters*. From the Vector menu apply Collect Geomtetries from the
Geometry Tools menu. This returns a MULTIPOINT layer rather than the
original POINT layer.
* *Located Centroi*d. The centroid of each cluster can be found by applying
Centroid from the Geometry Tools Menu. This latter layer can be saved as a
geojson file for use in JOSM (or iD or Potlatch).

In JOSM:

* Open the geojson of clustered UPRNs.
* Download existing solar data using an overpass query within the viewport
of the LSOA data. Make sure this is a new layer as this is the layer used
for editing.
* Select all items in the UPRN layer and add them to the to do list.
* Activate the solar data layer.
* Step through each item in the todo list searching for buildings within a
few hundred metres of where JOSM zooms too. Add any rooftop or ground solar
panels missing. Using nodes minimises likely conflicts as not all OSM data
is loaded.

In practice I'm very conservative with the first 10 or so items in the todo
list  and search in a bigger area, but as one steps through the items one
can have greater confidence in the localisation of each cluster.

I'm finding over 80% of installations predicted from FIT in very rural
LSOAs with this approach. It still needs a bit of refinement, but I think
30 minutes / LSOA is readily achievable. Taken together with the new
imagery this bodes much better for coverage of parts of Britain with
dispersed rural settlements.

Jerry
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/talk-gb/attachments/20200801/eeab32b7/attachment-0001.htm>