Mapping roof-top Solar Panels

We've all noticed that solar panels have become increasingly frequent on the roofs of residential buildings. It's one of the things I have taken note of ever since I started contributing to OpenStreetMap. However, I had never tried to add any to OSM. Until now!

Roof mounted solar PV panels on a semi-detached house.
There are 18 panel modules (in 2 rows of 7 and one of 4). Each module consists of a 12 by 6 array of solar cells (see below for further discussion).
Photo by Phil Sangwell on Flickr via Wikimedia Commons. CC-BY-SA

A couple of days ago Jack Kelly suggested that perhaps we could use OSM to capture the presence of solar panels across the UK.  A lively twitter discussion ensued.

How crazy would it be for national electricity grids to use @OpenStreetMap as *the* authoritative database of all solar photovoltaic systems?@mclduk @alex_rogers_cs @dekstop @EnergySRE

— jack_kelly (@jack_kelly) March 6, 2019

It seemed sensible to have a go at scoping what was involved. Over the past few months I've been improving the mapping of inter-war housing estates in Nottingham, with the current focus on the Aspley Estate which contains perhaps 2,400 houses. In the course of visits and scrutinising aerial imagery I already knew that there were a fair number of roof-top photovoltaic (PV) panels already installed. Unfortunately on my last visit to get representative photos of the buildings I only caught one house in the background.

Housing in the Aspley Estate: note the solar PV panel on house in background (this one on OSM).

Particularly useful is that the best quality imagery layer available for Aspley is Bing, and it shows the panels clearly. In fact sufficiently clearly that I decided to map them as areas.

In a relatively short time (15-20 minutes) I had found just over 200. Unfortunately I was also reminded that there were quite a few houses in the NW sector of the estate which I had not mapped, so I then spent a while adding houses and addresses, followed by fixing a lot of QA issues pointed out by the JOSM validator tool. Only then was I able to align the houses & solar panels, which took another hour.

This was a little long winded for first-cut mapping, so the following morning I gave myself 30 minutes and searched through adjacent housing estates which I suspected would have a similar density of panels as I had found in Aspley.

Solar Panels on new-build replacement housing at Rutland Close, The Meadows.
A recent example of housing I've surveyed without paying particular attention to solar.

One reason for this is that social housing often has a much higher density of solar panel installation than private housing. Firstly the housing stock is often of similar or identical buildings under one ownership enabling economies of scale. Secondly, owners of social housing are much exercised by fuel poverty of their tenants: reducing fuel costs through providing electricity from solar power therefore has much to commend it. Thirdly, Nottingham City Homes, the at-length housing provider for Nottingham City Council, has a great deal of expertise in applying greener energy polices to their housing. Through an odd coincidence I saw a tweet about what they have achieved as I was doing my second scoping task. This set a target of around 4,500 panels to find in the city.

With my second run just mapping each panel as a node I found 320, or just over 10 a minute. This was partially targeted because I pretty much restricted myself to examining areas of social housing. It therefore represents a rather efficient data acquisition rate.

Jack extrapolated this to the whole country mapped in 5 days by 33 people working full-time. This is of course highly optimistic because I was mapping areas I know well from having being mapping them for 10 years on OSM. However, the OSM community is full of people with detailed knowledge of their local areas, so this ought to apply or many parts of the country. even if it was 5-10 times as much effort, say 1 a minute, 100 people mapping for a couple of hours a week for a quarter might find 150k. This suggests solar panels may be a good subject for a Quarterly Project.

Solar PV panels mapped in Nottingham, via Overpass Turbo

Now, 48 hours after I started, I have added 1760 solar panels to OSM in Nottingham. It's time to summarise what I have learnt. In no particular order:

• All available imagery layers need to be searched. New installations are occurring all the time and it's unlikely that the better quality imagery layers will be recent enough to enable adequate coverage.
• Newer imagery, such as the Digital Globe layers can be quite grainy & hard to interpret. However once panels have been spotted it is usually possible to then find many more.
• The huge variability in available imagery is likely to make any attempt to use machine learning to identify targets is likely to be fraught. I would also expect things like glass roofed extensions would generate many false positives.
• Knowing where panels are likely to be installed helps a great deal: both at the neighbourhood and building level. Christian Quest's OpenSolarMap used some crowd-sourced information from aerial photos to train  system to identify buildings in France with potential for installation of photovoltaic panels. Such information for the UK could reduce the total number of buildings needing to be inspected.
• Larger detached houses with solar panels are very difficult (impossible?) to pick out from aerial imagery. Shadows from chimneys, and changing roof lines obscure the presence of the panels.
• Mapping panels as nodes is the best approach initially. I used ID which has a suitable preset (and checked what others had already done, for instance brianboru around Birmingham). Thereafter I just copied the original node.
• Adding a tag to show that they are roof-mounted is useful (particularly if the building has not been mapped yet). I've used generator:location=roof. Indicating domestic use might also be helpful. The basic tags I used are also used for complete solar farm installations & clearly it is important to distinguish them. (I've subsequently learnt that generator:place=roof is the established tag).
• Many installations are sufficiently clear on aerial imagery to allow estimation of the number of panel modules involved. Virtually all the ones at Aspley are 2 rows of 5 modules. Unfortunately I don't know the exact module size, but they are probably 10 or 12 by 6 cells. Tagging the module array explicitly is probably better than guesstimating the area (as I have done).
• If module size is known (see top photo) the array area can be calculated directly. Each solar cell is likely to be 156 mm square, so a 10 by 6 array will be 1.56 x 0.96 m (1.46 sq m).
• Cell size, module size and number of modules allow optimal power rating to be estimated. I think these arrays of roughly 15 sq m are around 3500-3700 W.
• Adding compass orientation of the array in degrees, and angle from the horizontal would also be helpful for using the data for estimating likely power output. (Both could be derived from simple 3D building tags, but adding these is much more complex).
• The last few items (no. of modules, module size, area, power rating, orientation & angle) represent data which can be added iteratively.
• It's worthwhile surveying at least some of those added from aerial imagery to capture other information.
• Even if the panel is mapped as an area, most of these tags are still useful as it is unlikely that enough information will be present on the underlying buildings to derive them.
• Surprisingly few public buildings, education establishments or industrial buildings have solar panel installations. I've only noticed a few on buildings of Derby Hall, a hall of residence at the University of Nottingham, and a couple of warehouses in Bulwell.

Other than quickly looking for whether anyone had mapped roof-mounted solar panels in the UK, I haven't looked at activity in other countries. There may be places with a more developed approach to mapping and tagging.

A couple of caveats:

• solar panel distribution is likely to be very patchy; 
• aerial imagery may not be good enough to pick out panels, or recent enough (many of the Nottingham panels have been installed since 2014). 

To help judge what the latter point may mean I provide below a selection of available aerial imagery of various locations in Nottingham, and Basingstoke (Hampshire Council Open Data). The latter includes false colour infra-red (FCIR). 

Aspley Estate (Bing Imagery), roughly here.

Aspley Estate (Bing Imagery) roughly here.

Aspley Estate (Digital Globe Standard Imagery) location as above

Aspley Estate (Digital Globe Premium and ESRI World Imagery) location as above

Broxtowe Lane (Bing Imagery), about here.

Broxtowe Lane (Digital Globe Premium Imagery), same location as above. Obviously newer as a solar panel can be made out on the terrace in the centre. Note on the next terrace down a dark area on the roof. This does not appear to have the same visual appearance as other solar panels, so may be a solar hot water system. 

Broxtowe Lane, as above (Digital Globe Standard imagery)

Deptford Crescent area, Highbury Vale (Bing imagery). Area bottom right is Highbury Hospital.
No solar panels visible

Deptford Crescent (ESRI World Imagery)

Deptford Crescent (Digital Globe Standard Imagery)

Astrid Gardens, Bestwood Estate (ESRI Imagery).
Just occasionally panels have very strong reflections as here.

Astrid Gardens, Bestwood Estate (Digital Globe Standard Imagery). 

Astrid Gardens, Bestwood Estate (Bing Imagery). 

Bilborough Estate (Digital Globe Standard Imagery)
Note the panel lower right which is pretty hard to pick out.

Britten Road, Basingstoke (Hampshire false-colour Infra-red imagery)
This estate has quite a few solar hot water installations but only a few solar PV. The hot water ones are smaller and less obviously modular.

Britten Road, Basingstoke (Hampshire visible spectrum RGB imagery)

So if there are any other takers I suggest this is a suitable topic for a future UK Quarterly Project, perhaps in association with Jez Nicholson's current interest in solar & wind farms.

Lastly I'd like to thank Jack Kelly & Dan Stowell for comments & ideas whilst mapping & writing this up.