Hi all - we’ve hit over 32,000 UK solar PV generator objects in OSM. Woo!
I did a little bit of data wrangling to inspect the data we’ve got.
21,000 are nodes (points).
11,000 are ways (areas/polygons).
A handful of complex multi-polygons too.
Some of them represent individual panels, some are arrays, some are actually entire solar farms as one polygon.
Here are a couple of plots of the areas of the OSM objects. Note that these are not the true surface areas of the panels - they’re the geographic area calculated from each OSM polygon’s shape. A simple histogram of them:
The thick bar on the left is the points (zero area). The thin spike at around 10 sq m is presumably rooftop panels, the bulge at around 200 sq m medium-sized commercial installations etc, and the right-hand bulge around 30,000 sq m is objects that each represent a whole array-within-a-solar-farm or even a whole solar farm.
The drawing of these objects is not always consistent. Some users choose to draw one object for a whole farm, while some go into mega detail with an object for each panel.
One thing I definitely notice: the histogram should have a much bigger bump on the domestic-size panels. Maybe many of them are represented as points rather than areas, though - I’d imagine many of the point items are small domestics, since the larger arrays are easy to draw shapes around.
This plot is the same data but as a scatter plot. I’ve put latitude on the y-axis just to help spread the points out. (The single-point items don’t appear on this plot, only the items with nonzero area.) It also gives you some visual clues about the clumpiness of the data.
In this analysis I haven’t looked at any of the metadata (tagging) people have added. Apart from the larger-scale objects, it’s quite rare for people to add metadata (such as capacity, orientation, tilt…)
I’ve done some further analysis of the OpenStreetMap data. In particular, I wanted to check up on “standalone” solar installations separately from solar-panels-within-solar-farms. (The basic “taginfo” stats don’t always make it clear which is which - you need to do some geo querying.)
As of 11th August we’ve currently got (in OSM UK data):
45,000 “standalone” solar PV generators (we can presume these mostly to be domestic/commercial)
5,500 solar PV generators which are contained within solar farms
490 solar farm objects totalling 3.8 GW (compare this against the REPD which lists approx a thousand farms and 8.1 GW - we’re half way!)
Now here’s a plot which concentrates purely on the standalones. I’m showing the progress so far this year, and I’m dividing them up into 4 size categories just for analysis purposes:
You’ll see a lot of the amazing progress in the past two months is the blue chunk - a massive accumulation of nodes, by which I mean single points as opposed to polygons.
(The other three categories are all polygons, arbitrarily chunked as: red = smaller than 10 sq m; yellow = 10 to 2000 sq m; green = larger than 3000 sq m. Those green ones are likely to be farms rather than single panels!)
A question that occurs to me is - who is adding that amazing number of nodes? The answer is: the OpenStreetMap UK community People are responding to the OSM UK quarterly project on solar power. There are 45 different users listed as contributing to these objects since 1st of July. There’s a long-tail effect though: 99% of these items were edited by the top-10 most active users.
On the topic of mapping individual PV panels: Can anyone think of a use-case for such detailed mapping? All the use cases I can think of only need “bounding box” polygons (where each polygon bounds an entitre PV array). We probably don’t need invididual PV panels to be mapped. If others agree with this, should we communicate to the community that they needn’t spend their valuable time mapping individual PV panels, and that it would be preferable to map lots of PV systems instead of detailed mapping of a small number of PV systems?
(that said, I’m hugely grateful for all the mapping work that’s been done - and those highly detailed maps certainly are still useful - it’s just that I’m not sure those detailed maps are more useful than simple bounding boxes)
One use-case would be to train a machine learning system that works “pixel-wise” on some imagery data - the system could train better since it wouldn’t be exposed to training pixels that were labelled as “panel” but were actually a gap between panels.
However I think that’s a minor concern.
I did consider saying to people “we don’t need the detail inside the solar farms, thanks” - but I don’t think it would make a big difference. Some people have cartographic motivations that are beyond the motivations of OpenClimateFix. And it seems clear that this is not having any large negative impact on the number of small installations captured.
OK, cool beans. Also, I guess that OSM’s tag is called “solar panels” not “solar systems”, so it’s entirely understandable that some people would think that they should map individual panels! (BTW, mapping PV panels is addictive - I’ve just added a few more in London!)
solar_panels is a little unfortunate as a tag because it is used for anything from a rooftop with 2 modules to a 5 MW system consisting of many aligned arrays of mounted modules. However it’s not confusing semantically (yet).
In terms of detail of how much gets mapped, I would favour the following:
Within a solar farm major groups of panels/arrays separated by, for instance, access tracks, ancillary power gear, footpaths, hedgerows etc. It is not uncommon for field boundaries to be retained. Mapping individual arrays is not detail which is really needed: it might be useful if we didn’t have external estimates of power output. We don’t AFAIK have solar farms consisting of heliostats in the UK, but they may need to be considered differently. Obviously power=plant does most of the work here.
larger (commercial) rooftop systems on warehouses, schools, supermarkets, etc. Map as areas. Module number can be estimated/counted as a proxy for output.
domestic rooftop systems as points. This is certainly how Gregory Williams & I are adding most systems right now. Many don’t (as yet) have the underlying building mapped on OSM. This may become significant if anyone wants to use building type either to find existing installations or identify new sites (as per OpenSolarMap). Recent work announced by MS & OSGB about using ML for building classification might be relevant.
I’d be interested what kind of information or coverage might be most useful for using manually identified rooftop solar as training data for ML identification. My current thought is that we should try and scour a few rural LAs (currently I have Rutland & Anglesey, but Isle of Wight fits too) as these seem to be hardest to find, although this might be imagery age as much as anything else.
Hey everyone, first time posting and looking forward to help where I can.
I was thinking of producing a GIF showing the growth in labelled UK solar and posting it alongside a mini video showing how to tag a panel with OSM. Does anyone know of a method to determine when a tag was made?
To determine when a tag was made… well, to determine when an object was last edited is pretty easy, since they all have timestamps on them, when you download them from the OSM API, or when you look at them on the website. (Here’s a simple example - see the “timestamp”.) However, that doesn’t quite tell you when an object was first created (to do that you’d need to query the object’s history from the API), and it’s a bit tricky to be sure when each tag on an object was added. I would suggest you use the timestamp (i.e. last-edited date) of each object, even though that’s not exactly what you’re wanting!
First iteration of the tagged PV growth GIF. Next steps are recording a video of the tagging process and then dissemination.Please let me know if you have any ideas as to how to improve this.
Regarding the PV panels which are tagged as points rather than areas, would there be any use for a classifier which estimates panel size so that retagging can be prioritised? Using distance to nearest DSO & TSO lines, or even roads through OSM, may go some way to helping classify them.
Points-vs-areas - that’s a really nice idea, to use known geographic features nearby to help guess the nature of items and which ones are in need of retagging. I like it!
I have planned, at some later time, to work through all point (node) solar installations which are within school and hospital grounds & those on buildings over a certain size. This ought to catch most of the bigger installations, provided buildings etc are mapped. There is also DECC data which I think was excluded on power output grounds (<1 MW) from the original data posted to the OSM Wiki, and it may be time to pull that in.
People are responding to 
