Skip to content

Judge large installations on your PC

If you run EarthPV yourself, building an atlas on your own PC or keeping one up to date, every country's run makes two review pages: one picture of each large outline the model drew where OpenStreetMap has no solar, and a few keys to say what it is. A verdict takes a few seconds, needs no OpenStreetMap account, and changes the country's capacity directly. It is the quickest way to make an atlas more right where it matters most, at its largest installations.

page what it shows what your verdict changes
Solar farm review outlines above 100,000 m2 (10 ha) nothing this size counts until a person judges it solar, so each park you confirm is added whole
Large installation review outlines from 10,000 m2 (1 ha) up to 10 ha, on roofs and on the ground these already count, each at the share its size class usually turns out real; your verdict replaces that share for this outline: solar counts it as real (on a roof, ½.5 of the outline, since the model draws a roof array about 2.5 times its size), not solar counts it as nothing

Both are drawn by default after a country's export step, largest first, and only from outlines that are still leads: no mapped solar within 100 m, inside the atlas's regions, and past the screens that take out steep slopes, open ground of low confidence and, where a country switches it on, water.

Open a page

data/oversize_review/index.html lists every country's solar farm page and data/large_review/index.html every country's large installation page, with how many outlines are left. Open one in a browser. Nothing is uploaded: the page and its pictures sit on your disk, and your progress is kept in the browser until you export it.

If a country has no page, draw it:

pixi run blob-review draw --aoi YOUR_AOI      # the solar farm review
pixi run large-review draw --aoi YOUR_AOI     # the large installation review

Drawing a page again keeps every verdict already given.

How to judge

Each outline is one picture of four panels, with the model's outline in yellow:

  • Sentinel-2 as the model saw it, in true colour, around the whole outline. This is the right date: it shows installations built after the high-resolution imagery was taken.
  • The same at its native 10 m pixels, close in.
  • A false colour (shortwave infrared, near infrared, red). Panels read dark purple; vegetation bright green; water dark in every panel.
  • Esri World Imagery with its capture date, for texture: rows of panels, a roof's edge.

Press 1 for PV, 2 for Not PV, 3 for unsure, 4 for bright roof, ← → to step, n for a note.

  • PV: panels inside or touching the outline. Judge whether solar is there, not how well the outline fits it: the model's outline is often larger than the array.
  • Not PV: something else, and the imagery is recent enough to show panels if they were there. The usual look-alikes are dark or blue metal roofs, wet or tarred roofs, greenhouses and shade netting, solar water heaters, car parks, dark ploughed fields, water and aquaculture ponds, burn scars, rock and shaded slopes.
  • Unsure: the four panels cannot settle it. Each picture links Google, Bing, Esri Wayback and JOSM at that spot; decide there, or leave it unsure. An unsure or skipped outline keeps the price it had and comes back on the next page.
  • Bright roof: not PV, and the outline sits on a very bright roof (white or light metal, a reflective coating). It counts exactly as Not PV, but apply keeps the label apart (flag in the country's review.csv), so these false alarms can be gathered and studied. A bright roof that does carry panels is PV.

A flat grey block in the Sentinel-2 panels is a cell nobody imaged, and a black block that reads dark red in the false colour is a cell that lost bands. Neither is ground: judge those parts on Esri.

Record your verdicts

Stop whenever you like. When you are done for the day, press the Export button at the top of the page and apply the file it saves under your name:

pixi run large-review apply --aoi YOUR_AOI --reviewer "Your name" \
    --verdicts YOUR_AOI_large_verdicts.csv
pixi run blob-review apply --aoi YOUR_AOI --reviewer "Your name" \
    --verdicts YOUR_AOI_blob_verdicts.csv

Any named reviewer's verdict counts. Your name is kept beside every verdict you give, and when a second person judges an outline again, apply reports where the two agree. Then recompute the country. This is much quicker than its first run, because every cell's cached result is reused and no imagery is read again:

pixi run earthpv density --aoi YOUR_AOI --districts
# then the atlas, with the same command that built it

pixi run large-review status and pixi run blob-review status show what is left per country. An outline you judged leaves the page the next time it is packaged.

What happens to an outline you judged

  • Mapped later. Once someone maps it in OpenStreetMap, the map counts it and your verdict rests. Mapping a confirmed outline is still worth it: the map gives its real shape, and the country's next model learns from it (map the leads).
  • A new model run. A new run numbers its outlines afresh, so a review belongs to the run it was drawn from; the next run starts a new page.
  • Not solar. Nothing in OpenStreetMap records a false alarm, so the verdict is the only place that knowledge lives: keep the review files with the run.