Research

How accurate is SolarPlan

We planned 50 real solar installations in 8 countries exactly as a person would in SolarPlan, then compared each plan with what was really built. This page shows what holds up, what does not, and what we changed.

SolarPlan engineering team · study of 29 September 2026 and re-run of 30 September 2026 · Updated 30 September 2026

The short answer

Solid on the basics, rough on anything that is not a simple house or field. Yearly production per installed kilowatt is accurate, the recommended roof face matches the real panels on most pitched roofs, and ground-mounted plants are close. Large flat, complex and industrial roofs are a sketch that can be 1.5 to 2 times off in either direction.

sites where our yearly production per kWp is within ±6% of the European Commission's long-term solar reference (median +3.4%): 48 of 50; sites where our yearly production per kWp is within ±6% of the European Commission's long-term solar reference (median +3.4%)
roofs where our recommended face is the face the real panels are on (10 of 20 before the v2 roof layout): 14 of 20; roofs where our recommended face is the face the real panels are on (10 of 20 before the v2 roof layout)
sites where the real system is within ±30% of our usable maximum (19 before v2): 22 of 50; sites where the real system is within ±30% of our usable maximum (19 before v2)
sites where the real system is bigger than our whole layout (9 before v2): 3 of 50; sites where the real system is bigger than our whole layout (9 before v2)

Method

We picked 50 buildings and plots in 8 countries (Germany 23, United Kingdom 7, France 6, Portugal 5, Italy 4, Netherlands 3, Ukraine 1, Spain 1) whose solar panels are mapped in the open community map: 14 houses, 7 apartment blocks, 10 commercial or industrial buildings, 5 farm buildings, 5 public buildings, 8 ground plants and 1 carport.
37 sites have their real capacity recorded. For the other 13 we computed it from the mapped panel area (0.19 kWp per m²). Every site was checked by eye on aerial imagery.
Each site went through the same planning request a person makes: the roof or field outline drawn on the map, the roof type a person would answer, and a yearly electricity use (houses 4,500 kWh; businesses and plants sized so the roof, not the bill, is the limit).
We compared our maximum layout, our recommended system and our yearly production with the real system and with the European Commission's long-term solar reference for the same place.
After the first run we changed the roof layout (v2) and re-ran the same 50 sites with the same method.

Results by building type, before and after the v2 roof layout

Usable maximum = our largest layout without north-facing roof faces, which an installer would not count. Real installations rarely fill a roof, so a real system below our maximum is expected; a real system above our whole layout means we under-count.

Building type	Sites	Real ÷ our usable maximum (median)	Within ±30%	Real bigger than our layout	Recommended face = real face
Houses	14	0.76 → 0.72	8 → 6	1	6/9 → 7/9
Apartment blocks	7	0.52 → 0.42	1 → 2	1	3/6 → 4/6
Commercial and industrial	10	0.70 → 0.66	3 → 5	3 → 1	0/3 → 2/3
Farm buildings	5	0.87 → 0.73	1 → 3	2 → 0	1/1
Public buildings	5	0.61 → 0.41	0	1 → 0	0/1
Ground plants and carport	9	0.81	6	1 → 0	—
All 50 sites	50	0.72 → 0.65	19 → 22	9 → 3	10/20 → 14/20

What we changed after the first run

Roof type is no longer guessed for large and non-house buildings: the plan asks pitched, flat or sawtooth with three pictures. Roofs wrongly drawn as flat went from 16 to 0.
Flat commercial roofs use east–west rows at about 85% coverage or low-tilt south rows, like real commercial practice, instead of widely spaced racks.
Complex buildings are planned wing by wing instead of the main block only.
Large flat roofs keep a realistic usable share for skylights and plant; carports get continuous canopies over the parking rows.

Honest limits

Recommended system sizes outside Spain are still small (median 0.63 × real) until each country has its own tariff module.
On large pitched and complex buildings our maximum is now generous: real systems use only part of such roofs.
We cannot see skylights or rooftop plant yet: one big box store gets about twice the real capacity.
Small hipped roofs still lose too much to setbacks (one cottage: 1.8 kWp against a real 6 kWp).
Only one site is in Spain: the open map has almost no Spanish rooftop systems with recorded capacity. Spanish houses should follow the house pattern above.
13 real capacities are computed from mapped panel area, which can be imprecise, and the sample leans German.
In the second run the reference service was overloaded, so production figures above come from the first run, where every site was compared.

What the comparison looks like

Left: the real panels as mapped (amber). Right: our first-run plan (solid = recommended system, faint = the rest of the maximum layout). Where v2 changed the result, the caption says so.

Good matches

German house: the real array fills the south face; we fill the same face (11.9 kWp against about 11.6 in v2).
German terraced house: east–west roof, real 6.7 kWp, our 8.4 kWp on the same two faces.
German warehouse: real 30 kWp on the south-west face, which is the face we recommend.
French ground plant: our 2.73 MWp against the real 2.59 MWp.
Dutch ground plant: our 2.11 MWp against the real 2.0 MWp.
UK house: the right face, but the L-shaped hipped roof is over-counted as one large roof.

Where the first run failed

French barn: pitched roof read as flat in v1 (76 kWp against a real 100); with the roof answer in v2, 90 kWp.
German flat apartment block: v1 used only the main block (60 kWp against a real 98); v2 plans both wings (89 kWp).
Dutch retail roof: dense east–west array of about 634 kWp; v1 south racks gave 422 kWp, v2 east–west rows 587 kWp.
German school: real 246 kWp on every wing; v1 used the main block only, v2 plans all wings.
Large store: real 1.21 MWp between skylights; we fill the whole roof (about twice the real figure). Still open.
Small UK hipped cottage: 3 panels per face in v1 (2.6 kWp) against a real 6 kWp. Still under in v2.

Download the results

All 50 sites, anonymised (building type, country and figures only; no map ids, names or locations), as CSV. Derived from open community map data, so the table is published under the Open Database Licence (ODbL 1.0).

Download CSV

Who did this

The SolarPlan engineering team ran the study on 29–30 September 2026 with the production engine, unchanged except for the improvements listed above. The full write-up, with every site, is kept with the product's research notes.

Sources and licences

OpenStreetMap contributors — solar generators, plants and building outlines · ODbL 1.0
Aerial imagery in the comparison images: Esri, Maxar, Earthstar Geographics · Imagery © Esri, Maxar, Earthstar Geographics