How accurate is Azima — 1,000 windows recalculated
Measured at 1,000 windows in all 23 districts, against the City of Vienna’s real roofs and the 2023 laser scan: how far the sun figure is off, where it scatters and what the check does not show.
A sun figure is only worth as much as its check — and a single window proves nothing about a city. So on 16 and 17 October 2026 the calculation was measured at 1,000 randomly drawn windows in all 23 districts: against an independent calculation with the real roof and wall surfaces from the City of Vienna’s 3D model, minute by minute, on three reference days.
How it was measured
The 1,000 windows are drawn at random, reproducibly, and stratified: every district equally often, within it evenly across four height classes of houses and three floor levels from the ground floor to below the roof. Every window sits on a real outside wall. The reference reads the City of Vienna’s official roof and wall surfaces within 300 m — courtyards as holes, not roofs —, the buildings beyond up to 2 km, and computes the sun with the NOAA formulas including refraction, one ray every minute. What is measured: how many minutes of direct sun Azima’s calculation returns, and how many the reference does.
The result at the window
| Day | Mean | ±30 min | ±1 h | 9 in 10 |
|---|---|---|---|---|
| 21 Mar | −2.2 min | 85.5 % | 95.8 % | ±38 min |
| 21 Jun | −3.2 min | 80.2 % | 94.5 % | ±47 min |
| 21 Dec | −0.2 min | 88.2 % | 96.1 % | ±33 min |
Mean = deviation of the sun minutes against the reference · ±30 min / ±1 h = share of windows within · 9 in 10 = the furthest nine in ten windows are off.
On average the calculation sits a few minutes below the reference — rather slightly too cautious than too generous. On the shortest day, nine in ten windows are within a good half hour.
Without a floor plan: why a range is shown
Entering only an address gives the calculation a point inside the house — it then does not know which wall the window is on. Recreated with a point 2 m behind each façade, compared with the real roofs at the window:
| Day | ±1 h | 9 in 10 |
|---|---|---|
| 21 Mar | 81.9 % | ±95 min |
| 21 Jun | 77.2 % | ±115 min |
| 21 Dec | 89.2 % | ±65 min |
That is why every number on the Azima page and in the report carries its range. With a floor plan and the window direction the calculation knows which wall the light arrives at — then the values of the first table apply.
The error the check found
The first measurement on 17 October showed: until then Azima computed on average 52 to 81 minutes too much sun. The cause was in the data, not the calculation — for buildings made of several parts the data held only a sliver of the first part, which concerns just over half of Vienna’s houses. Fixed the same day and re-measured at the same 1,000 windows; the table above is the state after that. Every further correction is measured at the same windows before it goes live, and its result appears on this page with a date — a worse one too.
Building heights against the laser scan
The heights of the neighbouring houses come from the City of Vienna’s roof model (2016). Checked against the 2023 laser scan of the BEV at 256 buildings from all parts of the city, part by part (478 parts): the eaves lie 0.01 m from the measured roof surface in the median, 0.26 m above it on flat roofs, 1.86 m below it on pitched roofs, because the ridge sits above. The calculation does not know this part of the roof; on average a cautious rule at the building edges cancels it out, which is also measured at the 1,000 windows. One part in ten deviates by more than 5 m: glass roofs over courtyards, overbuilds, errors in the model.
What moves the number most
One metre more or less on every neighbouring house shifts the sun time by at most 4 minutes for half of the windows, by about 20 for one in ten. One floor higher or lower shifts it many times more — by over an hour for one window in ten. Stating the floor matters more than any inaccuracy in the height model.
What this check does not show
- Window reveals, balconies and canopies — they are missing in the calculation and in the reference alike.
- Trees — missing in both. In summer they really do shade; in winter, with bare branches, hardly at all.
- Terrain — this check is on flat ground. Azima itself also adds the shade of hillsides.
- How well the 2016 roof model of the city matches today’s Vienna, apart from the 256 buildings measured against the 2023 laser scan.
- Clouds. Every number is clear sky — the upper limit of what can arrive.
State of the measurement: 17 October 2026, calculation as live. Method, scripts and raw data on request.
Write to me
Interested, a question or a conversation? Pick a topic, adjust the text, send it – straight to me.
The button copies your text and opens my LinkedIn profile in a new tab – tap “Message” there and paste. The email arrives filled in.
No form server, no tracking: the text stays in your browser until you send it yourself on LinkedIn or from your mail app.