Forecast accuracy
How accurate is the nano forecast?
nano is a weather API for Germany with its own forecast model, built on top of the DWD’s weather models. To show how good the forecast is, we check it every hour: at 100 places we compare what nano predicted with what the nearest weather station of the Deutscher Wetterdienst (DWD), Germany’s national weather service, actually measured. We check DWD forecast (MOSMIX) and MET Norway the same way, so you can compare.
- places across Germany
- 100
- forecast hours checked since 30 September
- 382k
- DWD weather stations provide the measurements
- 76
- new results every hour, last at 5 Oct, 14:00 CEST
- Hourly
Forecasts for the next 3 hours
nano gets the temperature right to within 1 °C in 67 % of hours, DWD forecast (MOSMIX) in 62 % and MET Norway in 52 %. In hours where the two were more than 1 °C apart, nano was closer 7 times out of 10.
Results for longer lead times and for sun, rain and wind follow below.
Our forecast model
nano’s forecast comes from our own model
It is built on top of the DWD’s weather models: we don’t simulate the weather ourselves. We combine the DWD’s forecasts, adapt them to every place and correct them with live measurements. We check the result every hour and keep improving the model.
1What goes in
Two forecasts of the Deutscher Wetterdienst: its high-resolution weather model ICON-D2 (a grid of about 2 km) and its station forecasts (MOSMIX). We combine them and calculate the forecast for every place in Germany, corrected for its altitude.
2Corrected on every request
For the next hours, each request adjusts the forecast with the latest DWD radar images for rain and the latest station measurements for temperature and clouds.
3Tested before every change
Right now 7 versions of the model run side by side. Only one of them delivers the forecast you receive. A new version runs next to it for at least three days, at the same places and hours, and replaces it only if it measures better. This is how the forecast improves over time.
Temperature · 1–3 hours ahead
Is the temperature right?
nano bestAt this lead time nano is more accurate than DWD forecast (MOSMIX) and MET Norway.
72 of 100
times nano was closer to the measurement than DWD forecast (MOSMIX).
nano closer: 2,305 hoursDWD forecast (MOSMIX) closer: 885 hours
Counted are hours in which the two forecasts were more than 1 °C apart. The measured temperature shows which one was closer. Data since 30 September, forecasts made 1–3 hours ahead. 3,190 such hours in total; the dots show them per 100.
All hours
Hours right to within 1 °C — higher is better
- nano67 %
- DWD forecast (MOSMIX)62 %
- MET Norway52 %
The verdict compares nano with each service only on the hours for which both had a forecast.
Sun & clouds · 1–3 hours ahead
Sun or clouds?
Not ahead yetMET Norway is more accurate at this lead time.
82 of 100
times nano was right and DWD forecast (MOSMIX) wrong.
nano right: 2,552 hoursDWD forecast (MOSMIX) right: 557 hours
Counted are daytime hours in which one forecast said sunny and the other cloudy. The measured sunshine shows which one was right. Data since 30 September, forecasts made 1–3 hours ahead. 3,109 such hours in total; the dots show them per 100.
All hours
Daytime hours with the sky (sunny, mixed or cloudy) right — higher is better
- MET Norway63 %
- nano51 %
- DWD forecast (MOSMIX)35 %
The verdict compares nano with each service only on the hours for which both had a forecast.
Rain · 1–3 hours ahead
Will it rain?
Catches the most rainnano forecasts more of the rainy hours than DWD forecast (MOSMIX) and MET Norway. It also says “rain” slightly more often when it then stays dry than DWD forecast (MOSMIX) and MET Norway.
47 of 100
times nano was right and DWD forecast (MOSMIX) wrong.
nano right: 285 hoursDWD forecast (MOSMIX) right: 322 hours
Counted are hours in which one forecast said rain and the other dry. The station’s rain gauge shows which one was right. Data since 30 September, forecasts made 1–3 hours ahead. 607 such hours in total; the dots show them per 100.
All hours
Rainy hours that were forecast — higher is better
- nano94 %
- DWD forecast (MOSMIX)91 %
- MET Norway82 %
The verdict compares nano with each service only on the hours for which both had a forecast.
False alarms — share of “rain” forecasts that stayed dry; lower is better
- DWD forecast (MOSMIX)52 %
- nano52 %
- MET Norway58 %
Wind · 1–3 hours ahead
How windy will it be?
Slightly behindOn average nano and DWD forecast (MOSMIX) are about equal. In the hours where the two differed clearly, DWD forecast (MOSMIX) was closer more often.
5 of 43
times nano was closer to the measurement than DWD forecast (MOSMIX).
nano closer: 5 hoursDWD forecast (MOSMIX) closer: 38 hours
Counted are hours in which the two forecasts were more than 5 km/h apart. The measured wind shows which one was closer. Data since 30 September, forecasts made 1–3 hours ahead. Only 43 such hours so far, too few for a firm conclusion.
All hours
Hours right to within 5 km/h — higher is better
- DWD forecast (MOSMIX)89 %
- nano89 %
- MET Norway82 %
The verdict compares nano with each service only on the hours for which both had a forecast.
How we test
How we test the forecast
- 1
We save every forecast
Every hour we store what nano and DWD forecast (MOSMIX) and MET Norway predicted for 100 places across Germany.
- 2
Stations measure what happened
Official weather stations of the Deutscher Wetterdienst measure the real weather near each place.
- 3
We compare
Once the hour is over, we compare each forecast with the measurement. A comparison between two services only uses hours for which both had a forecast.
Scientific background
The methods behind the model
nano combines established methods of weather forecasting. These are the six that matter most for the forecast you receive.
- Numerical weather predictionA physics-based weather model
- ICON-D2 is the DWD’s weather model for Germany. It calculates how the atmosphere develops on a grid of about 2 km and is run every 3 hours. It gives our forecast the small-scale differences between nearby places.
- Model Output Statistics (MOS)Statistically corrected station forecasts
- MOSMIX is the DWD’s forecast for fixed station points. It corrects weather model output with statistics learned from past forecasts and measurements. It is the base of our forecast.
- Inverse distance weighting, lapse rateFrom stations to every place
- For a place between stations we combine the eight nearest station forecasts, weighted by distance, and correct the temperature for altitude by 0.65 °C per 100 m.
- Radar nowcastingRain for the next two hours
- The DWD’s radar network measures rain across Germany. Its RV product extrapolates the measured rain two hours ahead and is updated every 5 minutes. We use it for rain in the next hours.
- NudgingCorrection with the latest measurements
- If the latest station measurement differs from the forecast, we shift the coming hours towards it. The correction fades out over a few hours, faster by day than at night.
- Forecast verificationStandard accuracy measures
- We score forecasts with the measures used in meteorology: mean absolute error, bias, and for rain the detection rate and the false alarm ratio. Two services are only compared on hours both forecast.
Sources. Station measurements and the DWD forecast (MOSMIX): Deutscher Wetterdienst, licensed CC BY 4.0. All scores are calculated by nano.
Details for expertsAverage errors, sample sizes, daily values and the exact method
Error by lead time
Mean absolute error and bias (forecast − observed) per provider, on the hours both nano and that provider answered. Rain: detection rate (POD), false alarm ratio (FAR) and missed rainy hours; wet = ≥ 0.1 mm. Sunshine: share of the hour vs 1 − cloud cover, sun ≥ 15° above the horizon. n = forecast/observation pairs.
Temperature
| Lead | Provider | MAE | Bias | n |
|---|---|---|---|---|
| 1–3 h | nano | 0.88 °C | -0.03 °C | 29,534 |
| DWD forecast (MOSMIX) | 1.00 °C | -0.16 °C | 29,489 | |
| MET Norway | 1.21 °C | +0.26 °C | 2,964 | |
| 4–6 h | nano | 1.03 °C | -0.02 °C | 28,776 |
| DWD forecast (MOSMIX) | 1.05 °C | -0.20 °C | 28,731 | |
| MET Norway | 1.15 °C | +0.30 °C | 2,280 | |
| 7–24 h | nano | 1.10 °C | +0.03 °C | 158,047 |
| DWD forecast (MOSMIX) | 1.10 °C | -0.18 °C | 157,777 | |
| MET Norway | 1.05 °C | -0.03 °C | 2,736 | |
| 25–48 h | nano | 1.20 °C | +0.13 °C | 166,065 |
| DWD forecast (MOSMIX) | 1.19 °C | -0.03 °C | 165,720 |
Sun vs cloud
| Lead | Provider | MAE | Bias | n |
|---|---|---|---|---|
| 1–3 h | nano | 27 pp | -17 pp | 6,010 |
| DWD forecast (MOSMIX) | 37 pp | -32 pp | 6,010 | |
| MET Norway | 23 pp | -16 pp | 693 | |
| 4–6 h | nano | 27 pp | -18 pp | 6,240 |
| DWD forecast (MOSMIX) | 37 pp | -32 pp | 6,240 | |
| MET Norway | 24 pp | -17 pp | 693 | |
| 7–24 h | nano | 28 pp | -17 pp | 31,912 |
| DWD forecast (MOSMIX) | 38 pp | -32 pp | 31,824 | |
| MET Norway | 28 pp | -23 pp | 1,168 | |
| 25–48 h | nano | 35 pp | -23 pp | 33,958 |
| DWD forecast (MOSMIX) | 44 pp | -37 pp | 33,870 |
Rain yes/no
| Lead | Provider | POD | FAR | Missed | n |
|---|---|---|---|---|---|
| 1–3 h | nano | 94 % | 52 % | 98 | 29,223 |
| DWD forecast (MOSMIX) | 91 % | 52 % | 153 | 29,178 | |
| MET Norway | 82 % | 58 % | 17 | 2,811 | |
| 4–6 h | nano | 94 % | 52 % | 113 | 28,539 |
| DWD forecast (MOSMIX) | 90 % | 52 % | 172 | 28,494 | |
| MET Norway | 81 % | 62 % | 10 | 2,127 | |
| 7–24 h | nano | 91 % | 53 % | 919 | 156,993 |
| DWD forecast (MOSMIX) | 88 % | 53 % | 1,225 | 156,723 | |
| MET Norway | 80 % | 90 % | 3 | 2,333 | |
| 25–48 h | nano | 86 % | 56 % | 1,459 | 164,849 |
| DWD forecast (MOSMIX) | 83 % | 55 % | 1,681 | 164,504 |
Wind speed
| Lead | Provider | MAE | Bias | n |
|---|---|---|---|---|
| 1–3 h | nano | 2.4 km/h | -0.2 km/h | 21,159 |
| DWD forecast (MOSMIX) | 2.4 km/h | -0.1 km/h | 21,129 | |
| MET Norway | 2.9 km/h | +0.2 km/h | 2,142 | |
| 4–6 h | nano | 2.5 km/h | -0.2 km/h | 20,664 |
| DWD forecast (MOSMIX) | 2.5 km/h | -0.1 km/h | 20,634 | |
| MET Norway | 2.9 km/h | -0.1 km/h | 1,647 | |
| 7–24 h | nano | 2.5 km/h | -0.2 km/h | 113,674 |
| DWD forecast (MOSMIX) | 2.5 km/h | -0.1 km/h | 113,494 | |
| MET Norway | 2.7 km/h | -0.5 km/h | 1,972 | |
| 25–48 h | nano | 2.4 km/h | -0.2 km/h | 119,695 |
| DWD forecast (MOSMIX) | 2.5 km/h | -0.1 km/h | 119,465 |
Day by day (1–3 h ahead)
nano across all DWD stations
nano alone against all DWD stations over the last 48 hours (mean absolute error) — the larger sample, without a comparison.
| Variable | 0–2 h | 3–6 h | 7–12 h | 13–24 h | 25–48 h | 49 h + |
|---|---|---|---|---|---|---|
| Temperature | 0.92 °Cn 45,791 | 1.00 °Cn 45,791 | 1.03 °Cn 68,688 | 1.08 °Cn 137,368 | 1.29 °Cn 271,978 | 1.57 °Cn 268,372 |
| Wind speed | 1.8 km/hn 20,653 | 1.8 km/hn 20,643 | 1.8 km/hn 30,975 | 1.9 km/hn 61,941 | 2.0 km/hn 122,646 | 2.2 km/hn 121,017 |
| Wind gusts | 2.5 km/hn 20,647 | 2.6 km/hn 20,639 | 2.6 km/hn 30,966 | 2.7 km/hn 61,923 | 2.9 km/hn 122,607 | 3.2 km/hn 120,983 |
Method
- Truth: the DWD station nearest each place (within 12 km and 150 m height difference; 76 stations). An hour counts once it is complete.
- nano = exactly what
/v2/weatherserved at that moment (stable product). Other providers: their forecast for the same place, fetched at the same time. “DWD forecast (MOSMIX)” is the raw DWD MOSMIX point forecast for the MOSMIX station nearest each place, retrieved via Bright Sky (which passes DWD open data on unchanged). It is not the DWD’s own refined app forecast. - Per place, hour and lead time only the newest forecast counts. Scores use only hours that both nano and the other provider answered.
- Clear disagreements: forecasts more than 1 °C, 5 km/h or 15 % sunshine apart, or wet vs dry — the station decides. Counted per lead time, like every other number on this page.
- Computed 5 Oct, 15:35 CEST; data until 5 Oct, 14:00 CEST.