A weather forecast that checks itself
Nine weather sources side by side, the middle value with the spread, and every forecast set against what really happened. So you know how far you can rely on the forecast.
A weather app shows a number. We show who agrees with whom.
Behind every forecast is a model, and models differ. FairGrow puts them side by side and takes the middle value, not the average. A source that is far off from the rest then stands out instead of skewing the result.
- The “who says what” table: the daily maximum of every source, side by side.
- For rain, a count, not a percentage. Three out of three dry is different from two out of three.
- Only real answers count. If a source falls back on another one, it does not count.
Where the sources disagree
How far ahead can you rely on it?
FairGrow sets the forecast made 1 to 7 days earlier against what actually happened. A year back, available straight away for every location. You see how fast the error grows, and so up to which day you can plan on it.
- Per source: ECMWF, NOAA GFS and DWD ICON, each measured separately. A blend of models would distort the curve.
- With a zero point. We also measure the difference on day 0, so you see the real forecast error and not a fixed difference between two models.
- Dry or wet, too: how often was the answer to “will it stay dry” right?
- Our own measurement alongside. Since 7 October 2026 we store the forecasts ourselves every day and check them afterwards, as a second check.
Error per day ahead
Looking further, without false certainty
After a week, a single forecast becomes guesswork. That is why FairGrow calculates days 7 to 14 from three ensembles: ECMWF with 51 members, NOAA GFS with 31 and the Canadian GEM with 21. Each model carries the same weight, however many members it has.
- The band gets wider the further ahead you look. That is how it should be.
- The far end drifts towards normal. Behind it are fifteen years of climate for your location, from ERA5.
- We check the band too: does the real weather fall inside it as often as it promises?
Days 7 to 14, deviation from normal
An ordinary weather app, and FairGrow
| An ordinary weather app | FairGrow | |
|---|---|---|
| Sources | One model, often without telling you which | Nine, by name, side by side |
| Certainty | A single number | The middle value and the spread |
| Rain | A percentage | How many sources expect rain |
| Afterwards | Forgotten | Checked a year back, per source and per day ahead |
| Your own plot | The nearest weather station | Your own sensors alongside |
Where to find it
A button at the bottom of the weather card. The middle value of all sources, the band around it and the who-says-what table, up to 14 days ahead.
Forecast against actual over a year, with a picker per model and a table per day ahead.
All sources next to your sensors in the comparison module, forty years of weather history from NASA POWER, and the weather behind your irrigation advice.
Answered briefly
Which nine sources are they?
By default Open-Meteo, NOAA GFS and DWD ICON: three independent weather services. You can also switch on Google, Visual Crossing, Xweather, Meteosource, WeatherAPI and OpenWeatherMap.
What do you check against?
Against ERA5, the reanalysis of the European weather centre: the best possible reconstruction of what really happened. Because ERA5 comes from the same centre as the ECMWF model, we also store the forecasts ourselves and set them against the same reality.
Do I have to wait a year first?
No. The year back is available straight away for every location, because the forecasts from back then have been kept. Our own measurement has been running since 7 October 2026 and grows richer every day.
Does my own sensor count?
Your sensors sit next to the weather sources on the card and in the comparison module. As a second source of truth next to ERA5 in the checking, they are on the way.