MetOry R&D
How MetOry reads the weather, explained simply, and the methods behind it.
Insights
Open a card to read it.
Photo: Pixabay / Pexels
MetOry reads the forecast in windows: the next 1, 3, 6, 12, 24, 48 and 72 hours. Each window shows the strongest value expected inside it, such as the highest wind or the heaviest rain.
A window never says that something will happen at a set hour. It tells you how far ahead you are looking.
The index places each hazard on one scale with four levels: low, watch, high and critical. It lets you compare places and days at a glance.
87 out of 100 means the forecast sits high on that scale, not that an event has an 87% chance. When data is missing we show it as unknown, never as safe.
Warnings from services such as AEMET in Spain or Météo-France are shown with the issuing authority’s own wording.
A model forecast never becomes an official warning. When a warning feed is not connected, we say so instead of showing “no warnings”.
Global models are run by centres such as ECMWF, NOAA, DWD and Météo-France. Services like Open-Meteo pass their output on: they are channels, not separate models.
When models agree, that is useful context, but agreement alone does not prove accuracy. Accuracy is measured against real observations.
Forecasts are saved before the weather happens and are never edited afterwards. Later, they are paired with station observations at the same hour.
Current work compares temperature forecasts from ECMWF, NOAA and DWD with station observations in Spain and France.
Each marina reads the nearest sea point of the model, up to 30 km away, and the distance is always shown. Inside harbours and bays the sea is usually calmer.
Rivers, canals, lakes and enclosed harbours are too small for the model, so they get no wave value rather than a guess. Official marine warnings come from the national authority.
Research
Technical work on how we verify forecasts and assess hazards.
Pairs ECMWF, NOAA and DWD 2 m temperature forecasts with AEMET and Météo-France station observations at the exact valid hour.
Bias, mean absolute error and root mean square error, computed per model, lead time and country.
How station data is checked and flagged before it is used.
How wave height, period and direction are read from the ECMWF wave model at the nearest sea point.
Sustained wind and gusts from three global models, assessed separately and never derived from each other.
Rainfall totals over 1 to 24 hour windows, counted only when the whole window is covered.
Universities and research groups interested in collaboration can write to info@metory.net.
