Forecast Methodology
Meteomat shows probabilistic weather forecasts anywhere in the world. Instead of a single number, it shows a range of likely outcomes so you can see not just what is most likely, but how confident the forecast is.
It is powered by ECMWF, the European Centre for Medium-Range Weather Forecasts, which produces what is widely considered the world's best operational weather model.
How to read the charts
Each chart shows a central line (the best estimate) surrounded by a shaded band. The band spans the 10th to 90th percentile of the forecast ensemble — meaning there is an 80% chance the true value falls within it. A narrow band means a confident forecast; a wide band means high uncertainty.
- Rainfall uses a square-root y-axis so that small amounts (drizzle) remain visible without large amounts dominating the chart.
- Wind and wave direction is shown as arrows anchored to the top of each panel.
- All panels share the same time axis — zoom or pan one and the rest follow.
Data Sources
Weather forecasts combine two ECMWF land products, while a third is used for the wave and sea surface temperature data. They are both accessed via the Open-Meteo API at native resolution:
| Product | Resolution | Members | Horizon |
|---|---|---|---|
| ECMWF IFS Ensemble (EPS) | ~18 km | 51 perturbed runs | 7 days |
| ECMWF IFS HRES | ~9 km | 1 deterministic run | 7 days |
| ECMWF Wave Model | ~28 km | — | 7 days |
Both IFS products share the same model physics and initial conditions. The ensemble samples uncertainty by perturbing the initial atmospheric state 51 times; HRES is the single highest-resolution deterministic run from the same system. ECMF Wave model is only shown for coastal locations.
Weather Variables
Uncertainty bands (10th–90th percentile)
For each variable, the shaded area represents the range between the 10th and 90th percentile across the 51 ensemble members. This means there is an 80% probability that the true value falls within the shaded band.
Temperature
- Best estimate (line): ECMWF HRES value, shifted from the ensemble median.
- Uncertainty bands: Ensemble 10th/90th percentiles, shifted by the same offset as the median, so the best estimate always sits at the same relative position within the band.
Wind speed & gusts
- Best estimate (line): ECMWF HRES value, shifted from the ensemble median.
- Uncertainty bands: Ensemble 10th/90th percentiles, shifted by the same offset.
- Gusts: Ensemble 90th percentile shifted by the same HRES offset, representing the upper range of expected wind speed.
Humidity
- Best estimate (line): ECMWF HRES value, shifted from the ensemble median.
- Uncertainty bands: Ensemble 10th/90th percentiles, shifted by the same offset.
Rainfall
Rainfall is treated differently from other variables because its distribution is non-Gaussian (many ensemble members predict zero), and because HRES has a meaningful resolution advantage for detecting localised precipitation features (orographic, coastal, convective).
Best estimate (line)
An asymmetric blend of the ensemble median and HRES, depending on whether HRES detects rain:
- When HRES detects rain (> 0.1 mm/h): 50% ensemble median + 50% HRES.
HRES gets equal weight because its higher resolution makes it more likely to be resolving a real precipitation feature. - When HRES detects no rain (≤ 0.1 mm/h): 75% ensemble median + 25% HRES.
The ensemble leads, but HRES pulls the estimate down — a displacement effect (where the ensemble places rain just over the location) is common, so no-rain from HRES is informative but not conclusive.
Lower uncertainty band (10th percentile)
Taken directly from the ensemble, unchanged. This correctly reaches zero when a significant fraction of ensemble members predict no rain, giving an honest picture of the chance of dry conditions.
Upper uncertainty band (90th percentile)
Taken from the ensemble, but raised when HRES predicts more rain than the ensemble median. The rationale: HRES's resolution advantage primarily helps detect precipitation-producing features that coarser ensemble members miss, so it should be able to raise the ceiling of possible rainfall. The upper band is never lowered by HRES.
Probability of rain
The probability that rainfall exceeds 0.1 mm/h, blended using the same asymmetric weights as the best estimate:
- When HRES detects rain: 50% × ensemble probability + 50% × 100%
- When HRES detects no rain: 75% × ensemble probability + 25% × 0%
This keeps the displayed probability internally consistent with the displayed amount.
Rainfall amounts are rounded to the nearest 0.1 mm, which reflects the true resolution of the forecast.
Marine Variables
Wave height, period, and direction
Sourced directly from the Open-Meteo Marine API. Total wave height is the combination of swell and wind-wave components. No post-processing is applied.
Sea surface temperature (SST)
Sourced from the Open-Meteo Marine API. A 3-hour centred rolling average is applied for visual smoothness. Note that SST reflects the model's boundary condition (typically derived from satellite composites) and may not capture short-lived phenomena such as coastal upwelling, which can cause real surface temperatures to differ by 1–3°C from the forecast.
Limitations
- Spatial resolution: HRES at ~9 km and the ensemble at ~18 km cannot resolve sub-grid features such as valley-scale orography, urban heat islands, or small coastal inlets. Two nearby locations within the same grid cell will receive nearly identical forecasts.
- Ensemble calibration: The ECMWF ensemble is one of the best-calibrated probabilistic forecast systems in the world, but calibration is not perfect. Stated probabilities are skill estimates, not guarantees.
- Rainfall uncertainty: Precipitation is the hardest variable to forecast deterministically. The blending methodology is a pragmatic combination of two imperfect sources and should be interpreted with appropriate caution.
- SST and upwelling: The marine SST forecast does not account for wind-driven upwelling events, which are common along the Cantabrian coast of Spain and can cause surface temperatures significantly colder than the model suggests.