GFS (USA Forecast Model)
77%
13 temperature samples · 0 matched rain periods
Temperature only · rainfall history pending · Higher is better
Actual lead: 8.0–10.0 hours
Marathon Park
What are GFS and ECMWF predicting at the time of the latest station reading?
← Live Weather map · Refresh comparisonSelected forecast: Latest available
Uses the forecast available when each sample was recorded. Select an earlier forecast age to evaluate predictions made days in advance.
Forecast Accuracy
ECMWF (European Forecast Model)
80%
13 temperature samples · 0 matched rain periods
Temperature only · rainfall history pending · Higher is better
Actual lead: 14.0–16.0 hours
A gradual agreement score: each temperature sample starts at 100 and loses 10 points per °C of error; each matched rain period loses 10 points per mm of error, with a minimum of zero. Temperature and rain averages carry equal weight when both are available. Wind is excluded. Rain requires matching interval boundaries and excludes counter resets and forecast snow. Until rainfall history is available, only temperature is scored. This is a site-defined comparison scale, not a statistical probability or a standard measure of forecast accuracy.
Observation: Sun 18/10 13:00 · Europe/Athens · 10 minutes old
38.148353° N, 24.009376° E · 10 m altitude. Checked Sun 18/10 13:10.
Observed
20.2 °C
Wind: E · 5 km/h
Humidity: 54 %
Pressure: 1020 hPa
GFS (USA Forecast Model)
17.8 °C
-2.4 °C versus observed
Wind: 2.2 km/h
Run: Sun 18/10 03:00
Valid time: Sun 18/10 13:00
Actual lead: 10.0 hours
Matched from Sun 18/10 12:00 to Sun 18/10 15:00
ECMWF (European Forecast Model)
18.7 °C
-1.5 °C versus observed
Wind: 2.2 km/h
Run: Sat 17/10 21:00
Valid time: Sun 18/10 13:00
Actual lead: 16.0 hours
Matched from Sun 18/10 12:00 to Sun 18/10 15:00
Rain at the station & in the models
Station rain rate: 0 mm/h · Station rain today: 0 mm
GFS: 0.0 mm total precipitation for Sun 18/10 12:00–Sun 18/10 15:00.
ECMWF: 0.0 mm total precipitation for Sun 18/10 12:00–Sun 18/10 15:00.
Model precipitation includes rain and snow water equivalent over the whole interval, which may extend into the future. It cannot be subtracted from an instantaneous rain rate or a partial-day station total.
FORECAST ONLY
Fresh snowfall
Model estimate for the forecast interval covering the observation time.
GFS
0.0 cm
No fresh snow forecast in this interval
Sun 18/10 12:00–Sun 18/10 15:00ECMWF
0.0 cm
No fresh snow forecast in this interval
Sun 18/10 12:00–Sun 18/10 15:00Snowfall is not measured by this station. These are the site's existing fresh-snow estimates, not measured snowfall or snow depth. The interval may extend into the future; no snowfall error or verification score is calculated.
| Difference | GFS | ECMWF |
|---|---|---|
| Temperature (°C) | -2.4 | -1.5 |
| Wind speed (km/h) | -2.8 | -2.8 |
This snapshot: the models differ by 0.8 °C. ECMWF is closer to the observed temperature. This is one comparison, not an overall model ranking.
The last 24 hours
A snapshot every ten minutes. The graphs show the last 24 hours; older records are retained for future analysis. Gaps longer than 30 minutes are left disconnected.
● Observed● GFS● ECMWF
13 recorded samples · Latest observation Sun 18/10 13:00
Each sample preserves the models available when it was collected; later runs do not rewrite older comparisons. Scores and model lines follow the selected forecast age. Days ahead are measured from model initialization to observation time; the actual lead is shown on each card. Snowfall is not measured by the station, so there is no observed snowfall comparison graph.
How to interpret this pilot
Temperature and wind speed use the site's existing coordinate interpolation and altitude adjustment. We linearly interpolate their forecast values between the steps surrounding the observation time; these are estimates between model steps, not new observations. Positive differences mean the model is warmer or windier.
Station wind exposure and averaging periods may differ from the models. Spatial resolution, terrain and sensor quality can also affect agreement. Latest available uses the run saved with each observation. Earlier lead times use immutable archived runs saved before the observation, with the same coordinates and altitude. Archive collection starts now; earlier missing forecasts are not reconstructed.
One matched sample cannot establish correlation, bias or forecast skill. Those require repeated observations paired with archived forecasts and defined lead times. No automatic forecast correction is applied.
Only fresh Marathon Park readings are used. Missing or suspect measurements remain unavailable. Results are cached for two minutes; use Refresh comparison for the latest snapshot.