Every day, a post-processing AI is compared with the official physical forecast on a panel of French wave buoys, tide gauges and wind stations. Results are published as they are, including where the AI loses.
Gap to the physical forecast
Forecast issued —
Scoreboard — Figures not displayed
Measurements are read when the page opens, from the pipeline’s public repository; they are never frozen into this page. They could not be obtained here — network unavailable, file not matching the contract, or page read without JavaScript. Nothing is shown rather than a doubtful figure.
Part of the displayed gap is only a correction of the baseline’s constant bias, and the two tide gauges are still trained on reanalysis wind although they serve forecasts: their figures are therefore rather optimistic. The limits are detailed on the method page. The 48 h exceedance probabilities of all stations are gathered on the alerts page.
Reading guide
What this table measures
Three clarifications without which the figures above mean nothing.
Comparison with the official forecast
The reference is not a convenient baseline: it is the physical forecast served operationally. For waves as for wind, each station is compared with the model that forecasts it best — Météo-France MFWAM, DWD EWAM or NCEP GFS-Wave for waves, Météo-France ARPEGE, ECMWF IFS or DWD ICON-EU for wind, depending on the site and named in the table. For water level, the harmonic tidal analysis.
Stations below the threshold stay displayed
A station where the AI does not beat the physics stays tracked and listed, without an AI figure. Removing it from the table would make the others unverifiable. The table above names the ones currently in this case.
Measured every day, never embellished retroactively
Each forecast is compared with the observations that arrive afterwards. Days without usable observations stay visible as such, and enter no average.
