Which weather model is best for sailing?

The short answer: there is no model that always wins. Which one comes closest depends on the sailing area, the weather pattern and how far ahead you are looking. If you know the strengths and limits of the models, you will read every forecast better.

Global models: ECMWF, GFS, ICON and AIFS

Global models cover the whole planet and run many days ahead.

• ECMWF IFS from the European Centre for Medium-Range Weather Forecasts is regarded as the benchmark for forecasts several days out. The open data, which Sailseer also uses, has a grid of 0.25 degrees, around 25 km.
• GFS from the US weather agency NOAA is freely available, runs four times a day and goes 16 days ahead. Many apps show it by default.
• ICON from the Deutscher Wetterdienst (DWD), Germany's national weather service, runs globally and, as ICON-EU, at higher resolution over Europe, at around 7 km and five days ahead.
• AIFS is the ECMWF's AI model. It is trained on many decades of weather data and runs much faster than a physics-based model. It does not output gusts, so in Sailseer it only counts towards wind speed and direction.

Regional models: fine, but short-range

Regional models have a finer grid and represent coasts, islands and wind funnelling better. In return they cover only a limited area and run only two to four days ahead.

• ICON-D2 from the DWD covers Germany and its neighbours at around 2 km, about two days ahead.
• Harmonie from the Danish Meteorological Institute (DMI) runs from Ireland across the North Sea and the Baltic to the Gulf of Lion, the Tyrrhenian Sea and the Adriatic, MEPS from MET Norway in Scandinavia, AROME from Météo-France on the Spanish Mediterranean coast and the Balearics, and ICON-2I from Italy around Sicily and in the Ionian Sea.
• AROME Antilles from Météo-France covers the Antilles at 2.5 km, two days ahead.

In the Aegean including Crete, on the Algarve, in the Canaries and the Azores none of these regional models runs, so only the global models remain. Sailseer only counts a regional model where and for as long as it actually runs, and does not quietly fill its gaps with another model.

What a model number means

A model calculates on a grid. The wind in a cell is an average over its whole area, so with 25 km grid spacing over an area in which small islands, headlands and bays disappear. It stands for the mean wind at 10 m height, not the gust that hits your rig. Along a steep coast, in a gap between two islands where the wind funnels, or in the lee of a mountain, the real wind therefore often differs markedly, up or down. A finer grid gets closer to such effects, but never captures them completely.

Gusts

The gust is a separate quantity that the model derives from wind, atmospheric stability and turbulence. When it comes to reefing, it matters more than the mean wind. Sailseer shows it next to every wind value and uses the upper end of the model gusts in its rating. More in the guide Gusts and thunderstorms.

Ensembles: how certain is the forecast?

An ensemble runs the same model many times with slightly altered starting conditions. If the runs stay close together, the situation is stable; if they diverge, it is uncertain. Sailseer uses the 51 members of the ECMWF ensemble and the 20 of ICON-D2 and shows a range and the probability of gusts above 20, 25 and 30 kn.

Why a consensus says more than one model

If you look at only one model, you cannot see whether it is out on its own. That is why Sailseer reads nine models for each waypoint at the expected time of arrival and takes the median for each hour, which a single outlier barely shifts. The models count as agreeing if, outliers aside, their wind speeds are no more than 5 kn apart and their directions differ by less than 45 degrees on average. If they do not agree, the hour is marked as uncertain and the recommendation works with the upper end instead of the median. The individual models remain visible in the “Who says what” table.

When the models disagree

Disagreement is information in itself: the situation is hard to forecast, often because of a low with an uncertain track or a front whose timing varies. Then it helps to:

• plan for the stronger scenario, not the average,
• wait for the next run and see which way the models move,
• know a port of refuge along the route,
• read the official marine forecast, in which meteorologists have already weighed up the models.

The best model from observations

Which model was right most recently can be measured. Sailseer compares the previous day's forecast with observations from the nearest stations over up to 72 hours and ranks the models by their mean error. The most recent winner gets a mark in the model table, and the “Best model” row takes, for each hour, the top-ranked model that is still running at that hour. Wind and sailing wind alerts can optionally be based on it instead of the consensus. The limits are part of the picture: two stations are not a sailing area, and three days are just one weather pattern. When the pattern changes, another model may take the lead. Tap a point on the map and you will find the comparison under “Observed now”.

Official forecasts come first

Models are raw material. The marine forecast and the warnings from the weather services are the official basis, and they take precedence over any app, including Sailseer.