To make their forecasts, meteorologists rely on different models from data from different countries. “Weather is a bit like cooking. Each meteorologist makes his own mix,” summarizes Paul Marquis, independent meteorologist from Marseille. As ingredients, weather models with somewhat barbaric acronyms: GFS, ECMWF, Icon, Arome, Arpege…
These models are produced by different countries and have their own uses: “The advantage is that they will have more or less fine meshes. But the finer the mesh, the less far in time the forecast goes,” explains Christophe Dumard, the weather man for the Transat Jacques Vabre race committee, whose departure was postponed for part of the fleet due to the Approaching Storm Ciaran.
Thus, Météo-France publishes two modeling systems: Arpege, with a mesh of 16 kilometers and Arome where the mesh is of the order of one kilometer. In fact, the forecasts established do not go as far back in time and do not have the same geographical coverage. The Arpege model models weather conditions over Europe for a period of four days, compared to 48 hours for Arome which covers France and neighboring countries. And for now, the only limit to the extent of these forecasts remains the calculation capabilities. “When the resolution goes from 20 to 10 kilometers, for example, it is not twice as much computing power but four times more that must be deployed,” indicates Christophe Dumard.
Today, weather models allow simulation for up to 16 days. To do this, meteorologists take data from the largest mesh models, which is the American GFS model, with a resolution of 22 kilometers. Followed by the German Icon model (13 kilometers of resolution) and the European ECMWF with a mesh size of 9 kilometers. And to make their forecasts, these models integrate a bunch of variations, the famous “butterfly effects”, then put them through the mill of their algorithm which will make them evolve over time. “The GFS calculates 30 possible scenarios, the ECMWF 50, and the Arome model 15,” continues Christophe Dumard. And it is from these 30, 50, or 15 different results that meteorologists will obtain their “confidence index”.
“The closer the results between the different models are, the more reliable the forecasts are,” indicates Paul Marquis. But not all global models such as the GFS or the ECMWF are equally favored by meteorologists, as Paul Marquis explains: “For two or three years, the American GFS model, which was the best, has given, I find, , exaggerated results, particularly on the calculation of wind gusts. Currently I prefer to work with ECMWF data.”
A gradual approach
And behind this race for reliability, there is a kind of arms race in the background. Thirty European countries came together to launch in 2006 a fleet of meteorological satellites managed by the intergovernmental organization Eumetsat. However, the data collected by these satellites, but also by the various observation stations and buoys are made public, although some are by subscription, so that meteorologists around the world can have access to them.
Overall, the forecasters’ working methodology consists of initially studying the different models from the highest mesh systems before refining as the storm approaches, for example, with models at smaller resolutions. What Christophe Dumard expresses for sailing races: “For coastal I start with the Arome model, for offshore I rely on the GFS or ECMWF and between the two, I take Arpege”.
Still, one might wonder, since data is shared and certain models are the result of intergovernmental cooperation, why does each country still need its own organization? Well, probably because it is the responsibility of governments to ensure the security of the goods and people of their countries. Thus, Météo-France placed three Breton departments on orange alert, while the SNCF immediately announced that it was suspending the circulation of its trains in Brittany, Normandy and Pays-de-la-Loire on Thursday.
We remember, for example, that Météo-France was strongly criticized after placing Corsica in simple yellow vigilance (before raising it to orange, but too late) in August 2022 during a violent storm episode in which five people lost their lives. .