ECMWF AIFS, a forecast learnt by machine

Classic weather models solve the equations of the atmosphere, and when they share the same blind spot they can be wrong together: four routes agreeing are not always four good reasons to believe them. What a skipper weighing a passage lacks is an opinion that reached its forecast another way.

ECMWF AIFS takes another road: the European Centre's model solves no equation, it learnt by machine, from decades of past weather, how the atmosphere moves on from one state to the next. ECMWF runs it beside its classic model, and Saily WR reads each new forecast soon after it is published.

It forecasts the whole world up to 15 days ahead, so as a global model it gives a route on every passage, beside those of the classic models covering the same water. Its strength is the large picture several days out: where the lows and highs will lie, when a front will arrive, which way a tropical storm will go.

That is what an ocean passage turns on: from the Azores to Brittany, choosing a route mostly means choosing a low to ride or a high to skirt. AIFS starts from the very same state of the atmosphere as ECMWF's classic model; where their two routes part, the difference comes from the models alone, a clear sign that the forecast is uncertain and deserves a second look before you commit. Where AIFS, ECMWF and a model from another centre such as GFS agree, the forecast stands on firmer ground.

Having learnt to keep its average error small, it smooths: gusts and wind peaks are blunted, the more so the further ahead it looks. In a gale or along a sharp front, read its wind as probably too light, and its timing and track as the useful part. Like the other global models, it sees no sea breeze and no wind shadow behind an island. And every model draws on the same observations of the sea and the sky: their agreement is a good sign, not a promise.