Researchers Test Whether Modern AI Could Have Nailed the D-Day Weather Forecast

A new film called Pressure has renewed interest in one of history's most consequential weather forecasts: the prediction that helped General Dwight D. Eisenhower decide when to launch the D-Day invasion of Normandy in June 1944. Meteorologist James Stagg and his teams had to judge whether conditions would let more than 150,000 Allied troops land safely, briefing Eisenhower at Southwick House near Portsmouth. A new study combined digitized historical weather observations, modern weather science and an AI forecasting system to ask whether today's technology would have gotten that forecast right.

The answer turned out to be surprisingly nuanced, the researchers found. Modern AI forecasts would probably have supported delaying the invasion by a day, to June 6, but would also have underestimated how severe conditions actually were that day. Even with today's technology, the study concluded, D-Day would not have been an easy call. Today's scientists also have fewer historical weather observations available than Stagg did, because many paper records from that period have not yet been digitized.

In the film, one forecasting team, an amalgamation of two real-life teams led by Sverre Pettersen and Geoffrey Wolfe, focused on the physical development of weather systems, while a second team led by American meteorologist Irving Krick searched for past examples of similar weather and predicted the future would develop the same way, an approach known as analog forecasting. Stagg had to combine these often-contradictory forecasts into simple guidance for Eisenhower.

The researchers tested a modern AI forecasting model called AIFS, developed by the European Centre for Medium-Range Weather Forecasts, producing a series of forecasts leading up to D-Day. The system captured the storm over the North Sea, but the researchers were surprised by how much uncertainty remained in its forecasts for the Normandy beaches, and the storm affecting the early landings moved through the North Sea faster in the AI's forecasts than it did in reality.

Even if Stagg's teams had seen these AI forecasts, the complexity of the decision would still have stood out: forecasts made on June 5 showed a 30% chance of strong winds and significant cloud cover for the beaches on June 6. The researchers concluded that while future forecasting may combine physical and AI models, the role of a human expert in interpreting and challenging those forecasts remains critical.