Forecast models repeatedly signal a deep low near Iceland

Meteorologist Einar Sveinbjörnsson estimates there is at least a 50 per cent chance that a deep low-pressure system will form in some shape by the end of next week, mbl.is reports. He says it is too early to assess its impact on Iceland or advise people to prepare, and stressed that the probability is his own estimate.

Sveinbjörnsson said the more striking point was how closely conventional weather models and artificial intelligence models agree on the possibility of a low about a week in advance. Forecasts have shown signs of its formation repeatedly in recent days.

There are still many uncertainties. The timing could shift by as much as 12 hours in either direction, and the biggest question is whether the low will form where current forecasts indicate. Its track will largely determine the weather in Iceland: it could make a decisive difference whether the centre passes west of the country or crosses it.

It is therefore too early to say which parts of Iceland might be worst affected or how severe the weather could become. Sveinbjörnsson said the forecast maps were not being shown to encourage people to prepare for a storm: “We are not at that point.”

He said weather forecasts had improved significantly in recent years. Two or three years ago, he said, he had viewed seven-to-10-day forecasts during periods of frequent winter low-pressure systems as “a kind of fantasy” that he took little notice of. Now, he said, such forecasts can be taken seriously with certain reservations when the same pattern appears repeatedly.

Sveinbjörnsson said progress in forecasting had come in leaps, and that weather forecasting may now be going through another such leap.

He also highlighted the agreement between the conventional model run by the European Centre for Medium-Range Weather Forecasts, known as ECMWF, and the artificial intelligence model AIFS. Conventional models calculate weather patterns using physical laws, while AI models learn from large volumes of weather data collected over decades. He said AI models had developed greatly in recent years.