Let us know about free updates
Improved weather forecasting using artificial intelligence promises to take a major step towards launching a new European system that could surpass traditional forecasting methods up to 15 days in the future.
While high-tech companies and weather offices around the world are already applying AI to weather, the European Medium-Distance Weather Forecast Centre (ECMWF) operates by making its operating model freely available to anyone at any time. He said the model has broken new grounds.
“This milestone transforms weather science and forecasting,” said Florence Rabier, director of the intergovernmental organization ECMWF. “By operating an AI prediction system, we use machine learning available to date to generate the widest range of parameters.”
The experimental versions tested over the last 18 months showed that the system was about 20% accurate with key predictions than the best traditional methods. It fed millions of global weather observations to supercomputers and adjusted them with physics-based equations.
According to Florian Pappenberger, ECMWF’s forecast director, the new European system is able to predict tropical cyclone tracks 12 hours ahead, giving valuable additional warning time to serious events.
The world recorded its hottest temperatures in 2024, and Europe became an extreme warming continent, causing extreme weather events. The agency is at the forefront of observation and public perception of the impacts of climate change.
Experts to study temperature ECMWF’s “weather room” is predicted extreme
Other medium-range AI prediction systems under development include Gencast and Graphcast from Google Deepmind, Pangue-Weather from Huawei, Fourcastnet from Nvidia, and Fuxi from Science and the AI Shanghai Academy at Fudan University. All were trained in a database of weather observations compiled by the ECMWF for 40 years.
Comparing the accuracy of competing AI prediction systems was difficult, Pappenberger said, as their relative performance differed depending on the variables evaluated and timescale. The scores published by ECMWF give you an idea of performance, but do not identify the entire champion.
However, Pappenberger noted that the system stands out because it predicts more features than standard temperatures, precipitation and wind. For example, solar radiation and wind speeds are predicted to be useful for the renewable energy sector at a typical turbine height of 100 meters.
ECMWF forecasts are freely available, but institutions do not issue harsh weather warnings or tailor-made forecasts to industry users, leaving specialized forecasts to national or local governments and private companies.
The ECMWF and the European group of National Met Offices have created an open source technology framework for AI weather systems called ANEMOI, after the Greek god of wind. The underlying machine learning architecture is based on the same “graph neural network” as Google Deepmind’s prediction model.
Deepmind’s research director Peter Battaglia said it was “impressive” to see how ECMWF adapted to the AI waves that have reshaped the field in recent years, with the latest open models being added to the pool of knowledge He said.
ECMWF increases spatial resolution, generates one prediction from the current version, and moves to “ensemble predictions” or creates a collection of 50 predictions at the same time, with slightly different starting conditions, allowing the system to be used either. We plan to improve it further. The range of possible outcomes.
In the future, Kirstine Dale, chief AI officer at Met Office in the UK, said, “Physics to provide the strengths of their combination to provide accurate, fast, reliable, reliable predictions.” He said it would be necessary to mix base and database simulations.
Today, the boundaries of reliable daily weather forecasts in Europe are 6-7 days off for precipitation and winds, with temperatures up to 14 or 15 days.
“There’s a good chance that machine learning models can extract something from data that is not currently well represented by physics-based models, so there’s a fair chance that they’ll expand on that.”
Video: Extreme Science of Climate Prediction | ft Climate Capital
