The new European model of meteorological conditions is faster, more intelligent and free -here’s what you know

The European Medium Range Forecast Center (ECMWF) has just launched AI forecasting model, which the center says is superior to the latest physics models by up to 20%.

The model is called the artificial intelligence (AIFS) prediction system. According to ECMWF, the new model works at faster speeds than physics-based models and takes approximately 1000 times less energy to make a forecast.

ECMWF, which is already 50 years old, produces ENS, one of the world’s leading models to forecast with medium range. Forecasting of average scope Includes weather forecasts made between three days and 15 days in advance, but ECMWF also predicts time up to one year ahead. Time forecasting models are essential for countries and local authorities to remain prepared for extreme meteorological events – as well as for more daily needs, such as knowing what the time of your upcoming vacation will be.

Traditional time forecasting models make forecasts by solving physical equations. The limit of these models is that they are the approximations of atmospheric dynamics. A captivating aspect of AI-managed models is that they could learn more complex relationships and dynamics in meteorological models directly from the data, not rely solely on known and documented equations.

ECMWF message comes on Google Deepmind’s heels Gencast model To predict AI power supply time, the next iteration of the Google weather forecasting software that includes Neuralgcm and GraphicsS Gencaste is superior EnsThe leading model of ECMWF time forecast, 97.2% of the goals in different variables of time. With time of performance, more than 36 hours, Gencast is more accurate than ENS at 99.8% of the goals.

But the European Center is also innovative. Starting AIFS-Single is just the first operational version of the system.

“This is a huge endeavor that guarantees that the models work in a stable and reliable way,” says Florian Papenberger, director of forecasts and services at ECMWF, in the Center’s edition. “Currently, the resolution of AIF is less than that of our model (IFS), which achieves a resolution of 9 km (5.6 miles) using an approach based on physics.”

“We see AIFS and IFS as complementary and part of the provision of a number of products to our consumer community that decide what best suits their needs,” added Papenberger.

The team will examine hybridizing data based and physics to improve the organization’s ability to predict time with accuracy.

“Physics based models are crucial for the current data assimilation process,” says Matthew Chantry, a strategic leading role in ECMWF machine training and the manager of the innovation platform, in an email to Gizmodo. “The same process of assimilation of data is also vital for the initialization of every day models for machine learning and allows them to make predictions.”

“One of the next limits to forecasting machine learning time is this step of assimilation of data, which, if resolved, would mean that the full -time forecasting chain can be based on machine learning,” Chantry added.

Chantry co -authored a survey in anticipation of an affiliate check that describes a system -controlled system, an end -to -end estimated system that does not rely on physics -based re -analysis.

Called Graphdop, the system uses observed quantities, such as brightness temperatures from polar orbits, “to form a coherent latent representation of the dynamics and physical processes of the earth system,” the team writes, “and is able to produce skillful forecasts for the respective meteorological parameters, to Five days in the future. “

The integration of artificial intelligence methods by modeling time forecasting managed by physics is a promising place for more precise prediction. Testing so far shows that AI power supply can surpass historical models, but so far, these models rely on resource data. On -site observations were essential to training models, and it remains to be seen how impressive the ability to predict technology would be when forced to go out of the script.