MOSCOW, December 3. /TASS/. Russian researchers developed an AI system capable of detailing data from global meteorological services and predicting dangerous vortices and storms in the Arctic about 50 times faster than complex physical models do. The development will improve weather forecasts in the Russian Arctic, press service of the Moscow Institute of Physics and Technology (MIPT) told TASS.
"Our development's main advantage is speed with minimal loss in quality. The neural network gives a highly detailed forecast for the Barents and Kara Seas more than 50 times faster than resource-intensive physical models. In addition, it shows storms that other approaches do not reproduce," the press service quoted head of the university's Machine Learning Laboratory in Earth Sciences Mikhail Krinitsky as saying.
The scientist and his colleagues have been working for a long time to create AI tools for more accurate and quick forecast of how the ocean behaves and how it contributes to the Earth climate's fluctuations. Currently used global meteorological models have low spatial resolution and can barely "see" small atmospheric vortices, and high-precision hydrodynamic models require enormous calculations and time.
Russian scientists hoped for a similar quality with significantly less time and resources if AI is trained based on results obtained from using the WRF (Weather Research and Forecasting) hydrodynamic model. Thus, the researchers prepared a set of calculated data on weather above the Barents and Kara Seas in 2015-2021, and trained a neural network, created by scientists of MIPT and the Shirshov Institute of Oceanology (the Russian Academy of Sciences).
The researchers calculated how weather was changing in that region in 2022 and 2023, and compared AI forecasts with real data and WRF and ERA5 (atmospheric reanalysis) calculations. The Russian AI system has accurately predicted general weather fluctuations in the Arctic during the period and also reproduced the so-called Novaya Zemlya 'bora' - an unusually strong cold wind that moves rapidly from the mountain ranges of Novaya Zemlya and poses a danger to shipping.
The neural network significantly surpassed the ERA5 global meteorological model in predicting vortex structures, where its forecasts almost completely coincided with the WRF reference model's calculation, the scientist said. In the future, specialists will predict extreme weather events more accurately and at lower cost to ensure the safety of navigation, ports and oil and gas platforms in the Arctic.