MOSCOW, June 15. /TASS/. Russian scientists developed an AI algorithm to predict accurately extreme weather events in the Russian Arctic and to study storms and various medium-scale vortices. The development is about five times higher in resolution than global climate models, the MIPT Center for Scientific Communication reported.

"By having tracked mesoscale vortices and compared their lifecycle statistics with a benchmark, we have proven that the neural network captures key properties of polar mesocyclones. Results of our calculations demonstrate that the intensity of these hazards is reproduced realistically, which is important for applications such as wind power or maritime safety," the university's Center for Scientific Communication quoted leader of MIPT's Machine Learning Laboratory in Geosciences Mikhail Krinitsky as saying.

The climate and weather in the Arctic are changing very rapidly due to global warming, which also contributes to an increase in numbers of extreme weather events in the polar regions, the scientist and his colleagues said. This encourages scientists to create new tools and approaches to predicting such anomalies, as well as to studying big, medium and small vortices and storms that occur in Russia's circumpolar and polar regions.

So far, for these purposes have been used various local and global climate models, but they require a huge amount of computing resources to obtain detailed high-resolution forecasts. The Russian scientists have suggested that these costs can be cut significantly by using a neural network trained on a big sample of results from similar supercomputer calculations.

Thus, the scientists have created an AI algorithm capable of using data from the ERA5 global weather archive for the recent eight decades, where the world map is split into squares 31 km long and wide. They managed to use the algorythm to make much more detailed forecasts, where environmental conditions are calculated with a resolution of 6 by 6 km. After 17 hours of training, the scientists compared results of AI calculations with accurate data obtained in using the WRF (Weather Research and Forecasting) model.

The quality of these forecasts turned out to be generally comparable, however, the process of calculating annual wind fields using AI takes only 10 minutes, whereas the reference model needs about 10 hours to complete that process. At the same time, the neural network is particularly good at predicting movement of squall winds and wave heights in the Barents Sea, which makes it possible to use AI to predict accurately dangerous waves and storm warnings, the researchers concluded.