Russian Scientists Develop Neural Network to Improve Arctic Storm Forecasts
Researchers from several Russian scientific centers have developed a neural network system called BERTUNet that significantly improves forecasts of dangerous weather conditions in the Arctic.
Global weather models share a common weakness: they smooth out small atmospheric vortices and temperature anomalies, precisely the phenomena responsible for sudden storms, including polar cyclones and the Novaya Zemlya bora wind. BERTUNet addresses large-scale forecasting errors while intentionally preserving small vortex structures.
The system is primarily designed to improve safety along the Northern Sea Route, as well as aviation operations and resource extraction projects in the Arctic region. Researchers from the Shirshov Institute of Oceanology of the Russian Academy of Sciences, MIPT, Skoltech and AIRI participated in developing the platform.
