The Yandex team has introduced a significant update to its Weather service, implementing the "Neurometeor" technology, which uses neural networks for forecasting. This technology is the first of its kind available to a wide audience. It is capable of modeling atmospheric processes across the entire planet.
Yandex noted that "Neurometeor" generates global forecasts in a matter of minutes, which significantly reduces the time compared to traditional methods that require several hours. The previously used "Meteum" technology was also based on neural networks, but it depended on the optimization of multiple weather models, with calculations performed on supercomputers.
Currently, "Neurometeor" helps predict precipitation, one of the most difficult parameters in meteorology, for the next 16 hours. In the future, it is planned to expand the functionality to predict temperature, direction and strength of wind, as well as other meteorological indicators. A transition to neural network forecasts for the next 10 days is planned.
According to Yandex representatives, compared to Google's WeatherNext2 model, "Neurometeor" demonstrates comparable or even higher quality precipitation forecasts for the next 12–18 hours, using only one model instead of an ensemble, and also calculates the volume of precipitation for the entire world 10 times more accurately compared to other neural network solutions.
Read more related articles: