Russia Unveils NeuroMeteum, a Neural Network-Based Weather Forecasting Technology
The system independently models atmospheric behavior across the entire planet.
Yandex has introduced a new weather forecasting technology called NeuroMeteum (neural weather forecasting system), the company’s press service said. The system uses neural networks to independently model atmospheric behavior on a global scale and generate worldwide forecasts in a matter of minutes, making it faster and more accurate than traditional calculation methods.
Unlike the earlier Meteum system, which focused on selecting optimal combinations of forecasts produced by meteorological centers, NeuroMeteum does not rely on supercomputer-based models that can take several hours to run. Instead, it performs end-to-end atmospheric modeling using neural networks.
Outperforming Existing Models
At the initial stage, NeuroMeteum is being used to forecast precipitation for the next 16 hours, one of the most challenging parameters in meteorology. Over time, the technology is expected to be adapted for predicting temperature, wind, and other weather indicators. The service is also set to transition to neural network-based forecasts for periods of up to 10 days.
According to the company, the Russia-developed NeuroMeteum outperforms the international WeatherNext2 model in precipitation forecast quality for the 12–18 hour range. The system achieves this using a single model rather than an ensemble approach. Overall, the domestic solution is capable of estimating global precipitation volumes with up to ten times greater accuracy than comparable neural network-based systems.
