The National Oceanic and Atmospheric Administration of the United States (NOAA) has begun a new stage in weather forecasting by betting on models based on artificial intelligence.
These systems, most of which are already operational, promise to generate faster, more efficient forecasts, using only a fraction of the computing power required by traditional models.
This advance has been possible thanks to the work of Environmental Modeling Center of NOAA, in coordination with the National Weather Service. As explained by the agency, artificial intelligence is not designed to replace traditional numerical models based on physical equations, but to complement them.
In fact, these classic systems remain one of the main data sources used to train machine learning models.
For decades, NOAA weather forecasts have been based on Global Forecast System (GFS), a physical model that simulates the atmosphere using complex mathematical equations to generate data on temperature, wind, precipitation and other parameters.
Subsequently, it was developed Global Ensemble Forecast System (GEFS), which runs multiple simulations with the aim of reducing biases and better representing the uncertainty of different meteorological scenarios.
As explained Daryl Kleistdeputy director of Environmental Modeling Centerthe new artificial intelligence models have been trained with decades of data from these traditional systems.
NOAA estimates that these AI systems will consume between 91% and 99% less computing resources than conventional models during their execution. In addition, they could extend the horizon of the usual forecasts by an additional 18 to 24 hours.