Weather forecast models incorporate many data points from satellites, radar, the upper atmosphere, and surface weather station information like temperature, humidity, windspeed, and direction. The models crunch the data with complex mathematical equations.
The new AI models use Google's DeepMind Graphcast model, an AI weather model, that has been fine-tuned by NOAA.
Such models use machine learning and train on historical weather data.
The new models will provide improved accuracy for large-scale weather systems.
Chris Nuttall is the warning coordination meteorologist for the National Weather Service in Shreveport, Louisiana.
"Something on the scale of like a hurricane or a large cold front or a large low pressure system moving through the area … it’s when we're looking at upper-level lows, upper-level troughs, what we call in meteorology, the synoptic scale. That's where this should really help us out. These three models that they've just debuted; those really aren't going to help us as far as with tornado warnings and getting a tornado warning out faster or sooner. They can't model down to that level of detail."
He said the new AI models will not take as long to run, and their performance is similar to that of traditional weather models.
"Weather is always changing. The GFS, for example, runs every six hours, because it takes 2-3 hours to run well. With it running over less than an hour, maybe we can run it every two hours. Maybe we can run it every three, maybe every hour. We'll just have to see."
