AI tools are taking off in the fields of meteorology, with the goals of greatly improving both accuracy and speed of predictions, and the potential to help identify the likelihood of extreme weather events like tornadoes, hail, and major storms. Recently, two studies published in Nature helped to quantify progress in those measures, in the form of the AI based systems Pangu-Weather, and NowcastNet.
For Pangu-Weather, the tool relies on 39 years of global weather data, allowing it to forecast temperature, pressure, and wind speed, but at a speed almost 10,000 times faster than conventional numerical weather prediction models. It also incorporates a 3D model, allowing it to examine atmospheric states at different pressure levels, and can predict weather patterns up to a week in advance. It cannot, however, predict rainfall.
NowcastNet, comparatively, specifically focuses on extreme rainfall patterns for more localized regions. Using radar observations from the US and Canada, the tool can predict precipitation rates up to 3 hours in advance, and is a reliable predictor of heavy rain, correctly identifying 71 per cent of events.
There are still concerns about these tools, however, as they’re not physics-based in their analysis, and are heavily dependent on the quality of data used. Says Amy McGovern, director of the National Science Foundation AI Institute for Research on Trustworthy AI in Weather, Climate, and Coastal Oceanography, “All of these generative AI models are promising.” She additionally notes, “Give it 5 to 10 years, we are going to be amazed at what these models can do.”
Source:
https://www.popsci.com/environment/ai-weather-prediction-accuracy/
