Two new papers published in the journal Nature highlight the potential of artificial intelligence (AI) in revolutionising weather forecasting, reports E&E News. The papers describe two AI weather forecasting models, Pangu-Weather and NowcastNet, which offer faster and potentially more accurate results compared to traditional models. Pangu-Weather, developed by Huawei Technologies, can accurately predict global weather variables up to a week in advance, 10,000 times faster than conventional models, and track the path tropical cyclones. It performs slightly better than the European Centre for Medium-Range Weather Forecasts. NowcastNet specializes in short-term rainfall forecasts with a lead time of up to 3 hours, and has outperformed many competitors in this area.
Whereas conventional forecasting systems represent the physics behind the movement of air in the atmosphere and oceans with mathematical equations, AI models use historical weather data and learn to recognize patterns and make predictions when fed current weather conditions. Pangu-Weather and NowcastNet are part of a wave of new AI weather models, many being developed by private corporations, marking a departure from traditional weather forecasting done by government entities.
There are, however, challenges that AI weather models face, particularly in the context of a changing climate. As climate change leads to more intense and unprecedented weather events like heatwaves, droughts and hurricanes, there are fewer examples of such events in the historical record. This makes it difficult for AI models to accurately simulate and predict these extremes. The behaviour of AI systems under such conditions may be unpredictable and erratic, according to experts.
The field of AI weather forecasting is evolving very fast: two years ago, in this opinion piece, scientists recognised that there “might be potential” for AI models to produce equal or better quality forecasts than numerical models for specific demands, adding “we think that it is not inconceivable that numerical weather models may one day become obsolete, but a number of fundamental breakthroughs are needed before this goal comes into reach.”
Researchers Imme Ebert-Uphoff and Kyle Hilburn from Colorado State University commented on the new models in Nature, acknowledging that the faster computational speed of models such as Pangu-Weather could “yield immense benefits”, but also that they may encounter problems when simulating extreme weather events as they become more intense due to climate change.
“The question of how AI models will perform in a warming climate is a very interesting one”, said Russ Schumacher, Colorado’s state climatologist and also a researcher at Colorado State University. He suggests that hybrid models, including AI and numerical model components, might do better to forecast record-breaking events, highlighting that meteorology can leverage the strengths of all approaches to enhance weather forecasting capabilities.
