Weather forecasting is the latest industry to flirt with an artificial intelligence makeover. But experts warn that climate change may pose a distinct challenge for AI weather models, which rely on historical data to produce forecasts, writes Chelsea Harvey. As the planet warms, heat waves and hurricanes are reaching unprecedented extremes. In the past week alone, the world experienced what scientists warned were likely the hottest days in recorded history. Unable to draw on similar trends from the past, AI may not be able to accurately forecast climate-fueled disasters. That could lead to a lack of preparedness for the worst of what nature has to offer. “The question of how AI models will perform in a warming climate is a very interesting one, and to my knowledge hasn’t been explored very thoroughly at this point,” Russ Schumacher, Colorado’s state climatologist and a scientist at Colorado State University, told Chelsea. AI is coming Researchers are already testing out new AI weather models from private companies. The results indicate a quickly developing technology that could infiltrate meteorology worldwide. For example, the Pangu-Weather model — developed by researchers at a Chinese technology company — forecasts global weather variables like temperature and wind speed up to 10,000 times faster than conventional models, according to a new paper published in the journal Nature. It’s even slightly more accurate than the models used by the European Centre for Medium-Range Weather Forecasts, one of the world’s leading weather centers. Another AI system, known as NowcastNet, is capable of outperforming many of its leading competitors in forecasting short-term rainfall, one of the trickiest weather variables to predict, according to another paper published in Nature. The climate catch But without historical or predictive data, AI systems may not be able to forecast climate change-fueled events, which are increasingly smashing records. In fact, if presented with entirely new weather conditions, AI models might churn out “highly erratic predictions,” Colorado State University researchers Imme Ebert-Uphoff and Kyle Hilburn said in a statement — a phenomenon observed in other incarnations of AI. That could undermine what meteorologists have worked years to achieve: accurate and reliable forecasting that allows governments to issue public safety announcements and facilitate evacuations with enough time to protect vulnerable populations.
| 
No comments:
Post a Comment