Can AI Models Handle Extreme Weather Events?

AI models are becoming a big deal in weather forecasting. They are supposed to help us predict extreme weather events. But there is a problem. These models are trained on past weather data. They struggle with events that are so rare, they weren't in the training data. These rare events are called "gray swan" events. They are possible, but we don't see them often. One big question is whether AI models can predict these gray swan events. Can they take what they know about smaller storms and apply it to bigger, rarer ones? To find out, researchers did an experiment. They trained different versions of an AI model called FourCastNet. Some versions were trained on all weather data from 1979 to 2015. Others were trained without data from the strongest tropical cyclones, called Category 3-5. Then, they tested these models on Category 5 tropical cyclones from 2018 to 2023. These are the gray swan events. The models that were trained without the strongest cyclones couldn't predict these events accurately. This shows that AI models can't just guess about stronger storms based on weaker ones. They need to see the stronger ones in their training data. However, there was a twist. The models that were trained without strong cyclones in one specific area could still predict strong cyclones in that area. This suggests that AI models can learn from one region and apply it to another. This is interesting because the models don't get specific regional data. They learn it on their own. This experiment shows that current AI models have a long way to go. They need new ways of learning to predict the rarest and most dangerous weather events. This is not just about tropical cyclones. It's about all kinds of extreme weather. We need AI models that can give us early warnings about these events. This way, we can be better prepared. It's clear that AI models have potential. But they also have limitations. We need to keep improving them. We need to find new ways for them to learn. Only then can they help us with the most challenging weather events. This is a big task. But it's an important one. Our safety depends on it.