Koen Hufkens, PhDNo surprises here. If it is out of distribution it will be hard to predict with ML (AI). Increasingly the weather will slide in a permanent state of being "out of distribution".
Obviously, you can always continuously fine tune the model but then it loses a part of its cost advantage. Augmentation techniques, based upon simulated scenarios of state of the art physical models could help.
All GNU social JP content and data are available under the