AI deep learning for climate forecasts

AI deep learning for climate science has attracted increasing attentions in recent years with rapidly expanded applications to many areas. In this talk, I will briefly present our recent progresses on using various deep learning methods for seasonal-to-multi-seasonal predictions of ENSO, the Indian Ocean Dipole (IOD), summer precipitation in China and East Africa, Arctic sea ice cover, ocean waves, as well as the bias correction and downscaling of dynamical model’s forecasts. The results suggest that many popular deep learning methods, such as convolutional neural networks, residual neural network, long-short term memory, ConvLSTM, multi-task learning, cycle-consistent generative adversarial networks and vision transformer, can be well applied to improve our understanding and predictions of climate. In addition, a brief introduction of AI large models for ensemble weather-subseasonal-seasonal-decadal forecasts, together with the perspective on the future development of AI methods, will also be presented.