Meteorological Institute, Ludwig-Maximilians-University
"Machine learning to represent atmospheric sub-grid processes"
The representation of sub-grid processes, especially clouds, remains the largest source of uncertainty for climate prediction. Cloud-resolving models alleviate many of the gravest problems but will remain too computationally expensive for climate predictions in the coming decades. In this talk I will discuss how machine learning, and deep learning specifically, can learn to parameterize atmospheric sub-grid processes from short-term high resolution simulations. Our results tie in with a recent push towards a more data-drive climate model development.