Using Deep Learning for coupled atmospere-ocean modeling
Dale Durran
Univ of Washington
3 Apr, 2pm
Abstract:
The deep learning weather prediction (DLWP) model of Weyn et al. (2021) is significantly improved by shifting its grid structure from the cube sphere to the Hierarchical Equal Area Pixelization (HEALPix) mesh, which is used extensively in astronomy. This is an easy-to-refine equal-area mesh whose cells lie along lines of constant latitude. The HEALPix mesh has unique properties that make it better suited for CNNs in weather forecasting applications than the cube sphere or alternative grid structures. Further improvements were obtained by refining the convolution neural network architecture and by introducing gated recurrent units.
The model remains parsimonious, using only eight 2D shells of prognostic data with an effective grid spacing of roughly 100 km. The model simulates realistic weather patterns at 3-hour time resolution while being recursively stepped forward over a full annual cycle.
The climatology and multi-year stability of the model is dramatically improved by coupling it to a prognostic ocean model that predicts sea-surface temperatures. The representation of SST features such as El Niņo is improved by adding an observed variable, out-going long-wave radiation (OLR) to the set of prognostic fields predicted in the coupled model. This use of OLR from the ISSCP dataset extends the type of training variables used in machine learning weather prediction beyond reanalysis and NWP-model-generated datasets.