Sub-Seasonal and Seasonal Forecasting with a Deep Learning Earth-System Model
, Professor, University of Washington, Department of Atmospheric Sciences
When forecasting the state of the atmosphere at lead times longer than two weeks, we must account for the influence of the ocean and land surface. These provide lower boundary conditions that evolve much more slowly than the day-to-day weather, and via their persistence, they imprint a signal on atmospheric circulations that's important for sub-seasonal and seasonal forecasts, and climate simulations. Some surface influences, such as atmosphere-biosphere exchange, are particularly difficult to express as fundamental equations suitable for numerical solution in a global model. This is one reason that deep learning provides an attractive alternative to conventional numerical simulation in earth-system modeling.
Here we present first results from a parsimonious deep learning earth-system model with coupled atmospheric and ocean modules. We examine its skill in ensemble forecasts of sub-seasonal weather anomalies and El Niño.