Accurate and Fast Neural Network Emulations and Parameterizations of Climate Model Physics
Abstract
The need to include more comprehensive and complex fast physical processes in atmospheric models leads to increasing computational costs and demands. The demand significantly exceeds currently available computational resources and grows faster than the resources do. We will briefly review applications of neural network (NN) techniques which allow us to introduce complex physical parameterizations of sub-grid scale physical processes without exceeding computational limitations. Two different approaches will be presented. The first approach is the development of a NN emulation of an existing state of the art, costly parameterization. The NN emulation approximates the parameterization with high accuracy. It provides a speed-up of calculations that is two or more orders of magnitude faster than the original parameterization. This approach is illustrated by NN emulations of the RRTM short and long wave radiative parameterizations developed for NCEP CFS. The second approach produces a new NN based parameterization developed using simulated and/or observed (ARM, etc.) data that contain information on fast and sub-grid scale physics. The NN based parameterization that is produced is a stochastic parameterization. It is implemented as an ensemble of NNs. This approach is illustrated by a prototype of a NN based convection parameterization developed for NCAR CAM using simulated data produced by the cloud resolving model called SAM (System for Atmospheric Modeling) developed by M. Khairoutdinov, D. Randall and collaborators. These NN approaches provide opportunities to use new and much more sophisticated state of the art model physics parameterizations which are computationally prohibited otherwise.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2010
- Bibcode:
- 2010AGUFM.A22C..02K
- Keywords:
-
- 0320 ATMOSPHERIC COMPOSITION AND STRUCTURE / Cloud physics and chemistry;
- 0360 ATMOSPHERIC COMPOSITION AND STRUCTURE / Radiation: transmission and scattering;
- 1626 GLOBAL CHANGE / Global climate models