(Very) Brief History of Global Atmospheric Modeling at GFDL:
Joseph Smagorinsky and Syukuro Manabe pioneered the development of numerical models of the atmosphere suitable for studying the Earth’s climate in the 1950’s and 1960’s, resulting by 1965 in a model with many characterstics still familiar today: a fluid dynamical core solving the primitive equations on the sphere with enough horizontal resolution to explicitly resolve the basic structure of midlatitude storms, but with most vertical fluxes of energy and water, especially in the tropics, treated with sub-grid parameterizations; relatively modest vertical resolution (9 levels in this particular case); and with a substantial fraction of the computation devoted to comprehensive radiative transfer.
In the following 2 to 3 decades, global atmospheric modeling at GFDL evolved in various directions, including the first efforts to develop atmospheric models suitable for coupling to ocean models (headed by Syukoro Manabe), models focusing on stratospheric dynamics and chemistry (headed by Jerry Mahlman) and models that explored the limits of deterministic weather forecasting (headed by Kikuro Miyakoda).
Many of these earliest models used a finite-difference horizontal discretization of the equations of motion, in which the evolution of the state of the atmosphere is followed on a set of horizontal grid points, but in the 1980’s much of this modeling effort moved to a spectral dynamical core, in which atmospheric fields are decomposed into spherical harmonics, with the amplitudes of these harmonics being the fundamental model variables, transforming back to physical space as needed. While spectral models have dominated atmospheric modeling, for both climate research and numercial weather prediction, for the past several decades, because they are very efficient and handle spherical geometry in a very natural way, they also have significant deficiencies, most notably a difficulty in treating the Earth’s topography without creating artificial wave-like features (Gibbs’ ripples) in the height of the surface.
In the late 1990’s, atmospheric modeling had become so demanding of diverse expertise, including software engineering, that GFDL decided to consolidate its atmospheric (and other) modeling activities to reduce the number of models being actively developed and to work within a coherent software framework so as to make it easier to take advantage of scalable computer architectures. The Flexible Modeling System was developed for this purpose, and a Global Atmospheric Model Development Team (GAMDT) constituted to design a new atmospheric model, for use in simulations that would be made available for the IPCCs 4th Assesment.
The AM2 model series:
The starting point for the AM2 development was an ongoing effort headed by Jeff Anderson to develop a gridpont model that had been in use for research in extended range forecasting by Kikuro Miyakoda and collaborators (a model that was denoted as AM1, in retrospect). The dynamical core -etc — by Bruce Wyman.
The model was evolved and was finalized under the GAMDT leadership of Steve Klein and Paul Kushner. etc
As AM2.0 was under development, a new gridpoint dynamical core, based on a “finite-volume” numerical framework, was undergoing final testing and showing a number of significant benefits, especially when coupled to an ocean model.
While ranked as one of the best of the world’s atmospheric models, a number of aspects of AM2 clearly needed to be improved. Many of these aspects relate to the simulation of clouds and of moist convection. In addition, there was no serious attempt to simulate the stratosphere in AM2, and there was no aerosol or gas phase chemistry incorporated into the model. AM3 is an amibitious attempt to move AM towards a more physically-based simulation of clouds and convection, allowing for the interaction between clouds and natural as well as anthropogenic aerosols, the simulation of stratospheric ozone (including the Antarctic ozone hole), as well as the tropospheric chemistry needded to discuss transcontinental transport of pollutants and global air quality. The model also takes advantage of developments in the finite volume dynamical core, sepcifically a new grid with the topology of a cube (the “cubed sphere” grid topology) that results in much improved performance on scalable computer architectures and more homogeneous simulations of flow over polar and non-polar regions.