The GENESIS Earth Systems Model (ESM) consists of an atmospheric general circulation model (AGCM) coupled to multi-layer models of
The atmospheric and oceanic components include large-scale dynamics, convection, and (for the atmosphere) detailed treatments of solar and infrared radiation, clouds and precipitation. Processes such as atmospheric chemistry, ocean biogeochemical cycles, soil microbial and trace-gas activity, net primary productivity, and climate-induced changes in vegetation are included in the various models. The main goal of the ESM is to simulate future changes in these systems on 50 to 200 year timescales due to greenhouse-gas emissions and other anthropogenic forcing.
The first version of the model, version 1.02, was the control version from 1992 to mid-1994 and found use in present-day and future simulations (Bonan et.al, 1992; Pollard and Thompson, 1994a,b; Thompson and Pollard, 1994a,b; Wang et.al, 1994), and many paleoclimatic applications (Barron et.al, 1993a,b; Crowley and Baum, 1994; Crowley et.al, 1993a,b, 1994; Crowley and Yip, 1994; Otto-Bliesner, 1993; Pollard and Schulz, 1994; Sloan, 1994; Wilson et.al, 1994).
The model formulation of version 1.02 is described briefly in Appendix A of Thompson and Pollard (1994a), and the land-surface model is presented in detail in Pollard and Thompson (1994a). Present-day results with version 1.02 have been compared to those of other GCMs by Robertson et.al (1993), Foster et.al (1994) and Hurrell et.al (1994). Also the Arctic climate of version 1.02 is included in the principal component analyses of McGinnis and Crane (1994).
The current version of GENESIS described here is designated version 2. The global grids of the models have various resolutions ranging from 350 km and 18 levels for the atmosphere to 100 km and 20 levels for the ocean. The choice of each grid is basically a compromise between the desire to resolve important physical processes and the ability of the complete model to perform 100 year simulations in reasonable time and within memory constraints on available computers. With a basic model timestep of 0.5 hours, the ESM uses 10 to 40 CPU hours to simulate one year and needs 10 to 50 Megawords of memory depending on its configuration.
The natural timescales of individual processes in the models range from minutes (for instance atmospheric convection, vegetation temperature, upper-soil moisture) to centuries (the response of deep-ocean temperature and salinity to changes in surface fluxes). One challenge of the ESM's design is to incorporate such diverse timescales in a time-marching framework, keeping all processes numerically stable yet keeping total CPU usage within reasonable bounds. There is a natural division between static 1-D components that only involve single grid points (for instance, LSX and all soil processes) and components involving horizontal dynamics that need to treat entire 2-D or 3-D fields at once (AGCM, OGCM, sea-ice advection). A central hub controls communication between the dynamical components and LSX, with the latter being a distributor to all the 1-D modules. The code for each model is modular, and is called individually by an overall driver. Communication between the models consists mainly of surface fluxes of thermal energy, momentum and water mass, and current state variables at a common two-dimensional boundary. For instance, to compute the surface heat flux over ocean the atmospheric model needs to know only the sea-surface temperature from the ocean model; the computed fluxes are then accumulated in time and passed as forcing to the ocean model the next time it is called. The central hub interpolates or extrapolates spatially between different horizontal grids, and accumulates fluxes in time between the different model timesteps.
The basic model timestep of 0.5 hours used by most processes is determined mainly by atmospheric dynamics. However, for many other processes a number of standard techniques (Washington and Parkinson, 1986) are employed for economy and stability: (i) computing more slowly varying but expensive processes at longer intervals (e.g. atmospheric infrared emissivities, solar radiation, sea-ice dynamics, ocean dynamics), (ii) treating fast processes as instantaneous and non-prognostic (e.g. plume convection), (iii) using time-implicit numerical methods (e.g. vegetation temperatures, soil-moisture flow, atmospheric gravity waves). In addition two tricks are used that attack the problem of the 500 year response time of the deep ocean, to avoid having to integrate the whole model for several thousand years just to achieve climatic equilibrium. One is called distorted physics, whereby the specific heat of water in the deep ocean is artificially reduced during the spin-up phase of a simulation (Bryan, 1984). The other is asynchronous coupling, whereby the atmospheric and oceanic models are run together for relatively short synchronous periods of about 10 years, separated by much longer asynchronous periods of about 100 years during which the ocean is run alone forced by surface fluxes extrapolated in time from the previous synchronous period(s) (Hasselman, 1988).
Subdirectories on the ftp site and their contents are as follows:
12/8/97