Model Structure

The atmospheric model formulation (HRES, ENS, etc) is based on a set of basic equations.  These are either:

• diagnostic and describe the static relationship between pressure, density, temperature and height.
• prognostic and describe the time evolution of the horizontal wind components, surface pressure, and the temperature and the water vapour content of the model atmosphere.

Additional equations describe physical processes within the atmosphere:

• changes in the hydrometeors (rain, snow, liquid water, cloud ice content etc.).  
• processes of radiation, gravity wave drag, vertical turbulence, convection, clouds, and surface interactions.
• climate fields of passive aerosols, reactive chemicals, and greenhouse gases together with their impact on the radiation or moisture balance in the atmosphere.
• effects of solar radiation on high altitude chemistry in the model stratosphere and above.

These processes are mostly unresolved due to their small scales compared to model resolution.  Because of this, they are handled by physical parameterisation in a statistical way that describes the mean effect of sub-grid processes.

The model equations are discretized in space and time and solved numerically by a semi-Lagrangian advection scheme.  This ensures stability and accuracy while using as large time-steps as possible to progress the computation of the forecast within an acceptable time.

Other models running at the same time describe energy, moisture and momentum fluxes from the underlying land or sea surfaces.  These are:

• a Land-Surface Model (HTESSEL), including a program for lakes and coastal waters (FLake) that itself provides information on temperature changes and the development, decay and movement of areas of ice which impact upon heat flux and albedo.
• the Wave Model (ECWAM)
• the Dynamic Ocean Model (NEMO), including a program (LIM2) which provides information on the development, decay and movement of areas of sea ice which impact upon heat flux and albedo.
  
  

Fig2.1.1: Sub-grid scale parameterised processes in the ECMWF model – Surface to Stratosphere.

Basic prognostic equations are efficiently processed in spectral space and this is reflected in a small proportion of computer time required.  However, many processes are computed in grid point space (e.g. rainfall) which requires a larger, but still relatively small, proportion of computer time.  This and processes in spectral space, together with the associated spectral transforms, are approximately similar.  The necessary transpositions between spectral and grid point spaces is a significant overhead in computer time.  Semi-lagrangian computations also take up an important proportion of processing time.

Fig2.1.2: A schematic pie-chart showing approximate proportions of computer processing time during execution of an atmospheric model forecast based on T799 (regular 25km resolution grid) and 91 levels (current resolution for ENS is T_co640 (18km resolution) and 137 levels).   Parameterised physical processes consume about 30% of computer processing time.  Computations in grid point space and spectral space together take about 20% of computer processing time while rather more time (~27%) is taken in transposing data from one space to the other.