GENIE:Igcm

GENIE-IGCM is an (I)ntermediate complexity (G)eneral (C)irculation (M)odel.

History
The ICGM was initially developed in the department of Meteorology, University of Reading. In Reading the IGCM went through three versions:


 * IGCM1 : Portable version of the original adiabatic baroclinic model (Hoskins & Simmons, 1975), with an option for Newtonian forcing and Rayleigh drag to obtain a stable climate.
 * IGCM2 : Includes advected trace species, simplified moist parameterizations and a cheap "radiation scheme" (constant tropospheric cooling!).
 * IGCM3 : Intermediate climate model which includes more sophisticated radiation and real-Earth surface conditions. IGCM3 was developed in the mid 1990s, mainly by Piers Forster, Robin Glover and Suzanne Rosier, for climate sensitivity studies.

IGCM3 is currently being used for research at Reading (sometimes refered to as RUGCM). Mike Blackburn is the current point of contact at Reading. Details of the Reading IGCM can be found at: http://www.met.rdg.ac.uk/~mike/dyn_models/igcm/

The PUMA model: http://www.mi.uni-hamburg.de/PUMA.215.0.html?&L=3 was also developed from the early versions of IGCM (at least).

In terms of GENIE, the GENIE-IGCM contracts have been to:


 * Dan Lunt (GENIE)
 * John Hughes (GENIEfy)

Though more people than this have been involved in working on it.

Summary of changes since IGCM3
GENIE-IGCM is based on IGCM3, with the following changes (up to July 2006). The changes are either scientific or structural. The structural changes do not affect the results (or only affect them at the last decimal place, due to rounding errors).

Structural:


 * Diabatic/Adiabatic/Boundary-Layer split into separate parts.
 * Surface fluxes split into a separate GENIE module.
 * Separation of the ocean and sea ice into separate GENIE modules.
 * NetCDF input/output.
 * All tunable parameters moved into the namelists.
 * Dimensionalisation of the land-surface and boundary layer schemes.
 * Added tests for water and energy conservation.

Scientific:


 * Addition of the Tiedtke convection scheme (as an option).
 * Input files added to run at T42 if required.
 * 7 vertical levels is now the default.
 * Seaice albedo corrected so that the albedo changes when seaice melts.
 * Small bug in the land surface scheme corrected.
 * Ordering of some of the calculations in the land surface scheme changed.
 * Timestep of ocean to 2 days and seaice to 6 hours (both were every atmosphere timestep previously).
 * Small bug in the vegetation-albedo lookup table corrected.
 * Small bug in the vegetation-type input file corrected.
 * Closure of the hydrological cycle by properly accounting for sublimation.
 * Adding cloud cover diagnostic to the Tiedtke convection scheme.

Model structure
The IGCM is a spectral model. Typically run at T21 there have been attempts to run it at T42 also.

In the code the main division in the dynamics is between diabatic (i.e. wet etc) and adiabatic (i.e. dynamics) processes. The top level structure of GENIE-IGCM is:



Convection
There are two different convection schemes available in GENIE-IGCM: the Betts Miller and the Tiedtke. At their hearts the Betts Miller scheme "adjusts toward neutrality by moist convection without needing to explicitly simulate clouds" (Emanuel, 1994, page 538) whilst the Tiedtke scheme is a mass flux scheme that calculates fluxes of moisture, heat etc in a more 'process-based' approach. A good reference for explaining the different ways of modelling convection is Emanuel (1994).

The original convection scheme used in GENIE-IGCM is the Betts Miller scheme and was developed by Mike Blackburn. A description of this scheme can be found here: http://www.met.reading.ac.uk/~mike/betts_miller.html. The main parameter in the Betts Miller scheme is the relaxation timescale towards standard atmospheric profiles, which is chosen pragmatically with a typical value of around 2 hours. Shallow (non-precipitating) convection must satisfy an analytical profile, but the deep convective profile is based on observations and is therefore not thought to be suitable for climate modelling (in contrast to regional forcasting; Emanuel, 1994, page 539).

In response to poor precipitation values in a tuning exercise (Annan et al, 2005) the Betts Miller scheme is no longer the only convection scheme in GENIE-IGCM. The second convection scheme (implemented by David Cameron and with a cloud diagnostic added by John Hughes) is the Tiedtke convection scheme (Molteni, 2003). The Tiedtke scheme, as described in Molteni (2003) is a simplified mass-flux scheme that is triggered by conditional instability. Molteni (2003) use the convection scheme within the SPEEDY intermediate GCM and present analysis of the resulting climatology. Online documentation for the SPEEDY model can be found at: http://www.ictp.trieste.it/~moltenif/speedy-doc.html. This link points to a pdf that discusses the convection scheme in more detail. Initial results using the GENIE-IGCM-tiedtke (presented at GENIEfy meeting October 11th 2006) look promising and may dramatically improve the prediction of precipitation.

Having two different convection schemes is also interesting since it allows comparison studies in which the only change is a change of the convection scheme.

igcm3_diab |                                |                               dgrmlt |                          ______|__________________                          |          |              |                        convec/convec_tiedtke     lscrn |                 |                        |               ___|_____                        |              |         |    ____________________|_________  convmf    vdifsc |      |      |        |      |  dryadj cbcon  cbadj(1)  cudif  cubm

For the Tiedtke scheme the two subroutines are the convective mass flux scheme and the large-scale condensation scheme that is modelled as a relaxation of humidity towards a reference value. (1) - deep convection subroutine.

In IGCM3 there are two sets of divisions in the cloud scheme. The first group of cloud types is the set { convective,large-scale } ; Within convective rain there is deep and shallow convection. In large-scale rainfall there is { high,medium,low }. Both Betts-Miller and Tiedtke schemes use the same large-scale rainfall/cloud scheme. Large-scale rainfall is diagnosed in lscrn.

Both convection schemes use the same models to represent cloud cover. Deep convection cloud cover is proportional to the logarithm of the precipitation rate, whilst shallow convection cloud is set using a pre-defined constant fraction. Large-scale rainfall cloud is a linear function of relative humidity, above a specified threshold humidity.

Published work completed using the IGCM
Annan et al (2005) present the results of a tuning exercise completed using GENIE-IGCM with fixed SSTs. Whilst the tuning significantly improves the temperature field it fails to improve the precipitation field. This study was the motivation behind implementing the tiedtke convection scheme.

History of IGCM tuning

 * Annan et al tuning study using EnKF to tune IGCM (pre-tiedtke) to get surface fluxes.


 * December 2006. Tuning towards pre-industrial HadCM3 ocean surface fluxes of energy and moisture fluxes.  Using a GA based approach.  John Hughes setting up tuning targets and performing initial sensitivity analysis, Andy Price performing the main GA optimisation.

Known issues
Given that staff may be transient and knowledge of bugs, issues and model suitability for different kinds of studies are easilly lost, it is important to build a repository of known bugs and issues.


 * The formatting of output files could be improved (e.g. better inclusion of units). A critical development would be the inclusion of time-stamping the files (which could then be processed easily using such software as NC operators).  Specifically _rg_ files do not have time stamping, _cl_ does (integer year number) but only the year number and not the month number, and _an_ files report "1" for all years.
 * A preliminary result suggests that when the Tiedtke scheme is used in conjunction with the original land surface scheme (i.e. not genie-land) the model crashes.