Dr. Gavin A. Schmidt

Project MUDDLE: Modularisation and Updating of Design, Diagnostics and Literature Effort

This is just a brief list of the main changes that we are planning to make. All suggestions are open to argument and any other suggestions will also be entertained. There are a couple of guiding ideas that should be kept in mind: we want to make the model more flexible, clearer, easier to modify and especially to make the model structure independent of the model physics and vice versa. To a large extent these aims should not be contradictory, however, there are also some limitations on what can be done due to a) the limited personnel and time available, and b) the need for some continuity for existing model users.

The areas that will be tackled fall into four main categories: the model, the diagnostics, the documentation and the runtime environment, and to a large extent they are independent of each other and can proceed in parallel. The model track will have to deal with the modularisation of the physics routines, removal of resolution dependent coding, etc. The diagnostics track will have to deal with the format, frequency and content of both the on-line output and the more efficient post-processing. The documentation aspect will have to deal with the cataloguing of all the variables, what they do, what units they are in, and the structure of the physics routines. This is best done by the people that wrote the code and so can be shared out. The standards for the documentation and the tools used to read and analyse are also being developed. The runtime environment aspect will deal with any changes to the methods of compilation, the setting of options and the execution of the model.

For each track, I have listed each of the suggested changes as either Level 1: absolutely necessary and uncontroversial, Level 2: optional but probably beneficial changes, and Level 3: changes that would either require a lot of work, or a big change in practice, and have benefits that may or may not make this worthwhile.

As parts of this list are achieved, they will be marked as DONE.

Model changes

    Level 1:
    • Everything compilable with FORTRAN 90 (DONE)
    • IMPLICIT NONE (no dimension statements) (DONE)
    • No -static requirements (DONE except radiation)
    • No explicit dependence on vertical/horizontal resolution
    • no magic numbers (DONE except radiation)
    • parameter statements for constants (DONE)
    • calculations should only be done once
    • include stratospheric model(DONE)/coupled model(DONE)/tracer model/parallelised code as standard add-on options.
    • Replacement of O/BL/G/GH/F-DATA arrays with meaningful names. (DONE)
    • atmospheric mass, P**kapa and P*T**kapa, should be standard model variables (DONE)
    • add JMONTH variable (DONE)
    • Lakes should be completely divorced from ocean variables (DONE)
    • Equivalence statements should be removed (DONE (except for radiation) )
    • saturation vapor pressure should be a standard centralized function (DONE)
    • a standard tri-diagonal solver included in the utilities (DONE)
    • standardization of looping variables and practice
    Level 2:
    • Use of modules instead of includes (DONE)
    • No references to all-model variables in local physics subroutines (DONE)
    • Force physics routines (incl. seaice) to only output fluxes (DONE)
    • Every physics routine should have an interface (no physics) + physics module (implies rewriting of SURFCE/GROUND (DONE)/PRECIP (DONE))
    • Separation of time-stepping features in MAIN from model overhead.
    • Common blocks should not be used to pass variables to physics routines, only to be used for common model variables.
    • standardize naming conventions(DONE)
    • removal of all ENTRY points (DONE except radiation)
    • no hard coding of array sizes or loop limits(DONE)
    • consistent initialization of radiation/ground/pbl etc. Possibly set as options in rundeck (DONE)
    • Paleo-year should be input variable (+automatic calculation/input of insolation parameters?)
    • Array indexing (for QUS moments, and large I,J,N arrays) should be optimized (DONE)
    • consistent use of QUS moments in pbl/diffus/radiation
    • advection of cloud liquid water mass, sub-gridscale pbl profiles
    Level 3:
    • Radiation code could be split to allow incremental addition of options (ghg/volcanic/aerosol/solar trends etc)
    • Use pre-processor options ( #ifdef - #endif directives)(DONE)
    • Use a specific file-manager system (a al GFDL) (DONE)
    • Re-institute partial land/ocean boxes for atmospheric model (DONE)
    • Include a flux-coupler for communication between different grids

Diagnostics

    Level 1:
    • netCDF/HDF output options (DONE)
    • self-contained diagnostic output files (i.e. boot-strapable acc files)
    • Option to change 'line printer' format for PRT files to standard text
    • include a surface pressure diagnostic
    • automatic averaging/collating of strings of acc files (DONE)
    Level 2:
    • No more unreferenced diagnostic array numbers (i.e. the accumulated sea level pressure diagnostic would be referenced as AIJ(I,J,IDIJSLP) or something similar. Requires a diagnostics module which sets all IDxxx numbers, readable by the post-processing routine.) Alternatively this could be replaced with a derived type for each group of diagnostics and direct references to fields witin them. (DONE for AJ/AIJ/AJL/AJK/ADIURN)
    • Easy definition and output of regional model diagnostics
    • Options for daily/6 hourly etc. diagnostics output
    • Options for control of diagnostic variables (DONE)
    • Extendable diagnostic routines to allow easy expansion for increases in ITYPEs (i.e. the budget pages) (DONE)
    • Direct inclusion of satellite diagnostics (i.e. MSU brightness, ICCP clouds, TOPEX SSH variability etc)
    Level 3:
    • Point and click interface for choosing/outputing and viewing standard model output

Documentation

    Level 1:
    • Self-documentation of every routine in a consistent format, including units (see appendix for details).(DONE)
    • An up-to-date manual on running the model (possibly derivable from the written specifications of this renovation project)
    • A comprehensive description of how the model works
    • Email listservers for model development issues (standards/functionality etc.), and for bug discoveries and fixes(DONE)
    Level 2:
    • A gcmdoc utility that reads all self-documentation and produces html reports (DONE)
    • a GISS-wide database that collects all pertinent info about simulations (with a short description).
    Level 3:

Run Environment

    Level 1:
    • use of makefiles to ensure up-to-date code (either in setup or as a separate make gcm step) (DONE)
    • Include declarations of compilation options in rundecks (DONE)
    • Month/day/year input should be used instead of Tau for model start/finish (DONE)
    • RANDSGI should be moved out of setup (DONE)
    Level 2:
    • Allow configuration files for diagnostic output and parameter changes (ie. make the NAMELISTs bigger and more sophisticated, and put them in separate files).(DONE)
    • Input all model code/rundecks into a real database (e.g. PostgreSQL) that is also web-accessible
    Level 3:
    • Implement standard revision control systems (CVS)(DONE)
    • Force files to only contain single routines, and abandon the 'priority' system currently used.
    • point and click interface for running standard model


gavin@isis