About the Programs

MLEF ALGORITHM FOR DATA ASSIMILATION AND MODEL ERROR ESTIMATION

The Maximum Likelihood Ensemble Filter (MLEF) data assimilation methodology is explained in:

• Zupanski, M., 2004: Maximum likelihood ensemble filter. Theoretical aspects (MWR)
• Zupanski, D., and M. Zupanski 2004: Model error estimation employing ensemble data assimilation approach. (MWR)

The algorithm explained below includes both the MLEF (mode) and sample EnKF (ensemble mean) options.

DIRECTORIES

Main directory is: $HOME/ensda Under this directory there are subdirectories with source files (../ensda/src), include files (../ensda/include), namelist files (../ensda/namelists), etc. User dependent scripts used to run the EnsDA jobs are in ../ensda/runscripts. All other scripts are kept in ../ensda/scripts.

COMPILING source codes: (Shallow-water model)

Main MODEL directory is $HOME/SWM_CSU. This directory has a structure as the original model directory. The difference is that now there is a need for TWO executables: SWM_fcst.x and SWM_get_init.x, both created in ../SWM_CSU/swm/compile.

Need to type:

• Make –f GNUMakefile_fcst => create SWM_fcst.x
• Make –f GNUMakefile_init => create SWM_get_init.x

Then, these executables should be copied to the main executable directory, usually /ptmp/usernm/exec/ensda

MLEF

To compile the rest of the MLEF codes copy $HOME/ensda/makefiles/Makefile_ensda.ibm to the executable directory ($work_root/$usernm/exec/ensda) and type:

• make -f Makefile_ensda.ibm

RUNNING MLEF Jobs

1. GENERATE simulated observations
   • To use model generated observations, one first needs to create observation files. This is done by running an interactive job script named runmakeobs, located in ../ensda/runscripts. In runmakeobs one needs to only change the usernm. Script runmakeobs is controlled by another shell script called envir_makeobs.sh which defines user-defined environmental variables. Most important variables are: cycle_start_date, N_obs_cycles, cycle_interval, etc. (explanations given in the job script itself).
2. Perform Data Assimlation/Ensemble Prediction
   • Job that performs data assimilation in cycled mode (one data assimilation cycle after another) is called runensda. As for the case of runmakeobs, only usernm needs to be changed in this script. More changes are needed in the script that controls runensda, that is called envir.sh. This script includes many user defined environmental variables, such as: cycle_start_date, START (COLD or WARM), N_cycles, cycle_interval, ens_size, etc. (explanations given in the script itself).
3. Post-processing (calculate RMS errors and innovation statistics)
   • LL script runpost needs to be submitted. Change the usernm, as before. Control job script is envir_post.sh, which includes user defined environmental variables.

1. runmakeobs - create model-produced "observations" = model + random perturbation
   • envir_makeobs.sh - environment variables, ("ideal" twin-model experiment only)
2. runnoobs - create no-observation forecast (what happens without data assimilation)
   • envir_noobs.sh - environment variables, ("ideal" twin-model experiment only)
3. runensda - Ensemble data assimilation/ensemble forecasting run. Includes postprocessing (statistics and RMS error calculation)
   • envir.sh - environment variables (any experiment - real or ideal)
4. runcov - Calculate analysis (posterior) error covariance matrix
   • envir_cov.sh - environment variables, used for specified window AFTER the data assimilation experiment (any experiment - real or ideal)
5. runpdf - Calculate innovation PDF histogram
   • envir_pdf.sh - environment variables, used for specified data assimilation cycles AFTER the data assimilation experiment (any experiment - real or ideal)

Go back to the main page | Go to Ensemble Site