ETKF Data Assimilation in WRFDA

The ETKF system updates the ensemble perturbations.

Please refer to Bishop et al. (2001) and Wang et al. (2003) for a detailed description of the methodology. 

a. Source Code

Three executables are used in the ETKF system:

WRFDA/var/build/gen_be_etkf.exe
   -Used to update the ensemble perturbations

WRFDA/var/build/gen_be_addmean.exe
   -Used to combine the emsemble mean (can be updated by Hybrid)

WRFDA/var/build/da_wrfvar.exe
   -Main WRFDA executable, used to create filtered observations and prepare formatted omb files for ETKF.

If one or more of these programs is missing, check your compilation log.

b. Running The ETKF System

A set of initial ensemble members must be prepared before the ETKF application can be started; this ensemble 
can be obtained from previous ensemble forecast. A tutorial case with a test ensemble can be found at 
http://www2.mmm.ucar.edu/wrf/users/wrfda/download/wrfda_hybrid_etkf_testdata.tar.gz. In this example, the 
ensemble forecasts were initialized at 2006102712 and valid 2006102800. ETKF will be performed using the 
ensemble valid 2006102800 as input. Once you have the initial ensemble, the ensemble perturbations can be 
updated by following the steps below:

  1)   Set an environment variable for your working directory and your data directory

  > setenv WORK_DIR_ETKF your_etkf_path
  > setenv DAT_DIR your_data_path
  > cd $WORK_DIR_ETKF

  2)   Prepare filtered observations

    a)   In your ETKF working directory, make a subdirectory to prepare the filtered observations and link 
         all the necessary files and directories as follows: 
	 
         > mkdir obs_filter
         > cd obs_filter
         > ln -fs $DAT_DIR/Hybrid/fc/2006102712/wrfout_d01_2006-10-28_00:00:00.mean ./fg  (first guess is the ensemble mean)
         > ln -fs $WRFDA_DIR/run/LANDUSE.TBL .
         > ln -fs $DAT_DIR/Hybrid/ob/2006102800/ob.ascii ./ob.ascii (or ob.bufr)
         > ln -fs $DAT_DIR/Hybrid/be/be.dat ./be.dat 
         > ln -fs $WRFDA_DIR/var/build/da_wrfvar.exe .
         > cp $DAT_DIR/ETKF/namelist.input .

    b)   Edit namelist.input, paying special attention to the following 'QC-OBS'-related settings:

         &wrfvar17
           analysis_type                       = 'QC-OBS',

    c)   Execute the WRFDA file, running in QC-OBS mode

         > ./da_wrfvar.exe >&! wrfda.log

    d)   Check the output files. The output file lists are the same as for running WRFDA-3DVAR, and 
         'filtered_obs_01' is a file containing the filtered observations.

  3)   Prepare omb files for ETKF

    a)   In your ETKF working directory, make a sub directory to prepare the omb files for each ensemble 
         member and link all the necessary files and directories as follows:
	 
         > cd $WORK_DIR_ETKF
         > mkdir -p omb/working.e001
         > cd omb/working.e001
         > ln -fs $DAT_DIR/Hybrid/fc/2006102712/wrfout_d01_2006-10-28_00:00:00.e001 ./fg  (first guess is the ensemble member) 
         > ln -fs $WRFDA_DIR/run/LANDUSE.TBL .
         > ln -fs $WORK_DIR_ETKF/obs_filter/filtered_obs_01 ./ob.ascii 
         > ln -fs $DAT_DIR/Hybrid/be/be.dat ./be.dat 
         > ln -fs $WRFDA_DIR/var/build/da_wrfvar.exe .
         > cp $DAT_DIR/ETKF/namelist.input .

    b)   Edit namelist.input, paying special attention to the following 'VERIFY'-related settings:

         &wrfvar17
           analysis_type                       = 'VERIFY',

    c)   Execute the WRFDA file, running in VERIFY mode

         > ./da_wrfvar.exe >&! wrfda.log

    d)   Check the output files. Again, the output file lists are the same as for running WRFDA-3DVAR/ The 
         'ob.etkf.0*' files contain the filtered observations.

    e)   Combine the ob.etkf.0* files and add the number index in the head of ob.etkf.e0*

         > cat ob.etkf.0* > ob.all
         > wc -l ob.all > ob.etkf.e001
         > cat ob.all >> ob.etkf.e001

    f)   Likewise, prepare ob.etkf.e0* files for other ensemble members

  4)   Run ETKF

    a)   Copy or link the ensemble mean and forecasts and ob.etkf.e0* files to your working directory.
         and make a directory to save the parameter files.
         > setenv PAR_DIR_ETKF your_parameter_path
         > cd $WORK_DIR_ETKF
         > ln -sf $DAT_DIR/Hybrid/fc/2006102712/wrfout_d01_2006-10-28_00:00:00.mean ./etkf_input
         > ln -sf $DAT_DIR/Hybrid/fc/2006102712/wrfout_d01_2006-10-28_00:00:00.e001 ./etkf_input.e001
           ...
         > ln -sf $DAT_DIR/Hybrid/fc/2006102712/wrfout_d01_2006-10-28_00:00:00.e010 ./etkf_input.e010

         > ln -sf omb/working.e001/ob.etkf.e001 .
           ...
         > ln -sf omb/working.e010/ob.etkf.e010 .

    b)   Provide template files. These files will be overwritten with the ensemble perturbations.

         > cp $DAT_DIR/Hybrid/fc/2006102712/wrfout_d01_2006-10-28_00:00:00.e001 ./etkf_output.e001
           ...
         > cp $DAT_DIR/Hybrid/fc/2006102712/wrfout_d01_2006-10-28_00:00:00.e010 ./etkf_output.e010

    c)   Copy gen_be_etkf_nl.nl (cp $DAT_DIR/ETKF/gen_be_etkf_nl.nl .) 
         You will need to set the information in this script as follows:

         &gen_be_etkf_nl
           num_members = 10,
           nv = 7,
           cv = 'U', 'V', 'W', 'PH', 'T', 'QVAPOR', 'MU',
           naccumt1 = 20,
           naccumt2 = 20,
           nstartaccum1 = 1,
           nstartaccum2 = 1,
           nout = 1,
           tainflatinput = 1,
           rhoinput = 1,
           infl_fac_file = '$PAR_DIR_ETKF/inflation_factor.dat',
           infl_let_file = '$PAR_DIR_ETKF/inflation_letkf.dat',
           eigen_val_file = '$PAR_DIR_ETKF/eigen_value.dat',
           inno2_val_file = '$PAR_DIR_ETKF/innovation_value.dat',
           proj2_val_file = '$PAR_DIR_ETKF/projection_value.dat',
           infl_fac_TRNK = .false.,
           infl_fac_WG03 = .false.,
           infl_fac_WG07 = .true.,
           infl_fac_BOWL = .false.,
           letkf_flg=.false.,
           rand_filt = .false.,
           rnd_seed = 2006102800,
           rnd_nobs = 5000
           etkf_erro_max = 20.
           etkf_erro_min = .00001
           etkf_inno_max = 20.
           etkf_inno_min = .00001
           etkf_erro_flg = .true.
           etkf_inno_flg = .true.
           etkf_wrfda = .false.

         'num_members' is the ensemble size (number of ensemble members)
         'nv' is the number of variables
         'cv' is a vector of strings containing the names of variables
         'naccumt1' and 'naccumt2' are number of previous cycles used to account for inflation and rho factor.
         'nstartaccum1' is the cycle from which naccumt1 cycle starts.
         'nstartaccum2' is the cycle from which naccumt2 cycle starts.
         'nout' is the cycle index. 
         'tainflatinput' and 'rhoinput' are prescribeld factors for inflation and rho factor.
         'infl_let_file', 'eigen_val_file', 'inno2_val_file', and 'proj2_val_file' are files to save template parameters
         'infl_fac_TRNK', 'infl_fac_WG03', 'nfl_fac_WG07', and 'nfl_fac_BOWL' are options for different adaptive 
           inflation schemes.
         'rand_filt', 'rnd_seed', and 'rnd_nobs' are options for using filtered and random observations.
         'etkf_erro_max', 'etkf_erro_min', 'etkf_inno_max', 'etkf_inno_min', 'etkf_erro_flg', 'etkf_inno_flg', and 
           'etkf_wrfda' are options to conduct further observation filtering. 

    d)   Link gen_be_etkf.exe to your working directory and run it.

         > ln -sf  $WRFDA_DIR/var/build/gen_be_etkf.exe . 
         > ./gen_be_etkf.exe

    e)   Check the output files. etkf_output.* files are updated ensemble pertubations.

 

  5)   Add updated ensemble perturbations to ensemble mean get new emsemble members
       > cd $WORK_DIR_ETKF

    a)   Copy add_mean_nl.nl (cp $DAT_DIR/ETKF/add_mean_nl.nl .) 
         You will need to set the information in this script as follows for each member:
         &add_mean_nl
           num_members = 10
           cv          = 'U', 'V', 'W', 'PH', 'T', 'QVAPOR', 'MU'
           nv          = 7
           path        = '$WORK_DIR_ETKF'
           file_mean   = 'etkf_input' 
           file_pert   = 'etkf_output.e001' (for each member, etkf_output.e0*...)

    b)   Run gen_add_mean.exe.
 
         > ln -sf WRFDA/var/build/gen_add_mean.exe  .
         > ./gen_add_mean.exe 

    c)   Check the output files. etkf_output.e0* files are the new ensemble members.