Chapter 6: WRF Data Assimilation

Table of Contents

• Introduction
• Installing WRFDA for 3D-Var Run
• Installing WRFPLUS and WRFDA for 4D-Var Run
• Running Observation Preprocessor (OBSPROC)
• Running WRFDA
• Radiance Data Assimilations in WRFDA
• WRFDA Diagnostics
• Updating WRF Boundary Conditions
• Running gen_be
• WRFDA with Multivariate Background Error (MBE) Statistics
• Additional WRFDA Exercises
• Hybrid Data Assimilation
• Description of Namelist Variables

Introduction

Data assimilation is the technique by which observations are combined with a NWP product (the first guess or background forecast) and their respective error statistics to provide an improved estimate (the analysis) of the atmospheric (or oceanic, Jovian, whatever) state. Variational (Var) data assimilation achieves this through the iterative minimization of a prescribed cost (or penalty) function. Differences between the analysis and observations/first guess are penalized (damped) according to their perceived error. The difference between three-dimensional (3D-Var) and four-dimensional (4D-Var) data assimilation is the use of a numerical forecast model in the latter.

The MMM Division of NCAR supports a unified (global/regional, multi-model, 3/4D-Var) model-space data assimilation system (WRFDA) for use by NCAR staff and collaborators, and is also freely available to the general community, together with further documentation, test results, plans etc., from the WRFDA web-page http://www2.mmm.ucar.edu/wrf/users/wrfda/Docs/user_guide_V3.3/users_guide_chap6.htm.

Various components of the WRFDA system are shown in blue in the sketch below, together with their relationship with rest of the WRF system.

x^b: first guess either from a previous WRF forecast or from WPS/REAL output.

x^lbc: lateral boundary from WPS/REAL output.

x^a: analysis from WRFDA data assimilation system.

x^f: WRF forecast output.

y^o: observations processed by OBSPROC. (note: PREPBUFR input, Radar and Radiance data donÕt go through OBSPROC)

B₀: background error statistics from generic BE data (CV3) or gen_be.

R: observational and representative error statistics.

In this chapter, you will learn how to run the various components of the WRFDA system. For training purposes, you are supplied with a test case including the following input data: a) observation file (in the format prior to OBSPROC), b) WRF netCDF background file (WPS/REAL output used as a first guess of the analysis), and c) Background error statistics (estimate of errors in the background file). You can download the test dataset from http://www2.mmm.ucar.edu/wrf/users/wrfda/download/testdata.html. In your own work, you have to create all these input files yourselves. See the section Running Observation Preprocessor for creating your observation files. See section Running gen_be for generating your background error statistics file if you want to use cv_options=5 or cv_options=6.

Before using your own data, we suggest that you start by running through the WRFDA related programs at least once using the supplied test case. This serves two purposes: First, you can learn how to run the programs with data we have tested ourselves, and second you can test whether your computer is adequate to run the entire modeling system. After you have done the tutorial, you can try running other, more computationally intensive, case studies and experimenting with some of the many namelist variables.

WARNING: It is impossible to test every code upgrade with every permutation of computer, compiler, number of processors, case, namelist option, etc. The ÒnamelistÓ options that are supported are indicated in the ÒWRFDA/var/README.namelistÓ, and these are the default options.

Running with your own domain, hopefully, our test cases will have prepared you for the variety of ways in which you may wish to run WRFDA. Please inform us about your experiences.

As a professional courtesy, we request that you include the following reference in any publications that uses of any component of the community WRFDA system:

Barker, D.M., W. Huang, Y.R. Guo, and Q.N. Xiao., 2004: A Three-Dimensional (3DVAR) Data Assimilation System For Use With MM5: Implementation and Initial Results. Mon. Wea. Rev., 132, 897-914.

Huang, X.Y., Q. Xiao, D.M. Barker, X. Zhang, J. Michalakes, W. Huang, T. Henderson, J. Bray, Y. Chen, Z. Ma, J. Dudhia, Y. Guo, X. Zhang, D.J. Won, H.C. Lin, and Y.H. Kuo, 2009: Four-Dimensional Variational Data Assimilation for WRF: Formulation and Preliminary Results. Mon. Wea. Rev., 137, 299–314.

Running WRFDA requires a Fortran 90 compiler. We have currently tested the WRFDA on the following platforms: IBM (XLF), SGI Altix (INTEL), PC/Linux (PGI, INTEL, GFORTRAN), and Apple (G95/PGI). Please let us know if this does not meet your requirements, and we will attempt to add other machines to our list of supported architectures as resources allow. Although we are interested to hear of your experiences on modifying compile options, we do not yet recommend making changes to the configure file used to compile WRFDA.

Installing WRFDA for 3D-Var Run

a. Obtaining WRFDA Source Code

Users can download the WRFDA source code from http://www2.mmm.ucar.edu/wrf/users/wrfda/download/get_source.html.

After the tar file is unzipped (gunzip WRFDAV3.3.TAR.gz) and untarred (untar WRFDAV3.3.TAR), the directory WRFDA should be created; this directory contains the WRFDA source, external libraries, and fixed files. The following is a list of the system components and the content for each directory:

Directory Name	Content
var/da	WRFDA source code
var/run	Fixed input files required by WRFDA, such as background error covariance, and radiance related files CRTM coefficients, radiance_info and VARBC.in.
var/external	Library needed by WRFDA, include crtm, bufr, lapack, blas
var/obsproc	Obsproc source code , namelist, and observation error file.
var/gen_be	Source code of generate background error
var/build	Build all .exe files.

b. Compile WRFDA and Libraries

Starting from V3.1.1, some external libraries (for example, lapack, blas, and NCEP BUFR) are included in the WRFDA tar file. To compile the WRFDA code, it is necessary to have installed the netCDF library, which is the only mandatory library if only conventional observational data from LITTLE_R format file is to be used.

> setenv NETCDF your_netcdf_path

If observational data in PREPBUFR format are to be used, an environmental variable needs to be set like (using the C-shell),

> setenv BUFR 1

To have NCEP BUFR library compiled and have BUFR-related WRFDA code generated and compiled after configure/compile.

If satellite radiance data are to be used, in addition to NCEP BUFR library, RTM (Radiative Transfer Model) is required. The current RTM versions that WRFDA uses are CRTM V2.0.2 and RTTOV V10. WRFDA can compile with CRTM only, or RTTOV only, or both CRTM and RTTOV together.

Starting from V3.2.1, CRTM V2.0.2 is included in the WRFDA tar file.

> setenv CRTM 1

To have CRTM library compiled and CRTM-related WRFDA code generated and compiled after configure/compile.

If RTTOV v10 is the RTM to be used for radiance data assimilation, for the user should have downloaded and installed the RTTOV library before compiling WRFDA.

RTTOV v10 can be downloaded from http://research.metoffice.gov.uk/research/interproj/nwpsaf/rtm

After following the RTTOV documentation to compile the RTTOV library, set the RTTOV environment variable to the path where lib/librttov10.1.0_*.a reside.

> setenv RTTOV /usr/local/rttov10/pgi (in this example, there should exist /usr/local/rttov10/pgi/lib/librttov10.1.0_*.a)

Note: Make sure the required libraries were all compiled using the same compiler that will be used to build WRFDA, since the libraries produced by one compiler may not be compatible with code compiled with another.

Assuming all required libraries are available and the WRFDA source code is ready, start to build the WRFDA using the following steps:

To configure WRFDA, enter the WRFDA directory and type

> ./configure wrfda

A list of configuration options for your computer should appear. Each option combines a compiler type and a parallelism option; since the configuration script doesnÕt check which compilers are actually available, be sure to select only among the options for compilers that are available on your system. The parallelism option allows for a single-processor (serial) compilation, shared-memory parallel (smpar) compilation, distributed-memory parallel (dmpar) compilation and distributed-memory with shared-memory parallel (sm+dm) compilation. For example, on a Macintosh computer, the above steps look like:

> ./configure wrfda

checking for perl5... no

checking for perl... found /usr/bin/perl (perl)

Will use NETCDF in dir: /users/noname/work/external/g95/netcdf-3.6.1

PHDF5 not set in environment. Will configure WRF for use without.

$JASPERLIB or $JASPERINC not found in environment, configuring to build without grib2 I/O...

------------------------------------------------------------------------

Please select from among the following supported platforms.

1. Darwin (MACOS) PGI compiler with pgcc (serial)

2. Darwin (MACOS) PGI compiler with pgcc (smpar)

3. Darwin (MACOS) PGI compiler with pgcc (dmpar)

4. Darwin (MACOS) PGI compiler with pgcc (dm+sm)

5. Darwin (MACOS) intel compiler with icc (serial)

6. Darwin (MACOS) intel compiler with icc (smpar)

7. Darwin (MACOS) intel compiler with icc (dmpar)

8. Darwin (MACOS) intel compiler with icc (dm+sm)

9. Darwin (MACOS) intel compiler with cc (serial)

10. Darwin (MACOS) intel compiler with cc (smpar)

11. Darwin (MACOS) intel compiler with cc (dmpar)

12. Darwin (MACOS) intel compiler with cc (dm+sm)

13. Darwin (MACOS) g95 with gcc (serial)

14. Darwin (MACOS) g95 with gcc (dmpar)

15. Darwin (MACOS) xlf (serial)

16. Darwin (MACOS) xlf (dmpar)

Enter selection [1-10] : 13

------------------------------------------------------------------------

Compile for nesting? (0=no nesting, 1=basic, 2=preset moves, 3=vortex following) [default 0]:

Configuration successful. To build the model type compile .

ÉÉ

After running the configuration script and choosing a compilation option, a configure.wrf file will be created. Because of the variety of ways that a computer can be configured, if the WRFDA build ultimately fails, there is a chance that minor modifications to the configure.wrf file may be needed.

Note: WRF compiles with –r4 option while WRFDA compiles with –r8. For this reason, WRF and WRFDA cannot reside and be compiled under the same directory.

Hint: It is helpful to start with something simple, such as the serial build. If it is successful, move on to build dmpar code. Remember to type Ôclean –aÕ between each build.

To compile the code, type

> ./compile all_wrfvar >&! compile.out

Successful compilation of Ôall_wrfvarÓ will produce 42 executables in the var/build directory which are linked in var/da directory, as well as obsproc.exe in var/obsproc/src directory. You can list these executables by issuing the command (from WRFDA directory)

> ls -l var/build/*exe var/obsproc/src/obsproc.exe

-rwxr-xr-x 1 users 435816 Mar 9 19:26 var/build/da_advance_time.exe

-rwxr-xr-x 1 users 1195264 Mar 9 19:27 var/build/da_bias_airmass.exe

-rwxr-xr-x 1 users 815088 Mar 9 19:26 var/build/da_bias_scan.exe

-rwxr-xr-x 1 users 780476 Mar 9 19:26 var/build/da_bias_sele.exe

-rwxr-xr-x 1 users 1120408 Mar 9 19:26 var/build/da_bias_verif.exe

-rwxr-xr-x 1 users 1627284 Mar 9 19:26 var/build/da_rad_diags.exe

-rwxr-xr-x 1 users 639940 Mar 9 19:26 var/build/da_tune_obs_desroziers.exe

-rwxr-xr-x 1 users 608912 Mar 9 19:27 var/build/da_tune_obs_hollingsworth1.exe

-rwxr-xr-x 1 users 377748 Mar 9 19:27 var/build/da_tune_obs_hollingsworth2.exe

-rwxr-xr-x 1 users 1600636 Mar 9 19:27 var/build/da_update_bc.exe

-rwxr-xr-x 1 users 1662172 Mar 9 19:27 var/build/da_verif_grid.exe

-rwxr-xr-x 1 users 535916 Mar 9 19:32 var/build/da_verif_obs.exe

-rwxr-xr-x 1 users 29399039 Mar 9 19:32 var/build/da_wrfvar.exe

-rwxr-xr-x 1 users 2014440 Mar 9 19:32 var/build/gen_be_cov2d.exe

-rwxr-xr-x 1 users 2027684 Mar 9 19:27 var/build/gen_be_cov2d3d_contrib.exe

-rwxr-xr-x 1 users 2017952 Mar 9 19:27 var/build/gen_be_cov3d.exe

-rwxr-xr-x 1 users 2027804 Mar 9 19:27 var/build/gen_be_cov3d2d_contrib.exe

-rwxr-xr-x 1 users 2023396 Mar 9 19:27 var/build/gen_be_cov3d3d_bin3d_contrib.exe

-rwxr-xr-x 1 users 2027468 Mar 9 19:27 var/build/gen_be_cov3d3d_contrib.exe

-rwxr-xr-x 1 users 2003888 Mar 9 19:32 var/build/gen_be_diags.exe

-rwxr-xr-x 1 users 2028372 Mar 9 19:32 var/build/gen_be_diags_read.exe

-rwxr-xr-x 1 users 2012816 Mar 9 19:27 var/build/gen_be_ensmean.exe

-rwxr-xr-x 1 users 2045908 Mar 9 19:27 var/build/gen_be_ensrf.exe

-rwxr-xr-x 1 users 2069376 Mar 9 19:32 var/build/gen_be_ep1.exe

-rwxr-xr-x 1 users 2059240 Mar 9 19:32 var/build/gen_be_ep2.exe

-rwxr-xr-x 1 users 2022588 Mar 9 19:32 var/build/gen_be_etkf.exe

-rwxr-xr-x 1 users 2027480 Mar 9 19:27 var/build/gen_be_hist.exe

-rwxr-xr-x 1 users 2093900 Mar 9 19:32 var/build/gen_be_stage0_gsi.exe

-rwxr-xr-x 1 users 2105344 Mar 9 19:32 var/build/gen_be_stage0_wrf.exe

-rwxr-xr-x 1 users 2036928 Mar 9 19:32 var/build/gen_be_stage1.exe

-rwxr-xr-x 1 users 2064784 Mar 9 19:32 var/build/gen_be_stage1_1dvar.exe

-rwxr-xr-x 1 users 2036036 Mar 9 19:32 var/build/gen_be_stage1_gsi.exe

-rwxr-xr-x 1 users 2036024 Mar 9 19:32 var/build/gen_be_stage2.exe

-rwxr-xr-x 1 users 2100760 Mar 9 19:32 var/build/gen_be_stage2_1dvar.exe

-rwxr-xr-x 1 users 566584 Mar 9 19:26 var/build/gen_be_stage2_gsi.exe

-rwxr-xr-x 1 users 2023600 Mar 9 19:32 var/build/gen_be_stage2a.exe

-rwxr-xr-x 1 users 2036060 Mar 9 19:32 var/build/gen_be_stage3.exe

-rwxr-xr-x 1 users 2013852 Mar 9 19:32 var/build/gen_be_stage4_global.exe

-rwxr-xr-x 1 users 2049676 Mar 9 19:27 var/build/gen_be_stage4_regional.exe

-rwxr-xr-x 1 users 2003608 Mar 9 19:32 var/build/gen_be_vertloc.exe

-rwxr-xr-x 1 users 2155760 Mar 9 19:32 var/build/gen_mbe_stage2.exe

-rwxr-xr-x 1 users 1752352 Mar 23 09:29 var/obsproc/src/obsproc.exe

da_wrfvar.exe is the main executable for running WRFDA. Make sure it is created after the compilation. Sometimes (unfortunately) it is possible that other utilities get successfully compiled, while the main da_wrfvar.exe fails; please check the compilation log file carefully to figure out the problem.

The basic gen_be utility for regional model consists of gen_be_stage0_wrf.exe, gen_be_stage1.exe, gen_be_stage2.exe, gen_be_stage2a.exe, gen_be_stage3.exe, gen_be_stage4_regional.exe, and gen_be_diags.exe.

da_updated_bc.exe is used for updating WRF low and lateral boundary condition before and after a new WRFDA analysis is generated.

da_advance_time.exe is a very handy and useful tool for date/time manipulation. Type Òda_advance_time.exeÓ to see its usage instruction.

In addition to the executables for running WRFDA and gen_be, obsproc.exe (the executable for preparing conventional data for WRFDA) compilation is also included in Ò./compile all_wrfvarÓ. da_advance_time.exe

Go to WRFDA/var/external/bufr and WRFDA/var/external/crtm to check if the libbufr.a and libcrtm.a were generated if you use BUFR and CRTM library.

c. Clean Compilation

To remove all object files and executables, type:

clean

To remove all build files, including configure.wrfda, type:

clean -a

The 'clean –a' command is recommended if compilation fails or configuration file is changed.

Installing WRFPLUS and WRFDA for 4D-Var Run

If you intend to run WRF 4D-Var, it is necessary to have WRFPLUS installed. From V3.3, we release a new version of WRFDA and WRFPLUS for 4D-Var. WRFPLUS contains the adjoint and tangent linear models based on a simplified WRF model, which only include dry dynamic processes. We are developing the tangent linear and adjoint codes of several simplified physical packages.

To install WRFPLUS V3.3:

• Get the WRFPLUS zipped tar file from:

http://www2.mmm.ucar.edu/wrf/users/wrfda/download/wrfplus.html

• Unzip and untar the file to WRFPLUS

> gzip -cd WRFPLUS3.3.TAR.gz | tar -xf -

> cd WRFPLUS

> ./configure wrfplus

serial means single processor

dmpar wrfplus.exe means Distributed Memory Parallel (MPI)

Note: For Version 3.3 WRFDA 4D-Var, parallel run is still under development, please compile WRFPLUS3.3 with serial mode.

• Compile WRFPLUS

> ./compile em_real

> ls -ls main/*.exe

You should see wrf.exe

To install WRFDA for 4D-Var run,

• Before you install WRFDA to run 4D-Var, environment variable should to be set with,

>setenv WRFPLUS_DIR ${your_source_code_dir}/WRFPLUS

• If you intend to use observational data with PREPBUFR format or if you intend to assimilate satellite radiance data, you need set environment variables for BUFR, CRTM, or RTTOV. This procedure is the same as installing WRFDA for 3D-Var run.

>./configure 4dvar

Note: Please compile WRFDA for 4D-Var run with serial mode.

>./compile all_wrfvar

>ls -ls var/build/*.exe

You should see da_wrfvar.exe.

Running Observation Preprocessor (OBSPROC)

The OBSPROC program reads observations in LITTLE_R format (a legendary ASCII format, in use since the MM5 era). The LITTLE_R format is also used in OBSGRID program. Please refer to the documentation at http://www2.mmm.ucar.edu/mm5/mm5v3/data/how_to_get_rawdata.html and Chapter 7 of this UserÕs Guide for LITTLE_R format description. For your applications, you will have to prepare your own observation files. Please see http://www2.mmm.ucar.edu/mm5/mm5v3/data/free_data.html for the sources of some freely available observations and the program for converting the observations to LITTLE_R format. Because the raw observation data files could be in any of formats, such as ASCII, BUFR, PREPBUFR, MADIS, HDF, etc. Furthermore, for each of formats, there may be the different versions. To make WRFDA system as general as possible, the LITTLE_R format ASCII file was adopted as an intermediate observation data format for WRFDA system. Some extensions were made in the LITTLE_R format for WRFDA applications. More complete description of LITTLE_R format and conventional observation data sources for WRFDA could be found from the web page: 2010 Summer Tutorial by clicking ÒObservation Pre-processingÓ. The conversion of the user-specific-source data to the LITTLE_R format observation data file is the usersÕ task.

The purposes of OBSPROC are to:

á Remove observations outside the time range and domain (horizontal and top).

á Re-order and merge duplicate (in time and location) data reports.

á Retrieve pressure or height based on observed information using the hydrostatic assumption.

á Check vertical consistency and super adiabatic for multi-level observations.

á Assign observational errors based on a pre-specified error file.

á Write out the observation file to be used by WRFDA in ASCII or BUFR format.

The OBSPROC program—obsproc.exe should be found under the directory WRFDA/var/obsproc/src if Òcompile all_wrfvarÓ was completed successfully.

a. Prepare observational data for 3D-Var

To prepare the observation file, for example, at the analysis time 0h for 3D-Var, all the observations between ±1h (or ±1.5h) will be processed, as illustrated in the following figure, which means that the observations between 23h and 1h are treated as the observations at 0h.

Before running obsproc.exe, create the required namelist file namelist.obsproc (see WRFDA/var/obsproc/README.namelist, or the section Description of Namelist Variables for details.

For your reference, an example file named Ònamelist_obsproc.3dvar.wrfvar-tutÓ has already been created in the var/obsproc directory. Thus, proceed as follows.

> cp namelist.obsproc.3dvar.wrfvar-tut namelist.obsproc

Next, edit the namelist file namelist.obsproc by changing the following variables to accommodate your experiments.

&record1

obs_gts_filename='obs.2008020512'

&record2

time_window_min = '2008-02-05_11:00:00',: The earliest time edge as ccyy-mm-dd_hh:mn:ss

time_analysis = '2008-02-05_12:00:00', : The analysis time as ccyy-mm-dd_hh:mn:ss

time_window_max = '2008-02-05_13:00:00',: The latest time edge as ccyy-mm-dd_hh:mn:ss

&record6,7,8

Edit all the domain settings to conform to your own experiment. You may pay special attention to NESTIX and NESTJX, which are described in the section Description of Namelist Variables for details.

&record9

use_for = '3DVAR', ; used for 3D-Var, default

To run OBSPROC, type

> obsproc.exe >&! obsproc.out

Once obsproc.exe has completed successfully, you will see an observation data file, obs_gts_2008-02-05_12:00:00.3DVAR, in the obsproc directory. This is the input observation file to WRFDA.

obs_gts_2008-02-05_12:00:00.3DVAR is an ASCII file that contains a header section (listed below) followed by observations. The meanings and format of observations in the file are described in the last six lines of the header section.

TOTAL = 9066, MISS. =-888888.,

SYNOP = 757, METAR = 2416, SHIP = 145, BUOY = 250, BOGUS = 0, TEMP = 86,

AMDAR = 19, AIREP = 205, TAMDAR= 0, PILOT = 85, SATEM = 106, SATOB = 2556,

GPSPW = 187, GPSZD = 0, GPSRF = 3, GPSEP = 0, SSMT1 = 0, SSMT2 = 0,

TOVS = 0, QSCAT = 2190, PROFL = 61, AIRSR = 0, OTHER = 0,

PHIC = 40.00, XLONC = -95.00, TRUE1 = 30.00, TRUE2 = 60.00, XIM11 = 1.00, XJM11 = 1.00,

base_temp= 290.00, base_lapse= 50.00, PTOP = 1000., base_pres=100000., base_tropo_pres= 20000., base_strat_temp= 215.,

IXC = 60, JXC = 90, IPROJ = 1, IDD = 1, MAXNES= 1,

NESTIX= 60,

NESTJX= 90,

NUMC = 1,

DIS = 60.00,

NESTI = 1,

NESTJ = 1,

INFO = PLATFORM, DATE, NAME, LEVELS, LATITUDE, LONGITUDE, ELEVATION, ID.

SRFC = SLP, PW (DATA,QC,ERROR).

EACH = PRES, SPEED, DIR, HEIGHT, TEMP, DEW PT, HUMID (DATA,QC,ERROR)*LEVELS.

INFO_FMT = (A12,1X,A19,1X,A40,1X,I6,3(F12.3,11X),6X,A40)

SRFC_FMT = (F12.3,I4,F7.2,F12.3,I4,F7.3)

EACH_FMT = (3(F12.3,I4,F7.2),11X,3(F12.3,I4,F7.2),11X,3(F12.3,I4,F7.2))

#------------------------------------------------------------------------------#

ÉÉ observations ÉÉÉ

Before running WRFDA, you may find it useful to learn more about various types of data that will be processed to WRFDA, e.g., their geographical distribution. This file is in ASCII format and so you can easily view it. For a graphical view about file's content, use the ÒMAP_plotÓ utility to see the data distribution for each type of observations. To use this utility, proceed as follows.

> cd MAP_plot

> make

We have prepared some configure.user.ibm/linux/mac/É files for some platforms, when ÒmakeÓ is typed, the Makefile will use one of them to determine the compiler and compiler option. Please modify the Makefile and configure.user.xxx to accommodate the complier on your platform. Successful compilation will produce Map.exe. Note: The successful compilation of Map.exe requires pre-installed NCARG Graphics libraries under $(NCARG_ROOT)/lib.

Modify the script Map.csh to set the time window and full path of input observation file (obs_gts_2008-02-05_12:00:00.3DVAR). You will need to set the following strings in this script as follows:

Map_plot = /users/noname/WRFDA/var/obsproc/MAP_plot

TIME_WINDOW_MIN = Ô2008020511Õ

TIME_ANALYSIS = Ô2008020512Õ

TIME_WINDOW_MAX = Ô2008020513Õ
OBSDATA = ../obs_gts_2008-02-05_12:00:00.3DVAR

Next, type

> Map.csh

When the job has completed, you will have a gmeta file gmeta.{analysis_time} corresponding to analysis_time=2008020512. This contains plots of data distribution for each type of observations contained in the OBS data file: obs_gts_2008-02-05_12:00:00.3DVAR. To view this, type

> idt gmeta.2008020512

It will display (panel by panel) geographical distribution of various types of data. The following graphic shows the geographic distribution of ÒsondeÓ observations for this case.

An alternative way to plot the observation is to use ncl script: WRFDA/var/graphics/ncl/plot_ob_ascii_loc.ncl. However, with this method, you need to provide the first guess file to the ncl script, and have ncl installed in your system.

b. Prepare observational data for 4D-Var

To prepare the observation file, for example, at the analysis time 0h for 4D-Var, all observations from 0h to 6h will be processed and grouped in 7 sub-windows from slot1 to slot7, as illustrated in following figure. NOTE: The ÒAnalysis timeÓ in the figure below is not the actual analysis time (0h), it indicates the time_analysis setting in the namelist file and is set to three hours later than the actual analysis time. The actual analysis time is still 0h.

An example file named Ònamelist_obsproc.4dvar.wrfvar-tutÓ has already been created in the var/obsproc directory. Thus, proceed as follows:

> cp namelist.obsproc.4dvar.wrfvar-tut namelist.obsproc

In the namelist file, you need to change the following variables to accommodate your experiments. In this test case, the actual analysis time is 2008-02-05_12:00:00, but in namelist, the time_analysis should be set to 3 hours later. The different value of time_analysis will make the different number of time slots before and after time_analysis. For example, if you set time_analysis = 2008-02-05_16:00:00, and set the num_slots_past = 4 and time_slots_ahead=2. The final results will be the same as before.

&record1

obs_gts_filename='obs.2008020512'

&record2

time_window_min = '2008-02-05_12:00:00',: The earliest time edge as ccyy-mm-dd_hh:mn:ss

time_analysis = '2008-02-05_15:00:00', : The analysis time as ccyy-mm-dd_hh:mn:ss

time_window_max = '2008-02-05_18:00:00',: The latest time edge as ccyy-mm-dd_hh:mn:ss

&record6,7,8

Edit all the domain settings according to your own experiment. You may pay special attention to NESTIX and NESTJX, which is described in the section Description of Namelist Variables for details.

&record9

use_for = '4DVAR', ; used for 3D-Var, default

; num_slots_past and num_slots_ahead are used ONLY for FGAT and 4DVAR:

num_slots_past = 3, ; the number of time slots before time_analysis

num_slots_ahead = 3, ; the number of time slots after time_analysis

To run OBSPROC, type

> obsproc.exe >&! obsproc.out

Once obsproc.exe has completed successfully, you will see 7 observation data files:

obs_gts_2008-02-05_12:00:00.4DVAR

obs_gts_2008-02-05_13:00:00.4DVAR

obs_gts_2008-02-05_14:00:00.4DVAR

obs_gts_2008-02-05_15:00:00.4DVAR

obs_gts_2008-02-05_16:00:00.4DVAR

obs_gts_2008-02-05_17:00:00.4DVAR

obs_gts_2008-02-05_18:00:00.4DVAR

They are the input observation files to WRF 4D-Var. You can also use ÒMAP_PlotÓ to view the geographic distribution of different observations at different time slots.

Running WRFDA

a. Download Test Data

The WRFDA system requires three input files to run:

a) A WRF first guess and boundary input files output from either WPS/real (cold-start)

or WRF forecast (warm-start)

b) Observations (in ASCII format, PREPBUFR or BUFR for radiance)

c) A background error statistics file (containing background error covariance)

The following table summarizes the above info:

Input Data	Format	Created By
First Guess	NETCDF	WRF Preprocessing System (WPS) and real.exe or WRF
Observations	ASCII (PREPBUFR also possible)	Observation Preprocessor (OBSPROC)
Background Error Statistics	Binary	WRFDA gen_be utility /Default CV3

In the test case, you will store data in a directory defined by the environment variable $DAT_DIR. This directory can be at any location, and it should have read access.  Type

> setenv DAT_DIR your_choice_of_dat_dir

Here, "your_choice_of_dat_dir" is the directory where the WRFDA input data is stored. If it does not exist, create this directory by typing

> cd $DAT_DIR

Download the test data for a ÒTutorialÓ case valid at 12 UTC 5^th February 2008 from http://www2.mmm.ucar.edu/wrf/users/wrfda/download/testdata.html

Once you have downloaded ÒWRFDAV3.3-testdata.tar.gzÓ file to $DAT_DIR, extract it by typing

> gunzip WRFDAV3.3-testdata.tar.gz
> tar -xvf WRFDAV3.3-testdata.tar

Now you should find the following three sub-directories/files under Ò$DAT_DIRÓ

ob/2008020512/ob.2008020512.gz # Observation data in Òlittle_rÓ format
rc/2008020512/wrfinput_d01 # First guess file
rc/2008020512/wrfbdy_d01 # lateral boundary file
be/be.dat # Background error file

......

You should first go through the section ÒRunning Observation Preprocessor (OBSPROC)Ó and have a WRF-3D-Var-ready observation file (obs_gts_2008-02-05_12:00:00.3DVAR) generated in your OBSPROC working directory. You could then copy or move obs_gts_2008-02-05_12:00:00.3DVAR to be in $DAT_DIR/ob/2008020512/ob.ascii.

If you want to try 4D-Var with Little-R format observations, please go through the section ÒRunning Observation Preprocessor (OBSPROC)Ó and have the WRF-4D-Var-ready observation files (obs_gts_2008-02-05_12:00:00.4DVAR,ÉÉ). You could copy or move the observation files to $DAT_DIR/ob using following commands:

> mv obs_gts_2008-02-05_12:00:00.4DVAR $DAT_DIR/ob/2008020512/ob.ascii+

> mv obs_gts_2008-02-05_13:00:00.4DVAR $DAT_DIR/ob/2008020513/ob.ascii

> mv obs_gts_2008-02-05_14:00:00.4DVAR $DAT_DIR/ob/2008020514/ob.ascii

> mv obs_gts_2008-02-05_15:00:00.4DVAR $DAT_DIR/ob/2008020515/ob.ascii

> mv obs_gts_2008-02-05_16:00:00.4DVAR $DAT_DIR/ob/2008020516/ob.ascii

> mv obs_gts_2008-02-05_17:00:00.4DVAR $DAT_DIR/ob/2008020517/ob.ascii

> mv obs_gts_2008-02-05_18:00:00.4DVAR $DAT_DIR/ob/2008020518/ob.ascii-

At this point you have three of the input files (first guess, observation, and background error statistics files in directory $DAT_DIR) required to run WRFDA, and have successfully downloaded and compiled the WRFDA code. If this is correct, you are ready to learn how to run WRFDA.

b. Run the Case—3D-Var

The data for this case is valid at 12 UTC 5^th February 2008. The first guess comes from the NCEP FNL (Final) Operational Global Analysis data, passed through the WRF-WPS and real programs.

To run WRF 3D-Var, first create and cd to a working directory, for example, WRFDA/var/test/tutorial, and then follow the steps below:

> cd WRFDA/var/test/tutorial

> ln -sf WRFDA/run/LANDUSE.TBL ./LANDUSE.TBL

> ln -sf $DAT_DIR/rc/2008020512/wrfinput_d01 ./fg (link first guess file as fg)

> ln -sf WRFDA/var/obsproc/obs_gts_2008-02-05_12:00:00.3DVAR ./ob.ascii (link OBSPROC processed observation file as ob.ascii)

> ln -sf $DAT_DIR/be/be.dat ./be.dat (link background error statistics as be.dat)

> ln -sf WRFDA/var/da/da_wrfvar.exe ./da_wrfvar.exe (link executable)

If you use PREPBUFR format data, please change the ob_format=1 in &wrfvar3 in namelist.input and link the data as ob.bufr,

> ln -fs $DATA_DIR/ob/2008020512/gds1.t12.prepbufr.nr ob.bufr

We will begin by editing the file, namelist.input, which is a very basic namelist.input for running the tutorial test case as shown below and provided as WRFDA/var/test/tutorial/namelist.input. Only the time and domain settings need to be specified in this case, if we are using the default settings provided in WRFDA/Registry/Registry.wrfvar)

&wrfvar1

print_detail_grad=false,
/
&wrfvar2
/

&wrfvar3

/

&wrfvar4

/

&wrfvar5

/

&wrfvar6

/

&wrfvar7

/

&wrfvar8

/

&wrfvar9

/

&wrfvar10

/

&wrfvar11

calculate_cg_cost_fn=.false.

/

&wrfvar12

/

&wrfvar13

/

&wrfvar14

/

&wrfvar15

/

&wrfvar16

/

&wrfvar17

/

&wrfvar18

analysis_date="2008-02-05_12:00:00.0000",

/

&wrfvar19

/

&wrfvar20

/

&wrfvar21

time_window_min="2008-02-05_11:00:00.0000",

/

&wrfvar22

time_window_max="2008-02-05_13:00:00.0000",

/

&wrfvar23

/

&time_control

start_year=2008,

start_month=02,

start_day=05,

start_hour=12,

end_year=2008,

end_month=02,

end_day=05,

end_hour=12,

/

&dfi_control

/

&domains

e_we=90,

e_sn=60,

e_vert=41,

dx=60000,

dy=60000,

/

&physics

mp_physics=3,

ra_lw_physics=1,

ra_sw_physics=1,

radt=60,

sf_sfclay_physics=1,

sf_surface_physics=1,

bl_pbl_physics=1,

cu_physics=1,

cudt=5,

num_soil_layers=5, (IMPORTANT: itÕs essential to make sure the setting here is consistent with the number in your first guess file)

mp_zero_out=2,

co2tf=0,

/

&fdda

/

&dynamics

/

&bdy_control

/

&grib2

/

&namelist_quilt

/

&perturbation

/

> da_wrfvar.exe >&! wrfda.log

The file wrfda.log (or rsl.out.0000 if run in distributed-memory mode) contains important WRFDA runtime log information. Always check the log after a WRFDA run:

*** VARIATIONAL ANALYSIS ***

DYNAMICS OPTION: Eulerian Mass Coordinate

alloc_space_field: domain 1, 606309816 bytes allocat

ed

WRF TILE 1 IS 1 IE 89 JS 1 JE 59

WRF NUMBER OF TILES = 1

Set up observations (ob)

Using ASCII format observation input

scan obs ascii

end scan obs ascii

Observation summary

ob time 1

sound 86 global, 86 local

synop 757 global, 750 local

pilot 85 global, 85 local

satem 106 global, 105 local

geoamv 2556 global, 2499 local

polaramv 0 global, 0 local

airep 224 global, 221 local

gpspw 187 global, 187 local

gpsrf 3 global, 3 local

metar 2416 global, 2408 local

ships 145 global, 140 local

ssmi_rv 0 global, 0 local

ssmi_tb 0 global, 0 local

ssmt1 0 global, 0 local

ssmt2 0 global, 0 local

qscat 2190 global, 2126 local

profiler 61 global, 61 local

buoy 247 global, 247 local

bogus 0 global, 0 local

pseudo 0 global, 0 local

radar 0 global, 0 local

radiance 0 global, 0 local

airs retrieval 0 global, 0 local

sonde_sfc 86 global, 86 local

mtgirs 0 global, 0 local

tamdar 0 global, 0 local

Set up background errors for regional application for cv_options = 5

Using the averaged regression coefficients for unbalanced part

WRF-Var dry control variables are:psi, chi_u, t_u and ps_u

Humidity control variable is rh

Vertical truncation for psi = 15( 99.00%)

Vertical truncation for chi_u = 20( 99.00%)

Vertical truncation for t_u = 29( 99.00%)

Vertical truncation for rh = 22( 99.00%)

Scaling: var, len, ds: 0.100000E+01 0.100000E+01 0.600000E+05

Scaling: var, len, ds: 0.100000E+01 0.100000E+01 0.600000E+05

Scaling: var, len, ds: 0.100000E+01 0.100000E+01 0.600000E+05

Scaling: var, len, ds: 0.100000E+01 0.100000E+01 0.600000E+05

Scaling: var, len, ds: 0.100000E+01 0.100000E+01 0.600000E+05

Calculate innovation vector(iv)

Minimize cost function using CG method

Starting outer iteration : 1

Starting cost function: 2.53214888D+04, Gradient= 2.90675545D+02

For this outer iteration gradient target is: 2.90675545D+00

----------------------------------------------------------

Iter Cost Function Gradient Step

1 2.32498037D+04 2.55571188D+02 4.90384516D-02

2 2.14988144D+04 2.22354203D+02 5.36154186D-02

3 2.01389088D+04 1.62537907D+02 5.50108123D-02

4 1.93433827D+04 1.26984567D+02 6.02247687D-02

5 1.88877194D+04 9.84565874D+01 5.65160951D-02

6 1.86297777D+04 7.49071361D+01 5.32184146D-02

7 1.84886755D+04 5.41516421D+01 5.02941363D-02

8 1.84118462D+04 4.68329312D+01 5.24003071D-02

9 1.83485166D+04 3.53595537D+01 5.77476335D-02

10 1.83191278D+04 2.64947070D+01 4.70109040D-02

11 1.82984221D+04 2.06996271D+01 5.89930206D-02

12 1.82875693D+04 1.56426527D+01 5.06578447D-02

13 1.82807224D+04 1.15892153D+01 5.59631997D-02

14 1.82773339D+04 8.74778514D+00 5.04582959D-02

15 1.82751663D+04 7.22150257D+00 5.66521675D-02

16 1.82736284D+04 4.81374868D+00 5.89786400D-02

17 1.82728636D+04 3.82286871D+00 6.60104384D-02

18 1.82724306D+04 3.16737517D+00 5.92526480D-02

19 1.82721735D+04 2.23392283D+00 5.12604438D-02

----------------------------------------------------------

Inner iteration stopped after 19 iterations

Final: 19 iter, J= 1.98187399D+04, g= 2.23392283D+00

----------------------------------------------------------

Diagnostics

Final cost function J = 19818.74

Total number of obs. = 39800

Final value of J = 19818.73988

Final value of Jo = 16859.85861

Final value of Jb = 2958.88127

Final value of Jc = 0.00000

Final value of Je = 0.00000

Final value of Jp = 0.00000

Final value of Jl = 0.00000

Final J / total num_obs = 0.49796

Jb factor used(1) = 1.00000 1.00000 1.00000 1.00000 1.00000

1.00000 1.00000 1.00000 1.00000 1.00000

Jb factor used(2) = 1.00000 1.00000 1.00000 1.00000 1.00000

1.00000 1.00000 1.00000 1.00000 1.00000

Jb factor used(3) = 1.00000 1.00000 1.00000 1.00000 1.00000

1.00000 1.00000 1.00000 1.00000 1.00000

Jb factor used(4) = 1.00000 1.00000 1.00000 1.00000 1.00000

1.00000 1.00000 1.00000 1.00000 1.00000

Jb factor used(5) = 1.00000 1.00000 1.00000 1.00000 1.00000

1.00000 1.00000 1.00000 1.00000 1.00000

Jb factor used = 1.00000

Je factor used = 1.00000

VarBC factor used = 1.00000

*** WRF-Var completed successfully ***

A file called namelist.output (which contains the complete namelist settings) will be generated after a successful da_wrfvar.exe run. The settings appearing in namelist.output, but not specified in your namelist.input, are the default values from WRFDA/Registry/Registry.wrfvar.

After successful completion of job, wrfvar_output (the WRFDA analysis file, i.e. the new initial condition for WRF) should appear in the working directory along with a number of diagnostic files. Various text diagnostics output files will be explained in the next section (WRFDA Diagnostics).

To understand the role of various important WRFDA options, try re-running WRFDA by changing different namelist options, for example, making WRFDA convergence criteria more stringent. This is achieved by reducing the value of the convergence criteria ÒEPSÓ to e.g. 0.0001 by adding "EPS=0.0001" in the namelist.input record &wrfvar6. See section (WRFDA additional exercises) for more namelist options.

c. Run the Case—4D-Var

To run WRF 4D-Var, first create and enter into a working directory, for example, WRFDA/var/test/4dvar.

Note: If you want to setup your own directories to run 4D-Var, please make sure you follow the linkages and namelist.input under WRFDA/var/test/4dvar.

Assume the analysis date is 2008020512 and the test data directories are:

> setenv DATA_DIR /ptmp/$user/DATA

> ls –lr $DATA_DIR

ob/2008020512

ob/2008020513

ob/2008020514

ob/2008020515

ob/2008020516

ob/2008020517

ob/2008020518

rc/2008020512

be

Note: WRFDA 4D-Var is able to assimilate conventional observation data, satellites radiance BUFR data, radar data, and the input data format can be PREPBUFR format data or observation data processed by OBSPROC.

Assume the working directory is:

> setenv WORK_DIR $WRFDA_DIR/var/test/4dvar

Then follow the steps below:

1) Link the executables.

> cd $WORK_DIR

> ln -fs $WRFDA_DIR/var/da/da_wrfvar.exe .

2) Link the observational data, first guess, BE and LANDUSE.TBL, etc..

> cd $WORK_DIR

> ln -fs $DATA_DIR/ob/2008020512/ob.ascii+ ob01.ascii

> ln -fs $DATA_DIR/ob/2008020513/ob.ascii ob02.ascii

> ln -fs $DATA_DIR/ob/2008020514/ob.ascii ob03.ascii

> ln -fs $DATA_DIR/ob/2008020515/ob.ascii ob04.ascii

> ln -fs $DATA_DIR/ob/2008020516/ob.ascii ob05.ascii

> ln -fs $DATA_DIR/ob/2008020517/ob.ascii ob06.ascii

> ln -fs $DATA_DIR/ob/2008020518/ob.ascii- ob07.ascii

> ln -fs $DATA_DIR/rc/2008020512/wrfinput_d01 .

> ln -fs $DATA_DIR/rc/2008020512/wrfbdy_d01 .

> ln -fs wrfinput_d01 fg

> ln -fs $DATA_DIR/be/be.dat .

> ln -fs WRFDA/run/LANDUSE.TBL ./LANDUSE.TBL

> ln -fs WRFDA/run/GENPARM.TBL ./GENPARM.TBL

> ln -fs WRFDA/run/SOILPARM.TBL ./SOILPARM.TBL

> ln -fs WRFDA/run/VEGPARM.TBL ./VEGPARM.TBL

> ln –fs WRFDA/run/RRTM_DATA_DBL RRTM_DATA

If you use PREPBUFR format data, please change the ob_format=1 in &wrfvar3 in namelist.input and link the data as ob.bufr,

> ln -fs $DATA_DIR/ob/2008020512/gds1.t12.prepbufr.nr ob.bufr

If you would like to assimilate PREPBUFR data at both12hr and 18hr for 4D-Var, you should linked it as follows,

> ln -fs $DATA_DIR/ob/2008020512/gds1.t12.prepbufr.nr ob01.bufr

> ln -fs $DATA_DIR/ob/2008020518/gds1.t18.prepbufr.nr ob02.bufr

Note: NCEP BUFR files downloaded from NCEPÕs public ftp server ftp://ftp.ncep.noaa.gov/pub/data/nccf/com/gfs/prod/gdas.${yyyymmddhh} are Fortran-blocked on big-endian machine and can be directly used on big-endian machines (for example, IBM). For most Linux clusters with Intel platforms, users need to download the byte-swapping code ssrc.c (http://www.dtcenter.org/com-GSI/users/support/faqs/index.php, the C code ssrc.c located in the /utils directory of the GSI distribution), and this code will convert a prepbufr file generated on an IBM platofrm to a prepbufr file that can be read on a Linux or Intel Mac platform. Compile ssrc.c with any c compiler (e.g., gcc -o ssrc.exe ssrc.c). To convert an IBM prepbufr file, take the executable (e.g. ssrc.exe), and run it as follows,

ssrc.exe <name of Big Endian prepbufr file> name of Little Endian prepbufr file

3) Run in single processor mode (serial compilation required for WRF 4D-Var)

Edit $WORK_DIR/namelist.input to match your experiment settings. The most important namelist variables related to 4D-Var are listed below, please refer to README.namelist under WRFDA/var directory.

&wrfvar1

var4d=true,

var4d_lbc=true,

var4d_bin=3600,

ÉÉ
/

ÉÉ

&perturbation

trajectory_io=true,

enable_identity=false,

jcdfi_use=true,

jcdfi_diag=1,

jcdfi_penalty=1000.0,

/

> cd $WORK_DIR

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

Radiance Data Assimilations in WRFDA

This section gives a brief description for various aspects related to radiance assimilation in WRFDA. Each aspect is described mainly from the viewpoint of usage rather than more technical and scientific details, which will appear in separated technical report and scientific paper. Namelist parameters controlling different aspects of radiance assimilation will be detailed in the following sections. It should be noted that this section does not cover general aspects of the WRFDA assimilation. These can be found in other sections of chapter 6 of this users guide or other WRFDA documentation.

a. Running WRFDA with radiances

In addition to the basic input files (LANDUSE.TBL, fg, ob.ascii, be.dat) mentioned in ÒRunning WRFDAÓ section, the following additional files are required for radiances: radiance data in NCEP BUFR format, radiance_info files, VARBC.in, RTM (CRTM or RTTOV) coefficient files.

Edit namelist.input (Pay special attention to &wrfvar4, &wrfvar14, &wrfvar21, and &wrfvar22 for radiance-related options. A very basic namelist.input for running the radiance test case is provided as WRFDA/var/test/radiance/namelist.input)

> ln -sf ${DAT_DIR}/gdas1.t00z.1bamua.tm00.bufr_d ./amsua.bufr

> ln -sf ${DAT_DIR}/gdas1.t00z.1bamub.tm00.bufr_d ./amsub.bufr

> ln -sf WRFDA/var/run/radiance_info ./radiance_info # (radiance_info is a directory)

> ln -sf WRFDA/var/run/VARBC.in ./VARBC.in

(CRTM only) > ln -sf WRFDA/var/run/crtm_coeffs ./crtm_coeffs #(crtm_coeffs is a directory)

(RTTOV only) > ln -sf your_path/rtcoef_rttov10/rttov7pred51L ./rttov_coeffs # (rttov_coeffs is a directory)

See the following sections for more details on each aspect.

b. Radiance Data Ingest

Currently, the ingest interface for NCEP BUFR radiance data is implemented in WRFDA. The radiance data are available through NCEPÕs public ftp server ftp://ftp.ncep.noaa.gov/pub/data/nccf/com/gfs/prod/gdas.${yyyymmddhh} in near real-time (with 6-hour delay) and can meet requirements both for research purposes and some real-time applications.

So far, WRFDA can read data from the NOAA ATOVS instruments (HIRS, AMSU-A, AMSU-B and MHS), the EOS Aqua instruments (AIRS, AMSU-A) and DMSP instruments (SSMIS). Note that NCEP radiance BUFR files are separated by instrument names (i.e., each file for one type instrument), and each file contains global radiance (generally converted to brightness temperature) within 6-hour assimilation window from multi-platforms. For running WRFDA, users need to rename NCEP corresponding BUFR files (table 1) to hirs3.bufr (including HIRS data from NOAA-15/16/17), hirs4.bufr (including HIRS data from NOAA-18/19, METOP-2), amsua.bufr (including AMSU-A data from NOAA-15/16/18/19, METOP-2), amsub.bufr (including AMSU-B data from NOAA-15/16/17), mhs.bufr (including MHS data from NOAA-18/19 and METOP-2), airs.bufr (including AIRS and AMSU-A data from EOS-AQUA) and ssmis.bufr (SSMIS data from DMSP-16, AFWA provided) for WRFDA filename convention. Note that airs.bufr file contains not only AIRS data but also AMSU-A, which is collocated with AIRS pixels (1 AMSU-A pixels collocated with 9 AIRS pixels). Users must place these files in the working directory where WRFDA executable is run. It should also be mentioned that WRFDA reads these BUFR radiance files directly without use if any separate pre-processing program is used. All processing of radiance data, such as quality control, thinning and bias correction and so on, is carried out inside WRFDA. This is different from conventional observation assimilation, which requires a pre-processing package (OBSPROC) to generate WRFDA readable ASCII files. For reading the radiance BUFR files, WRFDA must be compiled with the NCEP BUFR library (see http://www.nco.ncep.noaa.gov/sib/decoders/BUFRLIB/).

Table 1: NCEP and WRFDA radiance BUFR file naming convention

NCEP BUFR file names	WRFDA naming convention
gdas1.t00z.1bamua.tm00.bufr_d	amsua.bufr
gdas1.t00z.1bamub.tm00.bufr_d	amsub.bufr
gdas1.t00z.1bhrs3.tm00.bufr_d	hirs3.bufr
gdas1.t00z.1bhrs4.tm00.bufr_d	hirs4.bufr
gdas1.t00z.1bmhs.tm00.bufr_d	mhs.bufr
gdas1.t00z.airsev.tm00.bufr_d	airs.bufr

Namelist parameters are used to control the reading of corresponding BUFR files into WRFDA. For instance, USE_AMSUAOBS, USE_AMSUBOBS, USE_HIRS3OBS, USE_HIRS4OBS, USE_MHSOBS, USE_AIRSOBS, USE_EOS_AMSUAOBS and USE_SSMISOBS control whether or not the respective file is read. These are logical parameters that are assigned to .FALSE. by default; therefore they must be set to .TRUE. to read the respective observation file. Also note that these parameters only control whether the data is read, not whether the data included in the files is to be assimilated. This is controlled by other namelist parameters explained in the next section.

NCEP BUFR files downloaded from NCEPÕs public ftp server ftp://ftp.ncep.noaa.gov/pub/data/nccf/com/gfs/prod/gdas.${yyyymmddhh} are Fortran-blocked on big-endian machine and can be directly used on big-endian machines (for example, IBM). For most Linux clusters with Intel platforms, users need to download the byte-swapping code ssrc.c (http://www.dtcenter.org/com-GSI/users/support/faqs/index.php, the C code ssrc.c located in the /utils directory of the GSI distribution), and this code will convert a prepbufr file generated on an IBM platofrm to a prepbufr file that can be read on a Linux or Intel Mac platform. Compile ssrc.c with any c compiler (e.g., gcc -o ssrc.exe ssrc.c). To convert an IBM prepbufr file, take the executable (e.g. ssrc.exe), and run it as follows,

ssrc.exe <name of Big Endian prepbufr file> name of Little Endian prepbufr file

c. Radiative Transfer Model

The core component for direct radiance assimilation is to incorporate a radiative transfer model (RTM, should be accurate enough yet fast) into the WRFDA system as one part of observation operators. Two widely used RTMs in NWP community, RTTOV (developed by EUMETSAT in Europe), and CRTM (developed by the Joint Center for Satellite Data Assimilation (JCSDA) in US), are already implemented in WRFDA system with a flexible and consistent user interface. Selection a which RTM to use is controlled by a simple namelist parameter RTM_OPTION (1 for RTTOV, the default, and 2 for CRTM). WRFDA is designed to be able to compile with only one of two RTM libraries or without RTM libraries (for those not interested in radiance assimilation) by the definition of environment variables ÒCRTMÓ and ÒRTTOVÓ (see Installing WRFDA section).

Both RTMs can calculate radiances for almost all available instruments aboard various satellite platforms in orbit. An important feature of WRFDA design is that all data structures related to radiance assimilation are dynamically allocated during running time according to simple namelist setup. The instruments to be assimilated are controlled at run time by four integer namelist parameters: RTMINIT_NSENSOR (the total number of sensors to be assimilated), RTMINIT_PLATFORM (the platforms IDs array to be assimilated with dimension RTMINIT_NSENSOR, e.g., 1 for NOAA, 9 for EOS, 10 for METOP and 2 for DMSP), RTMINIT_SATID (satellite IDs array) and RTMINIT_SENSOR (sensor IDs array, e.g., 0 for HIRS, 3 for AMSU-A, 4 for AMSU-B, 15 for MHS, 10 for SSMIS, 11 for AIRS). For instance, the configuration for assimilating 12 sensors from 7 satellites (what WRFDA can assimilated currently) will be

RTMINIT_NSENSOR = 12 14 # 6 AMSUA; 3 AMSUB; 3 MHS; 1 AIRS; 1 SSMIS

RTMINIT_PLATFORM = 1, 1, 1, 1,9,10,,1, 1, 1, 1, 1,10, 9, 2

RTMINIT_SATID = 15,16,18,19,2, 2,15,16,17,18,19, 2, 2,16

RTMINIT_SENSOR = 3, 3, 3, 3,3, 3, 4, 4, 4,15,15,15, 11,10

The instrument triplets (platform, satellite, and sensor ID) in the namelist can be ranked in any order. More detail about the convention of instrument triplet can be found on the web page http://research.metoffice.gov.uk/research/interproj/nwpsaf/rtm/rttov_description.html

or the tables 2 and 3 in the RTTOV v10 Users Guide (http://research.metoffice.gov.uk/research/interproj/nwpsaf/rtm/docs_rttov10/users_guide_10_v1.3.pdf )

CRTM uses a different instrument naming method. A convert routine inside WRFDA is already created to make CRTM use the same instrument triplet as RTTOV such that the user interface remains the same for RTTOV and CRTM.

When running WRFDA with radiance assimilation switched on (RTTOV or CRTM), a set of RTM coefficient files need to be loaded. For RTTOV option, RTTOV coefficient files are to be copied or linked to a sub-directory Òrttov_coeffsÓ under the working directory; for CRTM option, CRTM coefficient files are to be copied or linked to a sub-directory Òcrtm_coeffsÓ under the working directory. Only coefficients listed in namelist are needed. Potentially WRFDA can assimilate all sensors as long as the corresponding coefficient files are provided with RTTOV and CRTM. In addition, necessary developments on corresponding data interface, quality control, and bias correction are also important to make radiance data assimilated properly. However, a modular design of radiance relevant routines already facilitates much to add more instruments in WRFDA.

The RTTOV package is not distributed with WRFDA due to licensing and supporting issues. Users need to follow the instructions

on http://research.metoffice.gov.uk/research/interproj/nwpsaf/rtm to download the RTTOV source code and supplement coefficient files and emissivity atlas dataset. Starting from WRFDA V3.3, only RTTOV v10 can be used in WRFDA.

Starting from V3.2.1, the CRTM package is distributed with WRFDA, which is located in WRFDA/var/external/crtm. The CRTM code in WRFDA is basically the same as the source code that users can download from the the following link:

ftp://ftp.emc.ncep.noaa.gov/jcsda/CRTM.

d. Channel Selection

Channel selection in WRFDA is controlled by radiance ÔinfoÕ files located in the sub-directory Ôradiance_infoÕ under the working directory. These files are separated by satellites and sensors, e.g., noaa-15-amsua.info, noaa-16-amsub.info, dmsp-16-ssmis.info and so on. An example for 5 channels from noaa-15-amsub.info is shown below. The fourth column is used by WRFDA to control when to use a corresponding channel. Channels with the value Ò-1Ó indicate that the channel is Ònot assimilatedÓ (channels 1, 2 and 4 in this case), with the value Ò1Ó means ÒassimilatedÓ (channels 3 and 5). The sixth column is used by WRFDA to set the observation error for each channel. Other columns are not used by WRFDA. It should be mentioned that these error values might not necessarily be optimal for your applications; It is the userÕs responsibility to obtain the optimal error statistics for your own applications.

Sensor channel IR/MW use idum varch polarisation(0:vertical;1:horizontal)

415 1 1 -1 0 0.5500000000E+01 0.0000000000E+00

415 2 1 -1 0 0.3750000000E+01 0.0000000000E+00

415 3 1 1 0 0.3500000000E+01 0.0000000000E+00

415 4 1 -1 0 0.3200000000E+01 0.0000000000E+00

415 5 1 1 0 0.2500000000E+01 0.0000000000E+00

e. Bias Correction

Satellite radiance is generally considered biased with respect to a reference (e.g., background or analysis field in NWP assimilation) due to system error of observation itself, reference field, and RTM. Bias correction is a necessary step prior to assimilating radiance data. There are two ways of performing bias correction in WRFDA. One is based on the Harris and Kelly (2001) method and is carried out using a set of coefficient files pre-calculated with an off-line statistics package, which will apply to a training dataset for a month-long period. The other is Variational Bias Correction (VarBC). Only VarBC is introduced here and recommended for users because of its relative simplicity in usage.

f. Variational Bias Correction

Getting started with VarBC

To use VarBC, set namelist option USE_VARBC to TRUE and have a VARBC.in file in the working directory. VARBC.in is a VarBC setup file in ASCII format. A template is provided with the WRFDA package (WRFDA/var/run/VARBC.in).

Input and Output files

All VarBC input is passed through one single ASCII file called VARBC.in file. Once WRFDA has run with the VarBC option switched on, it will produce a VARBC.out file which looks very much like the VARBC.in file you provided. This output file will then be used as the input file for the next assimilation cycle.

Coldstart

Coldstarting means starting the VarBC from scratch i.e. when you do not know the values of the bias parameters.

The Coldstart is a routine in WRFDA. The bias predictor statistics (mean and standard deviation) are computed automatically and will be used to normalize the bias parameters. All coldstarted bias parameters are set to zero, except the first bias parameter (= simple offset), which is set to the mode (=peak) of the distribution of the (uncorrected) innovations for the given channel.

A threshold of number of observations can be set through a namelist option VARBC_NOBSMIN (default = 10), under which it is considered that not enough observations are present to keep the Coldstart values (i.e. bias predictor statistics and bias parameter values) for the next cycle. In this case, the next cycle will do another Coldstart.

Background Constraint for the bias parameters

The background constraint controls the inertia you want to impose on the predictors (i.e. the smoothing in the predictor time series). It corresponds to an extra term in the WRFDA cost function.

It is defined through an integer number in the VARBC.in file. This number is related to a number of observations: the bigger the number, the more inertia constraint. If these numbers are set to zero, the predictors can evolve without any constraint.

Scaling factor

The VarBC uses a specific preconditioning, which can be scaled through a namelist option VARBC_FACTOR (default = 1.0).

Offline bias correction

The analysis of the VarBC parameters can be performed "offline", i.e. independently from the main WRFDA analysis. No extra code is needed, just set the following MAX_VERT_VAR* namelist variables to be 0, which will disable the standard control variable and only keep the VarBC control variable.

MAX_VERT_VAR1=0.0

MAX_VERT_VAR2=0.0

MAX_VERT_VAR3=0.0

MAX_VERT_VAR4=0.0

MAX_VERT_VAR5=0.0

Freeze VarBC

In certain circumstances, you might want to keep the VarBC bias parameters constant in time (="frozen"). In this case, the bias correction is read and applied to the innovations, but it is not updated during the minimization. This can easily be achieved by setting the namelist options:

USE_VARBC=false

FREEZE_VARBC=true

Passive observations

Some observations are useful for preprocessing (e.g. Quality Control, Cloud detection) but you might not want to assimilate them. If you still need to estimate their bias correction, these observations need to go through the VarBC code in the minimization. For this purpose, the VarBC uses a separate threshold on the QC values, called "qc_varbc_bad". This threshold is currently set to the same value as "qc_bad", but can easily be changed to any ad hoc value.

g. Other namelist variables to control radiance assimilation

RAD_MONITORING (30)

Integer array of dimension RTMINIT_NSENSER, where 0 for assimilating mode, 1 for monitoring mode (only calculate innovation).

THINNING

Logical, TRUE will perform thinning on radiance data.

THINNING_MESH (30)

Real array with dimension RTMINIT_NSENSOR, values indicate thinning mesh (in KM) for different sensors.

QC_RAD

Logical, control if perform quality control, always set to TRUE.

WRITE_IV_RAD_ASCII

Logical, control if output Observation minus Background files which are in ASCII format and separated by sensors and processors.

WRITE_OA_RAD_ASCII

Logical, control if output Observation minus Analysis files (including also O minus B) which are ASCII format and separated by sensors and processors.

USE_ERROR_FACTOR_RAD

Logical, controls use of a radiance error tuning factor file Òradiance_error.factorÓ, which is created with empirical values or generated using variational tunning method (Desroziers and Ivanov, 2001)

ONLY_SEA_RAD

Logical, controls whether only assimilating radiance over water.

TIME_WINDOW_MIN

String, e.g., "2007-08-15_03:00:00.0000", start time of assimilation time window

TIME_WINDOW_MAX

String, e.g., "2007-08-15_09:00:00.0000", end time of assimilation time window

USE_ANTCORR (30)

Logical array with dimension RTMINIT_NSENSER, control if performing Antenna Correction in CRTM.

AIRS_WARMEST_FOV

Logical, controls whether using the observation brightness temperature for AIRS Window channel #914 as criterium for GSI thinning.

USE_CRTM_KMATRIX

Logical, controls whether using CRTM K matrix rather than calling CRTM TL and AD routines for gradient calculation.

USE_RTTOV_KMATRIX

Logical, controls whether using RTTOV K matrix rather than calling RTTOV TL and AD routines for gradient calculation.

RTTOV_EMIS_ATLAS_IR

integer, control the use of IR emissivity atlas.

Emissivity atlas data (should be downloaded separately from the RTTOV web site) need to be copied or linked under the sub-directory emis_data in the working directory if RTTOV_EMIS_ATLAS_IR is set to 1.

RTTOV_EMIS_ATLAS_MW

integer, control the use of MW emissivity atlas.

Emissivity atlas data (should be downloaded separately from the RTTOV web site) need to be copied or linked under the sub-directory emis_data in the working directory if RTTOV_EMIS_ATLAS_MW is set to 1 or 2.

h. Diagnostics and Monitoring

(1) Monitoring capability within WRFDA.

Run WRFDA with the rad_monitoring namelist parameter in record wrfvar14 in namelist.input.

0 means assimilating mode, innovations (O minus B) are calculated and data are used in minimization.

1 means monitoring mode: innovations are calculated for diagnostics and monitoring. Data are not used in minimization.

Number of rad_monitoring should correspond to number of rtminit_nsensor. If rad_monitoring is not set, then default value of 0 will be used for all sensors.

(2) Outputing radiance diagnostics from WRFDA

Run WRFDA with the following namelist variables in record wrfvar14 in namelist.input.

write_iv_rad_ascii=.true.

to write out (observation-background) and other diagnostics information in plain-text files with prefix inv followed by instrument name and processor id. For example, 01_inv_noaa-17-amsub.0000 (01 is outerloop index, 0000 is processor index)

write_oa_rad_ascii=.true.

to write out (observation-background), (observation-analysis) and other diagnostics information in plain-text files with prefix oma followed by instrument name and processor id. For example, 01_oma_noaa-18-mhs.0001

Each processor writes out information of one instrument in one file in the WRFDA working directory.

(3) Radiance diagnostics data processing

A Fortran90 program is used to collect the 01_inv* or 01_oma* files and write out in netCDF format (one instrument in one file with prefix diags followed by instrument name, analysis date, and suffix .nc)) for easier data viewing, handling and plotting with netCDF utilities and NCL scripts.

(4) Radiance diagnostics plotting

NCL scripts (WRFDA/var/graphics/ncl/plot_rad_diags.ncl and WRFDA/var/graphics/ncl/advance_cymdh.ncl) are used for plotting. The NCL script can be run from a shell script, or run stand-alone with interactive ncl command (need to edit the NCL script and set the plot options. Also the path of advance_cymdh.ncl, a date advancing script loaded in the main NCL plotting script, may need to be modified).

Step (3) and (4) can be done by running a single ksh script (WRFDA/var/scripts/da_rad_diags.ksh) with proper settings. In addition to the settings of directories and what instruments to plot, there are some useful plotting options, explained below.

export OUT_TYPE=ncgm	ncgm or pdf pdf will be much slower than ncgm and generate huge output if plots are not split. But pdf has higher resolution than ncgm.
export PLOT_STATS_ONLY=false	true or false true: only statistics of OMB/OMA vs channels and OMB/OMA vs dates will be plotted. false: data coverage, scatter plots (before and after bias correction), histograms (before and after bias correction), and statistics will be plotted.
export PLOT_OPT=sea_only	all, sea_only, land_only
export PLOT_QCED=false	true or false true: plot only quality-controlled data false: plot all data
export PLOT_HISTO=false	true or false: switch for histogram plots
export PLOT_SCATT=true	true or false: switch for scatter plots
export PLOT_EMISS=false	true or false: switch for emissivity plots
export PLOT_SPLIT=false	true or false true: one frame in each file false: all frames in one file
export PLOT_CLOUDY=false	true or false true: plot cloudy data. Cloudy data to be plotted are defined by PLOT_CLOUDY_OPT (si or clwp), CLWP_VALUE, SI_VALUE settings.
export PLOT_CLOUDY_OPT=si	si or clwp clwp: cloud liquid water path from model si: scatter index from obs, for amsua, amsub and mhs only
export CLWP_VALUE=0.2	only plot points with clwp >= clwp_value (when clwp_value > 0) clwp > clwp_value (when clwp_value = 0)
export SI_VALUE=3.0

(5) evolution of VarBC parameters

NCL scripts (WRFDA/var/graphics/ncl/plot_rad_varbc_param.ncl and WRFDA/var/graphics/ncl/advance_cymdh.ncl) are used for plotting evolutions of VarBC parameters.

Updating WRF Boundary Conditions

There are three input files: WRFDA analysis, wrfinput, and wrfbdy files from WPS/real.exe, and a namelist file: param.in for running da_update_bc.exe for domain-1. Before running NWP forecast using the WRF-model with WRFDA analysis, update the values and tendencies for each predicted variable in the first time period in the lateral boundary condition file. For domain-1 (wrfbdy_d01) must be updated to be consistent with the new WRFDA initial condition (analysis). This is absolutely essential. Moreover, in the cycling run mode (warm-start), the low boundary in the WRFDA analysis file also needs to be updated based on the information from the wrfinput file generated by WPS/real.exe at analysis time.

For the nested domains, domain-2, domain-3É, the lateral boundaries are provided by their parent domains, so no lateral boundary update is needed for these domains; but the low boundaries in each of the nested domainsÕ WRFDA analysis files still need to be updated. In these cases, you must set the namelist variable, domain_id > 1 (default is 1 for domain-1), and no wrfbdy_d01file need to be provided to the namelist variable: wrf_bdy_file.

This procedure is performed by the WRFDA utility called da_updated_bc.exe.

Note: Make sure that you have da_update_bc.exe in WRFDA/var/build directory. This executable should be created when you compiled WRFDA code,

To run da_update_bc.exe, follow the steps below:

> cd WRFDA/var/test/update_bc

> cp –p $DAT_DIR/rc/2008020512/wrfbdy_d01 ./wrfbdy_d01 (IMPORTANT: make a copy of wrfbdy_d01 as the wrf_bdy_file will be overwritten by da_update_bc.exe)

> vi parame.in

&control_param

wrfvar_output_file = './wrfvar_output'

wrf_bdy_file = './wrfbdy_d01'

wrf_input = '$DAT_DIR/rc/2008020512/wrfinput_d01'

cycling = .false. (set to .true. if WRFDA first guess comes from a previous WRF forecast.)

debug = .true.

low_bdy_only = .false.

update_lsm = .false.

/

> ln –sf WRFDA/var/da/da_update_bc.exe ./da_update_bc.exe

> ./da_updatebc.exe

At this stage, you should have the files wrfvar_output and wrfbdy_d01 in your WRFDA working directory. They are the WRFDA updated initial condition and boundary condition for any subsequent WRF model runs. To use, link a copy of wrfvar_output and wrfbdy_d01 to wrfinput_d01 and wrfbdy_d01, respectively, in your WRF working directory.

As of V3.2, some changes were made to da_update_bc to address some issues that are related to sea-ice and snow change during cycling runs and radiance data assimilation. The new settings in parame.in are introduced as follows. (However, for backward compatibility, the pre-V3.2 parame.in mentioned above still works with V3.2+ da_update_bc)

&control_param

da_file = '../tutorial/wrfvar_output'

wrf_bdy_file = './wrfbdy_d01'

wrf_input = '$DAT_DIR/rc/2008020512/wrfinput_d01'

domain_id = 1

debug = .true.

update_lateral_bdy = .true.

update_low_bdy = .true.

update_lsm = .false.

iswater = 16

/

update_lateral_bdy is required only for domain 1.

update_low_bdy is needed for all domains if running in cycling mode.

iswater (water point index) is 16 for USGS LANDUSE and 17 for MODIS LANDUSE.

Running da_update_bc.exe is recommended,

update_lateral_bdy = .false.

update_low_bdy = .true.

before running WRFDA, if in cycling mode (especially if you are doing radiance data assimilation and there is SEA ICE and SNOW in your domain) to get low-bdy updated first guess (da_file will be overwritten by da_update_bc).

Next run da_update_bc.exe with

update_lateral_bdy = .true.

update_low_bdy = .false.

after WRFDA to get updated lateral boubdary conditions (wrf_bdy_file will be overwritten by da_update_bc).

Running gen_be

Starting with WRFDA version 3.1, users have three choices to define the background error covariance (BE). We call them CV3, cv5, and CV6 respectively. With CV3 and CV5, the background errors are applied to the same set of the control variables, stream function, unbalanced potential velocity, unbalanced temperature, unbalanced surface pressure, and pseudo relative humidity. However, for CV6 the moisture control variable is the unbalanced part of pseudo relative humidity. With CV3, the control variables are in physical space while with CV5 and CV6 the control variables are in eigenvector space. So, the major differences between these two kinds of BE is the vertical covariance. CV3 uses the vertical recursive filter to model the vertical covariance but CV5 and CV6 use empirical orthogonal function (EOF) to represent the vertical covariance. The recursive filters to model the horizontal covariance are also different with these BEs. We have not conducted the systematic comparison of the analyses based on these BEs. However, CV3 (a BE file provided with our WRFDA system) is a global BE and can be used for any regional domains while CV5 and CV6 BEÕs are domain-dependent, which should be generated based in the forecasts data from the same domain. At this time, it is hard to tell which BE is better; the impact on analysis may vary case by case.

CV3 is the NCEP background error covariance, it is estimated in grid space by what has become known as the NMC method (Parrish and Derber 1992) . The statistics are estimated with the differences of 24 and 48-hour GFS forecasts with T170 resolution valid at the same time for 357 cases distributed over a period of one year. Both the amplitudes and the scales of the background error have to be tuned to represent the forecast error in the guess fields. The statistics that project multivariate relations among variables are also derived from the NMC method.

The variance of each variable and the variance of its second derivative are used to estimate its horizontal scales. For example, the horizontal scales of the stream function can be estimated from the variance of the vorticity and stream function.

The vertical scales are estimated with the vertical correlation of each variable. A table is built to cover the range of vertical scales for the variables. The table is then used to find the scales in vertical grid units. The filter profile and the vertical correlation are fitted locally. The scale of the best fit from the table is assigned as the scale of the variable at that vertical level for each latitude. Note that the vertical scales are locally defined so that the negative correlation further away in the vertical direction is not included.

Theoretically, CV3 BE is a generic background error statistics file which, can be used for any case. It is quite straightforward to use CV3 in your own case. To use the CV3 BE file in your case, set cv_options=3 in $wrfvar7 and the be.dat is located in WRFDA/var/run/be.dat.cv3.

To use CV5 or CV6 background error covariance, it is necessary to generate your domain-specific background error statistics with the gen_be utility. The background error statistics file supplied with the tutorial test case can NOT be used for your applications.

The Fortran main programs for gen_be can be found in WRFDA/var/gen_be. The executables of gen_be should be created after you have compiled the WRFDA code (as described earlier). The scripts to run these codes are in WRFDA/var/scripts/gen_be.

The input data for gen_be are WRF forecasts, which are used to generate model perturbations, used as a proxy for estimates of forecast error. For the NMC-method, the model perturbations are differences between forecasts (e.g. T+24 minus T+12 is typical for regional applications, T+48 minus T+24 for global) valid at the same time. Climatological estimates of background error may then be obtained by averaging such forecast differences over a period of time (e.g. one month). Given input from an ensemble prediction system (EPS), the inputs are the ensemble forecasts, and the model perturbations created are the transformed ensemble perturbations. The gen_be code has been designed to work with either forecast difference, or ensemble-based perturbations. The former is illustrated in this tutorial example.

It is important to include forecast differences from at least 00Z and 12Z through the period, to remove the diurnal cycle (i.e. do not run gen_be using just 00Z or 12Z model perturbations alone).

The inputs to gen_be are netCDF WRF forecast output ("wrfout") files at specified forecast ranges. To avoid unnecessary large single data files, it is assumed that all forecast ranges are output to separate files. For example, if we wish to calculate BE statistics using the NMC-method with (T+24)-(T+12) forecast differences (default for regional) then by setting the WRF namelist.input options history_interval=720, and frames_per_outfile=1 we get the necessary output datasets. Then the forecast output files should be arranged as follows: directory name is the forecast initial time, time info in the file name is the forecast valid time. 2008020512/wrfout_d01_2008-02-06_00:00:00 means a 12-hour forecast valid at 2008020600 initialized at 2008020512.

Example dataset for a test case (90 x 60 x 41 gridpoints) can be downloaded from http://www2.mmm.ucar.edu/wrf/users/wrfda/download/testdata.html, untar the gen_be_forecasts_20080205.tar.gz, you will have:

>ls $FC_DIR

-rw-r--r-- 1 users 11556492 2008020512/wrfout_d01_2008-02-06_00:00:00

-rw-r--r-- 1 users 11556492 2008020512/wrfout_d01_2008-02-06_12:00:00

-rw-r--r-- 1 users 11556492 2008020600/wrfout_d01_2008-02-06_12:00:00

-rw-r--r-- 1 users 11556492 2008020600/wrfout_d01_2008-02-07_00:00:00

-rw-r--r-- 1 users 11556492 2008020612/wrfout_d01_2008-02-07_00:00:00

-rw-r--r-- 1 users 11556492 2008020612/wrfout_d01_2008-02-07_12:00:00

In the above example, only 1 day (12Z 05 Feb to 12Z 06 Feb. 2002) of forecasts, every 12 hours are supplied to gen_be_wrapper to estimate forecast error covariance. It is only for demonstration. The minimum number of forecasts required depends on the application, number of grid points, etc. Month-long (or longer) datasets are typical for the NMC-method. Generally, at least a 1-month dataset should be used.

Under WRFDA/var/scripts/gen_be, gen_be_wrapper.ksh is used to generate the BE data, following variables need to be set to fit your case:

export WRFVAR_DIR=/users/noname/work/code/trunk/phoenix_g95_opt/WRFDA

export START_DATE=2008020612 # the first perturbation valid date

export END_DATE=2008020700 # the last perturbation valid date

export NUM_LEVELS=40 # e_vert - 1

export BIN_TYPE=5

export FC_DIR=/users/noname/work/exps/friendlies/expt/fc # where wrf forecasts are

export RUN_DIR=/users/noname/work/exps/friendlies/gen_be${BIN_TYPE}

Note: The START_DATE and END_DATE are perturbation valid dates. As show in the forecast list above, when you have 24-hour and 12-hour forecasts initialized at 2008020512 through 2008020612, the first and final forecast difference valid dates are 2008020612 and 2008020700 respectively.

Note: The forecast dataset should be located in $FC_DIR. Then type:

> gen_be_wrapper.ksh

Once gen_be_wrapper.ksh run is completed, the be.dat can be found under $RUN_DIR directory.

To get a clear idea about what are included in be.dat, the script gen_be_plot_wrapper.ksh may be used to plot various data in be.dat, for example:

Additional WRFDA Exercises:

a. Single Observation response in WRFDA:

With the single observation test, you may get some ideas of how the background and observation error statistics work in the model variable space. Single observation test is done in WRFDA by setting num_pseudo=1 along with other pre-specified values in record &wrfvar15 and &wrfvar19 of namelist.input,

With the settings shown below, WRFDA generates a single observation with pre-specified innovation (Observation – First Guess) value at desired location e.g. at (in terms of grid coordinate) 23x23, level 14 for ÒUÓ observation with error characteristics 1 m/s, innovation size = 1.0 m/s.

&wrfvar15

num_pseudo = 1,

pseudo_x = 23.0,

pseudo_y = 23.0,

pseudo_z = 14.0,

pseudo_err = 1.0,

pseudo_val = 1.0,

/

&wrfvar19

pseudo_var = ÒuÓ, (Note: pseudo_var can be u, v, t, p, q.
If pseudo_var is q, then the reasonable values of pseudo_err and pseudo_val are 0.001)

/

Note: You may wish to repeat this exercise for other observations like temperature (ÒtÓ), pressure ÒpÓ, specific humidity ÒqÓ and so on.

b. Response of BE length scaling parameter:

Run single observation test with following additional parameters in record &wrfvar7 of namelist.input

&wrfvar7

len_scaling1 = 0.5, # reduce psi length scale by 50%

len_scaling2 = 0.5, # reduce chi_u length scale by 50%

len_scaling3 = 0.5, # reduce T length scale by 50%

len_scaling4 = 0.5, # reduce q length scale by 50%

len_scaling5 = 0.5, # reduce Ps length scale by 50%

/

Note: You may wish to try the response of an individual variable by setting one parameter at a time. See the spread of analysis increment.

c. Response of changing BE variance:

Run the single observation test with following additional parameters in record &wrfvar7 of namelist.input

&wrfvar7

var_scaling1 = 0.25, # reduce psi variance by 75%

var_scaling2 = 0.25, # reduce chi_u variance by 75%

var_scaling3 = 0.25, # reduce T variance by 75%

var_scaling4 = 0.25, # reduce q variance by 75%

var_scaling5 = 0.25, # reduce Ps variance by 75%

/

Note: You may wish to try the response of individual variable by setting one parameter at one time. See the magnitude of analysis increments.

d. Response of convergence criteria:

Run the tutorial case with

&wrfvar6

eps = 0.0001,

/

You may wish to compare various diagnostics with the earlier run.

e. Response of outer loop on minimization:

Run the tutorial case with

&wrfvar6

max_ext_its = 2,

/

With this setting Òouter loopÓ for the minimization procedure will be activated. You may wish to compare various diagnostics with earlier run.

Note: The Maximum permissible value for ÒMAX_EXT_ITSÓ is 10.

f. Response of suppressing particular types of data in WRFDA:

The types of observations that WRFDA gets to use actually depend on what is included in the observation file and the WRFDA namelist settings. For example, if you have SYNOP data in the observation file, you can suppress its usage in WRFDA by setting use_synopobs=false in record &wrfvar4 of namelist.input. It is OK if there is no SYNOP data in the observation file and use_synopobs=true.

Turning on and off certain types of observations is widely used for assessing the impact of observations on data assimilations.

Note: It is important to go through the default Òuse_*Ó settings in record &wrfvar4 in WRFDA/Registry/Registry.wrfvar to know what observations are activated in default.

WRFDA with Multivariate Background Error (MBE) Statistics

A new control variable option to implement multivariate background error (MBE) statistics in WRFDA has been introduced. It may be activated by setting the ÒnamelistÓ variable Òcv_options=6Ó. This option introduces six additional correlation coefficients in the definition of balanced part of analysis control variables. Thus with this implementation, moisture analysis is multivariate in the sense that temperature and wind may lead to moisture increments and vise-versa. The Ògen_beÓ utility has also been updated to compute the desired MBE statistics required for this option. The updates include basic Òsource codeÓ, ÒscriptsÓ and ÒgraphicsÓ to display some important diagnostics about MBE statistics. Further details may be seen at:

https://wiki.ucar.edu/download/attachments/60622477/WRFDA__update_for_cv6.pdf

a. How to generate multivariate background error statistics for WRFDA?

Multivariate background error statistics for WRFDA is generated by executing a top-level script Ògen_be/wrapper_gen_be_gsi.kshÓ residing under ÒSCRIPTS_DIRÓ via a suitable wrapper script. The rest of the procedure remains the same as with normal running of the Ògen_beÓ utility. A successful run will create a Òbe.datÓ file in ÒRUN_DIRÓ directory.

b. How to run WRFDA with multivariate background error statistics?

After successfully generating multivariate background error statistics file Òbe.datÓ the procedure for running WRFDA is straight. If WRFDA is run through ÒwrapperÓ script, declare suitably the namelist variable ÒNL_CV_OPTIONS=6Ó in the ÒwrapperÓ script. If WRFDA is run directly (by executing Òda_wrfvar.exeÓ) then, include Òcv_options=6Ó in Ònamelist.inputÓ file under Òwrfvar7Ó list of namelist options.

c. How to tune multivariate background error statistics in running WRFDA?

Following is the list of nine tuning parameters available in WRFDA. Default values for these variables are set as Ò1.0Ó. By setting corresponding values > 1.0 (< 1.0) will increase (decrease) the corresponding contributions as described in the following Table.

Variable name	Description
psi_chi_factor	Parameter to control contribution of stream function in defining balanced part of velocity potential
psi_t_factor	Parameter to control contribution of stream function in defining balanced part of temperature
psi_ps_factor	Parameter to control contribution of stream function in defining balanced part of surface pressure
psi_rh_factor	Parameter to control contribution of stream function in defining balanced part of moisture
chi_u_t_factor	Parameter to control contribution of unbalanced part of velocity potential in defining balanced part of temperature
chi_u_ps_factor	Parameter to control contribution of unbalanced part of velocity potential in defining balanced part of surface pressure
chi_u_rh_factor	Parameter to control contribution of unbalanced part of velocity potential in defining balanced part of moisture
t_u_rh_factor	Parameter to control contribution of unbalanced part of temperature in defining balanced part of moisture
ps_u_rh_factor	Parameter to control contribution of unbalanced part of surface pressure in defining balanced part of moisture

WRFDA Diagnostics

WRFDA produces a number of diagnostic files that contain useful information on how the data assimilation has performed. This section will introduce you to some of these files, and what to look for.

Having run WRFDA, it is important to check a number of output files to see if the assimilation appears sensible. The WRFDA package, which includes lots of useful scripts may be downloaded from http://www2.mmm.ucar.edu/wrf/users/wrfda/download/tools.html

The content of some useful diagnostic files are as follows:

cost_fn and grad_fn: These files hold (in ASCII format) WRFDA cost and gradient function values, respectively, for the first and last iterations. However, if you run with PRINT_DETAIL_GRAD=true, these values will be listed for each iteration; this can be helpful for visualization purposes. The NCL script WRFDA/var/graphics/ncl/plot_cost_grad_fn.ncl may be used to plot the content of cost_fn and grad_fn, if these files are generated with PRINT_DETAIL_GRAD=true.

Note: Make sure that you removed first two lines (header) in cost_fn and grad_fn before you plot. Also, you need to specify the directory name for these two files.

gts_omb_oma_01: It contains (in ASCII format) information on all of the observations used by the WRFDA run. Each observation has its observed value, quality flag, observation error, observation minus background (OMB), and observation minus analysis (OMA). This information is very useful for both analysis and forecasts verification purposes.

namelist.input: This is the WRFDA input namelist file, which contains all the user defined non-default options. Any namelist defined options that do not appear in this file, should have their names checked against values in WRFDA/Registry/Registry.wrfvar.

namelist.output: A consolidated list of all the namelist options used.

rsl*: Files containing information of standard WRFDA output from individual processors when multiple processors are used. It contains host of information on number of observations, minimization, timings etc. Additional diagnostics may be printed in these files by including various ÒprintÓ WRFDA namelist options. To learn more about these additional ÒprintÓ options, search Òprint_Ó string in WRFDA/Registry/Registry.wrfvar.

statistics: Text file containing OMB (OI), OMA (OA) statistics (minimum, maximum, mean and standard deviation) for each observation type and variable. This information is very useful in diagnosing how WRFDA has used different components of the observing system. Also contained are the analysis minus background (A-B) statistics i.e. statistics of the analysis increments for each model variable at each model level. This information is very useful in checking the range of analysis increment values found in the analysis, and where they are in the WRF-model grid space.

The WRFDA analysis file is wrfvar_output. It is in WRF (NetCDF) format. It will become the input file Òwrfinput_d01Ó of any subsequent WRF runs after lateral boundary and/or low boundary conditions are updated by another WRFDA utility (See section ÒUpdating WRF boundary conditionsÓ).

A NCL script WRFDA/var/graphics/ncl/WRF-Var_plot.ncl, is provided for plotting. You need to specify the analsyis_file name, its full path etc. Please see the in-line comments in the script for details.

As an example, if you are aiming to display U-component of the analysis at level 18, execute the following command after modifying the script ÒWRFDA/var/graphcs/ncl/WRF-Var_plot.nclÓ, and make sure the following pieces of codes are uncommented:

var = "U"

fg = first_guess->U

an = analysis->U

plot_data = an

When you execute the following command from WRFDA/var/graphics/ncl.

> ncl WRF-Var_plot.ncl

The plot should look like:

You may change the variable name, level etc in this script to display the variable of your choice at the desired eta level.

Take time to look through the text output files to ensure you understand how WRFDA works. For example:

How closely has WRFDA fitted individual observation types? Look at the statistics file to compare the O-B and O-A statistics.

How big are the analysis increments? Again, look in the statistics file to see minimum/maximum values of A-B for each variable at various levels. It will give you a feel for the impact of input observation data you assimilated via WRFDA by modifying the input analysis first guess.

How long did WRFDA take to converge? Does it really converge? You will get the answers of all these questions by looking into rsl-files, as it indicates the number of iterations taken by WRFDA to converge. If this is the same as the maximum number of iterations specified in the namelist (NTMAX) or its default value (=200) set in WRFDA/Registry/Registry.wrfvar, then it means that the analysis solution did not converge. If so, you may need to increase the value of ÒNTMAXÓ and rerun your case to ensure that the convergence is achieved. On the other hand, a normal WRFDA run should usually converge within 100 iterations. If it still doesnÕt converge in 200 iterations, that means there might be some problem in the observations or first guess.

A good visual way of seeing the impact of assimilation of observations is to plot the analysis increments (i.e. analysis minus first guess difference). Many different graphics packages (e.g. RIP4, NCL, GRADS etc) can do this. The plot of level 18 theta increments below was produced using the particular NCL script. This script is located at WRFDA/var/graphics/ncl/WRF-Var_plot.ncl.

You need to modify this script to fix the full path for first_guess & analysis files. You may also use it to modify the display level by setting ÒklÓ and the name of the variable to display by setting ÒvarÓ. Further details are given in this script.

If you are aiming to display the increment of potential temperature at level 18, after modifying WRFDA/var/graphcs/ncl/WRF-Var_plot.ncl, make sure following pieces of codes are uncommented:

var = "T"

fg = first_guess->T ;Theta- 300

an = analysis->T ;Theta- 300

plot_data = an - fg

When you execute the following command from ÒWRFDA/var/graphics/nclÓ.

> ncl WRF-Var_plot.ncl

The plot created will looks as follows:

Note: Larger analysis increments indicate a larger data impact in the corresponding region of the domain.

Hybrid Data Assimilation in WRFDA

The WRFDA system also includes a hybrid data assimilation technique, which is based on the existing 3DVAR. The difference between hybrid and 3DVAR schemes is that 3DVAR relies solely on a static covariance model to specify the background errors, while the hybrid system uses a combination of 3DVAR static error covariances and ensemble-estimated error covariances to incorporate a flow-dependent estimate of the background error statistics. Please refer to Wang et al. (2008a,b) for a detailed description of the methodology used in the WRF hybrid system. The following section will give a brief introduction of some aspects of using the hybrid system.

a. Source Code

Three executables are used in the hybrid system. If you have successfully compiled the WRFDA system, you will see the following:

WRFDA/var/build/gen_be_ensmean.exe

WRFDA/var/build/gen_be_ep2.exe

WRFDA/var/build/da_wrfvar.exe

gen_be_ensmean.exe is used to calculate the ensemble mean, while gen_be_ep2.exe is used to calculate the ensemble perturbations. As with 3DVAR/4DVAR, da_wrfvar.exe is the main WRFDA program. However, in this case, da_wrfvar.exe will run in the hybrid mode.

b. Running The Hybrid System

The procedure is the same as running 3DVAR/4DVAR with the exception of some extra input files and namelist settings. The basic input files for WRFDA are LANDUSE.TBL, ob.ascii or ob.bufr (depending on which observation format you use), and be.dat (static background errors). Additional input files required by the hybrid are a single ensemble mean file (used as the fg for the hybrid application) and a set of ensemble perturbation files (used to represent flow-dependent background errors).

A set of initial ensemble members must be prepared before the hybrid application can be started. These ensembles can be obtained from other ensemble model outputs or you can generate them yourself, for example, adding random noise to the initial conditions at a previous time and integrating each member to the desired time. Once you have the initial ensembles, the ensemble mean and perturbations can be calculated following the steps below.

1) Calculate ensemble mean

Copy or link the ensemble forecasts to your working directory. In this example, the time is 2006102712.

< ln -sf /wrfhelp/DATA/VAR/Hybrid/fc/2006102712.e0* .

Next, copy the directory that contains two template files (ensemble mean and variance files) to your working directory. In this case, the directory name is 2006102712, which contains the template ensemble mean file (wrfout_d01_2006-10-28_00:00:00) and the template variance file (wrfout_d01_2006-10-28_00:00:00.vari). These template files will be overwritten by the program that calculates the ensemble mean and variance as discussed below.

< cp -r /wrfhelp/DATA/VAR/Hybrid/fc/2006102712 .

Edit gen_be_ensmean_nl.nl (or copy it from /wrfhelp/DATA/VAR/Hybrid/gen_be_ensmean_nl.nl). You will need to set the following information in this script as follows:

< vi gen_be_ensmean_nl.nl

&gen_be_ensmean_nl
directory = './2006102712'
filename = 'wrfout_d01_2006-10-28_00:00:00'
num_members = 10
nv = 7
cv = 'U', 'V', 'W', 'PH', 'T', 'MU', 'QVAPOR'
/

Here,

ÒdirectoryÓ is the folder you just copied,

ÒfilenameÓ is the name of the ensemble mean file,

Ònum_membersÓ is the number of ensemble members you are using,

ÒnvÓ is the number of variables, which must be consistent with the next ÒcvÓ option, and

ÒcvÓ is the name of variables used in the hybrid system.

Next, link gen_be_ensmean.exe to your working directory and run it.

< ln –sf WRFDA/var/build/gen_be_ensmean.exe .
< ./gen_be_ensmean.exe

Check the output files.

2006102712/wrfout_d01_2006-10-28_00:00:00 is the ensemble mean

2006102712/wrfout_d01_2006-10-28_00:00:00.vari is the ensemble variance

2) Calculate ensemble perturbations

Create another sub-directory in which you will be working to create ensemble perturbations.

< mkdir -p 2006102800/ep
< cd 2006102800/ep

Next, run gen_be_ep2.exe.

gen_be_ep2.exe requires four command-line arguments (DATE, NUM_MEMBER, DIRECTORY, FILENAME) as shown below:

< ln –sf WRFDA/var/build/gen_be_ep2.exe   .
< ./gen_be_ep2.exe    2006102800   10 ../../2006102712    wrfout_d01_2006-10-28_00:00:00

Check the output files.

A list of binary files will be created under the 2006102800/ep directory. Among them, tmp.e* are temporary scratch files that can be removed.

3) Run WRFDA in hybrid mode

In your hybrid working directory, link all the necessary files and directories as follows:

< ln -fs 2006102800/ep ./ep (ensemble perturbation files should be under the ep subdirectory)
< ln -fs 2006102712/wrfout_d01_2006-10-28_00:00:00 ./fg (first guess is the ensemble mean)
< ln -fs WRFDA/run/LANDUSE.TBL .
< ln -fs /wrfhelp/DATA/VAR/Hybrid/ob/2006102800/ob.ascii ./ob.ascii (or ob.bufr)
< ln -fs /wrfhelp/DATA/VAR/Hybrid/be/be.dat ./be.dat
< ln –fs WRFDA/var/build/da_wrfvar.exe .
< cp /wrfhelp/DATA/VAR/Hybrid/namelist.input .

Edit namelist.input and pay special attention to the following hybrid-related settings:

&wrfvar7
je_factor = 2.0
/
&wrfvar16
ensdim_alpha = 10
alphacv_method = 2
alpha_corr_type=3
alpha_corr_scale = 1500.0
alpha_std_dev=1.000
/

Next, run hybrid in serial mode (recommended for initial testing of the hybrid system), or in parallel mode

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

Check the output files.

The output file lists are the same as when you run WRF 3D-Var.

c. Hybrid namelist options

1) je_factor : ensemble covariance weighting factor. This factor controls the weighting component of ensemble and static covariances. The corresponding jb_factor = je_factor/(je_factor - 1).

2) ensdim_alpha: the number of ensemble members. Hybrid mode is activated when ensdim_alpha is larger than zero.

3) alphacv_method: 1=perturbations in control variable space (ÒpsiÓ,Óchi_uÓ,Ót_uÓ,ÓrhÓ,Óps_uÓ); 2=perturbations in model space (ÒuÓ,ÓvÓ,ÓtÓ,ÓqÓ,ÓpsÓ). Option 2 is extensively tested and recommended to use.

4) alpha_corr_type: correlation function. 1=Exponential; 2=SOAR; 3=Gaussian.

5) alpha_corr_scale: hybrid covariance localization scale in km unit. Default value is 1500.

6) alpha_std_dev: alpha standard deviation. Default value is 1.0

Description of Namelist Variables

WRFDA namelist variables.

Variable Names	Default Value			Description
&wrfvar1
write_increments	false			.true.:  write out a binary analysis increment file
var4d	false			.true.: 4D-Var mode
var4d_lbc	true			.true.: on/off for lateral boundary control in 4D-Var
var4d_bin	3600			seconds, observation sub-window length for 4D-Var
multi_inc	0			> 0: multi-incremental run
print_detail_radar	false			print_detail_xxx: output extra (sometimes can be too many) diagnostics for debugging; not recommended to turn them on for production runs
print_detail_xa	false
print_detail_xb	false
print_detail_obs	false
print_detail_grad	false			.true.: to print out detailed gradient of each observation type at each iteration
check_max_iv_print	true			obsolete (used only by Radar)
&wrfvar2
analysis_accu	900			seconds, if the time difference between the namelist setting (analysis_date) and date info read in from first guess is larger than analysis_accu, WRFDA will issue a warning message ("=======> Wrong xb time found???"), but won't abort.
calc_w_increment	false			.true.:  the increment of the vertical velocity W will be diagnosed based on the increments of other fields. If there is information of the W from observations assimilated, such as the Radar radial velocity, the W increments are always computed, no matter calc_w_increment=true. or .false. .false.: the increment of the vertical velocity W is zero if no W information is assimilated.
dt_cloud_model	false			Not used
&wrfvar3
fg_format	1			1: fg_format_wrf_arw_regional (default) 2: fg_format_wrf_nmm_regional 3: fg_format_wrf_arw_global 4: fg_format_kma_global
ob_format	2			1: ob_format_bufr (NCEP PREPBUFR), read in data from ob.bufr (not fully tested) 2: ob_format_ascii (output from obsproc), read in data from ob.ascii (default) 3: ob_format_madis (not tested)
num_fgat_time	1			1: 3DVar > 1: number of time slots for FGAT and 4DVAR
&wrfvar4
thin_conv	true			for ob_format=1 (NCEP PREPBUFR) only. thining is mandatory for ob_format=1 as time-duplicate data are "thinned" within thinning routine, however, thin_conv can be set to .false. for debugging purpose.
thin_mesh_conv	20. (max_instruments)			for ob_format=1 (NCEP PREPBUFR) only. km, each observation type can set its thinning mesh and the index/order follows the definition in WRFDA/var/da/da_control/da_control.f90
use_synopobs	true			use_xxxobs - .true.: assimilate xxx obs if available .false.: not assimilate xxx obs even available
use_shipsobs	true
use_metarobs	true
use_soundobs	true
use_pilotobs	true
use_airepobs	true
use_geoamvobs	true
use_polaramvobs	true
use_bogusobs	true
use_buoyobs	true
use_profilerobs	true
use_satemobs	true
use_gpspwobs	true
use_gpsrefobs	true
use_qscatobs	true
use_radarobs	false
use_radar_rv	false
use_radar_rf	false
use_airsretobs	true
; use_hirs2obs, use_hirs3obs, use_hirs4obs, use_mhsobs ; use_msuobs, use_amsuaobs, use_amsubobs, use_airsobs, ; use_eos_amsuaobs, use_hsbobs, use_ssmisobs are ; radiance-related variables that only control if reading ; in corresponding BUFR files into WRFDA or not, but ; do not control if assimilate the data or not. ; Some more variables have to be set in &wrfvar14 in order ; to assimilate radiance data.
use_hirs2obs	fasle			.true.: to read in data from hirs2.bufr
use_hirs3obs	false			.true.: to read in data from hirs3.bufr
use_hirs4obs	false			.true.: to read in data from hirs4.bufr
use_mhsobs	false			.true.: to read in data from mhs.bufr
use_msuobs	false			.true.: to read in data from msu.bufr
use_amsuaobs	false			.true.: to read in data from amsua.bufr
use_amsubobs	false			.true.: to read in data from amsub.bufr
use_airsobs	false			.true.: to read in data from airs.bufr
use_eos_amsuaobs	false			.true.: to read in data from airs.bufr
use_hsbobs	false			.true.: to read in data from hsb.bufr
use_ssmisobs	false			.true.: to read in data from ssmis.bufr
use_obs_errfac	false			.true.: apply obs error tuning factors if errfac.dat is available for conventional data only
&wrfvar5
check_max_iv	true			.true.: reject the observations whose innovations (O-B) are larger than a maximum value defined as a multiple of the observation error for each observation. i.e., inv > (obs_error*factor) --> fails_error_max; the default maximum value is 5 times the observation error ; the factor of 5 can be changed through max_error_* settings.
max_error_t	5.0			maximum check_max_iv error check factor for t
max_error_uv	5.0			maximum check_max_iv error check factor for u and v
max_error_pw	5.0			maximum check_max_iv error check factor for precipitable water
max_error_ref	5.0			maximum check_max_iv error check factor for gps refractivity
max_error_q	5.0			maximum check_max_iv error check factor for specific humidity
max_error_p	5.0			maximum check_max_iv error check factor for pressure
max_error_thickness				maximum check_max_iv error check factor for thickness
max_error_rv				maximum check_max_iv error check factor for radar radial velocity
max_error_rf				maximum check_max_iv error check factor for radar reflectivity
&wrfvar6
max_ext_its	1			number of outer loops
ntmax	200			maximum number of iterations in an inner loop
eps	0.01 (max_ext_its)			minimization convergence criterion (used dimension: max_ext_its); minimization stops when the norm of the gradient of the cost function gradient is reduced by a factor of eps. inner minimization stops either when the criterion is met or when inner iterations reach ntmax.
&wrfvar7
cv_options	5			3: NCEP Background Error model 5: NCAR Background Error model (default) 6: Use of multivariate background error statistics
as1(3)	-1.0			tuning factors for variance, horizontal and vertical scales for control variable 1 = stream function. For cv_options=3 only. The actual default values are 0.25, 1.0, 1.5.
as2(3)	-1.0			tuning factors for variance, horizontal and vertical scales for control variable 2 - unbalanced potential velocity. For cv_options=3 only. The actual default values are 0.25, 1.0, 1.5.
as3(3)	-1.0			tuning factors for variance, horizontal and vertical scales for control variable 3 - unbalanced temperature. For cv_options=3 only. The actual default values are 0.25, 1.0, 1.5.
as4(3)	-1.0			tuning factors for variance, horizontal and vertical scales for control variable 4 - pseudo relative humidity. For cv_options=3 only. The actual default values are 0.25, 1.0, 1.5.
as5(3)	-1.0			tuning factors for variance, horizontal and vertical scales for control variable 5 - unbalanced surface pressure. For cv_options=3 only. The actual default values are 0.25, 1.0, 1.5.
rf_passes	6			number of passes of recursive filter.
var_scaling1	1.0			tuning factor of background error covariance for control variable 1 - stream function. For cv_options=5 only.
var_scaling2	1.0			tuning factor of background error covariance for control variable 2 - unbalanced velocity potential. For cv_options=5 only.
var_scaling3	1.0			tuning factor of background error covariance for control variable 3 - unbalanced temperature. For cv_options=5 only.
var_scaling4	1.0			tuning factor of background error covariance for control variable 4 - pseudo relative humidity. For cv_options=5 only.
var_scaling5	1.0			tuning factor of background error covariance for control variable 5 - unbalanced surface pressure. For cv_options=5 only.
len_scaling1	1.0			tuning factor of scale-length for stream function. For cv_options=5 only.
len_scaling2	1.0			tuning factor of scale-length for unbalanced velocity potential. For cv_options=5 only.
len_scaling3	1.0			tuning factor of scale-length for unbalanced temperature. For cv_options=5 only.
len_scaling4	1.0			tuning factor of scale-length for pseudo relative humidity. For cv_options=5 only.
len_scaling5	1.0			tuning factor of scale-length for unbalanced surface pressure. For cv_options=5 only.
je_factor	1.0			ensemble covariance weighting factor
&wrfvar8 ;not used
&wrfvar9				for program tracing. trace_use=.true. gives additional performance diagnostics (calling tree, local routine timings, overall routine timings, memory usage) It does not change results, but does add runtime overhead.
stdout	6			unit number for standard output
stderr	0			unit number for error output
trace_unit	7			Unit number for tracing output note that units 10 and 9 are reserved for reading namelist.input and writing namelist.output respectively.
trace_pe	0			Currently, statistics are always calculated for all processors, and output by processor 0.
trace_repeat_head	10			the number of times any trace statement will produce output for any particular routine. This stops overwhelming trace output when a routine is called multiple times. Once this limit is reached a 'going quiet' message is written to the trace file, and no more output is produced from the routine, though statistics are still gathered.
trace_repeat_body	10			see trace_repeat_head description
trace_max_depth	30			define the deepest level to which tracing writes output
trace_use	true			.true.: activate tracing
trace_use_frequent	false
trace_use_dull	false
trace_memory	true			.true.: calculate allocated memory using a mallinfo call. On some platforms (Cray and Mac), mallinfo is not available and no memory monitoring can be done.
trace_all_pes	false			.true.: tracing is output for all pes. As stated in trace_pe, this does not change processor statistics.
trace_csv	true			.true.: tracing statistics are written to a xxxx.csv file in CSV format
use_html	true			.true.: tracing and error reporting routines will include HTML tags.
warnings_are_fatal	false			.true.: warning messages that would normally allow the program to continue are treated as fatal errors.
&wrfvar10 ; for code developer
&wrfvar11
cv_options_hum	1			do not change
check_rh				0 --> No supersaturation check after minimization. 1 --> supersaturation (rh> 100%) and minimum rh (rh<10%) check, and make the local adjustment of q. 2 --> supersaturation (rh> 95%) and minimum rh (rh<11%) check and make the multi-level q adjustment under the constraint of conserved column integrated water vapor
sfc_assi_options	1			1 --> surface observations will be assimilated based on the lowest model level first guess. Observations are not used when the height difference of the elevation of the observing site and the lowest model level height is larger than 100m. 2 --> surface observations will be assimilated based on surface similarity theory in PBL. Innovations are computed based on 10-m wind, 2-m temperature and 2-m moisture.
calculate_cg_cost_fn	false			conjugate gradient algorithm does not require the computation of cost function at every iteration during minimization. .true.: Compute and write out cost function and gradient of each iteration into files called cost_fn and grad_fn. false.: Only the initial and final cost functions are computed and output.
lat_stats_option	false			do not change
&wrfvar12
balance_type	1			obsolete
&wrfvar13
vert_corr	2			do not change
vertical_ip	0			obsolete
vert_evalue	1			do not change
max_vert_var1	99.0			specify the maximum truncation value (in percentage) to explain the variance of stream function in eigenvector decomposition
max_vert_var2	99.0			specify the maximum truncation value (in percentage) to explain the variance of unbalanced potential velocity in eigenvector decomposition
max_vert_var3	99.0			specify the maximum truncation value (in percentage) to explain the variance of the unbalanced temperature in eigenvector decomposition
max_vert_var4	99.0			specify the maximum truncation value (percentage) to explain the variance of pseudo relative humidity in eigenvector decomposition
max_vert_var5	99.0			for unbalanced surface pressure, it should be a non-zero positive numer. set max_vert_var5=0.0 only for offline VarBC applications.
psi_chi_factor	1.0			Parameter to control contribution of stream function in defining balanced part of velocity potential
psi_t_factor	1.0			Parameter to control contribution of stream function in defining balanced part of temperature
psi_ps_factor	1.0			Parameter to control contribution of stream function in defining balanced part of surface pressure
psi_rh_factor	1.0			Parameter to control contribution of stream function in defining balanced part of moisture
chi_u_t_factor	1.0			Parameter to control contribution of unbalanced part of velocity potential in defining balanced part of temperature
chi_u_ps_factor	1.0			Parameter to control contribution of unbalanced part of velocity potential in defining balanced part of surface pressure
chi_u_rh_factor	1.0			Parameter to control contribution of unbalanced part of velocity potential in defining balanced part of moisture
t_u_rh_factor	1.0			Parameter to control contribution of unbalanced part of temperature in defining balanced part of moisture
ps_u_rh_factor	1.0			Parameter to control contribution of unbalanced part of surface pressure in defining balanced part of moisture
&wrfvar14						Parameter to control contribution of unbalanced part of surface pressure in defining balanced part of moisture
the following 4 variables (rtminit_nsensor, rtminit_platform, rtminit_satid, rtminit_sensor) together control what sensors to be assimilated.
rtminit_nsensor	1				total number of sensors to be assimilated
rtminit_platform	-1 (max_instruments)				platforms IDs array (used dimension: rtminit_nsensor); e.g., 1 for NOAA, 9 for EOS, 10 for METOP and 2 for DMSP
rtminit_satid	-1.0 (max_instruments)				satellite IDs array (used dimension: rtminit_nsensor)
rtminit_sensor	-1.0 (max_instruments)				sensor IDs array (used dimension: rtminit_nsensor); e.g., 0 for HIRS, 3 for AMSU-A, 4 for AMSU-B, 15 for MHS, 10 for SSMIS, 11 for AIRS
rad_monitoring	0 (max_instruments)				integer array (used dimension: rtminit_nsensor); 0: assimilating mode; 1: monitoring mode (only calculate innovations)
thinning_mesh	60.0 (max_instruments)				real array (used dimension: rtminit_nsensor); specify thinning mesh size (in KM) for different sensors.
thinning	false				.true.: perform thinning on radiance data
qc_rad	true				.true.: perform quality control. always .true.
write_iv_rad_ascii	false				.true.: output radiance Observation minus Background files, which are in ASCII format and separated by sensors and processors.
write_oa_rad_ascii	false				.true.: output radiance Observation minus Analysis files (Observation minus Background information is also included), which are in ASCII format and separated by sensors and processors.
use_error_factor_rad	false				.true.: use a radiance error tuning factor file "radiance_error.factor", which can be created with empirical values or generated using variational tuning method (Desroziers and Ivanov, 2001)
use_antcorr	false (max_instruments)				.true.: perform Antenna Correction in CRTM
rtm_option	1				what RTM (Radiative Transfer Model) to use 1: RTTOV (WRFDA needs to compile with RTTOV) 2: CRTM (WRFDA needs to compile with CRTM)
only_sea_rad	false				.true.: assimilate radiance over water only
use_varbc	false				.true.: perform Variational Bias Correction. A parameter file in ASCII format called VARBC.in (a template is provided with the source code tar ball) is required.
freeze_varbc	false				.true: together with use_varbc=.false., keep the VarBC bias parameters constant in time. In this case, the bias correction is read and applied to the innovations, but it is not updated during the minimization.
varbc_factor	1.0				for scaling the VarBC preconditioning
varbc_nobsmin	10				defines the minimum number of observations required for the computation of the predictor statistics during the first assimilation cycle. If there are not enough data (according to "VARBC_NOBSMIN") on the first cycle, the next cycle will perform a coldstart again.
airs_warmest_fov	false				.true.: uses the observation brightness temperature forAIRS Window channel #914 as criterion for GSI thinning (with a higher amplitude than the distance from the observation location to the nearest grid point).
use_crtm_kmatrix	false				.true. use CRTM K matrix rather than calling CRTM TL and AD routines for gradient calculation, which reduces runtime noticeably.
use_rttov_kmatrix	false				.true. use RTTOV K matrix rather than calling RTTOV TL and AD routines for gradient calculation, which reduces runtime noticeably.
rttov_emis_atlas_ir	0				0: do not use IR emissivity atlas 1: use IR emissivity atlas (recommended)
rttov_emis_atlas_mw	0				0: do not use MW emissivity atlas 1: use TELSEM MW emissivity atlas (recommended) 2: use CNRM MW emissivity atlas
&wrfvar15 (needs to be set together with &wrfvar19)
num_pseudo	0			Set the number of pseudo observations, either 0 or 1 (single ob)
pseudo_x	1.0			Set the x-position (I) of the OBS in unit of grid-point.
pseudo_y	1.0			Set the y-position (J) of the OBS in unit of grid-point.
pseudo_z	1.0			Set the z-position (K) of OBS with the vertical level index, in bottom-up order.
pseudo_val	1.0			Set the innovation of the ob; wind in m/s, pressure in Pa, temperature in K, specific humidity in kg/kg
pseudo_err	1.0			set the error of the pseudo ob. Unit the same as pseudo_val.; if pseudo_var="q", pseudo_err=0.001 is more reasonable.
&wrfvar16 (for hybrid WRFDA/ensemble)
alphacv_method	2			1: ensemble perturbations in control variable space 2: ensemble perturbations in model variable space
ensdim_alpha	0			ensemble size
alpha_corr_type	3			1: alpha_corr_type_exp 2: alpha_corr_type_soar 3: alpha_corr_type_gaussian (default)
alpha_corr_scale	1500.0			km
&wrfvar17
analysis_type	Ò3D-VARÓ			"3D-VAR": 3D-VAR mode (default); "QC-OBS": 3D-VAR mode plus extra filtered_obs output; "VERIFY": verification mode. WRFDA resets check_max_iv=.false. and ntmax=0; "RANDOMCV": for creating ensemble perturbations
&wrfvar18 (needs to set &wrfvar21 and &wrfvar22 as well if ob_format=1 and/or radiances are used)
analysis_date	Ò2002-08-03_00:00:00.0000Ó			specify the analysis time. It should be consistent with the first guess time. However, if time difference between analysis_date and date info read in from first guess is larger than analysis_accu, WRFDA will issue a warning message ("=======> Wrong xb time found???"), but won't abort.
&wrfvar19 (needs to be set together with &wrfvar15)
pseudo_var	ÒtÓ			Set the name of the OBS variable: 'u' = X-direction component of wind, 'v' = Y-direction component of wind, 't' = Temperature, 'p' = Prerssure, 'q' = Specific humidity "pw": total precipitable water "ref": refractivity "ztd": zenith total delay
&wrfvar20
documentation_url	Òhttp://www2.mmm.ucar.edu/people/wrfhelp/wrfvar/code/trunkÓ
&wrfvar21
time_window_min	"2002-08-02_21:00:00.0000"			start time of assimilation time window used for ob_format=1 and radiances to select observations inside the defined time_window. Note: Start from V3.1, this variable is also used for ob_format=2 to double-check if the obs are within the specified time window.
&wrfvar22
time_window_max	"2002-08-03_03:00:00.0000"			end time of assimilation time window used for ob_format=1 and radiances to select observations inside the defined time_window. Note: Start from V3.1, this variable is also used for ob_format=2 to double-check if the obs are within the specified time window.
&perturbation (settings related to the 4D-Var)
jcdfi_use		false	.true.: Include JcDF term in cost function. .False.: Ignore JcDF term in cost function.				.true.: Include JcDF term in cost function. .False.: Ignore JcDF term in cost function.
jcdfi_diag		1	0: Doesn't print out the value of Jc. 1:Print out the value of Jc.
jcdfi_penalty		10	The weight to Jc term.
enable_identity		.false.	.true.: use identity adjoint and tangent linear model in 4D-Var. .false.: use full adjoint and tangent linear model in 4D-Var.
trajectory_io		.true.	.true.: use memory I/O in 4D-Var for data exchange .false.: use disk I/O in 4D-Var for data exchange

OBSPROC namelist variables.

Variable Names	Description
&record1
obs_gts_filename	name and path of decoded observation file
fg_format	'MM5' for MM5 application, 'WRF' for WRF application
obserr.txt	name and path of observational error file
first_guess_file	name and path of the first guess file
&record2
time_window_min	The earliest time edge as ccyy-mm-dd_hh:mn:ss
time_analysis	The analysis time as ccyy-mm-dd_hh:mn:ss
time_window_max	The latest time edge as ccyy-mm-dd_hh:mn:ss ** Note : Only observations between [time_window_min, time_window_max] will kept.
&record3
max_number_of_obs	Maximum number of observations to be loaded, ie in domain and time window, this is independent of the number of obs actually read.
fatal_if_exceed_max_obs	.TRUE.: will stop when more than max_number_of_obs are loaded .FALSE.: will process the first max_number_of_obs loaded observations.
&record4
qc_test_vert_consistency	.TRUE. will perform a vertical consistency quality control check on sounding
qc_test_convective_adj	.TRUE. will perform a convective adjustment quality control check on sounding
qc_test_above_lid	.TRUE. will flag the observation above model lid
remove_above_lid	.TRUE. will remove the observation above model lid
domain_check_h	.TRUE. will discard the observations outside the domain
Thining_SATOB	.FALSE.: no thinning for SATOB data. .TRUE.: thinning procedure applied to SATOB data.
Thining_SSMI	.FALSE.: no thinning for SSMI data. .TRUE.: thinning procedure applied to SSMI data.
Thining_QSCAT	.FALSE.: no thinning for SATOB data. .TRUE.: thinning procedure applied to SSMI data.
&record5
print_gts_read	TRUE. will write diagnostic on the decoded obs reading in file obs_gts_read.diag
print_gpspw_read	.TRUE. will write diagnostic on the gpsppw obs reading in file obs_gpspw_read.diag
print_recoverp	.TRUE. will write diagnostic on the obs pressure recovery in file obs_recover_pressure.diag
print_duplicate_loc	.TRUE. will write diagnostic on space duplicate removal in file obs_duplicate_loc.diag
print_duplicate_time	.TRUE. will write diagnostic on time duplicate removal in file obs_duplicate_time.diag
print_recoverh	.TRUE will write diagnostic on the obs height recovery in file obs_recover_height.diag
print_qc_vert	.TRUE will write diagnostic on the vertical consistency check in file obs_qc1.diag
print_qc_conv	.TRUE will write diagnostic on the convective adjustment check in file obs_qc1.diag
print_qc_lid	.TRUE. will write diagnostic on the above model lid height check in file obs_qc2.diag
print_uncomplete	.TRUE. will write diagnostic on the uncompleted obs removal in file obs_uncomplete.diag
user_defined_area	.TRUE.: read in the record6: x_left, x_right, y_top, y_bottom, .FALSE.: not read in the record6.
&record6
x_left	West border of sub-domain, not used
x_right	East border of sub-domain, not used
y_bottom	South border of sub-domain, not used
y_top	North border of sub-domain, not used
ptop	Reference pressure at model top
ps0	Reference sea level pressure
base_pres	Same as ps0. User must set either ps0 or base_pres.
ts0	Mean sea level temperature
base_temp	Same as ts0. User must set either ts0 or base_temp.
tlp	Temperature lapse rate
base_lapse	Same as tlp. User must set either tlp or base_lapse.
pis0	Tropopause pressure, the default = 20000.0 Pa
base_tropo_pres	Same as pis0. User must set either pis0 or base_tropo_pres
tis0	Isothermal temperature above tropopause (K), the default = 215 K.
base_start_temp	Same as tis0. User must set either tis0 or base_start_temp.
&record7
IPROJ	Map projection (0 = Cylindrical Equidistance, 1 = Lambert Conformal, 2 = Polar stereographic, 3 = Mercator)
PHIC	Central latitude of the domain
XLONC	Central longitude of the domain
TRUELAT1	True latitude 1
TRUELAT2	True latitude 2
MOAD_CEN_LAT	The central latitude for the Mother Of All Domains
STANDARD_LON	The standard longitude (Y-direction) of the working domain.
&record8
IDD	Domain ID (1=< ID =< MAXNES), Only the observations geographically located on that domain will be processed. For WRF application with XLONC /= STANDARD_LON, set IDD=2, otherwise set 1.
MAXNES	Maximum numbe of domains as needed.
NESTIX	The I(y)-direction dimension for each of the domains
NESTJX	The J(x)-direction dimension for each of the domains
DIS	The grid size for each of the domains. For WRF application, always set NESTIX(1),NESTJX(1), and DIS(1) based on the infomation in wrfinput.
NUMC	The mother domain ID number for each of the domains
NESTI	The I location in its mother domain of the nest domain's low left corner -- point (1,1)
NESTI	The J location in its mother domain of the nest domain's low left corner -- point (1,1). For WRF application, NUMC(1), NESTI(1), and NESTJ(1) are always set to be 1.
&record9
prepbufr_output_filename	Name of the prebufr OBS file.
prepbufr_table_filename	'prepbufr_table_filename' ; not change
output_ob_format	output 1, prebufr OBS file only; 2, ASCII OBS file only; 3, Both prebufr and ASCII OBS files.
use_for	'3DVAR' obs file, same as before, default 'FGAT ' obs files for FGAT '4DVAR' obs files for 4DVAR
num_slots_past	the number of time slots before time_analysis
num_slots_ahead	the number of time slots after time_analysis
write_synop	If keep synop obs in obs_gts (ASCII) files.
write_ship	If keep ship obs in obs_gts (ASCII) files.
write_metar	If keep metar obs in obs_gts (ASCII) files.
write_buoy	If keep buoy obs in obs_gts (ASCII) files.
write_pilot	If keep pilot obs in obs_gts (ASCII) files.
write_sound	If keep sound obs in obs_gts (ASCII) files.
write_amdar	If keep amdar obs in obs_gts (ASCII) files.
write_satem	If keep satem obs in obs_gts (ASCII) files.
write_satob	If keep satob obs in obs_gts (ASCII) files.
write_airep	If keep airep obs in obs_gts (ASCII) files.
write_gpspw	If keep gpspw obs in obs_gts (ASCII) files.
write_gpsztd	If keep gpsztd obs in obs_gts (ASCII) files.
write_gpsref	If keep gpsref obs in obs_gts (ASCII) files.
write_gpseph	If keep gpseph obs in obs_gts (ASCII) files.
write_ssmt1	If keep ssmt1 obs in obs_gts (ASCII) files.
write_ssmt2	If keep ssmt2 obs in obs_gts (ASCII) files.
write_ssmi	If keep ssmi obs in obs_gts (ASCII) files.
write_tovs	If keep tovs obs in obs_gts (ASCII) files.
write_qscat	If keep qscat obs in obs_gts (ASCII) files.
write_profl	If keep profile obs in obs_gts (ASCII) files.
write_bogus	If keep bogus obs in obs_gts (ASCII) files.
write_airs	If keep airs obs in obs_gts (ASCII) files.