source: lmdz_wrf/trunk/WRFV3/test/em_real/README.obs_fdda @ 667

General description
-------------------
Features and advantages of observational nudging are discussed in (*) below. 
The method uses relaxation terms based on the model error at observational 
stations, and the relaxation is such as to reduce this error.
Each observation has a radius of influence, a time window, and a relaxation 
time scale determined by user-specified input. These determine where, when, 
and how much it affects the model solution.  Typical model grid points may 
be within the radius of influence of several observations, and their 
contributions are weighted according to the distance from the observation(s). 
Before performing obs-nudging, you will need to generate an observation 
input file for each WRF domain. The observation file(s) contain chronological 
lists of the 3D positions and values of each observation, in a specific format.
It is critical that your observations be listed in chronological time order!

*  Liu, Y., A. Bourgeois, T. Warner, S. Swerdlin and J. Hacker, 2005: An
  implementation of obs-nudging-based FDDA into WRF for supporting
  ATEC test operations. 2005 WRF user workshop. Paper 10.7.


How to use the obs-data converter
-------------------------------------
A utility program for converting observation data to the format required by 
WRF has been provided (RT_fdda_reformat_obsnud.pl). The converter assumes 
that your observation data is in standard LITTLE_R format. 

To convert your data that is in LITTLE_R format:

  RT_fdda_reformat_obsnud.pl yourfilename

where "yourfilename" is the obs-data in LITTLE_R format. The converter will 
produce a file named yourfilename.obsnud, in the format required by the WRF
model.

Note that during the conversion process:

  1). P,T,U,V and RH fields are extracted.

  2). U and V are assumed to be the wind components rotated to
      the model map-projection (see 3DVAR and MM5 Little_R).

  3). SPD, DIR and Td fields are ignored.

  4). For upper-air data, currently WRF nudging only takes
      those data with valid pressure records. For obs with
      height levels (e.g. wind profilers data), users need to
      calculate or estimate the pressure value. Inaccurate
      estimate of pressure will lead to bad data assimilation.


Naming your obs-nudge input files
---------------------------------
After you have converted your obs data file to the proper format for WRF, 
you will need to rename it according to the naming convention for the WRF
domain on which the obs-nudging is to be performed. For example, for
observations to be used in Domain 1, use the naming convention OBS_DOMAIN101,
for Domain 2, OBS_DOMAIN201, etc. 

These files must be present in your WRF run directory, along with the usual
WRF input and boundary files.



How to activate obs-nudging
---------------------------
To activate the observational nudging option in WRF, you will need to set
the obs_nudge_opt flag(s) in the WRF "fdda" namelist. Note that there is
a unique flag for each WRF domain in which you want to activate obs-nudging.

To activate the print statements within the obs-nudging subroutines ERROB,
NUDOB, and IN4DOB, set the respective print flags obs_ipf_errob, 
obs_ipf_nudob, and obs_ipf_in4dob to ".true." You can then easily verify that 
you have activated observational nudging by observing text in your WRF
"standard out" that tell you how many obs stations are being processed at
given model timesteps. This information will look something like:

0****** CALL IN4DOB AT KTAU =     8 AND XTIME =      24.00:  NSTA =   11040 ******
++++++CALL ERROB AT KTAU =     8 AND INEST =  1:  NSTA = 11040 ++++++

These lines will print out for each nest in which you have activated nudging, 
while nudging is active on that domain.

Below is an example of a namelist set up to activate obs-nudging on domains
1, 2, and 3:

&fdda
obs_nudge_opt                       = 1,1,1,0,0   
max_obs                             = 150000,
fdda_start                          =     0.,     0.,     0.,     0.,     0.
fdda_end                            = 99999., 99999., 99999., 99999., 99999.
obs_nudge_wind                      = 1,1,1,1,1
obs_coef_wind                       = 6.E-4,6.E-4,6.E-4,6.E-4,6.E-4
obs_nudge_temp                      = 1,1,1,1,1
obs_coef_temp                       = 6.E-4,6.E-4,6.E-4,6.E-4,6.E-4
obs_nudge_mois                      = 1,1,1,1,1
obs_coef_mois                       = 6.E-4,6.E-4,6.E-4,6.E-4,6.E-4
obs_rinxy                           = 240.,240.,180.,180,180
obs_rinsig                          = 0.1,
obs_twindo                          = 0.6666667,0.6666667,0.6666667,0.6666667,0.6666667,
obs_npfi                            = 10,
obs_ionf                            = 2, 2, 2, 2, 2,
obs_idynin                          = 0,
obs_dtramp                          = 40.,
obs_prt_freq                        = 10, 10, 10, 10, 10,
obs_prt_max                         = 10
obs_ipf_errob                       = .true.
obs_ipf_nudob                       = .true.
obs_ipf_in4dob                      = .true.
obs_ipf_init                        = .true.

In addition, add the following in &time_control:

auxinput11_interval_s               = 180, 180, 180, 180, 180,
auxinput11_end_h                    = 6, 6, 6, 6, 6,


*********************
    NEW FOR V3.1
*********************

1) Enhanced diagnostics
  --------------------
  For version 3.1, diagnostics have been enhanced to allow the user to verify
  grid placement for observations throughout the model run. For v3.1, the fdda namelist
  variable "nobs_obs_prt" is obsolete, and has been replaced by the two namelist
  variables:

  obs_prt_max  - maximum allowed obs entries in diagnostic printout (integer)
  obs_prt_freq - frequency in obs index for diagnostic printout (max_domains integer)

  For example, specifying:

  obs_prt_max         = 5,
  obs_prt_freq        = 1000, 500, 100,

  allows up to 5 observations and their locations to be reported for each model timestep
  at which the obs are read and weights calculated (see obs_ionf). For this example, the
  obs are reported for domain 1 with an obs-index frequency of 1000, on domain 2 with a
  frequency of 500, and on domain 3 with a frequency of 100. Below is an example of the
  initial obs diagnostic report produced for each nest, using the namelist values above.

  ++++++CALL ERROB AT KTAU =     0 AND INEST =  1:  NSTA = 17090 ++++++

  REPORTING OBS MASS-PT LOCS FOR NEST    1 AT XTIME=     0.0 MINUTES
  FREQ=1000, MAX=    5 LOCS, NEWLY READ OBS ONLY, -999 => OBS OFF PROC

      OBS#     I       J       K     OBS LAT  OBS LON   XLAT(I,J)  XLONG(I,J)  TIME(hrs)
          1   5.282   2.658   1.000   27.580  -97.220     27.580    -97.220     0.00
       1001  47.851   9.468  32.377   30.380  -84.360     30.380    -84.360     0.00
       2001  40.728  19.834  26.618   33.160  -86.700     33.160    -86.700     0.00
       3001  60.866  30.631 -99.000   36.080  -79.950   -999.000   -999.000     0.00
       4001  73.216  38.315 -99.000   37.930  -75.480   -999.000   -999.000     0.00
 ...

  ++++++CALL ERROB AT KTAU =     0 AND INEST =  2:  NSTA =  3504 ++++++

  REPORTING OBS MASS-PT LOCS FOR NEST    2 AT XTIME=     0.0 MINUTES
  FREQ= 500, MAX=    5 LOCS, NEWLY READ OBS ONLY, -999 => OBS OFF PROC

      OBS#     I       J       K     OBS LAT  OBS LON   XLAT(I,J)  XLONG(I,J)  TIME(hrs)
          1  62.717   3.505   1.000   34.600  -78.580   -999.000   -999.000     0.00
        501  45.014  30.857  32.102   37.200  -80.410     37.200    -80.410     0.00
       1001  15.585  54.082 -99.000   39.410  -83.810   -999.000   -999.000     0.00
       1501  18.822  78.673   1.000   41.690  -83.400   -999.000   -999.000     0.00
       2001  67.015 103.930   1.000   43.830  -77.150   -999.000   -999.000     0.00
 ...

  ++++++CALL ERROB AT KTAU =     0 AND INEST =  3:  NSTA =   606 ++++++

  REPORTING OBS MASS-PT LOCS FOR NEST    3 AT XTIME=     0.0 MINUTES
  FREQ= 100, MAX=    5 LOCS, NEWLY READ OBS ONLY, -999 => OBS OFF PROC

      OBS#     I       J       K     OBS LAT  OBS LON   XLAT(I,J)  XLONG(I,J)  TIME(hrs)
          1  48.735   3.730   1.000   38.220  -76.040   -999.000   -999.000     0.00
        101  10.646  25.032  34.295   38.980  -77.460     38.980    -77.460     0.00
        201  25.010  28.538   9.296   39.050  -76.880     39.050    -76.880     0.00
        301  28.675  46.438   1.000   39.590  -76.670   -999.000   -999.000     0.15
        401  12.030  38.804   1.000   39.400  -77.360     39.400    -77.360     0.30


  With this report, the user can verify the WRF mapping of each reported observation.
  The report shows:

  (1) the real-valued WRF grid (I,J,K) location to which the obs is mapped,
  (2) the input obs latitude and longitude coordinate (OBS LAT, OBS LON), and
  (3) the corresponding model lat, lon coordinate (XLAT(I,J), XLONG(I,J)) for (I,J,K) 
  (4) the input obs time in hours into the run

  Note that each processor produces a report (rsl.out.0000, rsl.out.0001, etc) for the
  same set of observations, but that full information for an individual observation is
  only available in the report from the processor whose grid "patch" contains that
  observation. A -999.000 entry in the XLAT and XLONG columns indicates that the
  observation is not located on the processor producing the report. In the example
  above, the rsl.out.0000 file indicates that OBS#1 for nest 3 is not located on the
  WRF patch handled by process 0. OBS#1 falls on the patch for process 1, and the
  report for nest 3 from the rsl.out.0001 file looks like:

  ++++++CALL ERROB AT KTAU =     0 AND INEST =  3:  NSTA =   606 ++++++

  REPORTING OBS MASS-PT LOCS FOR NEST    3 AT XTIME=     0.0 MINUTES
  FREQ= 100, MAX=    5 LOCS, NEWLY READ OBS ONLY, -999 => OBS OFF PROC

      OBS#     I       J       K     OBS LAT  OBS LON   XLAT(I,J)  XLONG(I,J)  TIME(hrs)
          1  48.735   3.730   1.000   38.220  -76.040     38.220    -76.040     0.00
        101  10.646  25.032 -99.000   38.980  -77.460   -999.000   -999.000     0.00
        201  25.010  28.538 -99.000   39.050  -76.880   -999.000   -999.000     0.00
        301  28.675  46.438   1.000   39.590  -76.670   -999.000   -999.000     0.15
        401  12.030  38.804   1.000   39.400  -77.360   -999.000   -999.000     0.30

  which confirms this.

  Note that the namelist variable obs_prt_max can be declared as large as desired, but
  a value larger than 10^4 might significantly affect performance.

  CAUTION! Depending on how many observations are in your obs-nudge input file, be
  aware that your diagnostic output can potentially produce obs_prt_max/obs_prt_freq
  lines of output for each domain, for each obs input step!

2) Option to input wind vectors in Earth coordinates

  In previous WRF versions, input wind vectors were required to be in WRF grid-relative 
  coordinates (that is, already rotated from Earth coordinates to the WRF grid). In
  WRFV3.1, the user may specify winds in Earth coordinates and have the WRF model
  internally rotate them to the WRF grid. To activate this capability, the user must
  specify a u- and v-component QC flag value of 129 for each wind vector that is to be
  rotated. Otherwise, the model assumes the wind vector to be WRF grid-relative. For
  future releases, this option will possibly be activated by a namelist variable.

3) Option for Obs-in-height

  I previous WRF versions, the vertical model coordinate for an upper air observation
  is determined by its input pressure field. In WRFV3.1, the obs height field may be
  used instead. In the new implementation, if the obs pressure field contains "missing
  value" -888888 and the obs height field good, the vertical model coordinate for the 
  obs is determined using a geopotential height calculation.

4) U-, V-, and T-ratios are calculated in all surface schemes

  In previous WRF versions, calculations for u-,v-, and t- ratios (from 10 meter winds
  and 2 meter temperatures) are only done in the SFCLAY surface scheme. These ratios
  are now calculated for all surface scheme options. (The ratios are used in the
  obs-nudging routine to correct obs to model sigma level using reverse similarity
  theory.)