Skip to content

Commit

Permalink
Merge pull request #732 from NCAR/GOES_obs_converter
Browse files Browse the repository at this point in the history
Remove offset on GOES obs converter
  • Loading branch information
hkershaw-brown authored Oct 22, 2024
2 parents 5c4c880 + a58694a commit e218864
Show file tree
Hide file tree
Showing 7 changed files with 264 additions and 166 deletions.
1 change: 1 addition & 0 deletions .gitignore
Original file line number Diff line number Diff line change
Expand Up @@ -168,6 +168,7 @@ L1_AMSUA_to_netcdf
convert_airs_L2
convert_amsu_L1
convert_L2b
convert_goes_ABI_L1b

# Test programs built by developer_tests
rttov_test
Expand Down
5 changes: 5 additions & 0 deletions CHANGELOG.rst
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,11 @@ individual files.

The changes are now listed with the most recent at the top.

**October 22 2024 :: Bug-fixes: WRF and GOES. Tag 11.8.2**

- Force THM to be the WRF-DART temperature variable
- Remove offset on GOES observation converter

**September 27 2024 :: MOM6 mask bug-fix. Tag 11.8.1**

- Fix for MOM6 CESM3 workhorse 2/3 degree grid TL319_t232 to
Expand Down
2 changes: 1 addition & 1 deletion conf.py
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,7 @@
author = 'Data Assimilation Research Section'

# The full version, including alpha/beta/rc tags
release = '11.8.1'
release = '11.8.2'
root_doc = 'index'

# -- General configuration ---------------------------------------------------
Expand Down
135 changes: 118 additions & 17 deletions observations/forward_operators/obs_def_rttov_mod.rst
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@ MODULE ``obs_def_rttov_mod``
Overview
--------

DART RTTOV observation module, including the observation operators for the two primary
The DART RTTOV observation module, including the observation operators for the two primary
RTTOV-observation types -- visible/infrared radiances and microwave
radiances/brightness temperatures.

Expand Down Expand Up @@ -46,25 +46,124 @@ Although a model may not have the necessary inputs by itself,
the defaults in RTTOV based on climatology can be used.
The impact on the quality of the results should be investigated.

The quanities for each observation type are defined in obs_def_rttov{13}_mod.f90, like so:
The quanities for each observation type are defined in obs_def_rttov{13}_mod.f90, for example:

.. code::
! HIMAWARI_8_AHI_RADIANCE, QTY_RADIANCE
! HIMAWARI_9_AHI_RADIANCE, QTY_RADIANCE
If you want to change the quantity associated with an observation, for example, if you want
to assimilate HIMAWARI_8_AHI_RADIANCE as QTY_BRIGHTNESS_TEMPERATURE, edit the QTY
in obs_def_rttov{13}_mod.f90 and rerun quickbuild.sh.
to assimilate HIMAWARI_9_AHI_RADIANCE as QTY_BRIGHTNESS_TEMPERATURE, edit the QTY
in obs_def_rttov{13}_mod.f90 and rerun quickbuild.sh. Although both spectral radiance
(mW/cm/m^2/sr) and brightness temperature (Kelvin) quantify the same emitted/reflected
radiance from the atmosphere, the tendency for brightness temperatures to adhere closer
to a gaussian distribution may improve the quality of the assimilation if using
a DART filter type that depends on Gaussian assumptions (e.g. EAKF). This is
an ongoing area of research.


Known issues:

- DART does not yet provide any type of bias correction
- Cross-channel error correlations are not yet supported. A principal component approach has been discussed. For now,
the best bet is to use a subset of channels that are nearly independent of one another.
- Vertical localization will need to be tuned. Turning off vertical localization may work well if you have a large
number of ensemble members. Using the maximum peak of the weighting function or the cloud-top may be appropriate.
There are also other potential approaches being investigated.
RTTOV Metadata
---------------

The RTTOV module ingests metadata from the ``obs_seq.out`` file in order to calculate the
expected observed radiance. For example, a single ``HIMAWARI_9_AHI_RADIANCE``
observation in units of brightness temperature (Kelvin) looks like the following:

.. code::
OBS 1
288.370817896852
0.000000000000000E+000
-1 2 -1
obdef
loc3d
1.766273140907288 0.1535889655351639 34000.00000000000 2
kind
304
visir
100.500000000000 46.6700000000000 -888888.000000000
-888888.000000000
31 9 56 8
-888888.000000000
1
0 154166
1.00000000000000
Please note, that in this example the radiance observation was assigned a vertical level (34000 Pa)
with the ``VERTISPRESSURE`` (integer = 2) vertical coordinate.
Although radiance/BT observations are technically representative of the entire atmospheric
column and not a single vertical level, for some applications where specific channels (wavelength)
are especially sensitive to constituents at particular atmospheric levels, assigning
a vertical level to the observation may improve the skill of the assimilation forecast. This is an ongoing
area of research. As an alternative, it is also common to leave the vertical level
as undefined (VERTISUNDEF, integer = -2), however, this limits the ability to vertically
localize the impact of the observation on the model state.

In this example where the observation is infrared (IR) radiance, the metadata is located after
the ``visir`` line (Note: for microwave observations the metadata would follow ``mw``).
The metadata includes the azimuth and elevation angle of the satellite and the sun respectively. In this instance the sun azimuth/elevation are given missing values (-888888) because
solar reflectance has no impact on an IR radiance observation. Also note, the observation
provides a 4 integer description (31/9/56/8) of the platform/satellite/sensor/channel
combination specific to this satellite observation. For more information on this
metadata refer to this GOES observation converter example here:
:doc:`../../observations/obs_converters/GOES/README`

.. Important ::
It is important that the user confirms the satellite integer metadata within
the obs_seq.out file matches the metadata within rttov_sensor_db.csv. Furthermore,
confirm that the channel as defined in the obs_seq.out file matches the channel
available in the RTTOV coefficient file (.dat). See next section for more information.
RTTOV coefficient files
-----------------------

The RTTOV coefficent file (.dat) contains the appropriate parameter values for a specific satellite
radiance observation. The DART file (``rttov_sensor_db.csv``) refers to the RTTOV coefficent
file. For the ``HIMAWARI_9_AHI_RADIANCE`` observation type, for example, the following information
is provided within ``rttov_sensor_db.csv``:

.. code::
HIMAWARI_9_AHI 31 9 56 ir rtcoef_himawari_9_ahi.dat
The coefficent file (.dat) is included with the RTTOV installation and can be found at the
path ``${RTTOV_install}/rtcoef_rttov13/rttov9pred54L/rtcoef_himawari_9_ahi.dat``. This file
should be included in your run folder at runtime. Additional coefficent files for a given
satellite sensor may be required.

It is good practice to always view your coefficent file (.dat) to confirm that the
channels listed in the file match the channel from the ``obs_seq.out`` file. The coefficent
file will include a list of channels (wavebands) with the associated wavelength (microns).


.. Important ::
The RTTOV package includes multiple coefficent files (e.g. all wavelengths, IR only, etc.) that
contain the appropriate parameter data for each satellite/sensor/channel combination. Whether
the file contains all wavelengths versus only IR wavelengths is **extremely important** because
it will shift the value of the channel number. Recommended practice is to choose a coefficient file
with all channels included. If, on the other hand, you subset your coefficent file to only include
IR channels, you should edit your observation converter such that the channels match.
If RTTOV always returns expected observations of radiance = 0, or if the prior expected radiance
is unusually biased from your prior, this could be a sign there is a mismatch between the
obs_seq.out channel and the coefficient file channel.
Known issues:
-------------
- DART does not provide any type of observation bias correction. It may be appropriate to preprocess your radiance
observations to remove systematic bias before assimilation, using techniques such as cumulative distribution
function (CDF) matching.
- Cross-channel error correlations are not supported. A principal component approach has been discussed. For now,
we recommend to use a subset of channels that are nearly independent of one another.
- Vertical localization will need to be tuned based on the research application. Turning off vertical localization
may work well if you have a large number of ensemble members. Using the maximum peak of the channel weighting
function or the cloud-top height to set a vertical location for an observation may be appropriate.


The namelist ``&obs_def_rttov_mod_nml`` is read from file ``input.nml``. Namelists start with an ampersand '&'
Expand Down Expand Up @@ -159,11 +258,13 @@ RTTOV v12 Namelist
+------------------------+--------------------+----------------------------------------------------------------------+
| Item | Type | Description |
+========================+====================+======================================================================+
| rttov_sensor_db_file | character(len=512) | The location of the RTTOV sensor database. The format for the |
| | | database is a comma-separated file. The columns of the database are |
| | | the DART observation-kind, the platform/satellite/sensor ID, the |
| | | observation type, the coefficient file, and a comma-separated list |
| | | of RTTOV channels to use for this observation type. |
| rttov_sensor_db_file | character(len=512) | The location of the DART file with RTTOV sensor metadata. The format |
| | | is a comma-separated file. The columns are the DART |
| | | observation type, the platform/satellite/sensor ID, the |
| | | wavelength band, the coefficient file, and a comma-separated list |
| | | of RTTOV channels to use for this observation type. The default file |
| | | does not provide a list of channels, thus default behavior is to |
| | | make all channels available. |
+------------------------+--------------------+----------------------------------------------------------------------+
| first_lvl_is_sfc | logical | Whether the first level of the model represents the surface (true) |
| | | or the top of the atmosphere (false). |
Expand Down
Loading

0 comments on commit e218864

Please sign in to comment.