Switching from assimilating microwave antenna temperatures to brightness temperatures #68
Comments
|
Plotted are assimilated brightness temperatures that were converted from antenna temperatures (AMSU-A channel 9).
|
|
Antenna temperature conversion appears to be working. Will verify for other satellites, sensors, channels. Will investigate outliers. |
|
Issues have been found with metop-a AMSU-A (figure 1). Strings of swaths have extreme differences between native and converted antenna/brightness temperatures. These observations are clumped in the same swath, in consecutive swaths, and exhibit a scan/zenith angle dependence of the exact difference. The scan/zenith angle dependence resembles the pattern of the difference between antenna and brightness temperatures from ATMS, which provides both. (figure 2) The working theory is that there are errors in the reporting of satellite and sensor parameter values used in the conversion between antenna and brightness temperature. |
|
This is to expound upon the previous comment. In the one 6-hour period we have investigated, there are instances of a significant inconsistency between co-located observations across the EARS direct broadcast and the normal feed, exclusively for amsua-a on metop-a (all other amsu-a/-b, mhs and atms appear fine in this 6-hour period). These inconsistencies exist whether we convert the normal feed antenna temperature to brightness temperature (figure 1; channel 11), or convert the EARS direct broadcast brightness temperature to antenna temperature. There are patterns to which observations have the significant inconsistency. The vast majority of observations are in a swath in which all other observations have the inconsistency. The vast majority of swaths with inconsistent observations are adjacent to other such swaths. There is also a distinct pattern of the magnitude of the inconsistency relating to the field of view or scan angle; at least for channel 11, they are higher at the the moderate scan angles. It was this that prompted experimenting with messing with the FOV in the conversion from normal feed antenna temperature to brightness temperature. We set the FOV index to always be 1 instead of the actual value. This of course changes the cold space, earth view, and platform efficiency values used in the antenna/brightness temperature conversion. The results are shown in the figure below (channel 11). Looking at southern Indonesia, Australia, and the Pacific Ocean as areas which were showing inconsistencies, we see that the magnitude of the inconsistencies are substantially reduced across the observations at moderate scan angle, generally between -0.1 K and 0.1 K. We hope that these results would help with any efforts in identifying a cause of the problem. |
|
O minus B suggest that there is an with the EARS direct broadcast. In the aforementioned regions where the EARS brightness temperatures are greater than the normal feed converted brightness temperatures, there is an overall average positive O-B (figure below). This feature is not present with the normal feed observations that have been converted in the GSI to brightness temperature (figure below). |
|
|
|
Metop-A AMSU-A disparities have been confirmed to be due to v1/v2 satellite antenna correction versions. The value of BUFR mnemonic SACV has been used to decide whether to use v1 coefficients from CRTM SpcCoeff file and v2 coefficients (for channel 11, which were hard-coded) for a given observation. The read_bufrtovs subroutine has been modified to support to read and store multiple sets of antenna correction patterns from respective SpcCoeff files, using naming convention <Sensor_ID>.SpcCoeff.bin and <Sensor_ID>_v#.SpcCoeff.bin with # able to range from 2 to 9 (though only a version 2 is ever likely). The functionality of reading multiple files been tested successfully except actually having a unique <Sensor_ID>_v#.SpcCoeff.bin file.
|
|
We are awaiting SpcCoeff files from JCSDA to complete the assessment and proceed with a parallel experiment |
|
The metop-a amsua SpcCoeff version 2 file from JCSDA was delivered on 13 April 2021. The GSI has been successfully modified to read and use multiple versions of the SpcCoeff file as appropriate. Normal feed observations are converted to brightness temperature with the latest version of antenna correction coefficients. Direct broadcast observations with SACV not of the latest version are converted to antenna temperature with the SACV version of ACCs, and then converted back to brightness temperature with the latest version of ACCs. As an example, shown are channel 9 brightness temperature differences between converted normal feed observations and direct broadcast observations. |
|
The parallel experiment is near completion. There are some short-term forecast degradation in upper-level winds shown in the "sanity check" plots, but longer term forecasts show improvement. |
|
The original TaTb experiment was extended to mid-October. The upper-level wind degradation has been lessened somewhat as well as the 8- to 10-day height anomaly correlation improvement.
A new experiment has been started with resetting the bias correction coefficients of all satellite sensors to 0, and using CRTM v2.4.0. Compared to the original experiment, results so far are distinctly different and mixed in improvement/degradation |
|
The parallel experiment with bias correction coefficients of all satellite sensors to 0, and using CRTM v2.4.0, is near completion. Compared to the original experiment, results are distinctly different and mixed in improvement/degradation, though perhaps more improvement. |
|
The CRTM 2.4.0 experiment also has overall reduced biases, especially in temperature. |
|
The experiment with resetting the bias correction coefficients of all satellite sensors to 0, and using CRTM v2.4.0 (TaTb4b_240) was stopped on October 2. Fit-to-obs show neutral to positive results in Z (pictured), W, Q and T. In TaTb4b_240, the GFDL cloud fraction was calculated and input into the CRTM. Especially without the inclusion of precipitation in this experiment, these cloud fractions are generally small. We have begun a new experiment in which CRTM 240 is used, but all clouds with water content greater than zero have the cloud fractions set to one. |
|
An experiment was run in which CRTM 240 is used, but all clouds with water content greater than zero have the cloud fractions set to one. Results as shown by the sanity check plots are similar to using the calculated cloud fractions. The experiment was shortened to just a few days.
|
|
We await results from experiments which built-upon these methods to implement this into v16.x. |
|
Regression test results: Test project /scratch1/NCEPDEV/da/Scott.Sieron/GDAS/GSI/build 32% tests passed, 13 tests failed out of 19 Total Test time (real) = 4509.56 sec The following tests FAILED: |
|
@ScottSieron-NOAA , do we understand why these failures occurred? I assume these are because we expect the results to change for the microwave sounders (and only the microwave sounders), but can you confirm that the differences are as expected? |
|
Hi @ScottSieron-NOAA. Since these changes were merged to the authoritative repository at c13361c with PR #350, I am checking to see if it is alright to close this issue. |
|
@MichaelLueken-NOAA, @ScottSieron-NOAA is not longer with EMC. Please feel free to close this issue. |
|
@ADCollard Thanks for the update. Closing issue now. |
This is to switch from assimilating satellite microwave antenna temperature observations to brightness temperatures.
As it is now, the bias between the observed antenna temperature and the simulated brightness temperature is being handled by the variational bias correction. For the platforms and feeds which have only antenna temperature, an accurate formula for converting antenna temperatures to brightness temperatures will be used. Doing this and assimilating brightness temperatures removes a burden of the variational bias correction.
Many (or all) other centers are assimilating brightness temperatures.
The text was updated successfully, but these errors were encountered: