Test the UFO of HDOB with HAFS

[DanWu-HRD]

This is to record the process of assimilating HDOB (high density recconnaissance obs) using JEDI 3DEnVar with HAFS background.

[DanWu-HRD]

1. Use high level python API to create ioda nc format obs file from hdobbufr: hafs.t12z.hdob.tm00.bufr_d

add decomposed wind component windEastward and windNorthward
add specificHumidity
add quality control marker for each ObsValue
Used python script and yaml files:
/scratch2/AOML/aoml-hafs1/Dan.Wu/test/Test_hdob2ioda/bufr2ioda_aircft_hdob.yaml
/scratch2/AOML/aoml-hafs1/Dan.Wu/test/Test_hdob2ioda/bufr2ioda_hdobbufr_fl.py

The output ioda file:
hafs.t12z.hdob.tm00.nc

[DanWu-HRD]

The comparison of hofx generated by GSI and JEDI.
Currently simulated variables: flight level t, q and uv.

1. flight level air temperature
   
   

2. flight level q
   
   

3. flight uv
   
   

[DanWu-HRD]

For air temperature, there is an obvious difference in the input obs value between GSI and JEDI. Looking into the GSI source code, and found that the air temperature obs is a combination of the observed temperature (t) and the virtual temperature (tv). When there's moisture observation (td or RH), the air temperature is replaced by tv.

So, the new variables virtualTemperature and virtualTemperatureFlag were calculated and added in the output ioda file.
The virtualTemperatureFlag was set to 0 if tv exists.

[DanWu-HRD]

When the simulated variable is sensible temperature (airTemperature), the hofx differences are small for data labeled as t, but has a negative bias for data labeled as tv.

When the simulated variable is virtual temperature, the hofx differences are small for data labeled as tv, but has a positive bias for data labeled as t.

A simulation strategy needs to be designed to reconcile these two variables.

[DanWu-HRD]

For wind component, the hofx differences between GSI and JEDI are unreasonably large. And these data are all in the inner core region. Vertically they are at around the flight level and the model top.

[DanWu-HRD]

Comparing the geovals of the data in the red circle. The pressure differences in the vertical profile are very small. However, the geovals have large difference at the observation level. The guess is that, the hofx differences come from the geovals difference.

For some other windNorthward data points:
The red dots are the hofx difference, and the vertical profiles are the geovals difference between JEDI and GSI. The the hofx differences are comparable to the geovals difference at the obs levels.

[ShunLiu-NOAA]

@DanWu-HRD Do you get a chance to check the difference in QC between GSI and JEDI?

[ShunLiu-NOAA]

@JingCheng-NOAA Do you have the same issue in your geovals?

[JingCheng-NOAA]

@JingCheng-NOAA Do you have the same issue in your geovals?

The geovals difference are largest near model top in my case. But it is not as big as in the HDOB data.

[DanWu-HRD]

@JingCheng-NOAA Do you have the same issue in your geovals?

The geovals difference are largest near model top in my case. But it is not as big as in the HDOB data.

Is the geovals difference comparable to the hofx difference in your case?

[DanWu-HRD]

The large hofx-difference is located in a high-gradient region.

[DanWu-HRD]

Changing the grid_ratio_fv3_region from 2 to 1, can reduce the hofx differences between gsi and jedi, but not totally eliminate them.

[DanWu-HRD]

The distribution of hofx difference for grid_ratio 2 v.s. grid_ratio 1.

Eastward wind

virtual Temperature

specific Humidity