there is a relative large gap with LBLRTM when MODIS was simulated

[Sherry Chen]

Hello,
I am using ERA5 data to simulate MODIS 11 micron channels, and the difference between RTTOV and LBLRTM is quite large. It’s about 0.4K, and I’ve tested a lot of points, the difference is always in that range, like a systematic error. However, when I use the same profile information to simulate VIIRS ’11 micron channel, the difference was less than 0.1K.
The input files I used and the output files are as follows:
https://github.com/yunzhiya/RTM-test
LBLRTM output is radiance. When calculating the channel brightness temperature of LBLRTM, the spectral response function I used was downloaded from the RTTOV website(https://nwp-saf.eumetsat.int/downloads/rtcoef_rttov13/ir_srf/rtcoef_eos_1_modis_srf.html) and the formula I used was ∑ (SRF * RADIANCE)/(∑SRF). After obtaining channel radiance, it was converted to brightness temperature according to the look up table.(The difference in radiance between the two models is large(93.99 vs 94.66), even without the conversion to brightness temperature. I don’t think there is mistake in this part, because the simulation of VIIRS through the same process gets very consistent results.)
Because the satellite zenith Angle was set to 0 when I simulated, the gap should be very small. Moreover, using the same profile information and the same processing flow, the gap of VIIRS 11 micron channel is very small. I’ve been working with this for a long time and I don’t know what’s wrong. I would be grateful if you could give me some help.

Thanks a lot!
Sherry

[James Hocking]

Hi Sherry,

I have a few questions about how you are configuring your simulations:

– which version of LBLRTM are you using? And which RTTOV coefficients are you using? For a good comparison it is best to use the same version of LBLRTM with consistent spectroscopy as that used to train the rtcoef file (see the file headers) otherwise you are seeing differences in spectroscopy as well as RT modelling errors.

– are you specifying the same surface emissivity in RTTOV and LBLRTM? How does LBLRTM treat surface reflection of downwelling atmospheric emission? By default RTTOV assumes specular reflection. To rule out any impact from this you might try re-running the simulations assuming a surface emissivity of 1.0.

– just to clarify, you are computing the channel-averaged (SRF-integrated) radiance first, and then calculating the brightness temperature (BT) from that? If so that is correct (rather than computing an SRF-integrated BT).

– when you compute BT from the radiances, are you applying the band correction? This is done within RTTOV if you are using the RTTOV output BTs. To compare these with BTs computed from the LBLRTM radiances you must apply the band correction to the latter to account for the finite spectral bandwidth. This may not have large impact, but should strictly be taken into account.

Best wishes,
James

• This reply was modified 1 year, 7 months ago by James Hocking. Reason: typo RTTOV->LBLRTM

[Sherry Chen]

Hi James,
Thank you for your prompt reply.
I ran these modes with the same emissivity. There was still a 0.4K difference when it was set to 1.0, so it felt like a systematic error. I compute the channel-averaged (SRF-integrated) radiance first, and then calculating the brightness temperature (BT) from that. The version of LBLRTM I use is 12.13, and the corresponding RTTOV coefficient is generated by version 12.8. I think this may be the source of error, and band correction has not been carried out. I will download the corresponding version of LBLRTM and carry out band correction to re-confirm.
I have another problem, that is, when using the data of ERA5 to operate the mode, I set the number of pressure levels as 37 in the input file of RTTOV, and directly used the pressure value, temperature, water vapor ozone and other values on the 37 pressure levels data of ERA5. The following surface data are obtained from the single level data of ERA5, and 2m p is the value of surface pressure.
! Near-surface variables:
! 2m T (K) 2m q (kg/kg) 2m p (hPa) 10m wind u (m/s) 10m wind v (m/s) wind fetch (m)
!
290.6833 0.0101798 1004.30 1.146 2.3662 100000.
!
! Skin variables:
! Skin T (K) Salinity FASTEM parameters for land surfaces
!
288.3574 35.0 3.0 5.0 15.0 0.1 0.3

Then when I wrote LBLRTM TAPE5 input files, I typed the following:
HI=1 F4=1 CN=1 AE=0 EM=1 SC=3 FI=0 PL=0 TS=0 AM=1 MG=0 LA=0 OD=0 XS=0 00 00
781.646 1400.894 -0.001 REJ=0
288.4 -1.000 -1.000
0 2 38 1 1 3 1 00 0.000 0.000
824.0000 0.0000 180.0000
0.000 0.117 0.336 0.558 0.785 1.017 1.254 1.497
1.745 2.000 2.261 2.528 3.089 3.685 4.321 5.005
5.740 6.538 7.406 8.361 9.430 10.654 11.344 12.108
12.972 13.965 15.132 16.545 18.790 20.922 24.192 26.840
31.457 33.897 36.266 39.968 43.030 48.355
38 INPUT FOR CAMEX
0.000 1004.303 288.357 AA L CAC
1.01798220e+01 3.68500000e+02 7.00065575e-05
0.117 1000.000 292.868 AA L CAC
1.01798220e+01 3.68500000e+02 7.00065575e-05
…

I set the number of atmospheric profile boundaries to 38, and the boundary temperature at H2 (0.00km) used the skin temperature from ERA5 single level data. On the basis of using the pressure level data of ERA5 in layer 37, a layer of surface boundary is added.(The height is 0.00km. The pressure is the surface pressure from the ERA5 single level data; the temperature is the skin temperature; the other absorbed gases used the data of previous layer.)

However, it is stated in lblrtm instruction that these inputs are the corresponding data at layer boundary. It seems that there is also a difference between level and layer in CRTM. I am quite confused about this part, and do not know whether the input file I wrote was correct. So if you have used ERA5 data to run these model, could you please give me some suggestions on input files or point out my mistakes?

Best Regards,
Sherry

[James Hocking]

Hi Sherry,

Sorry for the delay in replying. The differences in spectropscopy could potentially explain the differences. The band correction may only have a small effect in these channels.

I have only a small amount of experience running LBLRTM, so I have asked my colleagues who use LBLRTM for generating the VIS/IR RTTOV coefficients to have a look at your question. They will reply when they are able.

Best wishes,
James

[Author]

Posts