CO2 optical depths

Sergio Machado

1 Location 1 : Code

/asl/data/hitran/H2016/LineMix/LM_calc_CO2_2017_kcarta_5ptboxcar_newdir.for or earlier versions.

2 Location 2 : Results


3 Overview of CO2 databases

Our CO2 spectroscopy (line mixing + Birnbaum) was developed around 2000, and may be out of date. The Birnbaum and line-mixing parameters were fit to HITRAN 1998 CO2. We are no longer pursuing this approach.

J.M. Hartmann made his line-mixing code available in 2008 or so. Our evaluation of it produced similar results to kCARTA, but the differences did not seem to warrant our adoption of it.

We now have LBLRTM versions 12.4 12.8 running at UMBC. More recently we started evaluating LBLRTM and decided to use it for CO2 optical depths in our more recent version of SARTA. In addition, LBLRTM also implemented CH4 line-mixing, which we have never worked on, so the more recent kCARTA CO2 and CH4 database uses the LBLRTM CO2 CH4 ods, and our run8 optical depths for every other gas (using the HITRAN 2016 database).

In 2016, the group working on the HITRAN database at the Harvard University Center For Astrophysics released a new LineMixing code, based on the latest coefficients by Niro and Lamoreaux. This webpage mainly details our ongoing comparisons to the latest version of this database, and the accompanying LM code.

As with all the monochromatic optical depths used by kCARTA, both the original HITRAN LM code, and my modifications to the code, have been run through the US Standard profile for 11 temperature offsets, and then compressed so they can be used with kCARTA.

4 Overview of UMBC Changes to the HITRAN LM code

A : Simple changes (should not significantly affect output ODs)

  1. File access changed from /data_new to ../LineMix2/Updates_LMCO_{2}_HITRANonline2018/data_new
  2. In updated the Qtips coeffs and the isotope masses
  3. Now using QT_CO_{2}_sergio instead of QT_CO_{2}

B : Physics changes (could affect output ODs if not done correctly)

  1. Changed code so it not only uses bands between SigMinR and SigMaxR but also uses sigMinR-xfar to SigmaxR+xfar, xfar=500.0 in run8, but I used 1250 here, maybe overkill?
  2. Changed rdmult from 30 to 1000000 (distance in multiples of GamD after which you switch from voigt to lorentz). See /home/sergio/SPECTRA/doppler_widths_wavenumber.m (typical doppler widths ~ 1e-3) at 667 cm-1, typical width = 5e-4 ==> 5e-4 * 1e6 = 500 cm-1
  3. The original LM code only used \(\nu \times (1-exp(-c_2 \nu/T))\) as an adjustment. I updated code so that every line in the CO2 database has the following radiation term
\begin{equation} \nu_i \times \frac{(1-e^{-c_2 \nu_i/T})} {(1-e^{-c_2 \nu_i/T_o})} \end{equation}

where \(T_o\) is 296 K.

The comparisons in Figure 1 below show comparisons between the original and modified HITRAN LM codes, compared to LBLRTM 12.8. The results are averaged over the UMBC 49 regression profiles

The upper panels of Figure 1 are comparisons using my modifications to the LM package, while the lower panels are comparison using the original LM package.


Figure 1: UL : LBLRTM 12.8 and modified HITRAN LM code. UR : bias/std dev (LBLRTM 12.8-modified HITRAN LM code). LL : LBLRTM 12.8 and orig HITRAN LM code. UR : bias/std dev (LBLRTM 12.8-orig HITRAN LM code).

Note that while the original code looks better, it only used the bands that lay within the 25 cm-1 chunks I was building. The modified code looks far worse in the 8-10 um window region, but it used bands that also lay outside the 25 cm-1 chunks I was building (see change number 4), so that should technically be more correct. So it would appear that changes 5 or 6 are the likely cause of the increased OD in the 8-10 um region which cause a positive bias.

Figure 2 I is a slide from J.M. Hartmann's presentation at the HITRAN 2018 meeting (Cambridge, MA in Summer 2018) depicting the CO2 collision-induced absorption that he and Ha Tran have been working on. See also /home/sergio/PAPERS/AIRS/airs_stm_oct18/SARTA/sergio_rta.pdf


Figure 2: CO2 CIA from Hartmann and Tran code : CO2-N2/H2O CIA.

For completeness, Figure 3 shows comparisons of LBLRTM 12.8 versus LBLRTM 12.4 and versus our older UMBC CO2 line mixing. The UMBC CO2 linemixing is what is in the current AIRS SARTA, while the LBLRTM 12.4 is in the most recent CrIS FSR SARTA.


Figure 3: LBLRTM12.8 vs LBLRTM 12.4 and UMBC CO2 ODs

Figure 4 shows the effects of the various CIA (from JM Hartmann and H.Tran) and are always (BT without- BT with CIA), for the 49 regression profiles. The baseline CO2 ods were LBLRTM12.8 but since we are always adding on CIA, the plots should not depend on this.


Figure 4: Adding on the CIA optical depths in the 4.3 um region. From L to R, add on (L) WV/CO2, then (C) WV/N2) and (R) both WV/CO2 and WV/N2

The three graphs are presented in a slightly different way in Figure 5. Here the BT difference between (basic - cia) at 2400 cm-1 is shown as a function of column water amount for the 49 regression profiles, showing that indeed the Hartmann CIA lineshape(s) are dependent on water amount.


Figure 5: CIA effects as a function of column water, at 2400 cm-1

5 To do list (as of Nov 2018)

  1. kcarta now has the JM Hartmann CIA (co2/n2 and co2/h2o) terms. They are independent of any compressed database and have to be tuned on. I plan to include them in comparisons soon (done 11/12/2018)
  2. Collect actual "clear sky" observations, match to sondes (or NWP) and then compare to calculations made from the various CO2 databases we have : UMBC 2000 linemixing, LBLRTM 12.4, LBLRTM 12.8 and HITRAN LM
  3. In the meantime I can individually test some of my code changes to the LM code. The three main changes to test are points 4,5 and 6 in Section (4) above.

6 Conclusions

  1. For now, continue use the CO2 optical depths generated using LBLRTM 12.8
  2. Continue testing the LM code and the CIA optical depths, against LBLRTM 12.8 as outlined above in Section 5.

7 Recent Updates

  1. Update 11/27/2018 : went back to the orig LM radiation term (Sect 4, point 6) and turned rdmult down to 1000.0; this had the effect of removing the bias at 8-10 um, as hoped/expected.

Notice the the difference between the radiance calcs using "original LM code" and this latest incarnation, look like a 2 ppm offset (see Figure 6). Indeed, in the /home/sergio/KCARTA/WORK/RUN_TARA/GENERIC_RADSnJACS_MANYPROFILES/template_Qrad.nml file I see a comment that I used a multiplier such that the CO2 was adjusted to 383 ppm instead of 385 ppm.


Figure 6: Comparing original LM code versus my latest updates, were I reverted back to the original radiation term. Notice the difference mimics a 2 ppmv offset, explained in main body.

  1. Have been using pretty old (about 2006) O2-O2 and N2-N2 continuum. Have now upgraded them to use

the MT-CKD3.2 continuums. The code has been incorporated as the default version of run8.m. Note I just wrote a matlab wrapper that shell escapes and calls the MT-CKD3.2 code, rather than copying the code and mexing it. Data is in /home/sergio/KCARTA/WORK/RUN_TARA/GENERIC_RADSnJACS_MANYPROFILES/JUNK_HITRAN_370ppmCO2_PERTURBS_forITOVS/N2_O2_MTCKD3.2. Figure 7 shows the difference between (new-old) N2 continuums, in delta BT at TOA space after averaging over the 49 regression profiles.


Figure 7: Effect of new N2 continuum (from CKD 3.2) compared to what we have been using since about 2005. This is averaged over our 49 regression profiles.

  1. For tropical profile, ran off kcarta to see how CIA affects the TOA radiance, and also to see how large


Figure 8 shows the results. The left panel shows the Gnd2Space optical depth for N2, CO2/WV CIA and N2/WV CIA. Work by Ha tran focuses on the 2380 cm-1 bandhead, and there (in blue) is the CO2/WV CIA. The panel on the right shows how the TOA BT changes. Note for the "default" case we used CKD3.2 while for the CIA cases we used CKD 1.0


Figure 8: Seeing how the CIA changes the ODs and TOA BTs for a tropical profile. The left panel shows the ODs for N2 (black), and CIA N2/WV (red) and CIA CO2/WV (blue). The right panel shows how these change the TOA BT; note we also changed the water continuum from CKD3.2 to CKD 1.0.