Really small SOA formation in CMAQ


I am running CMAQ version 5.3.2 for SOA simulation. However, I am seeing extremely low SOA concentrations generated from conventional anthropogenic SOA precursors (TOL, XYL, BENZ) compared to CAMx results (by a factor of 3-4 lower). I know that adding S/IVOCs emissions would increase SOA formation but I want to make sure that anthropogenic VOCs emissions are giving reasonable SOA formation. Both CAMx and CMAQ simulations were conducted using identical meteorology and emission inventory. I want to confirm there is nothing wrong with my CMAQ SOA simulation and post-processing.

For CMAQ SOA simulation, this is what I did:

(1) Provide the model with emissions of species including “BENZENE”, “TOL”, “SOAALK” (=0.108PAR), “XYLMN” (=0.998XYL), “NAPH” (=0.002*XYL).
(2) I summed the following species to obtain model simulated ASOA (anthropogenic SOA):

Assuming my above two steps are correct, I am getting very small SOA formation over China (up to ~2ug/m3 in July).

I check the SOA yields implemented in CMAQ v532’s source code and seems to be inconsistent with what I found online.

This is what I found online ( CMAQ/ at main · USEPA/CMAQ · GitHub

However, this is what I see in the source code:

The numbers before “SVAVB2”, “SVAVB3”, etc. represent the SOA yields, correct? Apparently, the values in the source code is much lower than the table above. I am wondering did I look at the wrong place?

Another question is what is difference between this “Mass-based SOA yields by C*” and the “Alpha” value for TOLNRXN, XYLNRXN, etc. in SOA_DEFN.F? Is the “Alpha” value for TOLNRXN and XYLNRXN valid for two-product scheme, which is not being used in the VBS scheme?

Thank you very much!

Hi Ling,

You have found nearly all of the relevant documentation to answer your question, and we can help fill in some of the connections.

  1. Difference between Table 1 and source code coefficients: this is due to data in Table 1 being mass-based, while the source code values are mole based. Since the values are multiplied by MW_reactant / MW_product and the SOA-relevant products are assumed to have larger MW than the reactants, the values decrease as expected.

  2. You are right that TOLNRXN and XYLNRXN are not being used for the VBS scheme.

  3. Comparing CMAQv5.3.x to CAMx for anthropogenic SOA is tricky. According to the CAMx version 7.10 manual, anthropogenic SOA species are aged continually to form lower volatility species (Fig. 5-2). Oxidized POA species also contribute to these species concentrations. These aging schemes are parameterized to yeild acceptable performance for OA in the United States (Koo et al., 2014; 1.5-Dimensional volatility basis set approach for modeling organic aerosol in CAMx and CMAQ - ScienceDirect).
    In CMAQv5.3.x, oxidized POA is tracked separately from anthropogenic SOA. Also, ASOA compounds are not aged in the same way; they form oligomers (AOLGA), as you have correctly accounted for, but with less efficiency than the aging in the CAMx approach. Finally, CMAQv5.3.x mechanisms employ the potential-combustion SOA species, which is highly parameterized for OA performance in the U.S. (Murphy et al., 2017; ACP - Semivolatile POA and parameterized total combustion SOA in CMAQv5.2: impacts on source strength and partitioning).

I’m not sure if a factor of 3-4 lower would be expected, but it doesn’t sound unreasonable to me, given the impact of SOA aging mechanisms I have seen in the past.

CMAQv5.4 (expected release Sept 2022) will include a new chemical mechanism (CRACMM) that will be specifically targeted for improved representation of the contribution of individual VOC precursors to the SOA system using the community’s most recent knowledge of important SOA pathways. Another option contributed by collaborators at Tsinghua University implements the two-dimensional VBS for SOA predictions.

Best wishes,
Ben Murphy


Hi Ben,

Really appreciate the reply!

(1) Got the idea of conversion between molar and mass yields. However, these values seem to be much lower than the corresponding SOA yields implemented in CAMx. For instance, on molar basis and under high NOx conditions, TOL yields 0.016 SVAVB2 (C*=1), 0.051 SVAVB3 (C*=10) and 0.047 SVAVB4 (C*=100). Whereas in CAMx, under the same condition, TOL yields 0.006 VAS1 (C*=1), 0.145 VAS2 (C*=10), 0.281 VAS3 (C*=100), and 0.432 VAS4 (C*=1000). These numbers seem to be different by an order of magnitude. Is there any reference for CMAQ’s SOA yields for TOL/XYL/BENZ? These values seem to be inconsistent (again lower) with Hodzic et al 2015 paper (Table 1) acp-16-7917-2016.pdf (

  1. You pointed out an important point that I didn’t realize earlier, that is in CMAQ, there is no such continuous aging from high volatility bins into low volatility bins (for example, from SVAVB4 → SVAVB3, from SVAVB3 → SVAVB2) for anthropogenic VOCs as implemented in CAMx. This might also cause smaller SOA formation in CMAQ, compared to CAMx?

  2. As you mentioned the potential-combustion SOA species, I have been really confused with this. I looked into the CMAQ source code. It seems to me that “PCVOC” represents IVOC emissions and “PCSOA” represents the SOA formation from PCVOC emissions, correct? If I have separate estimate of IVOCs emissions, I should rename my IVOCs emissions as “PCVOC” in CMAQ so that CMAQ could incorporate SOA formation from IVOCs, correct?


Hello Ling,
The CMAQv5.3+ BTX, PAH, and alkane SOA yields were implemented in CMAQ by Qin et al. 2021 (Criteria pollutant impacts of volatile chemical products informed by near-field modelling | Nature Sustainability). The v5.3+ release notes provide both the scientific basis and the work in which they were implemented (CMAQ/ at main · USEPA/CMAQ · GitHub). The underlying laboratory basis of the CMAQ yields is Ng et al. 2007 (as implemented in

CAMx may be using a different experimental basis resulting in different fitted coefficients. Note that CMAQv5.3+ BTX SOA is not wall loss corrected for vapors. CRACMM will include the wall loss corrections.

If CAMx has a larger pool of S/IVOC products than CMAQ, the aging will be particularly effective in making more SOA in CAMx.

1 Like

Dear Dr. Pye,

Thank you so much for the reply! This is really helpful. Now I know where did the CMAQ v5.3+ SOA yields come from. Based on my current understanding, CAMx VBS has a stronger aging effect than CMAQ, given that the volatility bins in CMAQ do not age over the time.

Just to double check. You said “CMAQv5.3+ BTX SOA is not wall loss corrected for vapors”. Does that mean the SOA simulated by CMAQv5.3+ is biased low?

And I am really looking forward to the CRACMM mechanism.


Hi Ling,
Yes, the SOA yield is biased low because instead of condensing to particles or remaining in the chamber to further react, some gas-phase mass condensed to the walls. The issue can be corrected. For BTX in particular, see the work of Zhang et al. 2014 PNAS (