The concentration of secondary organic aerosol (SOA) is underestimated in
current model studies. Recent research suggests that the reactive uptake of
dicarbonyls contributes to the production of SOA, although few models have
included this pathway. Glyoxal, an important representative component of
dicarbonyls in models, is significantly underestimated. We therefore
incorporated the reactive uptake of dicarbonyls into the regional air
quality modeling system RAMS-CMAQ (the Regional Atmospheric Modeling
System-Community Multiscale Air Quality) to evaluate the contribution of
dicarbonyls to SOA, and we then assess the impact of the underestimation of
glyoxal on the production of SOA in China during two time periods: 3 June to
11 July 2014 (episode 1) and 14 October to 14 November 2014 (episode 2).
When the reactive uptake process was added, the modeled mean concentration
of SOA in episode 1 increased by 3.65
The fine particle fraction of aerosols (PM
It has been reported that the concentration of SOA in the models is
underestimated by 1 to 2 orders of magnitude (de Gouw et al.,
2005; Volkamer et al., 2006). These results have motivated researchers to
investigate why these models are predicting SOA concentrations so poorly.
Traditionally, improvements in models have mainly concentrated on the
gas-phase and derived heterogeneous formation processes, such as the
formation of SOA from aromatic compounds under low- and high-
Many studies have suggested that glyoxal and methylglyoxal can form SOAs through chemical reactions in cloud or fog water (e.g., Warneck, 2003; Ervens et al., 2004; Lim and Ziemann, 2005; Carlton et al., 2006; Loeffler et al., 2006), or by reactive uptake on the surface of cloud droplets and aqueous aerosols (e.g., Liggio et al., 2005; Corrigan et al., 2008; Galloway et al., 2009; Ervens and Volkamer, 2010; Lim et al., 2010), which is probably a significant source of SOA (Ervens et al., 2014; Curry et al., 2018). A few studies (e.g., Carlton et al., 2008, 2010; Fu et al., 2009; Lin et al., 2012, 2014; Li et al., 2013; Woo and McNeill, 2015) have incorporated aqueous SOA formation pathways into atmospheric models. Several of these studies have shown that chemical reactions only in cloud or fog water make a negligible contribution to near-surface SOA relative to reactive uptake on the surface of cloud droplets and aqueous aerosols, and that the aqueous SOA formation cannot completely explain the gaps between the observations and simulations. There are still considerable uncertainties in our knowledge of the formation of SOA.
A series of studies (Fu et al., 2008; Myriokefalitakis et al., 2008; Liu et
al., 2012; Li et al., 2018) has shown that there is a substantial
underestimation in the modeled vertical column densities (VCDs) of glyoxal,
but few studies have considered the impact of this underestimation on
simulations of SOA concentrations. Xu et al. (2017) and
Ervens et al. (2011) showed that the aqueous SOA formation
depends on the liquid water content (LWC), which varies between seasons.
Previous studies, such as those of Fu et al. (2009) and
Li et al. (2013), have only considered the contribution from SOA
derived from the reactive uptake of dicarbonyls (pathway
J. Li et al. (2017) used the RAMS-CMAQ modeling system to investigate the effects of underestimated aromatic VOC emissions and yields of SOAs from different gas precursors on the concentration of SOAs. Similarly, we used Version 4.7.1 of the CMAQ, which is coupled with the gas-phase photochemical mechanism SAPRC99 (1999 Statewide Air Pollutant Research Center; Carter, 2000) and Version 5 of the aerosol module (AERO5; Byun and Schere, 2006; Foley et al., 2010). There are three major formation pathways for SOAs in this version. Based on the two-product approach, the first pathway is the equilibrium partition of semivolatile products formed from the oxidation of seven VOC precursors: long-chain alkanes, benzene (BNZ), high-yield aromatics (mainly toluene, ARO1), low-yield aromatics (mainly xylene, ARO2), isoprene (ISOP), monoterpene (TRP) and sesquiterpenes (SESQ). The second pathway is the oligomerization of semivolatile SOAs formed through the first pathway (Kalberer et al., 2004), namely the aging process. The third pathway is the formation of SOA via the in-cloud oxidation of glyoxal (GLY) and methylglyoxal (MGLY) (Carlton et al., 2008), both of which represent dicarbonyls in the model. The details of these formation pathways are given in Carlton et al. (2010).
The meteorological fields used to drive CMAQ are obtained from RAMS, which has been described in detail by Cotton et al. (2003). The National Centers for Environmental Prediction reanalysis datasets are served as the initial and lateral boundary meteorological conditions input into RAMS. The boundary conditions used for the RAMS computations include the weekly average sea surface temperature and the monthly measured snow cover. The final modeled results are output through the four-dimensional data assimilation mode using nudging analysis.
The emission sources are derived from several different inventories.
Anthropogenic emissions (M. Li et al., 2017) – including
The model domain is divided into
Geographical locations of the measurement stations in the model domain. 1: Beijing; 2: Tianjin; 3: Datong; 4: Jinan; 5: Qingdao; 6: Hohhot; 7: Fuzhou; 8: Guangzhou; 9: Urumqi; 10: Nanchang; 11: Guilin; 12: Guiyang; 13: Xinglong; 14: Yucheng. The different colors denote which variables are compared at each station. Purple: meteorological parameters; green: concentrations of aromatics; red: both meteorological parameters and aromatic compound concentrations.
There is a standard reaction probability formulation for reactive uptake of
gases by aerosols and clouds in Jacob (2000). In this formulation, the
first-order rate constant
We implement the reactive uptake of dicarbonyls by cloud droplets following
the standard Eq. (1). In a similar manner to Fu et al. (2008) and Li et
al. (2013), the cloud droplet surface area is calculated from the LWC in the
cloudy fraction of the model grid by assuming an effective droplet radius of
6
To assess the contribution of dicarbonyl species to the concentration of SOA,
we used the observed hourly concentration of SOA in Beijing during the summer
and fall of 2014 measured by Xu et al. (2017) with an Aerodyne
high-resolution time-of-flight aerosol mass spectrometer. Samples were taken
from 3 June to 11 July 2014 (episode 1) and 14 October to 14 November 2014
(episode 2) at the Institute of Atmospheric Physics in China. More detailed
information about the data has been reported by Xu et al. (2017). The
simulation periods are from 22 May to 11 July and from 1 October to
14 November, with the first 12 d as the spin-up time. To evaluate the
reasonability in simulating the formation processes of aqueous SOAs and
analyze the relative causes, the corresponding cloud water path (CWP) and
cloud fraction data measured by the MODerate Resolution Imaging
Spectroradiometer (MODIS) were obtained from the website
Both Liu et al. (2012) and Li et al. (2018) have suggested that the underestimation of glyoxal concentrations in simulations is related to the underpredicted emission of aromatic compounds. Therefore, the biases in the emission of aromatic compounds need to be evaluated through a comparison of the observed and simulated concentrations of aromatic compounds. The observed data were collected at 14:00 local standard time every Thursday by gas chromatography and mass spectrometry in Beijing, Xinglong and Yucheng (Fig. 1). Sun et al. (2016) and Wu et al. (2016) have presented the detailed information.
To evaluate the performance of our model, we also compared the simulated
PM
Table 1 shows statistical results of the meteorological parameters (i.e.,
temperature, relative humidity, wind speed and wind direction) and PM
Statistics for the meteorological variables and PM
There are inevitably some biases in the simulated meteorological parameters
relative to the observations due to the limited model resolution and system
errors. Nevertheless, the model reproduces the magnitude and variation trend
of the temperature and relative humidity fairly well (Figs. S1 and S2 in the
Supplement), with IOAs for temperature of 0.91 and 0.95 in the two episodes
and IOAs of 0.91 and 0.88 for the relative humidity, comparable with the
results of Li et al. (2012) and Wang et al. (2014). The mean biases of
temperature (
The modeled PM
The modeled and satellite-observed mean CWPs were compared in Fig. 2 during
the two analyzed episodes to evaluate the reasonability in simulating the
aqueous SOA formations. Figure 2a and b show that in episode 1, during one
summer period, the highest observed CWP (400–500 g m
Observed and simulated distributions of the mean cloud water path
(CWP) during the two episodes (
Overall, there are obvious biases in the numerical values between the observed and simulated CWPs. According to the comparisons between the observed and modeled cloud fraction and precipitation (shown in Figs. S6 and S7), this is a result that has a lot to do with the uncertainties in the cloud fraction estimations, which indirectly lead to uncertainties in the simulations of the concentrations of SOAs. However, the mean distribution patterns of the simulated CWP during the two episodes are similar to the observational results, indicating few impacts on the simulated distribution of SOA. Both the simulated and observed CWPs in episode 1 are higher than in episode 2, also implying differences in the simulation of SOA between the two episodes.
As described in Sect. 2.3 and with reference to previous research, the underestimation of glyoxal concentrations may partly result from the underprediction of the emissions of aromatic compounds. It is therefore necessary to evaluate the emissions of aromatic compounds during the analyzed episodes with the base model before designing and implementing the sensitivity case studies. For this purpose, comparisons were made between the simulated and observed concentrations of aromatic compounds in a similar manner to the work of Zhang and Ying (2011).
Figure 3 presents the ratio of the observed-to-predicted (
Box–whisker plot of the observed-to-predicted (
Three sensitivity simulation case studies are designed based on these
results. Case 0 is run with the three default SOA formation pathways included
in the standard model. According to Fu et al. (2008), the default aqueous SOA
formation pathway is included in the pathway
Figure 4 compares the hourly concentrations of the simulated and observed
SOAs during the two analyzed episodes; the corresponding statistical
parameters are listed in Table 2. The concentrations of SOAs are measured
from PM
Hourly concentrations of the observed and simulated near-surface SOA
concentrations in episode 1
Performance statistics of the modeled and observed SOA
concentrations (
In case 0, the SOA concentrations in episode 1 (Fig. 4a) are significantly
underestimated by an average factor of 5.7, with the differences being as
high as a multiple of
Figure 5 shows the mean contributions from different sources of SOAs in the
three sensitivity case studies during the two analyzed episodes. AAQ,
dicarbonyl-derived SOAs, contributes little in case 0 during the two
episodes. The mean contribution of AAQ to the total concentration of SOAs in
episode 1 is 2.39 %, larger than in episode 2 (0.89 %). As a result of
the greater amount of emissions from biogenic sources in summer, SOAs formed
from biogenic precursors (AISO
Mean contributions from different sources of SOAs in the three sensitivity case studies during the two analyzed episodes. The compositions represent the SOA formed from long-chain alkanes (AALK), high-yield aromatic compounds (ATOL), low-yield aromatic compounds (AXYL), benzene (ABNZ), monoterpenes (ATRP), isoprene (AISO), sesquiterpenes (ASQT) and dicarbonyls (AAQ), as well as aged anthropogenic (AOLGAJ) and biogenic SOA (AOLGBJ). Case 0 is the base run; case 1 is run with the incorporation of the reactive uptake of dicarbonyls (excluding the default in-cloud dicarbonyl oxidations); and case 2 is based on case 1, but taking into consideration the underestimation of glyoxal.
It is clear that the biases between the observed and simulated concentrations
of SOAs decrease when the contributions of unaccounted-for dicarbonyls to the
concentrations of SOAs are considered, especially in summer. However, the
sources of unaccounted-for SOAs cannot be explained completely. As a result of
uncertainties in the description of known SOA formation processes and missing
pathways that are not included in the model, for example, there are many
uncertainties in glyoxal simulations (Li et al., 2018). There are also many
uncertainties in incorporating pathway
To distinguish the contribution of dicarbonyls to the concentration of SOA
over China in case 2 from that in case 0, the distributions of
dicarbonyl-derived SOAs and their contributions to SOAs (AAQ
Figure 6a–b and c–d show the mean concentration of AAQ in cases 0 and 2,
respectively, during the two episodes. For the base case, in episode 1
(Fig. 6a), the concentration of AAQ over China is
Modeled distributions of the mean
Figure 6e–f and g–h show the spatial distribution of the mean AAQ
The RAMS-CMAQ modeling system was used to assess the contributions of
dicarbonyls from the reactive uptake process and unaccounted-for sources of
glyoxal to the concentrations of SOAs during the two episodes from 3 June to
11 July 2014 and 14 October to 14 November 2014. Comparisons between the
observed and simulated concentrations of SOAs from three sensitivity groups
showed different improvements in the SOA simulations with the inclusion of
pathway
The mean AAQ in case 2 during the two episodes was clearly improved over
China relative to case 0 and was generally higher in the east than in the
west. In episode 1, the highest value (10–15
It is clear that the contributions of dicarbonyls from pathway
Besides, the aqueous SOA formation is not only relative to the distributions of dicarbonyl concentrations, but also depends on the liquid water content (LWC). Due to the large space and time dependence, one single station measurement of SOA concentration is not enough to evaluate the model performance over China, especially the impacts of glyoxal underestimations on dicarbonyl-derived SOA. Thus, more observed SOA data from different stations need to be collected and used for comparisons to reduce the uncertainties in the conclusions.
The datasets of measured SOA concentrations, observed aromatic compound concentrations and corresponding modeled results used in this study can be accessed by contacting sunyele@mail.iap.ac.cn, tgq@dq.cern.ac.cn and mgzhang@mail.iap.ac.cn, respectively.
The supplement related to this article is available online at:
In this study, JL designed the sensitivity experiments, developed the model code and performed the corresponding simulations. MZ co-designed the experiments and provided valuable advice about the model operations. YS carried out the measurements of SOA and provided the corresponding data for evaluating the modeled results. GT and FW carried out the measurements of aromatic compound concentrations and provided the corresponding data for evaluating the simulated results. YX provided valuable advice on model result analysis.
The authors declare that they have no conflict of interest.
This study was supported by the National Key R&D Programs of China (no. 2017YFC0209803), the National Natural Science Foundation of China (91544221), and the Beijing Municipal Science and Technology Project (ZL171100000617002). We thank Qi Ying from Texas A&M University for helping to incorporate the reactive uptake of dicarbonyl pathways into the model.
This paper was edited by Xiaohong Liu and reviewed by three anonymous referees.