The spatial distribution and properties of submicron organic aerosol (OA)
are among the key sources of uncertainty in our understanding of aerosol
effects on climate. Uncertainties are particularly large over remote regions
of the free troposphere and Southern Ocean, where very few data have been
available and where OA predictions from AeroCom Phase II global models span 2 to 3 orders of magnitude, greatly exceeding the model spread over
source regions. The (nearly) pole-to-pole vertical distribution of
non-refractory aerosols was measured with an aerosol mass spectrometer
onboard the NASA DC-8 aircraft as part of the Atmospheric Tomography (ATom)
mission during the Northern Hemisphere summer (August 2016) and winter
(February 2017). This study presents the first extensive characterization of
OA mass concentrations and their level of oxidation in the remote
atmosphere. OA and sulfate are the major contributors by mass to submicron
aerosols in the remote troposphere, together with sea salt in the marine
boundary layer. Sulfate was dominant in the lower stratosphere. OA
concentrations have a strong seasonal and zonal variability, with the
highest levels measured in the lower troposphere in the summer and over the
regions influenced by biomass burning from Africa (up to 10
Organic aerosol (OA) is a complex mixture of directly emitted primary OA
(POA) and chemically produced secondary OA (SOA) from anthropogenic and
biogenic emission sources. It is associated with adverse health effects
(Mauderly and Chow, 2008; Shiraiwa et al., 2017) and contributes radiative
forcing to the climate system (Boucher et al., 2013). The currently limited
understanding of processes involved in the formation, aging, and removal of
organic compounds results in large uncertainties in (i) the predicted global
OA burden, (ii) relative contributions of emissions vs. chemistry to OA
formation, (iii) spatial distribution, and (iv) impacts on radiation and
clouds (Kanakidou et al., 2005; Hallquist et al., 2009; Heald et al., 2011;
Spracklen et al., 2011; Tsigaridis et al., 2014; Hodzic et al., 2016;
Shrivastava et al., 2017; Tsigaridis and Kanakidou, 2018; Zhu et al., 2019).
The uncertainties are particularly large in the estimated global burden of
SOA, which ranges from 12 to 450 Tg yr
Model performance has been especially poor in the remote regions of the atmosphere where OA measurements available for model evaluation have been sparse (especially aloft). Using data from 17 aircraft campaigns mostly located in the Northern Hemisphere, Heald et al. (2011) showed that the skill of the global GEOS-Chem model in predicting the vertical distribution of OA was significantly decreased in remote regions compared to polluted near-source regions. The study pointed out the limitations of commonly used SOA formation mechanisms that are based on chamber data; these have the tendency to underpredict OA in source regions and overpredict OA in the remote troposphere. For a subset of nine recent aircraft campaigns, Hodzic et al. (2016) showed that OA is likely a more dynamic system than represented in chemistry–climate models, with both stronger production and stronger removals. These authors suggested that additional removal mechanisms via, e.g., photolytic or heterogeneous reactions of OA particles are needed to explain low OA concentrations observed in the upper troposphere where direct cloud scavenging is less efficient. The recent global multi-model comparison study (Tsigaridis et al., 2014) within the AeroCom Phase II project illustrates the amplitude of model uncertainties simulating OA mass concentrations and the contrast in model performance between near-source and remote regions. The results indicate that model dispersion (the spread between the models with the lowest and highest predicted OA concentrations) increases with altitude from roughly 1 order of magnitude near the surface to 2–3 orders of magnitude in the upper troposphere. Our own analyses of the AeroCom-II models shown in Fig. 1a indicate that model dispersion (quantified as the ratio of the average concentration of the highest model to that of the lowest one in each region) increases not only with altitude but also with distance from the northern midlatitude source (and data-rich) regions. The model spread is a factor of 10–20 in the free troposphere between the Equator and northern midlatitudes and increases to a factor of 200–800 over the Southern Ocean and near the tropopause. It is not surprising that model spread is lower closer to source regions where it is mostly driven by uncertainties in emissions and SOA production yields. Spread is expected to be larger in remote regions where models are also impacted by uncertainties in transport, chemical aging, and removal. The lowest model dispersion also coincides with the regions of the Northern Hemisphere (NH) and the African biomass burning outflow where models have been evaluated the most (Fig. 1b), emphasizing the need for further model–observation comparison studies in remote regions (of the Southern Hemisphere (SH) in particular).
Here, we present a unique dataset of airborne aerosol mass spectrometer
measurements of OA mass concentrations collected onboard the NASA DC-8 as
part of the Atmospheric Tomography (ATom) mission. The aircraft sampled the
vertical structure of the atmosphere from the near-surface (0.2 km) to
lower stratosphere (LS) regions (12 km of altitude) over both the Pacific and
Atlantic basins (to limit the influence of source regions) with a
quasi-global spatial coverage from 82
The modeling framework is described in Sect. 2. Section 3 describes the ATom dataset and the spatial and vertical distributions of OA over the Atlantic and Pacific regions. Section 4 presents the comparisons of ATom-1 and ATom-2 data to multi-model predictions from both the AeroCom-II models and the ensemble of eight current model simulations of the ATom campaign. Section 5 presents the conclusions of the study and discusses its implications.
ATom measurements were compared with the results of eight global models that
simulated the time period of the ATom-1 and ATom-2 campaigns (August 2016 and
February 2017) using the emissions and reanalysis meteorology corresponding
to this period (and a spin-up time of at least 6 to 12 months). These
are referred to hereafter as ATom models and include the NASA global Earth
system model GEOS5, the aerosol–climate model ECHAM6-HAM, three versions of
the NCAR Community Earth System Model (CESM), and three versions of the
global chemistry GEOS-Chem model. Simulations were performed at various
horizontal resolutions ranging from relatively high at
ATom global model configurations and their treatment of the most important processes affecting organic aerosols.
GEOS5 was run in a configuration similar to Bian et al. (2019) using the
anthropogenic emissions from HTAP v2 (Janssens-Maenhout et al., 2015) and
biomass burning emissions from the Quick Fire Emission Dataset (QFED v2.54).
Aerosols are simulated within the GOCART bulk aerosol module and include
externally mixed particles of black carbon (BC), organic carbon (OC),
sulfate, ammonium, nitrate, dust, and sea salt (Colarco et al., 2010; Bian et
al., 2017). The formation of SOA is based on a prescribed 10 % formation
yield from the monoterpene emissions. The primary emitted OC and SOA are
separated into hydrophobic (50 %) and hydrophilic (50 %) species, with a
2.5 d
The ECHAM6.3-HAM2.3 standard version (Tegen et al., 2019) was run using updated anthropogenic emissions (Schacht et al., 2019) combining the ECLIPSE (Klimont et al., 2017) emissions with Russian anthropogenic BC emissions from Huang et al. (2015). For biomass burning the Global Fire Assimilation System (GFAS; Kaiser et al., 2012) emissions are used but without the scaling factor of 3.4 suggested by Kaiser et al. (2012). Aerosol composition and processes are simulated using the Hamburg Aerosol Model (HAM2; Zhang et al., 2012) that considers an aerosol internal mixture of sulfate, BC, OC, sea salt, and mineral dust. The aerosol population and microphysical interactions are simulated using seven lognormal modes, including nucleation, soluble and insoluble Aitken, accumulation, and coarse modes. In the model configuration used in this publication the formation of SOA is based on a prescribed 15 % mass yield from monoterpene emissions only (Dentener et al., 2006). Aerosol particles are removed by dry and wet deposition. The wet deposition includes below-cloud scavenging by rain and in-cloud cloud scavenging for large-scale and convective systems (Croft et al., 2010).
The two simulations with the GEOS-Chem 12.0.1 global chemistry model (Bey et
al., 2001) use emissions based on the CMIP6 global inventory (CEDS historical
emissions up to 2014 and future emissions based on climate scenarios; Hoesly
et al., 2018; Feng et al., 2020) with regional improvements for
anthropogenic sources and on GFED v.4 for biomass burning emissions (Giglio
et al., 2013). Both simulations use the bulk aerosol representation and
differ only in the treatment of SOA formation and removal. The first
configuration (called hereafter GC12-REF) includes the default (
GC10-TOMAS is based on the GEOS-Chem version 10.01 coupled with the TwO Moment Aerosol Sectional microphysics scheme (TOMAS) and run in a similar configuration to that described in Kodros et al. (2016). The model computes the evolution of sulfate, sea salt, primary and secondary OA, BC, and dust aerosols described by 15 internally mixed size bins (six of which were analyzed for these comparisons; see Table 1). Anthropogenic emissions are based on the EDGAR v4 global inventory with regional improvements, while the biomass burning emissions are from GFED v3. SOA is irreversibly made from the emitted parent precursor, considering a 10 % mass yield from monoterpene emissions and an emission flux of 0.2 Tg of SOA per teragram of CO for the anthropogenic CO emissions. The removal of gases and aerosols is treated similar to the GEOS-Chem 12.0.1 model (GC12-REF; see above).
Simulations based on the CESM2.0 Earth system model use the standard version
of the Whole Atmosphere Community Climate Model (WACCM6; Gettelman et al.,
2019; Emmons et al., 2020). Details on the specifics of the model
configurations are described in detail in Tilmes et al. (2019); i.e.,
CESM2-SMP and CESM2-DYN correspond to the specified dynamics of the WACCM6-SOAG and
WACCM6-VBSext simulations described in that work, respectively. Emissions
are based on the CMIP6 global inventory for the year 2014 for anthropogenic
sources and on the QFED version 2.4 for the wildfires inventory. Aerosols
are represented with the modal aerosol scheme (MAM4, Liu et al., 2012) that
includes BC, primary and secondary OA, sulfate, dust, and sea salt. Four
modes are considered, including Aitken, accumulation, and coarse modes,
and an additional primary carbon mode. Only the accumulation mode was used
in this work. The CESM2-SMP and CESM2-DYN simulations differ in their
treatment of OA. CESM2-SMP forms OA directly using fixed mass yields from
primary emitted precursors (isoprene, monoterpenes, aromatics) without
explicitly simulating their oxidation and partitioning. These mass yields
are increased by a factor of 1.5 to match the anthropogenic aerosol indirect
forcing (Liu et al., 2012). The second configuration (referred to as
CESM2-DYN) includes the formation and removal parameterizations of organics
of Hodzic et al. (2016), as implemented into CESM2 by Tilmes et al. (2019)
for all species based on low-
CESM1-CARMA simulations use the configuration described in Yu et al. (2019),
which is based on CESM1 and the sectional Community Aerosol and Radiation
Model for Atmospheres (CARMA v3.0). Anthropogenic emissions are those from
the Greenhouse gas–Air pollution Interactions and Synergies (GAINS) model,
and biomass burning emissions are from the Global Fire Emission Database
(GFED v3; van der Werf et al., 2010). In CARMA, 20 size bins are used for
both pure sulfate particles (bins from 0.2 nm to 1.3
The ATom measurements are also compared to the global model OA predictions generated within the Phase II Aerosol Comparisons between Observations and Models (AeroCom-II) project (Schulz et al., 2009). We consider the monthly average results of 28 global models, which is a subset of those presented in Tsigaridis et al. (2014), based on the availability of model results. It should be noted that the meteorological forcing used in these models is mostly based on the year 2006, while the anthropogenic and biomass burning emissions are mostly representative of the year 2000. For comparison purposes, the monthly mean model outputs for the months of August (ATom-1) and February (ATom-2) are interpolated along the flight path (latitude, longitude, and altitude) and averaged the same way as the measurements (see Sect. 3.2).
The measurements of non-refractory submicron aerosols were performed onboard the NASA DC-8 aircraft as part of the ATom field study (Wofsy et al., 2018) using the University of Colorado Aerodyne high-resolution time-of-flight aerosol mass spectrometer (AMS in the following; Canagaratna et al., 2007; DeCarlo et al., 2006).
We use measurements from both the NH summer (August 2016, ATom-1) and winter
(February 2017, ATom-2) deployments. Figure 2a shows the flight path and the
vertical extent of the ATom-1 dataset colored by OA mass concentrations (see
Fig. S1 in the Supplement for ATom-2). The aircraft performed systematic vertical sampling
with
For ATom, the AMS reported the standard non-refractory aerosol species OA,
sulfate, nitrate, ammonium, and chloride, with the response for all the
nominally inorganic species characterized by in-field calibrations. In
addition, it also reported methanesulfonic acid (MSA; Hodshire et al., 2019a
describes the AMS MSA methods and calibrations for ATom) and sea salt
for
For ATom the AMS measured particles with geometric diameters (based on the
campaign-wide average density of 1640 kg m
Refractory and non-refractory aerosol composition was also measured using the NOAA Particle Analysis by Laser Mass Spectrometry (PALMS) instrument. PALMS classifies individual aerosol particles into compositional classes including biomass burning (Hudson et al., 2004), sea salt (Murphy et al., 2019), mineral dust (Froyd et al., 2019), and others. Mass concentrations for these particles types are derived by combining PALMS composition data with aerosol size distribution measurements (Froyd et al., 2019). Good agreement overall was found for OA, sulfate, and sea salt between the two particle mass spectrometers during ATom once the AMS and PALMS instrument transmissions were accounted for (Jimenez et al., 2019b). For all PALMS data used in this work (biomass burning fraction and dust) the AMS transmission function was applied to ensure that both instruments were characterizing approximately the same particle range.
For a particular air mass, the mass fraction of biomass burning (BB) aerosol
reported by the PALMS instrument,
For model evaluation purposes, it is important to know whether the source of
OA is primary or secondary. For ground studies close to sources (e.g.,
Jimenez et al., 2009), positive matrix factorization of AMS mass spectra
(PMF; Ulbrich et al., 2009) can be used to estimate the contribution of
primary sources (mostly from transportation, heating, cooking, and biomass
burning) to total OA. This approach is not suitable for ATom. To accurately
resolve a minor factor such as POA in an AMS dataset, there needs to be a
combination of (a) sufficient OA mass concentration so that the
signal-to-noise ratio of the spectra is sufficient; (b) enough fractional mass for
the factor to be resolved (
Instead, in this work we have estimated POA based on the fact that it is
co-emitted with BC as part of the combustion processes releasing both
species in source regions and that BC is not impacted by chemical aging
processes over the lifetime of the air mass. Note that BC can physically age
but it is not lost in any significant amount to the gas phase due to
chemical processes in the atmosphere. We assume non-differential removal
(and transport) of the BC fraction relative to the rest of the POA (the two
are generally internally mixed; Lee et al., 2015). Table S1 in the Supplement summarizes
recent POA
The PALMS-determined mass fraction of biomass-impacted aerosol
(
Further detail is provided in Table S2, which summarizes the POA
PALMS detection efficiency increases with size across the accumulation mode,
and therefore the
The contribution of POA from sea spray is difficult to constrain. As an
order-of-magnitude estimate, marine POA is roughly calculated based on
preliminary calibrations of OA on mineral dust particles from the PALMS
instrument (Karl Froyd, personal communication, 2019). Using this calibration, the
average OA by mass on sea salt was
For comparisons between the measurements and the various global models,
data were averaged both vertically and zonally to minimize the impact of
smaller plumes or vertical gradients in aerosol concentrations that might
not be captured by coarse-resolution models. For the same reason, all data
near airports were removed from the datasets prior to analysis (up to about 3 km on the climb in and out). In order to restrict this analysis to the remote
troposphere, the last leg of the ATom-1 mission (over the continental US)
was taken out of the dataset as well. Data were binned into five large latitude
regions as shown in Fig. 2a, including southern polar (55–80
Some of the performed analysis required separating the dataset into vertical
subsets. In this paper, we define the marine boundary layer (MBL) as
the region below 1.5 times the calculated boundary layer height in the NCEP
global model reanalysis. The free troposphere (FT) includes all data points
between the top of the MBL and the NCEP tropopause height, and the LS region
includes all points above the NCEP tropopause height. The tropopause height
varied during ATom between 8 and 16.5 km; given the DC-8 ceiling (12.8 km) the stratosphere was only sampled at latitudes higher than 30
Figure 2b shows that during both NH summer and winter ATom deployments, OA
is one of the three dominant components of the measured submicron aerosol in
the remote troposphere, together with sulfate and sea salt. During ATom-1,
average submicron aerosol concentrations were close to 0.8
OA is found to be a major constituent (
Figure 2a (and Fig. S1) shows the spatial and vertical distribution of OA
mass concentrations measured during the ATom-1 (and ATom-2) campaigns. Most data
were taken over remote oceanic regions (and a few remote continental
regions, primarily over the Arctic). The measured OA varies between
extremely clean conditions (
The measured OA is characterized by a strong latitudinal gradient. Figure 2c
shows the average vertical profiles of measured OA over the selected
latitudinal bands during August 2016. The cleanest air masses are observed
over the remote oceanic regions of the Southern Hemisphere (SH,
25–80
Figure 2d shows that the Atlantic Basin is often more polluted than the Pacific Basin, not only because of the African biomass burning influence but also due to the contribution of anthropogenic pollution in the lower troposphere of the NH. It should be noted that Asian pollution was likely an important contributor to the North Pacific basin, especially between 2 and 6 km, in both ATom deployments (see Figs. 2a and S1). Several-fold higher OA concentrations are found near the surface (below 1 km) over the southern Pacific compared to that same location in the southern Atlantic, which could be indicative of the stronger emission of marine OA in the Pacific Basin.
In addition to spatial gradients, a strong summer-to-winter contrast is
observed in OA concentrations. Figure 2e shows the ratio between OA vertical
profiles measured in the NH summer ATom-1 vs. in the NH winter ATom-2. The
NH is more polluted during the NH summer due to the photochemical production
of SOA and biomass burning emissions, leading to the tripling of OA
concentrations in the extratropical regions (25–80
Prior to evaluating model performance in simulating OA, we assessed the
ATom models' ability to simulate sulfate aerosols. According to the model
evaluation shown in Table S3, the predicted sulfate concentrations are
generally within 40 % of the measured values, which is comparable to the
AMS measurement uncertainties. The only exception is found for the
ECHAM6-HAM model, which overestimates sulfate aerosols by a factor of 2.
These results imply that most ATom models capture the
overall sulfate burden relatively well. However, the large root mean square error (RMSE
For OA, model evaluation metrics for the entire ATom-1 and ATom-2 campaigns
are given in Table 2 for the eight ATom models and their ensemble, as well
as the AeroCom-II ensemble. The results show that the normalized mean bias
is substantially lower for the ATom model ensemble compared to AeroCom-II,
decreasing from 74 % to 4 % for ATom-1 and from 137 % to 23 % to
ATom-2, which is within the measurement uncertainty range. The mean temporal
correlations are substantially improved from 0.31 (0.38) for AeroCom-II to
0.66 (0.48) for the ATom model ensemble during ATom-1 (ATom-2). However, results
vary strongly among ATom models. Models using prescribed emissions of
non-volatile SOA have the tendency to overestimate the OA concentrations
during both NH summer and winter deployments (with
Comparison of observed and simulated OA concentrations along ATom-1
and ATom-2 flights for eight global model simulations and their ensemble.
The results of the model ensemble are also indicated. The statistical
indicators are calculated as the normalized mean bias (NMB; %)
Figure 3 compares the average median ratios between modeled and observed OA
concentrations for the ATom and AeroCom-II model ensembles for different
regions (BB, MBL, FT, LS). The results show that the median ratio for the
ATom model ensemble is close to unity in all regions. This is at least a
factor of 2 improvement compared to AeroCom-II models, which were almost
always biased high for the remote regions sampled in ATom. The model spread
has also been reduced by a factor of 2–3 in all regions. This reduction in
the ensemble spread may partially be explained by a smaller size of the ATom
model ensemble (see Fig. S2), which also includes models with a more
up-to-date OA representation. In order to explore this point further,
results for a subset of AeroCom-II models (using earlier versions of models
in the ATom ensemble) show only a slight reduction (
Ratios between predicted and observed OA concentrations for all ATom-1 flights as calculated for the ATom and AeroCom-II model ensembles in different regions (BB: biomass-burning-influenced regions; MBL: clean marine boundary layer; FT: clean free troposphere; LS: lower stratosphere). The median of the ensemble ratio is shown as a horizontal line, while the boxes indicate the 25th and 75th percentiles. Medians for the individual models included in the current ATom model ensemble are also shown as blue lines.
Figure 4 compares the mean vertical profiles of OA measured during ATom-1
and ATom-2 with the predictions of the model ensemble average based on the eight
ATom models (Table 1) and 28 AeroCom-II models for the different latitudinal
regions of the Pacific and Atlantic basins. Note that the use of a wide
logarithmic scale (to be able to span all the observations) may make the
observed differences appear small, although they often reach factors of 2–10
and larger (Fig. S5 shows the results on a linear scale). For AeroCom-II,
large latitudinal differences exist in the results, with a better performance
closer to source regions and large disagreement in the lower stratosphere
and remote regions, as already suggested by the mission medians shown in
Fig. 3. The best AeroCom-II model performance is found over the Equator in
both basins, where the model ensemble captures within a factor of 2 the
observed OA concentrations throughout the troposphere in the Pacific Basin
and matches the observations remarkably well in the lower troposphere of the
Atlantic Basin that is heavily influenced by biomass burning emissions.
Reasonable agreement is found for the OA vertical distribution over the NH
Atlantic and Pacific oceans, especially in the lower troposphere (
Comparison of latitude-averaged predicted OA vertical profiles
with ATom-1 and ATom-2 measurements taken over the Pacific
By comparison, the results of the ATom model ensemble show much better agreement with observations. The model spread is still substantial but mostly below a factor of 5. Figures S6 and S7 show OA vertical profiles for individual ATom models and the spread in their results. In most regions, the ATom model ensemble captures both the absolute concentrations and the shape of the vertical profiles reasonably well. In the biomass burning outflow and NH midlatitude regions, the ATom ensemble average better captures the higher OA concentrations in the boundary layer and lower OA concentrations in the lower stratosphere than the AeroCom-II ensemble. We note that using the ensemble median OA profiles instead of ensemble mean OA profiles (as shown in Figs. 5 and S7) results in a slightly lower values of OA but does not change the conclusions of the model–measurement comparisons (Fig. S18).
In addition to OA mass concentrations, we also evaluate the model's ability
to simulate their degree of oxygenation, an indicator of their oxidation and
aging (Aiken et al., 2008; Kroll et al., 2011). Ambient measurements of the
oxidation level of organic particles are limited (Aiken et al., 2008;
Canagaratna et al., 2015), and the ATom dataset provides the first global
distribution of
Distribution of the OA
Note that for organosulfates (R-O-
Importantly, this ratio is also used to calculate the total OA mass
concentration for models that provide their outputs in terms of organic
carbon concentrations ([OA]
These results demonstrate that current global chemistry–climate models use
unrealistically low OA
We further assess whether global models can adequately predict the relative
contributions of primary and secondary OA. We strive to quantify these
fractions with the most straightforward methods (with the fewest
assumptions) for both models and measurements. POA concentrations were
estimated from the BC measurements by using an emission ratio appropriate for
the air mass origin (biomass burning vs. anthropogenic), as quantified by the
Figure 6 compares the vertical profiles of measurement-derived POA during ATom-1 and predicted by the CESM2-DYN model over clean remote regions of the Pacific Basin and northern polar Atlantic that are not influenced by biomass burning. Comparisons for other models are similar (not shown). Observations show that POA is extremely small in remote regions, whereas the model predicts that about half of the OA is made of POA in those areas. Although the model reproduces the measured total OA quite well, it tends to severely overpredict the amount of POA and underpredict that of SOA over clean remote regions (with the two errors canceling each other when it comes to total OA). Over biomass burning regions (not shown here) it can be difficult to directly quantify POA and SOA with this method, as total OA remains about constant, while POA decreases with aging and SOA increases (Cubison et al., 2011; Jolleys et al., 2015; Hodshire et al., 2019b). However, given this evolution the method used here would lead to an overestimate of POA for this reason.
Comparison of averaged POA and SOA vertical profiles as observed during ATom and as predicted by the CESM2-DYN model over the non-BB-influenced Pacific and Atlantic basins. The comparison is not shown for the strongly biomass-burning-influenced regions as all the OA is conservatively allocated to POA in those regions.
A more general comparison is made in Fig. 7 using the frequency
distributions of the measured and simulated fraction of POA
Frequency distribution of the observed and simulated ratio of POA to total OA in the free troposphere during ATom-1 and ATom-2 as computed by the GC12, CESM2, and CESM1-CARMA models.
The differences are so large that they are pretty insensitive to the details of
the POA estimation method from the measurements, mostly because for the vast
majority of the ATom track BC
Additional sensitivity tests were performed to investigate the impact of
noisy data and uncertainties of
A comparison between simulations that have the same treatment of POA, and only differ in their chemistry and removal of SOA (e.g., CESM2-SMP vs. CESM2-DYN; GC12-REF vs. GC12-DYN), indicates that a more complex SOA treatment does not always result in a more accurate simulation of the primary–secondary character of OA, a result that was also found in the AeroCom-II multi-model intercomparison (Tsigaridis et al., 2014).
Finally, we have examined whether the non-volatile treatment of POA in models could lead to these unrealistically high POA fractions in remote regions. Figure S16 shows a comparison of POA vertical profiles as predicted by the GC12-REF simulations that use non-volatile POA and a sensitivity simulation, GC12-REF-SVPOA, that uses semi-volatile POA similar to the standard treatment in GEOS-Chem as described in Pai et al. (2020). Note, however, that Pai et al. (2020) included marine POA emissions, used different reanalysis meteorology, and used a different model version (12.1.1 rather than 12.0.1 here), so their resulting comparisons to ATom measurements are somewhat different than found here for GC12-REF-SVPOA. The comparison indicates that the POA concentrations increase substantially in most regions when the semi-volatile POA parameterization is used. These results suggest that the non-volatile treatment of POA is not responsible for the model bias.
In this section, we further investigate some of the possible reasons for the incorrect model predictions of the relative contributions of POA and SOA in remote regions. Given the tendency of models to underestimate OA close to anthropogenic source regions and overestimate OA downwind in past studies (e.g., Heald et al., 2011; Tsigaridis et al., 2014; Hodzic et al., 2016), in this section we investigate the sensitivity of OA to increasing sources and increasing removals. We have performed two additional model simulations to test the sensitivity of the POA–SOA fractions to uncertainties in the representation of (i) wet scavenging based on the CESM1-CARMA simulations in which we have removed the improvements in the aerosol removal by the convective updrafts (Yu et al., 2019) and of (ii) SOA formation based on the GC12-REF simulations in which we have replaced the SOA formation VBS mechanism (Pye et al., 2010) by an updated VBS mechanism that uses chamber wall-loss-corrected SOA yields (Hodzic et al., 2016; the same formation scheme that is used in GC12-DYN and CESM2-DYN runs, but with removals kept identical to GC12-REF). The results of these two sensitivity simulations are displayed in Fig. 8, which shows measured and predicted mass concentrations of OA, POA, SOA, and sulfate for ATom-1 as a function of the number of days since the air mass was processed through convection. One should keep in mind that this is an averaged plot that includes air masses from various regions and altitudes, and it is not a Lagrangian plot following the same air mass.
Measured and predicted mass concentrations of POA, SOA, OA, and
sulfate aerosols during ATom-1 as a function of the number of days since the
air mass was processed through convection (based on a trajectory model from
Bowman, 1993, and satellite cloud data from NASA Langley;
Inefficient wet removal of primary OA could contribute to the POA
overprediction in global models, especially in the tropics. Previous global
model studies have reported overestimation by 2 to 3 orders of magnitude of
primary carbonaceous species such as BC in the free troposphere when
removal in convective updrafts was not included (e.g., Schwarz et al.,
2013; Yu et al., 2019). A strong reduction due to convective removal is also
expected for POA concentrations, as POA is a primary species co-emitted with
BC at the surface and internally mixed with it (Lee et al., 2015) and that
is typically coated by secondary inorganics and organics over short
timescales (Petters et al., 2006; Mei et al., 2013; Wang et al., 2010).
Figures 7a and 8 compare the simulations of CESM1-CARMA with and without
improved convective in-cloud scavenging during ATom-1. The improved in-cloud
scavenging scheme considers aerosol activation into cloud droplets from
entrained air above the cloud base, which is more realistic and results in a
more efficient removal of aerosols in the upper troposphere by convection. For example, a 2-orders-of-magnitude reduction in BC in the upper FT was reported
by Yu et al. (2019), resulting in much improved agreement with observations.
Similar results were observed for sea salt aerosols in Murphy et al. (2019).
Figure 8 shows that all submicron aerosol species simulated in CESM1-CARMA
are strongly impacted by in-cloud removal above the cloud base. POA
concentrations are reduced by an order of magnitude, while sulfate is reduced
by 30 %, leading in both cases to much-improved agreement with
observations. SOA is reduced by
For the CESM2-DYN model that does not have improved in-cloud removal, the reasonable agreement (within 20 %) with the observed OA concentrations thus results from coincidental error compensation between the overpredicted POA and underpredicted SOA. The prescribed SOA formation and the artificial 50 % adjustment of SOA emissions based on Liu et al. (2012) in CESM2-SMP lead to an overestimation of observed SOA in aged remote air masses.
In addition, we have also
tested the sensitivity of the OA composition to the choice of the SOA
formation mechanism. Figure 8 compares the results of the GC12-REF model
that uses SOA formation yields derived from traditional chamber experiments
(Pye et al., 2010) and those corrected for losses of organic vapors onto
chamber walls as proposed in Hodzic et al. (2016). Previous studies have
reported that chamber wall losses could lead to the underprediction of formed SOA by up to a factor of 4 (Zhang et al., 2014; Krechmer et al., 2016).
It should be noted that, in both cases, isoprene SOA is formed in aqueous
aerosols following Marais et al. (2016). The comparison shows a factor of 3
increase in SOA concentrations when the updated SOA formation is considered,
leading to much better agreement with the observed SOA and the
observed total OA. GC12-REF predicts the amount of POA well and somewhat overpredicts
the amount of sulfate aerosols, which is expected as it already
includes the improved aerosol removal in convective updrafts (Wang et al.,
2014). Figure S6 also shows that the POA vertical distribution is well captured
in GEOS-Chem in most regions, except over the polar North Pacific. It should
be noted that these results are consistent with the POA
These sensitivity simulations suggest that a stronger convective removal of POA and a stronger production of SOA might be needed to correctly represent not only the total OA concentrations but also its primary and secondary nature in the remote free troposphere and remote ocean regions. Accurate predictions of the OA concentration, composition, and source contributions for the right reasons are key for accurately predicting their life cycle and radiative impacts. Only when there is confidence that the sources are accurately predicted can we have confidence in OA predictions for preindustrial and future conditions, as well as evaluating PM mitigation strategies.
Finally, we assess the model ability to predict relative amounts of OA and sulfate in the free troposphere where they are the two major constituents of submicron aerosol (Fig. 2b). Accurate predictions of their relative contributions are crucial to determine the hygroscopicity of submicron aerosol and its ability to serve as cloud condensation nuclei (CCN) in the remote free troposphere (Carslaw et al., 2013; Brock et al., 2016).
Figure 9a compares the average measured relative fractions of sulfate
(36 %) and carbonaceous aerosols (OA
Comparison of the measured and predicted composition of submicron
aerosols as a function of altitude over the remote Southern Ocean region
during NH Winter (ATom-2). For models that do not calculate ammonium in the
aerosol (such as CESM1-CARMA, CESM2-SMP, CESM2-DYN, and ECHAM6-HAM), ammonium
was estimated from the sulfate mass assuming the formation of ammonium
sulfate. Note that while the modeled and measured submicron sea salt size
ranges agree fairly well (Table 1), this is not quite the case for dust.
Given that the accumulation-mode dust in the models presented contains
larger sizes than the AMS range (
Figure 9b shows the frequency distribution of the observed and predicted
fractions of OA relative to sulfate during ATom-1 and ATom-2 in the free
troposphere. Most models fail to reproduce the relatively uniform nature of
the observed distributions during ATom-1, with typically narrower model
shapes around a preferred ratio. The NH summer measurements indicate that OA
is greater than sulfate in
The discrepancies between the observed and predicted composition of submicron aerosol over remote regions can be quite large for other constituents as well. Figure 10 shows the comparison of the measured and predicted composition of submicron aerosol over the Southern Ocean (during the NH winter) where the disagreement in simulated sea salt, nitrates, ammonium, and MSA often exceeds the contribution of OA. While the observations show a more uniform distribution of non-marine aerosol with higher values in the middle and upper troposphere, respectively, most models tend to simulate the highest fractions of OA (and sulfate) towards the tropopause. This may also be explained by the uncertainties in the modeled wet removal of aerosol discussed above. Specific studies have discussed and continue to investigate the ATom measurements and simulations of different components in more detail, including particle number (Williamson et al., 2019), black carbon (Katich et al., 2018; Ditas et al., 2018), MSA (Hodshire et al., 2019a), sulfate–nitrate–ammonium (Nault et al., 2019), and sea salt (Yu et al., 2019; Bian et al., 2019; Murphy et al., 2019).
Our understanding and representation in global models of the life cycle of OA remain highly uncertain, especially in remote regions where constraints from measurements have been very sparse. We have performed a systematic evaluation of the performance of eight global chemistry–climate models and of 28 AeroCom-II models in simulating the latitudinal and vertical distribution of OA and its composition in the remote regions of the Atlantic and Pacific marine boundary layer, free troposphere, and lower stratosphere using the unique measurements from the ATom campaign. Our simulations are conducted for both ATom-1 and ATom-2 deployments that took place in August 2016 and February 2017, respectively. The main conclusions of the comparison are as follows.
The AeroCom-II ensemble average tends to be biased high by a factor of 2–5 in comparison to measured vertical OA profiles in the remote atmosphere during both NH summer and NH winter. The ensemble spread increases from a factor of 40 in the NH source regions to a factor of 1000 in remote regions of the Southern Ocean. The evaluation of AeroCom-II models in remote regions provides an extension of the previous evaluation with continental ground data by Tsigaridis et al. (2014). We note that the data from the AeroCom-II models were based on monthly mean values from a different simulated year than the ATom campaigns; however, the consistent model biases are strong enough that we would not expect our conclusions to change for a different modeled year.
The results of the ATom model ensemble used in this work show much better agreement with the OA observations in all regions and reduced model variability. However, some of the agreement is for the wrong reasons, as most models severely overestimate the contribution of POA and underestimate the contribution of SOA to total OA. Sensitivity simulations indicate that the POA overestimate in CESM could be due to an inadequate representation of primary aerosol removal by convective clouds (additional convective removal per Yu et al., 2019, in CESM1-CARMA led to better agreement with observations). Most models have insufficient production of SOA, and sensitivity studies indicate that a stronger production of SOA is needed to capture the measured concentrations. The photochemical aging of POA, which was not considered here (unlike for SOA), could also contribute to the model overestimation. The non-volatile POA treatment in models is consistent with the assumption of inert POA particles used to estimate POA from measurements and cannot explain the model bias. Indeed, sensitivity simulations with semi-volatile POA lead to a much larger model bias for OA in the upper troposphere and remote regions. The compensation between errors in POA and SOA in remote regions is, however, a recurring issue in OA modeling (de Gouw and Jimenez, 2009). For instance, it was found in urban outflow regions such as Mexico City during the MILAGRO 2006 field campaign (Fast et al., 2009; Hodzic et al., 2009), Paris during MEGAPOLI 2009 (Zhang et al., 2013), the Los Angeles area during CalNex-2010 (Baker et al., 2015; Woody et al., 2016), and the NE US outflow during WINTER 2015 (Schroder et al., 2018; Shah et al., 2019).
Additional errors in simulated OA concentrations can arise from the use of
OA
The results also show that in most models (except CESM2) the predicted OA contribution to the total submicron aerosol is underestimated relative to sulfate in the remote FT where OA and sulfate are the dominant submicron aerosols (important for climate). Accurate predictions of the composition of submicron particles remains challenging in remote regions and should be the topic of future studies.
The key implications of our results are the following: (i) model errors on the relative contribution of POA and SOA to OA reduce our confidence in the ability to simulate radiative forcing over time or OA health impacts; (ii) model errors for the relative contributions of sulfate and organics to the submicron aerosol in the free troposphere could lead to errors in the predicted CCN or radiative forcing of aerosols as inorganics are more hygroscopic than OA; and (iii) the OA system seems to be more dynamic with a need for an enhanced removal of primary OA and a stronger production of secondary OA in global models to provide better agreement with observations.
Data can be obtained from the ATom data repository at the NASA/ORNL DAAC:
All Global Modeled and HR-AMS Measured OA concentrations and related properties data for ATom used in this publication is archived at ORNL DAAC, Oak Ridge, Tennessee, USA:
The supplement related to this article is available online at:
AH, PCJ, and JLJ performed the measurement–model comparisons and wrote and revised the paper. PCJ, DAD, BNN, JCS, DTS, and JLJ performed and analyzed the AMS measurements. KDF and GPS performed and analyzed the PALMS measurements. JPS and JMK performed the BC measurements. HB, MC, PRC, BH, AH, DSJ, JKK, JRP, ER, JS, IT, ST, KT, and PY provided model output. All authors provided comments on the paper.
The authors declare that they have no conflict of interest.
This paper has not been reviewed by the EPA, and no endorsement should be inferred.
The authors want to thank the ATom leadership team and the NASA logistics and flight crew for their contributions to the success of ATom. The authors acknowledge Rebecca Buchholz (NCAR) for providing the emissions used for the CESM2 simulations. We thank Charles Brock (NOAA), Christina Williamson (NOAA), and Agnieszka Kupc (U. of Vienna, Austria) for the aerosol volume data, Paul Wennberg (Caltech) for HCN data, and Eric Apel and Rebecca Hornbrook (NCAR) for the CH3CN data used in Fig. S20. We thank Daniel Murphy (NOAA) for useful discussions. We would like to acknowledge high-performance computing support from Cheyenne provided by NCAR's Computational and Information Systems Laboratory.
This research was supported by the National Center for Atmospheric Research, which is operated by the University Corporation for Atmospheric Research on behalf of the National Science Foundation, NASA (grant nos. NNX15AH33A, NNX15AJ23G, and 80NSSC19K0124), the DOE (grant nos. DE-SC0016559, DE-SC0019000), the ERC (grant no. 819169), and EPA STAR (grant no. 835877010).
This paper was edited by Sergey A. Nizkorodov and reviewed by three anonymous referees.