Concentrations of POA and SOA
Hourly POA and SOA concentrations in multiple size fractions were calculated
throughout the 9-year simulation period and then averaged to daily and
monthly average concentrations. Although the focus of the current study is
PM0.1 POA and SOA, the predicted PM2.5 OA concentrations were also
calculated and compared to measurements as a confidence-building exercise
(since PM0.1 measurements are not routinely available). Model
calculations predict organic matter (OM) concentrations while ambient
measurements quantify organic carbon (OC) concentrations. Simulated OM
concentrations are converted to OC concentrations using an OM / OC ratio of
1.6 for POA (Turpin and Lim, 2010) and species-specific OM / OC ratios
for SOA species taken from Table 1 in Carlton et al. (2010). Detailed evaluation of the model performance for
PM2.5 OC (and other PM/gaseous species) has been presented in the
first paper in the series (Hu et al., 2015). In summary, predicted
monthly average PM2.5 OC has a mean fractional bias (MFB) of -0.32 and a mean
fractional error (MFE) of 0.43. Monthly MFB and MFE (Eqs. S1 and S2 in the Supplement)
calculated using daily average OC generally meet the model performance
criteria proposed by Boylan and Russell (2006) (Eqs. S3 and S4).
Monthly source contributions to PM2.5 total OC at seven urban
sites. Observed total OC concentrations are indicated by the circles with dots,
and predicted OC concentrations from different sources are indicated by the
colored areas.
Figure 1 illustrates the time series of the predicted and measured
monthly average total PM2.5 OC concentrations at seven major urban
locations: (a) Sacramento, (b) San Jose, (c) Fresno, (d) Bakersfield, (e) Los
Angeles, (f) Riverside, and (g) El Cajon. At each site, daily average
measured concentrations of the PM2.5 total mass and OC were obtained
from California Air Resources Board (CARB) (CARB, 2011) “1 in 3”
sampling network and averaged over the 9-year period. Measured PM2.5 OC
concentrations at all sites show strong seasonal variation with higher
concentrations in winter months and lower concentrations in summer months.
OC concentrations predicted by the UCD/CIT model generally capture the
monthly average concentrations and seasonal variations with MFB ranging from
-0.31 to 0.19 and MFE ranging from 0.4 to 0.59. However, the model predicts
much weaker trends of PM2.5 OC over the 9 years at Los Angeles and
Riverside, indicating that the declining emission trends might not be well
represented in the inventory. At Sacramento and Fresno, the measured monthly
average OC concentrations frequently exceeded 10 µg m-3 in winter
and the maximum monthly OC concentrations reached or exceeded
∼ 25 µg m-3. Wood smoke is predicted to be the
dominant OC source in winter at the two locations, contributing over 70 %
of the total OC concentrations on average. Wood smoke is also predicted to
be the dominant OC source in winter at San Jose and Bakersfield. Model
calculations tend to overpredict the winter OC concentrations at San Jose,
indicating that the wood smoke emissions are likely overestimated in this area.
This is consistent with more recent surveys of home heating fuels conducted
by the Bay Area Air Quality Management District (BAAQMD). Model calculations
generally underpredict OC in summer when concentrations are lower. Meat
cooking and other anthropogenic sources are predicted to be the largest
sources in summer at Sacramento, San Jose, Fresno, and Bakersfield. Together
these two categories contribute over 86 % of the total predicted OC in
summer. Both measured and predicted seasonal variation is weaker at Los
Angeles and Riverside than in Northern California due to smaller wood smoke
contributions. Meat cooking and other anthropogenic sources make the largest
predicted contributions to OA at these two Southern California locations.
Mobile sources (gasoline and diesel engines) also contribute approximately
30 % of the total PM2.5 OC at Los Angeles. Model calculations tend to
underpredict PM2.5 OC concentrations in all seasons in 2000–2006 at
Riverside (approximately 80 km downwind of the Los Angeles urban center).
Intense emissions transported from the upwind Los Angeles areas along with
the meteorology and topography enhances photo-oxidation of volatile organic
compounds (VOCs) and formation of SOA at this location. A measurement study
of organic aerosols at Riverside in summer indicated high SOA fraction of
the total OA (TOA) with an average SOA / OA ratio of 0.74
(Docherty et al., 2008). The PM2.5 OC
underprediction at Riverside during summer and the general underprediction
in summer at other sites may indicate that some important precursors and
pathways of PM2.5 SOA are missing or only partially included in the
current SOA module, such as SOA formation from glyoxal and methylglyoxal
(Ervens and Volkamer, 2010; Fu et al., 2008; Ying et al., 2015) and from
aerosol aqueous-phase chemistry (Volkamer et al., 2009), the conversion of intermediate-volatility compounds to SOA (Jathar
et al., 2014; Zhao et al., 2014), or SOA forming with higher yields than
included in the module (Zhang et al., 2014; Cappa et al., 2016).
Observed (obs) and predicted (model) OC / mass ratios
in (a) PM2.5 and (b) ultrafine and quasi-ultrafine PM. In (a), a sensitivity
analysis is conducted by removing the dust concentration from the PM2.5
total mass (model_no_dust). The ultrafine and
quasi-ultrafine data in (b) are extracted from published literature as
indicated in the figure.
POA and SOA concentrations estimated by the CMB method (left gray
columns) and predicted by the UCD/CIT model (right dark columns). Error bars
represent the standard deviation of concentrations estimated during the
sampling periods by both methods. The uncertainties of CMB-derived SOA range
from 1 to 22 % (Daher et al., 2012). The data are for sampling periods
in 2005–2007 at four sites in Southern California.
Predicted 9-year average (a) PM0.1 total OA (TOA)
concentration and (b) PM0.1 SOA / TOA ratio in California.
The 9-year average PM0.1 SOA concentrations derived
from (a) AALK, (b) AXYL, (c) ATOL, (d) ABNZ, (e) AISO, (f) ATRP, (g) ASQT, (h) AOLGA,
and (i) AOLGB. Note that AXYL and ATOL are actually derived from lumped aromatics species
ARO2 (groups of aromatics with kOH > 2 × 104 ppm-1 min-1, including xylenes and other di- and
polyalkylbenzenes) and ARO1 (groups of aromatics with kOH < × 104 ppm-1 min-1, including toluene and
monoalkylbenzenes). The color scales (shown in the last panel in unit of %) indicate the ratios of the concentrations to the maximum 9-year
average values, which are shown in the panels under species names with a
unit of ng m-3.
Figure 2a compares the average PM2.5 OC / mass ratios estimated from
ambient measurements and the values predicted by the UCD/CIT model over the
9-year study period at seven representative urban locations. Predicted
concentrations on the corresponding days were extracted and averaged for the
comparison. The average OC / mass ratios were then calculated. The observed
average OC / mass ratios vary in the range of 0.24 (at Riverside) to 0.45 (at
Sacramento). The predicted average OC / mass ratios are in relatively good
agreement with measured values at Los Angeles, Riverside, and Bakersfield
(difference < 20 %) but not at Sacramento, San Jose, Fresno, and
El Cajon (difference > 35 %). The predicted average OC / mass
ratios are consistently lower than observed ratios, by 0.01 (3 % at Los
Angeles) to 0.22 (48 % at Sacramento). This underprediction is partly
attributed to the underprediction of OC concentrations, especially the SOA
concentrations, as well as to the overprediction of total mass concentrations
due to overestimated dust emissions (Hu et al., 2014a, 2015).
The seasonal average dust emissions used in the current study were not
adjusted based on wind speed and soil moisture. A sensitivity analysis was
conducted by removing the dust concentrations from the predicted PM2.5
mass (Fig. 2a). The average predicted OC / mass ratio increased from 0.22 to
0.29 (average across the seven sites), compared to the observed ratio of
0.33. Omission of dust from the model predictions improves agreement with
OC / mass measurements at all sites except central Los Angeles, although
OC / mass without dust is still lower than measurements at four sites
(Sacramento, San Jose, Fresno, and El Cajon), indicating OC predictions are
likely biased low at these locations.
Figure 2b compares the predicted and observed OC / mass ratios in the
ultrafine (PM0.1) or quasi-ultrafine (PM0.18, PM0.25)
particles. The ultrafine/quasi-ultrafine measurement data were compiled in a
previous study (Hu et al., 2014a) from published literature
(Herner et al., 2005; Kim et al., 2002; Krudysz et al., 2008; Sardar et al., 2005a, b). The ultrafine or quasi-ultrafine data are more
sparse than the PM2.5 data, but they still cover a sufficient total number
of days to allow for robust comparison. The observed OC / mass ratios in
ultrafine/quasi-ultrafine sizes vary from 0.43 (at Modesto) to 0.71 (at
USC). The predicted ultrafine/quasi-ultrafine OC / mass ratios generally agree
well with observed values at all sites. The generally better agreement of
OC / mass ratios in the ultrafine/quasi-ultrafine size range compared to the
PM2.5 size range reflects the fact that SOA formation and dust
emissions make limited contributions to ultrafine/quasi-ultrafine
concentrations. Condensation of the semivolatile products to form SOA
mostly takes place in the particle accumulation mode and is generally not
dominant in the ultrafine size range due to the increase in the saturation
vapor pressure above small particles (Kelvin effect). Dust components mainly
contribute to coarse and fine particles but make little contribution to the
ultrafine particles.
The primary and secondary fraction of total OA cannot be directly measured
in ambient OA samples. A few indirect methods have been developed to
estimate the POA and SOA concentrations, such as molecular marker-based
method (Daher et al., 2011, 2012; Ham and Kleeman, 2011;
Kleindienst et al., 2007), elemental carbon (EC) tracer method (Cabada et al., 2004; Lim et al., 2003; Polidori et al., 2006, 2007;
Turpin and Huntzicker, 1995), water-soluble OC content method
(Weber et al., 2007), aerosol mass spectrometry factorization method
(Aiken et al., 2008; Lanz et al., 2007; Ulbrich et al., 2009), and the
unexplained fraction of OA by tracers for major POA categories (Chen et al., 2010; Schauer and Cass, 2000). In the current study, PM2.5 SOA
concentrations were estimated by the molecular marker chemical mass balance
(CMB) method (Daher et al., 2012) during sampling periods in
2005–2007 at four locations. PM2.5 POA concentrations were then
estimated by subtracting PM2.5 SOA concentrations estimated by the CMB
method from the total measured OA concentrations. Figure 3 shows the
PM2.5 POA and SOA concentrations predicted by the UCD/CIT model (right
dark columns) compared to the PM2.5 POA and SOA concentrations
estimated using the CMB method (left gray columns). Error bars represent the
standard deviation of concentrations estimated during the sampling periods.
The total PM2.5 OA (i.e., POA + SOA) concentrations predicted by the
UCD/CIT model generally agree with measured values (with fractional bias
within ±35 %) except at the Riverside site (with a fraction bias of
-63 %). However, the PM2.5 SOA concentrations predicted by the UCD/CIT
model appear to be a factor of 2∼ 3 lower than the SOA
concentrations estimated by the CMB method (ratio ranging from 2.2 at
Riverside to 2.8 at WSanG). The PM2.5 POA concentrations predicted by
the UCD/CIT model are higher than those estimated by the CMB method at WSanG
and ESanG1. This may reflect the effects of POA volatility. Studies have
indicated that some fraction of POA emissions will evaporate, and this
material may undergo photo-oxidation and condense back to particle phase
(Robinson et al., 2007). In the current model, POA is treated as
nonvolatile. Thus, no such evaporation occurs. However, the substantial
underprediction of PM2.5 SOA at all sites suggests that some SOA
precursors and pathways are likely missing from the current SOA mechanism.
Both PM2.5 POA and SOA are underpredicted at Riverside, indicating
that some important sources are likely missing in that area.
Monthly source contributions to PM0.1 SOA at six urban sites.
Predicted PM0.1 SOA concentrations from different sources are indicated
by the colored areas.
Predicted source contributions to 9-year average PM0.1 POA
concentrations. The color scales (shown in the last panel in unit of %)
indicate the ratio of the concentrations to the maximum 9-year average
concentration values, which are shown in the panels under source names with
a unit of ng m-3.
Figure 4 illustrates the predicted total PM0.1 OA concentrations
(Fig. 4a) and the predicted ratios of SOA to total OA averaged over the 9-year modeling period (Fig. 4b). High total PM0.1 OA concentrations
with maximum concentrations > 2 µg m-3 are located in
urban areas where the POA emissions are large due to human activities.
Predicted PM0.1 SOA generally accounts for less than 10 % of total
PM2.5 OA at urban areas, but predicted SOA contribute to
10–20 % of total OA in suburban areas and to
20–50 % in rural areas. The spatial distribution of
PM2.5 SOA concentrations and the ratios of SOA to total OA (shown in
Fig. S1 in the Supplement) are generally similar to those of PM0.1, but PM0.1 OA
has sharper spatial gradients and the PM0.1 SOA fraction is lower than
that in PM2.5 in urban areas, indicating POA contributes more in the
ultrafine size range.
Figure 5 shows the contributions from the nine precursor species to the
PM0.1 SOA concentrations (results of PM2.5 SOA are shown in Fig. S2).
Maximum SOA concentrations are located in southern part of the San Joaquin Valley (SJV).
Monoterpenes, sesquiterpenes, oligomers, and long alkanes are the most
important precursors, contributing over 90 % of the total SOA in most
areas, while other precursors (xylene, toluene, and benzene) in total
contribute less than 10 ng m-3 to SOA concentrations. These finding are
very dependent on the treatment of vapor wall losses during the formulation
of the SOA model. The contributions from different precursors to SOA
concentrations have very different spatial distributions. Long-chain alkanes
form SOA mainly in the urban areas of Southern California and in the
middle-southern portion of the SJV. Isoprene, monoterpenes, and
sesquiterpenes form SOA at coastal and foothill locations where the biogenic
emissions are greatest. The longer lifetime of long-chain alkanes than
isoprene leads to a broader spatial distribution for the SOA derived from
alkanes. The spatial distribution of oligomers of anthropogenic SOA
(Oligomer_A) and biogenic SOA (Oligomer_B)
reflects the patterns of SOA derived from long-chain alkanes and the total
biogenic species. The relative spatial patterns associated with each
precursor are generally not sensitive to the exact formulation of the SOA
model (see Sect. 3.3).
Sources of POA and SOA
Figure 6 displays the time series of monthly average PM0.1 SOA source
contributions at the six major urban locations. PM0.1 SOA
concentrations are high in summer (100–300 ng m-3) and
low (20–50 ng m-3) in winter, reflecting the seasonal
variation in photochemistry. PM0.1 SOA concentrations are higher at
Fresno and Bakersfield than other sites due to larger biogenic source
contributions. Biogenic emissions are the largest source of PM0.1 SOA
across all sites, followed by the other anthropogenic sources (mainly
solvent usage and waste disposal emissions, see Fig. S3). On-road gasoline
engines are an important source of SOA at Los Angeles and Riverside. Similar
source contributions to PM2.5 SOA are found and shown in Fig. S4.
Predicted source contributions to 9-year average PM0.1 SOA
concentrations. The definition of the color scales is the same as in Fig. 7.
(a) Predicted population weighted concentrations (PWCs) of SOA in
six counties in Southern California. Two sets of simulations (scenarios)
conducted by Cappa et al. (2016) were used, one with the low-NOx,
high-yield parameters (denoted as “highyield”) and the other with
high-NOx, low-yield parameters (denoted as “lowyield”). Each set
of simulations included three vapor wall loss cases, i.e., no considering of
vapor wall losses (denoted as “base”), low vapor wall loss rates (denoted
as “lowwallloss”), and high vapor wall loss rates (denoted as
“highwallloss”). (b) Normalized PWCs of SOA in all counties to the PWC of
SOA in Orange County. (c) Changes in the normalized PWCs of SOA in all
counties by accounting for vapor wall losses.
Figure 7 shows the predicted regional source contributions of PM0.1 POA
averaged over the 9-year modeling period. The important regional sources of
PM0.1 POA over the entirety of California are predicted to be other
anthropogenic sources (contributing 39.6 %), wood smoke (37.1 %),
on-road gasoline (9.1 %), and meat cooking (5.8 %). Wood smoke is the
dominant POA source especially in Northern California, with the maximum
PM0.1 POA contribution exceeding 1 µg m-3. Meat cooking and
mobile (on-road and off-road) sources are the major sources in urban areas,
especially in metropolitan areas such as the Greater Los Angeles Area and the
San Francisco Bay Area. Other anthropogenic sources from another major
category in the urban centers in the SJV and also the Los Angeles areas.
High-sulfur-content fuel sources are mainly located around the ports in the
Los Angeles and San Francisco Bay areas. The regional source contributions
of PM0.1 POA are quite different from those of PM2.5 POA (shown in Fig. S5). The PM2.5 POA source contributions are much more widespread
than the PM0.1 POA sources contributions because PM2.5 has a
longer lifetime due to slower deposition and coagulation compared to
PM0.1. For example, the mobile sources and the other anthropogenic
sources contribute greatly to PM2.5 POA throughout the entire SJV but
only contribute to PM0.1 POA in urban centers.
Figure 8 shows the predicted regional source contributions of PM0.1 SOA
averaged over the 9-year modeling period (and Fig. S6 shows the PM2.5
SOA results). Biogenic emission is predicted to be the single largest
PM0.1 SOA source in the present study, contributing 63.7 % of the
PM0.1 SOA over the entire California. The maximum biogenic PM0.1
SOA concentration is up to 0.1 µg m-3 around Bakersfield in the
southern SJV. Other anthropogenic sources (22.2 %) and on-road gasoline
engines (10.8 %) are predicted to be the most important anthropogenic
sources of PM0.1 SOA in California. The spatial distribution of
PM0.1 SOA concentrations from these anthropogenic sources are similar
(but different from the spatial distribution of SOA from biogenic sources)
with high concentrations in Southern California. PM0.1 SOA formation
from on-road diesel engines, off-road diesel engines, wood smoke, meat
cooking, and high-sulfur fuel combustion are small, with PM0.1 SOA
contributions generally less than a few ng m-3. A recent epidemiological
study has revealed that anthropogenic PM0.1 SOA is highly associated
with ischemic heart disease mortality (Ostro et al., 2015).
Therefore, the results in this study suggest that control of solvent usage,
waste disposal, and mobile emissions should be considered to protect public
health in California, but the exact determination of source controls will
need to be evaluated after the SOA formation mechanism is updated.
Influence of vapor wall losses on SOA exposure in California
The SOA concentrations predicted in the current study are based on the SOA
yield data measured in chamber experiments. A recent study has demonstrated
that organic vapors can be lost to chamber walls during SOA formation
experiments, resulting in SOA yields that are biased low (Zhang et al., 2014). Efforts have been carried out to parameterize the effect of vapor
wall losses on SOA formation in the UCD/CIT air quality model to account for
this effect when predicting ambient SOA concentrations in Southern
California (Cappa et al., 2016). SOA concentrations are
predicted to increase by factors of 2–5 with low vapor wall loss rates and
by factors of 5–10 with high vapor wall loss rates compared to the
concentrations in the simulations with no consideration of vapor wall
losses. Due to low SOA / TOA fractions (< 10 %) at the observation
sites located in urban areas (Figs. 4 and S1), the substantial
increase of SOA by the vapor wall loss corrections does not strongly change
the total OA concentrations and therefore does not significantly affect the
model evaluation results shown in Fig. 1. Here we further analyzed the
changes in the population weighted concentrations (PWCs) of SOA when vapor
wall losses are accounted for. Two sets of simulations (scenarios) conducted
by Cappa et al. (2016) are considered, one with the low-NOx, high-yield
parameters (denoted as “highyield”) and the other with high-NOx,
low-yield parameters (denoted as “lowyield”). Each set of simulations
included three vapor wall loss cases, i.e., no consideration of vapor wall
losses (denoted as “base”), low vapor wall loss rates (denoted as
“lowwallloss”), and high vapor wall loss rates (denoted as
“highwallloss”). PWCs of SOA are calculated for six counties in the
Southern California for the six scenarios. Spatial difference
in exposure is important in cohort studies; therefore the relative changes
of PWCs among counties are examined. Figure 9 shows the PWCs of SOA and
their relative changes in different scenarios in the six counties. The
results indicate that PWCs of SOA increase substantially by accounting for
vapor wall losses in all counties (panel a). However, the spatial pattern of
SOA PWC, as characterized by normalizing the PWC for each location by the
PWC in Orange County, is very similar in all scenarios (panel b).
Consequently, accounting for vapor wall losses changes the SOA exposure
ratio in different counties by only a small extent of < 15 % for
most scenarios/counties (panel c). These results suggest that future
simulations that account for vapor wall losses in SOA simulations will yield
increased absolute values of concentrations but will have spatial patterns
that are similar to the base case results in the current paper when used for
epidemiology studies.
Figure 9 suggests that associations between anthropogenic SOA and health
effects identified in previous epidemiological studies will prove robust to
future updates in SOA models. This finding also extends to the spatial
pattern of individual SOA precursors. The influence of vapor wall losses on
exposure to SOA formed from different precursors (i.e., long alkanes,
aromatics, isoprene, sesquiterpenes, and monoterpenes) is shown in Figs. S7–S11. In all cases, the spatial pattern of PWC for SOA derived from each
precursor is similar under all treatments of wall losses. Long alkanes and
aromatics are mainly from anthropogenic sources, and isoprene,
sesquiterpenes, and monoterpenes are mostly from biogenic sources. Further
detailed interpretation of source contributions to SOA and associated health
effects should only be carried out after new exposure fields are calculated
using the latest SOA models.