Emission sources and the relation of ambient to emission trends
This section incorporates previously published analyses by reference,
extends them through 2013, and integrates findings. Results related to
emission changes are compared with those obtained using other approaches in
Sect. 3.6 (Synthesis).
Southeastern emissions in 2013 are shown by source category in Table S2;
comparison with 2008 emissions reported in Blanchard et al. (2013) indicates
reductions since 2008. Statistically significant (p < 0.001)
relationships were found between mean annual PM2.5 EC and OC
concentrations at SEARCH sites and PM2.5 EC and OC emissions between
1999 and 2013 (Hidy et al., 2014). Ambient EC trends were significantly
related to both mobile source and total EC emissions, whereas ambient OC
trends were significantly related to mobile source OC emissions but not to
total OC emissions (Hidy et al., 2014). PM2.5 EC emissions in the
southeastern US declined by approximately half between 1996 and 2013 due
to reductions of on-road and non-road motor vehicle emissions (Hidy et al.,
2014). Corresponding declines occurred in on-road and non-road motor vehicle
PM2.5 OC emissions, but total PM2.5 OC emissions showed little
trend due to the dominance of relatively constant biomass burning emissions
(Hidy et al., 2014). Mobile source OC emissions represent less than 10 %
of OC emissions in the southeast (Blanchard et al., 2013; Hidy et al., 2014)
and only 4 % as of 2013 (Table S2), with biomass burning accounting for
∼ 75 % of OC emissions in emission inventories.
Using a receptor modeling approach, Blanchard et al. (2013) showed that
PM2.5 EC emissions generally account for reported mean annual EC
concentrations and trends in the SEARCH network (Fig. S2). Although the
receptor model overpredicted EC concentrations at the Jefferson Street (JST)
site in Atlanta, Georgia, and underpredicted EC concentrations at other
sites, the EC trends predicted by the model from the inventory agreed with
observed EC trends. Larger observed ambient EC decreases at SEARCH sites
coincided with an EC emission decline occurring between 2005 and 2013 that
resulted from new Environmental Protection Agency (EPA) standards for diesel
engines and fuels (effective in 2007 for on-road vehicles, in mid-2010 for
non-road mobile sources, and in mid-2012 for rail and marine sources) (Hidy
et al., 2014). Mobile sources account for over 50 % of EC emissions in the
southeast prior to 2007 (Blanchard et al., 2013) and decline to
∼ 40 % by 2013 (Table S2), so ambient EC concentrations are
expected to decrease with declining mobile source EC emissions.
Contrasting with results for EC (as well as carbon monoxide (CO), NOx, and SOx),
greater differences between receptor model-predicted OC and measured OC
trends were observed (Fig. S3). These differences occurred even when
comparing model predictions to the fraction of measured OC that was not
associated with O3 and SO4 (inventory OC emissions do not represent
SOA deriving from biogenic emissions of isoprene and other gases). Ambient OC
trends were more pronounced than trends predicted by the model from the
inventory (Fig. S3). However, the receptor model reproduces observed OC
trends more readily for sites where the mobile source contribution is
greatest (Fig. S3). Receptor-modeling studies have consistently identified
mobile source contributions to ambient PM2.5 mass concentrations in
Atlanta and Birmingham (e.g., Zheng et al., 2002; 2006; Baumann et al., 2008;
Lee et al., 2009); a recent analysis indicated that mobile sources
contributed 0.8 to 2.8 µg m-3 to 2006–2010 PM2.5 mass
concentrations (between 6–7 and 19–21 % of PM2.5 mass) in Atlanta
and Birmingham (Watson et al., 2015). Measured ambient concentrations of
non-polar PM2.5 OC species associated with motor vehicles, such as
hopanes and steranes, declined substantially (> 50 %) at Birmingham, Alabama (BHM)
and JST between 2006 and 2010, linking mobile source emission reductions
during those years with observed decreases in urban OC concentrations
(Blanchard et al., 2014a). As noted in the Appendix, emitted OC is not
conservative, but is affected by evaporation and possibly recondensation as
secondary species, or by augmentation by SOA derived from gas-phase
emissions. A possible explanation for the observed OC trends is that diesel
SOA concentrations (which were not incorporated in the receptor model
predictions) were greater prior to adoption of new diesel emission
regulations beginning in 2007. In addition, changes in gasoline-engine SOA
concentrations may have occurred. Reductions of SO2 emissions are also
thought to have changed SO4-associated SOA concentrations over time (Xu
et al., 2015a, b), but the chemical mass balance (CMB) model is set up to
predict OC that is not associated with O3 and SO4 (Blanchard et
al., 2013).
Trends in mobile source VOC emissions paralleled trends in mobile source
PM2.5 OC and EC emissions (Hidy et al., 2014; Blanchard et al., 2013).
Similar to OC emissions, mobile source VOC emissions in the southeastern
US declined by approximately half between 1996 and 2013 due to reductions
of on-road and non-road motor vehicle emissions (Hidy et al., 2014), but
total VOC emissions showed little trend due to dominance by relatively
constant VOC emissions from vegetation and soils (Table S2).
In summary, emission trends partially explain observed ambient EC and OC
trends. For OC, the link between inventory emissions and ambient
concentrations is less definitive than is the case for links between
reductions of EC, CO, NOx, and SO2 emissions and observed trends
in ambient EC, CO, NOy, SO2, and PM2.5 SO4
concentrations (Hidy et al., 2014).
Ambient EC and OC concentrations and trends
Trends and spatial variations are evident for mean annual and seasonal EC
and OC concentrations (Table 1 and Fig. 1). Mean EC concentrations were
2.0 to 3.5 times greater at JST than at CTR, thereby indicating two- to
threefold greater influence of combustion sources within Atlanta compared
to rural CTR because EC is a tracer of combustion (Appendix). Mean OC
concentrations were 1.0 to 1.8 times greater at JST than at CTR, indicating
urban sources of OC possibly superimposed on a relatively high regional
baseline. The ratio of JST EC to CTR EC declined from 2.8 : 1 to 2.1 : 1 between
the first and third 5-year periods, while the JST OC to CTR OC ratio
decreased from 1.5 : 1 to 1.2 : 1 between the first and third 5-year periods.
Since the ratio of JST EC / CTR EC declined by 25 % and the ratio of JST
OC / CTR OC declined by 20 %, the decreases are comparable but the
difference is consistent with a greater mobile source influence at JST than
at CTR. Both EC and OC concentrations exhibit decreasing trends at all
SEARCH sites (Hidy et al., 2014), particularly after 2007 but with a
possible rise between 2009 and 2011 (Fig. 1). Higher mean monthly
concentrations in 2011 were followed by further decline in 2012 and 2013
(Fig. 1). Whereas long-term ambient EC and OC trends are predicted by EC
and OC mobile source emission reductions (Sect. 3.1), changes between 2008
and 2013 are predicted from the emission inventory for EC (Fig. S2) but
not for OC (Fig. S3).
Seasonal mean EC and OC concentrations at CTR and JST. All
correlations among the four time series are statistically significant (p < 0.05): CTR EC and OC, r=0.68 (95 % CI 0.52–0.80); JST EC
and OC, r=0.87 (95 % CI 0.79–0.92); CTR EC and JST EC, r=0.76
(95 % CI 0.62–0.85); CTR OC and JST OC, r=0.68 (95 % CI 0.51–0.79).
No season consistently exhibits the highest mean EC and OC concentrations
but the CTR mean OC concentrations and OC / EC ratios are highest during
summer, interpreted as the influence of aging and SOA formation during
warmer months. In contrast, JST mean OC and EC concentrations tend to be
higher during autumn and winter (Table 1). In 2013, the ratios of OC to
total carbon (TC = EC + OC) in daily-average filter samples were
greatest at CTR during the SOAS campaign (Fig. S4). This result suggests
that rates of SOA formation at CTR during SOAS exceed SOA formation rates at
other sites in the region and at other times of the year. The differences
between JST and CTR mean summer OC concentrations decline from 1.1 µg m-3 in 1999–2003 to less than 0.1 µg m-3 in 2009–2013,
interpreted as reductions of urban OC concentrations toward a regional
baseline level (Table 1).
In total, 5-year seasonal mean EC and OC concentrations at CTR and JST with
mean OC / ECa.
Periodb
CTR EC
CTR OC
CTR OC / EC
JST EC
JST OC
JST OC / EC
1999–2003 W
0.490 ± 0.025
2.615 ± 0.154
5.34
1.725 ± 0.068
4.801 ± 0.153
2.78
1999–2003 Sp
0.607 ± 0.030
3.411 ± 0.154
5.62
1.410 ± 0.037
4.465 ± 0.096
3.17
1999–2003 Su
0.537 ± 0.020
3.541 ± 0.100
6.59
1.439 ± 0.035
4.664 ± 0.090
3.24
1999–2003 A
0.684 ± 0.026
3.814 ± 0.145
5.58
1.808 ± 0.060
5.264 ± 0.150
2.91
2004–2008 W
0.538 ± 0.036
2.348 ± 0.167
4.37
1.319 ± 0.050
4.099 ± 0.125
3.11
2004–2008 Sp
0.556 ± 0.029
3.269 ± 0.199
5.88
1.173 ± 0.034
4.283 ± 0.135
3.65
2004–2008 Su
0.528 ± 0.030
3.267 ± 0.151
6.19
1.292 ± 0.032
4.114 ± 0.077
3.19
2004–2008 A
0.551 ± 0.024
2.850 ± 0.120
5.17
1.375 ± 0.049
3.852 ± 0.093
2.30
2009–2013 W
0.402 ± 0.027
2.066 ± 0.136
5.14
0.859 ± 0.060
2.828 ± 0.155
3.29
2009–2013 Sp
0.354 ± 0.018
2.243 ± 0.117
6.34
0.699 ± 0.039
2.774 ± 0.128
3.97
2009–2013 Su
0.357 ± 0.017
2.818 ± 0.112
7.89
0.723 ± 0.024
2.870 ± 0.095
3.97
2009–2013 A
0.437 ± 0.024
2.579 ± 0.105
6.31
0.926 ± 0.042
2.934 ± 0.110
3.05
a Uncertainties are 1 standard error of the means.
OC / EC is computed as ratios of means. Propagation of errors yields 1
standard error of OC / EC ranging from 0.30 to 0.49 for CTR (mean 0.41)
and 0.10 to 0.29 for JST (mean 0.16).
b W is December, January, February; Sp is March, April, May; Su is June, July, August; A is September, October, November.
Mean OC / EC ratios are higher at CTR than at JST, again consistent with
regional-scale aging of ambient aerosol and a relatively greater influence
of SOA at CTR. The period mean OC / EC ratios at JST range from 2.3 : 1 to
4.0 : 1, suggesting variable contributions from multiple sources. For
comparison, typical OC / EC ratios are ∼ 1 in freshly emitted
motor vehicle emissions (Chow et al., 2004), with important differences
among vehicle types (McDonald et al., 2015), ∼ 5 : 1–20 : 1 in
near-source biomass burning plumes (Andreae et al., 1996; Andreae and
Merlet, 2001; Hobbs et al., 1996; Lee et al., 2005), and potentially much
greater than unity as oxidation and SOA formation proceed (Robinson et al.,
2007).
Temporal trends in ambient EC and OC correlated within individual sites and
across the SEARCH domain (e.g., CTR and JST, Fig. 1), indicating regional
coherence of trends and seasonal variations for both EC and OC. The strong
correlation of EC and OC at all SEARCH sites, averaging times (annual,
seasonal, monthly, daily), and seasons (Table S3, Fig. S5) indicates that
combustion processes are a major source of OC. However, significant
correlations of SO4 with both EC and OC during summer suggest the
influence of SO4 on SOA formation in summer, consistent with results
from SOAS (Xu et al., 2015a, b; Budisulistiorini et al., 2015). OC
correlates with both EC and SO4, but for different reasons.
Consequently, EC and SO4 also correlate, but not as strongly and not as
consistently across timescales. In summary, the EC and OC measurements
indicate influence of multiple emission sources or atmospheric processes
affecting all SEARCH sites, though differently at urban and rural locations.
OM / OC ratios
More oxygenated OA has higher ratios of OM / OC, so OM / OC potentially serves
as an indicator of atmospheric aging (Turpin and Lim, 2001). A low value
(e.g., OM / OC ∼ 1.4 to 1.6) suggests little aging (i.e., POA is
a large fraction of OA), whereas a high value (e.g., > 2)
suggests more aging (SOA is a large fraction of OA). For comparison, OM / OC
ratios are 1.2 for pentane (and higher molecular weight alkanes), 1.1 for
isoprene, and 2.0 for isoprene epoxydiol (IEPOX) (a gas-phase intermediate
of isoprene oxidation, yielding SOA). The average motor vehicle OM / OC ratio
is ∼ 1.2 to 1.4 (Landis et al., 2007) while biomass burning
OM / OC averages ∼ 1.4 to 1.8 (Reid et al., 2005).
We estimate the OM / OC ratio for the urban and rural SEARCH sites using
a mass balance computation based on particle composition. The sum of species
concentrations, including estimated particle-bound water (PBW) at laboratory
temperature and relative humidity (RH), is
Sumofspecies=f1×SO4+f2×NO3+f3×NH4+EC+OC+MMO+Na+Cl
(inorganic species concentrations are from ion measurements). PBW at
laboratory RH of < 38 % is represented by the coefficients
f1 (1.28), f2 (1.15), and f3 (1.25) (Tombach, 2004, as derived
from Tang, 1996). The coefficient f1 is
an average of the coefficients for NH4HSO4 (1.27) and
(NH4)2SO4 (1.29), f2 is the coefficient for
NH4NO3, and f3 is a weighted average reflecting higher
SO4 than NO3 concentrations. MMO is the sum of the concentrations
of six crustal elements (Al, Ca, Fe, K, Si, Ti) (X-ray fluorescence (XRF)
spectroscopy measurements), expressed as oxides (Hansen et al., 2003). This
estimate of crustal mass is likely conservative, since it does not include Mg
or Mn and the assumed Ca mass (as CaO) would be less than the mass of
CaCO3 (if present). The carbon components, metals, and chloride are not
adjusted for retained water at laboratory temperature and humidity. This
creates a potential for uncertainty in the calculation, especially in the
case of OC. Atmospheric OC is known to be hygroscopic at elevated humidity,
but experimental data suggest that water retention is minimal at
< 38 % RH for laboratory filter analysis (e.g., Malm et al.,
2005; Taylor et al., 2011). Measurements made during SOAS indicate that
organic-associated water was less than ∼ 25 % of total particle
water in mid-day ambient samples when ambient RH was less than
∼ 50 % (Guo et al., 2015). We estimate an OC PBW uncertainty in
Eq. (1) by assuming that OC PBW is 10 % of OC (fOC= 1.1),
which would increase the calculated sum of species by 3 % and decrease
the OM / OC (calculated below) by 0.1 units on average.
The difference between PM2.5 mass and the sum of species concentrations
is denoted as non-measured (NM) mass:
NMmass=PM2.5mass-Sumofspecies.
An upper bound for OM is calculated as OM* = OC + NM mass, which assumes
that all NM mass is associated with OA. Any mass that is missing from the
computed sum of species would bias NM mass high, thereby also causing OM* to
be higher than the true OM. Similarly, underestimation of PBW would bias NM
mass and OM* high. We estimate the combined effect of missing species and
PBW to result in possible overestimation of OM* / OC by up to 0.2 units on
average. An opposing bias potentially arises in Eq. (2), because the FRM
sampler that is used by SEARCH to provide the PM2.5 mass measurement is
known to lose volatile species (e.g., inorganic particle NO3). We
recalculated Eq. (1) by replacing the measured NO3 and Cl
concentrations (which are the sum of a Teflon front filter and a nylon
backup filter located in the SEARCH PCM sampler) with the Teflon filter
concentrations. The effect was to reduce the calculated sum of species,
which then increased the calculated OM* / OC by 0.2 units on average.
Therefore, we estimate the uncertainty in the calculated OM* / OC ratios as
±0.2 units. If no PBW is associated with inorganic species (SO4,
NO3, NH4), Eq. (1) would underestimate OM* / OC by 0.5 units on
average. However, inorganic PBW is expected even at RH < 38 %, so
this potential bias appears less plausible than the documented bias in FRM
PM2.5 mass concentrations.
At all SEARCH sites, NM mass concentrations averaged 1.5 to
1.9 µg m-3 (interquartile range ∼ 0.5 to
∼ 2.5 µg m-3 at all except (rural) Yorkville, Georgia (YRK)) during the most recent
5-year period (2009 to 2013; Na and Cl ions were not measured prior to
2008) (Fig. S6). Daily NM mass correlated with daily OC to varying degrees:
r2 was 0.2–0.3 at BHM, Gulfport, Mississippi
(GFP), (rural) Oak Grove, Mississippi (OAK), and 0.4–0.5 at CTR, JST,
YRK, (suburban) Outlying Landing Field,
Pensacola, Florida (OLF), and Pensacola, Florida (PNS). The average
OM* / OC varied by site from 1.5 (BHM) to 2.0 (YRK) (Fig. 2) (ΔOM* /ΔOC regression slopes of 1.6 to 1.9 without intercept terms,
Fig. S7), which suggests a regionally characteristic but spatially and
temporally variable mix of less-oxidized and more-oxidized OA. The
consistency of the mean values in the range of 1.5 to 2.0 (±0.2)
indicates that relatively fresh emissions contribute a major portion of OA at
both urban and rural sites with variations in the degree of oxidation or SOA
mass. However, higher OM* / OC and OM* / EC occur in the warmest
months (Fig. 2), consistent with seasonal SOA formation and the seasonal
variations discussed above. Our mean OM* / OC is lower than reported in
SOAS research – for example, mean CTR OM / OC of 2.16 from aerosol mass
spectroscopy (AMS) measurements (Xu et al., 2015a). For identical sampling
periods, our spring 2012 mean OM* / OC was 1.34 at JST and 1.80 at YRK,
which is lower than mean OM / OC of 1.93 at JST and 1.98 at YRK reported
by Xu et al. (2015b). Our winter 2012–2013 mean OM* / OC was 1.51 at JST
and 1.56 at YRK, which is higher than mean OM / OC of 1.40 at JST and
1.31 at YRK reported by Xu et al. (2015b). Comparisons are discussed further
in Sect. 3.5.3.
Statistical distributions of the ratio OM* / OC computed for
daily-average measurements at SEARCH sites, 2009–2013. The distributions
show the 10th, 25th, 50th, 75th, and 90th
percentiles. OM* is the sum of measured OC and the computed difference of
PM2.5 mass minus the sum of measured species concentrations.
Biomass burning
Emission inventories indicate that biomass burning, including prescribed
burns, wildfires, agricultural burns, and domestic heating, is the largest
source of PM2.5 OC emissions in the southeast on an annual-average basis
(Hidy et al., 2014). Prescribed burns are the largest source of biomass
burning OC emissions, again on an annual basis (Hidy et al., 2014). In the
southeast, prescribed burns are employed to manage roughly
4 million hectares (ha) (∼ 10 million acres) of land every year,
primarily between January and April; wildfires may occur year-round but are
more frequent in warmer months (Wade et al., 2000; Haines et al., 2001).
Nearby (e.g., ∼ 10 km) biomass burning plumes are sometimes evident in
CTR hourly data and substantially affect observed concentrations of EC, OA,
CO, NOy, NH3, and O3 (Fig. S8). However, the cumulative effect
of widespread and potentially wide-ranging biomass burning on long-term
ambient OA concentrations is more difficult to determine. The available data
record does not include organic biomass burning tracers, such as
levoglucosan, except during special studies such as the 6-week SOAS
campaign. Alternatively, non-soil potassium (K) has been used as an indicator
of biomass combustion in previous studies (Calloway et al., 1989; Lewis et
al., 1988; Lewis, 1996; Pachon et al., 2010, 2013) and can be determined from
K measurements reported in the long-term SEARCH data. Using a single tracer
species to identify and quantify biomass burning contributions to ambient OA
is subject to important uncertainties, and potassium is an imperfect tracer
of biomass burning. Zhang et al. (2010), for example, showed that
water-soluble K and levoglucosan correlate in winter (when more biomass
burning occurs in the southeastern US) but not in summer. However,
levoglucosan and its associated AMS markers may persist in the atmosphere for
less than a day (May et al., 2012; Bougiatioti et al., 2014). Instability of
organic marker species could lead to differences in AMS biomass burning OA
compared with estimates made using K as a tracer.
Non-soil K (nsK) is estimated from coarse PM (PMcoarse or PMcrs,
PM between 2.5 and 10 µm) and PM2.5 XRF measurements of K and Si
following the K tracer approach of Pachon et al. (2013). Briefly, the method
regresses measured K against species X concentrations: K =α+β×X (where X derives primarily from crustal material). Si
measurements are used to represent the crustal species, X, because Si
concentrations are routinely well above the limits of detection and the
correlations of Si with Al and other known crustal elements indicate few or
no interfering sources of Si. The correlations of PMcoarse XRF K and Si
are very strong, with consistent values of the slope β=ΔK /ΔSi of 0.10 to 0.13 and r2∼ 0.8 at all sites
(Figs. S9–S11). These slopes therefore define the expected ratio of
K / Si in crustal material in the region. The ratios are lower than, but
consistent with, a value of 0.15 ± 0.01 reported for data from
Phoenix, AZ (Lewis et al., 2003). In contrast to PMcrs, PM2.5
measurements exhibit large excesses of K over the expected K / Si ratios,
indicating the presence of one or more non-crustal sources of PM2.5 K
(Figs. S9 to S11). For each plot, fine-particle K vs. Si forms one
branch that falls on the line defined by coarse-particle K vs. Si,
indicating similar relationships between K and Si within fine and coarse
fractions of crustal PM. High fine-particle K concentrations also occur at
lower-than-average fine Si concentrations. We apply the slopes β to
compute nsK = K -β*Si from PM2.5 data (Fig. S9). The
agreement between computed nsK and measured water-soluble K (K ion, KI;
measured beginning 2008) supports the interpretation of non-soil K as an
indicator of biomass burning K (Kbb) (which is water soluble) at rural
inland sites such as CTR (Fig. S9). Although computed Kbb exceeds measured
K ion concentrations by ∼ 0.02–0.03 µg m-3 on
average, the difference may relate to the differing resolutions and
sensitivities of the XRF and ion measurements (Fig. S9). Later analyses
(Sect. 3.5.2) link CTR Kbb with species (including CO and EC) deriving
from combustion. Possibly, both K ion and computed nsK could also have a
marine origin at some coastal sites (e.g., OLF) or an industrial process
origin at some urban sites (e.g., BHM). Detailed review of computed nsK
indicated that all nsK > 0.4 µg m-3 occurred on or
1
day after the 4 July and 1 January US holidays, and only at urban
sites (Fig. S11). This result appears to indicate fireworks as a source of
nsK on such occasions. Other than samples from 1 and 2 January and from
4 and 5 July, we identify nsK as biomass burning K (Kbb), recognizing some
uncertainty in this identification for BHM and coastal sites. Since nsK can
be computed from the XRF measurements of K and Si for the full SEARCH record
(1999 to 2013), whereas K ion measurements commenced in 2008, we use nsK as
our biomass burning tracer. After exclusion of obvious high-K events
(holiday fireworks), our identification of nsK as an indicator of biomass
burning (Kbb) could introduce a bias toward overestimation in the
calculation of OCbb, discussed below, if other sources of water-soluble K
are important.
The ratio of TC to Kbb (TCbb / Kbb) in biomass burning is known to vary
widely among fire types (e.g., wildfires differ from prescribed burns) and
among fire stages (e.g., temperature, or flaming vs. smoldering). The
variability of emissions among and within fires implies that biomass burning
tracers are more useful for estimating average impacts than for quantifying
burn contributions during individual events. We use a single average scaling
factor based on consideration of emissions information (Hidy et al., 2014),
which we check using the correlation of modern C with non-soil K (Fig. S12).
Inventory annual-average TCbb / Kbb for fires is in the range
28 : 1–36 : 1 (Blanchard et al., 2013). For fires (prescribed burns,
wildfire, agricultural field burns) plus area sources (largely open burning
from agricultural, construction, and yard waste), the 2013 ratio of
TCbb / Kbb is ∼ 23 : 1 (Table S2). The ratio varies among
emission source profiles from lower values of 5 : 1 and 7 : 1 (solid
waste combustion and agricultural burning, respectively) to an intermediate
value of 19 : 1 (wildfire) and higher values of 43.5 : 1 (slash burning)
and 61 : 1 (residential wood combustion) (EPA,
http://www.epa.gov/ttn/chief/software/speciate/; Reff et al., 2009). Our assumed fixed scaling
factor of 32 for TCbb / Kbb is similar to carbon-isotope data from CTR
winter samples (Fig. S12, CTR regression slope ΔTCmodern/ΔKbb = 43), when prescribed burns are more
common and SOA formation rates are lower. The higher slope of ΔTCmodern/ΔKbb = 71 : 1 at JST could reflect a
different type of biomass burning (e.g., residential wood combustion), while
the lesser correlation of modern TC with non-soil K (assumed to represent
Kbb) at BHM and higher slope at PNS potentially reflect source variability or
the confounding influence of industrial (BHM) or marine (PNS) sources of
non-soil K. The mean ratio of wood-burning OC concentrations determined by
Kleindienst et al. (2010) using organic tracers (20 samples, collected in May
and August 2005, paired by site and date) to Kbb concentrations is 20 : 1
(varying by site from 16 : 1 at BHM and CTR to 18 : 1 at PNS and 29 : 1
at JST). This result agrees with the inventory TCbb / Kbb of
∼ 23 : 1 averaged across fires and area sources (Table S2) assuming
that the corresponding ratio of wood-burning TC / Kbb is ∼ 10 %
higher than wood-burning OC / Kbb. Since prescribed burns and residential wood
combustion (higher TCbb / Kbb) generally occur during winter months,
whereas wildfires, agricultural field burns, and waste burning (lower
TCbb / Kbb) may occur during warmer months, our assumed fixed scaling
factor of 32 : 1 for TCbb / Kbb likely fails to capture some of the
seasonal variability in TCbb. A higher scaling ratio (e.g., ΔTCbb /ΔKbb = 32 rather than 20) would yield higher computed
TCbb and therefore higher OCbb. Based on ΔOC /ΔEC in actual
biomass burning events observed at SEARCH sites (Fig. S8), we compute
OCbb = 0.9 × TCbb (the ratio
OCbb / TCbb could be higher in some burn events). Considering the range
among SEARCH sites of winter ΔTCmodern/ΔKbb
(22 : 1 to 82 : 1, Fig. S12), the variability of TCbb / Kbb among
source types, and the possibility that Kbb could be overestimated if there
are sources other than biomass burning that contribute to nsK, we estimate
the uncertainty range for OCbb as -50 % to +100 % (factor of 2)
subject to the constraint that OCbb < OC.
CTR monthly average concentrations indicate a downward trend in OC but not
in computed OCbb, so that OCbb has become a larger fraction of OC at CTR
since 2007 (Fig. 3). The absence of trend in computed OCbb reflects the
absence of trend in measured K and computed nsK. OCbb tends to be higher in
winter months, when prescribed burns are more common and residential heating
needs are greatest, but OCbb is present during all seasons (Fig. 3) and at
all SEARCH sites (Fig. S13). Retene, a tracer of coniferous wood
combustion, is evident at the sites where it was measured (urban BHM and
JST) with a pronounced seasonal cycle (Fig. S14). This seasonality could
indicate that the summer OCbb has been overestimated, or it could indicate
that retene loss rates are greater during warmer months. Retene emissions
from prescribed burning in the southeast are highly variable and depend
largely on the amount of softwood present in the fuel. Since historical fire
suppression has led to the accumulation of significant amounts of hardwood
in a thick mid-story of pine-dominated forests (e.g. Provencher et al., 2001;
Varner et al., 2005), retene is not considered a unique indicator for
prescribed burning emissions in the southeast.
Monthly average measured OC (solid blue line) and computed biomass
burning OCbb (solid green line with surrounding shaded area indicating
estimated uncertainty) at CTR. Trends in OC (dashed blue line) are
statistically significant (p < 0.05); trends in OCbb (dashed green
line) are not statistically significant.
The analysis of K measurements from the SEARCH data reinforces the
conclusion that biomass burning is an important component of
combustion-related OA in the SEARCH domain, at all sites and in all seasons.
The contribution is especially important for regional-scale OA, as suggested
by the CTR data. Uncertainties in the estimation procedure and scaling
factors imply that our computed CTR mean OCbb (1.6 µg m-3, 1999–2013 average) could be up to twice as high as the true mean OCbb
concentration. If so, actual mean 1999–2013 OCbb would be 0.8 µg m-3, which is higher than AMS mean biomass burning OA (10 %, or
∼ 0.25 µg m-3 OC) at CTR during the 6-week SOAS
period (Xu et al., 2015a, b). Although the majority of brown carbon aerosol
mass during SOAS is attributed to biomass burning rather than to SOA,
biomass burning did not contribute the majority of OA (Washenfelder et al.,
2015). As previously noted, more biomass burning occurs in the southeastern
US during cooler months than during mid-summer (Zhang et al., 2010), so
the SOAS campaign is expected to show less biomass burning than during other
months. Reported AMS mean biomass burning OA concentrations were higher at
JST (∼ 0.5 µg m-3 OC during May and December 2012)
and at YRK (∼ 0.6 µg m-3 OC during December 2012 and
January 2013) (Xu et al., 2015a, b). Due to the loss of organic tracers on a
timescale of about a day or less, the biomass burning OA that is estimated
using AMS is thought to yield an estimate of relatively fresh burning as
compared to aged regional burning levels (Xu et al., 2015a, b). Estimates of
a regional pool of more aged biomass burning OA are not available. If the
reported AMS biomass burning OA concentrations are, e.g., ∼ 50 % lower than the sum of fresh and aged biomass burning OA, the
resulting sum (1 µg m-3 OC) would fall within our OCbb uncertainty
range. The lack of long-term trend in OCbb (Fig. 3) occurs regardless of
scaling uncertainties (assuming constant scaling of OCbb to Kbb), because no
trend exists in either K or Kbb concentrations.
Principal component analysis
Important insight into the origins of ambient aerosol can be obtained with
multivariate statistical methods, such as PCA, which is a well-established
method for PM source apportionment (Dattner and Hopke, 1982). PCA generates
mathematically independent groupings of measurements based on the
correlations among the measured variables (classically, the groups are
geometrically orthogonal to one another). The number of groups reproduces as
large a fraction of the total variance of a data set as possible subject to
optimization criteria, typically explaining ∼ 75 to 80 % of the
variance of, e.g., ∼ 20 to 25 air pollutant species concentrations with
∼ 5 to 10 groups, also known as factors or components. Although PCA
factors may be identifiable with emission sources in some applications,
factors fundamentally represent correlations among species and potentially
reflect a variety of aerometric processes (e.g., secondary species formation,
meteorological effects). In our application, we interpret PCA factors as
associations among species that are indicative of variations in the chemical
environment, and refer to such species associations as components for
brevity. A related methodology, PMF (EPA, 2014), differs in part from PCA in
that PMF constrains factors to positive values. This constraint is physically
realistic if PCA factors are interpreted as unique emission source
contributions. The negative values permitted by PCA are in fact meaningful
and informative if, in addition to emissions, factors represent a larger
suite of physical and chemical processes (e.g., deposition; chemical loss
processes; contrasts between inland versus marine air mass transport) as well
as species origins.
Application
We report two main versions of PCA, with additional versions used for
sensitivity tests and auxiliary information. PCA1 is applied to
measurements made at SEARCH sites from 2008 through 2013. The 23 gas and
PM2.5 measurements comprise daily-average concentrations of PM2.5
EC and OC (thermal–optical reflectance, TOR), daily averages of gases
NH3 (measured continuously or at 24-hour resolution) and continuous
NOx and NOz, secondary species (daily peak 8 h O3, plus
PM2.5 SO4, NH4, and NO3), and PM2.5 crustal
elements (XRF measurements of Al, Si, and Fe), species associated with salts
(PM2.5 Na, Cl, Mg, and Ca ions), and trace metals (PM2.5 Zn, Cu).
Both daily averages and daily 1-hour maxima of gases (CO and SO2) are
included to match the temporal resolution of the other daily data while also
potentially capturing shorter-duration plumes. Water-soluble PM2.5 K (K
ion) is included as a potential indicator of biomass combustion. Because
some species used in PCA1 were not measured throughout the 15-year SEARCH
program, PCA2 is carried out to interpret long-term OC trends
from 1999 through 2013. PCA2 excludes measurements that commenced in 2008
(water-soluble Ca, Mg, K, Na, and Cl). XRF Ca and nsK are used instead of
water-soluble Ca and K, respectively. Without Na and Cl in PCA2, salt is not
detectable, as will be discussed. NH3 is excluded from PCA2, since
those measurements began in 2004. Daily-average O3 is included in PCA2
to complement daily peak 8 h O3.
The sensitivity of our results to the choice of statistical method is
examined by comparing PCA1 and PCA2 and by using additional PCA and PMF
applications. As described in Sect. 3.5.3, the range of results obtained from
PCA1, PCA2, other PCAs, and PMF is used to estimate uncertainty. The
additional PCA applications are carried out by using special data, different
suites of measurements, or different measurement periods. Non-methane organic
compound (NMOC) measurements made every day at JST from 1999 through 2008 are
incorporated to generate PCA3 as a modification of PCA2 (no ions and only XRF
elements, and shorter time period). Alternate versions of PCA2 are carried
out for 2004–2013 CTR data to see if factor loadings are robust and
relatively insensitive to the choice of seasonal indicators (PCA4 and PCA5).
The EPA PMF model (version 5; EPA, 2014) was applied to the same CTR and JST
measurements used in PCA2. PMF requires estimates of measurement uncertainty,
which may be species-specific or even sample-specific. Two sets of
uncertainty estimates were employed: uniform (10 % of species
concentrations) (PMF1), and species-specific (incorporating detection limits
and species uncertainties of 5 to 25 % of measured concentrations)
(PMF2).
For PCA applications, the daily OC concentrations at each site are
apportioned using daily PCA factor scores. The OC apportionment is carried
out by multiple regression of daily OC concentrations against daily factor
scores, retaining those that are statistically significant (p < 0.05). Since the PCA components are orthogonal, the regression coefficients
are more stable than would be the case for multiple regression against
various tracer species, which are typically intercorrelated. The PMF model
generates source contributions internally.
Species associated with each PCA factor (component). Component
names are keyed to the species. Three species are listed in decreasing order
of association for associations of 0.6 or greater (or -0.6 or smaller).
Negative values indicate anti-correlation. COx and SO2x are
1 h daily maximum CO and SO2, respectively. O3x is 8 h
daily maximum O3. PCA1, 2008–2013; PCA2, 1999–2013. N is number
of days.
PCA
Site
N
Combustion
Crustal
Seasonal
SO2
SO4
Metals
Salt
Other
1
BHM
364
CO, NOx, OC
Al, Si
NO3
SO2x, SO2
NH4, SO4
Zn, Cu, Fe
K
NOz
1
CTR
383
EC, OC, CO
Si, Fe, Al
NH3
SO2, SO2x
SO4, NH4
Cu, Zn
Na, Cl, Mg
1
GFP
100
CO, COx, NOx
Si, Fe, Al
O3x, NH3
SO2, SO2x
NH4, SO4
Cl, Na, Mg
Ca
1
JST
516
CO, NOx, EC
Si, Al, Fe
O3x, NH3, -NO3
SO2x, SO2
NH4, SO4
K, Cl, Mg
Na
1
OAK
100
COx, CO
Fe, Al, Si
NH3, O3x
SO2x, SO2
SO4, NH4
NOx, Cu, NOz, Zn
Na, Cl, Mg
1
OLF
327
NOx, CO, EC
Si, Al, Fe
NH3, O3x
SO2, SO2x
SO4, NH4
Cu
Na, Mg, Cl
1
PNS
44
CO, NOx, EC
Si, Al, Fe
O3x
SO2, SO2x
NH4, SO4
Na, Mg, Cl
1
YRK
426
NOx, NO3, CO
Si, Fe, Al
O3x, OC, EC
SO2x, SO2
SO4, NH4
Cu
Na, Cl, Mg
Zn
2
BHM
1513
CO, COx, NOx
Al, Si
O3, O3x, -NO3
SO2x, SO2
NH4, SO4
Zn, Cu, Fe
2
CTR
1258
OC, EC, COx
Si, Fe, Al
O3x, O3
SO2, NOx, SO2x
SO4, NH4
Cu
2
GFP
376
COx, CO, NOx
Si, Fe, Al
O3, O3x
SO2x, SO2
NH4, SO4
Cu, Zn
2
JST
2593
CO, COx, NOx
Si, Al, Fe
NO3, -O3x, -O3
SO2x, SO2
NH4, SO4, O3x
Cu
2
OAK
707
COx, CO, EC
Si, Fe, Al
O3, O3x
SO2x, SO2
SO4, NH4
Cu, Zn
2
OLF
948
COx, CO, NOx
Si, Fe, Al
O3, O3x
SO2x, SO2
NH4, SO4, NOz
Zn, Cu
2
PNS
445
EC, CO, NOx
Si, Al, Fe
O3x, O3, SO4, NH4
SO2x, SO2
Cu
NOz
2
YRK
1435
COx, NOx, NO3
Si, Fe, Al
O3x, SO4, O3, NH4
SO2x, SO2
Cu
Zn
Mean OC concentrations associated with components identified by
PCA1 (2008–2013) and PCA2 (1999–2013). NS is not statistically
significant, n/a is not applicable (component not present in PCA). Units are
µg m-3. Standard errors of the means ranged from 0.003 to 0.09 µg m-3 (up to 0.25 µg m-3 for PNS PCA1).
PCA
Site
N
Combustion
Crustal
Seasonal
SO2
SO4
Metals
Salt
Other
1
BHM
364
1.36
0.09
0.40
0.35
0.45
0.15
0.14
0.10
1
CTR
383
1.28
0.26
0.56
NS
0.33
NS
NS
n/a
1
GFP
100
0.95
NS
0.45
0.15
0.25
NS
-0.41
0.62
1
JST
516
1.09
0.16
0.49
0.07
0.27
n/a
0.64
0.11
1
OAK
100
0.40
NS
0.50
0.37
0.53
0.32
-0.27
n/a
1
OLF
327
0.74
-0.08a
0.52
0.16
0.27
NS
-0.09a
n/a
1
PNS
44
1.95
NS
0.33
NS
0.56
n/a
-0.63
n/a
1
YRK
426
0.14
0.14
1.09
NS
0.26
0.16
0.15
0.47
2
BHM
1513
1.60
0.19
0.47
0.38
0.57
0.48
n/a
n/a
2
CTR
1258
1.50
0.12
0.69
NS
0.66
NS
n/a
n/a
2
GFP
376
0.72
0.14
0.37
0.21
0.50
0.25
n/a
n/a
2
JST
2593
2.58
0.32
0.06b
NS
1.01b
0.13
n/a
n/a
2
OAK
707
1.50
NS
0.47
NS
0.59
NS
n/a
n/a
2
OLF
948
0.81
-0.06a
0.25
0.09
1.02
0.20
n/a
n/a
2
PNS
445
1.55
NS
0.45c
0.17
n/a
NS
n/a
0.36
2
YRK
1435
0.76
0.20
1.40c
0.05
n/a
0.39
n/a
0.29
a OLF PCA1 and PCA2 crustal and PCA1 salt OC mean concentrations are
negative due to inverse associations of OC with crustal and salt components
at OLF (Tables S7 and S12).
b JST PCA2 seasonal OC is associated with NO3; JST PCA2 SO4
component includes OC associated with O3 (Table 2).
c PNS and YRK PCA2 seasonal components include OC associated with SO4
(Table 2).
PCA Components
PCA1 and PCA2 reveal consistent sets of species associations, resulting in 6–8 principal components at each SEARCH site (Table 2). For clarity, we
designate the components as (1) combustion, (2) crustal, (3) seasonal, (4) SO2, (5) SO4, (6) metals, (7) salt, and (8) other. These names are
used as descriptors, rather than as designated emission sources. Component
characteristics are discussed below. The full orthogonal solutions are shown
in the Supplement (Tables S4 to S11). The values in Tables S4 to S11 are the
coefficients of the linear combinations of standardized species
concentrations (daily concentration less mean divided by standard
deviation); each tabled value is also the correlation (r) between a given
species and a particular component. High (∼ 1) or low
(∼ -1) values indicate high correlation or anti-correlation,
respectively; both are meaningful. A value near zero indicates little or no
correlation, so values in the range of ∼ -0.5 to 0.5 represent
associations ranging from moderate anti-correlation (-0.5) to zero
correlation to moderate correlation (0.5).
The OC apportionments indicate statistically significant relationships
between OC and four to seven PCA components (Tables S12 and S13). Mean
contributions of each statistically significant component to daily OC at
each site using both PCA1 and PCA2 are summarized in Table 3; these
contributions are expressed as percentages of total OC in Table S14. PCA1
and PCA2 each indicate that OC is associated with multiple components at all
sites. Except at YRK and OLF (PCA2 only), the overall OC associations are
strongest for the combustion component (Tables S4 to S11).
The PMF source profiles varied depending on the choice of uncertainty
inputs, but yielded average OC apportionments that were qualitatively
comparable to PCA2 (Fig. S15). The PMF crustal OC and SO4-associated
OC concentrations were comparable to PCA (Fig. S16). However, PMF source
profiles combined CO and O3, whereas PCA tended to separate O3
from CO, leading to differences in the apportionment of OC to combustion and
seasonal components (Fig. S16). Differences between PCA and PMF occur in
part because the PCA seasonal component generally comprised contrasts (e.g.,
positive O3, negative inorganic particulate NO3) whereas PMF
forced positive solutions. In these applications, PCA predicted high OC
concentrations more accurately than PMF did (Fig. S17).
Combustion
All sites exhibit a suite of species associated
with combustion processes (EC, OC, CO, Kbb or K ion, NOx or NOz).
The variations in combustion associations among sites suggest different
source mixes, differences in air mass ages (e.g., fresh emissions at urban
sites, more aged emissions at rural sites), or differing transport of
polluted air masses. For example, NOz is more strongly associated than
NOx with the combustion component at the two most rural sites, CTR and
OAK. OC associated with the combustion factor could therefore comprise
material that would be classified as either POA or SOA by other analytical
approaches (e.g., hydrocarbon-like OA (HOA) or more-oxidized OA (MO-OOA) by AMS).
Mean combustion OC ranges from 0.7 to 1.6 µg m-3 for PCA1 (2008–2013) and from 1.5 to 2.6 µg m-3 for PCA2 (1999–2013), except
at YRK (Table 3). Daily PCA1 and PCA2 combustion OC concentrations are
correlated at all sites (Fig. S18). Mean absolute differences between PCA1
and PCA2 computed combustion OC range from 0.1 to 0.7 µg m-3 (not
tabled). However, the mean PCA1 and PCA2 combustion OC concentrations are
averaged over different time periods, so the differences in their averages
are partly due to declining EC, CO, and NOx concentrations (Table 1).
Trends in OC components are discussed in Sect. 3.4.4. Mean PCA2 combustion
OC ranged from 25 to 63 % of mean OC concentrations (Table 3).
Various combustion processes are expected to influence individual SEARCH
sites to different degrees. CTR OCbb correlates significantly
(p < 0.0001) with PCA1 and PCA2 combustion-associated OC (r2=0.54 and 0.58, respectively, Fig. S19a), suggesting that the combustion
component at CTR is primarily associated with biomass burning. Whereas OCbb
is computed from Kbb (Sect. 3.4), PCA1 and PCA2 combustion OC concentrations
are determined from the principal component association of CO, EC, and either
K ion (PCA1) or Kbb (PCA2) (Table S5). The association of K ion and Kbb with
CO and EC at CTR links potassium with a combustion process. At BHM and JST,
multiple regression of combustion OC against NO, gas-phase organic species
(Blanchard et al., 2010), and non-polar OC compounds (including polycyclic
aromatic hydrocarbons, PAHs, and iso/anteisoalkanes, or hopanes and steranes)
(Blanchard et al., 2014a) indicates an association of fresh emissions (NO)
and non-oxidized organic compounds with PCA combustion OC (Fig. S19b). The
urban PCA combustion factor associates CO, EC, and daily-average NOx;
1 h maximum NO and non-oxidized organic compound concentrations were not
used in determining either PCA1 or PCA2 (Tables S4 and S7). Urban PCA
combustion OC is more likely attributable primarily to motor vehicle exhaust
emissions than to biomass burning.
Crustal
A crustal component is present at all sites,
associated with Al, Si, Fe, and, to varying degrees, Ca. At BHM, Fe
associates more prominently with a metals component, consistent with
previous studies indicating the impact of industrial facilities (including
metals fabrication) on PM2.5 at BHM (Baumann et al., 2008; Blanchard et
al., 2014b).
The mean crustal-associated OC concentrations vary from 0.1 to 0.3 µg m-3 at inland sites (Table 3). Coastal sites exhibit non-significant,
minor, or inverse associations of OC with crustal elements (-0.1 to 0.1 µg m-3, Table 3). Inverse associations indicate that OC concentrations
at coastal sites are lower than average when Al, Si, and Fe concentrations
are elevated. PCA1 and PCA2 crustal OC concentrations correlate (Fig. S20)
and crustal OC correlates with Si (Fig. S21). Crustal-associated OC could
derive from region-wide phenomena (e.g., transport of Saharan dust), but may
also stem from ubiquitous and widely distributed activities that suspend
crustal material. Potential sources include soil-derived OC (e.g.,
agricultural activities, construction, or road dust), or biomass burning
that lofts crustal material (e.g., through plowing material into debris
piles). Road dust is known to include OC among its constituents (e.g.,
McDonald et al., 2013). There are two episodes with high crustal OC at CTR
during June 2013. Elevated concentrations of Al, Si, and Fe co-occurred at
all SEARCH sites during 9–13 and 23–28 June 2013, thus
suggesting region-wide events. Back-trajectory calculations indicate
southerly air flow during these times. Trajectories arrived at CTR and JST
after ∼ 24 h overland transport from the Gulf coast,
whereas trajectories arrived at OLF from overwater transport. At other
times, elevated concentrations of crustal elements occur at single sites,
indicating more local events.
Seasonal
A seasonal component is present at all sites, but in two forms: positive
O3 and NH3 (if measured), along with negative inorganic particulate
NO3, at BHM PCA2, CTR, GFP, JST PCA1, OAK, OLF, PNS, and YRK, or with
reverse signs (e.g., relatively weak negative O3) at BHM PCA1 and JST
PCA2. As noted, sign reversals represent a change in coordinate directions
and need not have physical significance; however, the association of OC with
the seasonal component may differ depending on sign (discussed below). We
denote this component seasonal rather than photochemical. While this factor
has photochemical properties, it is comprised of species with seasonality
variations that result from multiple processes: emissions (NH3),
photochemistry (O3), and temperature- and RH-sensitive thermodynamic
equilibrium (inorganic particulate NO3). The seasonal component
evidently represents seasonal variations not otherwise described by the
seasonal variations of the crustal, SO4, and other components. Because
of the strong connection of the seasonal component to O3, seasonal OC is
plausibly related to the less-oxidized oxygenated organic aerosol (LO-OOA)
component reported by Xu et al. (2015a, b). LO-OOA exhibits a strong diurnal
pattern, with night maxima and day minima (Xu et al., 2015a, b). However, the
LO-OOA diurnal variation is opposite to O3 diurnal variations, which
exhibit daytime maxima. Since PCA was applied to daily-resolution data, it is
not possible to directly compare the PCA seasonal OC to time-resolved LO-OOA.
We note that meteorological conditions that result in high peak daily O3
concentrations (with higher seasonal OC concentrations) are also conducive to
nitrate radical formation, which exhibits nighttime maxima and is associated
with LO-OOA (Xu et al., 2015a, b). Further comparisons are provided in
Sect. 3.5.3.
The mean PCA1 seasonal-component OC (OC associated with higher O3,
higher NH3, lower NOx, or lower PM2.5 inorganic NO3)
ranges from 0.4 to 0.6 µg m-3 at all sites (e.g., 23 % of OC at
CTR, 13 % at BHM and JST, 28 % at OLF), except at YRK where the average
is 1.0 µg m-3. The positive association with O3 suggests that
this OC component represents SOA formation from either or both anthropogenic
and biogenic precursors. PCA2 seasonal OC correlates with PCA1 seasonal OC,
except at JST. The JST PCA2 seasonal OC shows an inverse correlation (Fig. S22), indicating that the seasonal component represents higher winter (lower
O3, higher NO3) OC concentrations, possibly pointing to an
influence from domestic wood combustion for heating. The positive
association of OC with O3 is quantified within the JST PCA2 SO4
component. The mean absolute differences between PCA1 and PCA2
seasonal-component OC concentrations range from 0.2 to 0.5 µg m-3.
Sulfate
SO4 and NH4 are always associated and
usually represented by a single component, denoted SO4. However,
SO4 and NH4 are part of the seasonal component for PNS and YRK
PCA2, suggesting that differentiation of the SO4 and seasonal
components is subject to uncertainty. O3 is associated with both
seasonal and SO4 components.
All SEARCH sites show an association of OC with SO4 ranging from 0.3 to
0.6 µg m-3 on average for PCA1 and from 0.5 to 1.0 µg m-3 on average for PCA2 (Table 3), with PCA2 SO4 OC representing
15 to 44 % of the 1999–2013 mean OC concentrations (15 % at BHM; 22–25 % at CTR, GFP, JST, and OAK; 44 % at OLF). Mean PCA1 associations
of OC with SO4 were 14 % of OC at CTR, 15 % at BHM, 18 % at OLF,
10 % at JST, and 11 % at YRK. PCA1 and PCA2 SO4 OC concentrations
are correlated (Fig. S23) with mean absolute differences in PCA1 and PCA2
SO4-associated OC concentrations of 0.2 to 0.5 µg m-3; PCA2
did not separate the seasonal and SO4 components at PNS and YRK (Tables S10 and S11). The mass of OC associated with SO4 averages 20 to 30 %
of the SO4 concentrations (Fig. S24); therefore, that SO4-associated OC
concentrations decline over time along with decreasing SO4
concentrations. The presence and relative importance of SO4-associated
OC is consistent with research indicating the role of SO4 in
transferring isoprene gas-phase reaction products to the condensed phase
(e.g., Surratt et al., 2007; Xu et al., 2015a, b). Seasonal variations,
discussed below, also support biogenic origins of SO4–OC. The
quantitative relationship of our SO4-associated OC factor to SO4
is the same as the relationship between isoprene OA and SO4, which Xu
et al. (2015a, b) reported as 0.42 µg m-3 isoprene OA per 1 µg m-3 SO4. Based on their reported OM / OC for isoprene OA (1.97),
their result is 0.21 µg m-3 isoprene OC per 1 µg m-3
SO4. For CTR (2008–2013, n=383 days), we obtain 0.216 (±0.008, 1 SE) µg m-3 SO4-associated OC per 1 µg m-3
SO4 (PCA1), 0.190 (±0.004, 1 SE) µg m-3
SO4-associated OC per 1 µg m-3 SO4 (PCA2), 0.213
(±0.003, 1 SE) µg m-3 SO4-associated OC per 1 µg m-3 SO4 (PMF1), and 0.211 (±0.001, 1 SE) µg m-3
SO4-associated OC per 1 µg m-3 SO4 (PMF2).
SO2
The SO2 component, present at all sites,
identifies influences of relatively fresh plumes, whether from electric
generating units (EGUs), industrial, or other SO2 sources. At CTR,
NOx is more strongly associated with the SO2 component than with
the combustion component, consistent with relatively less aged plumes and
more aged general combustion influence. Differences between urban and rural
sites are evident; for example, OC at CTR and YRK is not significantly
related to the SO2 factors, but OC is related to the SO2 factors
at urban sites. This difference indicates the influence of SO2 emission
sources within urban areas, consistent with visual observations and
measurements made near emission sources in Birmingham (Blanchard et al.,
2014b).
OC associated with SO2, indicative of fresh emissions, accounted for
0.07 to 0.37 µg m-3 on average (12 % of OC at BHM, 2 % at JST,
7 % at GFP, 20 % at OAK, and 10 % at OLF, none at other sites using
PCA1) (Tables 3 and S14). PCA1 and PCA2 SO2 OC concentrations are
correlated (Fig. S25).
Salt
A salt (or salt-like) component (of marine or other
origins) is evidenced by Na, Cl, and Mg in PCA1. Na appears as a separate
other component for JST PCA1, suggesting multiple urban sources of one
or more of these species, while JST PCA1 salt is defined by K, Cl, and
Mg. These species are not necessarily unique marine tracers; for example, various
combustion processes generate Cl emissions.
Coastal sites show an inverse association of OC with Na and Cl (sea salt)
(Table S14) and a negative mean OC contribution from salt (Table 3). We
interpret this result as evidence that OC concentrations are lower at
coastal sites when marine salt species concentrations are higher (i.e.,
anti-correlated), indicating that marine air masses are not an important
source of OC. In contrast, mean salt-associated OC ranges from 0.14 to 0.15 µg m-3 (BHM and YRK) to 0.64 µg m-3 (JST). The species
associations for the BHM and JST salt components suggest urban influences
precluding identification of the salt component with marine air masses.
Because K is associated with the JST salt component (and not with the
JST combustion component) and Na is associated with the JST other
component, it is possible that JST salt OC represents biomass combustion,
while the JST combustion component primarily represents motor vehicle
exhaust.
Metals
Cu and Zn appear on a metals component at six sites –
BHM, CTR, GFP (PCA2), JST (PCA2), OAK, and OLF (PCA2); otherwise, Cu and
Zn are associated with combustion or are split between the metals and
other components. The Cu and Zn correlations range from r=0.1 to 0.3
in the full 1999 to 2013 data set, which does not suggest a simple or strong
association between these two species. At JST, Cu correlates with Pb.
Other
A component designated as other is present for BHM
PCA1, GFP PCA1, JST PCA1, PNS PCA2, and YRK PCA1 and PCA2, indicating
variability at urban and near-urban (YRK) sites not otherwise represented by
the major components (Table 2).
Intercomparisons and uncertainty
For PCA3 (Table S15), the sum of alkanes, sum of aromatics, and α-pinene are associated with the combustion component, whereas isoprene is
associated predominantly with the SO4 component. The measured alkane
and aromatic species are known constituents of motor vehicle exhaust
(Blanchard et al., 2010), consistent with a mobile source contribution to
the JST combustion component. Correlations between α-pinene and CO,
EC, and NOx range from r=0.5 to 0.6, mathematically associating
these species, but the physical processes underlying the correlation are
ambiguous (e.g., seasonal or meteorological versus common source emissions).
Isoprene and pinenes can be factors in O3 formation, and the
association of isoprene with SO4 could arise from a common seasonality
or from atmospheric chemical processes generating SOA from isoprene (Surratt
et al., 2007; Xu et al., 2015a, b). Additional work is needed to more fully
interpret VOC species associations.
PCA4 and PCA5 yield consistent results when NH3 or daily-average
O3 are either included or excluded from the analysis (Table S16).
The ranges of mean OC concentrations associated with each PCA component as
obtained from the various applications are listed for CTR, JST, and YRK in
Table 4. Uncertainties in the mean OC concentrations associated with each
PCA component are estimated as one-half the ranges for CTR and JST
(comprising both PCA and PMF applications) and the full ranges for YRK (PCA
applications only), which generally yield comparable uncertainties.
Ranges of mean OC concentrations associated with each PCA
component. The time period is 2009–2013. For each site, multiple methods
were compared using a common set of days. For CTR (six methods), both the
standard deviation and one-half the range of component mean concentrations
are shown. For JST (three methods), one-half the range of component mean
concentrations is shown. For YRK (two PCA methods), ranges are shown. YRK
ranges are smaller than ranges for CTR and JST because no PMF analyses were
carried out for YRK. The larger ranges for CTR and JST compared with YRK
reflect the larger differences between PCA and PMF.
CTRa
JSTb
YRKc
Range/2
Range/2
SD
SD
Range/2
Range/2
Range
Range
Component
(µg m-3)
(% of mean)
(µg m-3)
(% of mean)
(µg m-3)
(% of mean)
(µg m-3)
(% of mean)
Combustion
0.44
18
0.30
12
0.34
12
0.34
14
Crustal
0.09
4
0.07
3
0.11
4
0.09
4
Sulfate
0.15
6
0.11
4
0.19
7
0.26
11
Seasonal
0.36
15
0.25
10
0.43
15
0.12
5
SO2
0.03
1
0.03
1
Metals
0.07
2
0.04
2
Salt
0.33
12
0.16
7
Other
0.05
2
0.16
7
a Mean OC = 2.43 µg m-3, n=383 days, number of methods = 6
(4 PCA, 2 PMF).
b Mean OC = 2.85 µg m-3, n=398 days, number of methods = 3
(2 PCA, 1 PMF).
c Mean OC = 2.40 µg m-3, n=426 days, number of methods = 2
(2 PCA).
A summary of our PCA1 results compared to the 2012–2013 source
apportionments reported by Xu et al. (2015a, b) is shown in Tables S17
through S19. For these comparisons, we determined the PCA1 means by matching
days to each of the Xu et al. (2015a, b) multi-week study periods. The PCA1
combustion OC tends to compare in magnitude to AMS HOA, biomass burning OA (BBOA), cooking OA (COA) (when one or more such
factors are found) or to MO-OOA. The last correspondence
would be expected to the extent that MO-OOA includes oxidized motor vehicle
exhaust, other anthropogenic combustion emissions, or biomass burning (Xu et
al., 2015a, b). As previously noted for CTR, PCA SO4-associated OC
concentrations and AMS isoprene OA concentrations exhibit nearly identical
regression relationships with SO4 concentrations. Some differences
between mean PCA SO4-associated OC and mean AMS isoprene OA (converted
to OC) percentage apportionments are evident in Tables S17 through S19,
however. Such differences appear to result from ambiguities in linking PCA
elements with AMS designations, different numbers of factors (affecting the
percentages), and differences in mean observed OA (OC) concentrations. The
SEARCH and AMS mean OC concentrations are comparable for the CTR (SOAS) and
YRK (winter) data. For JST (summer), JST (winter), and YRK (summer), the
mean AMS OC concentrations exceed the mean SEARCH OC concentrations by
40, 49, and 85 %, respectively. The reasons for these differences
are unknown, but operationally could be related to sampling and analytical
methods. The SEARCH mean OC concentrations during the multi-week comparison
periods are consistent with longer-term averages from 2012, 2013, and 2008–2013 (Tables S17–S19). Since SEARCH reports PM2.5 size fractions
and AMS is based on PM1 size fractions, higher AMS PM mass
concentrations are not expected. No AMS component appears to correspond to
the PCA crustal OC, which could relate to the difference in size fractions
sampled. The PCA crustal OC concentrations are generally small except during
occasional events, as previously noted.
Comparisons of our results to results reported by Kleindienst et al. (2010)
are shown in Fig. S26. Kleindienst et al. (2010) determined organic tracer
concentrations on archived samples from the SEARCH Carbonaceous Aerosol
Characterization Experiment (CACHE) archive. Twenty samples were analyzed,
five each from BHM, CTR, JST, and PNS, collected during May (7, 13, 22, 28)
and August (17) 2005 (Kleindienst et al., 2010). SEARCH OC measurements are
made on filters from a sampler with a denuder placed upstream to remove
organic gases, whereas the CACHE sampler was not denuded. The CACHE OC
concentrations were ∼ 50–100 % higher than the SEARCH OC
concentrations (Fig. S26), with a CACHE average OC concentration of 7.32 µg m-3 compared with SEARCH OC average of 4.62 µg m-3
when restricted to the 17 samples that had both CACHE and SEARCH OC
concentrations. Kleindienst et al. (2010) accounted for ∼ 70 % of the measured CACHE OC concentration using 11 source types, so
their apportioned OC concentrations are roughly comparable to measured
SEARCH OC concentrations. The sum of the Kleindienst et al. (2010) diesel
and wood-burning OC concentrations correlates highly (r2=0.83) and
agrees in magnitude (intercept = 0; slope = 1) with PCA2 combustion OC
(Fig. S26). Wood-burning OC concentrations also correlate highly (r2 = 0.76) with PCA2 combustion OC,
corresponding to ∼ 50–70 % of the combustion OC concentrations (Fig. S26). OCbb (Sect. 3.4)
correlates less well (r2=0.38) with wood-burning OC, and OCbb
concentrations average ∼ 30 % higher than wood-burning OC
concentrations (Fig. S26). As previously noted, the fixed scaling factor
used for estimating OCbb from Kbb does not reflect the variability among
source types in OCbb / Kbb, nor does it reflect seasonal variability in their
source contributions (Sect. 3.4). The modest correlation of OCbb with
organic tracer-based wood-burning OC concentrations is expected due to
emission source variability; overprediction of wood-burning OC by OCbb is
expected during summer months when the principal biomass burning sources
(agricultural field burns, open burning of wastes) likely predominate and
have lower OCbb / Kbb. The sum of the Kleindienst et al. (2010) diesel and
wood-burning OC concentrations correlates highly (r2=0.92) with
OCbb and CO in a multiple regression model (Fig. S26), supporting
combustion origins of OCbb.
Temporal variations
Temporal variations of the 1999–2013 PCA2 results are described here
primarily for CTR and JST, representing (as in Table 1) one rural and one
urban location having extensive SEARCH data records. At JST, day-of-week
variations are evident for the combustion-derived OC and for the OC
associated with crustal species (Fig. S27), consistent with the occurrence
of weekly activity cycles for driving, construction, and other anthropogenic
emission sources. Day-of-week variations are not apparent for other OC
associations at JST or for any OC factors at CTR. Seasonal and
SO4-associated OC exhibit pronounced monthly variations at both CTR
and JST, with higher values of SO4-associated OC and of CTR seasonal
OC occurring during warmer months (Figs. S28 and S29). The patterns for
CTR SO4-associated OC (highest in July and August) and seasonal OC
(higher in spring and autumn than in July) are not independent.
Trends in source contributions to OC at CTR and JST determined
from PCA2 for 1999–2013.
Mean annual combustion-derived OC concentrations decline from 3.8 ± 0.2 to 1.4 ± 0.1 µg m-3 between 1999 and 2013 at JST
(Figs. 4, S30) and from 2.9 ± 0.4 to 0.7 ± 0.1 µg m-3 between 2001 and 2013 at BHM (not shown). Declining combustion OC
concentrations at the urban JST and BHM sites coincide with reductions of
motor vehicle emissions during this time period (Sect. 3.1), though these
urban sites may also be affected by industrial emissions. BHM additionally
benefits from a decline in OC associated with SO2 from 0.4 ± 0.04 µg m-3 in 2001 to 0.2 ± 0.03 µg m-3 in 2013,
probably as a reflection of declining emissions from industrial sources
within Birmingham. In contrast, combustion-derived OC at CTR does not
exhibit a statistically significant decline, equaling 1.5 ± 0.1 µg m-3 in 1999 and 1.3 ± 0.1 µg m-3 in 2013 (Figs. 4,
S30). At CTR, downward OC trends are evident only for SO4 and seasonal
OC (mean decreases of 0.6 and 0.7 µg m-3,
respectively) (Fig. S31). The OC associated with SO4 at CTR exhibits
declines during all seasons, with the weakest such change in winter (Fig. S32).
The trend results are consistent with the combined effects of (1) regional-scale reductions of ambient SO4 and O3 concentrations,
(2) reductions of urban OC due to declining mobile source OC and VOC
emissions, and (3) likely predominance of biomass burning OC at CTR (Hidy et
al., 2014). Carbon-isotope measurements from 2004 show that fossil carbon
represented ∼ 20 % of CTR TC that year (Blanchard et al.,
2011), indicating that mobile source or other fossil fuel emissions affect
CTR to some extent. Enhanced hourly concentrations of EC, OC, and CO at CTR
are associated with winds from the directions of Birmingham, Tuscaloosa, and
Montgomery (Hidy et al., 2014). EC declined by ∼ 0.3 µg m-3 at CTR between 1999 and 2013 (Fig. 1), suggesting an influence of
mobile source emission reductions that is possibly too modest to detect
using our PCA methods or is masked by annual variability in biomass burning
emissions. For comparison, mean EC concentrations at JST decrease by
∼ 1.4 µg m-3 (Fig. 1), and the overall mean EC at
JST (1.35 µg m-3) is ∼ 4 times the overall mean EC
at CTR (0.35 µg m-3).
The trends in mean annual OC from each identified species association
indicate that anthropogenic emission reductions decreased mean annual urban
combustion OC concentrations by 2.4 µg m-3 at JST and at BHM (and,
by inference, other metropolitan areas in the southeast), and indirectly
decreased SO4 and seasonal OC by ∼ 1.1 to 1.3 µg m-3 throughout the southeastern US between 1999 and 2013 (Fig. 4).
As of 2013, the overall mean annual combustion-derived OC is 1.3 to 1.4 µg m-3 at CTR and JST, whereas the sum of the mean annual SO4 and
seasonal-component OC is 0.4 to 0.8 µg m-3 at CTR and JST (Fig. 4).
Mean OC concentrations determined for the period 2008–2013 using
four analytical approaches: (1) multivariate regression (POC and
SOC; Blanchard et al., 2008), (2) calculation of OCbb from Kbb tracer,
(3) PCA and PMF analysis, and (4) CMB receptor modeling (Blanchard et al.,
2013, updated). Row indentations indicate subcategories. Units are µg m-3 unless specified as %.
Component
BHM
CTR
GFP
JST
OAK
OLF
PNS
YRK
Unca
OC (mean measured)
2.91
2.41
1.91
2.86
1.84
1.81
2.06
2.33
0.05
POCb
1.85
1.78
1.40
2.12
1.35
1.36
1.57
1.61
25 %
OCbb
1.58
1.60
1.64
1.40
1.63
1.62
1.77
1.37
2X
PCA1 combustion
1.36
1.28
0.95
1.07
0.40
0.85
1.95
0.13
0.3–0.6
PCA2 combustion
1.09
1.47
0.45
1.83
0.87
0.60
1.26
0.49
0.3–0.6
PMF combustion
NA
1.03
NA
1.22
NA
NA
NA
NA
0.3–0.6
CMB combustion total
2.54
1.52
1.33
2.16
1.42
1.47
1.89
1.49
0.87
CMB area sources
2.01
1.44
1.15
1.50
1.35
1.34
1.68
1.35
20–3 %
CMB mobile diesel
0.20
0.02
0.05
0.27
0.01
0.05
0.04
0.04
13–31 %
CMB mobile gas
0.29
0.03
0.10
0.34
0.03
0.05
0.15
0.06
17–41 %
CMB point sources
0.05
0.02
0.02
0.05
0.02
0.03
0.03
0.04
5–6 %
PCA1 crustal
0.09
0.26
0.00
0.15
0.00
-0.14
0.00
0.14
0.09–0.11
PCA2 crustal
0.20
0.12
0.17
0.35
0.00
-0.06
0.00
0.22
0.09–0.11
PMF crustal
NA
0.09
NA
0.17
NA
NA
NA
NA
0.09–0.11
CMB dust
0.09
0.02
0.04
0.03
0.03
0.02
0.04
0.01
9–22 %
SOCb
1.10
0.66
0.56
0.75
0.48
0.48
0.50
0.77
25 %
PCA1 seasonal + sulfate
0.85
0.90
0.70
0.76
1.03
1.00
0.90
1.26
0.3–0.5
PCA1 seasonal
0.39
0.57
0.45
0.49
0.50
0.71
0.33
1.00
0.1–0.4
PCA1 sulfate
0.45
0.33
0.25
0.27
0.53
0.29
0.56
0.26
0.2–0.3
PCA2 seasonal + sulfate
0.92
0.95
0.81
0.76
0.99
0.93
0.41
0.86
0.3–0.5
PCA2 seasonal
0.51
0.53
0.40
0.05
0.40
0.28
0.41
0.86
0.1–0.4
PCA2 sulfate
0.42
0.42
0.41
0.71
0.58
0.65
0.00
0.00
0.2–0.3
PMF seasonal + sulfate
NA
0.77
NA
1.32
NA
NA
NA
NA
0.3–0.5
PMF seasonal
NA
0.49
NA
0.86
NA
NA
NA
NA
0.1–0.4
PMF sulfate
NA
0.28
NA
0.46
NA
NA
NA
NA
0.2–0.3
N days (2008–2013,
366–1313
383–606
100–280
443–787
100–206
327–598
44–162
426–585
varies by analysis)
a Uncertainty for mean measured OC is 1 standard error of the mean.
Uncertainties for PCA and PMF are taken from Sect. 3.5.3. Uncertainty for
CMB combustion total is RMSE across sites and years, where error is defined
as the difference between predicted and observed concentrations. Uncertainty
for CMB components is based on uncertainties in inputs and across
alternative versions of the model expressed as 1-sigma % of prediction
(Blanchard et al., 2013).
b POC is the sum of OC associated with EC, CO, and Kbb. SOC is the sum
of OC associated with O3, and SO4. POC is used as a fitting
species in CMB.
Synthesis
Various apportionments of PM2.5 OC concentrations are presented in
Sect. 3.1, 3.4, and 3.5. These apportionments are compared and contrasted
in this section. Although the apportionments utilize different methods,
there is overlap of inputs. For example, Kbb is used as an input in the
multivariate regressions that generate primary organic carbon (POC)
and secondary organic carbon (SOC) (Blanchard et al., 2008, not
discussed here), and POC is a fitting species used in the CMB receptor
modeling. As shown in Table 5, the apportionments exhibit areas of agreement
as well as certain differences. Both are summarized using ratios of the
values listed in Table 5. We report averages and ranges across the sites.
Computed POC represents 72 % (64–76 %) of mean OC
concentrations, whereas SOC represents 29 % (25–38 %). As
noted, SOC is the OC that is associated with O3 and SO4, which
constitutes a portion of SOA. POC is associated with EC, CO, and Kbb,
but may include oxidized OC that would be identified as SOA in other
analyses. For the CMB analysis, OC derived from area sources (primarily
biomass burning), mobile sources, and point sources is summed to generate
combustion OC. CMB combustion OC is 97 % (73–118 %) of POC;
this level of agreement presumably is because the CMB receptor model of
Blanchard et al. (2013) used POC as a fitting species. The largest PCA1
and PCA2 OC components are combustion, seasonal, and SO4-associated OC.
The sum of these three components is, for PCA1, 87 % (60–139 %) of
mean measured OC (the overestimate, at PNS, is balanced by negative crustal
and salt components there). For PCA2, the sum of combustion, seasonal, and
SO4-associated OC is 81 % (58–101 %) of mean measured OC.
Other PCA OC components contribute smaller amounts (Table 5).
PCA1 and PCA2 combustion each represent 57 % (8–103 and 33–85 %, respectively) of CMB combustion. Other PCA factors, including
SO2, metals, and salts (possibly denoting biomass burning when
represented by K) may be related to specific types of combustion sources.
These comparisons suggest that the OCbb concentrations are likely biased
high by ∼ 10 % or more, with less evident biases at inland
sites. Specifically, OCbb is 99 % (66–121 %) of POC and
109 % (79–142 %) of CMB area-source OC concentrations. At inland
sites, OCbb is 96 % (79–111 %) of CMB area-source OC
concentrations, indicating approximate agreement. Although multiple analyses
(OCbb, POC, PCA2) used Kbb as an input variable, OCbb is calculated
using a fixed scaling factor between OC and Kbb. As described, uncertainty
in this scaling factor is estimated to generate a factor-of-2 uncertainty
in OCbb.