Introduction
Sea spray aerosols (SSAs) contribute significantly to the global aerosol
burden, both in terms of mass (Lewis and Schwartz, 2004) and cloud
condensation nuclei concentration (Murphy et al., 1998; Pierce and Adams,
2006; Clarke et al., 2006; Blot et al., 2013). The chemical composition of
SSAs (e.g., major ions: Na+, Mg2+, Ca2+, K+, Cl-,
SO42-; Tang et al., 1997) is affected
by atmospheric processing, with the uptake of nitric acid (Gard et al.,
1998, and references therein), sulfuric acid (McInnes et al., 1994),
dicarboxylic acids (Sullivan and Prather, 2007), and methylsulfonic acid
(Hopkins et al., 2008) shown to be important processes. Sea spray aerosols
also influence gas-phase atmospheric chemistry via displacement of chlorine
and bromine from the particle phase and subsequent impacts on ozone
formation and destruction (Yang et al., 2005; Long et al., 2014). Despite
this importance, much uncertainty remains in the factors affecting the
size-dependent production flux per whitecap area, which drives the emission
rates in most chemical transport models (de Leeuw et al., 2011).
An active area of recent research has been in the determination of the SSA
size distribution. The size distribution of particles influences their
atmospheric lifetime, surface area available for heterogeneous reactions,
cloud condensation nuclei efficiency, and optical properties. A widely used
SSA emission parameterization in early chemical transport models was
described by Monahan et al. (1986), which predicts the size distribution
between 0.8 and 8 µm in dry diameter based on laboratory
measurements. To address the overpredicted SSA emission rate when the
parameterization from Monahan et al. (1986) was extended to aerosol dry
diameters < 0.2 µm (Andreas, 1998; Vignati et al., 2001),
Gong (2003) revised the Monahan et al. (1986) parameterization to match the
SSA size distribution observed in the North Atlantic (O'Dowd et al., 1997)
down to a 0.07 µm dry diameter. Since the publication of
Gong (2003), several studies have examined the size distribution of SSAs
generated in the laboratory and measured in field campaigns (Mårtensson
et al., 2003; Clarke et al., 2006; Sellegri et al., 2006; Keene et al., 2007;
Tyree et al., 2007; Norris et al., 2008; Fuentes et al., 2010). In a review
of SSA emission measurements from both laboratory- and field-based studies,
de Leeuw et al. (2011) showed a broad range (0.05–0.1 µm in dry
diameter) of particle sizes having the maximum number production flux. Recent
SSA production parameterizations (see Grythe et al., 2014) reflect these
measurements, with most having a production rate maximum at aerosol sizes
lower than the lower cutoff (0.07 µm dry diameter) of Gong (2003).
Recent updates to the SSA emission parameterization in the Weather Research
and Forecasting model coupled with chemistry (WRF–Chem) increased predicted
submicron sodium mass concentrations over the northeastern Atlantic Ocean by
up to 20 % (Archer-Nicholls et al., 2014). Due to the lack of detailed
submicron measurements at the time, the Gong (2003) parameterization was
given as
dFdr=1.373U103.41r-(4.71+Θr-0.017r-1.44)(1+0.057r3.45)×101.607e-((0.433-logr)/0.433)2,
where dFdr is the SSA number flux with units of
m-2 s-1 µm-1, r is the particle radius in µm at
80 % relative humidity, U10 is the 10 m wind speed in m s-1,
and Θ is an adjustable shape parameter that controlled the submicron
size distribution. Gong (2003) tested Θ values between 15 and 40,
suggesting (with limited observational evidence) a Θ value of 30.
Seawater temperature can increase or decrease SSA number emissions by up to
∼ 100 % due to the temperature dependency of surface tension,
density, viscosity, and air entrainment (Mårtensson et al., 2003;
Sellegri et al., 2006; Zábori et al., 2012a; Ovadnevaite et al., 2014;
Callaghan et al., 2014). Mårtensson et al. (2003), Sellegri et
al. (2006), and Zábori et al. (2012a) all observed a negative temperature
dependence for the production flux of SSAs < 70 nm diameter in synthetic
seawater laboratory experiments. Similar negative temperature dependencies
are measured in SSAs generated from Arctic Ocean seawater (Zábori et al.,
2012b). Mårtensson et al. (2003) and Sellegri et al. (2006) also reported
positive temperature dependencies for the SSA production flux for particles
larger than 70 nm in diameter. This difference in the temperature dependence
of small and large SSA emissions is likely due to their bubble size
dependence and impact of SSTs on small and large bubbles (Sellegri et al.,
2006). Sofiev et al. (2011) developed a size-dependent temperature correction
factor for SSA emissions reflecting the different temperature dependencies of
fine- and coarse-mode aerosols. A global comparison of observed and
model-predicted coarse-mode sea salt concentrations in Jaeglé et
al. (2011) led to the development of a third-order polynomial function for
the SST dependence of the Gong (2003) SSA emission parameterization. Grythe
et al. (2014) compares the Jaeglé et al. (2011) and Sofiev et al. (2011)
temperature dependencies, finding that the Jaeglé et al. (2011) function
gives the best model improvement to the observed temperature dependence.
Modeling studies implementing the Jaeglé et al. (2011)
temperature-dependent SSA emissions have shown improved prediction of surface
sea salt mass concentration (Spada et al., 2013; Grythe et al., 2014)
relative to temperature-independent emissions. Using a process-based approach
incorporating seawater viscosity and wave state, Ovadnevaite et al. (2014)
found a positive temperature dependence of SSA emissions similar to
Jaeglé et al. (2011) but resembling a linear (rather than third-order
polynomial) relationship.
In addition to bubble bursting in the open ocean, SSAs can be emitted via
wave breaking in the surf zone covering an area roughly 20 to 100 m
from the coastline (Petelski and Chomka, 1996; Lewis and Schwartz, 2004).
Surf-zone SSA emissions have been shown to be enhanced relative to the open
ocean, resulting in higher sea salt concentrations near the coast (de Leeuw
et al., 2000). Vignati et al. (2001) concluded that surf-zone SSA emissions
provide additional surface for heterogeneous reactions and impact the
atmospheric chemistry of coastal areas. There are limited observations and
large uncertainties in the surf-zone SSA emissions related to the zone width
and whitecap coverage, with de Leeuw et al. (2000) observing a 30 m wide
surf zone with an assumed 100 % whitecap fraction on the California coast
and Clarke et al. (2006) observing a mean whitecap fraction in the 35 m
wide surf zone of 40 % in Hawaii. The inclusion of surf-zone emissions
increases sodium and chloride concentrations by a factor of 10 and improves
the predicted concentration of particulate matter (PM) < 10 µm
in diameter (PM10) by up to 20 % in the eastern Mediterranean (Im,
2013).
The current SSA treatment in the Community Multiscale Air Quality (CMAQ)
model version 5.0.2 is described by Kelly et al. (2010) and includes the
open-ocean emissions of Gong (2003), surf-enhanced emissions similar to de
Leeuw et al. (2000) in which a fixed whitecap coverage of 100 % is applied
to the Gong (2003) parameterization for a 50 m wide surf zone, and dynamic
transfer of HNO3, H2SO4, HCl, and NH3 between coarse-mode particles and the gas phase. Based on comparison with observations from
three Tampa, Florida sites at different distances from the coastline, Kelly
et al. (2010) found that enhancing sea spray emissions in surf-zone-containing grid cells by assuming a 50 m wide surf-zone width and
100 % whitecap coverage improved CMAQ model underprediction of sodium,
chloride, and nitrate concentrations (particularly at the coastal site)
relative to a simulation with only the Gong (2003) open-ocean emissions. The
dynamic transfer of HNO3, H2SO4, HCl, and NH3 between
coarse particles and the gas phase as implemented by Kelly et al. (2010)
further improves predicted concentrations of semi-volatile species like
chloride and nitrate. Despite these improvements, persistent
underpredictions of sodium, chloride, and nitrate concentrations at the
inland site remain unresolved. In this work, we expand upon the Kelly et al. (2010) CMAQ SSA emission treatment by updating the fine-mode size
distribution, SST dependence, and surf-enhanced emissions to reflect recent
SSA research. Due to the advanced treatment of SSA chemistry in CMAQ, their
emissions can be evaluated using concentrations of the directly emitted
sea salt components such as sodium and species such as nitrate that react
with sea salt components in the atmosphere. Specifically, we hypothesize
that the improved prediction of sodium will correspond to improvements in
the gas-particle partitioning of nitrate aerosol as suggested by Kelly et
al. (2014). The goal of this work is to improve the size distribution,
magnitude, and spatiotemporal variability of CMAQ-predicted SSA emissions
and the resulting impacts on atmospheric chemistry in coastal and inland
areas.
Methods
Observational data sets
Change in the (a) fine mode and (b) total surface sodium
concentration between the revised and baseline simulations for May 2002 over
the continental US and BRACE domains with sites from left to right:
Azalea Park, Gandy Bridge, and Sydney as green dots.
![]()
Two field campaigns with different meteorology, atmospheric chemistry, and
SSA sources from oceans having distinct surface temperatures and bathymetry
were used to evaluate the updated emissions. The Bay Regional Atmospheric
Chemistry Experiment (BRACE) (Atkeson et al., 2007; Nolte et al., 2008) was
conducted from May to June 2002 at three sites (Azalea Park: 27.78∘ N,
82.74∘ W;
Gandy Bridge: 27.89∘ N, 82.54∘ W; and Sydney: 27.97∘ N, 82.23∘ W) around Tampa Bay,
FL (see Fig. 1). These three sites represent coastal (Azalea Park),
bayside (Gandy Bridge), and inland (Sydney) regions, and roughly 1, 25, and
50 km from the Gulf of Mexico coastline. Size-resolved measurements of
inorganic PM composition were made with four micro-orifice cascade
impactors, which operated for 23 h per sample at ambient relative humidity
(Evans et al., 2004). The cascade impactors had 8–10 fractionated stages
ranging from 0.056 to 18 µm in aerodynamic diameter, and two cascade
impactors were collocated at the Sydney site. Additionally, particulate
nitrate and nitric acid were measured under ambient relative humidity
conditions at a high temporal resolution (≤ 15 min) using a soluble
particle collector employing ion chromatography (Dasgupta et al., 2007) and
denuder difference (Arnold et al., 2007).
The California Research at the Nexus of Air Quality and Climate Change
(CalNex) 2010 field project was conducted from May to July 2010 throughout
California. The goal of the study was to simultaneously measure variables
affected by emissions, atmospheric transport and dispersion, atmospheric
chemical processing, and cloud–aerosol interactions and aerosol radiative
effects (Ryerson et al., 2013). The South Coast portion of the CalNex
campaign included continuous ground-based measurements of
PM < 2.5 µm
in diameter (PM2.5) composition using particle-into-liquid
sampling and ion chromatography (Weber et al., 2001) and the mixing ratio of
many gases at Pasadena, CA (34.14∘ N, 118.12∘ W;
∼ 35 km from the Pacific coast). Here, we evaluated CMAQ for
June 2010 to coincide with surface concentrations of sodium and nitrate
measured continuously at Pasadena and as daily averages every 3 days at
sites operated by the national Chemical Speciation Network (CSN) within the
South Coast, San Francisco Bay, and San Diego air basins. Hereafter, these
CSN sites and the Pasadena site will collectively be referred to as the
coastal CalNex sites. Although the CalNex campaign also included ship-based
measurements of aerosol composition in conjunction with the Sea Sweep (Bates
et al., 2012; Crisp et al., 2014), the portion of the cruise that took place
in June 2010 was mainly in the vicinity of San Francisco Bay in close
proximity to several CSN sites already included in the evaluation. For the
CalNex comparison, the sum of the Aitken and accumulation modes was used as
the model comparison. However, a comprehensive evaluation of size-resolved
inorganic particle composition from Nolte et al. (2015) shows that the
difference in the sum of the Aitken and accumulation modes and PM2.5
values is < 10 %.
In addition to local field campaigns, we evaluated SSA emissions in CMAQ
against surface PM2.5 concentrations of sodium and nitrate measured
throughout the continental US (CONUS) as part of the Interagency Monitoring
of Protected Visual Environments (IMPROVE) for remote/rural locations and CSN
for urban locations during the May 2002 BRACE time period. Daily average
sodium mass concentrations in the IMPROVE and CSN networks were measured once
every 3 days via tube-generated X-ray fluorescence (XRF) (White, 2008).
Nitrate concentrations for both the IMPROVE and CSN networks are determined
by ion chromatography. During the May 2002 period, the IMPROVE network
consisted of ∼ 160 sites while the CSN network consisted of
∼ 230 sites. Although we use the filter-based measurements from the
IMPROVE and CSN networks and BRACE campaign for direct model evaluation, we
acknowledge that they have uncertainties related to instrument sensitivity
and volatility (White, 2008).
Model configuration
In this work, we used the CMAQ model v5.0.2 to simulate the impact of updated
sea spray aerosol emissions on surface aerosol concentrations/size
distribution and gas-particle partitioning. CMAQ represents the aerosol size
distribution using three modes (Aitken, accumulation, and coarse) and
simulates inorganic aerosol thermodynamics using ISORROPIA II (Binkowski and
Roselle, 2003; Fountoukis and Nenes, 2007). Kelly et al. (2010) further
enhanced the SSA chemical treatment in CMAQ by allowing dynamic transfer of
HNO3, H2SO4, HCl, and NH3 between coarse particles and
the gas phase. For comparison with the CONUS observational data sets such as
IMPROVE and CSN, we used a model domain covering the continental US at
12 km × 12 km horizontal resolution and 41 vertical layers with a
surface layer up to 20 m a.g.l. The simulation time period (1 May 2002 to
3 June 2002 with an 11 day spin-up) was made to coincide with the BRACE
campaign to enable additional evaluation of the coastal-to-inland changes in
the aerosol composition/size distribution and gas-particle partitioning.
Meteorological parameters were generated by the Weather Research Forecasting
model (WRF) version 3.1 (Skamarock et al., 2008), with initial and boundary
conditions generated from a previous CMAQ simulation and a GEOS-Chem global
model simulation, respectively. Detailed meteorological and emission inputs
can be found in Bash et al. (2013). For the CalNex comparison, we used a
model domain covering nearly all of California and Nevada as well as parts of
the Pacific Ocean, Mexico, and Arizona with 4 km horizontal resolution and
34 vertical layers. Chemical boundary conditions were derived from a
GEOS-Chem simulation (Henderson et al., 2014), and prognostic meteorological
fields used to drive CMAQ were generated with WRF version 3.4. Detailed
description of the meteorological and emission inputs can be found in Baker
et al. (2013) and Kelly et al. (2014). SST was taken from the Moderate
Resolution Imaging Spectroradiometer (MODIS) composite for all simulations.
As the Θ value primarily affects the fine-mode size distribution of
the Gong (2003) SSA production parameterization, adjusting Θ allows
the user to change the (1) number flux without affecting the mass flux and (2) peak aerosol
size emitted (see Fig. S1 in the Supplement). These two changes can result in
higher downwind concentrations of sea salt components due to the reduced dry
deposition velocities of fine-mode aerosols relative to the coarse mode and
resulting increase in atmospheric lifetime. The higher downwind
concentration of sodium aerosol can increase the concentration of nitrate
aerosol by affecting the gas-particle partitioning of total inorganic
nitrate (NO3- + HNO3). This increase, in turn, can increase
the nitrate lifetime as fine-mode NO3- has a longer atmospheric
lifetime than gaseous HNO3. Both the sea salt and nitrate aerosol
concentrations at the Sydney inland site were found to be underpredicted in
CMAQ (Kelly et al., 2010). For this study, we used Θ values of 30
(consistent with the current CMAQ representation, given as CMAQv5.0.2a or
“baseline”), 20 (CMAQv5.0.2b), 10 (CMAQv5.0.2c), and 8 (CMAQv5.0.2d),
which were expected to result in progressively higher emission rates of fin-
mode SSAs (see Fig. S1). For the simulations using Θ values ≤ 20, the lower limit of the SSA dry diameter is decreased to 10 nm to better
reflect changes in the emitted number size distribution (which peaks at
∼ 170, 140, 80, and 60 nm dry diameter for Θ values of
30, 20, 10, and 8, respectively). This decrease was consistent with
measurements of Aitken mode SSAs (Clarke et al., 2006) and a recent global
modeling study evaluating different SSA emission parameterizations (Grythe
et al., 2014). The radius of peak emissions at 80 % relative humidity (RH)
from the Gong (2003) parameterization with a Θ value of 8 was
∼ 60 nm; this value was similar to the radius of maximum
production flux from several parameterizations reviewed in de Leeuw et al. (2011).
Including the positive temperature dependence for SSA emissions in CMAQ was
expected to affect the seasonality and spatial distribution of predicted
concentrations. The Jaeglé et al. (2011) third-order polynomial function
of SST dependence for SSA emissions (CMAQv5.0.2e) increases the
summertime/tropical concentrations, decreases wintertime/polar
concentrations, and leaves mid-latitude/spring/autumn concentrations
relatively unchanged. The surf-zone width used in parameterizing the
surf-enhanced emissions was decreased from 50 to 25 m (CMAQv5.0.2f),
reflecting both the uncertainty in the width distance and whitecap fraction
within the surf zone. As SSA emissions from surf-zone-containing grids
impact a narrow region, adjusting the surf-zone width was expected to
strongly affect coastal concentrations while having a relatively minor
effect on downwind concentrations. We conducted two simulations to test the
combined effect of setting Θ=8, SST dependence, and
surf-enhanced emissions (surf zone = 25 m), with CMAQv5.0.2g using
the Jaeglé et al. (2011) third-order SST dependence and CMAQv5.0.2h
using a hybrid of the Jaeglé et al. (2011) third-order SST dependence
and the Ovadnevaite et al. (2014) process-based linear SST dependence (see
Fig. 12 from Ovadnevaite et al., 2014) for open-ocean emissions as follows:
dFdr=(0.38+0.054×SST)×1.373U103.41r-(4.71+8r-0.017r-1.44)(1+0.057r3.45)×101.607e-((0.433-logr)/0.433)2,
where SST has units of ∘ C. The updated SSA emission
parameterization given in Eq. (2) was mapped to the CMAQ aerosol modes as
a function of relative humidity following Zhang et al. (2005, 2006). A
summary of the different CMAQ model simulations in which SSA emissions were
changed is given in Table 1. The approach used in CMAQv5.0.2h, hereafter
referred to as the “revised” simulation, is planned to be included in the
next public release of CMAQ (version 5.1).
Differences in the CMAQ model versions used in this study.
Simulation
Θ
SST dependence
Surf
Zone (m)
Baseline1
30
None
50
CMAQv5.0.2b
20
None
50
CMAQv5.0.2c
10
None
50
CMAQv5.0.2d
8
None
50
CMAQv5.0.2e
30
Jaeglé et al. (2011)
50
CMAQv5.0.2f
30
None
25
CMAQv5.0.2g
8
Jaeglé et al. (2011)
25
Revised2
8
Jaeglé et al. (2011),
25
Ovadnevaite et al. (2014)
1 This simulation is also referred to as the CMAQv5.0.2a simulation.
2 In this simulation, which is also referred to as the CMAQv5.0.2h
simulation, the SST dependence of Jaeglé et al. (2011) has been
linearized following Ovadnevaite et al. (2014).
A potential limitation of this study is the reliance on ambient surface
concentrations in the evaluation of modeled SSA emissions. Although all
model processes other than SSA emissions are left constant for the CMAQ
simulations listed above, the selection of deposition, transport, and
chemistry parameterizations within the model can affect the predicted
concentrations. Nolte et al. (2015) found that constraining the aerosol mode
widths and enabling gravitational settling for all model layers in CMAQ
affected the predicted coarse-mode sodium at the BRACE sites. Although
changes in the model chemistry would likely have a minor impact on the
Na+ evaluations, future diagnostic evaluations that account for
deposition and transport uncertainties are advised.
Results
BRACE
Comparison of the mean and Pearson's correlation coefficient (r) of
total observed and model-predicted inorganic particle concentrations
(µg m-3) at three Bay Regional Atmospheric Chemistry Experiment (BRACE)
sites near Tampa, FL.
Species
Obs.
Baselinea
v5.0.2b
v5.0.2c
v5.0.2d
v5.0.2e
v5.0.2f
v5.0.2g
Revisedb
Mean
r
Mean
r
Mean
r
Mean
r
Mean
r
Mean
r
Mean
r
Mean
r
Azalea Park
NO3-
1.96
0.74
0.34
0.72
0.33
0.73
0.34
0.76
0.35
0.92
0.30
0.65
0.45
0.74
0.45
0.79
0.43
Cl-
1.93
2.41
0.17
2.33
0.15
2.36
0.15
2.49
0.18
3.69
0.19
1.55
0.31
1.92
0.38
2.15
0.42
Na+
1.62
1.62
0.19
1.61
0.18
1.62
0.18
1.71
0.21
2.39
0.22
1.11
0.33
1.38
0.41
1.52
0.44
Na+c
0.13
0.11
0.38
0.16
0.42
0.15
0.41
0.16
0.42
0.15
0.42
0.10
0.43
0.16
0.53
0.18
0.58
Gandy Bridge
NO3-
1.74
1.32
0.55
1.03
0.54
1.03
0.54
1.07
0.55
1.32
0.51
0.93
0.60
1.09
0.61
1.17
0.61
Cl-
1.72
1.57
0.71
1.51
0.71
1.53
0.71
1.63
0.71
2.53
0.68
1.32
0.81
1.91
0.81
2.26
0.81
Na+
1.46
1.17
0.67
1.17
0.67
1.17
0.67
1.24
0.67
1.78
0.65
1.01
0.79
1.41
0.81
1.62
0.80
Na+c
0.13
0.09
0.51
0.13
0.54
0.12
0.53
0.13
0.54
0.12
0.51
0.09
0.56
0.14
0.60
0.17
0.63
Sydney
NO3-
1.51
0.73
0.58
0.71
0.57
0.72
0.57
0.75
0.58
0.88
0.59
0.68
0.60
0.78
0.63
0.84
0.64
Cl-
1.31
0.82
0.35
0.78
0.35
0.79
0.35
0.86
0.36
1.32
0.30
0.71
0.49
1.02
0.50
1.26
0.53
Na+
1.14
0.67
0.44
0.66
0.45
0.67
0.45
0.72
0.46
0.98
0.41
0.59
0.55
0.82
0.57
0.98
0.61
Na+c
0.11
0.09
0.19
0.12
0.27
0.11
0.25
0.12
0.27
0.11
0.21
0.08
0.23
0.13
0.33
0.16
0.40
a This simulation is also referred to as the CMAQv5.0.2a simulation.
b This simulation is also referred to as the CMAQv5.0.2h simulation.
c Na+ predicted for the sum of Aitken and accumulation modes
(approximating PM2.5; Nolte et al., 2015) and observed for aerosols
< 1.8 µm in diameter.
The total particulate (PMtot) nitrate, chloride, and sodium
concentrations observed at the three sites during the BRACE campaign and
corresponding CMAQ-predicted concentrations for the baseline and sensitivity
simulations (v5.0.2b–h) are summarized in Table 2. The baseline simulation
predicted the magnitude of chloride and sodium at the coastal site (Azalea
Park) relatively well with normalized mean biases (NMBs) between 0 and
25 %. However, it increasingly underpredicted chloride and sodium as the
distance from the shore increased (at the inland Sydney site the sodium NMB
was -41 %). The baseline simulation overestimated by approximately a
factor of 2 the observed decrease in PMtot chloride and sodium
between the coastal Azalea Park and inland Sydney sites. The average
fine-mode sodium concentration (given as PM1.8 for the measurements and
the sum of the Aitken and accumulation approximating PM2.5 (Nolte et
al., 2015) for the model predictions) was consistently underpredicted by the
baseline simulation for the BRACE sites with an NMB of -21.6 %. The
baseline simulation underpredicted nitrate concentrations for all sites with
a NMB of -46.4 %. As the Θ value was changed from 30 to 20
(v5.0.2b), the predicted PMtot chloride and sodium (and nitrate via
secondary processes) at the coastal Azalea Park site decreased slightly
(< 0.1 µg m-3)
despite an increase (by 0.05 µg m-3) in fin-
mode sodium concentrations. This surprising result was due to slight
differences in the fitting of coarse-mode SSA emissions to CMAQ's aerosol
modes. The transition of Θ values from 20 to 10 to 8 led to small
(∼ 0.05 to 0.1 µg m-3, or 10 %) increases in the
nitrate, chloride, and sodium concentrations relative to the baseline
simulation for all sites. Although it slightly overestimated chloride and
sodium at the coastal Azalea Park site, the v5.0.2d simulation with a
Θ value of 8 had the best prediction (both in terms of magnitude and
correlation according to Table 2) of concentrations at the Gandy Bridge and
Sydney sites.
The modeled chloride and sodium aerosol concentrations were much more
sensitive to the implementation of SST-dependent SSA emissions (v5.0.2e) and
reduction of the surf-zone width used for surf-enhanced SSA emissions
(v5.0.2f) than the changing of the Θ values. With the positive
temperature dependence of the Jaeglé et al. (2011) sea spray aerosol
emissions and warm (25 ∘C) Gulf of Mexico surface waters in May
(see Fig. S2), concentrations of nitrate, chloride, and sodium were
predicted to be higher (> 20 %) in the v5.0.2e simulation than
the baseline for all sites. The reduction in surf-enhanced emissions in the
v5.0.2f simulation had a more site-specific impact on surface
concentrations, with the coastal Azalea Park site having a 0.4–0.5 µg m-3
(30 %) decrease in predicted chloride and sodium concentrations
and the bayside (Gandy Bridge) and inland (Sydney) sites having only a
10–15 % decrease relative to the baseline simulation. Figure S3 shows the
model grid cells in the vicinity of Tampa Bay (including the Gandy Bridge
site) have a representation of the open-ocean fraction but not the surf-zone
fraction used for surf-enhanced SSA emissions. The predicted 50 % decrease
in the chloride and sodium surface concentrations from Azalea Park to Sydney
in the v5.0.2f simulation was more similar to the observed 30 % decrease
than the 60 % decrease predicted by the baseline simulation.
In general, the best model performance at the BRACE sites occurred with SSA
emissions having a Θ value of 8, SST dependence, and a reduced surf
enhancement as implemented in the v5.0.2g and revised simulations. While both
the v5.0.2g and revised simulations severely underpredicted nitrate
concentrations (by up to 1.2 µg m-3) at all sites, the
chloride and sodium concentrations were consistently improved both in
magnitude and correlation compared to the baseline simulation (see Table 2).
The largest improvement occurred at the inland Sydney site, where substantial
underpredictions of chloride and sodium in the baseline simulation were
largely eliminated in the revised simulations (chloride and sodium NMBs
improved from -37/-41 % to -4/-14 %, respectively). Comparison of the
simulations with the third-order polynomial (v5.0.2g) and linear (revised)
SST dependence of SSA emissions revealed that the linear dependence led to
slightly improved prediction of chloride and sodium at the Azalea Park and
Sydney sites (Pearson's correlation coefficients jumped from 0.57 to 0.61 and
biases went from -0.32 to -0.16 µg m-3 for sodium in
Sydney) and similar performance at the Gandy Bridge site. Improved prediction
of chloride and sodium concentrations at these sites was not surprising as
the linear temperature dependence was adapted from a process-based
parameterization incorporating seawater viscosity and wave state (Ovadnevaite
et al., 2014) as opposed to the top-down, model-specific third-order
polynomial parameterization developed for GEOS-Chem in Jaeglé et
al. (2011).
Time series of the observed and predicted daily PM10 and
PM1.8 Na+ concentration at the three BRACE sites. Note that the
PM1.8 Na+ concentration predicted by CMAQ is represented by the
sum of the Aitken and accumulation modes.
![]()
The statistical improvement in the revised simulation relative to the
baseline simulation is reflected in the time series of sodium concentrations
at the three sites (Fig. 2). Besides showing the generally higher
PMtot sodium concentrations at the bayside and inland sites and higher
PM1.8 sodium concentrations at all sites, Fig. 2 also shows that the
revised simulation diverges most from the baseline during periods of high
SSA concentration episodes (15, 22 May 2002). This suggests that the revised
simulation better replicated the sea spray aerosol emissions during periods
with strong onshore flow compared to the baseline simulation. The range of
PM1.8 sodium concentrations predicted by the revised simulation was
more consistent with observations than the baseline simulation, especially
at the Sydney site which has observed concentrations of 0.05–0.27 µg m-3
and predicted concentrations of 0.02–0.16 µg m-3 and
0.03–0.25 µg m-3 for the baseline and revised simulations. The
PM1.8 sodium concentrations at the BRACE sites were lower than
PM2.5 sodium measured at a nearby CSN site (located at 28.05∘ N,
82.378056∘ W) averaging 0.34 µg m-3 during the same period but
well correlated (correlation coefficients ranging from 0.65 to 0.90) for the
5–6 days of coincident measurements. This CSN site is part of the CONUS
model evaluation described in Sect. 3.3.
Observed and predicted size distributions of Na+ at the three
Tampa-area sites averaged over 15 sampling days (14 at Sydney) during 2 May–2 June 2002.
![]()
Comparison of the predicted and observed size distribution of sodium at the
three sites (see Fig. 3) showed that much of the observed and predicted
decrease in the sodium mass concentration in the transition from coastal to
inland sites occurred within the coarse mode. The baseline simulation
overpredicted/underpredicted coarse-mode sodium at the coastal/inland sites,
while the revised simulation well predicted the coarse-mode sodium at both
the coastal and inland sites. At the bayside Gandy Bridge site, the high SSTs
in Tampa Bay resulted in an increase in the bias from the baseline
simulation due to the revised simulation overestimating coarse-mode
observations. Both the baseline and revised simulations predict a second
submicron mode for the three sites that is not evident in the observations;
it is unclear whether this discrepancy is related to inaccuracies in the
size-resolved emissions or the modal distribution of the model.
Fine-mode (Aitken + accumulation) sodium concentrations increased
throughout the BRACE domain in the revised simulation relative to the
baseline simulation with larger changes (up to 0.1 µg m-3)
offshore and smaller changes (0.05 µg m-3) inland as shown in
the right column of Fig. 1a. The total (sum of Aitken, accumulation, and
coarse modes) sodium concentrations over the open ocean increased in the
warmer southern waters of the Atlantic and Pacific oceans and decreased in
the cooler waters off New England and the Pacific Northwest. Grid cells
directly adjacent to the coast experienced concentration decreases of up to
1 µg m-3, with the largest decreases occurring for cells
with large surf zones due to irregular coastlines (barrier islands,
peninsulas, etc.). These coastline-centered decreases were limited
spatially, as adjacent cells just offshore had large increases in sodium
concentration. Like the fine-mode changes, the largest total sodium
concentration increases occurred offshore while more modest increases were
predicted for inland locations. The coastal-inland concentration gradients
were stronger for the total concentration changes due to the faster
deposition velocity of coarse-mode aerosols (relative to the fine mode)
that comprise most of the total mass.
Time series of observed and modeled fraction of total nitrate in
the particle phase
[NO3-/ (HNO3+ N3-)]
at the Sydney, FL, site for May 2002. Tick marks represent 00:00 local
standard time on each day.
![]()
The hourly time series of observed and predicted nitrate gas/particle
partitioning from the Sydney site for May 2002 (Fig. 4) shows that the
revised simulation pushes the partitioning towards the particle phase
(relative to the baseline simulation) and closer to observations. The
average observed fraction of nitrate in the particle phase was 0.51, while
the predicted fractions from the baseline and revised simulations were 0.36
and 0.42, respectively. Figure 4 indicates that the largest difference in
the nitrate partitioning between the baseline and revised simulations
occurred during the daytime, when higher concentrations of inorganic ions
like sodium prevented some of the nitric acid evaporation from the particle
phase during the hot afternoon period. Despite improvement in the daytime
partitioning, the revised simulation continued to overpredict the nighttime
nitrate fraction and daytime nitric acid fraction. This impact on
partitioning is consistent with Kelly et al. (2014), which suggested that
improving CMAQ prediction of sodium concentration and relative humidity
would improve gas-particle partitioning of nitrate in the CalNex model
domain.
CalNex
Change (µg m-3) in the fine-mode (Aitken + accumulation)
surface sodium concentration between the revised and baseline
simulations for June 2010 over the CalNex domain surrounded by time series
plots of the observed and predicted daily and/or hourly PM2.5-sodium
concentration at the coastal CalNex sites.
![]()
Similar to results for the BRACE sites, the predicted fine-mode sodium
surface concentrations were improved in the revised simulation relative to
the baseline for sites examined during the CalNex simulation period (see
Fig. 5). Surface sodium concentrations were underpredicted by both the
baseline and revised simulations for all the coastal CalNex sites,
especially in the 11–16 June time period when high sodium concentrations at
several of the sites were not well captured by either the revised or
baseline simulation. It is worth noting that a sensitivity test in which the
surf-enhanced emissions were increased (using a surf-zone width of 100 m
rather than 25 m as in the revised simulation) did not
substantially improve the sodium underpredictions at the coastal CalNex
sites. Monthly average (June 2010) sodium concentrations predicted in the
revised simulation increased by up to ∼ 0.25 µg m-3
off the California coast relative to the baseline simulation, with increases
between 0.05 and 0.1 µg m-3 widespread in the San Francisco, Los
Angeles, and San Diego air basins (Fig. 5). Hourly or daily average
increases between the revised and baseline simulations were even higher in
these urban areas, with the time series plots in Fig. 5 showing increases
up to 0.2 µg m-3. The spatial patterns of impacts on sodium in the
Central Valley and South Coast air basin matched those of tracers released
from San Francisco and LAX airport that are drawn inland on the sea breeze
(Baker et al., 2013).
Change (µg m-3) in the fine-mode (Aitken + accumulation)
surface nitrate concentration between the revised and baseline
simulations for June 2010 over the CalNex domain surrounded by time series
plots of the observed and predicted daily and/or hourly PM2.5 nitrate
concentration at the coastal CalNex sites.
![]()
Improving the sodium underprediction at the coastal CalNex sites in the
revised simulation had the effect of improving the frequent nitrate aerosol
underprediction at the same sites (see Fig. 6). Unlike the sodium
concentration changes, the largest (0.5 µg m-3) increases in
monthly average nitrate aerosol concentration occurred over the Los Angeles
air basin well inland from the coast. The increase of nitrate largely
occurred in inland areas where nitric acid was produced downwind of urban
centers with large NOx emissions. For conditions unfavorable for
ammonium nitrate formation (e.g., high temperature, low RH, low NH3),
nitrate may still form in sea spray particles through replacement reactions
(e.g., NaCl(p) + HNO3(g) → NaNO3(p) + HCl(g)). Since such
pathways involve pollution derived from urban emissions (HNO3) in
addition to sea salt (NaCl), the highest nitrate increases occurred inland
despite the relatively small increases in sodium compared to the baseline
simulation in these areas. Similarly, polluted sites such as Pasadena and
Riverside had larger increases in nitrate concentrations than cleaner sites
in the San Francisco air basin despite having similar sodium concentration
changes. This behavior suggested that these SSA emission updates had the
largest air quality impact in coastal urban areas with mixtures of marine
and polluted air masses. Note that the nitrate-to-sodium ratio of molar
masses is about 2.7, and so a 1:1 increase in the moles of sodium and
nitrate according to NaNO3 stoichiometry would lead to a greater
increase of nitrate than sodium mass. The nitrate underpredictions in Fig. 6
were not resolved entirely by improved sodium predictions. In Riverside,
for example, nitrate underpredictions in the revised simulation were likely
due to a combination of persistent sodium underpredictions and an
underestimate of ammonia emissions from upwind dairy facilities (Nowak et
al., 2012; Kelly et al., 2014).
Continental US
Unlike the PM1.8 or PM2.5-sodium concentrations evaluated using
the BRACE and CalNex observations, the total sodium surface concentration
changes shown in Fig. 1b both increased and decreased in the CONUS domain
due to the variability in coastal and oceanic SSTs. The distribution of fine-mode
(Aitken + accumulation) concentration changes (Fig. 1a) had some
similar features to the total concentration changes (Fig. 1b), with the
largest increases occurring over areas with high (> ∼ 20 ∘C) SSTs. Differences between the fine mode and
total concentration changes were most notable for regions with low
(< ∼ 10 ∘C) SSTs (Pacific and northeastern US
coasts) and for inland regions. Because fine-mode particles have a low dry
deposition velocity, offshore increases in the fine-mode sodium
concentrations were able to extend inland and lead to increased deposition
(see Fig. S4a). The flat topography and large offshore concentration
increases in the southeastern US resulted in concentration increases of up to
0.25 µg m-3 hundreds of kilometers from the coast. While
reductions in fine-mode SSA emissions due to low SSTs were balanced by
increased emissions from changing Θ, cold seawater temperatures off
the Pacific coast and northeastern US led to large decreases in total sodium
concentration of up to -0.5 µg m-3. As in the BRACE domain, the
decrease in surf-enhanced emissions led to localized decreases in PMtot
sodium concentration for grid cells immediately adjacent to the coastline
throughout the CONUS domain. Regions with rugged coastlines and barrier
islands experienced the largest concentration decreases because of the large
surf-zone area.
Model bias of PM2.5-sodium concentration predicted by the
revised simulation compared to observations from the IMPROVE (triangles) and
CSN (squares) networks for May 2002 segregated by an (a) increase or (b)
decrease in the error relative to the baseline simulation. The map only
includes data where the model percentage difference between the revised and
baseline simulations is > 5 %.
![]()
Statistical comparison of the mean and Pearson's correlation
coefficient (r) between observed and model-predicted sodium, nitrate and
PM2.5 surface concentrations (µg m-3) for the continental
US in May 2002 from the IMPROVE and CSN networks.
Specie
Obs.
Baseline1
v5.0.2g
Revised2
Mean
r
Mean
r
Mean
r
IMPROVE
Na+
0.44
0.16
0.11
0.16
0.17
0.19
0.20
NO3-
0.61
0.23
0.28
0.26
0.26
0.26
0.27
PM2.5
5.98
4.24
-0.04
4.16
-0.01
4.30
0.04
CSN
Na+
0.34
0.11
0.59
0.14
0.62
0.15
0.62
NO3-
1.94
0.61
0.76
0.68
0.76
0.68
0.75
PM2.5
9.74
6.04
0.74
6.29
0.74
6.48
0.74
1 This simulation is also referred to as the CMAQv5.0.2a simulation.
2 This simulation is also referred to as the CMAQv5.0.2h simulation.
Model comparison of PM2.5-sodium concentrations from the IMPROVE and
CSN networks revealed improvement from the baseline to revised simulation
(see Fig. 7). For both the IMPROVE and CSN networks, far fewer sites had
an increased error (Fig. 7a) in the revised simulation relative to the
baseline than had reductions in the model error (Fig. 7b). Sites where the
model error increased in the revised simulation were widely scattered across
the CONUS domain and typically overpredicted concentrations. The sites where
model error was reduced in the revised simulation were in the southeastern and
mid-Atlantic US and typically underestimated concentrations. Sodium
concentrations at numerous sites were underpredicted by > 0.1 µg m-3 in the revised simulation, suggesting that the SSA
emission changes were insufficient to bring the model into agreement with
most observations. Despite cold waters off the Pacific coast leading to
lower emissions (relative to the warmer Gulf of Mexico) in the revised
simulation, there were more sites in California that had an error reduction
in the predicted concentrations than had increased model error. Cold waters
in the Gulf of Maine and the associated lower emissions/concentrations in
the revised simulation had the effect of reducing the overprediction of
sodium at several sites in coastal New England. Table 3 shows that the
average bias for sodium concentrations for all stations in the IMPROVE and
CSN networks was reduced from the baseline to revised simulation (NMB: -63.7
to -57.6 % and -67.2 to -54.9 % for the IMPROVE and CSN networks,
respectively) with small improvements in the correlation. Predicted nitrate
concentrations improved in the revised simulation relative to the baseline,
with slight reductions in the large model underpredictions for the IMPROVE
(NMB: -62.7 to -56.8 %) and CSN (NMB: -68.6 to -65.0 %) networks.
Despite similar changes in average sodium concentrations between the
baseline and revised simulations for the IMPROVE and CSN networks, the
average change in PM2.5 between the two simulations was much higher for
the CSN (+0.42 µg m-3) than the IMPROVE (+0.06 µg m-3)
network. Predominantly comprised of urban sites, CSN sites
are located in more polluted regions where changes in sodium concentrations
were more likely to have an impact on the partitioning of HNO3, HCl,
and NH3 between gas and particle phases leading to increases in nitrate
aerosol concentrations (see Fig. 6 for an example). The enhanced
partitioning of nitrate to the particle phase in the revised simulation also
led to decreased deposition of total nitrate inland because of the lower dry
deposition velocity of nitrate aerosol relative to nitric acid (see Fig. S4b).
Conclusions
In this study, the size distribution, temperature dependence, and surf-zone
enhancement of sea spray aerosol (SSA) emissions were updated in the
Community Multiscale Air Quality (CMAQ) model version 5.0.2. Increasing fine-mode emissions, including temperature dependence, and reducing the
surf-enhanced emissions from the “baseline” to the “revised” simulation
collectively improved the summertime surface concentration predictions for
sodium, chloride, and nitrate at three Bay Regional Atmospheric Chemistry
Experiment (BRACE) sites near Tampa, Florida. Surface concentrations at the
inland site near Tampa were particularly affected by these emission changes,
as low dry deposition velocities for the fine-mode aerosols increased the
atmospheric lifetime and inland concentrations. The coastal–inland
concentration gradient was also affected by the updated emissions, as the
reduction in surf-zone width used to enhance surf-zone emissions brought the
revised simulation in closer agreement with observations. These SSA emission
updates led to increases in the fine-mode sodium surface concentrations
throughout coastal areas of the continental US, with the largest increases
occurring near the southeastern US coast where sea surface temperatures (SSTs)
were high. Decreases in the total sodium concentration were predicted for
oceanic regions with low SSTs such as the Pacific and northern Atlantic
coasts. Comparison of the baseline and revised simulation with sodium
observations from the IMPROVE and CSN networks showed that the updated
emissions reduced the widespread underprediction of concentrations,
especially in the southeastern and mid-Atlantic US. Non-linear responses
between changes in total and sea salt PM2.5 concentrations indicated
that the impacts of these emissions changes on aerosol chemistry were
enhanced in polluted coastal environments. The revised simulation had
increased sodium and nitrate aerosol concentrations at most CalNex sites,
slightly reducing the underprediction from the baseline simulation.
Potential future work includes treating the organic fraction of SSAs (Gantt
et al., 2010), implementing the Group for High Resolution Sea Surface
Temperature (GHRSST) data set (Donlon et al., 2007), and linking the SSA
emissions to marine boundary layer halogen chemistry via debromination (Yang
et al., 2005). Episodic high SSA concentrations are not well captured at any
of the coastal CalNex sites in the revised simulation, suggesting that other
factors not accounted for in our updated SSA emission parameterization such
as wind history, wave state, ocean biology, solar radiation, whitecap
timescales, or the limited ocean surface area in the modeling domain
(Callaghan et al., 2008, 2014; Ovadnevaite et al., 2014; Long et al., 2014) may play an important role. Additional model
developments focused on the South Coast region of California are warranted
considering the impact on nitrate discussed above as well as the impact that
reactive chlorine atoms derived from sea spray particles can have on ozone
in this region (Simon et al., 2009; Sarwar et al., 2012; Riedel et al.,
2014). As the fine-mode size distribution has a far greater impact on the
number concentration than the mass concentration, the changes described in
this study likely impact other model parameters such as aerosol radiative
feedbacks, which are included in the coupled WRF–CMAQ modeling system (Gan et
al., 2014).
Code availability
The updated code is available upon request prior to the public release of
CMAQ v5.1. Please contact Jesse Bash at bash.jesse@epa.gov for more
information.