Introduction
Aerosol acidity is a driver of many important atmospheric processes (Guo et
al., 2015; Weber et al., 2016), catalyzing the conversion of isoprene
oxidation products to form secondary organic aerosol (SOA) (Xu et al., 2015;
Pye et al., 2013; Surratt et al., 2010; Eddingsaas et al., 2010) and driving
the semivolatile partitioning of key aerosol species (Guo et al., 2015; Weber
et al., 2016). High acidity can also lead to the solubilization of iron,
copper and other trace metals in aerosol, which may serve as nutrients for
ecosystems (Meskhidze et al., 2003), but also prove toxic for humans (Ghio et
al., 2012; Fang et al., 2017). Significant reductions in primary pollutant
emissions over the last decades have greatly improved air quality in the
developed world and are also thought to fundamentally affect aerosol acidity.
SO2, an important aerosol precursor and a major driver of its
acidity, has seen decreases of about 6 % yr-1 over the 2001–2011
period alone in the USA, with a continued anticipated downward trend (Pinder
et al., 2007, 2008). Emissions of NOx and the resulting
acidic HNO3, are also declining. In contrast, ammonia, the primary
alkaline fine-mode aerosol precursor, was either constant or increasing
during this period (Pinder et al., 2007, 2008; Heald et al., 2012), owing to
intensified agricultural activity and livestock farming from the demands of
population growth. These trends have created the expectation that the aerosol
has and will become increasingly less acidic (West et al., 1999; Pinder et
al., 2007, 2008; Heald et al., 2012; Tsimpidi et al., 2007; Saylor et al.,
2015), with ammonium sulfate being replaced, at least in part, by ammonium
nitrate (West et al., 1999; Bauer et al., 2007; Bellouin et al., 2011; Li et
al., 2014; Goto et al., 2016).
The concept of nitrate substitution of sulfate has largely been based on the
notion that nitrate is volatile when the aerosol is acidic, and in turn
aerosol is acidic when insufficient amounts of total ammonia (i.e.,
gas+aerosol) or dust nonvolatile cations (NVCs) exist to neutralize aerosol
sulfate. Based on this conceptual model, aerosol ionic molar ratios have
largely been used as proxies of aerosol acidity (pH), so that when the
aerosol ammonium to sulfate molar ratio approaches 2 (the composition of
ammonium sulfate), aerosol is assumed neutral and only then can nitrate
aerosol form (Fisher et al., 2011; Hennigan et al., 2015; Wang et al., 2016;
Silvern et al., 2017). Modeling studies have corroborated this view,
predicting that nitrate substitution may be prevalent in the future,
including in the southeastern USA (SE USA) (Heald et al., 2012; Bauer et al.,
2007; Bellouin et al., 2011; Li et al., 2014; Goto et al., 2016; Vayenas et
al., 2005; Karydis et al., 2016). A more careful analysis, however (Guo et
al., 2015, 2016; Weber et al., 2016; Hennigan et al., 2015), reveals that
this conceptual model of aerosol acidity and conditions for nitrate
substitution fails; thermodynamic analysis of SE US aerosol observations
instead show that fine-mode aerosol remains strongly acidic, despite a
70 % reduction in sulfates and more than a sufficient amount of total
ammonia to neutralize it. The strong acidity is maintained by the large
difference in volatility between sulfate and ammonia (Guo et al., 2015; Weber
et al., 2016), so large changes in total ammonia concentrations are required
for a notable change in aerosol acidity, of about 1 order of magnitude
increase in NH3 concentration per unit increase in aerosol pH (Guo
et al., 2015, 2017c). However, ammonia gas deposits relatively rapidly,
limiting its build up except in high-emission regions. Throughout the decade,
the levels of aerosol nitrate have remained relatively constant throughout
the USA (Guo et al., 2015; Weber et al., 2016; Pye et al., 2009). The
persistent strong aerosol acidity in turn explains why nitrate aerosol has
not considerably increased over the last decades and is unlikely to appear in
the immediate future in the SE USA. These findings constitute a paradox, as
the same thermodynamic models (e.g., ISORROPIA-II, Fountoukis and Nenes,
2007) used to demonstrate the aerosol tendency for strong acidity in the SE
USA (Guo et al., 2015; Weber et al., 2016) using ambient data are also used
in 3-D modeling studies (Pye et al., 2009; Heald et al., 2012) for the region
that predicts nitrate substitution as a possible aerosol response.
Reconciling the nitrate substitution paradox requires a careful examination
of aerosol thermodynamics and the conditions under which nitrate partitioning
to the aerosol is favored. Meskhidze et al. (2003) and later Guo et
al. (2016) showed that, for aerosol nitrate formation to occur, aerosol pH
needs to exceed a certain characteristic value (depending on the temperature
and the amount of liquid water it ranges between a pH of 1.5 and 3; Guo et
al., 2017b). If aerosol pH is therefore high enough (typically above a pH of
2.5 to 3), a behavior consistent with nitrate substitution emerges, because
any inorganic nitrate forming from NOx chemistry mostly
resides in the aerosol phase. When pH is low enough (typically below 1.5 to
2), nitrate remains exclusively in the gas phase (as HNO3),
regardless of the amount produced, and nitrate substitution is not observed.
Between these high and low pH values, a sensitivity window emerges (of
typically 1–1.5 pH units), at which partitioning shifts from nitrate being
predominantly found as a gas to being mostly found as an aerosol. Therefore,
if a model is for any reason biased in its prediction of aerosol pH, it may
be preconditioned towards nitrate prediction biases. The sensitivity to pH
biases is strongest when the aerosol lies in the pH sensitivity window, which
is often the case for atmospheric aerosol (Guo et al., 2015, 2016, 2017b;
Bougiatioti et al., 2016). When below this pH sensitivity window, aerosol
nitrate is almost nonexistent and relatively insensitive to emissions (and pH
biases); when above the window, almost all nitrate resides in the aerosol
phase and directly responds to NOx emission controls.
If aerosols were composed only of nonvolatile sulfate, semivolatile
nitrate and ammonium, prediction
biases in pH could result only from errors in RH, and large errors (e.g.,
order of magnitude) of NH3, NOx and SO2
because pH is relatively insensitive to changes in these aerosol precursors
(Hennigan et al., 2015; Guo et al., 2017c). Acidity, however, can also be
modulated by other soluble inorganic cations from sea salt and mineral dust,
such as K+, Na+, Ca+2 and Mg+2. The
low volatility of these cations allows them to preferentially neutralize
sulfates over NH3, and, even in small amounts, elevate particle pH
to levels that can promote the partitioning of nitrates to the aerosol phase
(Fountoukis and Nenes, 2007; Guo et al., 2017a). NVCs tend to reside in the
coarse-mode aerosol, with a fraction found in smaller particles, while
sulfate tends to reside in the fine mode (e.g., West et al., 1999; Vayenas et
al., 2005; Guo et al., 2015); the degree to which NVCs can affect fine-mode
pH therefore lies in the degree at which the two types of species mix across
different particle sizes. Potential interactions between inorganics and
organics can also affect aerosol acidity. However, recent studies driving
thermodynamic models utilizing water associated with organics find only
minimal differences in pH predictions (Guo et al., 2015; Bougiatioti et al.,
2016; M. Liu et al., 2017; Pye et al., 2018; Song et al., 2018). In the
presence of very high NVCs (for example in sea-spray aerosol), where the
aerosol has much higher pH, the pH can approach the pKa of organic acids,
leading to conditions in which their dissociation can contribute to aerosol
acidity (Laskin et al., 2012).
Although aerosol models are evaluated in terms of their ability to predict
the concentration of aerosol species (including across size), no studies to
date focus on their ability to predict aerosol pH across size, even though it
is known to potentially vary up to 6 units (Fang et al., 2017; Bougiatioti et
al., 2016; Li et al., 2017). Evaluation of models in this context is
challenging, since there is no established data set of aerosol acidity –
although that is rapidly changing, with pH estimates derived from a
combination of observations and models (e.g., Guo et al., 2015, 2017b, c;
Bougiatioti et al., 2016; Y. Liu et al., 2017; Song et al., 2018).
Furthermore, given that most of this pH variability occurs in the PM1 to
PM2.5 range (Fang et al., 2017), it is quite likely that model
assumptions on how aerosol species interact within a mode (degree of internal
mixture) may lead to pH prediction biases that drive model behavior,
especially for particles in the 1–2.5 µm range.
The aim of this study is to address the underlying reasons for the nitrate
substitution paradox, and in the process, provide a conceptual framework for
quantifying and understanding the importance of aerosol pH biases. The
guiding hypothesis of this work is that aerosol pH prediction bias
fundamentally changes predicted aerosol behavior and is the underlying cause
of the paradox. The approach is demonstrated with the Community Multiscale
Air Quality (CMAQ) model (Byun and Schere, 2006) and is based on predictions
of pH over the 2001–2011 period in the southeastern and eastern USA, which
is the region in which aerosol pH trends are constrained by observations. The
role of internally mixed nonvolatile cations in PM2.5 as a source of the
pH bias is then assessed.
Results and discussion
Predicted sulfate, ammonium and nitrate
For the main inorganic aerosol species (SO4-2, NO3-
and NH4+), CMAQ captures the observed trends, as seen in the
literature (Park et al., 2006; Hand et al., 2012; Blanchard et al., 2013a, b;
Kim et al., 2015; Saylor et al., 2015) over the CONUS throughout
decade (Fig. S1 in the Supplement).
As expected, sulfate over the entire USA drops significantly between 2001 and
2011 (∼30 %), with major decreases in the eastern USA (∼2 µg m-3). Areas impacted the most by these reductions are
places of significant industrial activity or coal-fired electricity
generating units (EGUs), such as the Ohio River Valley, Baton Rouge in
Louisiana and South Carolina. Ammonium levels only experience small
reductions, which are consistent with a buffered
response to the decrease in sulfate levels, and minimal changes in emissions.
Local reductions (∼20 %) in ammonia are seen over North Carolina
and Louisiana. Aerosol nitrate concentrations remain constant on average over
the domain, with small increases over the eastern USA. The highest levels of
ammonium are observed in areas with significant livestock, such as North
Carolina and the Midwest; sulfate concentrations are the highest around the
Ohio River Valley due to SOx emissions and so is nitrate due
to significant NOx and ammonia emissions.
Predicted annual and seasonal pH
Figure 1 depicts the annual average pH fields over the USA for 2001 and 2011,
calculated using the annual average PM2.5 concentrations, with the study
domain of the outlined eastern USA. Simulations show that there are noticeable
differences between the two years, localized mainly in desert regions along
the US–Mexico border, southern Texas and the eastern USA. The sulfate
reductions in the eastern USA appear to have a major impact on model results,
leading to significant increases in aerosol pH in the area. For 2001, the
average yearly pH for the eastern USA is 1.6, consistent with recent
literature and observations from the WINTER campaign (Guo et al., 2015, 2016;
Weber et al., 2016) (Fig. 1a). For 2011, however, predicted pH increases to
about 2.5 – almost a unit higher (Fig. 1b).
pH diurnal profiles for May (a), August (b),
September (c) and November (d) at JST/RS/GT,
July (e) and December (f) at YRK and for the SOAS campaign
period (g). Blue and red lines are the offline ISORROPIA simulated
pH using CMAQ concentrations for 2001 and 2011 respectively, while the shaded
areas are 1 model standard deviation. The green line represents the pH
calculated through the thermodynamic analysis of the measurements (found in
Guo et al., 2015) and the shaded area is the standard error.
Seasonal pH trends are also positive over the eastern USA, with the
summertime (Fig. S2f) experiencing stronger increases than in the winter
(Fig. S2c), being 0.5–1.5 for winter and 0.5–2 for summer. Much of the
seasonal variability is driven by changes in temperature and relative
humidity: increased relative humidity (RH) leads to less acidic aerosol,
since liquid water content and pH are inversely related (Guo et al., 2015,
2016), while increased temperatures promote low RH nitrate partitioning and therefore more acidic aerosol. The desert areas of the
western USA, southern Texas, Florida, SW Alabama and Mississippi are the most
sensitive in the wintertime (Fig. S2a, b), while the central USA is mostly
unaffected. During the summer, the entire central USA is much more strongly
impacted, while the wintertime sensitive areas exhibit only minor pH
increases (Fig. S2d, e).
Model evaluation of pH
Model results for both simulation years were compared to thermodynamic
analysis of measurements from three urban sites in Atlanta, Georgia
(Jefferson Street, JST; Georgia Tech, GT; Atlanta Road-Side, RS) and two
rural (Yorkville, Georgia – YRK; and Centerville, Alabama – CTR) SEARCH
network sites. Measurements for the urban sites and the YRK site were taken
between May and December 2012 for the SCAPE study, while measurements from
the CTR site were for the SOAS campaign period (1 June to 15 July 2013) (Guo
et al., 2015; Xu et al., 2015). The three urban sites are contained within
the same CMAQ grid cell. All urban sites (Fig. 2a, b, c, d) exhibit an early
morning/late night pH maximum and an afternoon minimum throughout the year
(Guo et al., 2015). This a combination of two factors: RH being highest
during the early morning/late night, which increases water uptake and hence
decreases acidity (Guo et al., 2015) (Fig. S3), and the presence of crustal
elements in significant quantities during that time (Fig. S4). The model pH
closely tracks the diurnal profile of predicted cations (Fig. S4), indicating
that they have an important impact on predicted pH, which, however, is not
seen in the measurements (Fig. 2), since they make up a much smaller
percentage of observed PM2.5. Despite the presence of NVCs, the pH
remains low for both simulation years but it tends to be higher in 2011,
because of sulfate levels that are approximately half of those in 2001 across
all sites, leading to the increased relative effect of NVCs (Weber et al.,
2016). Removal of all NVCs from the thermodynamic calculations (Fig. S5),
significantly reduces the pH differences between 2001 and 2011 while removing
some of the increased variability introduced by NVCs. At the same time, a
negative bias is introduced to the simulated pH, which is more prominent for
the urban sites, even after the sulfate reductions.
The increase in pH is not proportional to the reduction in sulfate, since
aerosol responds nonlinearly to such reductions through the volatilization
of ammonia (Weber et al., 2016). Depending on location, sulfate reductions
range from 38 % to 55 %, while the corresponding pH increase is much
lower, pointing to the fact that cations, although small in amount, tend to
have a disproportionately strong impact on acidity. For the SOAS campaign
period (Fig. 2g), pH is underestimated, especially for 2001. The biases follow
the pattern of NVCs present, by being negatively biased until noon and
positively biased for the rest of the day (Figs. 2 and S4). The bias is
particularly evident in the early morning hours, when NVC concentrations
reach a maximum (Fig. S4). For the YRK site (Fig. 2b, e), pH is
underestimated overall during the winter and overestimated during the summer.
Similarly to the urban sites, the predicted RH agrees well with the
measurements (Fig. S3), albeit with a positive afternoon bias during the
summer. The diurnal profile of pH closely tracks the one of cations, further
suggesting they may be directly related to the bias.
When evaluating the predicted pH trend for CTR, the model results exhibit a
clear, increasing trend of 0.6 pH units per decade (Fig. 3). This trend is
inconsistent with recent thermodynamic analysis of observations, suggesting a
slight decrease in pH over the same time period for the SE USA (Guo et al.,
2015, 2016; Weber et al., 2016). If this bias in predicted pH trend
continues, it can have profound implications for future regulatory modeling,
since the increased pH can lead to elevated levels of model nitrate,
reproducing nitrate substitution (Bauer et al., 2007; Bellouin et al., 2011;
Li et al., 2014; Goto et al., 2016). Possible reasons behind this pH bias
could be overestimated ammonia emissions, underestimated sulfate or the
presence of NVCs in PM2.5. The first two possibilities are unlikely,
given the agreement of predicted ammonium and sulfate with previous studies
(Park et al., 2006; Hand et al., 2012; Blanchard et al., 2013a, b; Kim et
al., 2015; Saylor et al., 2015), and the relative insensitivity of pH to
ammonia and sulfate (Weber et al., 2016; Silvern et al., 2017). However,
NVCs, if inappropriately distributed in PM2.5, can exert significant
biases on pH (Meskhidze et al., 2003; Karydis et al., 2016; Guo et al.,
2017b). Indeed, offline calculations of aerosol pH excluding the influence of
NVCs mitigates most of the predicted positive trend of 0.6 pH units per
decade when all the aerosol species are considered (Fig. 3), while also
reducing the standard error. The remaining bias may arise from errors in model
RH, given that it controls water uptake and drives much of the diurnal
variability in pH (Guo et al., 2015). Usage of observed (instead of
predicted) RH in the thermodynamic calculations did not impact the predicted
pH more than 0.1 units (Fig. S6). A more thorough evaluation of the remainder
of the pH bias, as well as the impact of NVCs when included in appropriate
quantities, requires a far more extensive analysis of the emissions profiles
– especially regarding its diurnal variability – and observational data set
than the one available for this study (Henneman et al., 2017; Guo et al.,
2017c).
Decadal pH trends from the thermodynamic analysis of the
measurements from Weber et al. (2016) (blue line), default CMAQ (black line)
and CMAQ results at the Centreville grid cell without crustal elements (green
line). Also shown is the pH for the SOAS campaign and the CMAQ-predicted pH for 1 June–15 July 2001 and 2011. CMAQ exhibits a clear
positive trend, with pH increasing throughout the decade, both due to sulfate
reductions and the increasingly important role of NVCs. Standard error is
also plotted for all data points.
The pH bias becomes negative for most of the CMAQ eastern USA when removing
all NVCs from the calculations (Fig. S5). This, combined with the
considerable model skill for sulfate, nitrate and ammonium when compared to
the literature (Henneman et al., 2017), implies that pH biases are not related to
errors in the major inorganic ions or biases in meteorological parameters
(humidity and temperature), but rather in the NVCs, which are minor
contributors to PM2.5 and hence poorly constrained. For the SEARCH
sites, NVCs comprise 5 % to 10 % of the total inorganic PM2.5 (Guo et
al., 2015), which is significantly less than the model-predicted values
that are a factor of 4 higher than the measurements. The most important
result, therefore, is that NVCs are a considerable source of pH prediction
uncertainty when not accounted for correctly (Supplement: The role of NVCs in
PM2.5 pH). It should be noted that, for the summertime at the CTR
location, the ammonium and sulfate values are biased low in CMAQ by a factor
of 3 using the Weber et al. (2016) data. These biases, however, are consistent
with literature and typical of CTMs (Henneman et al., 2017).
The SEARCH sites have been thoroughly studied in the previous literature (Guo
et al, 2015, 2017a; Xu et al., 2015; Weber et al., 2016) and given the high
concentrations of organic mass observed throughout the year, they present an
excellent case study for the potential impact of organics on pH. Recent
studies indicate that organic aerosol can have a secondary, but still
quantifiable impact on aerosol pH, especially when allowed to interact with
inorganics (Pye et al., 2018). Most 3-D models do not account for potential,
nonideal interactions between the two, in addition to not including organics
in thermodynamic calculations, which, if the above statement is true, can
lead to significant predictive pH errors. To investigate the role of organics
on pH we used the E-AIM model (Wexler and Clegg, 2002; Friese and Ebel, 2010;
Clegg et al., 1992) (http://www.aim.env.uea.ac.uk/aim/aim.php, last
access: 1 April 2018) on our model results for the SEARCH sites to calculate
partitioning with the considered organic–inorganic interactions. We tested a
variety of organic compounds under different scenarios to determine the
potential of organics to influence pH (see Supplement: Organic acids and pH).
We find that the addition of organic compounds to the model did not have a
significant impact on acidity (≤2 % pH deviation from the baseline
value) compared to the baseline run, apart from the cases in which RH was
higher than 80 % and the mole fraction of organic acids in the aqueous
phase is greater than 25 % (Supplement: Organic acids and pH). We conclude that
the maximum impact of organics on aerosol pH can likely result from the
effects of liquid–liquid phase separation (Pye et al., 2018) but are of
insufficient magnitude to sustain a positive aerosol pH trend as observed in
our base case simulation.
CMAQ-derived nitrate partitioning ratio for the eastern USA and select
months of 2001. The black squares denote the average pH values for each
month. Note the insensitivity of nitrate partitioning to pH biases in the
summer for pH values of less than 1 ∂εNO3∂pH∼0, which is not the case
for colder months.
Increase in aerosol nitrate corresponding to a one-unit positive
change in pH for (a) January and (b) July. Emissions for
2011 are assumed, but to account for pH prediction biases from NVCs, they are
removed from the thermodynamic calculations. Plots are on different scales
due to the large differences in predicted nitrate increases.
The impact of pH biases on nitrate partitioning and sulfate–nitrate
substitution
To understand the importance of pH biases on nitrate partitioning and the
potential for predicting a behavior consistent with nitrate substitution, we
express the CMAQ output in each grid cell in terms of the nitrate
partitioning ratio, εNO3=NO3-HNO3+NO3-. It can be shown that εNO3
follows a simple sigmoidal curve (Meskhidze et al., 2003; Guo et al., 2016),
εNO3=1-H+H++L⋅T⋅Ψ, where L is the liquid water
content, T is ambient temperature, [H+] is the concentration of
H+ in the aerosol aqueous phase, and Ψ=R⋅HNO31000⋅P0 is a fitting parameter that
depends on the universal gas constant (R), the effective Henry's law
constant for nitric acid in the aerosol aqueous phase (HNO3) and
the standard pressure (P0).
Depending on the value of pH, nitrate partitioning in CMAQ can either be
insensitive ∂εNO3∂pH∼0 or sensitive ∂εNO3∂pH∼0.5 to pH biases,
depending on the month of the year considered (Fig. 4). We generally find
that nitrate partitioning is insensitive ∂εNO3∂pH∼0 and heavily shifted to
the gas phase (εNO3∼0) during the summer and
spring (Fig. 4), while it becomes quite sensitive to pH errors
∂εNO3∂pH∼0.5 in the winter and fall. For the latter case, this means that
small pH perturbations in either direction can strongly affect the amount of
nitrate that partitions in the aerosol phase; if the weather is sufficiently
cold and NOx emissions and pH predictions are biased
sufficiently high, εNO3∼1, meaning that all
nitrates are partitioned to the aerosol phase and the emergence of nitrate
substitution behavior.
To exemplify the above, we determine the amount of excess nitrate from pH
prediction biases as follows. Perturbing the acidity by ΔpH from
the monthly mean value along the εNO3 curves (Fig. 4)
gives the corresponding change in the partitioning ratio (ΔεNO3). Multiplying ΔεNO3
with the total nitrate (HNO3(g)+NO3) predicted in CMAQ
in each grid cell gives the total nitrate response (ΔNO3)
to ΔpH. Applying the eastern USA to ΔpH=+1 (the
average pH impact of including NVCs in the PM2.5 calculations over the
entire eastern USA) gives the ΔNO3 distributions shown in
Fig. 5 for the winter (Fig. 5a) and the summer (Fig. 5b). The predicted
wintertime nitrate bias tends to be higher than in the summer, owing to the
lower temperatures and higher aerosol pH levels present (which shift
εNO3 towards higher values; Fig. 4) and the higher
values of total available nitrate in the wintertime. The combination of both
factors (available nitrate and high pH) is necessary for appreciable
quantities of nitrate to partition, but in general the locations with a pH of
between 0.5 and 1 are the most susceptible to positive pH biases, since a
unit increase places nitrate partitioning into the ascending part of the
S-curve (Fig. 4), rapidly increasing the amount of aerosol nitrate that can
form. During both seasons, areas rich in total nitrate, and a pH between 0.5
and 1.5, such as the Ohio River Valley, New York, New Jersey and South
Louisiana (Figs. 1, S1e, f), exhibit the largest increases in aerosol
nitrate. Other locations that have low pH and low total nitrate such as West
Virginia see minimal changes. A notable exception is North Carolina which has
a higher pH than the aforementioned locations – mainly due to the high
NH3 emissions from livestock – which pushes the partitioning
beyond the sensitive regime, where increases in pH do not drive additional
nitrate in the particle phase.
CMAQ predicted nitrate substitution
NO32011-NO32001SO42001-SO42011
over the decade, when NVCs are accounted for (a), and when they are
removed from the thermodynamic calculations (b).
To investigate the potential of NVCs and sulfate reductions to induce nitrate
substitution, the sensitivity of the nitrate increase, ΔNO3, to the corresponding sulfate reduction, ΔSO4, was quantified for the eastern USA, both when NVCs are
included in the calculations and when they were not (Fig. 6). Over the
decade, nitrate has seen increases in the eastern USA (Fig. S11) ranging from
0.5 to 2.5 µg m-3, and NVCs can have a profound impact on how
these increases are distributed across the domain (Fig. S11a, b). In the
presence of NVCs (Fig. 6a), there is a 1 µg m-3 increase in
nitrate for a sulfate reduction of the same value over the eastern USA. For
this case, substitution is predicted across the entire eastern USA, with only
a few grid cells in southern Georgia, Mississippi and North Carolina
exhibiting the opposite trend (nitrate reduction), attributed to the
formation of insoluble CaSO4, which reduces the availability of
aerosol water, and in turn inhibits the formation of NO3 with the
co-condensation of NH3. When NVCs are removed (Fig. 6b), the
corresponding nitrate increase is much less (0–0.2 µg m-3
per 1 µg m-3 of sulfate), especially in the eastern USA,
while substitution is still predicted in the northern parts of the domain,
such as Ohio, Indiana and Michigan. The aforementioned areas tend to have
higher seasonal pH values than the SE USA (Fig. 1), and the removal of NVCs
reduces the pH to values at which nitrate partitioning is more sensitive to
small pH perturbations (Fig. 4), leading to a higher predicted sensitivity to
sulfate reductions. This analysis suggests that nitrate substitution is of a
much smaller magnitude than expected (West et al., 1999; Heald et al., 2012;
Bauer et al., 2007; Bellouin et al., 2011; Li et al., 2014; Goto et al.,
2016; Vayenas et al., 2005; Karydis et al., 2016) and heavily impacted by pH
biases introduced by NVCs.
Given the importance of aerosol acidity for almost any aerosol-related
process and impact, it is imperative that aerosol studies evaluate acidity
inferred from thermodynamic analysis of ambient data as presented here. We
demonstrate that, in the case of nitrate substitution, the distribution of
nonvolatile cations over particle size can have a profound impact on model
behavior and requires better constraints from emissions to observations (or
at least appropriate sensitivity studies, such as those carried out here, to
unravel the potential impact of nonvolatile cations). Understanding aerosol
pH and the drivers thereof is a powerful tool for evaluating model
performance that has never been used before. Usage of molar ratios, ion
balances and other conceptual models of aerosol acidity (Hennigan et al., 2015; Wang et al., 2016; Silvern et al., 2017) provide limited insights
into aerosol pH and should be strongly avoided to limit incorrect conclusions.