Introduction
Numerous federal (e.g., United States Environmental Protection
Agency, USEPA), state and private entities rely on numerical model
simulations of atmospheric chemistry, transport and deposition of airborne
emissions as well as the resulting pollutants as part of their decision-making
process for air quality management and mitigation (e.g., Scheffe et al.,
2007). Chemical transport models (CTMs), such as the Community Multiscale Air
Quality (CMAQ) model (Byun and Schere, 2006), are often employed to provide
information about the potential effects of emission control strategies (e.g.,
Fann et al., 2009) and climate change (e.g., Nolte et al., 2008), and to provide
next-day air quality forecasts (e.g., Eder et al., 2006) in order to inform
and protect the public from potentially harmful air pollutants. Since these
models are often used to inform the standard setting and implementation for
criteria pollutants (e.g., ozone (O3) and fine particulate matter (PM2.5)),
they must be maintained at the state-of-the-science level. New
versions of the CMAQ model have been released periodically over the past
15 years, with each new version consisting of numerous updates to the
scientific algorithms within the model, while also improving the quality of
the input data used. Collectively, these updates are aimed at improving the
underlying science of atmospheric dynamics and chemistry represented in the
model, extending the capabilities for emerging applications, and reducing
systematic biases in the modeling system. Every new release of the CMAQ model
undergoes extensive evaluation in order to establish its credibility (e.g.,
Mebust et al., 2003; Appel et al., 2007, 2008, 2013; Foley et al., 2010) and
documents its performance relative to previous versions. Most recently, the
CMAQ modeling system version 5.1 (v5.1) has been tested and evaluated against
observations and was publicly released in December 2015
(http://www.cmaq-model.org/).
The scientific upgrades in the CMAQv5.1 modeling system include major
revisions to the Pleim–Xiu land-surface model (PX-LSM; Pleim and Xiu, 1995)
and the asymmetric convective mixing version 2 (ACM2; Pleim, 2007a, b)
planetary boundary layer (PBL) model in the Weather Research And Forecast
(WRF) model version 3.7 (Skamarock et al., 2008), which required revisions to
the ACM2 scheme in CMAQ to maintain consistency. Corrections were also made
to the Monin–Obukhov length (MOL) calculation in CMAQv5.1 to make it
consistent with the calculation in the WRF model. The changes to the PX-LSM,
ACM2 and MOL calculations in CMAQ had significant impact on the mixing within
both WRF and CMAQ, and hence large impacts on the pollutant concentrations in
CMAQ. These updates are described in Sect. 2.1. A new explicit treatment of
secondary organic aerosol (SOA) formation from isoprene, alkenes and polycyclic aromatic hydrocarbons (PAHs) was
also added in CMAQv5.1. Additionally, two aerosol mechanisms are now
available in v5.1, AERO6 and AERO6i (with isoprene extensions), which include
updates to the SOA and ISORROPIA algorithms (Nenes et al., 1998, 1999). The AERO5 mechanism has been deprecated and is no longer
available. The updates to the aerosol treatment in v5.1 are described in
Sect. 2.2. Significant changes were also made to the in-line calculation of
photolysis rates (described in Sect. 2.3). The photochemistry in
v5.1 underwent major changes; specifically, the photochemical cross sections
and quantum yields for the carbon bond 2005 e51 (CB05e51) chemical mechanism
were updated, along with updates to inorganic and organic chemical reaction
rates and products to ensure consistency with the International Union of Pure
and Applied Chemistry (IUPAC). Finally, the representation of organic nitrate species in CB05e51 was added. These updates are described in Sect. 2.4.
Section 2 provides a brief description of the major scientific and structural
improvements included in v5.1. The model configuration and observational data
sets used in the model evaluation are provided in Sect. 3. The evaluation
of v5.1 is then presented in two parts. Section 4 documents the evaluation of
several specific changes that were isolated as part of the overall testing of
the model. Specifically, Sect. 4.1 evaluates the meteorological updates in
WRF and CMAQ; Sect. 4.2 evaluates the aerosol updates; Sect. 4.3
evaluates the changes to the inline photolysis calculation and the
representation of clouds within CMAQv5.1; and Sect. 4.4 evaluates the
updates to the CB05e51 chemical mechanism. These increments were chosen as
the focus of this paper because they represent a fundamental change from the
previously released model version and had the propensity to impact model
performance for criteria pollutants. The second portion of the evaluation,
presented in Sect. 5, summarizes the overall change in PM2.5 and
O3 model performance with v5.1 compared to the previously released
version (CMAQ version 5.0.2: v5.0.2). Section 6 provides a discussion of the
model response of O3 and PM2.5 to hypothetical reductions in
emissions. Finally a summary discussion in provided in Sect. 7.
Review of scientific improvements in CMAQv5.1
Improvements to the v5.1 modeling system are the result of many years of
scientific advancements derived from laboratory, field and numerical
experiments as well as the efforts of a relatively small group of model developers
that both investigate avenues for model improvements and then update the
model (i.e., write code). Given the large community of CMAQ model users, there
are never sufficient resources to diagnose and address every issue in the
modeling system that has been reported. As such, it is necessary to
prioritize updates to the model based on many different factors, including
results from evaluations of past model versions, existing and upcoming
regulatory needs, emerging scientific issues, requests from the CMAQ user
community, and the expertise within the model developer group to meet those
needs and requests. The updates presented herein represent the “major” updates
made to the CMAQ modeling system from the previous model version, and
therefore do not constitute a fully comprehensive description of all the
changes made to the system. This section briefly describes these “major”
updates to CMAQ, providing the reader with an understanding of what was
updated in the model and why. A comprehensive description of all the updates
made in v5.1 and in-depth technical documentation of those changes can be
found on the CMAS Center website for the CMAQv5.1 release at
https://cmaswiki-cempd.vipapps.unc.edu/index.php/.
WRF and CMAQ meteorological and transport updates
The WRF and CMAQ models were updated to improve the representation of
land-surface processes and vertical mixing. There were two changes made to
the PX-LSM in WRF. First, the stomatal conductance function for
photosynthetically active radiation (PAR) was revised based on measurements
of net photosynthetic rate as a function of PAR for cotton plants, reported by
Echer and Rosolem (2015). The new functions yield a significantly lower
magnitude when shortwave radiation is less than 350 W m-2. This in turn
results in reduced latent heat flux and enhanced sensible heat flux, causing
a delay in surface stabilization (prolongs mixing) during evening transition
hours (i.e., sunset). This reduces overestimations (reduced positive bias) in
water vapor mixing ratios, which are common in the WRF–CMAQ modeling system
during the evening transition. Similarly, overestimation of concentrations of
surface-emitted species (e.g., NO, NO2, CO and elemental carbon (EC)) are also reduced
during the evening transition. This change was released in WRFv3.7 and
further revised in WRFv3.8. The second change made to the PX-LSM is an
increase of the coefficient to the surface energy forcing in the soil
temperature force-restore equation (Cv), which is related to
volumetric heat capacity (cv) and heat conductivity (λ)
(Pleim and Gilliam 2009) as follows:
Cv=2πcvλτ12,
where τ is 1 day (86 400 s), from the previous value of 8 ×10-6 K m2 J-1 recommended by Giard and Bazile (2000) to 1.2 ×10-5 K m2 J-1. The new value for Cv results from updated
values for cv and λ or vegetation based on measurements
of various leaves by Jayakshmy and Philip (2010) (cv =
2.0 ×106 J m-3 K-1, λ=0.5 W m-1 K-1).
These changes reduce overestimations of minimum 2 m temperature (i.e.,
warmer surface temperatures) during the early morning (dawn) hours while also
reducing underestimations of 2 m temperature during the post-dawn hours.
There were also two major revisions made to the ACM2 vertical mixing scheme
in both WRF and CMAQ. In WRF, the ACM2 was updated to estimate and apply
different eddy diffusivities for momentum (Km) and heat
(Kh) so that the Prandtl number (Pr) is no longer
assumed to be unity (Pr =Km/Kh≠1). The second major modification to ACM2 is the implementation of new
stability functions for both heat and momentum for stable conditions, which
allows for more mixing in the stable regimes, particularly moderately stable
conditions that often occur in the early evening hours. CMAQv5.1 has also
been modified to include the same stability functions that are used in WRFv3.7, and therefore, for consistency, WRFv3.7 (or newer) and CMAQv5.1
should be used together. Both of these revisions to the ACM2 are described in
Pleim et al. (2016).
The MOL values used in the ACM2 model in CMAQ were
found to differ from the MOL values used in the ACM2 model in WRF.
Specifically, the output from WRF was for a preliminary estimate of MOL that
was computed in the surface layer model in WRF (module_sf_pxsfclay.F). The
MOL was later re-computed in ACM2 in WRF but not loaded into the output
array. This inconsistency has been fixed in v5.1 by re-computing the MOL in
CMAQ exactly as it is computed in ACM2 in WRF. However, starting with WRFv3.8, this re-computed MOL value will be available in the WRF output, and
therefore it will be unnecessary to re-compute the MOL value in CMAQ.
New and revised SOA species in the CMAQv5.1 AERO6 mechanism.
Aerosol species
Change since v5.0.2
Applicable mechanism
Description of modification
AH3OP
added
all
Hydronium ion (predicted by ISORROPIA for I + J modes); used for IEPOX uptake
APAH1,2
added
cb05e51, saprc07tb, saprc07tc, saprc07tic, racm
Naphthalene aerosol from RO2 + NO reactions
APAH3
added
cb05e51, saprc07tb, saprc07tc, saprc07tic, racm
Naphthalene aerosol from RO2 + HO2 reactions
AISO1,2
updated
cb05e51, saprc07tb, saprc07tc*, racm
Aerosol from isoprene reactions NO3 added to existing OH (all yields follow the OH pathway)
AISO3
updated
cb05e51, saprc07tb, saprc07tc*, racm
Aerosol from reactive uptake of IEPOX on aqueous aerosol particles. Specifically intended to be the sum of 2-methyltetrols and IEPOX-derived organosulfates
AALK1,2
added
cb05e51, saprc07tb, saprc07tc, saprc07tic, racm
Alkane aerosol
AALK
removed
all
Deprecated alkane aerosol
* AERO6i does not include SOA from isoprene + NO3 in
AISO1,2 (it is included in AISOPNNJ). AERO6i does not include IEPOX SOA in
AISO3 (it is included in AITETJ, AIEOSJ, AIDIMJ, etc.). AISO3 is
approximately zero in AERO6i.
Scientific improvements in the CMAQv5.1 aerosol treatment
CMAQ has historically underestimated SOA in both urban (Woody et al., 2016)
and rural (Pye et al., 2015) locations. Thus, improvements to the
representation of aerosol from anthropogenic and biogenic hydrocarbons were
needed. The updates to SOA formed from anthropogenic volatile organic
compounds (VOCs) focus on VOCs in existing emission inventories, such
as the EPA National Emissions Inventory (NEI), that are likely to fall in the
intermediate VOC (IVOC) range. These include long-chain alkanes such as
heptadecane and PAHs such as naphthalene.
Since these compounds are much less volatile than traditional VOCs, they
readily form aerosol in high yields. Long-chain alkanes and PAHs were
included in other VOC categories in CMAQ versions prior to v5.1, but were
lumped with smaller, more-volatile compounds that did not form SOA with the
same efficiency. By separating long-chain alkanes and naphthalene at the
emission processing step, CMAQ can better account for their higher yields.
Several studies (e.g., Pye and Pouliot, 2012; Jathar et al., 2014) have
indicated that a large fraction of VOC emissions, particularly IVOC-type
compounds, may not be characterized in emission inventories, which limits how
much SOA can be formed from anthropogenic VOCs in current CTMs.
Several new SOA species were introduced in v5.1 AERO6, specifically AALK1 and
AALK2 (from long-chain alkanes) and APAH1, APAH2, and APAH3 (from
naphthalene). CMAQv5.1-predicted alkane SOA is responsible for ∼ 20 to
50 % of SOA from anthropogenic VOCs, with the largest absolute
concentrations occurring during summer in urban areas. Naphthalene oxidation
is predicted to produce more modest amounts of SOA (Pye and Pouliot, 2012).
Note that PAH SOA in v5.1 only considers naphthalene as the parent
hydrocarbon, which is about half of the PAHs considered as SOA precursors in
Pye and Pouliot (2012). This approach was used since naphthalene is a high-priority hazardous air pollutant (HAP) and necessary in the model for
purposes other than SOA formation.
CMAQv5.1 has been updated to include the isoprene epoxydiol (IEPOX) SOA
resulting from aqueous reactions for most chemical mechanisms including CB05
and SAPRC07, as described in Pye et al. (2013). Later-generation isoprene
oxidation products formed under low-NOx conditions, specifically IEPOX,
are recognized as a significant source of SOA based on laboratory (Surratt et
al., 2010), field (Hu et al., 2015) and modeling (McNeill et al., 2012; Pye et
al., 2013; Marais et al., 2016) studies. This SOA is linked to sulfate and
acidity and thus represents an anthropogenically controlled source of
biogenic SOA.
In addition to the SOA updates for anthropogenic VOCs, AISO3 (acid-catalyzed isoprene epoxide aerosol) was also revised in CMAQv5.1 to represent
SOA from IEPOX. For the CB05tucl, CB05e51 and SAPRC07 chemical mechanisms
with IEPOX formation in the gas phase, heterogeneous uptake of IEPOX on
acidic aerosol results in SOA (Pye et al., 2013). This IEPOX SOA replaces the
AISO3 treatment based on Carlton et al. (2010). The AISO3J species name
is now retained for IEPOX SOA and represents the sum of IEPOX-derived
organosulfates and 2-methyltetrols. Explicit isoprene SOA species including
2-methyltetrols, 2-methylglyceric acid, organosulfates, and oligomers (e.g.
dimers) are available in the SAPRC07tic with AERO6i mechanism now available in
CMAQv5.1. See Table 1 for more information regarding these new SOA species.
Improvements to the CMAQv5.1 in-line photolysis and cloud model
The in-line calculation of photolysis rates in CMAQ has undergone significant
changes. The calculation of photolysis rates in v5.1 still uses the same
approach for calculating actinic fluxes by solving a two-stream approximation
of the radiative transfer equation (Binkowski et al., 2007; Toon et al.,
1989) over wavebands based on the FAST-J photolysis model (Wild et al.,
2000). Each layer includes scattering and extinction using simulated air
density, cloud condensates, aerosols and trace gaseous such as O3 and
NO2. The first area changed in v5.1 is how clouds are described in the
actinic flux calculation. In v5.0.2, a vertical column had a single cloud
deck with constant cloud fraction, liquid water content and water droplets as
the source of scattering and extinction from clouds. These cloud parameters
were diagnosed from humidity and air temperature predicted by the
meteorological model (e.g., WRF). CMAQv5.1 uses additional information
available from WRF that describes the resolved cloud cover, which allows the
vertical column to have multiple cloud decks with variable cloud fractions
and multiple types of water condensates. In addition to the resolved cloud
cover, v5.1 also includes the radiative effect from CMAQ's subgrid
convective clouds in the calculation of actinic fluxes. CMAQ uses the ACM
cloud model to describe subgrid convective clouds based on convective
precipitation rates from WRF. These updates to the clouds used in the
photolysis rates improved CMAQ's internal consistency between cloud mixing,
aqueous chemistry and gas-phase chemistry.
The second area of change to the in-line photolysis calculation addressed the
radiative effect from aerosols. The mixing model used to compute the
refractive indices of aerosol modes (an internal-volume weighted average
model) allows the refractive index of each aerosol component to depend on
wavelength. Most importantly, the refractive index for elemental (black)
carbon reflects the current scientific consensus (Bond and Bergstrom, 2006;
Chang and Charalampopoulos, 1990; Segelstein, 1981; Hess et al., 1998) and
increases its absorptive capacity from the v5.0.2 value. Additionally,
estimating aerosol optical properties includes new options to solve Mie
scattering theory, or the option to use the core–shell model with an
elemental carbon core (Bohren and Huffman, 2004). A user can choose to use
these options by setting environment variables before executing the CMAQ
model
(http://www.airqualitymodeling.org/).
By default, v5.1 uses approximate solutions to Mie scattering and the
internal-volume weighted average model (Binkowski et al., 2007). Third,
several new variables (e.g., resolved cloud fraction, subgrid cloud fraction,
resolved cloud water content) have been added to the cloud diagnostic file
that describe the optical properties of aerosol and clouds and their
radiative effects.
Improvements in CMAQv5.1 atmospheric chemistry
Several changes were made to the CB05TUCL chemical mechanism in v5.1 (Whitten
et al., 2010; Sarwar et al., 2012), which is now referred to as CB05e51.
These changes include updates to reactions of oxidized nitrogen (NOy)
species, incorporation of new research on the atmospheric reactivity of
isoprene photooxidation products, addition of several high-priority HAPs to
the standard CB05e51 mechanism (following the protocol in the multipollutant
version of CMAQ), and other changes to update the mechanism and make it
compatible with updates to the aerosol chemistry, but overall retaining the
fundamental core of the CB05 mechanism. A more detailed explanation of the
changes made in the CB05e51 mechanism is provided below.
NOy updates and additions
The most extensive changes made consisted of updates and extensions of the
NOy species, including peroxyacyl nitrates, alkyl nitrates, and NOx
reactions with HOx. The thermal formation and degradation of
peroxyacetyl nitrate (PAN) were modified to correct the parameters that
describe the rate constant pressure dependence in the fall-off region between
the high-pressure limit and the low-pressure limit based on the values
determined by Bridier et al. (1991). An additional species, MAPAN, was added
to explicitly represent PANs from methacrolein because these are a possible
contributor to SOA formation. The OH + NO2 reaction rate was updated
based on Troe (2012), and a small yield of HNO3 (< 1 % at standard
temperature and pressure, varying with temperature and pressure) was added to
the reaction of HO2 + NO (Butkovskaya et al., 2007). The single
alkyl nitrate species in CB05, NTR, was replaced with seven species to better
investigate the variety of chemical and physical fates of alkyl nitrates. The
first-generation monofunctional alkyl nitrates and difunctional hydroxy
nitrates were assigned Henry's law constants of 6.5×10-1 and
6.5×103 M, respectively, while second-generation carbonyl nitrates
were assigned 1.0×103 M and multifunctional hydroxy nitrates were
assigned a value of 1.7×104 M. Five species are predominantly from
anthropogenic sources, with the relative distribution of mono-functional
(alkyl nitrates) and multifunctional (hydroxy, carbonyl, hydroxycarbonyl,
and hydroperoxy) nitrate products determined based on the nitrates produced
from the five alkanes and alkenes, with the largest emissions as listed in
the NEI (Simon et al., 2010). The other two nitrate species represent first-generation and later-generation nitrates from biogenic (isoprene and terpene)
sources. Biogenic nitrate products were based on reaction products from Lee
et al. (2014), with NOx recycling from secondary biogenic nitrate
products (Jenkin et al., 2015) and photolysis rates with quantum yields of
unity. Finally, a heterogeneous hydrolysis rate of alkyl nitrates was added
(Hildebrandt-Ruiz et al., 2013), with a 6 h lifetime on aerosol at high
relative humidity (Liu et al., 2012; Rollins et al., 2013). Additional
details can be found in the CMAQv5.1 release documentation
(http://www.airqualitymodeling.org/).
Other changes
The high HOx pathways for isoprene oxidation have been modified to
explicitly account for production of IEPOX, which can form SOA and modify the
gas-phase concentrations. The high-NOx pathways have been modified to
explicitly produce methacrolein PAN (MAPAN, described in Sect. 2.4.1)
because it reacts faster with OH than other PAN species. Several high-priority HAPs were added to the standard version of CB05e51 as either active
species or reactive tracers, specifically acrolein, 1,3-butadiene (which
produces acrolein), toluene, xylene isomers, α- and β-pinene,
and naphthalene, using reaction pathways and rates as defined by IUPAC. Refer
to the CMAQv5.1 release documentation for additional details on these
updates.
Several other, smaller changes were made to the chemistry to either improve
consistency with IUPAC, enhance the integration with heterogeneous chemistry,
or for numerical consistency. These include the following: the updates to the products of
ethanol reaction with OH using recommended yields from IUPAC
(http://iupac.pole-ether.fr; accessed 11 May 2016); updates to the reactions
of acylperoxy radicals with HO2 to include a 44 % yield of OH; the
addition of a new species, SOAALK, to account for SOA formation from alkanes;
and the addition of gas-phase and heterogeneous nitryl chloride formation
(ClNO2) and CINO2 photolysis as described by Sarwar et al. (2012).
Updates to air–surface exchange processes in CMAQv5.1
Meteorologically dependent emissions and deposition, hereafter referred to as
air–surface exchange, were extensively updated in v5.1. A data module was
developed to share meteorological and calculated atmospheric transport
environmental variables between vertical diffusion, deposition and
meteorologically dependent emissions to more consistently represent processes
common to both deposition and emissions. Additionally, sea-salt and biogenic
emissions as well as dry deposition routines were updated.
Sea-salt aerosol emission
The sea-salt aerosol emissions module was updated to better reflect emission
estimates from recent field observations and to incorporate ocean
thermodynamic impacts on emissions. The size distribution of sea-salt aerosol
was expanded to better reflect recent fine-scale aerosol measurements in
laboratory and field studies (de Leeuw et al., 2011) by modifying the O
parameter of Gong (2003) from 30 to 8. A sea-surface temperature (SST)
dependency to the sea-salt aerosol emissions following Jaeglé et al. (2011)
and Ovadnevaite et al. (2014) was also added, which increased
accumulation and coarse-mode sea-salt emissions in regions with high SSTs and
reduced the emissions in regions with low SSTs. Finally, the surf-zone
emissions of sea-salt aerosol were reduced by 50 %, assuming a decrease in
the surf-zone width from 50 to 25 m to address a systematic overestimation
of near-shore coarse sea-salt aerosol concentrations (Gantt et al., 2015).
Biogenic emissions
There were also several updates to the calculation of non-methane biogenic
volatile organic carbon (BVOC) emissions in v5.1. The Biogenic Emissions
Inventory System (BEIS;
https://www.epa.gov/)
model was updated to include the implementation of a dynamic two-layer, sun
and shaded, vegetation canopy model, while the PAR response function was
integrated into the canopy model following Niinemets et al. (2010) for each
canopy layer. In earlier versions of BEIS, emissions were a function of the
2 m temperature which was inconsistent with measured emission factors that
were empirically correlated with leaf temperature. BEIS 3.6.1, released with
v5.1, was updated to model emissions as a function of the leaf temperature
rather than 2 m temperature to be more consistent with how BVOC emission
factors are typically estimated. For additional details see Bash et
al. (2016). Finally, the Biogenic Emission Land-use Data (BELD) version 4.0
and emission factors for herbaceous wetlands were updated to address
overestimates of BVOCs at coastal sites (Guenther et al., 2006), and the BELD
land-use and vegetation species were updated using high-resolution satellite
data and in situ survey observations from 2002 to 2012 (Bash et al., 2016).
Dry deposition
There were two important updates to the dry deposition calculation in v5.1.
First, the dry deposition of O3 over oceans was updated to include the
additional sink due to interaction with iodide in the seawater (marine
halogen chemistry), with the iodide concentrations estimated based on
sea-surface temperature (Sarwar et al., 2015), which increased the O3
deposition velocity over oceans. Second, over vegetative surfaces, the wet
cuticular resistance was updated following Altimir et al. (2006) (385 s m-1),
and dry cuticular resistance was set to the value of Wesley (1989)
for lush vegetation (2000 s m-1). These changes resulted in an
approximately 2.0 ppbv reduction in the modeled O3 mixing ratios, with
the largest reductions, ∼ 10 %, occurring during the nighttime and
early morning hours, and approximately a 2 % reduction in the modeled
midday O3 mixing ratio. It was later discovered (after the release of
v5.1) that the 385 s m-1 value represents a canopy resistance rather
than a leaf resistance, and therefore should be closer to a value of 1350 s m-1
following Altimer et al. (2006). The value will be corrected in the
next CMAQ model release.
Gravitational settling
Previous evaluations of the ground-level coarse particle (PM10–PM2.5)
concentrations in CMAQ have shown that the model significantly
underestimated the total PM10 concentrations (Appel et al., 2012).
Contributing to this underestimation is the fact that CMAQ previously did not
have a mechanism in place to allow coarse particles to settle from upper
layers to lower layers (although coarse particles in layer one can settle to
the surface). As a result, large particles that would normally settle to
lower layers in the model could remain trapped in the layers in which they
were emitted or formed. To account for this deficiency in the model, the
effects of gravitational settling of coarse aerosols from upper to lower
layers have been added to v5.1 to more realistically simulate the aerosol mass
distribution. The net effect of this update is an increase in ground-level
PM10 concentrations in v5.1 compared to v5.0.2, particularly near
coastal areas impacted by sea spray (Nolte et al., 2015).
As stated in the beginning of this section, but is useful to reiterate here,
the information provided in this section only covers a portion of the vast
number of updates that went into v5.1, and was intended to make the reader
aware of the more significant changes made and why, but often avoids
including the very specific detailed code changes that were made to the
model. Those seeking a complete detailed list of all the changes made to the
model should refer to the v5.1 technical documentation using the link
provided at the beginning of this section.
Modeling setup and observational data sets
The modeling setup for the evaluation of v5.1 utilizes a domain covering the
entire contiguous United States (CONUS) and surrounding portions of northern
Mexico and southern Canada, as well as the eastern Pacific and western Atlantic
oceans. The modeling domain consists of 299 north–south by 459 east–west grid
cells utilizing 12 km × 12 km horizontal grid spacing, 35 vertical
layers with varying thickness extending from the surface to 50 hPa and an
approximately 10 m midpoint for the lowest (surface) model layer. The
simulation time period covers the year 2011, which is a base year for the
EPA's NEI and also a period during which specialized measurements from a
variety of trace species are available from the Deriving Information on
Surface Conditions from Column and Vertically Resolved Observations Relevant
to Air Quality (DISCOVER-AQ;
http://www.nasa.gov/)
campaign.
All the CMAQ simulations presented here employed the Euler backward iterative
(EBI) solver. The v5.0.2 simulations utilized the windblown dust treatment
available, while the v5.1 simulations did not due to errors in the
implementation of the windblown dust model in v5.1. However, the overall
contribution of windblown dust to PM2.5 is small on a seasonal average
and does not affect the seasonal comparisons shown in Sect. 5. Additional
details regarding the options employed in the CMAQ simulations are available
upon request from the corresponding author. For the annual simulations, a
10-day spin-up period in December 2010 was used (and then discarded) to
reduce the effects of the initial conditions, after which the model was run
continuously for the entire year 2011 (one continuous simulation stream). For
the 1-month January and July sensitivity simulations presented, 10-day
spin-up periods in the previous month were used and then discarded. Boundary
conditions for the 12 km CMAQ simulations are provided by a 2011 hemispheric
GEOS-Chem (Bey et al., 2001) with the chemical species mapped to the
corresponding CMAQ species.
Several sets of CMAQ simulations were performed to help thoroughly evaluate
both the overall change in model performance between v5.0.2 and v5.1 and to
examine the individual impact of specific model process changes on the model
performance. As such, different input data sets were used for the
v5.0.2 and v5.1 simulations. The base v5.0.2 simulation (CMAQv5.0.2_Base)
utilized WRFv3.4 meteorological input data, while WRFv3.7-derived
meteorological data were used for all the v5.1 simulations presented here. As
stated previously, different versions of WRF were used for the v5.0.2 and
v5.1 simulations due to the updates made in both WRF and CMAQ (Sect. 2.1)
that would have made performing the CMAQ simulations with output from the
same version of WRF difficult and introduced some inconsistencies. While there
were other updates made to WRF between versions 3.4 and 3.7, those changes
were minor and did not impact the WRF results significantly for the
configuration of the model used here.
Both WRF simulations employed the same options, which include the Rapid
Radiation Transfer Model Global (RRTMG) long-wave and shortwave radiation (Iacono
et al., 2008), Morrison microphysics (Morrison et al., 2005), and the
Kain–Fritsch version 2 cumulus parametrization (Kain, 2004). For the LSM and PBL
models, the PX-LSM and ACM2 were used. Four-dimensional data assimilation
(FDDA) was also employed in the WRF simulations. The name lists used for each
WRF simulation are provided in the Supplement (see Sects. 4
and 5). Model-ready meteorological input files were created using version
4.1.3 of the meteorology–chemistry interface processor (MCIP; Otte and Pleim,
2010) for the WRFv3.4 data and MCIP version 4.2
(https://www.cmascenter.org/)
for the WRFv3.7 data.
Description of the CMAQ model simulations utilized.
CMAQ simulation name
CMAQ
WRF
NEI
Photolysis
Chemical
Simulation period
version
version
version
scheme
mechanism
(all 2011)
CMAQv5.0.2_Base
v5.0.2
v3.4
v1
v5.0.2
CB05TULC
Annual
CMAQv5.0.2_WRFv3.7
v5.0.2
v3.7
v1
v5.0.2
CB05TUCL
January and July
CMAQv5.1_Base_NEIv1
v5.1
v3.7
v1
v5.1
CB05e51
Annual
CMAQv5.1_Base_NEIv2
v5.1
v3.7
v2
v5.1
CB05e51
Annual
CMAQv5.1_Retrophot
v5.1
v3.7
v2
v5.0.2
CB05e51
January and July
CMAQv5.1_TUCL
v5.1
v3.7
v2
v5.1
CB05e51
January and July
Two sets of emission input data were utilized for the analysis presented
here. Both sets of emission data were based on the 2011 NEI, with version 1
(v1) of the 2011 NEI modeling platform developed by the USEPA from regulatory
applications
(https://www.epa.gov/)
utilized for the majority of the simulations, while version 2 (v2) of the
2011 modeling platform was utilized for one set of sensitivity simulations.
However, all the comparisons of model simulations presented here are shown
with simulations that utilized the exact same emissions inventory, and as
such any differences in model performance are not the result of differences
in emissions. See Table 2 for information regarding which version of the
emission inventory was utilized for each simulation.
The raw emission files were processed using versions 3.5 (v1 emissions) and
3.6.5 (v2 emissions) of the Sparse Matrix Operator Kernel Emissions (SMOKE;
https://www.cmascenter.org/smoke/) program to create gridded speciated hourly
model-ready input emission fields for input to CMAQ. Electric generating unit
(EGU) emissions were obtained using data from EGUs equipped with a continuous
emission monitoring system (CEMS). Plume rise for point and fire sources were
calculated in-line for all simulations (Foley et al., 2010;
https://www.cmascenter.org/).
Biogenic emissions were generated in-line in CMAQ using BEIS versions 3.14
for v5.0.2 and 3.61 (Bash et al., 2016) for v5.1. All the simulations
employed the bidirectional (bi-di) ammonia flux option for estimating the
air–surface exchange of ammonia, as well as the in-line estimation of
NOx emissions from lightning strikes.
Output from the various CMAQ simulations is paired in space and time with
observed data using the atmospheric model evaluation tool (AMET; Appel et
al., 2011). There are several regional and national networks that provide
routine observations of gas and particle species in the US. The national
networks include the EPA's Air Quality System (AQS; 2086 sites;
https://www.epa.gov/aqs) for hourly and daily gas and aerosol PM
species; the Interagency Monitoring of Protected Visual Environments
(IMPROVE; 157 sites; http://vista.cira.colostate.edu/improve/) and
Chemical Speciation Network (CSN; 171 sites;
https://www3.epa.gov/ttnamti1/speciepg.html) for daily average
(measurements typically made every third or sixth day) total and speciated
aerosol PM species; and the Clean Air Status and Trends Network (CASTNET; 82
sites; http://www.epa.gov/castnet/) for hourly O3 and weekly
aerosol PM species. In addition to these routinely available observations,
the DISCOVER-AQ campaign
(https://www.nasa.gov/mission_pages/discover-aq/) during July 2011
provides additional ground-based gas and aerosol PM measurements, along with
unique aloft measurements made by aircraft, vertical profilers (e.g., light
detection and ranging (lidar) measurements), ozonesondes and tethered
balloons (not utilized in this analysis, however).
Evaluation of major scientific improvements
In this section we evaluate the impact that several of the major scientific
improvements in v5.1 have on the operational model performance. Unlike Foley
et al. (2010), in which several individual major scientific improvements in
CMAQ v4.7 were evaluated incrementally (e.g., each subsequent improvement is
evaluated against the previous improvement), here we examine each scientific
improvement separately by comparing simulations with the specific improvement
removed (i.e., as it was in v5.0.2) to the base v5.1 simulation
(CMAQv5.1_Base_NEIv1) which includes all the updates. While this has the
disadvantage of not showing the incremental change in model performance due
to each improvement, it does limit the number of simulations that need to be
performed. In addition, it allows for easier examination of the effect of
nonlinear increments on total model performance, as some updates to the
modeling system may be affected by updates to other parts of the model, the
effects of which on model performance may not be captured in an incremental
testing format. Note that while some attempt is made to broadly identify the
processes involved that cause the observed changes in model performance
between v5.0.2 and v5.1, it would be too laborious (both to the reader and to
the investigators) to comprehensively describe and investigate in depth the
processes involved that result in each observed difference in model
performance described in this section. Where appropriate, the analyses
presented in this section use the v5.0.2 base simulation (CMAQv5.0.2_Base)
for comparison to the scientific improvement while for other improvements the
v5.1 base simulation is used for comparison. In each case, the simulations
being compared are noted. Table 2 provides a description of the CMAQ model
simulations referred to in the following sections.
Monthly average difference in O3 (ppbv) for
(a) January and (b) July and PM2.5
(µg m-3) for (c) January and (d) July
between CMAQv5.0.2 using WRFv3.4 (CMAQv5.0.2_Base) and CMAQv5.0.2 using
WRFv3.7 (CMAQv5.0.2_WRFv3.7) (CMAQv5.0.2_WRFv3.7–CMAQv5.0.2_Base). Note
that the scales between each plot may vary.
WRF and CMAQ meteorological updates
As discussed in Sect. 2.1, there were several significant
corrections and improvements made to the meteorological calculations in both WRF
and CMAQ. While the focus of this work is on updates to the CMAQ model,
certain options within WRF and CMAQ are linked, and therefore it is necessary
to discuss the WRF model updates alongside the corresponding CMAQ model
updates.
Figure 1 shows the cumulative impact that all the meteorological changes in
WRF and CMAQ (i.e., changes to ACM2 and MOL) had on O3 and PM2.5 in
January and July by comparing the CMAQv5.0.2_Base simulation to a CMAQv5.0.2
simulation using WRFv3.7 (CMAQv5.0.2_WRFv3.7) which includes the ACM2 and
MOL updates. The effect of the changes on O3 in January is mixed, with
some areas (e.g., Florida, Chicago and the northwest) showing a relatively
large (2.5 ppbv) increase in O3, while other areas (e.g., the southwest and
Texas panhandle) show a relatively large decrease (-2.5 ppbv) in O3.
For PM2.5, the differences in January are generally small and isolated;
however, there is a relatively large increase in PM2.5 (> 2.5 µg m-3)
in the San Joaquin Valley (SJV) of California due to
the updates, which, combined with the decrease in O3 there as well,
indicates a likely reduction in PBL height and mixing as the cause. There are
also some relatively large decreases (1.5–2.0 µg m-3) in
PM2.5 in the northeast and around in the Great Lakes region (i.e.,
Chicago). Otherwise, most of the remaining impacts on PM2.5 are
relatively small (< 1.0 µg m-3).
For July, the meteorological updates in WRF and CMAQ result in exclusively
increased O3 mixing ratios over land, which are considerably larger than
the impacts observed in January. The largest increases (4.0–10.0 ppbv)
occur in the eastern US, particularly in the southeast. Smaller increases
of 2.0–4.0 ppbv occur across much of the US, while in the Gulf of Mexico
and the Caribbean O3 mixing ratios decrease roughly 2.0–6.0 ppbv
across a large area. The difference in PM2.5 in July is similar to that
in January, with mostly small, isolated increases or decreases occurring in
the eastern US. The largest increase (2.0–2.5 µg m-3)
occurs in the southern Ohio Valley (Kentucky and West Virginia), while the
largest decreases (> 2.5 µg m-3) occur in Louisiana and
Texas (i.e., Houston).
It makes intuitive sense to see summertime O3 mixing ratios increasing
due to the meteorological changes in WRF and CMAQ, since the net effect of
those changes was to increase mixing, particularly in the late afternoon and
early evening, which in turn decreases the amount of NO titration of O3
that occurs in the model, and ultimately results in higher O3 mixing
ratios on average. Conversely, PM2.5 concentrations would be expected to
decrease due to the increased mixing in the model, which would effectively
decrease the concentrations of primary emitted pollutants (e.g., EC and organic carbon (OC)),
which was observed in areas with the largest emissions (i.e., urban areas). In
addition, changes in the oxidant (i.e., OH) concentrations would also
potentially affect PM2.5 concentrations through increased or decreased
SOA formation (spatial heterogeneity of PM2.5 formation), which results
in spatially varying increases and decreases in PM2.5 concentrations.
Monthly average sum total of AALK1, AALK2, APAH1, APAH2 and APAH3
for (a) January and (b) July (upper right) and the monthly
average difference is the sum total of AISO1, AISO2, AISO3 and AOLGB for
(c) January and (d) July between the aerosol treatments in
CMAQ v5.0.2 and v5.1 (v5.1–v5.0.2). All plots are in units of micrograms per
cubic meter (µg m-3). Note that the scales between each plot may vary.
Aerosol updates
Several new SOA species from anthropogenic VOCs (i.e., AALK1, AALK2, APAH1,
APAH2 and APAH3; Table 1) were added to AERO6 in v5.1 that were not present
in v5.0.2. Figure 3 shows the difference in the monthly average sum total
concentration of these five species for January and July 2011 between the
CMAQv5.0.2_Base and CMAQv5.1_Base simulations. Since none of these species
were present in v5.0.2, the difference totals in Fig. 2 represent the
additional SOA mass that these five species contribute to the total
PM2.5 mass in v5.1. For both January and July, the monthly average
concentration of these species is small, ranging between
0.0 and 0.1 µg m-3, with the largest concentrations in the
eastern half of the US, particularly in the upper Midwest. However, the
concentration of these new species during shorter time periods and smaller,
isolated regions would be larger. In addition, the inclusion of these new
species is potentially important for health-related studies on the impact of
PAHs. Overall, however, these new species represent a small addition to the
total PM2.5 concentration in the model.
Along with the introduction of the new SOA species above, the pathways for
the formation of acid-enhanced isoprene SOA were also updated. The bottom
panels in Fig. 2 show the monthly average difference in the sum of the
species containing isoprene SOA (AISO1, AISO2, AISO3 and AOLGB) between v5.1
and v5.0.2. For January, the difference in the sum of these
species is relatively small, with minimum and maximum values peaking around
±0.5 µg m-3, consistent with the fact that isoprene
emissions are low in winter. For July the difference is always positive (v5.1
higher than v5.0.2) and much larger compared to January, with peak
differences exceeding 2.5 µg m-3, primarily in the areas with
the highest aerosol SO42- concentrations (i.e., Ohio Valley).
Therefore, the updated IEPOX-SOA formation pathways in v5.1 represent a
potentially significant contribution to the total PM2.5, particularly
during the summer. Increased isoprene emissions in v5.1 with BEISv3.61
compared to v5.0.2 with BEISv3.14 also contribute to the larger contribution
of isoprene SOA in v5.1.
Cloud model and in-line photolysis updates
Changes in the photolysis and cloud model treatment in v5.1 have potentially
significant impacts on the O3 and PM2.5 estimates from the model.
Figure 3 shows the difference in O3 and PM2.5 for the
CMAQv5.1_Base simulation and the CMAQv5.1_RetroPhot simulation (see Table 2
for simulation description). The CMAQv5.1_RetroPhot simulation is the same
as the CMAQv5.1_Base simulation except it employs the same (old)
photolysis and cloud model treatment as in v5.0.2. For January, O3 mixing
ratios (Fig. 3a) and PM2.5 concentrations (Fig. 3c) are both higher
across the southeast and portions of California in the v5.1 simulation,
indicating that v5.1 has much less photolysis attenuation due to the updates
in the representation of cloud effects on photolysis.
Difference in the monthly average O3 for (a) January
and (b) July and PM2.5 for (c) January and
(d) July between CMAQv5.1_base and v5.1_RetroPhot (v5.1_Base -
v5.1_RetroPhot). O3 plots are in units of parts per billion by volume (ppbv) and PM2.5 plots
are in units of micrograms per cubic meter (µg m-3). Note that the scales between each plot
may vary.
The impact of the updated photolysis in v5.1 is considerably larger in July
(when there is more convection) than in January. Peak O3 differences in
January were around 2.0 ppbv, whereas in July peak differences of greater
than 5.0 ppbv (Fig. 3b) occur over the Great Lakes (where low PBL heights
can enhance the impact of changes in O3). However, in general the
difference in O3 mixing ratios is larger in both magnitude and spatial
coverage in July compared to January, indicating that the updated
photolysis and cloud model treatment in v5.1 increases O3 to a greater
extent in July compared to January, as expected due to increased photolysis
rates in the summer compared to winter. Overall, differences in O3 in
July range on average from 1.0 to 3.0 ppbv, with larger differences
occurring in the major urban areas (e.g., Atlanta, Charlotte and Los Angeles)
and off the coast of the northeast corridor. The change in PM2.5 is also
larger (both in magnitude and spatial coverage) in July than January
(Fig. 3d). The greatest change is primarily confined to the eastern US,
resulting in a roughly 0.1 to 0.5 µg m-3 increase in
PM2.5 in v5.1, with the maximum increase located over the Great Lakes
region and areas to the south, the result of increased SOA and gas-phase
production of SO42- due to greater OH- concentrations in v5.1.
Additional diagnostic evaluation of photolysis and cloud model treatment in CMAQ
was conducted based on the model-predicted cloud albedo at the top of the
atmosphere. The predicted cloud albedo from WRFv3.7, CMAQv5.0.2 and CMAQv5.1
were evaluated against cloud albedo from NASA's Geostationary Operational
Environmental Satellite (GOES) Imager product. This evaluation was used to
qualitatively determine whether one CMAQ version better considers how clouds
affect calculated photolysis rates. The GOES product has a 4 km horizontal
resolution and was re-gridded to the 12 km grid structure used in the WRF
and CMAQ simulations using the Spatial Allocator utility
(https://www.cmascenter.org/sa-tools/). The satellite data are
available at 15 min prior to the top of the hour during daytime hours and
were matched to model output at the top of the hour. There were 301 h with
available satellite data across the domain in July 2011. Figure 4 shows the
average cloud albedo (i.e., reflectivity at the top of the atmosphere) during
these 301 h in July derived from the GOES 35 satellite product (Fig. 4a),
and the cloud parameterizations within: WRF3.7 (Fig. 4b), CMAQv5.1_RetroPhot
(Fig. 4c) and CMAQv5.1_Base_NEIv2 (Fig. 4d). Comparison of Fig. 4b and c
shows the dramatic differences between the clouds predicted by WRFv3.7 and
the predictions from the cloud parameterization in v5.0.2. Most of these
large differences, particularly over land, are now gone in model predictions
from the CMAQv5.1_Base simulation, which uses resolved clouds from WRF and
subgrid clouds from the convective cloud model within CMAQ (compare Fig. 4b
to d).
The average cloud albedo during daytime hours in July 2011 with
available satellite data (n=301 h total) derived from (a) the
GOES satellite product, (b) WRF3.7, (c) CMAQv5.1 with
photolysis and cloud model treatment from v5.0.2 and WRF3.7 inputs
(CMAQv5.1_RetroPhot), and (d) CMAQv5.1 using WRF3.7 inputs
(CMAQv5.1_Base_NEIv2).
Two notable issues remain with the v5.1 modeled cloud parametrization. The
photolysis cloud parameterization in v5.1 produces more clouds over water
compared to the WRF parameterization, which is itself biased high for some
parts of the Atlantic Ocean compared to GOES. This issue will be addressed by
science updates planned for the CMAQ system, and evaluation results are
expected to improve in the next CMAQ release. A more significant issue, from
an air quality perspective, is the underprediction of clouds over much of
the eastern and west-central US in the WRF-predicted clouds, which is now
directly passed along to CMAQ. This misclassification of modeled clear sky
conditions can contribute to an overprediction of O3 in these regions.
Resolving this issue will require changes to the WRF cloud parameterization.
Future research will also include changing the subgrid cloud treatment
currently used in the CMAQ system to be consistent with the subgrid
parameterization used in WRF. Section S1 in the Supplement
provides a table with additional evaluation metrics of the modeled clouds
over oceans vs. over land and also describes how cloud albedo was calculated
for the three model simulations.
Atmospheric chemistry updates
As detailed in Sect. 2.4, numerous updates were implemented in the
representation of atmospheric chemistry in v5.1. It would be extremely
cumbersome to attempt to isolate the impact of each chemistry update
individually. Instead, in order to assess the overall impact that the
combined chemistry changes have on the model results, model comparisons are
conducted using the CMAQv5.1_Base simulation, which employs the CB05e51
chemical mechanism (the v5.1 default chemical mechanism) and the
CMAQv5.1_TUCL simulation (see Table 2 for description). The CMAQv5.1_TUCL
simulation is the same as the CMAQv5.1_Base simulation except that it
employs the CB05TUCL chemical mechanism (Whitten et al., 2010; Sarwar et al.,
2012), the default mechanism in v5.0.2. Note that the aerosol updates
discussed in Sect. 4.2 were incorporated into the CB05e51 chemical mechanism
(in the past that portion of the aerosol chemistry was separate from the
gas-phase chemical mechanism). As such, differences between the
CMAQv5.1_TUCL and CMAQv5.1_Base_NEIv2 simulations include impacts from
those changes (i.e., Fig. 2). In order to isolate primarily just the effect on
PM2.5 from the atmospheric chemistry changes, the organic matter (AOMIJ;
see Sect. 2 and 3 for species definition descriptions) mass has been removed
from the comparisons of total PM2.5 mass discussed below.
Difference in the monthly average O3 for (a) January
and (b) July and PM2.5 (with organic matter mass removed) for
(c) January and (d) July between CMAQv5.1_Base_NEIv2 and
v5.1_TUCL (CMAQv5.1_Base_NEIv2–CMAQv5.1_TUCL). O3 plots are in
units of parts per billion by volume (ppbv) and PM2.5 plots are in units of micrograms per cubic meter (µg m-3).
Note that the scales for each plot can vary.
Figure 5 shows the difference in monthly average O3 and PM2.5 for
January and July between the CMAQv5.1_Base_NEIv2 and CMAQv5.1_TUCL
simulations. For January, O3 mixing ratios are higher in the simulation
using the CB05e51 mechanism (CMAQv5.1_Base_NEIv2 simulation); however, the
overall impact of CB05e51 on O3 is generally small (∼ 2–4 %),
with maximum differences of only approximately 1.0 ppbv (∼ 6 %),
primarily along the southern coastal areas of the US. PM2.5 is also
higher in January in the simulation using the CB05e51 mechanism
(CMAQv5.1_Base_NEIv2 simulation), with the largest changes in PM2.5 of
0.2–0.4 µg m-3 (∼ 2–6 %) primarily occurring in
the eastern US and greater than 1.0 µg m-3
(∼ 6–8 %) in the SJV of California.
For July, O3 mixing ratios are higher across most areas in the
CMAQv5.1_Base_NEIv2 simulation, primarily across northern portions of the
US, the Great Lakes region and in California (i.e., Los Angeles and the SJV).
Most increases in O3 in the CMAQv5.1_Base simulation range between 0.6
and 1.2 ppbv (∼ 2–4 %); however, larger increases of over
3.0 ppbv (∼ 4–8 %) occur in southern California and over Lake
Michigan (likely influenced in part by low PBL heights over the lake). A
small area of lower O3 mixing ratios occurs off the eastern coast of the
US. For July, the difference in PM2.5 due to the CB05e51 chemical
mechanism is relatively small, with differences in concentrations generally
ranging from ±0.50 µg m-3 (∼ 2–4 %) across the
eastern US.
Evaluation of CMAQv5.1
In this section, comparisons are made of the performance of the
CMAQv5.0.2_Base and CMAQv5.1_Base_NEIv1 simulations by initially comparing
the simulations to each other (model to model) and then evaluating them
against a wide variety of available air quality measurements (see Sect. 3).
Several common measurements of statistical performance are used, namely mean
bias (MB), mean error (ME), root mean square error (RMSE) and Pearson
correlation. Note that representativeness (incommensurability) issues are
present whenever gridded values from a deterministic model such as CMAQ are
compared to observed data at a particular point in time and space, as
deterministic models calculate the average outcome over a grid for a certain
set of given conditions, while the stochastic component (e.g., subgrid
variations) embedded within the observations cannot be accounted for in the
model (Swall and Foley, 2009). These issues are somewhat mitigated for
networks that observe for longer durations, for example the CSN and IMPROVE
networks, which are daily averages, and the CASTNET observations, which are
weekly averages. The longer temporal averaging helps reduce the impact of
stochastic processes, which can have a large impact on shorter (e.g., hourly)
periods of observation (Appel et al., 2008).
Difference in the seasonal average PM2.5 for
(a) winter (DJF), (b) spring (MAM), (c) summer
(JJA) and (d) fall (SON) between CMAQv5.0.2_Base and
CMAQv5.1_Base_NEIv1 (CMAQv5.1_Base_NEIv1–CMAQv5.0.2_Base). All plots
are in units of micrograms per cubic meter (µg m-3).
There are a couple of important differences to keep in mind between the
comparison of the CMAQv5.0.2_Base and CMAQv5.1_Base_NEIv1 simulations
beyond the obvious changes to the model process representations discussed in
the previous sections. First, the simulations use different versions of WRF
(as discussed in Sects. 2.2 and 4.1). This was intentional, as it was
determined that the changes made from WRFv3.4 (used in the CMAQv5.0.2_Base
simulation) to WRFv3.7 (used in the CMAQv5.1_Base_NEIv1 simulation) and
subsequent required changes made to the CMAQ code represent a change to the
overall WRF–CMAQ modeling system and therefore should be evaluated together.
It should also be noted that the windblown dust treatment was employed in the
CMAQv5.0.2_Base simulation but not in the CMAQv5.1_Base_NEIv1 simulation.
This was due to issues with the implementation of the updated windblown dust
treatment in v5.1 that were not discovered until after the model was released
and the CMAQv5.0.2_Base simulation was completed. However, the contribution
of windblown dust to total PM2.5 in v5.0.2 tends to be small and
episodic and therefore should not constitute a significant impact to the
performance differences between v5.0.2 and v5.1, especially for the monthly
averages generally shown here. However, we make an attempt to note when and
where the impact from windblown dust is apparent. For reference, the v5.0.2-simulated seasonal average values of PM2.5 and maximum daily 8 h
average (MDA8) O3 are provided in Figs. S1 and S2 in the Supplement,
respectively.
PM2.5
Figure 6 shows the seasonal average difference in model-simulated PM2.5
between v5.0.2 and v5.1 (CMAQv5.1_Base_NEIv1–CMAQv5.0.2_Base), with cool
colors indicating a decrease in PM2.5 in v5.1 (vs. v5.0.2) and warm
colors indicating an increase in PM2.5. Figure 7 shows the seasonal MB for PM2.5 for the CMAQv5.1_Base_NEIv1 simulation, while
Figure 8 shows the change in the absolute value of the seasonal MB in PM2.5 between the CMAQv5.0.2_Base and CMAQv5.1_Base_NEIv1
simulations. Cool colors indicate smaller PM2.5 MB in the
CMAQv5.1_Base_NEIv1 simulation (vs. the CMAQv5.0.2_Base simulation), while
warm colors indicate larger MB in the CMAQv5.1_Base_NEIv1 simulation.
Seasonal average PM2.5 mean bias (µg m-3) at
IMPROVE (circles), CSN (triangles), AQS Hourly (squares) and AQS Daily
(diamonds) sites for (a) winter (DJF), (b) spring (MAM),
(c) summer (JJA) and (d) fall (SON) for the CMAQv5.1_Base simulation.
Difference in the absolute value of seasonal average PM2.5 mean
bias for (a) winter (DJF), (b) spring (MAM),
(c) summer (JJA) and (d) fall (SON) between CMAQ
v5.0.2_Base and v5.1_Base_NEIv1 (CMAQv5.1_Base_NEIv1–CMAQv5.0.2_Base).
All plots are in units of micrograms per cubic meter (µg m-3). Cool colors indicate a
reduction in PM2.5 mean bias in v5.1, while warm colors indicate an
increase in PM2.5 mean bias v5.1.
During winter, v5.1 simulates lower PM2.5 concentrations in the eastern
US and portions of central Canada compared to v5.0.2, and higher PM2.5
concentrations in the SJV (Fig. 6). PM2.5 is largely overestimated in
the eastern US and underestimated in the western US (the exception being portions
of the northwest) in the winter in the CMAQv5.1_Base_NEIv1 simulation
(Fig. 7a). The change in MB between v5.0.2 and v5.1 is negative (reduced MB
in v5.1) across the majority of the sites, with relatively large reductions
(3–5 µg m-3) in MB in the northeast, upper Midwest (i.e.,
Great Lakes region) and the SJV (Fig. 8a). Figure S3 presents a histogram of
the change in PM2.5 MB using the same data and color scale as in Fig. 8.
It is clear from the histogram that there is a large percentage (72.3 %) of sites
where the MB decreases in the CMAQv5.1_Base_NEIv1 simulation in the winter
(Fig. S3a), demonstrating a widespread improvement in the PM2.5
performance for v5.1 vs. v5.0.2.
Diurnal time series of seasonal PM2.5 (µg m-3)
for AQS observations (gray), CMAQv5.0.2_Base simulation (blue) and
CMAQv5.1_Base_NEIv1 simulation (red) for (a) winter,
(b) spring, (c) summer and (d) fall.
The diurnal profile of PM2.5 for winter (Fig. 9a) indicates a relatively
large decrease in MB throughout most of the day with v5.1 vs. v5.0.2,
particularly during the overnight, morning and late afternoon hours. A
similar improvement is seen in the RMSE, and the correlation also improves
for all hours (Fig. S5). Figure 10 shows seasonal and regional stacked bar
plots of PM2.5 composition (SO42-, NO3-, NH4+,
EC, OC, soil, NaCl, NCOM and Other). Soil is based on the IMPROVE soil
equation and contains both primary and secondary sources of soil (Appel et
al., 2013), while Other represents the unspeciated PM mass in the inventory
(see Appel et al., 2008) The five regions shown in Fig. 10 are the northeast
(Maine, New Hampshire, Vermont, Massachusetts, New York, New Jersey,
Maryland, Delaware, Connecticut, Rhode Island, Pennsylvania, District of
Columbia, Virginia and West Virginia), Great Lakes (Ohio, Michigan, Indiana,
Illinois and Wisconsin), Atlantic (North Carolina, South Carolina, Georgia
and Florida), south (Kentucky, Tennessee, Mississippi, Alabama, Louisiana,
Missouri, Oklahoma and Arkansas) and west (California, Oregon, Washington,
Arizona, Nevada and New Mexico). These regions are derived from principle
component analysis to group states with similar PM2.5 source regions
together. For winter, the total PM2.5 high bias is reduced across all
five regions, with most of the improvement coming from reductions in OC,
non-carbon organic matter (NCOM; see Sect. 2 or 3 for definition) and Other,
indicating that improvements in the representation of mixing under stable
conditions helped in reducing the high bias. Still, a large bias remains for
OC, which may be due in part to an overestimation of the residential wood
combustion in the NEI.
Regional and seasonal stacked bar plots of PM2.5 composition at
the CSN sites (left), CMAQv5.0.2_Base simulation (middle) and
CMAQv5.1_Base_NEIv1 simulation (right). In order from top to bottom are
(a) winter, (b) spring, (c) summer and
(d) fall seasons and left to right the northeast, Great Lakes,
Atlantic, south and west regions. The individual PM2.5 components (in
order from bottom to top) are SO42- (yellow), NO3- (red),
NH4+ (orange), EC (black), OC (light gray), soil (brown), NaCl
(green), NCOM (pink), other (white), blank adjustment (dark gray) and
H2O / FRM adjustment (blue).
For spring, the changes in PM2.5 are much more isolated than in winter
(Fig. 6b), with the largest decreases occurring around Montreal (Canada) and
portions of the Midwest and desert southwest (lack of windblown dust in v5.1
contributes to the decrease in the desert southwest). The MB for PM2.5
in the spring is relatively small, with most sites (75 %) reporting a MB
between ±3.0 µg m-3, with some larger underestimations in
Texas and larger overestimations in the northeast, Great Lakes and northwest
(Fig. 7b). As expected with the relatively small change in modeled PM2.5
concentrations with v5.1 in the spring (Fig. 6b), the difference in MB
between v5.0.2 and v5.1 is relatively small, with most differences in MB less
than ±1.0 µg m-3 (Fig. 8b). Some slightly larger
decreases in MB occur in the northeast and northwest, while some larger
increases in MB occur in the Midwest and Texas. A little more than half
(53.0 %) of the sites report an improvement in MB (Fig. S3b). The diurnal
profile of PM2.5 for spring shows a consistent underestimation of
PM2.5 throughout most of the day in the v5.0.2 simulation, which becomes
larger in the CMAQv5.1_Base_NEIv1 simulation, with an overall decrease in
PM2.5 in the spring (Fig. 9b). However, the RMSE is lower during the
overnight, morning and afternoon hours in the CMAQv5.1_Base_NEIv1
simulation, and the correlation improves throughout most of the day as well
(Fig. S5). Total PM2.5 MB improves in three of the five regions shown in
Fig. 10, with most of the improvement coming from lower concentrations of OC
and NCOM.
Difference in the monthly average hourly O3 (ppbv) for winter
(DJF; top left), spring (MAM; top right), summer (JJA; bottom left) and fall
(SON; bottom right) between CMAQ v5.0.2_Base and v5.1_Base_NEIv1
(CMAQv5.1_Base_NEIv1–CMAQv5.0.2_Base). Note that the scales between each
plot may vary.
In the summer, PM2.5 is considerably higher
(> 5.0 µg m-3) across a large portion of the eastern US in
the CMAQv5.1_Base_NEIv1 simulation, particularly in Mississippi, Alabama,
Georgia and portions of the Ohio Valley (Fig. 6c). The increase in PM2.5
is primarily due to the updates to the IEPOX-SOA chemistry in v5.1 (Fig. 2),
updates to BVOC emissions in BEISv3.61 (approximately
1.0 µg m-3 increase PM2.5 in the southwestern US), and the
ACM2 and MOL updates in WRF and CMAQ (Fig. 1), with smaller contributions from
the updates in CB05e51 chemical mechanism (Fig. 5) and updates to the
clouds/photolysis (Fig. 3). Despite the increase in PM2.5 with v5.1,
PM2.5 still remains largely underestimated in the summer, with the
largest underestimations in the southeastern US, Texas and California (Fig. 7c).
However, the result of the widespread increase in PM2.5 with v5.1 is a
similar large, widespread reduction in the MB across the eastern US,
particularly in the southeast and the Ohio Valley, where reductions in MB
range from 3.0 to 5.0 µg m-3 (Fig. 8c). Smaller increases in
the MB (typically less than 2.0 µg m-3) occur in Florida and
isolated areas in the western US. Of all the sites, 69.8 % report an
improvement in MB, with a number of sites showing reductions in MB greater
than 5.0 µg m-3 (Fig. S3c). PM2.5 is underestimated
throughout the day in both the v5.0.2 and v5.1 simulations (Fig. 9c) in
summer, with the underestimation improving slightly with v5.1, particularly
during the afternoon and overnight hours. RMSE improves during the daytime
hours with v5.1, while correlation is considerably higher with v5.1 than
v5.0.2 throughout the entire day (Fig. S6). Total PM2.5 is
underestimated in the CMAQv5.0.2_Base simulation in four of the five regions
(the west region being the exception), which improves in the
CMAQv5.1_Base_NEIv1 simulation (Fig. 10). The overestimation in the west
region with v5.0.2 also improves with v5.1. Small increases in SO42-
and NH4+ and larger increases in OC and NCOM contribute to the
improvement.
For the fall, the difference in PM2.5 between v5.0.2 and v5.1 is again
small, with the largest increases occurring in isolated portions of the
eastern US and California, and the largest decreases occurring in Montreal
and isolated areas in the western US (Fig. 6d). The MB pattern in the fall
(Fig. 7d) is similar to the one in the spring as well (Fig. 7b), with
relatively small MBs in the eastern US (±2.0 µg m-3) and
larger MBs along the west coast (underestimated in California and
overestimated in the northwest). As expected, the change in the MB between
v5.0.2 and v5.1 is also relatively small in the fall, with the majority of
the sites reporting a change in MB of less than
±2.0 µg m-3 (Fig. 8d), and 68.1 % of the sites
reporting a reduction in MB (Fig. S3d). The average diurnal profile of
PM2.5 in the fall (Fig. 9d) is similar to the spring, with improved MB
with v5.1 during the overnight, morning and late afternoon or evening hours and
reduced RMSE and improved correlation throughout the entire day (Fig. S8).
Total PM2.5 is overestimated in all five regions in the fall (Fig. 10),
but improves with v5.1 in all of those regions (albeit only very slightly for
the south region), with decreases in EC and OC responsible for most of the
improvement.
Seasonal average hourly O3 (ppbv) mean bias at AQS sites for
(a) winter (DJF), (b) spring (MAM), (c) summer
(JJA) and (d) fall (SON) for the CMAQv5.1_Base_NEIv1 simulation.
Difference in the absolute value of monthly average O3 (ppbv)
mean bias for (a) winter (DJF), (b) spring (MAM),
(c) summer (JJA) and (d) fall (SON) between CMAQ
v5.0.2_Base and v5.1_Base_NEIv1 (CMAQv5.1_Base_NEIv1–CMAQv5.0.2_Base).
Cool colors indicate a reduction in O3 mean bias in v5.1, while warm
colors indicate an increase in O3 mean bias v5.1.
Diurnal time series of seasonal O3 (ppbv) for AQS observations
(gray), CMAQv5.0.2_Base simulation (blue) and CMAQv5.1_Base_NEIv1
simulation (red) for (a) winter, (b) spring,
(c) summer and (d) fall.
Diurnal time series of seasonal NOx (ppbv) for AQS observations
(gray), CMAQv5.0.2_Base simulation (blue) and CMAQv5.1_Base_NEIv1
simulation (red) for (a) winter, (b) spring,
(c) summer and (d) fall.
Ozone
For the winter, O3 widely decreases in the CMAQv5.1_Base_NEIv1
simulation vs. the CMAQv5.0.2_Base simulation across the western US, with
the seasonal average decreases ranging between 1.0 and 3.0 ppbv, and several
areas where decreases exceed 3.0 ppbv, primarily over the oceans (Fig. 11a).
In the eastern US, the change in O3 is relatively small and isolated.
Ozone is underestimated at most sites across the northern portion of the US,
with the largest underestimations occurring in Colorado, Wyoming and Utah.
Despite the decreases in O3 with v5.1, O3 is still overestimated in
the southwestern US and California (Fig. 12a). There is a widespread reduction
in the O3 MB in California and increased MB in the upper Midwest with
v5.1, while across the rest of the domain the change in MB is relatively
small (Fig. 13a). The majority of the change in O3 falls between
±5.0 ppbv, with 56.5 % reporting a reduction in MB (Fig. S8a). The
average diurnal profile of O3 in the winter (Fig. 14a) shows slightly
lower mixing ratios during most of the day with v5.1, the exception being the
late afternoon and early evening hours, when mixing ratios are slightly
higher. The result is reduced MB and RMSE, and higher correlation throughout
the day with v5.1 vs. v5.0.2 (Fig. S9). The NOx diurnal profile also
generally improves throughout the day in winter (Fig. 15a), with decreased MB
and RMSE in the afternoon or early evening and increased correlation throughout
the day (Fig. S11).
The pattern of change in O3 between v5.0.2 and v5.1 in spring is similar
to winter, with lower O3 mixing ratios in the western US and higher
mixing ratios in the eastern US in v5.1 compared to v5.0.2 (Fig. 11b).
Decreases in O3 mixing ratios in the western US in v5.1 range from
roughly 1.0 to 3.0 ppbv (similar to winter), while in the eastern US the
increases generally range from 1.0 to 2.0 ppbv, with isolated areas of larger
increases. The MB of O3 for the v5.1 simulation primarily ranges from
slightly overestimated to slightly underestimated across most the sites, with
larger overestimations along the Gulf Coast and larger underestimations in
the western US (Fig. 12b). The change in MB between v5.0.2 and v5.1 shows
mixed results (Fig. 13b), with slight increases and decreases across much of
the eastern US and a relatively large increase in MB in the Midwest (i.e.,
Colorado and Wyoming). The MB mostly improved across the Gulf Coast and in
California due to reduced O3 mixing ratios from the new marine halogen
chemistry and enhanced O3 deposition to ocean surfaces. Half of the
sites reported a reduction in MB (Fig. S8b) with v5.1. The diurnal profile of
O3 for spring (Fig. 14b) shows a relatively large increase in mixing
ratios in the late afternoon and evening (04:00 to 22:00 LST),
resulting in a large improvement in MB during that time (Fig. S11). Similar
improvements are noted in RMSE and correlation in the afternoon and evening
hours. The NOx diurnal profile also shows a large decrease in the late
afternoon and early evening mixing ratios (Fig. 15b), with a large decrease
in both MB and RMSE during that time, and improved correlation throughout the
day (Fig. S12).
For the summer, the pattern of change in O3 is markedly different from
the winter and spring, with a large widespread increase in O3 mixing
ratios across the eastern US and decreases in the Gulf of Mexico, southern
Florida and over the eastern Atlantic Ocean (Fig. 11c). Increases in O3
in the eastern US range from 2.0 to 10.0 ppbv, with isolated areas of larger
increases in the major urban areas (e.g., Chicago, Illinois, and Atlanta,
Georgia) that can be largely attributed to the updates to the ACM2 and the
MOL calculation in WRF and CMAQ (Fig. 2b) as well as increased photolysis in
v5.1 vs. v5.0.2 (Fig. 1). Smaller increases in O3 occur in the western
US, particularly southern California and the SJV. Large decreases in O3
over the oceans are likely the result of the inclusion of the marine halogen
chemistry in v5.1, with some decreases exceeding 10.0 ppbv. The MB of
O3 for the v5.1 simulation shows widespread overestimations in the
eastern US, particularly along the Gulf of Mexico, while in the western US
the MB is mixed, with the largest overestimations occurring along the
California coast (Fig. 12c).
As expected, the consequence of the widespread increase in O3 in the
eastern US in v5.1 is a corresponding widespread increase in the MB compared
to v5.0.2, particularly in the mid-Atlantic and southeast (Fig. 13c). Ozone
MB decreases along the coast of Florida and along the Gulf of Mexico, the
result of decreased O3 over the water. The change in MB in the western
US is mixed, with some areas showing improved MB (e.g., SJV), while others
show increased MB (e.g., southern California). The diurnal profiles of O3
show that mixing ratios increase throughout most of the day in v5.1
(the exception being 00:00–05:00 LST) (Fig. 14c), resulting in increased MB
and RMSE throughout the morning and early afternoon hours (Fig. S13).
However, RMSE decreases substantially during the late afternoon and overnight
hours, and the correlation improves throughout the entire day. The NOx
concentrations are lower throughout the day with v5.1 compared to v5.0.2
(Fig. 15c), which results in large improvements in the MB in the morning and
afternoon periods and slightly increased MB in the middle of the day, while
RMSE and correlation improve throughout the day (Fig. S14).
For the fall, the pattern of change in O3 for v5.1 vs. v5.0.2 is nearly
identical to spring (Fig. 11d), with widespread decreases in O3 in the
western US (possibly due to reduced cloud mixing and entrainment from the
free troposphere) and mostly small increases in O3 in the eastern US,
with the exception of larger increases in several of the major urban areas
(e.g., St. Louis, Missouri and Atlanta, Georgia). The changes are generally
small, between ±2.0 ppbv, with isolated areas of larger increases or
decreases. Ozone is also lower over the Pacific and Atlantic oceans and the
Gulf of Mexico. While the change in O3 between v5.0.2 and v5.1 is very
similar to the spring, the MB pattern for v5.1 is not. Unlike the spring,
where O3 was underestimated in many areas, in the fall O3 is
overestimated for almost all the sites (Fig. 12d). Sites in the Midwest have
the lowest overall MB, while the east and west coasts show large
overestimations of O3. The increased O3 in the eastern US with v5.1
results in generally higher MB compared to v5.0.2, while in the western US
the result is slightly lower MB on average, the exception being southern
California (Fig. 13d). As was the case in the spring, slightly less than half
the sites (48.4 %) report a reduction in MB, with the majority of the
change falling between ±5.0 ppbv (Fig. S8d). The diurnal profile of
O3 in the fall shows increased mixing ratios with v5.1 during the
daytime hours and slightly decreased mixing ratios overnight (Fig. 14d),
resulting in increased MB during the daytime and lower MB overnight. However,
the RMSE is reduced and the correlation is higher throughout the day
(Fig. S15). Similar to the other seasons, the diurnal profile of NOx in
the fall shows lower mixing ratios throughout the day (Fig. 15d),
particularly in the early morning and late afternoon hours, resulting in
higher MB in the morning and lower MB in the afternoon, while RMSE is reduced
and correlation is higher throughout the entire day with v5.1 (Fig. S16).
Observed (black) and CMAQ-simulated vertical profiles of
(a) O3, (b) NO2, (c) NOy,
(d) alkyl nitrates (ANs), (e) peroxy nitrates (PNs) and
(f) HNO3 for the Edgewood site in Baltimore, MD, on 5 July 2011.
CMAQv502_Base simulation profiles are shown in green and
CMAQv51_Base_NEIv1 simulation profiles are shown in red. Altitude (km) is
given on the y axis, while mixing ratio (ppbv) is given on the x axis.
ANs, PNs and NOy
Previous studies have shown that CMAQ can significantly overestimate NOy
mixing ratios (e.g., Anderson et al., 2014). To help address the NOy
overestimation in CMAQ, updates were made to the atmospheric chemistry in
v5.1 pertaining to the formation and cycling of alkyl nitrates (ANs), peroxy
nitrates (PNs) and NOy in the model (Sect. 2.4.1). Overall, monthly
average hourly NOy mixing ratios at AQS sites decreased between
approximately 13 % (January) and 21 % (July) in the
CMAQv5.1_Base_NEIv1 simulation vs. the CMAQv5.0.2_Base simulation. The
result is a slight decrease in the normalized mean error (NME) in January
from 70 % (v5.0.2) to 61 % (v5.1), but a much larger decrease in NME
in July from 151 % (v5.0.2) to 101 % (v5.1). Mixing ratios of ANs and
PNs are not routinely measured; however, the DISCOVER-AQ campaign
(https://www.nasa.gov/mission_pages/discover-aq/) that took place over
the Baltimore, Maryland and Washington DC area in July 2011 provides
aircraft measurements of PNs and ANs, along with NO2, NOy,
HNO3 and O3. The National Oceanic and Atmospheric Administration
(NOAA) P3B aircraft performed measurement flights on a number of days during
the DISCOVER-AQ campaign. Those flights included vertical spirals over
several locations, one of which was Edgewood, MD (39.41∘ N,
76.30∘ W; 11 m above sea level), a site that often measures very
high O3, and in recent years has measured some of the highest O3 in
the eastern US.
Figure 16 shows vertical profiles of observed and CMAQ (v5.0.2 and v5.1)-simulated O3, NO2, NOy, ANs, PNs and nitric acid (HNO3)
for the Edgewood site on 5 July 2011. Several spirals were performed over the
Edgewood site that day, roughly taking place in the late morning, early
afternoon and late afternoon, so the profiles shown represent an average
profile throughout the day. While O3 is underestimated throughout the
PBL by both versions of the model on that day, the underestimation improves
significantly in the v5.1 simulation. NO2 and NOy are overestimated
throughout the PBL by both versions of the model, but again, the
overestimation is greatly improved in the v5.1 simulation. The PNs, ANs and
HNO3 show mixed results, with the AN performance improving, the PN
performance degrading and the HNO3 performance relatively unchanged with
v5.1. Note that there has been an update in the recommended PAN formation and
degradation equilibrium constant (http://iupac.pole-ether.fr) which
lowers the predicted PAN concentrations in CMAQ and is currently being
examined for its impact on other species. On this particular day, v5.1
generally shows a large improvement in performance over v5.0.2; however, the
results on other days may be different, but it does highlight the large
change in NOy mixing ratios that can be expected with v5.1.
Difference in MDA8 O3 daily ratios (cut scenario / base)
for CMAQv5.0.2 and v5.1 (v5.0.2–v5.1) for a 50 % cut in anthropogenic
NOx (a–b) and VOC (c–d) for January (a, c)
and July (b, d) binned by the modeled MDA8 O3 mixing ratio
(ppbv). Values greater than 1 indicate v5.1 is more responsive than v5.0.2
to the emissions cut, while values less than 1 indicate v5.0.2 is more
responsive. Given above the x axis is the number of model grid cells in
each bin.
Modeled response to emission changes
One of the primary applications of air quality models is to determine the
impact that changes (e.g., reductions from abatement strategies) in emissions
have on ambient air quality. Examples of this type of application include
federal rules and state implementation plans (SIPs) which aim to reduce
emissions (through regulations) in order to meet mandated air quality
standards. In this type of application, the air quality model is run using
both baseline (often current year) and future year emissions (when emissions
are typically lower due to state and national regulatory efforts) and then
the change in criteria pollutant (e.g., O3 and PM2.5) concentrations
between the two simulations is quantified in order to assess the impact
(benefit) that emission reductions will have on future ambient air quality.
As such, it is important to establish the ability of the model to accurately
simulate the future ambient air quality given a known change in emissions,
which here is referred to as the model responsiveness (to emission changes).
Some previous analyses comparing observed changes in ambient air quality
(over periods witnessing large reductions in emissions) to CMAQ-estimated
changes in ambient air quality (with estimated reductions in emissions)
during the same period have shown that the model tends to underestimate the
observed change in ambient O3, suggesting the model may be
underresponsive to the emission reductions impacting O3 (Gilliland et
al., 2008; Foley et al., 2015). The over- or underresponsiveness of the model to
emission projections can have implications in the planning process for
determining the extent to which emissions must be reduced in order to meet
future air quality standards. In the following sections, we examine the model
responsiveness to emission reductions in CMAQ v5.0.2 and v5.1 by computing
the ratio of maximum daily 8 h average (MDA8) O3 mixing ratios and
total PM2.5 (and select PM2.5 component species) between
simulations using the base emission inventories and those employing 50 %
reductions in NOx, VOC and SOx emissions in order to estimate a
model responsiveness to the emission reductions for each version of the
model. The model responsiveness for v5.1 is then compared to that of v5.0.2
to determine whether the model responsiveness increased, decreased or was
unchanged in the new version of the model.
O3
Figure 17 shows the difference in the ratio (emission cut
simulation / base simulation) of MDA8 O3 for the 50 % cut in
anthropogenic NOx and VOC scenarios, binned by model MDA8 O3 mixing
ratio. Values greater than zero indicate v5.1 is more responsive to the
NOx or VOC cut than v5.0.2, while values less than zero indicate v5.1 is
less responsive than v5.0.2. For both January and July, the median difference
in ratio values for all bins for the 50 % NOx cut scenario are
greater than zero, indicating that v5.1 is more responsive than v5.0.2 to the
cut in NOx. For the 50 % cut in VOC emissions, the difference in the
ratio values is mixed across the 2 months and the different bins. For
January, all of the bins indicate that v5.0.2 is more responsive than v5.1 to
the 50 % VOC cut, with the greatest difference occurring for MDA8 O3
mixing ratios greater than 65 ppbv. For July, v5.1 is slightly more
responsive to the VOC cut for MDA8 O3 mixing ratios less than 75 ppbv
and less responsive for MDA8 O3 mixing ratios greater than 85 ppbv.
Box plots of monthly average ratio values (cut / base) of PMIJ
(total PM2.5), ASO4IJ, ANO3IJ, ANH4IJ, AECIJ, ANCOMIJ, AUNSPECIJ, AOMIJ,
APOAIJ, AORGAJ, AORGBJ and AORGCJ for v5.0.2 (blue) and v5.1 (red) for a
50 % cut in anthropogenic NOx (a, d), VOC (b, e)
and SOx (c, f) for January (a–c) and
July (d–f).
PM2.5
Figure 18 shows the difference in the ratio (emission cut
simulation / base simulation) of PM2.5 and select PM2.5
component species between v5.0.2 and v5.1 for January and July for a 50 %
cut in anthropogenic emissions of NOx, VOC and SOx. For January,
the overall response of modeled PM2.5 (PMIJ) to a 50 % reduction in
NOx is primarily driven by a decrease in nitrate and its associated
ammonium. CMAQv5.1 PM2.5 is slightly less responsive to NOx
reductions compared to v5.0.2, but is still overall quite similar. The VOC
cut shows greater response with v5.1 than v5.0.2 in January in ANCOMIJ
(non-carbon organic matter attached to primary organic carbon; Simon and
Bhave, 2012), AUNSPECIJ (unspeciated PM), AOMIJ (all organic matter), AORGAJ
(SOA from anthropogenic VOCs), AORGBJ (SOA from biogenic VOCs) and total
PM2.5 (see Sects. 2 and 3). Note that the letters I and J after the
species name indicate which CMAQ modal distributions are being included in
the total species mass, with I indicating the Aitken mode and J indicating
the accumulation mode. Since NCOMIJ is nonvolatile, its change reflects how
reducing VOCs changes oxidants such as OH. In general, the model PM2.5
is not very sensitive to VOC cuts in January. And finally, for the 50 %
SOx cut scenario PM2.5 is only slightly less responsive with v5.1,
with all the species being similarly responsive to the SOx cut using
v5.1 compared to v5.0.2.
For July, the NOx cut scenario with v5.1 shows greater responsiveness
for the ASO4IJ (sulfate), ANH4IJ (nitrate), AECIJ (elemental carbon), APOAIJ
(primary organic aerosol) and AORGCJ (SOA from glyoxal and methylglyoxal
processing in clouds) species and total PM2.5 vs. v5.0.2. For the VOC
cut scenario, the AORGAJ species show increased responsiveness with v5.1.
CMAQv5.1 alkane SOA is not dependent on NOx levels or HO2 : NO
ratios, so the decrease in VOC precursors have a more direct effect than for
the aromatic systems (the only AORGAJ in v5.0.2), where decreasing the VOC
precursors can also modify the HO2 : NO ratio and thus yields. CMAQv5.1 PMIJ becomes slightly more responsive to SOx cut as a result of an
increased sensitivity of biogenic SOA to sulfur containing compounds. This
link results from the IEPOX acid-catalyzed SOA in the model which has been
shown to be correlated with sulfate (Pye et al., 2016).
Summary
A new version of the CMAQ model (v5.1) containing numerous scientific updates
has been released and evaluated in terms of the change in performance against
the previous version of the model (v5.0.2), performance compared to
observations, and response to changes in inputs (i.e., emissions).
Specifically, updates were made to the ACM2 scheme in both WRF and CMAQ to
improve the vertical mixing in both models, along with updates to the MOL
calculation, which also directly impacted the vertical mixing in the WRF–CMAQ
system. The overall net effect of these updates was to increase the
ventilation in the model, particularly during the transition periods (morning
and evening), which in turn reduced the concentration of primary emitted
species (e.g., NOx and OC) and consequently increased simulated O3
(a result of reduced NOx titration) and decreased PM2.5
concentrations due to greater dilution. Several new SOA formation pathways
and species were added to v5.1, resulting in increased SOA, particularly in
the southeastern US, and improved PM2.5 performance in the summer, as
PM2.5 is generally underestimated by CMAQ during the summer in the US.
The in-line photolysis model within CMAQ was updated in v5.1. Cloud cover for
the photolysis model in v5.0.2 used a single cloud deck with a constant cloud
fraction and water droplet mixing ratio. In v5.1, multiple cloud decks with
variable cloud fractions and multiple types of water condensates are used in
the photolysis model to be more consistent with the WRF meteorological model
and the CMAQ cloud model. The net effect of this change was to decrease the
amount of subgrid clouds in the photolysis calculation in v5.1, which in
turn results in higher photolysis rates and thus higher predicted O3
mixing ratios on average. In addition to the change to the photolysis model,
the refractive indices for aerosol species are now both wavelength- and
composition-dependent. Changes in aerosol scattering and extinction also
introduce options for how to calculate their optical properties and allow the
user to specify which aerosol mixing model and method to use to solve the Mie
scattering theory. The atmospheric chemistry in the model has also been
updated from CB05TUCL to CB05e51 in v5.1, which includes, among other things,
updates to the NOy reactions, additional isoprene extensions, explicit
representation of several HAPs and a simple parameterization of the effects
of halogens on O3 in marine environments. The net effect of going from
CB05TUCL to CB05e51 was to increase O3 in the winter and summer, while
increasing PM2.5 slightly in the winter and increasing or decreasing
PM2.5 slightly in the summer.
Overall, the scientific updates in v5.1 resulted in improved model
performance for PM2.5 in the winter and summer and a very small overall
change in performance for the spring and fall. Wintertime PM2.5
concentrations are considerably lower with v5.1 vs. v5.0.2, a season when
PM2.5 is typically overestimated by CMAQ over the US. Conversely, during
the summer when PM2.5 is largely underestimated by CMAQ over the US,
PM2.5 concentrations are typically higher with v5.1 vs. v5.0.2,
particularly in the southeastern US. The change in O3 mixing ratios in
v5.1 resulted in mixed improvement in MB, both spatially and temporally, with
the summer showing the largest increase in MB. However, RMSE largely improved
regardless of season and showed a larger improvement spatially across the
sites than MB, and the correlation was almost always higher with v5.1.
Comparisons of vertical profiles of several species taken over Edgewood, MD,
during the DISCOVER-AQ campaign showed improved performance with v5.1
throughout the PBL for O3, NO2, NOy, ANs and CO, with the PNs
being the only species to show degraded performance on that day.
The response of the model to changes in emission inputs was examined by
comparing the ratio of the base v5.0.2 and v5.1 simulations to sensitivity
simulations with 50 % cuts each to anthropogenic NOx, VOC and
SOx emissions. CMAQv5.1-simulated MDA8 O3 exhibited more
responsiveness (greater reduction) to the 50 % NOx cut in January
and July than v5.0.2, which is considered an improvement as previous studies
suggested CMAQ O3 to be underresponsive to large changes in emissions.
The responsiveness of PM2.5 to the emission cuts is more complicated
than for O3 since there are many more species comprising PM2.5 and
some of those have greater or smaller response with v5.1. However, the new
pathways of formation for several PM2.5 components in v5.1 generally
result in greater responsiveness in v5.1 compared to v5.0.2 for the various
emission cut scenarios.
Finally, a number of important science updates are in development and will be
available in the next release of CMAQ (v5.2), which improve upon the existing
science in the model. These updates include a new version of the windblown
dust treatment (Foroutan et al., 2017), the carbon bond 6 (CB6) chemical
mechanism (Ramboll Environ, 2016), enhancements to the calculation of
semi-volatile primary organic aerosol (POA) and SOA from combustion sources
in CMAQ (Pye et al., 2016), and additional updates to the calculation of
clouds. In addition to the model updates, a number of instrumented versions
of the model (e.g., decoupled direct method, sulfur tracking) will also be
released with v5.2. These updates represent potentially significant
improvements over the current options in v5.1 (specifically the updated
windblown dust treatment) and therefore are being made available to the
community more quickly than they might have in the past.