Overview and evaluation of the Community Multiscale Air Quality (CMAQ) model version 5.1

The Community Multiscale Air Quality (CMAQ) model is a comprehensive multi-pollutant air quality modeling system developed and maintained by the U.S. Environmental Protection Agency’s (EPA) Office of Research and Development (ORD). 15 Recently, version 5.1 of the CMAQ model (v5.1) was released to the public which incorporates a large number of science updates and extended capabilities over the previous release version of the model (v5.0.2). These updates include improvements in the meteorological calculations in both CMAQ and the Weather Research and Forecast (WRF) model used to provide meteorological fields to CMAQ; updates to the gas and aerosol chemistry; revisions to the calculations of clouds and photolysis; and improvements to the dry and wet deposition in the model. Sensitivity simulations isolating several of the major updates to the modeling system 20 show that changes to the meteorological calculations generally result in greater afternoon and early evening mixing in the model, times when the model historically underestimates mixing. The result is higher ozone (O3) mixing ratios on average due to reduced NO titration and lower fine particulate matter (PM2.5) concentrations due to greater dilution of primary pollutants (e.g. elemental and organic carbon). Updates to the clouds and photolysis calculations greatly improve consistency between the WRF and CMAQ models and result in generally higher O3 mixing ratios, primarily due to reduced cloudiness and reduced attenuation of photolysis 25 in the model. Updates to the aerosol chemistry results in higher secondary organic aerosol (SOA) concentrations in the summer, thereby reducing PM2.5 bias, while updates to the gas chemistry result in generally increased O3 in January and July (small) and slightly higher PM2.5 concentrations on average in both January and July. Overall, seasonal variation in simulated PM2.5 generally improves in the new model version, as concentrations decrease in the winter (when PM2.5 is overestimated by CMAQ v5.0.2) and increase in the summer (when PM2.5 is underestimated by CMAQ v5.0.2). Ozone mixing ratios are higher on average with v5.1 30 versus v5.0.2, resulting in higher O3 mean bias, as O3 tends to be overestimated by CMAQ throughout most of the year (especially at locations where the observed O3 is low), however both the error and correlation are largely improved with v5.1. Sensitivity simulations for several hypothetical emission reduction scenarios showed that v5.1 tends to be slightly more responsive to reductions in NOX (NO + NO2), VOC and SOX (SO2 + SO4) emissions than v5.0.2, representing an improvement as previous studies have shown CMAQ to underestimate the observed reduction in O3 due to large, widespread reductions in observed 35 emissions. Finally, the computational efficiency of the model was significantly improved in v5.1, which keeps runtimes similar to v5.0.2 despite the added complexity to the model.

I bring here 3 points for the final review. Point 1 strikes me as critical and needs to be corrected because it leads to a (in my opinion) biased and unjustified piece of conclusion regarding the relative importance of the update in emissions and the scientific updates.
Point 2 brings back to consideration a remark from the initial review that I think has been too overlooked by the authors, Point 3 is a request for change in the color scale of a Figure so that the albedo is between 0 and 1 I nonetheless wish to thank the authors for the considerable work in the Review process, even though I still think that more written information about the model design could have been brought in this new CMAQ reference article.
Below, in green the Authors' text (either answers to my initial review or text from the manuscript), in black my text, in blue statements from my initial review.
Point 1 "Obviously it was not made clear in the manuscript that the overall impact from the emission platform change was small. Hopefully this is now made clear in the text. In addition, a figure showing the impact of the emissions platform change on ozone and PM2.5 in January and July has been added to the text to quantify to the reader the impact from the emissions platform change." The following statement is introduced in the revised version (it would be easier to find if the Authors had indicated explicitly where they had made such an adjustment): "However, based on sensitivity simulations performed for January and July 2011 where the only difference was the emissions platform used, the differences in O3 30 and PM2.5 between those two simulations used were generally small and isolated, suggesting there is minimal impact to the comparison between the v5.0.2 and v5.1 simulations from the change in the emissions platform used. Figure S1 shows the impact on winter (January) and summer (July) O3 and PM2.5 between simulations using the different emission platforms." I have several remarks on the Author's response and the corresponding changes that have been performed: -The Figure has not been added "in the text" but as a supplement S1 -The statement that "the differences between those two simulations were generally small and isolated" seem to me as very strange: if one looks at Fig. S1a along with Fig. 6a of the revised simulations for several hypothetical emission reduction scenarios show that v5.1 tends to be slightly more responsive to reductions in NO x (NO + NO 2 ), VOC and SO x (SO 2 + SO 4 ) emissions than v5.0.2, representing an improvement as previous studies have shown CMAQ to underestimate the observed reduction in O 3 due to large, widespread reductions in observed emissions.

Introduction
Numerous Federal (e.g. United States Environmental Protection Agency (USEPA)), State and private entities rely on numerical 25 model simulations of atmospheric chemistry, transport and deposition of airborne emissions and 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), climate change (e.g. Nolte et al., 2008), and provide next-day air quality forecasts (e.g. Eder et al., 2006) in order to inform and protect the public 30 from potentially harmful air pollutants. Since these models are often used to inform the standard setting and implementation for criteria pollutants (e.g. ozone (O 3 ) and fine particulate matter (PM 2.5 )), they must be maintained at the state-of-the-science.
New versions of the CMAQ model have been released periodically over the past fifteen 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 35 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., 2007Appel et al., , 2008Appel et al., , 2013Foley 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 publically released in December 2015 (http://www.cmaq-model.org/). 40 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, 2007ab) 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, 45 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 Section 2.1. A new explicit treatment of secondary organic aerosol (SOA) formation from isoprene, alkenes and 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;Nenes et al., 1999). The AERO5 mechanism has been deprecated and 50 is no longer available. The updates to the aerosol treatment in v5.1 are described in Section 2.2. Significant changes were also made to the in-line calculation of photolysis rates (described in Section 2.3). Finally, 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). And finally the additional representation of 55 organic nitrate species in CB05e51. These updates are described in Section 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 Section 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, Section 4.1 evaluates the meteorological updates in WRF and CMAQ; Section 4.2 60 evaluates the aerosol updates; Section 4.3 evaluates the changes to the inline photolysis calculation and the representation of clouds within CMAQv5.1; and Section 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 Section 5, summarizes the overall change in PM 2.5 and O 3 model performance with v5.1 compared to the previously released 65 version (CMAQ version 5.0.2 (v5.0.2)). Section 6 provides a discussion of the model response of O 3 and PM 2.5 to hypothetical reductions in emissions. And finally a summary discussion in provided in Section 7.
2 Review of scientific improvements in CMAQ v5.1 Improvements to the v5.1 modeling system are the result of many years of scientific advancements derived from laboratory, field and numerical experiments and the efforts of a relatively small group of model developers that both investigate avenues 70 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/requests. The updates presented herein represent the "major" 75 updates made to the CMAQ modeling system from the previous model version, and therefore does 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 CMAQ v5.1 release at https://cmaswiki-cempd.vipapps.unc.edu/index.php/CMAQ_version_5.1_(November_2015_release)_Technical_Do 80 2.1 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 short-wave radiation is less 85 than 350 Wm −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 transitions 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, NO 2 , CO and 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 90 the PX-LSM is an increase of the coefficient to the surface energy forcing in the soil temperature force-restore equation (C v ), which is related to volumetric heat capacity (c v ) and heat conductivity (λ) (Pleim and Gilliam 2009) as where τ is 1 day (86400 s), from the previous value of 8x10 −6 K m 2 J −1 recommended by Giard and Bazile (2000) to 1.2 x10 −5 K m 2 J −1 . The new value for C v results from updated values for c v and λ or vegetation based on measurements 95 of various leaves by Jayakshmy and Philip (2010) (c v = 2.0x10 6 J m −3 K −1 , λ = 0.5 W m −1 K −1 ). These changes reduce overestimations of minimum 2-meter temperature (i.e. warmer surface temperatures) during the early morning (dawn) hours while also reducing underestimations of 2-meter 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 (K m ) and heat (K h ) so that the Prandtl 100 number (Pr) is no longer assumed to be unity (Pr = K m /K h = 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. CMAQ v5.1 has also been modified to include the same stability functions that are used in WRF v3.7, and therefore, for consistency, WRF v3.7 (or newer) and CMAQ v5.1 should be used together. Both of these revisions to the ACM2 are described in Pleim et al. (2016).

105
The Monin-Obukhov length (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 WRF v3.8, this re-computed MOL value will be available in the WRF output, and 110 therefore it will be unnecessary to re-compute the MOL value in CMAQ.

Scientific improvements in the CMAQ v5.1 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 (VOC) focus on VOC compounds in existing emission invento-115 ries, 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 polycyclic aromatic hydrocarbons (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 120 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 chemical transport models.
Several new SOA species were introduced in v5.1 AERO6, specifically AALK1 and AALK2 (from long-chain alkanes) and 125 APAH1, APAH2, and APAH3 (from naphthalene). CMAQ v5.1 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 about half of the PAHs is 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 130 the model for purposes other than SOA formation.
CMAQv5.1 has been updated to include the isoprene epoxydiols (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-NO x 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, 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, AISO 3 (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 . This IEPOX SOA replaces the AISO 3 treatment based on Carlton et al. (2010). The AISO3J species name is now retained for IEPOX 140 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/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 145 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 O 3 and NO 2 . 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 150 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 sub-grid convective clouds in the calculation of actinix fluxes. CMAQ uses the ACM cloud model to describe sub-grid convective clouds 155 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 160 (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 value v5.0.2. 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/cmaqwiki/index.php?title=CMAQv5.1_In-165 line_Calculation_of_Photolysis_Rates). 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, sub-grid 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.

170
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 (NO y ) species; incorporation of new research on the atmospheric reactivity of isoprene photo-oxidation 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 175 of the CB05 mechanism. A more detailed explanation of the changes made in the CB05e51 mechanism is provided below.

NO y updates and additions
The most extensive changes made consisted of updates and extensions of the NO y species, including peroxyacylnitrates, alkyl nitrates, and NO x reactions with HO x . 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-180 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+NO 2 reaction rate was updated based on Troe (2012) and a small yield of HNO 3 (<1% at standard temperature and pressure, varying with temperature and pressure) was added to the reaction of HO 2 +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 alkylnitrates and difunctional hydroxy nitrates were assigned Henry's law constants of 6.5e-1 M and 6.5e3 M respectively, while second generation carbonyl nitrates were assigned 1.0e3 M and multifunctional hydroxynitrates were assigned a value of 1.7e4 M. Five species are predominantly from anthropogenic sources, with the relative distribution of mono-functional (alkyl nitrates) and multi-functional (hydroxy, carbonyl, hydroxycarbonyl, and hydroperoxy) nitrate products determined based on the nitrates produced from 190 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 NO x 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 six-hour lifetime on aerosol at high relative hu-195 midity (Liu et al., 2012;Rollins et al., 2013). Additional details can be found in the CMAQv5.1 release documentation (http://www.airqualitymodeling.org/cmaqwiki/index.php?title=CMAQ_version_5.1_(November_2015_release)_Technical_Documentation

Other changes
The high HO x 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 NO x pathways have been modified to explicitly produce 200 methacrolein PAN (MAPAN, described in Section 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, αand 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 integra-205 tion with heterogeneous chemistry, or for numerical consistency. These include the updates to the products of ethanol reaction with OH using recommended yields from IUPAC (http://iupac.pole-ether.fr; accessed May 11, 2016); updates to the reactions of acylperoxy radicals with HO 2 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 (ClNO 2 ) and CINO 2 photolysis as described by Sarwar et al. (2012). 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 meteorological dependent emissions to more consistently represent processes common to both deposition and emissions. Additionally, sea salt and biogenic emissions and dry deposition routines were updated.

Sea salt aerosol emission
The sea salt aerosol emissions module was updated to better reflect emissions 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 220 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 m to 25 m to address a systematic overestimation of near-shore coarse sea-salt aerosol concentrations (Gantt et al., 2015).

225
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/air-emissions-modeling/biogenic-emission-inventorysystem-beis) 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-meter temperature which was inconsistent with 230 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-meter 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

Dry deposition
There were two important updates to the dry deposition calculation in v5.1. First, the dry deposition of O 3 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 O 3 deposition velocity 240 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 :::: ppbv : reduction in the modeled O 3 mixing ratios, with the largest reductions, ∼10%, occurring during the nighttime and early morning hours, and approximately a 2% reduction in the modeled midday O 3 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 245 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.

2.5.4 Gravitational Settling
Previous evaluations of the ground-level coarse particle (PM 10 -PM 2.5 ) concentrations in CMAQ have shown that the model significantly underestimated the total PM 10 concentrations (Appel et al., 2012). Contributing to this underestimation is the 250 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 has 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 255 PM 10 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  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 PM 2.5 is small on a seasonal average and does not affect the seasonal comparisons shown in Section 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 275 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 one-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.

280
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/required for the v5.0.2 and v5.1 simulations. The base v5.0.2 simulation (CMAQv5.0.2_Base) utilized WRF v3.4 meteorological input data, while WRF v3.7 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 v5.1 285 simulations due to the updates made in both WRF and CMAQ (Section 2.1) that would have made performing the CMAQ simulations with output from the same version of WRF difficult and introduce 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) 290 long and short wave radiation (Iacono et al., 2008), Morrison microphysics (Morrison et al., 2005), and the Kain-Fritsch 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 namelists used for each WRF simulation are provided in the supplemental material (see S.4 and S.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 WRF v3.4 data and MCIP 295 version 4.2 (https://www.cmascenter.org/help/documentation.cfm?model=mcip&version=4.2) for the WRF v3.7 data.
All the simulations employed the bi-directional ammonia flux (bi-di) option for estimating the air-surface exchange of ammonia, as well as the in-line estimation of NO x 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 observa-320 tions of gas and particle species in the U.  It makes intuitive sense to see summertime O 3 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 O 3 that occurs in the model, and ultimately results in higher O 3 mixing ratios 370 on average. Conversely, PM 2.5 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 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 PM 2.5 concentrations through increased or decreased SOA formation (spatial heterogeneity of PM 2.5 formation) which results in spatially varying increases and decreases in PM 2.5 concentrations.

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 Figure 2 represent the additional SOA mass that these five 380 species contribute to the total PM 2.5 mass in v5.1. For both January and July, the monthly average concentration of these species is small, ranging between 0.0-0.1 µgm −3 , with the largest concentrations in the eastern half of the U.S., 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 PM 2.5 concentration in the model.

385
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 Figure 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 (v5.1 -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 µgm −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) 390 and much larger compared to January, with peak differences exceeding 2.5 µgm −3 , primarily in the areas with the highest aerosol SO 4 2− concentrations (i.e. Ohio Valley). Therefore, the updated IEPOX-SOA formation pathways in v5.1 represent a potentially significant contribution to the total PM 2.5 , particularly during the summer. Increased isoprene emissions in v5.1 with BEIS v3.61 compared to v5.0.2 with BEIS v3.14 also contribute to the larger contribution of isoprene SOA in v5.1.

Cloud model and in-line photolysis updates 395
Changes in the photolysis/cloud model treatment in v5.1 have potentially significant impacts on the O 3 and PM 2.5 estimates from the model. Figure 3 shows the difference in O 3 and PM 2.5 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 The impact of the updated photolysis in v5.1 is considerably larger in July (when there is more convection) than in January.
Peak O 3 differences in January were around 2.0 ppbv, whereas in July peak differences of greater than 5.0 ppbv (Figure 3b) occur over the Great Lakes (where low PBL heights can enhance the impact of changes in O 3 ). However, in general the 405 difference in O 3 mixing ratios is larger in both magnitude and spatial coverage in July compared to January, indicating that the updated photolysis/cloud model treatment in v5.1 increases O 3 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 O 3 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 PM 2.5 is also larger (both in magnitude and spatial coverage) in July than January http://satdas.nsstc.nasa.gov/) ::::::: (GOES) ::::::: Imager :::::::: product. This evaluation was used to qualitatively determine if one CMAQ version better considers how clouds affect calculated photolysis rates. The GOES product has a 4km 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 minutes prior to the top of the hour during day-420 time hours and were matched to model output at the top of the hour(see section S.1 in the supplemental material for further information). : . ::::: There ::::: were :::: 301 :::::: hours :::: with :::::::: available :::::::: satellite :::: data :::::: across ::: the ::::::: domain ::: in :::: July ::::: 2011. : Figure 4 shows 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 430 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 under-prediction of clouds over much of the Eastern and West Central U.S. in the WRF predicted clouds, which is now directly passed along to CMAQ. This misclassification of modeled clear sky conditions can contribute to an over prediction of O 3 in these regions. Resolving this issue will require changes to the WRF cloud parameterization. Future research will 435 also include changing the sub-grid cloud treatment currently used in the CMAQ system to be consistent with the sub-grid parameterization used in WRF. Section S.1 in the supplemental material provides a table with additional evaluation metrics of the modeled clouds over oceans versus over land and also describes how cloud albedo was calculated for the three model simulations.

Atmospheric chemistry updates 440
As detailed in section 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 section 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. Figure 2). In order to isolate primarily just the effect on PM 2.5 from the atmospheric chemistry 450 changes, the organic matter (AOMIJ; See S.2 and S.3 for species definition descriptions) mass has been removed from the comparisons of total PM 2.5 mass discussed below. Figure 5 shows the difference in monthly average O 3 and PM 2.5 for January and July between the CMAQv5.1_Base ::::::: _NEIv2 and CMAQv5.1_TUCL simulations. For January, O 3 mixing ratios are higher in the simulation using the CB05e51 mechanism (CMAQv5.1_Base ::::::: _NEIv2 : simulation), however the overall impact of CB05e51 on O 3 is generally small (∼2-4%), with maxi- CMAQv5.1_Base simulation range between 0.6 and 1.2 ppbV :::: ppbv (∼2-4%), however larger increases of over 3.0 ppbV :::: 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 O 3 mixing ratios occurs off the eastern coast of the U.S. For July, the difference in PM 2.5 due to the CB05e51 chemical mechanism is relatively small, with differences in concentrations generally ranging from ±0.50 µgm −3

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 section ::::::: Section 3). Several common measurements of statistical performance are 470 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. sub-grid variations) embedded within the observations cannot be accounted for in the model (Swall and Foley, 2009). These issues are somewhat mitigated for networks 475 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).
The diurnal profile of PM 2.5 for winter ( Figure 9a) shows :::::::: indicates a relatively large decrease in MB throughout most 515 of the day in the :::: with : v5.1 versus v5.0.2, particularly during the overnight, morning and late afternoon hours. A similar improvement is seen in the RMSE, while ::: and : the correlation also improves for all hours ( Figure S5). Figure 10 shows seasonal and regional stacked bar plots of PM 2.5 composition (SO 4 2− , NO 3 − , NH 4 + , 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 Figure 10 are 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.

Ozone
For the winter, O 3 widely decreases in the CMAQv5.1_Base ::::::: _NEIv1 : simulation versus the CMAQv5.0.2_Base simulation 585 across the western U.S., with the seasonal average decreases ranging between 1.0 -3.0 ppbV ::::: ppbv, and several areas where decreases exceed 3.0 ppbV ::::: ppbv, primarily over the oceans (Figure 11a). In the eastern U.S., the change in O 3 is relatively small and isolated, the exception being along the coast of Louisiana (due to differences in emissions; see Figure S1c) and to 10pm LST), resulting in a large improvement in MB during that time ( Figure S12 :::: S11). Similar improvements are noted in RMSE and correlation in the afternoon and evening hours. The NO x diurnal profile also shows a large decrease in the late afternoon and early evening mixing ratios (Figure 15b), with a large decrease in both MB and RMSE during that time, and improved correlation throughout the day (Figure S13 :::: S12).
increases a ::::: large ::::::::::: widespread ::::::: increase : in O 3 mixing ratios across the eastern U.S. and decreases in the Gulf of Mexico(partially due to differences in emissions; see Figure S1d), southern Florida and over the eastern Atlantic (Figure 11c) ocean. Increases in O 3 in the eastern U.S. range from 2.0 -10.0 ppbV ::::: ppbv, with isolated areas of larger increases in the major urban areas (e.g. Chicago, Illinois and Atlanta, Georgia) that can largely be :: be :::::: largely : attributed to the updates to the ACM2 and the MOL calculation in WRF and CMAQ (Figure 2b) (Figure 14d), resulting in increased MB during the daytime and lower MB overnight. However, the diurnal profile of NO x in the fall shows lower mixing ratios throughout the day (Figure 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 lower ::::::: reduced and correlation is higher throughout the entire day with v5.1 ( Figure S17 ::: S16).

ANs, PNs and NO y
Previous studies have shown that CMAQ can significantly overestimate NO y mixing ratios (e.g. Anderson et al., 2014). To 655 help address the NO y 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 NO y in the model (Section 2.4.1). Overall, monthly average hourly NO y mixing ratios at AQS sites decreased between approximately 13% (January) and 21% (July) in the CMAQv5.1_Base ::::::: _NEIv1 simulation versus 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  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. O 3 and 685 PM 2.5 ) 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  For both January and July, the median difference in ratio values for all bins for the 50% NO x cut scenario are greater than 705 zero, indicating that v5.1 is more responsive than v5.0.2 to the cut in NO x . For the 50% cut in VOC emissions the difference in the ratio values is mixed across the two 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 O 3 mixing ratios greater than 65 ppbV :::: ppbv. For July, v5.1 is slightly more responsive to the VOC cut for MDA8 O 3 mixing ratios less than 75 ppbV :::: ppbv : and less responsive for MDA8 O 3 mixing ratios greater than 85 ppbV :::: ppbv.

710
22 Figure 18 shows the difference in the ratio (emission cut simulation / base simulation) of PM 2.5 and select PM 2.5 component species between v5.0.2 and v5.1 for January and July for a 50% cut in anthropogenic emissions of NO x , VOC and SO x . For January, the overall response of modeled PM 2.5 (PMIJ) to a 50% reduction in NO x is primarily driven by a decrease in nitrate and its associated ammonium. CMAQ v5 ::::::::: CMAQv5.1 PM 2.5 is slightly less responsive to NO x reductions compared to v5.0.2, 715 but is still overall quite similar. The VOC cut shows greater response with v5.1 than v5.0.2 in January in ANCOMIJ (noncarbon organic matter attached to primary organic carbon; Simon and Bhave, 2012), AUNSPECIJ (unspeciated PM), AOMIJ (all organic matter), AORGAJ (SOA from anthropogenic VOCs) and AORGBJ (SOA from biogenic VOCs) and total PM 2 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 735 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. NO x and OC) and consequently increased simulated O 3 (a result of reduced NO x titration) and decreased PM 2.5 concentrations due to greater dilution. Several new SOA formation pathways and species were added to v5.1, resulting in 740 increased SOA, particularly in the southeast U.S., and improved PM 2.5 performance in the summer, as PM 2.5 is generally underestimated by CMAQ during the summer in the U.S.
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 745 meteorological model and the CMAQ cloud model. The net effect of this change was to decrease the amount of sub-grid clouds in the photolysis calculation in v5.1, which in turn results in higher photolysis rates and thus higher predicted O 3 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 750 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 NO y reactions, additional isoprene extensions, explicit representation of several HAPs, and a simple parameterization of the effects of halogens on O 3 in marine environments. The net effect of going from CB05TUCL to CB05e51 was to increase O 3 in the winter and summer, while increasing PM 2.5 slightly in the winter and increasing/decreasing PM 2.5 slightly in the summer.