Description and evaluation of the Model for Ozone and Related chemical Tracers, version 4 (MOZART-4)

. The Model for Ozone and Related chemical Tracers, version 4 (MOZART-4) is an ofﬂine global chemical transport model particularly suited for studies of the troposphere. The updates of the model from its previous version MOZART-2 are described, including an expansion of the chemical mechanism to include more detailed hydrocarbon chemistry and bulk aerosols. Online calculations of a number of processes, such as dry deposition, emissions of isoprene and monoterpenes and photolysis frequencies, are now included. Results from an eight-year simulation (2000–2007) are presented and evaluated. The MOZART-4 source code and standard input ﬁles are available for download from the NCAR Community Data Portal (http://cdp.ucar.edu).


Introduction
MOZART-4 (Model for Ozone and Related chemical Tracers, version 4) is a global chemical transport model for the troposphere and includes a number of updates over the previous tropospheric version MOZART-2 . MOZART-3 is an extension of MOZART-2 with detailed stratospheric chemistry (Kinnison et al., 2007). The MOZART models are offline models, requiring meteorological fields from either climate models or assimilations of me-Correspondence to: L. Emmons (emmons@ucar.edu) teorological observations. MOZART is built on the framework of the Model of Atmospheric Transport and Chemistry (MATCH) (Rasch et al., 1997). Convective mass fluxes are diagnosed by the model, using the shallow and mid-level convective transport formulation of Hack (1994) and deep convection scheme of Zhang and MacFarlane (1995). Vertical diffusion within the boundary layer is based on the parameterization of Holstlag and Boville (1993). Wet deposition is taken from the formulation of Brasseur et al. (1998). Advective transport incorporates the flux form semi-Lagrangian transport algorithm of Lin and Rood (1996), as described in Horowitz et al. (2003). These physical processes have not been updated from MOZART-2. Details of the chemical solver numerics are given in the Auxiliary Material of Kinnison et al. (2007).
MOZART-4 can be driven by essentially any meteorological data set and with any emissions inventory, so there is not a unique standard simulation. The meteorological variables needed to run MOZART-4 are given in Table 1. Whereas the published evaluation of MOZART-2 was for a simulation driven by climate model meteorology , this work presents MOZART-4 driven by meteorological analyses. A MOZART-4 simulation of 2000-2007 is used here to illustrate components of the model and for evaluation. The model was driven by meteorology from the NCAR reanalysis of the National Centers for Environmental Prediction (NCEP) forecasts (Kalnay et al., 1996;Kistler et al., 2001), at a horizontal resolution of approximately 2.8 • ×2.8 • , Published by Copernicus Publications on behalf of the European Geosciences Union. with 28 vertical levels from the surface to approximately 2 hPa. This resolution is typically used for multi-year simulations. MOZART-4 can be run at essentially any resolution, depending on memory limitations, and typically matching the input meteorological fields. Currently the highest resolution that has been run with full chemistry is 0.7 • ×0.7 • .
MOZART-4 has been tested on computing platforms ranging from a single CPU on a desktop machine to more than one hundred CPUs on a state-of-the-art supercomputer. The following are required for running MOZART-4: the Linux or Unix operating system (including Apple OS X), the cpp utility (available on any Unix or Linux system), the gnu makefile utility, the netcdf utility, and a Fortran 90 compiler. The code has been parallelized using both OpenMP and MPI approaches, with a hybrid MPI-OMP configuration generally giving the fastest runtimes. For desktop systems, MOZART-4 has been compiled and is configured to use either the Portland Group, Lahey/Fujitsu, or Intel Fortran 90 or 95 compilers. All compilers produce similar execution times with normal optimizations. MOZART-4, configured at the standard horizontal resolution of 128×64 grid boxes with 28 vertical levels, requires at least 2 GB main memory per cpu, and can be run on single-, dual-, quad-or 8-CPU workstations. The source code is available for download from the NCAR Community Data Portal (http://cdp. ucar.edu), after registration and approval from the authors. Instructions on the building and running of MOZART-4 are included with the source code. Further information is posted on the MOZART-4 website (http://gctm.acd.ucar.edu/ mozart/models/m4/index.shtml). This paper presents a description of MOZART-4, providing details of the features of the model, and serving as a reference for past and future studies using this model. The following section describes the model updates included in MOZART-4 over MOZART-2. Section 3 summarizes previous applications of MOZART-4. The final sections give a description of the 2000-2007 model simulation presented here, and its evaluation through comparison to global networks and satellite observations.

Updates to model processes
Details of the improvements to MOZART-4 over MOZART-2 are described below. The chemical mechanism has been significantly expanded to include an improved representation of non-methane hydrocarbons and the online calculation of aerosols. An online photolysis scheme takes into account the impact of aerosols and clouds on the photolysis rates. Dry deposition velocities can be determined online in the model and an improved scheme for the determination of albedo has been implemented.

Chemical mechanism
The standard MOZART-4 mechanism includes 85 gas-phase species, 12 bulk aerosol compounds, 39 photolysis and 157 gas-phase reactions. The chemical mechanism includes an updated isoprene oxidation scheme and a better treatment of volatile organic compounds, with three lumped species to represent alkanes and alkenes with four or more carbon atoms and aromatic compounds (called BIGALK, BIGENE and TOLUENE). Oxidation products of these lumped species have also been added. MOZART-2 had a single lumped species ("C4H10") representing all hydrocarbons C 4 and larger. The full list of species and gas-phase, aerosol and photolytic reactions are given in Tables 2-6. Reaction rates are based on JPL 2002 (Sander et al., 2003). Discussion of the isoprene oxidation scheme is given in Pfister et al. (2008a). Table 7 gives an approximate mapping of the longer-lived VOCs in MOZART-4 to three other mechanisms commonly used in regional models, SAPRC-99, RADM, CBMZ (e.g., Tang et al., 2007;Grell et al., 2005;Fast et al., 2006). One application of this mapping is the matching of MOZART-4 results to regional model boundary conditions. The MOZART models make use of a chemical preprocessor, making it relatively easy to modify the chemical mechanism. This facilitates updating reaction rates, adding additional species, or running with simplified mechanisms or a few artificial tracers. MOZART-4 also has the capability of reading offline constituents, such as OH, and prescribed chemical production and loss rates, allowing the simulation of species with relatively simple chemistry (e.g., CO or CH 4 ) without running the full chemistry.

Reactants
Products Rate

Aerosols
The representation of tropospheric aerosols in MOZART-4 has been extended from the work of Tie et al. (2001Tie et al. ( , 2005, and includes the calculation of sulfate, black carbon, primary organic, secondary organic (SOA), ammonium nitrate, and sea salt . Sulfate aerosols are determined from emissions of SO 2 and DMS (Barth et al., 2000). DMS is included to provide an estimate of its source of SO 2 as a precursor to aerosols, and other minor products are ignored (Tie et al., 2001). Black carbon and organic carbon aerosols are emitted in a combination of hydrophobic and hydrophilic forms (80% and 50% hydrophobic, respectively), following Chin et al. (2002). Both black and organic carbon aerosols are converted from hydrophobic to hydrophilic with a rate constant of 7.1 × 10 −6 s −1 following Cooke and Wilson (1996), which is equivalent to a time constant of 1.6 days (Horowitz, 2006;Tie et al., 2005). Secondary organic aerosols are linked to the gas-phase chemistry through the oxidation of monoterpenes and toluene as in Chung and Seinfeld (2002). The ammonium nitrate distribution is determined from NH 3 emissions and the parameterization of gas/aerosol partitioning by Metzger et al. (2002), which is a set of approximations to the equilibrium constant calculation (Seinfeld, 1986) based on the level of sulfate present. The uptake of N 2 O 5 , HO 2 , NO 2 , and NO 3 on aerosols is included (Jacob, 2000), with details given in Table 5. Because only the bulk mass is calculated, a lognormal number distribution is assumed for all aerosols to calculate the surface area, using a different geometric mean radius and standard deviation for each type of aerosol, as listed in Table 6 (based on Chin et al., 2002). Sea salt aerosols are included in the model with four size bins (0.1-0.5, 0.5-1.5, 1.5-5, and 5-10 µm) and emissions are calculated online (Mahowald et al., 2006a). The distributions of four sizes of dust (0.05-0.5, 0.5-1.25, 1.25-2.5, and 2.5-5.0 µm) are set from monthly mean distributions taken from online calculations in the Community Atmosphere Model (CAM) (Mahowald et al., 2006b). Hygroscopic growth of the aerosols is determined from the ambient relative humidity, with different rates for each type of aerosol (Chin et al., 2002). Washout of all aerosols, except hydrophobic black carbon and organic carbon, is set to 20% of the washout rate of HNO 3 (Horowitz, 2006;Tie et al., 2005). Comparison of calculated Table 6. Bulk aerosol parameters used in calculation of surface area: number distribution mean radius (r m ), geometric standard deviation (σ g ) and density.

Photolysis
A significant improvement in the calculation of photolysis rates in MOZART-4 from MOZART-2 is the use of the online fast-TUV (FTUV) scheme, based on the TUV (Tropospheric Ultraviolet-Visible) model, that takes into account the impact of aerosols and clouds . The treatment of aerosols in determining photolysis frequencies and aerosol optical depth are from a lookup table based on the Mie calculations used in the NCAR Community Atmosphere Model (CAM3): soot, organic carbon and sea salt are from the Optical Properties of Aerosols and Clouds (OPAC) software package (Hess et al., 1998); ammonium sulfate is based on Tang and Munkelwitz (1994), Toon et al. (1976), and the appendix of Gong et al. (2003); dust optics are from Zender et al. (2003). A lookup table, based on 4-stream calculations from TUV and also used in MOZART-3, can be used instead of FTUV (Kinnison et al., 2007). The lookup table includes explicit calculation of photolysis frequencies for most of the MOZART-4 species, whereas FTUV includes mapping to a subset of the species. This configuration includes the influence of clouds (Chang et al., 1987), but does not account for the impact of aerosols.

Albedo
An improved scheme for the determination of albedo has been implemented, based on satellite observations (Laepple et al., 2005). Monthly snow and non-snow climatologies have been derived from MODIS observations of albedo, and are combined with snow and ice cover information from the model-driving meteorology.

Online dry deposition
Dry deposition velocities can be determined online in the model, based on the resistance-based parameterization of Wesely (1989), Walmsley and Wesely (1996), Wesely and Hicks (2000). The calculation of surface resistances uses the vegetation distribution of Bonan et al. (2003). If the online calculation is not selected at run-time, a monthly mean climatology, based on the same parameterizations using 10 years of NCEP meteorology, is used. In both cases, the deposition velocity calculation has been extended to take into account special cases for CO, H 2 and PAN. In the case of CO and H 2 , surface uptake is caused by the oxidation by soil bacteria or enzymes (Yonemura et al., 2000). This has been parameterized using the approach of Sanderson et al. (2003), which defines the deposition velocity by a linear or quadratic function in soil moisture content (or its logarithm), depending on the land cover type. In the case of PAN, new laboratory experiments have indicated a strong uptake of PAN by leaves (Teklemariam and Sparks, 2004). Using the results from that study, we have included a leaf uptake of PAN that is vegetation-dependent, based on Sparks et al. (2003). Results from this parameterization agreed with observations during a field experiment (Turnipseed et al., 2006).

Online water calculation
In MOZART-2 water vapor concentrations were taken from the meteorological fields. However, better agreement with observations of precipitation were found when H 2 O was calculated online from the surface moisture flux and all relevant physics parameters, as implemented in MOZART-3 (Kinnison et al., 2007). This online calculation is a runtime option, and the preferred mode of operation, in MOZART-4.

Emissions from vegetation based on MEGAN
Online calculation of biogenic emissions of isoprene and monoterpenes is based on the Model of Emissions of Gases and Aerosols from Nature (MEGAN)  using the implementation described by Pfister et al. (2008a), and in more detail below. While MEGAN parameterizations and emission factor maps have been developed for other species, these have not yet been implemented in MOZART-4. All other biogenic emissions, besides isoprene and monoterpenes, are taken from the POET inventory. These emissions include monthly variation, but are repeated annually for this simulation, without variation from year to year. Since Guenther et al. (2006) gives a comprehensive description of MEGAN, with various options for particular applications, the details of its implementation in MOZART-4, and suggested for other global models, are included here. The emission factor maps used in MEGAN are updated periodically, as more measurements are made and the algorithms are refined (available from http://bai.acd.ucar.edu/Megan). For the results shown here, version 2.1 of the emission factor maps, the most recent version currently available, are used. In other MOZART-4 studies to date, version 2.0 of the isoprene emission factor maps, and version 1.0 of the monoterpenes emissions, have been used, but they do not differ significantly from version 2.1. The MOZART-4 lumped monoterpene ("C10H16") emissions are the sum of the species alpha-pinene, beta-pinene, limonene, myrcene, ocimene, sabinene, and delta-3-carene in the MEGAN emission factor maps.
The MEGAN formulation also requires global maps of plant functional type (PFT) and the monthly leaf area index (LAI). The PFT and LAI maps used in MOZART-4 are based on AVHRR and MODIS data, as used in the NCAR Community Land Model (CLM) (Lawrence and Chase, 2007). The MEGAN emissions dependency on current and past surface air temperature and solar flux are considered using the model meteorology. Pfister et al. (2008a)

General formulation
The MEGAN emissions are based on the emission factor EF (the emissions of a compound at standard conditions), the emission activity factor γ (a normalized ratio that accounts for deviations from standard conditions), and a factor ρ accounting for the canopy production and loss (assume=1 because this is representative of typical conditions). Global, spatially varying emission factor (EF) maps for each compound (isoprene and monoterpenes for MOZART-4) are provided on a 0.5 • ×0.5 • grid for five plant functional types (PFTs): broadleaf trees (btr), needleleaf trees (ntr), shrubs (shr), agricultural crops (crp), and grasses (grs). Landcover maps of PFT fractions per grid box, and leaf area index (LAI) for each month for each PFT are needed. These same maps are used for dry deposition calculations in  PFTs are provided in these maps and must be combined to match the 5 MEGAN PFTs (see Table 8). The emission activity factor depends on the canopy environment (CE), leaf age and soil moisture: where the canopy environment factor depends on leaf area index (LAI), photosynthetic photon flux density (PPFD) and temperature: The soil moisture algorithm described in Guenther et al. (2006) was developed for a specific soil moisture database and wilting point database and may not be appropriate for other soil moisture and wilting point datasets, as pointed out by Müller et al. (2008). Since it is not clear that this algorithm can provide reasonable results with the MOZART driving variables, it has not yet been implemented (γ SM =1). This may result in an overestimation of global annual isoprene emissions by ≈7% .
Thus, for each compound, the emissions are determined by: where f PFT is the fraction of each grid box covered by each PFT, and γ LAI ,γ age ,γ P and γ T will vary by compound, and may depend on PFT.

Isoprene
The leaf age factor (γ age ) for isoprene emissions from evergreen canopies equals 1. For deciduous canopies, γ age is a weighted average of emissions from four ages of foliage (new, growing, mature and old): For isoprene, the relative weights of emissions from each canopy type are: A new =0.05, A gro =0.6, A mat =1.125 and A old =1. The leaf age fractions (F x , for new, growing, mature and old states) are calculated according to Eqs. (17-19), and accompanying text, of Guenther et al. (2006). These factors depend on the change in LAI between the current (LAI c ) and previous (LAI p ) timestep, which for MOZART-4 is a month, based on the time resolution of the LAI maps used. The temperature dependences of Eq. (18) of Guenther et al. (2006) are based on the average temperature of the previous month (cf., Guenther et al., 2006, and corrigendum).
Emission responses to LAI (γ LAI ) variations are estimated according to Eq. (15) of Guenther et al. (2006). The factor accounting for variation in solar radiation (γ P ) is a function of the solar angle, the previous month's average abovecanopy PPFD, and the above-canopy PPFD transmission φ (given by Eqs. (12-13) of Guenther et al., 2006). At low sun angles, φ can become greater than 1, so should be limited to a maximum value of 1.
The temperature dependence for isoprene emissions is determined from: where x=(T p −T opt )/(T p · T opt · 0.00831), C T 1 =80 and C T 2 =200, T opt =313+0.6·(T p −T clim ), E opt =1.75·exp(0.08· (T p −T clim )), with T p being the average temperature of the previous month and T clim the daily average climatological value (297 K). To be consistent with the dependence of γ age on the previous monthly average of LAI, T p refers to the monthly average temperature, as opposed to the previous hourly temperature used in Guenther et al. (2006).

Monoterpenes
MOZART-4 includes a simplified implementation of monoterpene emissions that does not depend on sunlight (γ P =1) or leaf age (γ age =1), and has simple LAI and temperature factors, with no dependence on past solar radiation: where T p is the average temperature of the previous month and T std =303.15.

Soil and lightning NO emissions
Soil NO emissions are a combination of interactive natural emissions (Yienger and Levy, 1995) and fertilizer use (Bouwman et al., 2002). The soil NO emissions are highly dependent on the degree of soil dryness; in order to keep track of this quantity, we have added to the model a bucketstyle parameterization of soil moisture, taking into account the model-calculated precipitation and input latent heat flux, as described by Yienger and Levy (1995). If the soil moisture is not available in the meteorological input dataset, then the model uses a 10-year climatology of monthly-averaged soil moisture from NCEP/NCAR-reanalysis meteorological fields. On average, for late 20th century conditions, the global annual average of soil NO emissions amount to approximately 7 Tg-N/year, with about half of that due to the effect of fertilization (see Yienger and Levy, 1995).
The lightning parameterization differs slightly from that used in MOZART-2 . The lightning strength still depends on cloud top height, with a stronger dependence over land than ocean (Price et al., 1997). The definition of ocean grid boxes has been refined to include only boxes surrounded by ocean, so that the land parameterization is extended one grid box beyond the continents (Price and Rind, 1992). Flash frequency is determined by area, not grid box. The vertical distribution of NO emissions has been modified from that given by Pickering et al. (1998), to have a reduced proportion of the emissions emitted near the surface, Geosci. Model Dev., 3, 43-67, 2010 www.geosci-model-dev.net/3/43/2010/ similar to that used by DeCaria et al. (2006). In addition, the strength of intra-cloud (IC) lightning strikes is assumed to be equal to cloud-to-ground strikes, as recommended by Ridley et al. (2005). The annual lightning emissions for 2006 from MOZART-4 driven by NCEP/GFS are shown in Fig. 2.

Surface boundary conditions
For some long-lived species such as CH 4 , H 2 , and N 2 O, their tropospheric concentrations are known more accurately than their emissions. Therefore, MOZART-4 simulations are generally run using fixed lower boundary conditions constrained by observations, instead of direct emissions for these species. CH 4 monthly zonal averages are based on the measurements by NOAA/ESRL Global Monitoring Division Cooperative Air Sampling Network (Dlugokencky et al., 2005(Dlugokencky et al., , 2008. H 2 is set to 530 ppbv, based on global average observations made by NOAA/ESRL/GMD for 1993-2003 (Novelli et al., 1999). N 2 O concentrations are taken from IPCC (2000).

Upper boundary conditions
Mixing ratios of several species (O 3 , NO x , HNO 3 , N 2 O 5 , CO, CH 4 ) are constrained in the stratosphere since MOZART-4 does not have complete stratospheric chemistry. These mixing ratios have been updated to zonal means from a MOZART-3 simulation. Model concentrations are set to the climatology values above 50 hPa, and relaxed to the climatology with a 10-day time scale down to the tropopause. The primary change from the climatology used in MOZART-2 is a reduction in the NO x and HNO 3 mixing ratios in the lower stratosphere. O 3 is still constrained to observations (from satellite and ozonesondes), as in MOZART-2 . A stratospheric aerosol surface area density climatology (SPARC, 2006) has been included in MOZART-4 so as to more accurately represent the fast NO y partitioning in the lower stratosphere. The gas-phase and heterogenous reactions operate at all model levels, without any additional parameterizations for stratospheric chemistry. It should be noted that MOZART-4 is generally not suitable for studies of the dynamics and detailed structure of the upper troposphere and stratosphere due to these constraints. For example, the impact of lightning and aircraft emissions in the lower stratosphere may not be accurately represented (Grewe et al., 2002). Reanalysis meteorological datasets generally result in a stratospheric flux that is too strong in offline chemical transport models (Van Noije et al., 2004, and references therein), resulting in errors in the tropospheric ozone budget and ozone mixing ratios that are too high in the upper troposphere. This can be the case for MOZART-4 when driven by NCEP/NCAR reanalyses (Kalnay et al., 1996;Kistler et al., 2001) and ECMWF reanalyses (ERA-40) (Simmons and Gibson, 2000). Preliminary simulations with these re- analyses resulted in ozone mixing ratios in the upper troposphere that were much higher than observations, such as ozonesondes.
In order to constrain the stratospheric flux of ozone independently of the meteorological dataset used, the synthetic ozone (SYNOZ) representation is used (McLinden et al., 2000). SYNOZ is a tracer with a specified source region (30S-30N, 10-70 hPa) and production rate (400-500 Tg/yr); after production, SYNOZ is advected as a passive tracer. In order to provide a tropospheric sink, the SYNOZ mixing ratio is relaxed to 25 ppbv below 500 hPa. O 3 is set to SYNOZ above the tropopause, if SYNOZ>100 ppbv; this limits the overwriting of the modeled ozone field by SYNOZ to the lower stratosphere. Because SYNOZ does not provide a realistic distribution of ozone in the stratosphere, an additional tracer, named O3RAD, is relaxed to the ozone climatology (mentioned above) in the stratosphere and to the model calculated ozone in the troposphere; this additional tracer is only used for photolysis calculations. To illustrate the relationship between the SYNOZ, O3RAD and O3 variables, a monthly  Fig. 3. Example profiles of the three ozone variables (O3, O3RAD and SYNOZ) used when including the SYNOZ parameterization. Shown is a monthly average profile for July 2006 at a point over the Pacific Ocean. mean profile is shown in Fig. 3. When SYNOZ is used, the O3 variable is appropriate only for the troposphere, and should not be used for studies of the tropopause region. For simulations using the NCEP/GFS analyses it was found not to be necessary to use SYNOZ to get accurate ozone amounts in the upper troposphere, probably due to the higher vertical resolution than the NCAR reanalysis (42 levels instead of 28) (Pfister et al., 2008b).

Previous MOZART-4 applications
MOZART-4 has already been used in several studies where it has been shown to reproduce observations well. Detailed comparisons have been made of MOZART-4 results to the aircraft, ozonesonde and ground-based observations from the International Consortium for Atmospheric Research on Transport and Transformation (ICARTT) during Summer 2004 (Pfister et al., 2005Lapina et al., 2006;Horowitz et al., 2007;Pfister et al., 2008b). MOZART-4 simulations have been used for boundary conditions in regional models Fast et al., 2009;Mena-Carrasco et al., 2009). Significant improvement in regional model results have been found when time-varying chemical boundary conditions, such as from MOZART-4, are used (e.g., Tang et al., 2009).
MOZART-4 results have also been included in multimodel comparisons, such as the study coordinated by the European Union project Atmospheric Composition Change: the European Network of excellence (ACCENT; http://www. accent-network.org) Stevenson et al., 2006;Shindell et al., 2006). Chemical forecasts have been produced with MOZART-4 and used in flight planning activities for aircraft experiments such as the NSF Megacities Impact on Regional and Global Environment (MIRAGE) Mexico City campaign in March 2006 (Fast et al., 2007), and the NASA Intercontinental Chemical Transport Experiment

MOZART-simulation of 2000-2007
A simulation of 2000-2007 is presented here to illustrate the capabilities of MOZART-4 and for its evaluation. This simulation was driven by NCEP/NCAR-reanalysis meteorological fields, at a horizontal resolution of 2.8 • ×2.8 • (28 levels). The artificial stratospheric ozone mechanism, SYNOZ, was used. Dry deposition velocities, water vapor concentrations, photolysis rates using FTUV, and biogenic emissions for soil NO, and for isoprene and monoterpenes from MEGAN, were calculated online using the new interactive schemes described above.

Emissions
A number of emissions inventories are suitable for use in MOZART-4, depending on the application. For this overview, several inventories have been used to cover all of the species and emissions types. These inventories, as used in MOZART-4, are provided with the MOZART-4 source code (http://cdp.ucar.edu).
The majority of the anthropogenic emissions used for this simulation came from the POET (Precursors of Ozone and their Effects in the Troposphere) database for 2000 (Granier et al., 2005;Olivier et al., 2003), which includes anthropogenic emissions (from fossil fuel and biofuel combustion) based on the EDGAR-3 inventory (Olivier and Berdowski, 2001). The anthropogenic emissions (from fossil fuel and biofuel combustion) of black and organic carbon determined for 1996 are from Bond et al. (2004). For SO 2 and NH 3 , anthropogenic emissions are from the EDGAR-FT2000 and EDGAR-2 databases, respectively (http://www.mnp.nl/edgar/). For Asia, these inventories have been replaced by the Regional Emission inventory for Asia (REAS) with the corresponding annual inventory for each year simulated (Ohara et al., 2007). Aircraft emissions of NO, CO and SO 2 from scheduled, charter, general aviation and military traffic for 1999 are also included (Baughcum Geosci. Model Dev., 3, 43-67, 2010 www.geosci-model-dev.net/3/43/2010/  Monthly average biomass burning emissions for each year are from the Global Fire Emissions Database, version 2 (GFED-v2), which is currently available for (van der Werf et al., 2006. For species not provided in GFED (e.g., individual volatile organic compounds as specified in MOZART-4, SO 2 , and NH 3 ), emissions are determined by scaling the GFED CO 2 emissions by the emission factors of Andreae and Merlet (2001) and updates (Granier et al., 2005), using the vegetation classification provided with GFED. While MOZART-4 provides the capability to use vertically distributed emissions (such as for biomass burning), for these simulations, all emissions, except aircraft and lightning, are emitted at the surface.
Emissions of isoprene and monoterpenes from vegetation, and NO from soil and lightning, are calculated online, as described above. The DMS emissions are monthly means from the marine biogeochemistry model HAMOCC5, representative of the year 2000 (Kloster et al., 2006). SO 2 emissions from continuously outgassing volcanoes are from the GEIA-v1 inventory (Andres and Kasgnoc, 1998).
The surface emissions totals for 2006, separated by category, are shown in Table 9, and summarized for each year in Table 10. NO emissions from lightning are included in Table 10.

Model evaluation
To illustrate the capability of MOZART-4 to reproduce the true atmosphere, a few examples of the evaluation of the model through comparisons with a standard OH climatology, network ground sites, ozonesondes and satellite observations of CO and aerosol optical depth are shown.

Hydroxyl radical
Simulated OH is compared to the climatology of Spivakovsky et al. (2000) in Fig. 4 in terms of the annual airmass-weighted burden, as recommended by Lawrence et al. (2001). For this MOZART-4 simulation, OH is somewhat lower than the climatology, and gives a methane lifetime of about 10.5 years (2000-2007 global average, for the troposphere, altitudes below 100 hPa), which is on the high side of previous model estimates. For example, the MOZART-2 standard simulation had a methane lifetime of 9.4 years , and the MATCH simulations of Lawrence et al. (2001) gave lifetimes between 7.8 and 10.3 years. The MOZART-4 distribution shows less of an increase with altitude from the lower troposphere to the upper troposphere than the climatology, but is similar to the shape of the MATCH simulations (Lawrence et al., 2001). This could indicate limitations in the climatology, which is

Carbon monoxide
Comparison of carbon monoxide (CO) in MOZART-4 to the NOAA ESRL Global Monitoring Division surface sites (Novelli et al., 2003) is shown in Fig. 5. The global distributions of CO from the Measurements of Pollution In The Troposphere instrument on the Terra satellite are a valuable data set for model evaluation (e.g., Shindell et al., 2006). The recently released MOPITT V4 retrievals (Deeter et al., 2010) are used for comparison to the MOZART-4 simulated CO. Figure 6 shows an example comparison of the total column CO for May 2003. The top panel shows the column average mixing ratio directly from the model output and the middle panel shows the MOPITT column retrieval, expressed as average mixing ratio. To properly compare these two products, the averaging kernel and a priori profile associated with each MOPITT retrieval must be applied to the model profiles, as described in Emmons et al. (2009). The model appears to overestimate the CO produced   Fig. 7 for various regions. The monthly mean retrievals and monthly mean model results (with MOPITT averaging kernels and a priori applied) have been averaged over various continental regions. The error bars on the MOPITT observations, and the ranges indicated for the model results, indicate the range of the data (10th to 90th percentiles) over the region. In general the agreement between MOZART-4 and MOPITT is quite good, as indicated by the correlation coefficients and mean biases (as indicated in Fig. 7). In the Northern Hemisphere, the model appears to be a bit low, indicating perhaps the anthropogenic emissions are underestimated, as that is the dominant source on average. Since MOPITT is most sensitive to the free troposphere, the differences in these comparisons could be a combination of transport (including convection and boundary layer venting) and emissions errors in the model. In the regions with significant biomass burning impact (such as Southeast Asia, Indonesia, and S. America), the model has a positive bias of 7-16 ppbv, indicating the CO emissions from biomass burning may be over-estimated. The positive bias over Australia is consistent with that seen in the NOAA surface sites, particularly Cape Grim (Fig. 5). Further evaluation of the model can be performed for case studies for which there are more detailed measurements, such as field campaigns.

Ozone
An evaluation of the distribution of ozone in MOZART-4 is made through comparison with ozonesondes, which are available from many sites over the past several decades. The observations used here are from the World Ozone and Ultraviolet Radiation Data Centre (Environment Canada, retrieved 25 March 2009 from http://www.woudc.org). The data originate from many sources, including the SHADOZ network (Thompson et al., 2003). The model-data comparisons are summarized in Fig. 8  Monthly mean model results have been extracted for each ozonesonde site, and binned as the observations. The sites of the ozonesonde measurements used in each latitude bin are given in Table 11.  Generally good agreement is seen between MOZART-4 and the observations (Fig. 8), but some systematic differences are apparent. In the lower troposphere of the Southern Hemisphere, MOZART-4 generally underestimates the ozonesondes, whereas it is slightly high in the Northern Tropics (Eq-30N, 900 and 650 hPa). Significant interannual variability is seen in the observations in the Southern Tropics in the mid-troposphere (30S-Eq 650hPa and 400 hPa) at the end of the year, but MOZART-4 underestimates the concentrations, as well as the variation. MOZART-4 also does not fully reproduce the spring-to-summer increase in ozone in the Northern Hemisphere mid-troposphere (30-90N, 650 and 400 hPa), most likely due to the constraint on the stratospheric input through the use of SYNOZ, and the low vertical resolution of this simulation with 28 levels.  Table 11. Each dot indicates the average over the month and all sites in the latitude bin, colored by year. The lines indicate MOZART-4 monthly means, averaged over the ozonesonde locations.

Aerosol optical depth
The aerosol optical depth (AOD) retrievals from MODIS are compared to those calculated by MOZART-4 from the simulated aerosol distributions. Fig. 9 shows averages of MODIS AOD, over oceans only, for various regions, with corresponding averages from MOZART-4. The model agrees very well with the MODIS retrievals over the N. Pacific, which is strongly influenced by the export of pollution from Asia each spring. In May 2003 exceptionally large fires were burning in eastern Russia, resulting in very high AOD values (e.g., Edwards et al., 2004) that the model did not capture. Despite having biomass burning emissions specific for the year (see Sect. 4.1), this indicates the fire emissions inventory underestimated the black carbon or organic carbon emissions. In contrast, the model somewhat over-estimates the AOD over the N. Atlantic Ocean, both in the mean, and in the variability. The AOD from MOZART-4 systematically under-estimates the observations in the Southern Hemisphere. Since the model is too low for all seasons, the washout rates for the aerosols may be too strong in the model. However, the seasonal variation is also under-estimated in these three regions. The annual maxima over the S. Pacific and S. Atlantic Oceans are driven by biomass burning in the second half of the year, with fires in S. America impacting the S. Atlantic, and in Africa and Indonesia impacting the S. Pacific. Therefore, it is likely the Southern Hemisphere biomass burning emissions of aerosols are too low in this simulation. An additional uncertainty in the calculation of AOD lies in the definition of the physical or radiative properties of aerosols, which requires evaluation of the model with simultaneous aerosol size distribution, number density and composition along with radiation measurements.

Conclusions
The offline global chemical transport model for the troposphere MOZART-4 includes a number of updates over the previous version MOZART-2, the most significant being the expansion of the chemical mechanism, with the inclusion of aerosols, and the online calculation of photolysis rates, biogenic emissions and dry deposition. Evaluation with several sets of observations shows that MOZART-4 can reproduce well tropospheric chemical composition. When driven with time varying emission inventories (particularly for biomass burning), MOZART-4 reproduces the spatial and temporal variability in observations, such as the NOAA GMD network and MOPITT CO, ozonesondes and MODIS aerosol optical depth measurements. MOZART-4 is now available to the community and is suitable for many tropospheric investigations on the regional to global scale. The MOZART-4 source code and standard input files are available for download from the NCAR Community Data Portal (http://cdp.ucar.edu).