We present a global chemical transport model called the Integrated Massively Parallel Atmospheric Chemical Transport (IMPACT) model. This model treats chemical and physical processes in the troposphere, the stratosphere, and the climatically critical tropopause region, allowing for physically based simulations of past, present, and future ozone and its precursors. The model is driven by meteorological fields from general circulation models (GCMs) or assimilated fields representing particular time periods. It includes anthropogenic and natural emissions, advective and convective transport, vertical diffusion, dry deposition, wet scavenging, and photochemistry. Simulations presented here use meteorological fields from the National Center for Atmospheric Research (NCAR) Middle Atmospheric Community Climate Model, Version 3 (MACCM3). IMPACT simulations of radon/lead are compared to observed vertical profiles and seasonal cycles. IMPACT results for a full chemistry simulation, with approximately 100 chemical species and 300 reactions representative of a mid-1990s atmosphere, are presented. The results are compared with surface, satellite, and ozonesonde observations. The model calculates a total annual flux from the stratosphere of 663 Tg O3/year, and a net in situ tropospheric photochemical source (that is, production minus loss) of 161 Tg O3/year, with 826 Tg O3/year dry deposited. NOx is overpredicted in the lower midlatitude stratosphere, perhaps because model aerosol surface densities are lower than actual values or the NOx to NOy conversion rate is underpredicted. Analysis of the free radical budget shows that ozone and NOy abundances are simulated satisfactorily, as are HOx catalytic cycles and total production and removal rates for ozone.
 From pre-industrial times, the concentrations of key greenhouse gases such as carbon dioxide (CO2), methane (CH4), nitrous oxide (N2O), and tropospheric ozone (O3) have increased. Simultaneously, concentrations of stratospheric ozone have decreased. While CO2 is an important greenhouse gas, the combined non-CO2 greenhouse gases are also important to the radiative balance of the atmosphere [Intergovernmental Panel on Climate Change (IPCC), 2001]. A key region of interaction between atmospheric chemistry and the climate is the region near the tropopause. This is especially true for ozone and its precursors [Lacis et al., 1990].
 Ozone in the stratosphere is beneficial to the biosphere because it absorbs a significant fraction of the sun's shorter wavelength ultraviolet radiation. Ozone in the troposphere is a pollutant (respiratory irritant in humans and acts to damage crops, vegetation, and many materials). It affects the Earth's energy balance by absorbing both incoming solar radiation and outgoing longwave radiation. Ozone is an important part of the oxidizing capacity of the atmosphere, through a photolysis pathway that leads to the hydroxyl radical (OH). Reaction with OH is the main sink of many atmospheric species, so its concentration controls the distributions of many radiatively important species.
 Ozone in the troposphere arises from both in situ photochemical production and transport from the stratosphere [Danielsen, 1968; Chung and Dann, 1985; Holton et al., 1995]. These two “sources” of ozone vary both spatially and temporally. Within the troposphere, ozone is formed in situ when carbon monoxide (CO), methane (CH4), and non-methane hydrocarbons (NMHCs) react in the presence of nitrogen oxides (NOx = NO + NO2) and sunlight. In contrast, stratospheric ozone formation is initiated by the photolysis of O2 and destroyed via catalytic reactions with NO, H (hydrogen), OH, Cl (chlorine) and Br (bromine), and self-photolysis. Transport of ozone-rich stratospheric air through the tropopause into the troposphere can occur during tropopause folding events [Danielsen, 1968], decay of cutoff weather systems [Bamber et al., 1984; Loring et al., 1996], stratospheric streamers [Appenzeller and Davies, 1992; Appenzeller et al., 1996; Langford and Reid, 1998], and transport across the subtropical jet [Langford, 1999].
 This paper presents a chemical transport model named IMPACT (version 2.0). IMPACT simulates chemical-transport processes, including the important chemical production and loss cycles throughout the troposphere, stratosphere, and tropopause region. This allows for a more physically realistic simulation of both the upper troposphere and lower stratosphere, which includes the chemically and climatically important region around the tropopause. Forthcoming modeling requires this capability to adequately simulate past and future scenarios. Because the computational needs of these tasks are substantial, IMPACT is designed to run on large parallel computers.
 This paper describes the scientific and computational formulation of IMPACT and simulation results for 222Rn/210Pb and photooxidants. The IMPACT model chemistry and physics is described in section 2. Tracer simulations are described and presented in section 3, while section 4 presents selected full photochemistry simulation results. Results and conclusions are discussed in section 5.
2. Model Description
 IMPACT is a global, three-dimensional, chemistry-transport model that contains both a prognostic troposphere and stratosphere. Its input meteorological fields are obtained from either a general circulation model (GCM) or assimilated data, such as that available from the Data Assimilation Office (DAO) at NASA-Goddard. IMPACT uses meteorology from a GCM to address historical, future, and climatological average studies. IMPACT uses assimilated data to simulate specific historical time periods, typically for particular regions of interest. This paper focuses on describing the scientific capabilities of the IMPACT model and general evaluation of model results; all simulations and results presented here were obtained using GCM (see next section on MACCM3) input data. The IMPACT grid resolution is dictated by the input meteorological data, although currently IMPACT is using a version of the advection code [Lin and Rood, 1996] that requires equal latitude gridding. Hence data not on equal latitudes (e.g., Gaussian data) is regridded before use.
 IMPACT is based on an operator-split method for emissions, advection, diffusion, deposition, convection, gravitational settling, photolysis and chemistry. Below, input meteorological fields and the model processes are described in greater detail.
2.1. Input Meteorological Fields
 Currently, IMPACT uses meteorological fields from either the NCAR MACCM3 GCM [Kiehl et al., 1998] or the NASA DAO (Data Assimilation Office) GEOS-STRAT (Goddard EOS Assimilation System-Stratospheric Tracers of Atmospheric Transport) [Coy and Swinbank, 1997; Coy et al., 1997] products. Table 1 lists the meteorological fields and their units used in IMPACT.
Table 1. Meteorological Input Fields Used in IMPACT
Surface air temperature
Downward solar flux at surface
Boundary layer height
Surface friction velocity
Vertical diffusion coefficient
Updraft convective mass flux
kg m−2 s−1
Downdraft convective mass flux
kg m−2 s−1
Entrainment into convective updraft
Entrainment into convective downdraft
Detrainment from convective updraft
Rainfall across cell edges
Total cloud fraction by random overlap clouds
Large-scale cloud fraction
Convective cloud fraction
Turbulent kinetic energy
Surface drag coefficient
Condensed water total
 The vertical structure of IMPACT is based on a hybrid sigma-pressure coordinate system,
in which P is local pressure, i, j, and k are the indices in longitude, latitude, and pressure, PT is a constant unique to the meteorological data set (1000 for MACCM3), Psfc is local surface pressure, and A and B are weighting factors dependent only on the vertical level index, k. The values for A and B are shown in Table 2. Such a general vertical system allows the use of pure sigma input data (such as DAO GEOS-STRAT), pure pressure data, or hybrid combinations of sigma and pressure (such as NCAR MACCM3).
Table 2. List of Coefficients, A and B, Used in the Hybrid Sigma-Pressure Coordinate Systema
Pressure (Psfc = 1000 mbar)
P(i, j, k) = A(k) * PT + B(k) * Psfc(i, j).
 The NCAR MACCM3 input meteorology used for this paper covered 1 year using conditions representing the mid-1990s. MACCM3 ran on a T42 Gaussian grid, giving approximately 3° latitude × 3° longitude horizontal resolution. It has 52 hybrid vertical levels, with the top at 0.006 hPa. The T42 Gaussian grid varies latitudinally from pole to pole and was regridded by NCAR to 4° × 5° horizontal resolution. The IMPACT simulations in this work use this 4° × 5° regular grid. Vertical resolution in the tropopause region is approximately 1.15 km. The fields are cell-centered (commonly referred to as “A-grid”) 3-hour averages.
2.2. Velocity/Pressure Adjustment
 Divergences in the wind field are inconsistent with time derivatives of the surface pressure field in both the DAO and NCAR meteorological fields. This inconsistency arises for two main reasons. First, we need time-averaged mass-fluxes rather than time-averaged or instantaneous winds. Second, the winds may have been re-gridded from the parent GCM grid system. Such inconsistencies between the wind and surface pressure fields inevitably lead to one of the following undesirable consequences: non-conservation of tracer mass, spurious changes in tracer concentration, or spurious changes to the surface pressure distribution. These problems do not arise from a deficiency in the advection routine, which conserves tracer mass and handles tracer concentrations properly. Rather, these problems are inherent in all chemistry tracer models using off-line meteorological fields with inconsistent winds and surface pressures, as discussed by Jöckel et al. .
 In IMPACT, a method related to Prather et al. [1987, Appendix B.2] and Heimann and Keeling [1989, Appendix B], treats the inconsistency between winds and surface pressure. The IMPACT algorithm is simple, fast, and ensures exact equality between the surface pressure change implied by the divergence of the modified winds and the change in the meteorological field's surface pressure for each time step.
 The starting point for pressure-fixer algorithms is the divergence equation,
where F is the vertically integrated mass-flux derived from the meteorological field winds by multiplying the wind speed (m s−1) and the pressure thickness of the layer (Pa) then summing in the vertical (note that this includes linear interpolation from grid box centered values to values on grid box boundaries). F has units of N m−1 s−1. PF (in Pa) is the associated surface pressure calculated via equation (2). An equivalent equation also applies to PG, the surface pressure from the meteorological field, and G, the true mass-flux that gives rise to PG. Ideally we would like to determine G, but it cannot be uniquely calculated because the homogeneous equation ∇ • G = 0 has an infinite number of solutions. Instead, we want to calculate f, the smallest change to F that will satisfy
Although equation (3) does not have a unique solution either, it is much easier to find a solution in which f is small, rather than to ascertain which of the infinite set of solutions for G is the most physically realistic. It is more convenient to re-express equation (3) in terms of the difference between the pressure tendencies of PF and PG, which we call Perr,
Perr can be calculated from the meteorological field variables PG and F via equations (2) and (4) and knowledge of the meteorological field time step. Equation (3) then becomes
IMPACT uses the algorithm described and analyzed by P. J. Cameron-Smith et al. (manuscript in preparation, 2003) to solve equation (5) in a manner that is precise and efficient. The algorithm is as follows:
 1. Subtract any global mean pressure change from Perr.
 2. Remove the zonal mean distribution of Perr through zonally uniform values of f in the meridional direction. Note that the zonal mean of Perr will now be zero for all latitude bands. Note too that this implies the meridional component of f at each pole is zero.
 3. In each latitude band, start with any grid box and set f on its eastern boundary to exactly remove the box's Perr. Then consider the next grid box to the east, and set f on its eastern boundary to remove its Perr plus the value of f on its western boundary, which was determined in the previous step. Iterate for the rest of the grid boxes in the latitude band. Finally, subtract off the zonal mean, so that the zonal mean of f in the zonal direction is zero. Note that this solution does not depend on which grid box is chosen to be the first box.
 4. Distribute the horizontal mass-flux, f, (which is a column integral) between the different horizontal levels in proportion to dB(k), the change in B(k) between the top and bottom of each level (see equation (1) in section 2.1). Since the divergence of f is equal to the needed correction to the surface pressure (by equation (5)), this method of apportioning f ensures that the additional air mass flowing into each grid cell is equal to the increase in its mass implied by the change in surface pressure (Perr), and hence there will be no change to the inferred air mass flowing through the top or bottom of any cell (i.e., no change to the vertical wind in hybrid coordinates).
 This algorithm guarantees that f will exactly satisfy equation (5), as opposed to the algorithm of Prather et al. , which finds an f that only approximately satisfies equation (5). This algorithm generally generates small values for f, as do the other algorithms, but none of the three algorithms guarantees to find the absolutely smallest f, with the caveat that there are several possible definitions for what constitutes the “size” of f.
 Step 1 removes any change in global air mass since equation (5) only has a solution if the global integral of Perr is zero. Strictly speaking then, the algorithm does not solve equation (5), but rather finds a solution that eliminates all of the inconsistency between the meteorological winds and pressures except for any change in global air mass. Hence, once changes due to advection have been calculated, IMPACT uses the modified mass-fluxes (F + f) to determine its surface pressure, rather than PG, in order to avoid causing residual mass-conservation/concentration errors.
 We ignore any changes to the air mass distribution due to sources and sinks of water (precipitation and evaporation from the surface) because the effect is small (water vapor is less than 1% of air mass, and change in column abundance is a small fraction of that) causing changes in the wind speed of a few millimeters per second.
 With this algorithm, tracer mass is now perfectly conserved by advection. We see significant changes for many species, especially in the lowermost stratosphere where ozone is 40–50% higher without the pressure-fixer (which is unreasonably high compared to ozonesonde data; see section 4.3 and Figures 11 and 12). We have not seen any significant changes in the model results attributable to altered horizontal advection patterns.
2.3. Photochemical Solution Technique and Reaction Mechanism
 The photochemical species equations within IMPACT are solved using the SMVGEAR II technique [Jacobson, 1995]. In particular, we use a species-by-species variable time step within our operator-split 1-hour time step to increase computational performance and control the relative and absolute numerical errors. SMVGEAR II also orders grid cells within each node's sub-region into blocks according to similarity in stiffness of the species ordinary differential equations (ODEs), which essentially optimizes the average time step. Block lengths are selected according to problem size and number of processors used.
 The photochemistry includes reactions for both the stratosphere and troposphere. Reactions occurring in the stratosphere include those for Ox, NOy, ClOy, HOy, BrOy, CH4, and their oxidation products. Reactions allowed in the troposphere include those for O3, OH, PAN, NO, NO2, CO, CH4, HNO3, isoprene, ethane, propane, ketones (including acetone), formaldehyde, acetaldehyde, higher aldehydes, and their products [Lurmann et al., 1986]. The mechanism also includes isoprene reactions [Paulson and Seinfeld, 1992], reactions in the remote troposphere [Jacob and Wofsy, 1988] and peroxy radical reactions [Kirchner and Stockwell, 1996]. Where applicable, absorption cross sections and reaction rate coefficients were taken from DeMore et al.  and Sander et al. . This version of IMPACT does not include extensive sulfur chemical reactions, although other versions do. Even though reaction rates may be small away from their region of importance, for example CFC photolysis in the troposphere, all reactive processes are allowed to occur throughout the model domain.
 Substantial laboratory kinetic experiments on isoprene oxidation have been conducted since 1992. In this work, the isoprene mechanism proposed by Paulson and Seinfeld  has been updated by incorporating recently updated isoprene-related reaction rate coefficients, products, and reaction yields, as denoted in Table A1.
 Simulating water vapor provides special challenges, since none of the input meteorological fields contain the important source of stratospheric water from methane oxidation and IMPACT does not include a full predictive hydrologic cycle. Within IMPACT, water is produced photochemically in the stratosphere from CH4 and H2. A local tropopause height is calculated (using local temperature and pressure information) to differentiate these regions (Jim Stobie, personal communication, 1999). This photochemically produced stratospheric water is transported as a separate tracer, thus enabling exchange into the troposphere, and loss by wet deposition. The total water vapor used in the photochemistry subroutines in the troposphere is equal to this water vapor tracer plus the water vapor read in from the meteorological data. The supply of water vapor from the troposphere to the stratosphere, through the so-called “cold trap,” is simulated by enforcing a climatological average 3 μmol mol−1 lower limit on stratospheric water vapor.
 This approach simulates fairly well the observed quantities of water in the stratosphere in the sense of vertical profile and meridional gradient (for example, in comparison with Plate 2b of Harries et al. , with the concentration increasing to about 6.5 μmol mol−1 near the stratopause and in polar air returning to the troposphere. Water vapor mole fraction in air moving from the stratosphere to the troposphere is tracked into the troposphere; this avoids the possibility of artificial drying of air in the lowest stratosphere were the tropopause to descend from one time step to the next. Because the model does not include the complete hydrologic cycle with phase transitions, it naturally does not represent features of observed water vapor fields such as thin laminar layers in the lowest stratosphere, seasonal dependence of the hygropause, and polar dehydration.
 Photolysis frequencies are obtained from a clear-sky lookup table developed using methodologies from Douglass et al. . The rates are adjusted in the troposphere depending on the presence of clouds and the archived cloud fraction in the meteorological fields. Photolysis rates are decreased by a factor of 1 to 0.5 for clear sky fractions of 1 to 0. Although this approach is not considered a replacement for a full radiative transport calculation, it is necessary to account for the global cloud-average albedo of 0.3 used in producing the clear-sky look-up table. Computational performance was also considered.
Table A1 lists the reactions and corresponding rate coefficients. Table A2 gives species names. Additionally, equilibrium constants are listed for six three-body reactions in which the products are thermally unstable at atmospheric conditions. Table A1 also includes six reaction rate coefficients that are expressed using complex mathematical functions (rather than Arrhenius or Troe expressions). For these, the reaction rate coefficient evaluated at 298K (k298K) and the temperature dependence of the activation energy is listed.
 Nine heterogeneous processes are included in the photochemical mechanism, representing hydrolysis of acid anhydrides and chlorine activation. These reactions are listed in Table A3. However, neither the processes affecting aerosol composition or state in the winter polar stratosphere, nor the dehydration or denoxification of the stratospheric polar vortex are included.
 The tropospheric aerosol surface area densities used in N2O5 and NO3 hydrolysis were interpolated from Chuang et al. . These surface area densities combine sulfate, biomass burning, and fossil fuel carbon-containing particles. Reaction probabilities for tropospheric processes do not depend on particle type and are assumed to represent surface interactions with a water coating. Hydrolyzes of N2O5 and NO3 on dilute tropospheric sulfate aerosol to produce nitric acid are simulated as pseudo-first-order processes proportional to aerosol surface area. The pure water reaction probabilities of = 0.05 and = 2.0 × 10−4 [DeMore et al., 1997] are used. These tropospheric aerosol surface densities are smoothly joined to the stratospheric aerosol loading discussed below. The tropospheric reaction representing the slow hydrolysis of NO2 to produce HONO and HNO3 is included (S. Sillman, personal communication, 1997) at a fixed second-order rate constant (4.0 × 10−24) relative to water vapor. This reaction is presumably heterogeneously catalyzed and is an important source for HONO to initiate radical production at sunrise.
 Six of the reactions in Table A3 represent stratospheric hydrolyzes of N2O5, ClONO2, and BrONO2, and chlorine activation through surface-mediated reactions of HCl with ClONO2, HOCl, and HOBr. Climatological distributions of stratospheric liquid binary sulfate aerosol surface area density based on 1995 SAGE II observations [Thomason et al., 1997; World Meteorological Organization (WMO), 1999] are used in the calculation of the collision rate of the gas with the surface. The surface reaction probability, representing the irreversible reactive update of the gas on the aerosol, is temperature and composition dependent. Values are tabulated by DeMore et al. , while the expressions implemented in IMPACT are derived from the experimental literature cited. Reactions between chlorine species, HCl + ClONO2 for example, are treated as “pseudo” second order, by dividing the bimolecular heterogeneous rate constant by the HCl concentration, thus assuming all HCl is in the aerosol droplet at the temperatures for which such reactions are important. This simple approach captures the midlatitude conversion of NOx to HNO3 in the lower stratosphere, as well as a portion of the NOx reduction and chlorine activation that drives the winter polar stratospheric ozone destruction. We achieve some degree of chlorine activation as a result of the nonlinear, negative temperature dependence of the heterogeneous reaction parameters. Rates of Cl activation can be fast enough at sufficiently low temperatures essentially to titrate the reaction partner in smaller abundance. Although presence of the enhanced surface area density provided by PSCs would speed this process, it can be effectively saturated to polar Cl-driven ozone loss.
2.4. Source Emission Inventories
 IMPACT includes monthly (or, in some cases, annual) average emissions of nitrogen oxides (NOx), carbon monoxide (CO), methane (CH4), non-methane hydrocarbons (NMHCs), nitrous oxide (N2O), and chlorofluorohydrocarbons (CFCs). These sources arise mainly from industrial activities including fossil fuel emissions (NOx, CO, CH4, NMHCs, N2O, CFCs), biomass burning (NOx, CO, NMHCs, N2O), vegetation (NMHCs), soils (NOx, N2O), and lightning (NOx). Emissions for the IMPACT model are listed in Table 3.
Table 3. Annual Source Emission Rates in the IMPACT Model
Annual Emission, Tg yr−1
Monthly (M) or Annual (A) Average
1398 Tg CO
506 Tg CH4
Loss via soil absorption
NOx (emitted as NO2)
38.4 Tg N
18.1 Tg N2O
0.115 Tg CFC11
0.15 Tg CFC12
568 Tg C5H8
135 Tg C10H16
40 Tg CH3COCH3
Industrial/fossil fuel combustion
Biogenic (primary and secondary)
15.9 Tg C2H6
Industrial/fossil fuel combustion
17.6 Tg C3H8
Industrial/fossil fuel combustion
2.4.1. Fuel Combustion and Industrial Activity Emissions
 Global emissions of NOx from fossil fuel combustion are compiled on a 1° × 1° horizontal resolution by the International Global Atmospheric Chemistry Project-Global Emissions Inventory Activity (IGAC-GEIA) [Benkovitz et al., 1996] and updated by E. C. Voldner et al. (1° × 1° global SOx and NOx two-level inventory resolved seasonally into emission sectors and point and area emission sources, 1997, available at http://www.ortech.ca/cgeic/poster.html). This inventory gives surface emissions at two vertical levels for four seasons, which we then interpolate to monthly averages. The two levels in the GEIA inventory correspond well with the two lowest vertical levels of the IMPACT model. Aircraft NOx emissions (0.51 Tg N yr−1) are the monthly mean emissions inventory of Baughcum et al.  and Metwally .
 CO emissions arise primarily from biomass burning, fossil fuel combustion, and industrial processes. The 525 Tg CO yr−1 fossil fuel combustion source [Dignon et al., 1998] is distributed seasonally with a 1° × 1° horizontal resolution. This source is consistent with recent fossil fuel combustion estimates of Khalil  and Pacyna and Graedel , which show CO emissions as 380–620 Tg CO yr−1 (best estimate of 500 Tg CO yr−1) and 440 ± 150 Tg CO yr−1, respectively. All CH4 emissions are from Goddard Institute for Space Studies' (GISS) 1° × 1° inventory [Fung et al., 1991; Lerner et al., 1988], with the exception of methane hydrates and clathrates, which were not included due to uncertainties in their source strengths and locations. The industrial sources of ethane and propane are from Watson et al. , as described by Atherton . They are scaled to give annual totals of 8 Tg yr−1 ethane and 12 Tg yr−1 propane [Kanakidou and Crutzen, 1993]. The industrial acetone source of 1 Tg yr−1 given by Singh et al.  is distributed in a manner similar to the CO industrial distribution.
 The annual emissions of CFC11, CFC12, and N2O (including natural N2O sources) are obtained from GEIA [McCulloch et al., 1994; Bouwman et al., 1995]. The mixing ratio of CH3CCl3 in the lowest two model layers is specified to be typical of 1998 conditions [WMO, 2002].
2.4.2. Biomass Burning Emissions
 The biomass burning sources of NOx, CO, and NMHCs are from Atherton  based on the work of Liousse et al. . The sources are appropriate for the early 1990 time period. Since Liousse's emissions are only for tropical latitudes, the biomass burning CO source (which includes biofuel other than the liquid mobile source fuels) also incorporates the work of Dignon et al.  for higher latitudes. Boreal forests fires are included although are a relatively small source compared to low and middle latitudes. The total value of our biomass burning source is 857 Tg CO yr−1. Recent estimates of biomass burning CO range include 260–930 Tg CO yr−1 [Olivier et al., 1996], while Pacyna and Graedel  gives a range of 700 ± 200 Tg CO yr−1. The IMPACT value lies toward the higher end of this range. Other published source estimates include 1000 ± 600 Tg CO yr−1 [Conrad and Seiler, 1986], 875 Tg CO yr−1 [Andreae, 1990], 300–900 Tg CO yr−1 [Bates et al., 1995], 520 Tg CO yr−1 [Wang et al., 1998a], and 661.8 Tg CO yr−1 [Brasseur et al., 1998]. Emissions of CO from biomass burning for the months of January and July are illustrated in Figure 1.
 Biomass burning emissions of acetone total 5 Tg yr−1 [Singh et al., 1994], and are spatially and temporally distributed according to the CO emissions shown in Figure 1. Ethane and propane biomass burning sources [Atherton, 1995] are similarly distributed.
2.4.3. Oceanic Emissions
 Carbon monoxide oceanic emissions are distributed temporally and seasonally based on work by Erickson and Taylor , but with a total global source of 16.5 Tg CO yr−1 rather than 165 Tg yr−1 given in the original work. This value is similar to the recent estimate of Bates et al.  of 13 Tg CO yr−1. Ocean sources of ethane and propane are 0.9 and 1.6 Tg C yr−1, respectively, and are distributed spatially and temporally as for carbon monoxide. These values are slightly larger than the emissions of 0.8 Tg C yr−1 for ethane and 1.1–1.4 Tg C yr−1 for propane estimated by others [Brasseur et al., 1998; IPCC, 2001]. N2O emissions, as stated earlier, are from GEIA, and originate from Bouwman et al. .
2.4.4. NOx From Soil and Lightning
 Nitrogen oxides are emitted during nitrification and denitrification activities by natural microbes that live in the soil. Monthly NOx emissions from soils are from Dignon et al.  and incorporate monthly emission fluxes as a function of vegetation type [Matthews, 1983], temperature, and soil moisture [Willmott et al., 1985]. Soil NOx emissions are shown for January and July in Figure 2. Totaling 5.5 Tg N yr−1, this estimate compares well with the inventory given by Yienger and Levy .
 Atmospheric production of NOx from high temperature N2 fixation during lightning strikes is also included in IMPACT. The lightning source is 5.0 Tg N yr−1 and is horizontally distributed with the location of convective cloud activity according to parameterization of Price and Rind . The vertical distribution of the lightning NOx is specified from cloud convection simulations of Pickering et al. . The emissions are input as monthly mean values.
 Vegetation is a large source of NMHC emissions; however, there is considerable uncertainty in this estimate. IMPACT's isoprene and terpene source distributions are from IGAC-GEIA [Guenther et al., 1995] and total 568 Tg yr−1 and 135 Tg yr−1, respectively. IMPACT uses a biogenic source of acetone of 34 Tg yr−1, of which 18 Tg yr−1 and 16 Tg yr−1 are distributed as isoprene and terpene emissions, respectively.
 Chemical species in IMPACT advect via the Flux Form Semi-Lagrangian Transport (FFSLT) algorithm of Lin and Rood . FFSLT uses upstream differencing to reduce phase errors and contains multiple monotonicity constraints to eliminate the need for a filling algorithm. The scheme's basic building blocks are one-dimensional operators based on high-order Godunov-type finite volume schemes (primarily the third-order piecewise parabolic method, PPM). Multidimensional transport includes explicit consideration of the fluxes associated with cross terms. FFSLT avoids the strict Courant stability problem at the poles by transitioning to a one-dimensional application of a modified semi-Lagrangian algorithm (SLT) for east-west advection (meridional and vertical transport always uses the PPM based scheme; see Lin and Rood  for details). This combination of SLT and PPM allows larger time steps, resulting in highly efficient advection.
 IMPACT can incorporate any linear combination of pure sigma and pure pressure vertical coordinates (as described by equation (1)). Vertical mass fluxes in these coordinates are derived from the three-dimensional horizontal winds in the meteorological fields, conservation of mass, and the hydrostatic equation. This occurs after the horizontal winds have been modified to ensure mass conservation by the pressure-fixer, but by construction our pressure-fixer does not alter the vertical mass fluxes derived here (see section 2.2).
 Mixing of trace species due to sub-grid-scale eddies is modeled only in the vertical direction. IMPACT uses the three-dimensional field of time-varying vertical diffusion coefficients supplied in the meteorological fields. Vertical diffusion coefficients in the troposphere may have values as high as 100 to 1000 m2 s−1, while stratospheric values are approximately 0.01 to 0.1 m2 s−1. These diffusion coefficients are applied via an implicit scheme on each column [Walton et al., 1988]. The scheme conserves mass, will maintain a flat field, and is not subject to time step constraints.
 Transport in convective updrafts is an important mechanism for moving material from near the Earth's surface in the planetary boundary layer (PBL) into the free troposphere. Convective mass-fluxes in each vertical column are supplied in the meteorological fields at interfaces between neighboring vertical layers. The magnitude and spatial variation of these fluxes vary among GCMs, but can be as large as 0.1 kg m−2 s−1 and extend as high as 150 mbar above the surface in the tropics.
 Detrainment is also supplied in the meteorological fields. There can be substantial detrainment at multiple levels in a single convective column or little detrainment, except at the top of the convective column. Entrainment is calculated, to ensure conservation of mass, from the difference between the vertical derivative of convective mass flux and detrainment. Trace species convective transport is carried out using a modified version of the CONVTRANS algorithm [Rasch et al., 1997]. Grid boxes within the PBL are considered well mixed for the convective scheme. The convection algorithm starts at the first grid box above the PBL and moves a fraction of each species upward into the convective updraft (planetary boundary layer height is provided in the input meteorological fields). The trace species mixing ratio within the convective updraft (different than the bulk grid box mixing ratio) is calculated. An equal amount of air is assumed to subside elsewhere in the grid box. Detrainment and entrainment terms then modify the mixing ratio both in the updraft and the bulk grid box. This algorithm marches upward in the column until there is no more convective mass-flux. The algorithm is fast, conserves mass, and will maintain any initial flat fields.
2.8. Wet Deposition
 Species are scavenged dependent on their solubility as described by their Henry's law coefficient. Highly soluble species such as HNO3 and less soluble species such as acetone are transported and removed from the troposphere through a variety of hydrological processes (wet scavenging). Trace species can be incorporated into drops and ice crystals within clouds (rainout), collected by falling raindrops (washout), or be entrained into wet convective updrafts. IMPACT uses the Harvard wet scavenging model [Mari et al., 2000; Liu et al., 2001] that enhances previous models [Giorgi and Chameides, 1986; Balkanski et al., 1993].
 Scavenging within convective updrafts is calculated within the convective transport algorithm. If these were independent operators, soluble species could be transported to the top of the convective column and then dispersed. In each convective column, beginning at the bottom grid box, the fraction of each species scavenged is calculated and directly deposited on the Earth's surface with no chance for re-evaporation [Mari et al., 2000].
 Rainout, washout, and re-evaporation are each calculated for stratiform and convective precipitation. The three-dimensional meteorological field supplies rainfall or precipitable condensation rate, which are separated into convective and stratiform components. When not supplied in this form, rain is separated into components by the three-dimensional fields of convective and stratiform clouds. This module also operates on a column, but unlike the convective updrafts, rainout and washout are calculated from the top down, with the top grid box experiencing precipitation down to the ground. The horizontal area-fraction of each grid box experiencing precipitation is estimated [Giorgi and Chameides, 1986]. A fraction of each species (Fi) lost to rainout is then calculated from this areal fraction (f) and a Henry's Law dependent loss rate (ki) in the form
Aerosols and HNO3 are assumed to be 100% in the cloud condensate phase with ki = 0.005 s−1, while loss rates for other gases are dependent on Henry's law values.
 Washout occurs in grid boxes with no formation of precipitation and where precipitation is liquid. Re-suspension is calculated in all grid boxes with no formation of precipitation.
 Wet deposition of nitric acid (HNO3) is shown in Figure 3. Nitric acid deposition is high near and downwind of NOx source regions that are typically over populated continents.
2.9. Dry Deposition
 IMPACT calculates dry deposition loss rates using the dry deposition algorithm of Wang et al. [1998a], which follows the methodology of Wesely et al. . At each time step and for each surface grid box, this algorithm computes an aerodynamic resistance to deposition, which is dependent on meteorological conditions and surface type. A surface resistance is also calculated which depends on the physical characteristics of each species (Henry's Law coefficient and molecular weight), the meteorological conditions, the surface type, and a seasonal leaf area index [Wang et al., 1998a]. Surface resistance components for the deposition land types are from Wesely , except for tropical forests [from Jacob and Wofsy, 1990] and for tundra [from Jacob et al., 1992]. The module then employs a resistance in series approach [Wesely et al., 1985] to calculate a dry deposition velocity for each species.
 Dry deposition velocities for ozone for July are shown in Figure 4. The highest deposition velocities of 0.5–1 cm s−1 occur in regions of heavy vegetation. This is consistent with measurements and other global models.
2.10. Gravitational Settling
 When modeling aerosol movement in the atmosphere (e.g., lead), gravitational settling can play an important role if the aerosols are relatively large, if the time integration spans several years, or if stratospheric aerosols are involved. The mass-weighted settling velocity for the aerosol distribution is
where the mass, m, is given by
and the velocity, v, is given by
where N and R represent the number and radius (centimeters) of a log normal aerosol size distribution, μ is the viscosity (g cm−1 s−1), Cc is the slip correction factor, and ρ is density (g cm−3) [Seinfeld and Pandis, 1998].
 For a given size distribution, the mass-weighted settling velocity is a function of pressure and temperature. IMPACT uses a lookup table of settling velocities as a function of temperature and pressure that are applied at each time step to move a fraction of the aerosol from the grid box above to the current grid box. The advantages of this method are that the lookup tables are relatively small (6 × 16 entries for each distribution), apply to an arbitrary meteorological data grid, and have a much lower computational cost.
2.11. Computational Description
 Three-dimensional atmospheric chemistry models require large amounts of computational time because of their complexity, fine grid resolution, short time steps, and the need to perform simulations of long duration. To enable multiyear chemistry simulations, IMPACT runs on massively parallel (MP) computational platforms.
 LLNL's computational framework [Mirin et al., 1994] uses a logically rectangular, two-dimensional longitude/latitude domain decomposition, with a computational processor attached to each subdomain. Each subdomain consists of a collection of full vertical columns, spread over a limited range of latitude and longitude. Message passing interfacing (MPI) is used to communicate information between subdomains as necessary. This two-dimensional domain decomposition is efficient because the chemistry and photolysis algorithms take up the vast majority of the computational cycles, are column based, and require no communication with neighboring grid zones. However, this decomposition does impose communication requirements for the horizontal advection operator.
 Nearly all of IMPACT's code is written in FORTRAN 77/95, with a small amount of C. FORTRAN 95's dynamic memory management capability and the MPI message passing interface enhance portability. IMPACT's conditional compilation at high levels in the code allows machine-specific constructs that optimize performance and still maintain portability across many architectures.
 The results presented in this paper were performed on a Compaq-SC1 computer using 36 processors, and required 48 hours to simulate a year. IMPACT's parallel computational efficiency (simulation time multiplied by the number of processors relative to a standard case with 36 processors) is nearly 70% up to 92 processors, yielding a year's simulation in 28 hours.
3. Model Calculations: 222Rn/210Pb
 The gas 222Rn escapes from the ground into the atmosphere and decays into 210Pb, which quickly attaches to any nearby aerosol. Comparing 222Rn model simulations to atmospheric measurements tests a model's ability to accurately move trace species from the surface into the upper troposphere, while comparing modeled 210Pb to observations tests a model's dry deposition, wet scavenging, and long-range transport representations.
 A 222Rn/210Pb simulation using MACCM3 4° × 5° meteorology was performed with the IMPACT model. Beginning with a clean atmosphere, 222Rn was emitted from land surfaces at a rate of 1.0 atom cm−2 s−1. This emission rate was reduced to 0.3 atom cm−2 s−1 when surface air temperature is below 230 K, based on measurements by Larson . This threshold differs from Jacob et al. , who assumed 222Rn emissions decreased to 0.005 atoms cm−2 s−1 between 60° to 70° in latitude, and to zero poleward of 70°. It also differs from the work of Rind and Lerner , who set land emissions to 0.313 atom cm−2 s−1 when surface air temperature was less than 273 K. In sensitivity studies, high northern latitude concentrations of 222Rn predicted by IMPACT were too low if 222Rn emissions were severely limited or zeroed poleward of 60°, in particular, measurements made by Larson in the Yukon Valley showed near-surface measurements of 200 pCi M−3 while IMPACT results showed 20 pCi/M−3. This conclusion is also supported by current non-zero 222Rn emission measurements from snowpack in Maine (C. T. Hess, personal communication, 2001). Emission of radon from the oceans was assumed to be 0.005 atoms cm−2 s−1. Global radon emissions were 15.5 kg yr−1.
 Except for a small loss of radon through dry deposition (about 3%), the bulk of 222Rn decays to 210Pb with a time constant of 5.5 days. 210Pb is stable, quickly attaches to existing atmospheric aerosols, and both dry and wet deposits. Because of its relatively short lifetime, 222Rn reaches a steady state in the atmosphere after only a month or two, while 210Pb requires a much longer simulation time for upper tropospheric and lower stratospheric concentrations to come to steady state. Gravitational settling is included for 210Pb aerosol, as this can influence predicted aerosol concentrations in the stratosphere. Results from the final year of a 4-year simulation are discussed and plotted below.
Figure 5 shows the seasonal variation in 210Pb surface concentrations for six sites, three each from the Northern and Southern Hemispheres [Larsen et al., 1995]. IMPACT captures the seasonal cycle and magnitudes of concentrations for Moosonee (Canada), New York City, Tutuila (Samoa), and Cape Grim (Tasmania). For Antofagasta (Chile), located directly on the western coast of South America, the model results are shown for both the grid box containing Antofagasta, as well as the grid box to the west. These model results bracket the observations, indicating that Antofagasta observations may represent a combination of oceanic and land influenced sampling conditions. Additionally, the exact location of a model grid box may affect predictions near continental boundaries.
 Model results for Mauna Loa, Hawaii, are lower than observations during the first half of the year, including spring, during which the region should experience maximum outflow from Asia [Hoell et al., 1997]. The Mauna Loa site, at an altitude of roughly 3400 m, should experience free tropospheric flow. Analysis of IMPACT results shows the transport of lead from the Asian continent is too weak in the early months of the year resulting in this under prediction of lead. Other global models that also underpredict radon concentrations in the free troposphere [Stockwell et al., 1998] hypothesize that they underestimate convective transport and/or vertical diffusion.
Figure 6a shows the June-July-August average of the zonal mean concentration of 222Rn predicted by IMPACT, which compares well with Figure 5 of Jacob et al.  (a model intercomparison study of the 222Rn cycle). Similar to other models, the IMPACT simulation shows a peak of 20–50 × 10−21 mol mol−1 between 800 and 1000 mbar over the northern midlatitudes, decreasing to roughly 5 × 10−21 mol mol−1 at 200 mbar. Small values of less than 1 × 10−21 mol mol−1 predicted for the middle to upper troposphere in the polar Southern Hemisphere compare well with other models [Jacob et al., 1997]. The asymmetry between the Southern and Northern Hemispheres, due to differences in this land-based source (the Southern Hemisphere has far less land than the Northern Hemisphere) and accentuated by convection in the tropics, is also clearly displayed.
Figure 6b shows the model-predicted, annual average, zonal mean concentration of 210Pb. Similar to 222Rn, there is an asymmetry between the Northern and Southern Hemispheres, due to the asymmetry in the land-based radon source. The model levels of 0.2–0.5 mBq per standard cubic meter (SCM) at 100–200 mbar agree quite well with the observed levels of 0.1–0.5 mBq SCM−1 and an annual average of 0.3–0.4 mBq SCM−1 [Environmental Measurements Laboratory, 2003] for altitudes between 12.2 and 19.2 km. The IMPACT model predicts a local minimum in the upper tropical troposphere, most likely due to scavenging in convective updrafts, as seen in other models [Guelle et al., 1998; Giannakopoulos et al., 1999; Liu et al., 2001].
Figure 6c show the simulated annual total deposition of 210Pb. The maxima are located either directly above or shortly downwind of continental regions, due to the land-based emission of 222Rn and precipitation patterns. The maximum values of 200–250 kg m−2 over eastern Asia and minima of 10–50 kg m−2 over large regions of the southern oceans are similar to those found by an earlier model [Feichter et al., 1991].
Figure 6d shows IMPACT-predicted vertical profiles of radon off the western coast of North America near Moffett Field, California (37.4°N, 122°W). IMPACT predicted 720 hourly profiles of radon for June in this region. The shaded area represents 10–90% of these 720 model-predicted hourly vertical profiles. Also shown are seven individual radon profiles measured by Kritz et al. . Much of the envelope of predicted concentrations lie within those observed. However, several measured profiles have high concentrations of radon between 4 and 12 km. As discussed above, these high measurements in the free troposphere may not be reproduced by this and other models [Stockwell et al., 1998, Barrie et al., 2002] due to the model's radon source emissions or convective transport of radon into the free troposphere.
4. Model Calculations: Full Stratospheric and Tropospheric Photochemistry
 The primary model result we report here is a multiyear, full stratospheric and tropospheric photochemistry simulation with emissions, boundary, and initial conditions representative of the current atmosphere and meteorological fields from MACCM3. Results are presented from the last 12 months of a 40-month run initiated on day 244 (September 1) of a 365-day year. The initialization and length of this run is a compromise among dynamical and photochemical time constants in the troposphere and stratosphere. We specify emissions for many source gases, which enables the model to simulate tropospheric distributions for comparison to observations. It is computationally prohibitive to achieve steady state atmospheric burdens for source gases such as N2O, CFCs, and others relative to their emissions. Thus, although the simulation is not sufficiently long to establish completely internally consistent stratospheric distributions of source gases, distributions and fluxes in the “middle world” of the lower stratosphere and the troposphere should be well represented.
 To help achieve these goals, the model is initialized with zonal mean distributions of source gases and product radical families from the LLNL two-dimensional, zonal model (for a recent description, see Park et al. ). The 2-D model solves the IMPACT photochemical mechanism presented here, but with CH4, N2O, and CFCs set to 1998 abundances at the surface, rather than current emission fluxes. Additionally, the ozone distribution in the 2-D calculation was constrained to the Logan [1999a] zonal climatology developed from sets of observations. Although differences in the mean circulation and eddy diffusivity between 2- and 3-D dynamics will cause a transient in the IMPACT solution at the commencement of the 3-D run, stratospheric net fluxes and photochemical rates should still be characteristic of the solution, without memory of the initial condition. After more than 2 years of spinup, tropospheric quantities other than integrated burden of the source gases should have little dependence on the details of initialization.
4.1. Hydroxyl Radical, OH
 Zonal average tropospheric [OH] predicted for January and July is shown in Figure 7. The OH distribution has a local maximum in the tropical troposphere. In this region, radiation levels and water vapor concentrations, both of which contribute to OH production, are higher. A peak of 1.5 × 106 molecules cm−3 is located over the southern tropics in January. Its size diminishes somewhat in July, as the Intertropical Convergence Zone (ITCZ), a region with high water vapor concentration, shifts northward. Peak OH concentrations in July in the tropical to midlatitude Northern Hemisphere (near 30°N–40°N) reach 2.0–2.5 × 106 molecules cm−3. These higher levels are due partially to emissions of NOx and other ozone precursors from the highly industrialized Northern Hemisphere continents, which lead to higher O3 and, ultimately, higher OH concentrations. The OH concentrations are in very good agreement with other recent global model values [Hauglustaine et al., 1998, Wang et al., 1998b; Lawrence et al., 1999; Bey et al., 2001], although the IMPACT January results tend to be slightly lower than those of Bey et al. .
 The atmospheric lifetime of methyl chloroform, CH3CCl3, can be used as a proxy for the globally averaged hydroxyl radical concentration. Methyl chloroform, an industrial product, is removed predominantly by reaction with tropospheric OH. Its distribution and budget has been carefully characterized [Montzka et al., 2000; Prinn et al., 2001, and references therein]. Atmospheric lifetime can be defined in a variety of ways, including the ratio of the atmospheric burden to the loss rate at steady state [WMO, 1999; IPCC, 2001]. It is determined in IMPACT as the ratio of the total atmospheric mass of CH3CCl3 to the sum of the modeled loss rate processes given the modeled distribution. Because the lifetime is around 5 years, the steady state distribution of CH3CCl3 in the troposphere is essentially well mixed and is not sensitive to the details of tropospheric motions. The IMPACT lifetime should thus be close to the true steady state value.
Montzka et al.  derived a global CH3CCl3 lifetime of 5.2 years (+0.2/−0.3), by observing the decay of the abundance after emissions were reduced due to the Montreal Protocol. The lifetime includes losses due to (1) tropospheric reaction with OH, (2) stratospheric reaction with OH, (3) stratospheric photolysis, and (4) oceanic loss. Values derived from other observational techniques fall within these uncertainty limits. The IMPACT simulation described here includes the first three processes above, but does not include an explicit oceanic loss term. If an oceanic loss lifetime of 94 years (based on work by Yvon-Lewis and Butler ) is assumed, the IMPACT calculated global CH3CCl3 lifetime is 5.3 years. The portion of the loss ascribed to reaction with tropospheric OH (the first process above) produces a lifetime of 6.5 years, while the loss ascribed to stratospheric reaction and photolysis (the second and third processes above) produces a lifetime of 41 years. The IMPACT OH tropospheric loss lifetime of 6.5 years is consistent with the 6.3-year value derived by Montzka et al. .
 This good comparison is not sufficient to “validate” the related chemical and physical processes in IMPACT. For example, IMPACT shows little interhemispheric asymmetry in OH, where the methyl chloroform decay over the last several years is more rapid south of the intertropical convergence zone [Montzka et al., 2000]. However, we can conclude that the IMPACT troposphere possesses a good representation of global average tropospheric OH and that related species and processes that depend on this quantity should also be quantitatively reasonable.
4.2. Ozone Budget
 Global models are useful tools for analyzing the budget of tropospheric ozone. In the troposphere, the main sources are ozone transport from the stratosphere (S) and local in situ photochemical production (P), with the major losses due to photochemical destruction (L) and dry deposition (D), where the bold quantities are global integrals over space and time. Other losses are minor. Within the troposphere, the term P-L represents a very small difference between two large quantities. If tropospheric ozone is in steady state, mass balance gives
 The stratospheric source (S) is essentially independent of the troposphere (depending primarily on stratospheric chemical production and cross-tropopause air mass flux). Therefore, in the atmosphere, the globally averaged deposition (D) and net chemical production in the troposphere (P-L) must respond so as to satisfy equation (10) at steady state (even though the distribution of ozone in the troposphere is strongly affected by in situ production and loss, which are much larger individually than the stratospheric source, deposition, and net chemical production). Thus it is vital for a model to have a good estimate of the stratospheric source (S) in order to correctly determine the net chemical effect of anthropogenic activities on ozone. Recent tropospheric models have varied widely in their stratospheric source (391–1440 Tg O3 yr−1 [IPCC, 2001]), which leads to the following observation: “the large differences in the stratospheric source are apparently the driving force behind whether a model calculates a chemical source or sink of tropospheric O3. Individual CTM studies of the relative roles of stratospheric influx versus tropospheric chemistry in determining the tropospheric O3 abundance. will not represent a consensus until all CTMs develop a more accurate representation of the stratospheric source consistent with observations [Murphy and Fahey, 1994]” [IPCC, 2001].
 Some models have achieved a reasonable stratospheric ozone source by fixing the magnitude of the stratospheric ozone flux in various ways. For IMPACT, in order to model past and future climates when the stratospheric source may be different, we allow the stratospheric source to be calculated prognostically and interactively through chemical production in the stratosphere and advection by the air mass fluxes given in the meteorological fields.
Table 4 lists the IMPACT calculated annual advective ozone flux across seven different pressure levels. The annual flux ranges from 725 Tg O3 yr−1 (across a pressure level of 109 mbar) to 945 Tg O3 yr−1 (across a pressure level of 288 mbar). These fluxes fall well with the range of 391–1440 Tg O3 yr−1 predicted by recent models [IPCC, 2001], although they are somewhat higher than the ranges of 550 ± 140 Tg O3 yr−1 inferred by Olsen et al.  and 450 (range: 200–870) Tg O3 yr−1 estimated by Murphy and Fahey .
Table 4. IMPACT Calculated Annual O3 Flux Advected Across a Single Pressure Surface
Advective Flux, Tg O3/yr
 The monthly total tropospheric O3 budget (S, P-L, D) calculated by the IMPACT model via two methods is shown in Table 5. For the first method (fixed tropopause), the tropopause is assumed to be 150 mbar globally. In the second method (latitudinally dependent tropopause), the tropopause is defined as 93 mbar for latitudes between 40°S and 40°N and 246 mbar for more poleward latitudes. Table 6 lists the annual tropospheric O3 budgets calculated recently by other global CTMs, many of which, however, had fixed stratospheric concentrations or fluxes.
Table 5. Annual Tropospheric Ozone Budget (Tg O3) Calculated Using IMPACT Model
Net In Situ Photochemical Change (Production − Loss)
 Note that IMPACT calculates a net positive term for annual P-L (+17 Tg O3 yr−1 for the fixed tropopause and +161 Tg O3 yr−1 for the latitudinally dependent tropopause cases). This lies well within the range of −855–+600 Tg O3 yr−1 shown in Table 6. However, the term P-L represents a very small difference between two large quantities. The P-L term calculated by the model may increase when higher hydrocarbons are included in the model. The smallest terms for P-L occur during October-April, while the largest occur for May through August. Since ozone is produced by the emission of many land-based precursors in the presence of sunlight, and the Northern Hemisphere has much more land than the Southern Hemisphere, it follows that the Northern Hemisphere summer is a peak time for in situ photochemical production. Dry deposition, which also peaks during Northern Hemisphere summer, is 826 Tg O3 yr−1 for IMPACT, again well within the range of 533 to 1178 Tg O3 yr−1 calculated by other models. Transport from the stratosphere peaks during March-May for IMPACT, corresponding to the traditional peak Northern Hemisphere spring maximum. Note that for the latitudinally dependent tropopause method, the combined sources of stratosphere-troposphere exchange and net in situ photochemical production balance dry deposition to within 1 Tg O3 yr−1 showing that the contribution of convection, diffusion, wet deposition, and convergence to steady state terms are very small.
4.3. Ozone, O3
 Zonal average stratospheric O3 predicted by IMPACT for January and July is plotted in Figure 8, along with UARS-HALOE O3 observations, version 18 [Bruhl et al., 1996; Randel et al., 1998]. The ozone data sets [Bruhl et al., 1996] are compiled in the method described by Randel et al. . The location of peak ozone mixing ratios, at roughly 10 mbar in the tropics, is reproduced well by the model in both January and July. IMPACT predicts a slightly higher peak concentration, by ∼10%. IMPACT also captures the slight northward migration of peak ozone from January to July. Regions of lower ozone concentrations (toward the summer poles, and at very high and very low altitudes) are also simulated well by the model. At the winter poles, the IMPACT/MACCCM3 model appears to isolate the polar mid-stratosphere meridionally more strongly than the UARS/HALOE data shows occurs in the real atmosphere. We believe downward motion at the poles is too dominant over poleward motions, particularly in the Northern Hemisphere winter. We see this feature in other species (e.g., N2O, not shown) as well.
 Surface ozone concentrations for the months of January and July are shown in Figure 9. Peak ozone concentrations are predicted over regions where emissions of ozone precursors (NOx, CO, CH4, NMHCs) and photochemical activity are highest. In January, this occurs primarily in the Northern Hemisphere tropics. In July, ozone peaks over the industrialized Northern Hemisphere continents (due primarily to industrialized emissions) and Southern Hemisphere continents (due primarily to biomass burning). During January, ozone in the Northern Hemisphere has a longer lifetime (decreased solar radiation decreases photochemical destruction) and is transported further across the oceans than during July. This is evident for both the Northern Pacific and Northern Atlantic Oceans.
 Monthly surface ozone measurements made at a number of remote locations [Oltmans and Levy, 1994] are compared to IMPACT predictions in Figure 10. IMPACT predicts concentrations within 10 ppbv and captures the seasonal cycle for the Southern Hemisphere sites of Samoa and the South Pole and Northern Hemisphere sites of Westman Island, Mace Head, Izaña, and Barbados. The model overpredicts ozone concentrations at Barrow during winter, spring, and fall, probably due to ozone depleting surface bromine reactions that are not adequately represented in the model [Foster et al., 2001; Oltmans and Levy, 1994; Barrie et al., 1988]. The model overpredicts ozone at Niwot Ridge during the summer, which is possibly caused by not resolving the typically pristine Niwot Ridge area within IMPACT's 4° × 5° grid. At Bermuda, high observed ozone levels can be associated with transport from the mid-troposphere over North America [Oltmans and Levy, 1994; Moody et al., 1996; Merrill and Moody, 1996]. IMPACT simulations do not match observations at Mauna Loa with the model showing an over prediction of ozone of 20 ppb over the summer/fall season. These elevated ozone levels near the Hawaiian Islands persist to approximately 300 mbar, as shown in the vertical profiles of ozone at Hilo, Hawaii, for summer and fall (see PEM-TROPICS-Hawaii figure in the auxiliary materials). Analysis of IMPACT's stratospheric ozone transport shows large downward transport during this region and period of time resulting in this enhanced tropospheric profile.
 Ozonesonde data for four sites in the NOAA/CMDL network are plotted with IMPACT results for each of the four seasons in Figure 11. Additional plots for other CMDL sites are shown in the auxiliary materials. Ozone mixing ratios are predicted usually within 5–10 nbar for much of the troposphere at the northern latitude site of Resolute during all four seasons. The model predicted stratospheric maximum location and magnitude does vary from observations, though, especially during March/April/May.
 A “midlatitude” site is operated by CMDL at Boulder. The IMPACT model predicts ozone above 100 nbar at the midlatitude site of Boulder to within 15% during all seasons except DJF where it over predicts ozone by 20–30 nbars. While the model predicts ozone levels within 5–10 nbar in the lower troposphere, it tends to overpredict ozone levels between 100 and 400 mbars. The cause of the general problem is likely vertical air mass fluxes from the meteorological fields.
 The IMPACT model predicts ozone at the tropical site of Samoa within 5 nbar for all four seasons throughout the troposphere and stratosphere. Much farther south, at Lauder, the IMPACT model predicted ozone mixing ratios in the lower troposphere agree with observations within 5 nbar for December/January/February and September/October/November. As with Boulder, the model over predicts O3 in the upper troposphere for all four seasons because of excessive transport of stratospheric ozone. The model probably overpredicts ozone at Lauder during September/October/November, however, because phenomena associated with the ozone hole are not completely resolved within the model.
 Ozonesonde data [Logan, 1999a, 1999b] at 600, 200, 100, and 50 mbar from four locations are plotted, together with IMPACT predictions, in Figure 12. Additional plots for other sites are presented in the auxiliary materials. For 600 mbar, at the higher northern latitude site of Resolute, predictions from the IMPACT model agree within 10 ppbv with the mean observations for February to June. The model, however, overpredicts ozone during July–December/January by up to 15 ppbv. At 600 mbar, the model predicts O3 within 10 ppbv at Boulder, Samoa, and Lauder.
 At 200 mbar the IMPACT model represents the seasonal cycle at Resolute well, but predicts ozone concentrations higher than observed by 10–25% during June, July, and August. Observations of ozone over Resolute at 300 mbar show large variability because of the movements in the tropopause which IMPACT is not resolving. Model predictions capture the seasonal cycle very well at Boulder and Samoa. At Samoa, the model and observations agree within 5 ppbv. The largest differences between the model and observations at 200 mbar tend to be at the Southern Hemisphere site of Lauder and across all latitudes at 100 mbar. As discussed above, this is likely due to enhanced transport of stratospheric ozone. At 50 mbar, the model captures the seasonality of ozone at all four sites, while generally predicting slighter higher ozone levels (by 10–15%) than observed.
4.4. Ozone Precursors
 In Figure 13 are plotted profiles of ozone and other species from four different sampling campaigns, ABLE-3A (Alaska: 7 July to 17 August 1988), PEM-West A (Pacific Rim: 16 September to 21 October 1991), TRACE-A (Africa/South Atlantic Ocean: 21 September to 26 October 1992), and PEM-West B (Pacific Rim: 7 February to 14 March 1994) [Emmons et al., 2000]. Additional plots from these and other campaigns are available in the auxiliary materials. The observations correspond to several particular sampling-intensive campaigns in specific seasons and years, while the IMPACT model results are monthly average concentrations obtained using MACCM3 meteorology, representing more of a climatological average.
 The IMPACT ozone concentrations match observations for ABLE3A, but the model slightly underpredicts ozone for TRACE-A, by 10–20 ppb, and overpredicts ozone for PEM-West B for altitudes above 4 km, by up to a factor of 2 above 10 km. This is likely due to the model's placement of the tropopause too low, although the observations do show large variations in ozone above 8 km. The IMPACT model is not able to reproduce the detailed structure in the ozone profile for PEM-West A. The model predicts mean NOx concentrations very well at all four locations. Model HNO3 levels compare well with observations in TRACE-A and PEM-West B, but are too high by a factor of up to 10 at PEM-West A (as are the PAN concentrations). At PEM-West B locations, the model underpredicts PAN levels below 4 km. Although the model captures the H2O2 concentrations well for PEM-West A and PEM-West B, it underpredicts by a factor of 2 the large H2O2 concentrations observed within the bottom 4–6 km for TRACE-A. Additionally, the model underpredicts CO concentrations for 2–4 km for TRACE-A, indicating the model may not mix some species up as high as they are actually lofted. CO concentrations tend to agree with observations for both PEM-West A campaigns, but are slightly low for ABLE and slightly high for PEM-West B. At most locations, the model underpredicts C2H6 concentrations between 30% and a factor of 2, indicating that the model emission source strength is low. A uniform trend is not apparent for C3H8.
4.5. Ozone-Controlling Radical Photochemistry in the Tropopause Region and Lower Stratosphere
 Data on radical abundances from instrumented aircraft flying in the lower stratosphere have yielded much important information on the specifics of this photochemistry. Wennberg et al.  and WMO  investigate ozone removal by radical families for three northern midlatitude locations and between 120 and 60 hPa, based on observations of the abundance of radical family members during the NASA SPADE ER-2 mission in 1993. Absolute removal rates and the relative contributions of radical families are inferred quantities because their derivation depends also on photochemical mechanism assumptions and laboratory-derived kinetic information. IMPACT includes updates to several of the rate constants in the Wennberg et al.  analysis, of which several act to increase somewhat the importance of the NOx radical cycle relative to HOx. Comparing IMPACT zonal monthly averages (Figure 14) for May to Figure 4 of Wennberg et al.  and Figures 7–16 of WMO , the IMPACT simulation produces total ozone loss removal rates (in per cent per month) that show the same trends and magnitudes as the observationally based inferences. While this may partly be the result of a buffering effect on ozone as its loss-controlling processes compete among themselves, it is a necessary precursor to predict ozone in this region.
 In comparing the relative contributions of the radical families in Figure 14 to the 1993 NASA/SPADE results, IMPACT shows a greater importance for NOx-modulated loss cycles, relative to HOx and halogen-modulated cycles. The HOx cycle contribution to the total is roughly the correct magnitude. Changes in preferred kinetic values since the Wennberg et al.  analysis may change the quantitative results of their work. The rate-determining step for NOx mediated ozone destruction is now recommended to be about 15% faster than the recommendation used by Wennberg et al. . While additional recent changes in kinetic parameters of NOx-NOy conversion would support a larger proportional abundance of NOx and therefore a larger role relative to the other cycles, direct observation of NO by Wennberg et al. establishes the NOx abundance independent of NOx-NOy partitioning kinetics in their analysis.
 Other parameters, such as aerosol loading, can still play a role, however. Wennberg et al.  point out that in May 1993, the presence of aerosol surface area from the recent eruption of Mount Pinatubo suppresses NOx abundance relative to available NOy and that NOx could be 20–50% higher in a cleaner stratosphere. The IMPACT simulation uses a relatively clean aerosol loading climatology based on 1995 SAGE II observations over a period with small volcanic perturbation. The sensitivity of the IMPACT simulation to enhanced aerosol loading is a subject of future studies.
Neuman et al.  characterize observations of NO, NOx, HNO3, NOy, and O3 in the lower stratosphere and upper troposphere around 30°N during the NASA ACCENT WB-57 mission in 1999, when the effects of Mount Pinatubo have receded into the background. Using correlations between O3 and HNO3, NOy, and NOx, they distinguish between chemical regimes above and below the tropopause between 7 and 18 km. Because ozone can act as a proxy for the vertical coordinate in these correlations, it is first important to investigate IMPACT's ozone profiles near the tropopause. IMPACT comparisons to ozonesonde profiles are discussed above (section 4.3), with the general result that IMPACT ozone abundance often exceeds the sonde value. These comparisons are performed on long-term temporal averages of both model output and atmospheric observation. Because the IMPACT MACCM3 simulation is not based on data assimilation, specific meteorological conditions, for example, those affecting the ACCENT mission, are not modeled.
 Various approaches to comparing model output to the 22 September 1999 mission results selected for discussion by Neuman et al.  can be taken. Noting that the IMPACT simulation is climatological in nature, zonal or temporal averaging of the model result at the appropriate season and latitude could be argued. IMPACT ozone results averaged in this manner tend to be somewhat larger near the tropopause than the ACCENT observations, which will affect IMPACT O3/NOy correlation plots such as Figure 2 of Neuman et al. . Because temporal and zonal averaging may contribute to making the profile curvature near the tropopause less sharp, synoptic comparisons, even in the absence of an attempt to simulate specific meteorological conditions, may be more appropriate. Figure 15 shows the comparison of IMPACT simulated ozone profiles at the four nearest longitude/latitude grid points (open squares) and their average (thick solid line) to the 22 September ACCENT WB-57 descent profile (jagged line). The vertical coordinate for this comparison is displacement from the local tropopause, reported as 15.3 km for the ACCENT data, and calculated as the minimum in the expression
for the IMPACT profiles (J. Stobie, personal communication, 1999). Very good agreement in ozone abundance and curvature at the tropopause is seen in this figure. The vertical resolution of the MACCM3 grid in this pressure region is shown by the stack of lines on the right of the figure, labeled by vertical index number.
 Using the synoptic output for 30 September (the closest model time saved as a “snapshot” rather than a monthly average) and the matching longitude/latitude grid points, the IMPACT correlations for O3 and HNO3, NOy, and NOy partitioning are shown in Figure 16 (compare to Figure 2 of Neuman et al. ). Figure 15 shows that the O3 values in this figure are in good agreement with the ACCENT observations, so the noticeably larger fraction of NOy that is contributed by NOx at a given O3 value is a strong indication that NOx is simulated to be larger in the lower stratosphere than is observed. This result is reflected in the smaller fraction of HNO3 and the lower HNO3 abundance relative to the observationally derived least squares fit shown in Figure 16b. Total NOy abundance is also lower than observed, so that the model's NOx problem appears to be the result of partitioning problems, rather than NOy source/sink terms. In this case, aerosol loading in 1999 has recovered from the high values following Mount Pinatubo, so, while the climatological values in IMPACT may or may not be too low, the effect should be much smaller than in the 1993 comparison to observations.
 Additional evidence that IMPACT stratospheric NOy is not overpredicted is shown in Figure 17, which shows the IMPACT N2O/NOy and N2O/HNO3 correlation, where the lines represent the linear fits from Figure 3 of Neuman et al. . The slopes represent conversion efficiency of N2O to NOy and IMPACT matches the observations in the lower stratosphere somewhat away from the tropopause. A more global representation is shown in Figure 18, including points from all longitudes between 26°N and 62°N, with pressure below 180 hPa, and N2O between 250 ppb and 310 ppb. The center black line is a fit to results from all months and has a slope of −0.067. The green lines represent the envelope to least squares fits for all months individually. The blue points are for January only and the red points for July only. Olsen et al.  suggest a slope from combined observations of −0.073, or somewhat greater net NOy production from N2O in the stratosphere than IMPACT simulates. We also note slightly smaller peak NOy values in the IMPACT middle stratosphere than satellite and in situ observations.
 This apparent overprediction of NOx leads to an overemphasis on NOx cycles in ozone removal rates. Higher NO2 levels will also suppress active chlorine by converting ClO to ClONO2, and diminish the simulated importance of ClOx in destroying ozone. These behaviors may be symptoms of aerosol surface area densities that are below actual characteristic levels in the lower stratosphere, or NOx conversion and Cl activation rates that are too slow because of errors in kinetic parameters or, possibly, temperatures higher than lower stratosphere ambient.
 In this paper, we present a description of the LLNL IMPACT atmospheric chemistry model, which treats chemical and physical processes in the troposphere, stratosphere, and the climatically critical tropopause region, allowing for physically based simulations of past, present, and future ozone and its precursors. Being able to model the effects of natural and anthropogenic perturbations on ozone in the tropopause region is important because ozone in this region exerts a disproportionately greater influence on radiative forcing than ozone in other atmospheric regions [Lacis et al., 1990; IPCC, 2001].
 IMPACT predicts global, three-dimensional species distributions in the troposphere and the stratosphere using a comprehensive chemical mechanism that includes thermal and photolytic reactions. Detailed modules address surface and elevated emissions, dry deposition, wet scavenging from both convective and large-scale clouds, vertical diffusion, convection, and advection. Stiff ordinary differential equation systems representing chemical equations and rates are solved using SMVGEAR II. The model was designed for, and is exercised on, multiple computer platforms such as UNIX workstations, vector supercomputers, and massively parallel computers (including the COMPAQ SC1, IBM SP, and Cray T3E).
 In model calculations of 222Rn and 210Pb, IMPACT captures the seasonal distribution and magnitude of surface observations of 210Pb at a number of sites. Differences between model predictions and observations at Mauna Loa result from insufficient transport of Asian outflows during the first half of the year.
 A model simulation of a full annual photochemical cycle in the troposphere and stratosphere shows predicted OH and O3 concentrations agree well with observations. Concentrations of OH vary, as expected, with proximity to large water vapor concentrations and sources of O3 precursors (such as NOx, NMHCs, etc.). The IMPACT calculated global methyl chloroform lifetime is 5.3 years, while that calculated using tropospheric OH loss is 6.5 years.
 The stratospheric O3 mixing ratios predicted by IMPACT compare favorably with UARS-HALOE O3 measurements, although polar winter ozone is too low in IMPACT caused by isolation of the polar region. Ozone levels predicted by the model at 200 mbar and 100 mbar, however, tend to be higher than observed by ozonesondes by 10–25%, and up to a factor of 2 in particular locations, but are usually within observational error bars. Within the troposphere, model-predicted ozone concentrations are higher than observed for some high northern and southern latitude sites primarily due to excessive transport of stratospheric ozone. The model predicted ozone concentration in the tropics agrees quite well with observations there.
 The total flux of stratospheric ozone advected through a vertical surface ranges from 663 Tg O3 yr−1 to 806 Tg O3 yr−1, depending on the definition of tropopause. The net annual in situ photochemical production term (production minus loss) is calculated to be 17–161 Tg O3 yr−1, with an annual dry deposition amount of 826 Tg O3 yr−1.
 Comparison to in situ aircraft observations of ozone-controlling radicals reveals that ozone and NOy abundance are simulated reasonably well, as are HOx catalytic cycles and total production and removal rates for ozone. NOx is, however, overpredicted in the lower mid latitude stratosphere, possibly as a result of underpredicting processes converting NOx to NOy. This could be a result of climatological aerosol surface area densities that are specified at levels below actual values, underpredicted conversion rates, or both.
 Here we present the IMPACT photochemical mechanism. Table A1 includes homogeneous gas phase thermal and photolytic reactions, their reactants, products, kinetic parameters, and the literature sources. Table A2 describes all chemical species included in the mechanism. Table A3 includes the heterogeneous thermal reactions, with gas phase reactants and products, which are moderated by the presence of aerosol surfaces.
Table A1. Photochemical Reactions (Gas Phase) Included in IMPACT
Represents a possible homogeneous component for N2O5 hydrolysis. The JPL 97-4 recommended upper limit is 2.0e-21, but Sverdrup et al.  have reported the upper limit chosen here. More recent work also points to a third-order component.
This reaction combines the major actual product from the reaction above, CH3, CH2OH, or HCO, with the fast subsequent reaction with O2 to form CH3O2, CH2O + HO2, or HO2 + CO. This is done to reduce the number of species and size of a mechanism without compromising its accuracy (an implicit photostationary state assumption).
Table A3. Aerosol Heterogeneous Processes Within the IMPACT Model
1. N2O5 → 2HNO3
2. ClONO2 → HOCl + HNO3
3. BrONO2 → HOBr + HNO3
4. HCl + ClONO2 → Cl2 + HNO3
5. HCl + HOCl → Cl2 + H2O
6. HOBr + HCl → BrCl + H2O
7. N2O5 → 2HNO3
8. NO3 → HNO3
9. NO2 + H2O → 0.5 HONO + 0.5 HNO3
 We would like to thank the reviewers of this paper for their many helpful suggestions and comments. This work has been supported in part by the Department of Energy Office of Science through the Climate Change Prediction Program (CCPP), Scientific Discovery through Advanced Computing (SciDAC), and Atmospheric Chemistry Program (ACP) programs, as well as by the Lawrence Livermore National Laboratory internal LDRD program. This work was performed under the auspices of the U.S. Department of Energy by the University of California, Lawrence Livermore National Laboratory, under contract W-7405-ENG-48.