We build our new tropospheric chlorine simulation capability onto the standard version 11-02d of GEOS-Chem (http://www.geos-chem.org, last access: 1 March 2019). The standard version includes a detailed tropospheric oxidant–aerosol–halogen mechanism as described by Sherwen et al. (2016b) and Chen et al. (2017). It includes 12 gas-phase Cly species: Cl, Cl2, Cl2O2, ClNO2, ClNO3, ClO, ClOO, OClO, BrCl, ICl, HOCl, and HCl. It allows for heterogeneous chemistry initiated by SSA Cl- but does not actually track the SSA Cl- concentration and its exchange with gas-phase Cly. Here we add two new transported reactive species to GEOS-Chem to describe Cl- aerosol, one for the fine mode (<1 µm diameter) and one for the coarse mode (>1 µm diameter). The standard GEOS-Chem wet deposition schemes for water-soluble aerosols (Liu et al., 2001) and gases (Amos et al., 2012) are applied to Cl- aerosol and Cly gases, respectively, the latter with Henry's law constants from Sander (2015). Dry deposition of Cl- aerosol follows that of SSA (Jaeglé et al., 2011), and dry deposition of Cly gases follows the resistance-in-series scheme of Wesely (1989) as implemented in GEOS-Chem by Wang et al. (1998). We also add to the model two SSA alkalinity tracers in the fine and coarse modes and retain the inert SSA tracer to derive local concentrations of non-volatile SSA cations (Sect. 2.3). SSA de-bromination by oxidation of Br- as described by Schmidt et al. (2016) was included only as an option in standard version 11-02d of GEOS-Chem because of concern over excessive MBL BrO (Sherwen et al., 2016b). However, Zhu et al. (2018) show that it in fact allows a successful simulation of MBL BrO when one accounts for new losses in the standard model from aqueous-phase oxidation of SO2 by HOBr (Chen et al., 2017) and oxidation of marine acetaldehyde by Br atoms (Millet et al., 2010). We include SSA de-bromination in this work.

We present a 1-year global simulation for 2016 driven by GEOS-FP (forward processing) assimilated meteorological fields from the NASA Global Modeling and Assimilation office (GMAO) with native horizontal resolution of 0.25∘×0.3125∘ and 72 vertical levels from the surface to the mesosphere. Our simulation is conducted at 4∘×5∘ horizontal resolution and meteorological fields are conservatively regridded for that purpose. Stratospheric chemistry is represented using 3-D monthly mean production rates and loss rate constants from a fully coupled stratosphere–troposphere GEOS-Chem simulation (Murray et al., 2012; Eastham et al., 2014).

Table 1 lists the global sources and sinks of tropospheric gas-phase inorganic chlorine (Cly) and reactive chlorine (Cl*) in our model. The main source is mobilization of Cl- from SSA. SSA emission is computed locally in GEOS-Chem separately for fine and coarse modes as the integrals of the size-dependent source function over two size bins, fine (0.2–1 µm diameter) and coarse (1–8 µm diameter). The source function depends on wind speed and sea surface temperature (Jaeglé et al., 2011). We obtain a global SSA source of 3230 Tg a-1 for 2016, corresponding to 1780 Tg a-1 Cl- (assuming fresh SSA to be 55.05 % Cl- by mass (Lewis and Schwartz, 2014), of which 16 % is in the fine mode and 84 % is in the coarse mode. Only 64 Tg Cl- a-1 (3.6 %) is mobilized to Cly by acid displacement and other heterogeneous reactions, while the rest is deposited. A total of 42 % of the mobilization is from fine SSA and 58 % is from coarse SSA. Details of this mobilization are in Sect. 2.3 and 2.4. About 80 % of the mobilization is by acid displacement to HCl, which is in turn efficiently deposited. Only 19 % of HCl is further mobilized to Cl* by reaction with OH to drive chlorine radical chemistry. Direct generation of Cl* from SSA through heterogeneous chemistry provides a Cl* source of comparable magnitude to HCl+OH, with dominant contributions from HOBr+Cl- and N2O5+Cl- (the latter in polluted high-NOx environments).

Cl* can also be produced in the model by atmospheric degradation of the organochlorine gases CH3Cl, CH2Cl2, CHCl3, and CH2ICl. These gases are mainly of biogenic marine origin, with the exception of CH2Cl2, which has a large industrial solvent source (Simmonds et al., 2006). Mean tropospheric lifetimes are 520 days for CH3Cl, 280 days for CH2Cl2, 260 days for CHCl3, and 0.4 days for CH2ICl. Emissions of CH3Cl, CH2Cl2, and CHCl3 are implicitly treated in the model by specifying monthly mean surface air boundary conditions in five latitude bands (60–90, 30–60, and 0–30∘ N, 0–30, and 30–90∘ S) from AGAGE observations (Prinn et al., 2018). Emission of CH2ICl is from Ordóñez et al. (2012), as described by Sherwen et al. (2016a). Tropospheric oxidation of hydrochlorofluorocarbons (HCFCs) is neglected as a source of Cl* because it is small compared to the other organochlorines. The stratospheric source of Cly from chlorofluorocarbons (CFCs), HCFCs, and CCl4 is included in the model on the basis of the Eastham et al. (2014) GEOS-Chem stratospheric simulation as described in Sect. 2.1. Tropospheric organochlorines give a global Cl* source of 3.3 Tg Cl a-1 in Table 1, smaller than that from heterogeneous SSA Cl- reactions (11.9 Tg Cl a-1) or oxidation of HCl by OH (9.7 Tg Cl a-1). The stratosphere is a minor global source of tropospheric Cl* (0.06 Tg Cl a-1) although it could be important in the upper troposphere (Schmidt et al., 2016).

We also include primary HCl emissions from open fires. We apply the emission factors of (HCl+Cl-) from Lobert et al. (1999) for different vegetation types to the GFED4 (Global Fire Emissions Database) biomass burned inventory (van der Werf et al., 2010; Giglio et al., 2013), resulting in a global source of 7.6 Tg Cl a-1 emitted as HCl.

Anthropogenic sources of HCl include coal combustion, waste incineration, and industrial activities. The only global emission inventory is that of McCulloch et al. (1999), which gives a total of 6.7 Tg Cl a-1. As shown in Sect. 4.2, we find that this greatly overestimates atmospheric observations of HCl over the US. National inventories of HCl from coal combustion available for China (236 Gg Cl a-1 in 2012; Liu et al., 2018) and the US (69 Gg Cl a-1 in 2014; US EPA, 2018) are respectively 6 and 7 times lower than estimates of McCulloch et al. (1999) for those countries. We therefore choose not to include anthropogenic HCl emissions in our standard simulation, as they are small in any case from a global budget perspective. We show in Sect. 4.2 that we can account for HCl observations in continental air largely on the basis of the SSA Cl-source. We also do not consider Cl* generation from snow/ice surfaces, which could be important in the Arctic spring MBL (Liao et al., 2014) but is highly uncertain and would only affect a small atmospheric domain.

We do not include the anthropogenic Cl- source from fugitive dust, although it might be important in contributing to chloride levels in continental surface air (Sarwar et al., 2012). The global source of anthropogenic fugitive dust is estimated to be less than 13 Tg a-1 (Philip et al., 2017), of which 0.3 % by mass is estimated to be chloride (Reff et al., 2009). This corresponds to a chloride source of less than 0.39 Tg Cl a-1, negligible on a global scale.

SSA Cl- can be displaced to HCl by strong acids (H2SO4, HNO3) once the SSA is sufficiently aged that its initial supply of alkalinity (≡HCO3-+2×CO32-) has been exhausted. The acid displacement is described by R1Cl-+HNO3⇄HCl+NO3-,R2Cl-+H2SO4⇄HCl+HSO4-, with equilibrium constants from Fountoukis and Nenes (2007). Reaction (R1) must be treated as an equilibrium because HNO3 and HCl have comparable effective Henry's law constants. H2SO4 has a much lower vapor pressure so that Reaction (R2) fully displaces HCl. Additional displacement of HCl by HSO4- does not take place because HSO4- is a much weaker acid than HCl (Jacob et al., 1985).

Alkalinity initially prevents any acid displacement in freshly emitted SSA. Alkalinity is emitted as 0.07 mol equivalents kg-1 of dry SSA (Gurciullo et al., 1999) and is transported in the model as two separate tracers for fine and coarse SSA. It is consumed over time by uptake of acids (SO2, H2SO4, HNO3, and HCl) as described by Alexander et al. (2005), and once fully consumed it is set to zero (titration). The SSA is then diagnosed as acidified, enabling acid displacement by Reactions (R1)–(R2). In our simulation, alkalinity is titrated everywhere shortly after emission except in some areas of the Southern Ocean, which is consistent with the model results of Alexander et al. (2005) and Kasibhatla et al. (2018).

Observations in the MBL indicate that fine SSA is usually internally mixed with sulfate–nitrate–ammonium (SNA) aerosols while coarse SSA is externally mixed (Fridlind and Jacobson, 2000; Dasgupta et al., 2007). Acid displacement for the acidified fine SSA is thus computed by adding HCl/Cl- to the SNA thermodynamics. The local thermodynamic gas–aerosol equilibrium for the resulting H2SO4-HCl-HNO3-NH3-NVC system is calculated with ISORROPIA II (Fountoukis and Nenes, 2007). The calculation is performed assuming an aqueous aerosol even if relative humidity is below the deliquescence point (metastable state). NVCs (non-volatile cations) describe the sum of cations emitted as SSA and is treated in ISORROPIA II using Na+ as proxy. Here NVCs are emitted as 16.4 mol kg-1 of dry SSA to balance the emission of SSA anions including Cl-, alkalinity, and sea salt sulfate. The NVC concentration is determined locally from the mass concentration of the inert SSA tracer.

Acid displacement for acidified coarse SSA is assumed to be driven by uptake of strong acids from the gas phase, mainly HNO3 (Kasibhatla et al., 2018). The ISORROPIA II calculation is conducted with two gas species (HNO3 and HCl) and four aerosol species (NVC, Cl-, SO42-, and NO3-). Here the sulfate includes only the emitted sea salt component and that produced by heterogeneous SO2 oxidation in coarse SSA (Alexander et al., 2005). In the case of coarse aerosols, there may be significant mass transfer limitation to reaching gas–aerosol thermodynamic partitioning (Meng and Seinfeld, 1996). To account for this limitation, the concentrations are adjusted after the ISORROPIA II calculation following the dynamic method of Pilinis et al. (2000). This two-step thermodynamics approach has been used in previous studies (Koo et al., 2003; Kelly et al., 2010).

Table 2 lists the heterogeneous reactions of Cl- other than acid displacement. The loss rate of a gas species X due to reaction with Cl- is calculated following Jacob (2000): 1dnXdt=-rDg+4cγCl--1AnX. Here nX is the number density of species X (molecules of X per unit volume of air), A is the aerosol or cloud surface area concentration per unit volume of air, r is the effective particle radius, Dg is the gas-phase molecular diffusion coefficient of X, c is the average gas-phase thermal velocity of X, and γ is the reactive uptake coefficient, which is a function of the aqueous-phase molar Cl- concentration [Cl-] (moles of Cl- per liter of water). Values of γ in Table 2 are mostly from recommendations by the International Union of Pure and Applied Chemistry (IUPAC) (Ammann et al., 2013).

The heterogeneous reactions take place in both clear-air aerosol and clouds. The GEOS-FP input meteorological data include cloud fraction and liquid/ice water content for every grid cell. Concentrations per cubic centimeter of air of aerosol-phase species (including fine and coarse Cl- and Br-) within a grid cell are partitioned between clear air and cloud as determined by the cloud fraction. Clear-air aqueous-phase concentrations for use in calculating heterogeneous reaction rates are derived from the relative-humidity-dependent liquid water contents of fine and coarse SSA using aerosol hygroscopic growth factors from the Global Aerosol Database (GADS, Koepke, 1997) with an update by Lewis and Schwartz (2006). In-cloud aqueous-phase concentrations are derived using liquid and ice water content from the GEOS-FP meteorological data. Values of r in Eq. (1) are specified as relative-humidity-dependent effective radii for the different clear-air aerosol components (Martin et al., 2003), and are set to 10 µm for cloud droplets and 75 µm for ice particles. These effective radii are also used to infer the area concentrations A on the basis of the mass concentrations. Heterogeneous chemistry in ice clouds is restricted to the unfrozen layer coating the ice crystal, which is assumed to be 1 % of the ice crystal radius (Schmidt et al., 2016).

The heterogeneous uptake of HOBr, HOCl, and ClNO2 as well as further aqueous-phase reaction with Cl- in Table 2 are pH dependent, with a higher efficiency in acidic solutions (Fickert et al., 1999; Roberts et al., 2008; Abbatt et al., 2012). They are considered only when SSA alkalinity has been titrated, and ClNO2+Cl- further requires pH <2, following laboratory results presented in Roberts et al. (2008). The pH of chloride-containing fine aerosol after alkalinity has been titrated is calculated by ISORROPIA II. Liquid cloud water pH is calculated in GEOS-Chem following Alexander et al. (2012), with update to include Cl- and NVC. Coarse-mode SSA and ice cloud pH are assumed to be 5 and 4.5, respectively (Schmidt et al., 2016).

Table 3 compares our simulated Cl- SSA deficits to an ensemble of marine air observations compiled by Graedel and Keene (1995). The Cl- deficit is relative to seawater composition and provides an indicator of the mobilization of Cl- through acid displacement and heterogeneous chemistry. The observations show a wide range from -50 % to +90 %, and Graedel and Keene (1995) emphasize that uncertainties are large. Slight negative deficits in the observations could be caused by titration of alkalinity by HCl but large negative deficits are likely due to error. Mean model deficits sampled for the regions and months of the observations range from +4 to +40 %, not inconsistent with the observations. The largest model deficits are in polluted coastal regions because of acid displacement and this is also where the measured deficits are largest.

Figure 4 compares simulated HCl and HNO3 mixing ratios to concurrent observations of gases both at coastal sites and over oceans. The data are arranged from left to right by increasing latitude. Mean HCl mixing ratios average 323 ppt in the model and 347 ppt in the observations for the ensemble of regions. The HCl source in the model is mainly acid displacement from SSA. A sensitivity simulation without acid displacement from SSA has less than 7 ppt HCl in all regions. The model captures the spatial variability of the mean HCl mixing ratios across locations (r=0.88), which largely reflects the HCl enhancement at polluted coastal sites and northern midlatitudes (Fig. 2). Simulated HNO3 mixing ratios average 190 ppt across locations compared to 137 ppt in the observations, again with good simulation of spatial variability (r=0.96) driven by NOx emissions. HNO3 mixing ratios are sensitive to acid displacement from SSA, as the sensitivity simulation without acid displacement shows mean values of 441 ppt that are much higher than observed. This could partly explain the general model problem of HNO3 overestimation in remote air (Bey et al., 2001).

Figure 5 shows 2016 annual mean observations of PM2.5 Cl- (mass concentration in particles less than 2.5 µm diameter) from the US Interagency Monitoring of Protected Visual Environments (IMPROVE) network (Malm et al., 1994). Corresponding model values shown as background contours for fine Cl- (<1 µm diameter and internally mixed with SNA aerosol) are for total Cl-. One would expect the IMPROVE concentrations to be higher than the model fine Cl- (because of the larger size cut) and lower than total Cl-, and this is generally the case. The model is consistent with observations in the continental interior, which we attribute to inland transport of marine HCl incorporated into SNA aerosol. Fine Cl- concentrations can actually be higher over the continent than over the ocean because of HCl displacement from the coarse SSA followed by recondensation on anthropogenic SNA aerosol. The model underestimates the observations over the Southwest US and this may be due to a missing dust source. We find that IMPROVE Cl- and dust concentrations are positively correlated (R=0.3–0.9) in this region.

A number of surface air measurements have been made of Cl* as the water-soluble component of Cly after removal of HCl (Keene et al., 1990), although most of these measurements are below the detection limit (Table 4, mostly from Keene et al., 2009). This Cl* has been commonly assumed to represent the sum of Cl2 and HOCl (Pszenny et al., 1993) but it would also include ClNO2, ClNO3, and minor components of Cl*. Table 4 shows that simulated Cl* mixing ratios are consistent with the measurements to the extent that comparison is possible. Simulated Cl* over remote oceans is dominated by HOCl, but ClNO2 is responsible for the high values over the Atlantic Ocean near Europe.

Lawler et al. (2009) measured Cl2 and BrCl mixing ratios at Cabo Verde in the tropical Atlantic for 5 days in May–June 2007, and Lawler et al. (2011) measured Cl2 and HOCl mixing ratios at the same site for 7 days in May–June 2009. The observations show a diurnal cycle with mixing ratios of Cl2 highest at night, and HOCl highest in the day, consistent with the model. Observed mixing ratios in background marine air are in the range 0–30 ppt for Cl2 and 0–2 ppt for BrCl at night and 0–5 ppt for HOCl in the daytime. Corresponding mean model values are 0.3 ppt for Cl2, 1.8 ppt for BrCl, and 5 ppt for HOCl, with little day-to-day variability. Lawler et al. (2011) also sampled long-range outflow from Europe for 3 days in 2009 with daytime HOCl and nighttime Cl2 mixing ratio ranges of 40–200 and 5–40 ppt, respectively, but the model does not capture these enhancements. Sommariva and von Glasow (2012) suggested that a lower aerosol pH and/or slower rate for HOCl+Cl- could explain the high HOCl in European outflow but this would also cause Cl2 to be lower. We have no explanation for the high Cl2 values observed by Lawler et al. (2009, 2011) in marine air or for the joint observed enhancements of HOCl and Cl2 in European outflow.

Many surface observations of ClNO2 have been made in nighttime urban environments. These are difficult to compare to the model because of the 4∘×5∘ grid resolution and because of nighttime stratification of the surface layer (the lowest model grid level extends up to 130 m above the surface). In addition, the publications usually report maxima instead of means. Table 5 shows a comparison for representative sites, indicating that the model offers a credible simulation within the above caveats. The previous GEOS-Chem simulation of Sherwen et al. (2016b) only considered ClNO2 production in SSA and as a result their ClNO2 mixing ratios were consistently below a few parts per trillion at continental sites. Our simulation can reproduce the observed >100 ppt concentrations at these sites because it accounts for HCl dissolved in SNA aerosol, allowing marine influence to extend further inland as also shown in the comparison to the IMPROVE Cl- data (Fig. 5).

The WINTER aircraft campaign over the eastern US and offshore in February–March 2015 provides a unique data set for evaluating our model. Measurements included HCl, ClNO2, HOCl, Cl2, and ClNO3 by iodide, time of flight, chemical ionization mass spectrometry (I-TOF-CIMS) (Lee et al., 2018). We focus on the first four measurements because calibration for ClNO3 needs further examination. The mean 1 s detection limits for HCl, ClNO2, HOCl, and Cl2 were 100, 2, 2, and 1 ppt, respectively (Lee et al., 2018). The estimated calibration uncertainty is ±30 % for all chlorine species. As discussed in Lee et al. (2018), labeled 15-N2O5 was added to the inlet tip during WINTER flights to quantify inlet production of ClNO2, which was found to be negligible (≪10 % of measured ClNO2), but inlet production of Cl2, for example from surface reactions of HOCl with adsorbed HCl, was not evaluated.

Figure 6 compares the observed median vertical profiles of HCl, ClNO2, HOCl, and Cl2 in WINTER to the model sampled along the flight tracks for the corresponding period. Figure 7 compares the median diurnal variations below 1 km in altitude, separately over ocean and land. We exclude daytime (10:00–16:00 local) data for ClNO2 in Fig. 6 because its mixing ratios are near zero (Fig. 7).

The WINTER observations of HCl show median values of 380 ppt near the surface, dropping to a background of 100–200 ppt in the free troposphere (Fig. 6). The model is lower than the observations in the lowest 2 km but within the calibration uncertainty. The free tropospheric background in the model is much lower than observed but the observations are near the 100 ppt detection limit. HCl mixing ratios in the lowest kilometer average 60 % higher over ocean than over land in both the observations and the model, reflecting the marine source.

Also shown in Fig. 6 is a sensitivity simulation including anthropogenic HCl emissions from McCulloch et al. (1999) as described in Sect. 2.2. The resulting model mixing ratios are too high though still within the calibration uncertainty. Based on sampling of power plant plumes during the WINTER campaign, Lee et al. (2018) inferred a HCl:SO2 emission mass ratio of 0.033 from power plants. Adding this emission to the standard simulation scaled to the SO2 emissions in GEOS-Chem (from the EPA National Emission Inventory over the US) increases modeled mixing ratios of HCl in the continental boundary layer by 18 % along the WINTER flight tracks, improving agreement with observations relative to the standard simulation but still representing a relatively minor source. ClNO2, HOCl, and Cl2 mixing ratios increase by 12 %, 8 %, and 4 %, respectively.

Figure 8 compares the model HCl vertical profiles to measurements by the Georgia Tech CIMS instrument during the SEAC4RS campaign over the Southeast US in August–September 2013 (Toon et al., 2016), the KORUS-AQ campaign over and around the Korean peninsula in May–June 2015, and the INTEX-B campaign over the North Pacific in May 2006 (Kim et al., 2008). The standard model simulation without anthropogenic chlorine successfully simulates the boundary layer HCl observations during SEAC4RS and KORUS-AQ but adding the McCulloch et al. (1999) anthropogenic inventory results in large overestimates. Boundary layer HCl mixing ratios over land are much lower in SEAC4RS than in WINTER and this is well reproduced by the model, for which the difference is due to seasonal contrast in the SSA source and in the inflow of marine air. The free tropospheric background observed in SEAC4RS, KORUS-AQ, and INTEX-B data is only ∼25 ppt, much lower than in WINTER (100–200 ppt), whereas the model free tropospheric background is consistently 20–50 ppt in all four campaigns. The WINTER observations are near their 100 ppt detection limit as pointed out above. The modeled vertical profile of HCl during INTEX-B is very similar to that of KORUS-AQ but overestimates observed boundary layer HCl by a factor of 5. The median value of observed surface HCl mixing ratio is 24 ppt during INTEX-B, which is much lower than all the surface air observations in Sect. 4.1.

Mixing ratios of ClNO2 observed in WINTER are above the detection limit only in the lowest kilometer of atmosphere at night and are much higher over the ocean than over land. This is well simulated by the model (Figs. 6 and 7) and reflects the nighttime source from the N2O5+Cl- heterogeneous reaction combined with the fast loss by photolysis in daytime. Previous studies have suggested that the Bertram and Thornton (2009) representation of the ClNO2 production yield from the N2O5+Cl- heterogeneous reaction in Table 2 is too high (Riedel et al., 2013; Wagner et al., 2013; McDuffie et al., 2018a). By using a box model applied to the WINTER observations, McDuffie et al. (2018a, b) found that both the N2O5 uptake rate and the ClNO2 production yield were overestimated by the Bertram and Thornton (2009) parameterization. One important difference with these previous studies is the assumption of aerosol mixing state. GEOS-Chem assumes that Cl- is present only in SNA and SSA when doing the calculation of N2O5 reactive uptake rates, assuming an external mixture of aerosol types (Martin et al., 2003; Evans and Jacob, 2005; Jaeglé et al., 2018). This decreases both the N2O5 uptake rate and the ClNO2 yield compared to using the parameterization of Bertram and Thornton (2009) on internally mixed aerosols. Jaeglé et al. (2018) reported that using this external mixing assumption leads to good agreement between modeled and observed N2O5 during the WINTER campaign. The mean ClNO2 yield (φ in Reaction R3) below 1 km in our model is 0.20 during the WINTER campaign, very close to that calculated from observation (0.218) by McDuffie et al. (2018a).

Nighttime Cl2 mixing ratios in WINTER are greatly overestimated by the model. Under polluted wintertime conditions such as in WINTER the ClNO2+Cl- reaction greatly enhances Cl2 production in the model:

R7ClNO2+Cl-→NO2-+Cl2. The reactive uptake coefficient for Reaction (R7) in Table 2 is based on a single laboratory study (Roberts et al., 2008). It requires an aerosol pH <2 and this condition is generally met for our model simulation of the WINTER environment, consistent with the observation-based analysis of aerosol pH by Guo et al. (2016) for the eastern US in winter. A sensitivity simulation without Reaction (R7) is shown as dashed red lines in Fig. 7 and can reproduce the low Cl2 mixing ratios observed over the ocean at night. The analysis of WINTER data by McDuffie et al. (2018b) finds that the correlation between particle acidity and Cl2 observations is opposite of the trend expected from Reaction (R7), though there may be no trend on sufficiently acidic (i.e., pH <2) aerosol. Further study of that reaction is needed. Similar reactions such as ClNO2+Br- (Reaction R8) and ClNO2+I- are also very uncertain as their reaction rate coefficients are very different in different studies (Fickert et al., 1998; Frenzel et al., 1998; Schweitzer et al., 1998).

The model underestimates the WINTER observations of HOCl and Cl2 in daytime, over the ocean as well as over land. These species have short lifetimes due to photolysis (less than a few minutes). Direct anthropogenic emission from coal combustion has been proposed (Chang et al., 2002) but would only be observed in plumes and not over the oceans. Matching the >1 ppt Cl2 observed during daytime is particularly problematic since it would require a large photochemical source absent from the model. Lawler et al. (2011) suggested a fast daytime HOCl source from a hypothetical light-dependent Cl- oxidation. The measurements of Cl2 are also possibly subject to a positive artifact from rapid heterogeneous conversion of chlorine species on the surface of the TOF-CIMS inlet (Lee et al., 2018).

The global mean pressure-weighted tropospheric Cl atom concentration in our simulation is 620 cm-3, while the MBL concentration averages 1200 cm-3 (Fig. 2). Our global mean is lower than the previous global model studies of Hossaini et al. (2016) (1300 cm-3) and Long et al. (2014) (3000 cm-3), which had excessive Cl* generation as discussed above. It is consistent with the upper limit of 1000 cm-3 inferred by Singh et al. (1996) from global modeling of C2Cl4 observations (C2Cl4 is highly reactive with Cl atoms). Isotopic observations of methane have been used to infer a Cl atom concentration in the MBL higher than 9000 cm-3 in the extratropical Southern Hemisphere (Platt et al., 2004; Allan et al., 2007), much higher than our estimate of 800 cm-3 over this region. More recently, Gromov et al. (2018) revisited these data together with added constraints from CO isotope measurements and concluded that extratropical Southern Hemisphere concentrations of Cl atoms in the MBL should be lower than 900 cm-3, consistent with our estimate.

Tropospheric oxidation by Cl atoms drives a present-day methane loss rate of 5.3 Tg a-1 in our model, contributing only 1.0 % of total methane chemical loss. It has a more significant impact on the oxidation of some other VOCs, contributing 20 % of the global loss for ethane, 14 % for propane, 10 % for higher alkanes, and 4 % for methanol.

Bromine radicals (BrOx≡Br+BrO) and iodine radicals (IOx≡I+IO) affect global tropospheric chemistry by depleting ozone and OH (Parrella et al., 2012; Sherwen et al., 2016b). Br atoms are also thought to drive the oxidation of elemental mercury (Holmes et al., 2006). Chlorine chemistry increases IOx mixing ratios by 16 % due to the reactions of HOI, INO2, and INO3 with Cl- (Reactions R11–R13), producing ICl, which photolyzes rapidly to I atoms (Fig. 1). The effect on bromine is more complicated. Bromine radicals originate from photolysis and oxidation of organobromines emitted by the ocean, as well as from SSA de-bromination (Yang et al., 2005). They are lost by conversion to HBr, which is efficiently deposited. Parrella et al. (2012) pointed out that heterogeneous chemistry of HBr (dissolved as Br-) is critical for recycling bromine radicals and explaining observed tropospheric BrO mixing ratios in the background troposphere: R14HOBr(aq)+Br-+H+→Br2+H2O,R15Br2+hν→2Br. Chloride ions and dissolved SO2 can however compete with Br- for the available HOBr (Chen et al., 2017): R5HOBr(aq)+Cl-+H+→BrCl+H2O,R16HOBr(aq)+HSO3-/SO32-→HBr+HSO4-/SO42-. Chen et al. (2017) pointed out that Reaction (R16) effectively decreases BrO mixing ratios by producing HBr, which is rapidly deposited instead of contributing to BrOx cycling. They found in a GEOS-Chem simulation that global tropospheric BrO mixing ratios decreased by a factor of 2 as a result. Reaction (R5) may however have a compensating or opposite effect. It propagates the cycling of BrOx if BrCl volatilizes: R17BrCl+hν→Br+Cl, but it may also generate new BrOx if BrCl reacts with Br- in the aqueous phase to produce Br2 (Wang et al., 1994): R18BrCl(aq)+Br-⇄Br2Cl-,R19Br2Cl-⇄Br2(aq)+Cl-. The sequence Reactions (R5) + (R18) + (R19) with Cl- as a catalyst has the same stoichiometry as Reaction (R14) and thus contributes to HBr recycling in the same way. We find in the model that it is globally 30 times faster than Reaction (R14) and therefore much more effective at regenerating bromine radicals. In GEOS-Chem, the rate of Reaction (R5) computed from Table 2 is applied to the following stoichiometry reflecting the ensemble of Reactions (R5), (R14), (R18), and (R19): HOBr(aq)+YBr-+1-YCl-+H+→YBr2R5,R14,R18,R19+(1-Y)BrCl+H2O, where Y is the yield of Br2 and 1-Y is the yield of BrCl. Y is calculated following the laboratory study of Fickert et al. (1999). Y=0.41log10([Br-]/[Cl-])+2.252for[Br-]/[Cl-]<5×10-43Y=0.90for[Br-]/[Cl-]>5×10-4 This mechanism was first included in GEOS-Chem version 11-02d by Chen et al. (2017), who did not however have an explicit SSA Cl- simulation (they instead assumed a fixed SSA [Cl-] =0.5 M, and considered only dissolved HCl in cloud).

Chen et al. (2017) found in their GEOS-Chem simulation that the global tropospheric BrO burden was 8.7 Gg without the HOBr+S(IV) Reaction (R16) and dropped to 3.6 Gg when that reaction was included. Previous GEOS-Chem model estimates of the global tropospheric BrO burden were 3.8 Gg (Parrella et al., 2012), 5.7 Gg (Schmidt et al., 2016), and 6.4 Gg (Sherwen et al., 2016b). Our simulation features many updates relative to Chen et al. (2017), including not only explicit SSA Cl- but also explicit calculation of aerosol pH with ISORROPIA II for the rates of reactions in Table 2. By including explicit SSA Cl-, the cloud water [Cl-] in our model is much higher than that in Chen et al. (2017) and more comparable to measurements (∼10-4 M in typical cloud; Straub et al., 2007). We find in our standard simulation a global tropospheric BrO burden of 4.2 Gg, 17 % higher than that of Chen et al. (2017).

Figure 9 shows the change in surface BrO mixing ratios due specifically to tropospheric chlorine chemistry, as obtained by difference with a sensitivity simulation including none of the Cly chemistry shown in Fig. 1. The inclusion of chlorine chemistry increases the global tropospheric BrO burden by 85 %. More than 80 % of this change is caused by the HOBr+Cl- reaction as discussed above. Other significant contributions include ClNO3+Br- and ClNO2+Br-. The largest BrO increases (1–2 ppt) are in surface air over the high-northern-latitude oceans where SSA emissions are high and acidic conditions promote HOBr+Cl- chemistry.

Figure 9 also shows the effects of chlorine chemistry on NOx, OH, and ozone concentrations. The global tropospheric burdens decrease by 5 % for NOx, 3 % for OH, and 7 % for ozone. The interhemispheric (N/S) ratio of tropospheric mean OH decreases from 1.14 to 1.12. Models tend to overestimate global mean tropospheric OH and its interhemispheric ratio relative to the constraint from methyl chloroform, which suggests a ratio of 0.85–0.98 (Naik et al., 2013; Voulgarakis et al., 2013). The effect of chlorine chemistry on the N/S ratio is slight but in the right direction.

The chlorine-induced decreases in Fig. 9 are mainly through bromine chemistry initiated by chlorine (Sect. 5.2) and have spatial distributions characteristic of bromine chemistry with maxima at high latitudes as discussed by Schmidt et al. (2016). There are specific chlorine mechanisms including catalytic ozone loss through HOCl formation and photolysis: R20Cl+O3→ClO+O2,R21ClO+HO2→HOCl+O2,R22HOCl⟶hνCl+OH,Net:O3+HO2→OH+2O2 and also loss of NOx: R23ClO+NO2+M→ClNO3+M,R24ClNO3+H2O→HOCl+HNO3. However, we find that the rates are very small compared to similar mechanisms involving bromine and iodine because the stability of HCl quenches Cl* radical cycling.

A particular situation arises over polluted continents due to ClNO2 chemistry. Production of ClNO2 at night from the N2O5+Cl- heterogeneous reaction, followed by photolysis in the morning to release Cl and NO2, provides a source of radicals and ozone. This explains the increases in OH over North America and Europe in Fig. 9. The effect is most important at high northern latitudes in winter due to the longer night. To isolate the impact on ozone we conducted a sensitivity simulation with no ClNO2 production, setting φ=0 for Reaction (R3) in Table 2. The surface air ozone enhancement due to ClNO2 chemistry is found to be the largest (∼8 ppb) in European winter, due to the large supply of Cl- from the North Atlantic combined with high NOx emissions. Other polluted continents see ozone increases of 1–5 ppb in winter. The effect in summer is less than 1 ppb. These results are similar to previous regional modeling studies by Sarwar et al. (2014) and Sherwen et al. (2017).

Figure S1 shows the differences of BrO, NOx, OH, and ozone concentrations between our model and the standard GEOS-Chem model version 11-02d including SSA de-bromination. Our explicit treatment of chlorine chemistry and thermodynamic representation of aerosol pH increases the global tropospheric BrO burden by 40 %. Most of this change is caused by faster HOBr+Cl- reaction at high latitudes, resulting from higher Cl- concentration in our model, particularly in cloud. The decrease in BrO in the tropical MBL is caused by an increase in aerosol pH (pH was previously assumed to be 0 for computation of bromine chemistry), which slows down the acid-catalyzed recycling of bromine by Reactions (R5) and (R14). Our computed global tropospheric burdens decrease by 4 % for NOx, 2 % for OH, and 4 % for ozone relative to version 11-02d, again due to the more active bromine chemistry. The increase in OH over continental regions is due to our accounting of HCl dissolved in SNA aerosol, allowing marine influence to extend further inland to drive ClNO2 chemistry.