Inﬂuences of oceanic ozone deposition on tropospheric photochemistry

. The deposition of ozone to seawater is an important ozone sink. Despite constituting as much as a third of the total ozone deposition, it receives signiﬁcantly less at-tention than the deposition to terrestrial ecosystems. Models have typically calculated the deposition rate based on a resistance-in-series model with a uniform waterside resistance. This leads to models having an essentially uniform deposition velocity of approximately 0.05 cms − 1 to seawater, which is signiﬁcantly higher than the limited observational dataset. Following from Luhar et al. (2018) we include a representation of the oceanic deposition of ozone in the GEOS-Chem model of atmospheric chemistry and transport based on its reaction with sea-surface iodide. The updated scheme halves the calculated annual area-weighted mean deposition velocity to water from 0.0464 cms − 1 (25th and 75th percentiles of 0.0461 cms − 1 and 0.0471 cms − 1 respectively) to 0.0231 cms − 1 (25th and 75th percentiles of 0.0121 cms − 1 and 0.0303 cms − 1 respectively). The calculated ozone deposition velocity varies from 0.009 cms − 1 in polar waters to 0.040 cms − 1 at the tropics. This improves comparisons to observations. The variability is driven mainly by the temperature-dependent rate constant for the reaction between iodide and ozone, the temperature dependence of the solubility, and variations in the ocean iodide concentration. The calculated annual deposition ﬂux of ozone to the ocean is reduced from 222 to 122 Tgyr − 1 , and overall deposition of ozone to all surface types reduces from 862 to Tgyr − 1 . Tropospheric ozone burdens and global mean OH increase from 324 to 328 and from . × . × . A total of 34 grid boxes experience a Comparisons between observations of surface and the are with the new parameterization notably around the Southern Process-level representation of of thus ap-pears essential for representing the concentration of surface ozone over the

R. J. Pound et al.: Influences of oceanic ozone deposition on tropospheric photochemistry dominant mechanism (Garland et al., 1980). The transport of ozone within the ocean surface also plays an important role in this process; a simplified version of the relevant processes is shown in Fig. 1.
In addition, dissolved organic carbon (DOC) has been shown to react with dissolved ozone and have an enhancing effect on ozone deposition similar to that of iodide (Martino et al., 2012;Shaw and Carpenter, 2013), but it is less well understood. Dimethyl sulfide (DMS) and bromide have also been shown to enhance ozone deposition velocity but by small amounts (Sarwar et al., 2016). The net flux of a gas to a surface F is calculated as the atmospheric concentration at the ocean surface C multiplied by the deposition velocity, v d , shown in Eq. (1).
The deposition velocity (v d ) in many models is calculated using the resistance-in-series scheme (Wesely and Hicks, 1977) shown in Eq.
(2). This describes the different limiting factors of the deposition: transport to the surface through turbulent transport (r a ); transport through the quasilaminar sub-layer, which is the air directly in contact with a surface (r b ); and the physical, chemical or biological loss of the molecule at the surface (the ocean in this case) (r c ).
v d = 1 r a + r b + r c The relative importance of the different resistances is dependent primarily on the gas being considered. Gases that are highly soluble (such as sulfur dioxide) will have a small r c , so their limiting factors are the atmospheric resistances (r a and r b ). Less soluble gases such as ozone are limited by the chemical loss at the surface (r c ). Wesely (1989) gives a value of r c = 2000 s m −1 for ozone in all water types, and this is used in most atmospheric chemistry models (Hardacre et al., 2015;Luhar et al., 2017Luhar et al., , 2018. This chemical loss of ozone is the limiting factor for ozone deposition (95 % of the sum of the resistances is the value of r c ; Chang et al., 2004) and so yields an almost constant (0.05 cm s −1 ) overall deposition velocity, with only small variation due to meteorological variation in r a and r b . However, observations of ozone deposition show significant variability. From the observations collated by Ganzeveld et al. (2009), fresh water deposition velocities range from 0.01 to 0.1 cm s −1 , with ocean observations ranging from 0.01 to 0.15 cm s −1 . The higher values of ocean observations are likely influenced by coastal effects such as those described by Bariteau et al. (2010), with the open-ocean observations being substantially lower (0.009-0.065 cm s −1 ) (Helmig et al., 2012). Given this observed variability, the fixed r c approach appears overly simple. Based on Fairall et al. (2007) andLuhar et al. (2017), Luhar et al. (2018) formulated a new scheme for calculating r c which explicitly takes into account the simultaneous effects of chemical reactions in the ocean with iodide and the physical processes of molecular diffusion and turbulent transfer in the ocean surface. This considers three oceanic layers (Fig. 1): a very shallow "surface reaction-diffusion" layer that represents the region of the ocean through which the O 3 can diffuse from the ocean before it reacts in the ocean, which lies above a thicker turbulent layer which is mixed by wind-stress-driven turbulence, which in turn lies above the "bulk" ocean. The loss of O 3 is determined by the chemical reactivity within the reactiondiffusion layer, which is supplied by I − from below. The resulting scheme, derived by Luhar et al. (2018), is based on solving the fundamental equation for the conservation of mass of a reacting and diffusing substance in water  and yields Eq. (3).
Here α is the dimensionless solubility, a is the chemical reactivity of O 3 with sea-surface iodide (the product of [I − ] and the second order rate-coefficient, k), D is the diffusivity of O 3 in water, is defined in Eq. (5) where u * w is the waterside friction velocity, δ m is the thickness of the reactiondiffusion layer of the sea-surface microlayer, κ is the von Kármán constant (≈ 0.4), ξ δ defined in Eq. (4), λ is defined in Eq. (6), and K 0 and K 1 are modified Bessel functions of the second kind with order zero and one respectively.
In this paper we include this description of ozone deposition to the ocean in the GEOS-Chem model and explore the impact on the composition of the troposphere. In Sect. 2 we describe the GEOS-Chem model and the implementation of the new scheme. In Sect. 3 we describe the impact of the new scheme on the deposition velocities of ozone to the ocean in the model and assess them against observations of deposition velocities. The impacts of the new deposition scheme on the composition of the troposphere are described in Sect. 4 together with comparison to observations of surface ozone. Finally we draw some conclusions in Sect. 5.

Modelling
We use here version 12.1.1 of the 3-D global chemical transport model GEOS-Chem "Classic" (Bey et al., 2001) (http://www.geos-chem.org, last access: 20 February 2020) driven by assimilated meteorology from the NASA Global Modeling and Assimilation Office. GEOS-Chem includes HO x −NO x −VOC ozone-halogen-aerosol tropospheric chemistry with the halogen (chlorine, bromine and iodine) chemistry being the most recent addition, as described by Sherwen et al. (2016b). In this work we use global simulations run at a spatial resolution of 2 • ×2.5 • with meteorological data from MERRA-2 (Gelaro et al., 2017). Whilst 2 • × 2.5 • is a relatively coarse model resolution, we do not believe that there is any significant sub-grid-scale correlation between tropospheric ozone concentration and seasurface I − concentration; therefore this should not result in a resolution dependence. We run simulations for [2006][2007][2008]2013 and 2014 so that field observations are compared with the appropriate meteorology. Analysis of the sensitivity of the ozone deposition velocity to its controlling factors uses model runs for 2014. For the analysis of the impact on atmospheric composition, a 1-year "spin-up" was used to allow the tropospheric composition to reach equilibrium before the subsequent analysis year. As with many other atmospheric chemistry and transport models, the dry deposition in GEOS-Chem uses a resistancein-series scheme based on that of Wesely (1989). The details of this implementation are described by Wang et al. (1998). For terrestrial land types, the dry deposition in GEOS-Chem is generally consistent with observations (Silva and Heald, 2018).
We follow the Luhar et al. (2018) methodology, and as shown in Eq. (3), this requires the calculation of α, D, k, [I − ] and δ m . Where these require the sea surface temperature (K), T , we use the skin temperature from the MERRA-2 meteorological fields.
We use the dimensionless solubility of ozone in water α from Morris (1988).
We use the diffusivity D (m 2 s −1 ) from Johnson and Davis (1996).
The temperature-dependent k (M −1 s −1 ) for the aqueous phase reactions between ozone and iodide is from Magi et al. (1997).
The reaction-diffusion sub-layer thickness (m) is defined as The waterside friction velocity u * w (m s −1 ) can be calculated from the MERRA-2 atmospheric friction velocity u * using Eq. (11), where ρ a and ρ w are the density of the atmosphere and seawater respectively. This assumes that drivers of atmospheric stress result in an equivalent oceanic stress R. J. Pound et al.: Influences of oceanic ozone deposition on tropospheric photochemistry . . We decide to use the variable definition in our work as this is more physically based and produces comparable results in our simulations. However, it should be noted that using this definition of δ m results in terms cancelling in Eq. (6) such that λ = 1. This thus simplifies Eq. (3) somewhat as sinh(1) ≈ 1.175 and cosh(1) ≈ 1.543. Some of the implications for different choices for δ m are explored in Luhar et al. (2018). Finally, we differentiate between salt and fresh water, using a salinity map from the World Ocean Atlas 2013 (Zweng et al., 2013). The new ozone deposition scheme is only applied to ocean water. Anywhere with water and a salinity below 20 PSU or no salinity value (fresh water) is assigned a constant r c = 2000 s m −1 . One further difference between this work and that of Luhar et al. (2018) is in the global chemistry transport model and its chemistry scheme, GEOS-Chem includes halogen chemistry which has a notable effect on tropospheric ozone (Sherwen et al., 2016b). Any additional computational expense of implementing this improved r c calculation will be small as the deposition velocity calculation remains a two-dimensional problem, unlike the chemistry or transport calculations which are three dimensional problems.
It would be possible to apply this method of calculating r c to other chemical species, if the appropriate sink processes were understood, chemical kinetics available and concentrations of reactant species known. For this to be useful, the species would need to have a high dependence on r c (rather than the physical resistances), but also for dry deposition to form a substantial part of the species budget. It is not clear whether any species, other than O 3 , would meet these requirements.
3 Impact of new parameterization on deposition 3.1 Change in global distribution of deposition velocities Figure 2 shows the annual average global distribution of oceanic ozone deposition velocity for both the standard model and the updated surface resistance scheme, along with the percentage difference between the two. Table 1 gives a statistical description of global ozone dry deposition in the model. The near-uniform value of v d with the standard uniform surface resistance can be observed in Fig. 2a. The small variability in deposition velocity seen is driven by differences in the meteorology. This contrasts with the variability calculated with the new scheme (Fig. 2b). The two schemes also differ in the magnitude of the deposition velocities. The largest change occurs in the coolest waters towards the poles, with the Southern Ocean having a reduction of over 90 % compared to the standard scheme, whereas the tropics can have as little as a 10 % reduction. The distribution of v d is similar to that shown in Luhar et al. (2018), despite our use of the variable thickness for the reactiondiffusion sub-layer and the use of the Sherwen et al. (2019) iodide. On an area-weighted basis, the deposition of ozone to the ocean surface is reduced from 0.0464 cm s −1 (25th and 75th percentiles of 0.0461 and 0.0471 cm s −1 respectively) to 0.0231 cm s −1 (25th and 75th percentiles of 0.0121 and 0.0303 cm s −1 respectively). This amounts to a halving of the mean ocean deposition velocity. The reduction of deposition velocity to the ocean results in a reduction of 17 % in the global average deposition velocity (Table 1). The total annual loss of tropospheric ozone to dry deposition decreases by 104 to 758 Tg yr −1 , substantially lower than the average of 978 ± 127 Tg yr −1 from the multi-model comparison found by Hardacre et al. (2015) but comparable to the value obtained by Luhar et al. (2018) of 722 ± 87.3 Tg yr −1 . The seasonal changes in ozone oceanic deposition velocities from the new annual mean are shown in Fig. 3. This shows the response of the ozone deposition velocity to changes in sea-surface temperature, with the highest value in the summer for each hemisphere and the lowest values occurring in the winter. In the extra-tropical oceans, deposition velocities are predicted to vary by roughly 50 % between summer and winter. Deposition velocities in the tropics remain relatively constant over the year.

Comparison to observations
Here we evaluate the modelled deposition velocities against the open-ocean measurements from Helmig et al. (2012), who measured ozone fluxes to the ocean surface using eddy covariance. These measurements are from a series of five cruises between 2006 to 2008 that took place in the Gulf of Mexico, eastern Pacific Ocean, western Atlantic Ocean and Southern Ocean (Fig. 4). These cruises were made in waters  of significantly different sea surface temperature (SST) and show a trend between deposition velocity and the SST. The comparisons between observations and model were made using daily average values with model output selected from grid boxes that the ship track passed through in that 24 h period. The old scheme (grey line) overestimates the rate of dry deposition substantially and fails to capture any of the temperature dependencies seen in the observations. The new scheme (black line) is a significant improvement, agreeing more with the magnitude and the temperature dependence of the observations. It should be noted that there are significant uncertainties in the measured deposition velocities at low values (Helmig et al., 2012). Combining all the measurements made by Helmig et al. (2012) and comparing to the model predictions for deposition velocity, the root mean square error for the model agreement was reduced from 0.04 cm s −1 using the default scheme to 0.01 cm s −1 using the new scheme. Whilst the overall agreement of the model with the observations has been improved, the model still fails to capture all of the variability of the deposition velocity measurements. This may be an issue with the resolution of the model (2 • × 2.5 • ), which may fail to capture local conditions. Uncertainties in sea-surface iodide concentration or the lack of other sea-surface reactions (reaction between ozone and DOC) may also contribute.

Sensitivity of new scheme
We explore here the sensitivity of the new scheme to our choice of parameterization for u * w , I − , k, D and α. Five model simulations were each run for a year, with only one of the parameters allowed to vary. When constrained, the value of each parameter was set to a representative value of the global average (α, D, k calculated with an SST of 289 K, sea-surface iodide concentration of 106 nM and u *  parameters kept constant at these representative values. The resulting dependence of deposition velocity for each simulation is shown in Fig. 5 as a function of sea surface temperature. If all of the terms needed to calculate r c are kept constant (top left) the oceanic deposition velocity does not vary with temperature. Similarly, if only the water side friction velocity is allowed to vary, no dependence on temperature is seen. Surprisingly the temperature dependence of the iodide concentration is not large, reflecting its square root dependence in the calculation of r c . The two most important  . The response of deposition velocity to different laboratory measurements of k. Three are constant with respect to temperature (Garland et al., 1980;Liu et al., 2001;Hu et al., 1995) and the temperature-dependent parameterization of Magi et al. (1997), with two additional cases of k based on the error range of the Magi et al. (1997) measurements (shown in Eqs. 13 and 12). Each function is produced from global values averaged into 1 K temperature bins, with the shaded region representing the 25th to 75th percentiles.
factors for giving the observed temperature dependence are k and α. Of these two terms, the dependence on rate coefficient carries the most uncertainty. Magi et al. (1997) is the only temperature-dependent rate constant in the literature. Other studies are at single temperatures and show differences . We explore the impact of these differences by running a number of simulations with different values of the rate constants (Fig. 6). We use the single temperature rate constants given by Garland et al. (1980) (2.0 × 10 9 M −1 s −1 at 298 K), Liu et al. (2001) (1.2 × 10 9 M −1 s −1 at 298 K) and Hu et al. (1995) (4.0 × 10 9 M −1 s −1 at 277 K). We also use the upper (Eq. 12) and lower (Eq. 13) estimates of Magi et al. (1997) Figure 6 shows that the uncertainties in k can substantially impact the modelled deposition velocity, with the difference between a temperature-invariant and temperature-dependent k most notable. The temperature-independent rate constants do not correctly simulate the observed temperature variability in deposition velocity. The higher estimate from Magi et al. (1997) overestimates the deposition velocity in warm waters, with the lower estimate underestimating in cold waters. As discussed in Sect. 1 iodide is the dominant but not the only removal mechanism for ozone at the ocean surface.
Given the upper and mid value of the Magi et al. (1997) rate constants there does not appear to be much potential for other oceanic components to play an important role. On the other hand, if the lower values of the Magi et al. (1997) rate constant were correct, this would allow for the inclusion of additional reactions (such as that of ozone with dissolved organic carbon) in the model parameterization without overestimating deposition velocities.
4 Atmospheric impact

Global impacts
The net decrease in deposition of ozone to the surface results in an increase in both surface and column ozone mixing ratios (Fig. 7). The greatest increase in ozone concentration occurs in the boundary layer, with the magnitude of the change decreasing with altitude through the troposphere. The largest increases in the ozone mixing ratio is above the oceans, most notably the extra-tropics, with the Southern Hemisphere extra-tropics being the area of greatest increase. The increase in surface ozone concentration becomes small over land. Surface grid boxes that experience a 10 % increase or greater in ozone mixing ratio represent 34 % of the total surface grid box count. Table 2 gives diagnostics on the oxidative capacity of the troposphere for both the old and new schemes. The increase in ozone mixing ratio shown in Fig. 7 equates to an increase in the tropospheric ozone burden of 4 Tg yr −1 (1.2 %). This affects the global chemical production and loss of O 3 ; however, these changes are globally minimal, at −0.6 % and 1.2 %, respectively. Another consequence of the increased ozone mixing ratio is a small increase in global mean OH concentration of 0.9 % (Table 2), resulting in a decrease in the tropospheric methane lifetime from 8.3 to 8.2 years. Seasonal variations are also observed in the changes in surface ozone mixing ratio due to the new scheme (Fig. 8). The largest increase is observed over the oceans during the winter of each hemisphere due to both the lower deposition velocity that occurs in colder waters and due to the dry deposition playing a larger role in the ozone budget when photolysis is at a seasonal low.

Regional impacts
To assess the predictions of surface ozone mixing ratios in the model, comparisons were made with surface ozone measurements from a number of World Meteorological Organization (WMO) Global Atmosphere Watch (GAW; http:// www.wmo.int/pages/prog/arep/gaw/gaw_home_en.html, accessed through EBAS: http://ebas.nilu.no/, last access: 20 Feburary 2020; the database infrastructure is operated by NILU -Norwegian Institute for Air Research) sites around the world (Fig. 9, shown south to north).
The largest area of change in surface ozone in the model is in the Southern Ocean. GAW sites in this region (Cape Grim, Ushuaia and Neumayer) show increases in ozone prediction during their winter-spring, with the increase most notable in   Schmidt et al. (2016) and Sherwen et al. (2016a) as well as inter-model comparison with ozonesonde observations by Young et al. (2013) show a low bias of GEOS-Chem and other models in the Southern Ocean and Antarctic region. The increased surface ozone mixing ratio brings the model predictions closer to the observations in the Southern Ocean region (Fig. 9), as well as the reductions in root mean square error (RMSE), a measure of disagreement between the model and observations, (Table 3) which is reduced by an average of 44 % across these three locations. Whilst there are considerable improvements in the Antarctic location of Neumayer, surface ozone demonstrate a "lag" in responding to Antarctic spring-summer. The model also fails to capture the springtime halogen-induced ozone depletion events that are observed at Neumayer.  A comparison to a clean tropical location is made using the GAW site in Cabo Verde. Tropical waters are where there has been the least change in ozone deposition velocity, as well as the least increase in ozone mixing ratio both annually and seasonally. Whilst there is a slight increase in predicted ozone compared to the observations at Cape Verde, both the model using the old and the model using new schemes for ozone deposition are within the error of the observations, and there is a small reduction in RMSE.
Mace Head, Ireland, offers an evaluation of model performance in a mid-latitude inflow region, and the inflow of air from the North Atlantic at this site is the dominant component into Europe. Comparing the increase to the observations at Mace Head the improvement is notable, with the model error reduced by approximately 30 %.
The most northerly of the GAW sites in this comparison is the Villum research station in Greenland. There is a minimal increase in predicted surface ozone (∼ 1 ppbv) at this site and the resulting RMSE (Table 3) shows for Villum an increase of 0.3 ppbv with the new parameterization. The observations at Villum also show springtime ozone depletion events and, as with Neumayer, the model fails to capture this.
Overall, the majority of GAW sites show improved comparisons with observations due to the implementation of the new r c scheme and, supporting that, this change is an improvement to the model.

Conclusions
We have implemented a new scheme for the deposition of ozone to the ocean into the GEOS-Chem chemistry transport model based on the work of Luhar et al. (2018). This con-siders the physical and chemical controls of ozone loss in the sea surface. In contrast to Luhar et al. (2018), our work has used a variable surface micro-layer depth and the higher ocean iodide concentrations from Sherwen et al. (2019). The new scheme results in a halving of the global mean ozone deposition velocity to the ocean, leading to a small increase in the global tropospheric ozone burden and some regional increases in ozone mixing ratios of up to 30 % in the high latitude boundary layer, notably around the Southern Ocean. The new scheme improves comparisons between the model and observations in oceanic regions. The increase in tropospheric ozone concentration also has a minor effect on the global mean OH and CH 4 lifetimes.
The new parameterization improves comparisons between the model and observed oceanic dry deposition velocities. However, no account has been made of potential additional processes such as the reaction of O 3 with DOC, DMS and bromide at the ocean surface. Uncertainties in the rate constant for the reaction between I − and O 3 could allow room for such additional reactions to play a role. Reduced uncertainty in the temperature-dependent rate constant for this reaction would be useful. In addition it seems likely that the interaction between DOC and ozone would be complex. It seems likely that some compounds will act as deposition enhancers, whilst others may act as inhibitors (Martino et al., 2012;Shaw and Carpenter, 2013). Further lab, field and modelling studies will be required to better constrain this.
Code availability. GEOS-Chem version 12.1.1 was used in this project: https://doi.org/10.5281/zenodo.2249246 (The International GEOS-Chem User Community, 2018). This code will be available from version 12.8 of GEOS-Chem onwards and can be found at https://github.com/geoschem/geos-chem/tree/master. Code is also available on request.