Monthly resolved modelled oceanic emissions of carbonyl sulfide and carbon disulfide for the period 2000-2019

. Carbonyl sulfide (OCS) is the most abundant, long-lived sulphur gas in the atmosphere and a major supplier of 10 sulfur to the stratospheric sulfate aerosol layer. The short-lived gas carbon disulfide (CS 2 ) is oxidized to OCS and constitutes a major indirect source to the atmospheric OCS budget. The atmospheric budget of OCS is not well constrained due to a large missing source needed to compensate for substantial evidence that was provided for significantly higher sinks. Oceanic emissions are associated with major uncertainties. Here we provide a first, monthly resolved ocean emission inventory of both gases for the period 2000-2019 (available at https://doi.org/10.5281/zenodo.4297010)(Lennartz et al., 2020a). Emissions are 15 calculated with a numerical box model (resolution 2.8° x 2.8° at equator, T42 grid) for the surface mixed layer. We find that interannual variability in OCS emissions is smaller than seasonal variability, and is mainly driven by variations in chromophoric dissolved organic matter (CDOM), which influences both photochemical and light-independent production. A comparison with a global database of more than 2500 measurements reveals overall good agreement. Emissions of CS 2 constitute a larger sulfur source to the atmosphere than OCS, and equally show interannual variability connected to variability 20 of

contexts: first, OCS is transported to the stratosphere due to its long tropospheric lifetime of 1.5 to 3 years (Montzka et al., 2007), where it is a major precursor of sulfate aerosols (Brühl et al., 2012;Kremser et al., 2016;Turco et al., 1980). The stratospheric sulfate aerosol layer influences the radiative budget by increasing the planetary albedo, and in addition provides surfaces for ozone catalysing reactions (Solomon et al., 2011(Solomon et al., , 2015. Second, OCS has been suggested as a promising proxy to constrain the terrestrial CO2 uptake on a global scale using inverse atmospheric modelling (Berry et al., 2013;Stimler et al., 35 2010;Whelan et al., 2018). In order to understand the dynamics of the sulfate aerosol layer and to apply OCS as a proxy for gross primary production, the quantification of OCS sources and sinks to the atmosphere on a global scale is required.
Currently, oceanic emissions are associated with the highest uncertainties among sources in the atmospheric OCS budget (Kremser et al., 2016;Whelan et al., 2018). Evidence for increasing the vegetation sink led to a missing source the budget (Suntharalingam et al., 2008), and oceanic emissions have been suggested to account for a gap of 600-800 Gg S yr -1 (Berry et 40 al., 2013;Glatthor et al., 2015;Kuai et al., 2015). Global oceanic emission estimates extrapolated from measurements range from -16 Gg S yr -1 (Weiss et al., 1995b) to 320 Gg S yr -1 (Rasmussen et al., 1982). Surface ocean models that are largely in agreement with observations report direct OCS emissions from the oceans of 41 Gg S yr -1 (Kettle et al., 2002) to 130 Gg S yr -1 (Lennartz et al., 2017). Generally, surface seawater concentrations of OCS are too low to sustain emissions that would close the budget (Lennartz et al., 2017(Lennartz et al., , 2020b. A detailed description of the marine emissions of OCS and its precursor CS2 can 45 serve as an input to modelling studies, and thus help to identify the missing source. However, current marine emission inventories rely on modelled climatological averages, and do not resolve interannual variability (Kettle et al., 2002;Lennartz et al., 2017).
Models resolving the marine cycling of both trace gases are powerful tools to assess interannual variability of marine emissions through variations in the factors influencing production and consumption of the gas in seawater. The processes determining 50 OCS concentration in the surface ocean are comparably well understood, and include a photochemical production process, a light-independent dark production term, degradation by hydrolysis and air-sea exchange. Diapycnal fluxes into and out of the mixed layer seem to be of minor importance, at least in tropical waters (Lennartz et al., 2019). The photochemical OCS production involves UV-radiation interactions with chromophoric dissolved organic matter (CDOM) (Ferek and Andreae, 1984;Modiri Gharehveran and Shah, 2018;Pos et al., 1998). Apparent quantum yields (AQY) decrease with increasing 55 wavelength, but show orders of magnitude differences between locations (Cutter and Radford-Knoery, 1993;Weiss et al., 1995a;Zepp and Andreae, 1994). Reaction mechanisms involving thiyl radicals have been identified from precursor molecules such as cysteine, cystine and methionine (Modiri Gharehveran and Shah, 2018;Pos et al., 1998). However, the complexity of the natural mixture of dissolved organic sulfur molecules in the ocean (Ksionzek et al., 2016) makes the determination of a photoproduction rate constant on a global scale difficult. Following an approach initially suggested by von Hobe et al., (2003), 60 the photoproduction rate constant was scaled according to the CDOM absorption coefficient at 350 nm (a350) in the global surface ocean box model used in this study (Lennartz et al., 2017). This approach led to good agreement of climatological mean modelled concentration with measured sea surface OCS concentrations. The mechanism for OCS dark production is not well known, and two not mutually exclusive hypotheses have been suggested, i.e. dark production being connected to abiotic https://doi.org/10.5194/essd-2020-389  (von Hobe et al., 2001) or microbial remineralisation processes (Cutter et al., 2004). The dependency of the 65 dark production rate on CDOM absorption and temperature shows good agreement across various biogeochemical regimes (Lennartz et al., 2019). Hydrolysis is the main chemical sink for OCS in the mixed layer. In both an acid and an alkaline reaction, OCS hydrolysis yields CO2 and sulfide (Elliott et al., 1987). This reaction is strongly temperature dependent, leading to e-folding lifetimes between several hours in warm waters and several days in cold, high latitude waters (Elliott et al., 1989).
The temperature dependency of this reaction has been reasonably well described by independent laboratory and field studies 70 (Cutter and Radford-Knoery, 1993;Elliott et al., 1989;Kamyshny et al., 2003).
CS2 is present in seawater in picomolar concentrations, and measurements are generally sparse (Lennartz et al., 2020b). A correlation between temperature and CS2 concentration in surface waters is evident across several datasets (Lennartz et al., 2019;Xie and Moore, 1999). CS2 is produced by photochemical reactions as well, following a similar shape of the AQYwavelength spectrum as OCS (Xie et al., 1998). Precursor molecules such as cysteine, cystine, methionine and DMS have been 75 identified, and photochemical CS2 production itself seems to be temperature dependent (Modiri Gharehveran and Shah, 2018).
Furthermore, there is evidence for a biological production of CS2 by phytoplankton species, with varying yield from different species (Xie et al., 1999), but the exact mechanism is unknown. Outgassing to the atmosphere is considered the most important sink process for CS2 in the mixed layer. The only chemical sink mechanism known so far is hydrolysis with a lifetime of several years (Elliott, 1990). However, a chemical sink process in addition to air-sea gas exchange was needed to explain 80 observations along an Atlantic transect, with an e-folding lifetime of ca. 10 days (Kettle et al., 2001).
Here, we use existing models that include parameterizations of processes known to be relevant for each gas, and apply them on a global scale, accounting for interannual variability in the forcing parameters. We present the first monthly resolved inventory for marine OCS and CS2 emissions for the period 2000-2019. We encourage the community to use these emissions for atmospheric modelling studies in order to elucidate the atmospheric budget of OCS, assess variability in the supply to the 85 sulfate aerosol layer and determine gross primary production on a global scale (available at: https://doi.org/10.5281/zenodo.4297010) (Lennartz et al., 2020a).

Model description
A model version as described in Lennartz et al. (2017) is used to model the interannual variability in oceanic emissions for OCS. A new model is developed to simulate oceanic emissions of CS2. In both models, the surface ocean is divided into grid 90 boxes of 2.8 x 2.8° at the equator (T42 grid, Gaussian grid with ~310 km resolution at equator (NCAR, 2017)) that comprise various depth layers of 1m thickness depending on the depth of the mixed layer in each grid box. Note that the model does not resolve exchange of gases horizontally between the boxes (see Lennartz et al., 2017, for details Photochemical production is calculated as the product of UV radiation UV [W m -2 ], the absorption coefficient of CDOM at 100 350 nm a350 [m -1 ], and the photoproduction rate constant p integrated over the mixed layer depth (MLD) according to equation (2): The photochemical rate constant p is scaled with a350, following a rational suggested by von Hobe et al., (2003), which reflects that a350 can be regarded as a proxy for both photosensitizer and sulfur source across large spatial scales. The linear dependence between a350 and p is calculated based on fits to observational data from three major ocean basins as described in Lennartz et al., (2017). This wavelength-integrated approach has been shown to reproduce both local measurements from several cruises as well as global OCS observations (von Hobe et al., 2003;Lennartz et al., 2017). UV radiation below the sea surface is 110 calculated according to solar radiation, zenith angle and wind speed following von Hobe et al. (2003) as described in Lennartz et al. (2017). The light field in each 1 m depth layer is calculated by reducing the incoming short-wave radiation depending on the local absorption coefficient a350. Photochemical production is then computed for each layer individually, followed by integration over the entire mixed layer. This integration inherently assumes a well-mixed surface layer.
Dark production is calculated according to Lennartz et al. (2019). This reaction rate is an update of the original formulation by 115 von Hobe et al. (2001), resulting in a semi-empirical relationship based on observations from a wider spatial range of observation than the initial study. In this formulation, the dark production rate depends on temperature and a350 (eq. 3): OCS hydrolysis is determined according to Elliott et al. (1989) and depends on temperature (SST), salinity (SSS) and the proton activity a[H + ], equivalent to 10 -pH , according to eq. (4) and eq. (5) Air-sea exchange is calculated as the product of the concentration gradient between water and equilibrium concentration ∆ and the transfer velocity k parametrized according to Nightingale et al. (2000): The equilibrium concentration is calculated according to de Bruyn et al. (De Bruyn et al., 1995). The transfer velocity is corrected for OCS with the Schmidt number, calculated based on the molar volume according to Hayduk and Laudie (1974). The model for CS2 includes the processes of photochemical production and a first order chemical sink, according to eq. (7).
Photochemical production is calculated in the same way as for OCS, with an additional reduction factor r applied (eq. 8). Xie et al. (1998) approximated that CS2 photoproduction rates are about a factor of five smaller than OCS photoproduction rates by comparing an experimentally derived AQY from CS2 and OCS (r=0.2 in eq. 8). The two AQY were not measured at 135 the same location, but in comparable water properties. Another study with simultaneous measurements of both gases reported varying factors between 0.2 and 0.014 (5 to 70 times smaller than OCS photoproduction) (Lennartz et al., 2019). Here, we scaled the reduction factor to obtain the best fit in the average concentration, resulting in a factor r=0.1 in eq. 8. Thus, the model reflects the similar shape of the AQY for both gases by assuming a constant ratio, but the scaling of the overall magnitude of the photoproduction rate constant is chosen to obtain the best fit to observations from the database in Lennartz 140 et al. (2020). A chemical sink according to the model formulation in Kettle (2000), i.e. with an e-folding lifetime of 10 days ( 1 ), was implemented according to eq. (9): Air-sea exchange was calculated as described for OCS, using the CS2 solubility according to De Bruyn et al. (1995).
As CS2 cycling in the water column is not yet well understood, this model should be understood as a base model to be extended 145 as soon as additional process rates and their dependencies become available.

Simulation set-up
Simulations are performed for the period 2000-2019. There are several changes in the forcing data compared to the climatological run in Lennartz et al. (2017). Here we use monthly resolved data for the period 2000-2019 for a350, surface shortwave radiation, surface temperature, wind speed and sea level pressure (Table 1). The meteorological data were obtained 150 from the ERA5 reanalysis (more specifically, its product line 'ERA5 hourly data on single levels from 1979 to present', Hersbach et al., 2018) through the Copernicus Climate Change Service (https://climate.copernicus.eu/). One file per year and parameter, containing hourly data on 0.25° x 0.25° resolution, was downloaded. For wind speed, the zonal and meridional components of wind speed at 10m altitude (u10 and v10, respectively) were downloaded separately and converted into wind The post-processing of the meteorological data was done using CDO tools (climate data operators, version 1.9.8) (Schulzweida, 2019) and comprised the following steps: a) the yearly files for each parameter were split into monthly files using the CDO flag 'splityearmon', resulting in 240 monthly 160 files covering the 20-year period 2000 to 2019 for each parameter; b) for each of the 240 months within the period, monthly-mean diel cycles of each meteorological parameter x were calculated using the CDO flag 'dhouravg', which calculates multi-day averages for every hour of a day as where m is the month (1 to 12), h is the hour of the day (1 to 24), d is the day of the month (1 to 28, 29, 30, or 31), and Nm is 165 the number of days within month m; c) the resulting fields were regridded from the regular 0.25° x 0.25° longitude-latitude grid into the spectral T42 grid (~2.8° x 2.8°) using the cdo flag 'remapcon2', which is a second-order conservative remapping method that takes into account all source grid points, both in longitude and latitude directions. Among the remapping methods available in CDO, 'remapcon2' was considered the most appropriate to interpolate the selected meteorological parameters from a fine grid to a much coarser grid. 170 Monthly forcing fields for CDOM are derived from Aqua MODIS satellite level 3 product 'absoprtion due to gelbstof and detritus at 443 nm' (NASA Goddard Space Flight Center, 2019), and converted to 350 nm with an exponential slope of 0.02 for the wavelength spectrum. Climatological values are used for salinity and mixed layer depth in a monthly resolution, which is the same for each month of the year throughout the simulation period, unchanged to Lennartz et al. (2017). The average diel cycle of meteorological data is set to the 15 th of each month (one value for every 2 hours). In between, data is interpolated 175 separately for each time of the day, resulting in a continuous change of the amplitude of the diel cycles. The initial concentration for both gases is a constant value of 8 pmol L -1 in all grid boxes. The model is spun up for one year, repeating the conditions of year 2000 prior to the simulation period.

Spatial and seasonal variability 180
Both gases show distinct spatial patterns in their annual concentration and emission averages, which reflect their marine cycling. For OCS, highest concentrations are present in cold, high latitude waters and shelf areas, whereas lowest concentrations prevail in warm, subtropical gyres where CDOM abundance in the water is low (Fig 2a). A latitudinal gradient with higher concentrations in high latitudes and low concentrations in tropical and subtropical waters reflects the temperaturedependent degradation by hydrolysis. The degradation is strongest in warm waters, where the lifetime of OCS is on the order 185 of hours, keeping concentrations low. This general pattern is in broad agreement with observations of the largest available database on seaborne OCS measurements (Lennartz et al., 2020). Annual mean emissions largely follow the spatial pattern of  (Fig 3b). In July and August, the globally integrated net emissions are close to zero, similar to a previous budget using a similar model (Kettle et al., 2002).

200
CS2 concentration show a different global pattern than OCS concentrations. CS2 concentrations and emissions have hot spots in coastal and shelf regions, as well as in tropical and subtropical oceans, reflecting photoproduction as the main production process in the model. The tropical and subtropical areas show comparably low CS2 concentrations (Fig. 4c), and their importance for globally averaged emissions mainly comes from the large oceanic surface area (Fig. 4d). Notably, CS2 emissions in the western Pacific, where inverse modelling studies have located the missing source, are relatively low (Glatthor 205 et al., 2015;Kuai et al., 2015). The hot spots being located in the tropical and subtropical regions with similar intensities of incoming radiation all year, leads to less seasonal variation in globally integrated emissions, i.e. an amplitude of 3.2 Gg S yr -1 . The ocean is a source of CS2 to the atmosphere over the entire year, since emissions are calculated with an atmospheric mixing ratio of 0 ppt. Highest emissions occur in boreal winter, and the lowest in boreal summer. (range of mean annual cycle of 21 Gg S yr -1 ) than the interannual variability (mean monthly variability of 8.4 Gg S month -1 ) (Fig. 3). Interannual variability of the emissions in each month is largest during boreal spring (April, May, June) and fall (October) (Fig. 3a). These months show the largest difference between minima and maxima during the whole period (grey 220 https://doi.org/10.5194/essd-2020-389  Fig 3a). The spatial pattern of interannual variability of OCS emissions shows highest variability, i.e. highest standard deviation among annual averages in each gridbox, at locations with high OCS concentrations and emissions (Fig. 2). These regions comprise the northern temperate and polar regions, the Southern Ocean, and shelf areas, especially those close to coastal upwelling regions and river plumes (Fig. 2). The standard deviation for OCS concentrations between annual averages ranges from 0.22 at the oligotrophic gyres to 143.8 pmol L -1 at the highly dynamic coast off Alaska, USA (average standard 225 deviation 3.4 pmol L -1 ). The interannual variability also shows latitudinal differences. Polar regions in both Arctic and Antarctic waters display the largest seasonal cycles in OCS concentration, i.e. the highest annual variability (Fig 4), and at the same time also display highest interannual variability. Differences in mean concentrations (area weighted) in summer range between 72.8 pmol L -1 in June 2011 to 91.6 pmol L -1 in July 2017, i.e. ca. 20 pmol L -1 in the Arctic ocean (Fig. 4) Carbon disulfide concentrations are highest in shelf areas in the tropics and subtropics, and generally decrease towards high 240 latitudes (Fig. 2c). The spatial pattern of the annually integrated emissions mirrors this picture (Fig. 2d). While the spatial pattern of concentrations and emissions is similar in each year, the absolute concentration and magnitude of emissions does show interannual variability (Fig. 3b). Emissions are calculated here with a boundary layer mixing ratio of zero (maximum possible emission) as is commonly done for other short-lived gases such as DMS (Lana et al., 2011), so the ocean is a CS2 source at every location throughout the year. A little less than half of the sulfur in CS2 is converted to OCS. to the interannual variability (3.2 Gg S yr -1 ). This is different to OCS, where annual variability was higher than interannual variability for globally integrated emissions. This difference is caused by the location of the respective hotspots of the produced 250 gases: As OCS has its concentration and emission hot spots mainly in high latitudes, which experience a very seasonal light regime, its annual variability is high. The low concentrations of OCS (and corresponding low emissions) in the tropics result from the fast degradation by hydrolysis. In contrast, CS2 has its concentration and emission hotspots mainly in low latitudes with more constant forcing, and hence displays smaller annual variability. The interannual variability of CS2 emissions among https://doi.org/10.5194/essd-2020-389 single months has a similar magnitude throughout the year (grey shaded area in Fig. 3d). Maximum monthly mean 255 concentrations of CS2 vary the most in the summer months of the northern temperate regions (23°-66°N) from 4.3 Gg S month -1 in June 2011 and 6.0 Gg S month -1 in June 2018, but show less variability in the winter months, i.e. between 0.8 and 1.2 Gg S month -1 in December. Due to their comparably low surface area and the relatively low concentrations, high latitude regions do not play a significant role in globally integrated CS2 emissions (Fig. 4). The dominance of southern temperate emissions of CS2, despite higher absolute mean concentrations in northern temperate regions is explained by the larger surface ocean area 260 in the southern temperate regions (Fig. 4c and d).

Main drivers of interannual variability
The interannual variability in OCS and CS2 concentrations and emissions is a result of the interannual variability in their production and consumption processes, which in turn depends on environmental conditions. Globally integrated annual 265 emissions of OCS correlate significantly with global annual averages of CDOM a350, sea surface temperature and wind speed (Tab. 3). CDOM a350 explains the largest variance, and sea surface temperature and wind speed explain less of the observed variance. Thus, CDOM a350 has the strongest influence on the variability of global scale OCS concentrations. The influence is not surprising, as CDOM a350 impacts both photochemical and dark production of OCS and modulates the light field in the water (at higher a350, photoproduction is higher but also more limited to the surface). The photochemical production rate is 270 second order dependent on CDOM a350, reflecting its double role as photosensitizer, i.e. those molecules absorbing light energy for photochemical reactions, and as a proxy for the amount of sulfur molecules able to form radicals in photochemical reactions.
As such, CDOM a350 exerts a strong, non-linear and positive influence on OCS concentration, and seems to be the main driver of its interannual variability. The overall strong influence of CDOM a350 on OCS interannual variability is also underlined by similarity in the spatial pattern of the standard deviation in annual average concentrations and emissions between OCS and 275 CDOM a350 (Fig. 5). Sea surface temperature strongly influences OCS hydrolysis, which leads to low concentrations in warm tropical and subtropical waters. Temperature also controls the solubility of the gas in water, i.e. the equilibrium water concentration is higher in colder waters. Variations in temperature explain a small part of interannual variations in OCS emissions. However, rising temperature towards the end of the period (Fig. 5) did not outweigh the increase in CDOM a350, which supports the above mentioned result that the observed changes in CDOM a350 had a stronger influence on overall OCS 280 production than observed temperature changes had on hydrolysis. Finally, wind speed imposes a nonlinear control on OCS emissions, but the impact is smaller than that of CDOM a350.
Globally integrated CS2 emissions correlate significantly with CDOM a350, with a substantial part of the variance in interannual variability (67%) explained by this single factor, although this is less than for OCS. Photochemical production of CS2 is 285 similarly calculated as that for OCS, and hence depends nonlinearly and positively on CDOM a350. The lesser amount of explained variance compared to OCS may result from the lack of a CDOM a350 dependent dark production process. Interestingly, CS2 emissions correlate with temperature, although temperature is not part of any production or consumption process in the model, and solely modulates the solubility of CS2. Increasing temperature decreases the solubility and would lead to a lower surface water concentration, hence, this effect cannot explain the correlation between temperature and CS2 290 surface concentrations in observations (Lennartz et al., 2020). Potentially, the co-variation of temperature with radiation dose might be responsible for the correlation of CS2 concentration and temperature that is evident across observational datasets (see Introduction). The spatial variation of the standard deviation of annual averages of CS2 concentration and emissions resembles that of CDOM a350, again underlining that this is a major factor for interannual variability of CS2 (Fig. 5).

Comparison to observations
The model output of the monthly resolved simulation for 2000-2019 is compared to the database compiled by Lennartz et al., (2020), which contains 2970 fully georeferenced OCS measurements and 501 fully georeferenced CS2 measurements in the period considered here. The model output is subsampled at time (including time of day) and location closest to the measurements in the respective period for a 1:1 comparison. 300 For OCS, the range of the subsampled model output agrees well with data from the database (7 cruises, n=2971), with a slight underestimation of measured concentrations by the model (average 40.1 pmol L -1 in the database, 38.4 pmol L -1 in the model, Fig. 6a). The direct comparison reveals remaining scatter around the 1:1 line, and a high bias in the model which grows with increasing OCS concentrations (Fig. 5b). The scatter and high bias in the data likely results from simplifications in the model.
The main simplifications, probably causing these discrepancies between observation and models, are the missing horizontal 305 transport, the use of averaged wind speed as forcing, the use of CDOM a350 as a proxy for photochemical production and the application of a climatological mean for the depth of the mixed layer.
Using CDOM a350 as a proxy for OCS photochemical production may introduce some scatter, but likely not a systematic bias.
The very complex nature of the dissolved organic matter pool in the ocean, which comprises CDOM as the optically active fraction, makes it difficult to assign one photoproduction rate constant or apparent quantum yield to all the reactions taking 310 place with different precursors. CDOM a350 has been shown to be a suitable proxy across three major ocean basins (Atlantic, Pacific, Indian Ocean), but the rate constant -CDOM a350 relationship showed some variability that might be improved when more data becomes available.
The missing horizontal transport can lead to a systematic model bias especially in cold waters where OCS lifetime increases to time scales (days) relevant for physical transport, but still this process is unlikely to decouple OCS concentrations from its 315 drivers like CDOM and temperature, that would be transported accordingly. The effect of horizontal transport is negligible in warm waters of the tropics, subtropics and most of the temperate regions. In regions with deep mixed layers such as the Southern Ocean, the assumption of a completely well mixed surface layer may be violated and cause discrepancies between the modelled value (average of mixed layer) and the measured value (close to surface, i.e. higher concentration that at bottom of the mixed layer). Since the modelled concentration depends on depth of the mixed layer and its relation to the photic zone, 320 https://doi.org/10.5194/essd-2020-389 a climatological average as used here will introduce biases, however, detailed information on mixed layer depth in monthly resolution is not available. This simplification mainly affects OCS concentrations in high latitudes, where concentrations are relatively high, and thus might be partly responsible for the systematic bias revealed by the scatter plot in Fig. 5b. Furthermore, averaging wind speed to a mean monthly cycle will most likely lead to an underestimation of emissions and, hence, an overestimation of concentrations. Due to the nonlinear relationship of the transfer velocity of the gas exchange with wind 325 speed, averaging disproportionally reduces the effect of increased emissions during high wind speeds. Still, given these simplifications and assumptions, the overall good agreement with the measurements underlines the applicability of the model for assessing the marine cycling of OCS and its emissions to the atmosphere.
The marine cycling of CS2 is less well understood than that of OCS. This relatively poorer process understanding is reflected 330 by the comparison of the modelled CS2 concentrations with those of the database (3 cruises, 501 measurements) (R²=0.04).
Modelled concentrations agree with observations on average (average database: 18.0 pmol L -1 , average subsampled model output: 18.2 pmol L -1 ). While the cruises Poseidon 269 and ASTRA-OMZ are relatively well represented by the model (colour code in Fig. 6d), the variability of the measurements during Transpegaso is not well captured. The model used here has some underlying assumptions and simplifications that call for refinement in the future when detailed process understanding is 335 available. For example, the model is based on the assumption of a constant ratio between the apparent quantum yields of OCS and CS2. It has been shown that this ratio is not always constant ( (Kettle, 2000;Lennartz et al., 2019), but as the production pathways of both gases show some similarities (Modiri Gharehveran and Shah, 2018), the model formulation with a constant ratio is a first approximation. Second, the presence of a chemical sink is rationalised by its necessity to explain observed concentrations along an Atlantic transect (Kettle, 2000;Lennartz et al., 2019), but has no mechanistic foundation so far. 340 Dedicated laboratory experiments disentangling the source and sink processes in the water column are needed to further resolve this issue and to improve modelling efforts. Finally, this model does not consider any biological production of CS2. This assumption is justified for a first approximation, as CDOM and primary production (photosynthesis) show similar global scale patterns. High CDOM will thus lead to high production of CS2 in the water, even though the scaling of the photoproduction rate constant (AQY) might inherently include biological production due to the covariation of photosynthesis patterns with 345 CDOM and radiation. Overall, the presented CS2 concentration and emissions are a first approximation, and more detailed process understanding is important to improve emission estimates.
The emission estimate of both gases includes further uncertainties introduced by the parameterizations of the transfer velocity used for calculation of air-sea exchange, which carry large uncertainties especially at high wind speeds (Wanninkhof, 2014).

Data and code availability
The code is available on github under https://github.com/Sinikka-L/OCS_CS2_boxmodel. The simulation output is available 355 at zenodo 10.5281/zenodo.4297010 (Link: https://doi.org/10.5281/zenodo.4297010) (Lennartz et al., 2020a). The output consists of one netCDF files for each gas, each of a size of ca. 444 MB with monthly averages of sea surface concentrations and emissions to the atmosphere, as well as a mean diel cycle for each month.

Summary and Conclusions
OCS and CS2 are climate relevant trace gases and OCS can also be used as a proxy to infer terrestrial gross primary production. 360 A missing source in the atmospheric OCS budget currently makes conclusions on the future impact on both gases and the application of this proxy on a global scale difficult. Since both gases contribute to the atmospheric OCS budget, their oceanic emissions have been suggested previously to account for that missing source. We provide monthly resolved OCS and CS2 concentration and marine emission data for the period 2000-2019 based on a mechanistic ocean box model. We show that interannual variability of OCS is smaller than its seasonal variability in globally integrated emissions, but that a significant 365 positive trend is evident across the period 2000-2019. The main driver for interannual variabilities is variation in CDOM a350.
The comparison of our data to a database with more than 2500 measurements reveals an overall good agreement. The CS2 model presented here for the first time is a first approximation and reveals stronger interannual variability than seasonal variability of emissions. Again, CDOM (or indirectly, biological production) seems to be strongly influencing concentration and emission patterns of CS2. Similarly, an increasing trend in CS2 emissions is significant for the period 2000-2019. We 370 encourage the use of the data provided here as input for atmospheric modelling studies to further assess the atmospheric OCS budget and the role of OCS in climate.