The contribution of coral reef-derived dimethyl sulfide to aerosol burden over the Great Barrier Reef: a modelling study

Coral reefs have been found to produce the sulfur compound dimethyl sulfide (DMS), a climatically relevant aerosol precursor predominantly associated with phytoplankton. Until recently, the role of coral reef-derived DMS within the climate system had not been quantified. A study preceding the present work found that DMS produced by corals had negligible longterm climatic forcing at the global-regional scale. However, at sub-daily time scales more typically associated with aerosol and cloud formation, the influence of coral reef-derived DMS on local aerosol radiative effects remains unquantified. The Weather 5 Research and Forecasting chemistry model (WRF-Chem) has been used in this work to study the role of coral reef-derived DMS at sub-daily time scales for the first time. WRF-Chem was run to coincide with an October 2016 field campaign over the Great Barrier Reef, Queensland, Australia, against which the model was evaluated. After updating the DMS surface water climatology, the model reproduced DMS and sulfur concentrations well. The inclusion of coral reef-derived DMS resulted in no significant change in sulfate aerosol mass or total aerosol number. Subsequently, no direct or indirect aerosol effects were 10 detected. The results suggest that the co-location of the Great Barrier Reef with significant anthropogenic aerosol sources along the Queensland coast prevents coral reef derived-aerosol from having a modulating influence on local aerosol burdens in the current climate. 1 https://doi.org/10.5194/acp-2021-507 Preprint. Discussion started: 29 July 2021 c © Author(s) 2021. CC BY 4.0 License.

more polluted urban environments. However, Muñiz-Unamunzaga et al. (2018) provide no evaluation of the impact of DMS itself on the local climate.
In this study, we aim to explore the extent to which coral reef-derived DMS can influence local aerosol burdens over the Great 85 Barrier Reef. We do this by evaluating the ability of WRF-Chem to simulate DMS processes and analysing what the impact of including an additional coral reef source of DMS is on aerosol processes. We evaluate WRF-Chem against observations from a major field campaign undertaken in the austral spring of 2016: 'GBR as a significant source of climatically relevant aerosol particles', nicknamed 'Reef to Rainforest' (R2R). The model setup, experiment design and field campaign details are provided in Section 2, while the results of this work are provided in Section 3 and summarised by Section 4. All WRF-Chem simulations have been meteorologically nudged to provide the best comparison to the R2R field campaign and to ensure that the responses found in the model are attributable to the DMS surface water concentration (DMS w ) perturba-100 tions and not internal model variability. The Australian Bureau of Meteorology (BoM) Atmospheric high-resolution Regional Reanalysis for Australia -Regional domain (BARRA-R) has been used to provide initial conditions and to perform nudging at six hourly intervals (Su et al., 2018). Nudging has been applied to temperature and water vapour above the planetary boundary layer and to horizontal wind above vertical level 19 (approximately 3km).
The model was restarted every 4 days (ingesting the previous day's chemical conditions), with a 12 hour spin up thrown out. 105 The chemical boundary and initial conditions are provided by the Model for Ozone and Related Chemical Tracers (MOZART-4, Emmons et al., 2010). WRF-Chem maps aerosol mass and number to the eight simulated bin sizes (Fast et al., 2006) from the bulk aerosol mass provided by MOZART-4, representing the Aitken mode through to the accumulation mode. A full description of the chemistry, aerosol and physics setup for these simulations can be found in Appendix A.

DMS climatologies 110
The default DMS w climatology provided by WRF-Chem is the outdated Kettle and Andreae (2000) climatology on a 1x1.25 • grid. The climatology used here is the updated Lana et al. (2011) DMS w (referred to henceforth as the L11 climatology). The interpolation performed by WRF-Chem (via Prep-Chem) was deemed unsatisfactory (creating unphysical patterns around the coastlines and generally a very coarse interpolation). For this reason, the L11 climatology for October was interpolated to each WRF-Chem domain using the python (v3.5) basemap bilinear interpolation, overriding the default WRF-Chem DMS w 115 climatology. Further smoothing around the coastlines was performed. All fields that pass through Prep-Chem underwent the same interpolation to a higher resolution for consistency.
After initial testing, it was found that simulations using L11 overestimated DMS w in comparison to observations taken during the R2R campaign (this finding will be described in Section 3). For this reason, a scaled DMS w climatology was created, where L11 was divided by 2.8 to match the average DMS w observations taken during the R2R campaign. The scaled 120 climatology is referred to as L11S henceforth.
To examine the impact of coral reef-derived DMS, a new source of DMS w was added to the L11S climatology. The coral reef source was determined by using the areal fraction of coral reefs per WRF-Chem grid box to add a weighted 50 nM of DMS w , matching that of the global analysis performed in Fiddes et al. (2021). The 50 nM value was chosen as a high range estimate in order to maximise any potential signal and response. In reality, this source varies considerably with time (Jones et al., 2007, eg. 125 up to as much as 54 nM), but is likely much smaller. The coral reef DMS w source added to the L11S climatology is referred to as L11SCR and is shown by the coloured contours in Figure 1 along side the R2R DMS w observations.

Experiment setup
Three simulations are analysed in this study. The first simulation uses the L11 climatology (and hence will be referred to as the L11 simulation), with no biomass burning or dust and with the Gong et al. (1997) Gong et al. (1997) parameterisation. Biomass and dust emissions are also included. The third simulation uses the L11SCR DMS w climatology and otherwise the same setup as L11S, to examine if coral reef-derived DMS w plays a 135 role in aerosol characteristics over the region.

Observations and evaluation methods
The R2R field campaign took place on two platforms, the first on board the CSIRO Marine National Facility RV Investigator (RVI) which navigated a path around the GBR from the 28th September -22nd October 2016. The ship track is given in  Experiments (AIRBOX) mobile atmospheric chemistry lab, stationed at Mission Beach, QLD from the 20th September to the 16th October 2016. The position of AIRBOX is given in Figure 1. Note that in both domains, WRF-Chem considers this grid point as ocean (the container was <50 m from the shoreline). Observations collected both on the RVI and at AIRBOX comprised measurements of meteorology and atmospheric chemical and aerosol composition. These observations aimed to capture each step of the DMS cycle over the GBR for the first time. A list of observations used in this study from the R2R 145 campaign can be found in the Appendix Tables B1 and B2 as well as a brief description of how the data was collected and processed. All observations are available from the relevant institutions upon request.
To compare the estimated flux DM S from the model and observations we used the Liss and Merlivat (1986) parameterisation given modelled and observed DMS w , SSTs and wind fields. To evaluate sulfate aerosol, particle size bins were linearly interpolated to observed particle sizes, assuming aerosol bins are internally mixed. In order to determine periods of time in 150 which the RVI observations were contaminated by ship exhaust, the hourly black carbon concentrations needed to be above 50 ng m −3 and the wind direction relative to the ship was > 120 and < 240 degrees. Any time stamp within ±five minutes of meeting these two criteria were also flagged. Air-masses were considered to have marine origins if radon concentrations were below 300 mBq cm −3 .
To aid the airmass characterisation, the Hybrid Single Particle Lagrangian Integrated Trajectory Model (HYSPLIT) was 155 used to perform back trajectories (Draxler andHess, 1997, 1998). The National Centers for Environmental Prediction (NCEP) Global Data Assimilation System (GDAS) 0.5 degree product was used to produce the back trajectories, where vertical motion was calculated using the model vertical velocity. Initial height was set at 100 m and the trajectories were run for 72 hours, every two hours.
For the WRF-Chem evaluation, time series comparisons, correlations and a bias factor metric have been used, evaluating 160 equivalent model fields to the observations taken during R2R. Evaluation focused on domain two. The Normalised Mean Bias Factor (NMBF, Yu et al., 2006) is used. The NMBF is a symmetric metric, i.e. negative biases are not bound by zero, and remains viable when measured values are much smaller than model values. This metric is an improvement on other model performance metrics, as described by Yu et al. (2006). To ensure clarity of this metric across both positive and negative biases, the NMBF has been converted into a bias factor (BF) by adding 1 if NMBF < 0, or subtracting 1 if NMBF > 1. Correlation 165 analysis has been performed using Spearman's rank correlation methods (Wilks, 2011). This method is a non-parametric test that quantifies the monotonicity of the relationship between two variables. Figure 2 shows the timeseries of DMS w , wind speed and the resultant flux DM S for the three WRF-Chem simulations and the 170 RVI observations. In Figure 2a, the effect of scaling the L11 DMS w climatology can clearly be seen, where L11S represents a more realistic value for the GBR region with a BF of 1.02 compared to L11 of 2.74. While the L11 climatology is the most up to date gridded data set of DMS w available (although is not the default climatology for WRF-Chem), we can see that for this region it significantly overestimates DMS. This result highlights that much greater sampling of DMS w variability over space and time is required, especially in regional studies.

Surface water DMS and the resulting sea-air flux
175 Figure 2b shows that overall the model predicts wind speed along the RVI path well, with L11S having a NMBF of 1.06, and a correlation coefficient of R=0.87 to the observations. The diurnal cycle is also well captured (not shown) although the model tends to underestimate and delay the peak wind speed. This is attributed to the poor simulation of the sea breeze structure 6 https://doi.org/10.5194/acp-2021-507 Preprint. Discussion started: 29 July 2021 c Author(s) 2021. CC BY 4.0 License.
(results not shown). Some model skill may be attributed to the fact that RVI observations are assimilated into the BARRA reanalysis product used to nudge the model, although we also note that model winds are free running below 3 km. Comparison Wind speed is a key factor in the Liss and Merlivat (1986) flux parameterisation. An estimate of the flux DM S along the RVI track is shown in Figure 2c. Here we can see that despite the model's constant DMS w , it is able to do a comparatively good job representing the average flux DM S , with a BF of 1.21 for L11S, compared to 2.32 for L11. This skill is predominantly due 185 to the well captured marine wind speeds discussed above. Visually, the L11S flux DM S timeseries appears to follow that of the observations relatively closely, although with a weak R value of 0.24 and underestimated variance (where σ L11S = 2.77, σ Obs = 3.11), likely due to the constant DMS w .
From Figure 2a, we can see where the ship is within a grid box that has an additional coral reef DMS w source included in L11SCR. The corresponding flux DM S is in general much larger than what was observed although we recognise that the ship 190 did not measure directly over coral reefs. Nevertheless, this is a clear demonstration of how additional coral reef DMS w is influencing the flux DM S and should subsequently influence DMS a . Figure 3a and b shows DMS a from the RVI and at AIRBOX. The average at both sites is moderately well captured. For the RVI, the BF for L11S compared to the observations is 1.29, while at AIRBOX this is 1.42. Weak (negative) and insignificant 195 correlations are found for both locations. The poorer performance of the model at AIRBOX may be due to the complexity of the location on the coast. However, the weak and negative correlations suggest that the model is missing an important aspect of DMS a variability, likely caused by the constant DMS w field or perhaps missing chemical sinks. While WRF-Chem has more complex DMS chemistry compared to other chemical models, comprising 30 DMS oxidation pathway reactions, it is possible that it is still missing important reactions. For example, the importance of DMS removal by BrO or Cl 2 has been highlighted 200 in the literature (Breider et al., 2010;Khan et al., 2016;Muñiz-Unamunzaga et al., 2018). Specifically, Khan et al. (2016) note that without these inclusions, the variability of DMS a is not well modelled.

Atmospheric DMS and sulfate aerosol mass
Nevertheless, significant improvement can be seen in DMS a in the L11S simulations compared to the standard L11 simulation. The addition of coral reef DMS w has made a small difference in DMS a when near reef regions. However, the significant changes to DMS w between L11, L11S and L11SCR have not had a significant impact on the sulfate aerosol mass, shown in For L11S, over the entire timeseries, the model underestimates the observations of sulfate aerosol mass by approximately 0.11 µg m −3 , with a BF of 1.3. A moderate correlation of R = 0.46 suggests that the model is capturing some of the observed trends and variability. For AIRBOX, the underestimation of the observations by L11 increases to 0.16 µg m −3 and the BF is 1.43. In addition, a statistically significant negative relationship is found between the two timeseries (R=-0.16). This is likely 210 due to a number of local sulfate sources observed by AIRBOX that were not included in the model, for example, vehicle emissions, that impact that variability of the timeseries. Importantly, we note that reducing DMS w by approximately 65% between L11 and L11S results in a decrease of only 10% in total surface sulfate aerosol mass along the RVI track. It is possible that this is because L11S also included biomass burning, while L11 did not. The mean difference of SO 2 -4 surface concentration for L11 and L11S over the entire RVI timeseries is 215 -0.039 µg m −3 . When periods of contaminated air (airmasses that contained high levels of black carbon, terrestrial influence or were flagged for other reasons) are filtered out of the calculations, the difference between L11 and L11S was -0.049 µg m −3 , or a -15% change. The difference between the filtered and unfiltered means suggests that the inclusion of biomass burning has offset the sulfate reduction caused by DMS w only marginally (by about 5%). Nevertheless, these results imply that DMS only plays a small role in the sulfate aerosol burden over the GBR.  Figure 4a shows the time series of radon and black carbon (for both the RVI and L11S), the mean wind direction in the model and three sets of HYSPLIT trajectories over Stations 3.1, 3.2 and 4. It is clear from the radon timeseries that over the campaign, the ship did not encounter what could be considered clean marine air often (defined as periods below 300 mBq cm −3 ), although we note that exposed coral reef atolls are also a source of radon. The radon time series is coloured by the exhaust contamination 225 flag and indicates that there were even fewer occasions in which conditions uninfluenced by ship exhaust (shown by the green colours) or terrestrial air mass were measured.

Terrestrial airmass influence
Comparing the RVI black carbon to the L11S concentrations, where ship exhaust from the RVI is not included, we can see some agreement in periods of terrestrial airmass (eg. between stations 3.1 and 3.2). While the L11S black carbon levels are lower than what was measured, the mean of of 0.07 µg m −3 over the time period shown is above that of which was considered 230 clean in this study (0.05 µg m −3 ) also implying a predominant terrestrial influence.
In Figure 4b, south easterly surface winds are shown to prevail, although some bias in the day to day wind direction was found compared to the RVI (not shown). Looking at this map, one may expect that this period did comprise clean marine airmasses, and, as an example, this time of year was in part chosen due to this prevailing wind direction. However, as the 72 hour HYSPLIT back trajectories in Figure  In addition to unfavourable synoptic conditions, one such cause of a lack of clean marine periods is the influence of sealand breeze coupling. Over coastal-marine regions recycling of airmasses over the land and ocean can occur and have far 240 reaching impacts. For example, a sea-land breeze circulation up to 150 km offshore under favourable conditions during the R2R campaign was detected in previous (unpublished) WRF simulations (personal communication C. Vincnet). Similarly, sea-land coupling was also observed at AIRBOX (not shown). This circulation between land and ocean can lead to terrestrial airmasses extending far offshore.

Dominant anthropogenic sources of sulfur
Now we consider not just how DMS has changed over the RVI track or at AIRBOX, but how it has changed over the entire WRF domain, and in particular over coral reef regions. A vertical transect of DMS a and the surface mean for the entire domain is shown in Figure 5a and b. In this figure, the source of coral reef DMS a is clearly evident in the boundary layer (c) and being blown with the prevailing winds at the surface in (d). Over the entire domain, at the surface, a significant increase in DMS a of 0.003 ppb is found, approximately 12%, with a mean increase of 40% found over coral reef grid points. These significant 250 increases in DMS a however, do not result in significant change in sulfate aerosol mass, as found along the RVI track and at AIRBOX. These two plots clearly demonstrate that anthropogenic emissions represent a significant source of sulfate aerosol (among other species) for the GBR region. In Figure 6a, over the Gladstone coal fire power station (brown triangle), a dominant plume in 255 sulfate aerosol can be seen, while in Figure 6b, numerous plumes, that align with known power generators, can be seen. Figure, 6c and d indicate no coherent change in total sulfate aerosol mass that could be robustly attributed to the inclusion of coral reef-derived DMS. The surface mean change between L11S and L11SCR over the entire domain is 0.0018 µg m −3 , or a change of 0.38%. Directly over coral reef grid points, an increase of 0.47% was found.

260
The prevalence of anthropogenic sulfur and the abundance of pre-existing particles suggests that the small addition of sulfate from DMS is unlikely to participate in new particle formation in the boundary layer. Rather, it is more likely that coral reefderived H 2 SO 4 would condense onto pre-existing particles, growing their mass. Below, we analyse column integrals of aerosol number and mass in the free troposphere and boundary layer to test this hypothesis. Figure 7 shows the transect of the boundary layer (left) and free troposphere (right) total column sulfate mass and aerosol 265 number for the bins representing Aitken and accumulation mode aerosol. We suggest that regions where the changes along the transects in the number and mass co-vary are likely due to internal model variability, rather than changes in the DMS a field.
In the larger bins (Figure 7e-j), this appears to be the case in most locations. However, in the smaller bins (Figure 7a-d), some instances where the mass has changed independently of the number are found.
In Figure 7a and c, the total boundary layer total column sulfate mass has increased in some areas, while the number has 270 not. These increases may be evidence of condensational growth. Importantly we note that the regions over which the increase in mass occur are not co-located with coral reefs, but may be due to advected DMS a and its oxidation products.
In the free troposphere, new particle formation is far more likely to occur. Examining the smallest bin size (Figure 7b) however, no clear evidence of the coral reef-derived sulfuric acid participating in new particle formation is found. Rather, directly over coral reefs, a relatively large decrease in sulfate aerosol mass is found, with a lesser reduction in particle number concen-275 tration. These decreases may suggest that the coral reef-derived precursors have grown existing small sized particles, causing them to shift into larger bin sizes. On average over the transects for the remaining bin sizes, increases in free tropospheric sul-fate mass was found, accompanied by decreases in number (keeping in mind that these changes are insignificant and less than 1%). This may further suggest greater coagulation rates, reducing the number, while increasing the mass. However, due to the very small and insignificant changes found, we have low confidence that these results are caused directly by coral reef-derived 280 DMS as opposed to model noise.
While locally, changes in sulfate mass in some cases appear to be up to 5%, on average the changes over the transects in either the boundary layer or free troposphere are less than 1%. Furthermore, the largest signals from coral reef-derived DMS appear to occur in the smallest size aerosol, with little discernible change in the larger aerosol sizes that are of greater climatic relevance. Examination of changes in cloud condensation nuclei (not shown) confirm this and indicate that the very small 285 addition of sulfate by corals is unlikely to have any direct or indirect aerosol effects over the GBR region. Further investigation of these aerosol effects has been carried out, and no significant changes in clear sky radiation, total radiation, cloud droplet number, liquid water path, cloud fraction or precipitation were found.

Conclusions
Coral reefs as an unaccounted-for source of DMS have gained attention over recent years, with numerous observational studies 290 suggesting they play an important and even regulatory role in local climate (Jones, 2013;Hopkins et al., 2016;Jones et al., 2017;Cropp et al., 2018;Jackson et al., 2018Jackson et al., , 2020b. While Fiddes et al. (2021) in a global modelling study found that coral reef derived DMS over the Maritime Continent and Australian region had little impact on long-term climate, no regional-scale modelling has been performed prior to this present study. This is particularly important if we are considering temporal and spatial resolutions relevant to bioregulatory feedbacks. In this work, we have evaluated the ability of WRF-Chem to simulate 295 DMS and sulfur processes and tested the sensitivity of these processes to perturbations in DMS w , with a particular focus on coral reef-derived DMS.
We find that, compared to observations taken during the R2R campaign, the Lana et al. (2011) climatology significantly overestimates DMS w and required reduction by 65% to be of a similar magnitude. This finding adds to a growing argument of a need for an updated and if possible, time-varying (beyond the fixed monthly climatology), DMS w climatology (e.g. as 300 suggested in Green and Hatton, 2014). Furthermore, this finding demonstrates that greater attention needs to be paid to the DMS w climatology within modelling systems, highlighted by the fact that the default WRF-chem climatology is the out of date Kettle and Andreae (2000) climatology.
With a DMS w field that aligns with observations, the Liss and Merlivat (1986) flux DM S calculated from both observations and the model agree reasonably well in magnitude. This result is in part due to the well-captured marine wind speeds along 305 the RVI track. Subsequently, DMS a is also reasonably well captured, although overestimated, as is sulfate aerosol mass over the ship track. The Liss and Merlivat (1986) flux parameterisation is considered a conservative parameterisation compared to other methods which provide much larger fluxes (see Appendix for details). Hence the overestimation of DMS a found here (despite matched DMS w and well captured wind speeds) further suggests that Liss and Merlivat (1986) is the most realistic parameterisation for calculating the flux DM S . Nevertheless, this evaluation gives us confidence that the model is able to capture the key processes in the DMS-aerosol system.
By comparing simulations with the original Lana et al. (2011) DMS w climatology to the scaled climatology, we find that DMS plays only a small role in sulfate aerosol burdens over the GBR. For a 65% reduction in DMS w , a subsequent 67% reduction in DMS a and a 10-15% reduction in sulfate aerosol mass was found at the surface. Examination of the background meteorological conditions indicate that influence from terrestrial airmasses occurred for the majority of the R2R campaign, 315 which was broadly reflected in the WRF-Chem model. Furthermore, we suggest that local anthropogenic sources of sulfur from fossil fuel power generation is likely to have a strong influence over the GBR airmass due to proximity, interaction of the sea breeze and synoptic conditions. We recommend further observational studies are carried out to confirm if this is the case for different times of year. Additionally, we note that major coral bleaching events occurred in the summer prior to this field campaign. While the coral reefs south of Cairns (the region of this campaign) were not as severely bleached, we cannot rule 320 out an impact on the production of DMS by coral reefs due to this event.
These results suggest that much smaller changes in DMS from coral reefs are unlikely to have a large impact on the aerosol burden. We find that by adding in a source of coral reef DMS, the total sulfate aerosol mass increases by less than 1%, while insignificant changes of a similar magnitude were found for the total aerosol number. Over the time period studied, no evidence of new particle formation was found, although condensational growth in boundary layer and free troposphere in the smallest 325 aerosol bins may have occurred. No evidence of direct or indirect aerosol effects were found in response to these very small changes in aerosol mass and number.
Whether the lack of influence from coral reef-derived DMS on the local aerosol burden was a result of unfavourable synoptic conditions is difficult to assess. However, the close proximity of anthropogenic aerosol emissions to the inner GRB suggests that this region should not be considered a 'clean marine' environment unless under very specific conditions, therefore limiting 330 the role that coral reef-derived DMS can play in aerosol formation and growth while these emissions continue. We suggest that the influence of anthropogenic aerosol is analysed in future work by exploring if an increase in ammonium was also found, associated with sulfate produced by a power station. While Modini et al. (2009) suggested that they had observed new particle formation in such 'clean marine' conditions, further work needs to be done to understand how often such conditions occur over the GBR before we can consider if these new particle formation events could have an influence on aerosol and weather.

335
This study indicates that it is more likely the small contribution of volatile sulfur compounds from the GBR contribute to aerosol growth via condensational pathways. While this is in agreement with the hypothesis presented in Jackson et al. (2020b), our results suggest that the growth in the smaller sized aerosol (Aitken mode) due to coral reef-derived DMS is still too small to have an impact on radiative or cloud processes. This finding is in agreement with the global simulation study described in  This suggestion supports that made by Zaveri et al. (2008) who note that the Wexler et al. (1994) homogeneous nucleation parameterisation used here underpredicts binary nucleation, while over predicting ternary nucleation in the boundary layer. An aerosol scheme that explicitly simulates smaller sized aerosol would be desirable for future studies.
A limitation of this study the nudged meteorology (every six hours), potentially limiting the ability of indirect aerosol 350 effects to occur and any possible feedbacks. As shown in Fiddes et al. (2021), large differences can be found in aerosol-climate processes between nudged and free running simulations, although differentiating between model noise and a real signal is difficult. Furthermore, the four day restarts of WRF-Chem, despite ingesting the previous day's chemistry, appeared to impact total aerosol numbers, including CCN, which were found to be strongly constrained by this set-up choice. This impact has limited the analysis of these fields in this study. The restarts were not thought to impact the DMS processes, in part due to the 355 lifetime of DMS in the atmosphere.
We suggest future studies should consider running both nudged and free simulations to ensure a full understanding of the aerosol-climate system. However, we do not expect that in a similar study to this, the results would be significantly changed in a free running simulation due to the very small changes in aerosol found in this work, the abundance of anthropogenic aerosol, and the fact that the meteorological nudging was not applied in the boundary layer.

360
While in the current climate, this study, in agreement with Fiddes et al. (2021), suggest that coral-reef derived DMS has only a minor influence on small sized (nucleation-Aitken mode) aerosol, we suggest future work focuses on what the influence may be under 'pristine' conditions. Planned work will target the 'clean marine' period identified in the R2R campaign and consider the downwind processes from the RVI to AIRBOX, where the airmass crosses coral reefs. As this work has shown, small increases in sulfate aerosol mass is found directly over coral reefs and how this evolves downwind is of interest. Of further 365 interest, and perhaps yielding more significant results, would be a study conducted under pre-or post-industrial emissions conditions. Simulations such as this become particularly relevant if we consider a post-anthropogenic aerosol emissions world, in which coral reefs such as the Great Barrier Reef may already be extinct.
Code and data availability. RVI data including ship location, meteorology, black carbon and radon are available on the CSIRO Marlin Metadata System: https://www.cmar.csiro.au/data/trawler/. Remaining RVI and AIRBOX data will be submitted to PANGAEA in the near future, and in the mean time is available upon request. WRF-Chem namelists are avialable upon request and data can be made upon reasonable request. WRF-Chem analysis was performed using the wrf-python software package (Ladwig, 2017).
the Model for Simulating Aerosol Interactions and Chemistry (MOSAIC) aerosol scheme. Dry deposition of gases and aerosol are turned on, as is wet scavenging (including convective wet scavenging). In-cloud chemistry, turbulent mixing and subgrid convective transport is also switched on. The FTUV (Fast Tropospheric Ultraviolet-Visible) photolysis scheme (Tie, 2003) is used.
MOSAIC represents aerosol via a sectional approach with eight discrete size bins. For each bin, the number and mass of 380 particles are simulated: defined by the lower and upper limit of the dry particle diameter (Zaveri et al., 2008). Particle growth is calculated in a Lagrangian manner and transfer of particles between bins is calculated using a two-moment approach (Simmel and Wurzler, 2006). Coagulation of aerosol is calculated according to Jacobson et al. (1994). Homogeneous nucleation of H 2 SO 4 -H 2 O in MOSAIC is calculated via the Wexler et al. (1994) scheme. In MOSAIC, growth to Aitken mode particles is simulated implicitly as newly nucleated particle sizes are smaller than the smallest simulated aerosol size in the model.

385
Heterogeneous nucleation in MOSAIC is partitioned into two schemes, treating condensation of non-volatile gases (H 2 SO 4 and methanesulfonic acid) and condensation and evaporation of semi-volatile gases (HNO 3 , HCl and NH 3 ) separately (Zaveri et al., 2008).
MOSAIC includes 11 specific aerosol species: sulfate (SO 2 -4 and HSO -4 ), methanesulfonate, nitrate, chloride, carbonate, ammonium, sodium, calcium, black carbon, primary organic matter plus water, and treats other unspecified aerosol species as a lumped mass or through substitutions of equivalent species. Some gas phase species, including sulfuric acid and MSA, are allowed to partition to the particle phase. Atmospheric DMS chemistry is not a part of the CBMZ scheme, but was added with the development and coupling of MOSAIC (Zaveri et al., 2008). The DMS chemistry is based on that of Zaveri (1997) and includes 11 species and 30 reactions.
The Liss and Merlivat (1986) parameterisation is used to calculate the flux DMS emissions in WRF-Chem. As noted in Fiddes et al. (2018) and (Fiddes et al., 2021), the Liss and Merlivat (1986) parameterisation is considered a conservative method but is believed to be the most realistic (Vlahos and Monahan, 2009;Bell et al., 2017). Sea salt emissions are calculated online via either the Gong et al. (1997) (Longo et al., Guenther scheme (Guenther et al., 1994;Simpson et al., 1995)

A2 Physics
All simulations use the Morrison double moment cloud microphysics scheme (Morrison et al., 2008). The RRTMG (Rapid Radiative Transfer Model for General circulation models) model for longwave and shortwave radiation is used (Iacono et al., 2008), including the Monte Carlo Independent Column Approximation method for random cloud overlap (Barker et al., 2003).
Cumulus parameterisation was performed using the Grell 3D scheme, similar to the Grell-Devenyi Ensemble Scheme (Grell 415 and Dévényi, 2002). In addition, cumulus radiation effects are switched on, allowing interaction of the radiation scheme and parameterised convective clouds (Gustafson et al., 2007). Cloud fractions are calculated using the Xu and Randall (1996) method, while cumulus and aerosol radiative feedbacks are permitted. Aerosol optical properties (Mie calculations) are approximated using the volume averaging method. The Noah Land Surface Model is used, with soil temperature and moisture in four layers, fractional snow cover and frozen soil physics (Chen and Dudhia, 2002). The boundary layer scheme used is 420 the Mellor-Yamada-Janjic scheme (Janjić, 1994) operating in conjunction with the surface layer physics scheme, Eta (Janjic, 1996).
In WRF-Chem, with the chemistry and physics options described above, aerosol direct and indirect effects are permitted via the radiative, photolysis and cloud microphysical schemes. For direct aerosol effects, the size, number and composition of aerosol and aerosol water, refractive indices of aerosol types (based on literature) and the Mie calculations (Bohren and 425 Huffman, 1998) update the AOD, the single scattering albedo and the asymmetry factor used in the RRTGM radiation scheme.
Aerosol water has a large impact, and hence the relative humidity must also be considered when direct effects are being examined. In this work, water vapour has been nudged to the BARRA reanalysis, hence should not change significantly between experiments.
For indirect aerosol effects, the CCN number and composition is used to calculate the CDN in aerosol activation modules 430 (Abdul-Razzak and Ghan, 2000Ghan, , 2002. Activation depends on the composition and size of the particle (i.e. their hygroscopic properties), as well as the vertical and turbulent velocities of the air mass. First, second and semi indirect effects are calculated between both the chemistry and microphysics modules. were pumped through to the wet chemistry laboratory as part of the ship's routine underway measurements every minute. For Table B1. List of RVI observations used in the WRF-Chem evaluation, the institute responsible for data collection and processing, details on the instrument used and literature relevant to the data processing methods gas chromatography) was used to detect DMS w . After the 14th October, due to competing demands for the GC, DMS w was measured using an Equilibrator Inlet (EI) -Proton Transfer Reaction Mass Spectrometry (PTRMS) system (Kameyama et al.,440 2009; Omori et al., 2013Omori et al., , 2017. The seawater samples pumped by the ship system were flowed continuously into a 10-L glass equilibrator at 1 L min −1 . Pure nitrogen flowed from the bottom to the upper outlet of the equilibrator at 120 sccm. Dissolved DMS was extracted into the N 2 gas phase and introduced into the PTRMS (Ionicon Analytik GmbH, Innsbruck, Austria). The mass signal of DMS in the sample gas was obtained at 10-s integration at 1-min intervals. The DMS w concentrations were calculated from concentrations in the sample gas extracted from the equilibrator with Henry's law constant (Kameyama et al.,445 2009). Comparison between the DMS w observations taken by the two techniques show excellent agreement (r 2 = 0.999) with high confidence (p < 0.001).
Accompanying the DMS w measurements on board the RVI were DMS a observations. A commercially available high sensi- and a average primary ion signal (m/z 19) of 1.78 E+07 cps. Data was filtered to remove periods of instrument instability.
The PTQRMS was operated with a CSIRO custom built auxiliary system which controlled whether the PTQRMS sampled 455 VOC-free air to determine instrument background, calibration gas to determine instrument response or ambient air. Zero air measurements occurred twice daily (1 -2 AM and 1 -2 PM UTC) and an interpolated zero signal was subtracted from the reported ambient and calibration measurement signals. Once per day (1400 to 1500 UTC) calibration measurements occurred by diluting a certified calibration gas standard containing 997 ppb DMS a (Apel -Reimer Environmental Inc, Colorado, USA) (stated accuracy ±5%) into VOC-free air. The instrument sensitivity to DMS a was 11.3 normalised cps ppb −1 (rel. stdev 460 ±3%). The minimum detectable limit (MDL) for a single 5 sec measurement at m/z 63 (DMS a ) was 0.029 ppb determined using principles of ISO 6879 (ISO 1995). Values less than the MDL were removed.
At AIRBOX, DMS a was measured using an automated gas chromatograph (GC) -pulsed flame photometric detector (PFPD).
Measurements were taken every 20 minutes via an auto-sampler programmed to control the GC-PFPD. Full details on the instrumentation and data processing, including uncertainties, can be found in Swan et al. (2015). For DMS a measurements at 465 both platforms, the closest measurement within ±10minutes to the hour was taken for comparison to the instantaneous hourly output from WRF-Chem.

B2 Aerosol fields and radon
Mass concentrations of black carbon (BC) at AIRBOX and on the RVI were obtained with a Thermo Scientific Model 5012 multi-angle absorption photometer (MAAP) and were used to help identify periods of air contaminated by ship exhaust. The

470
MAAP sampled through a dedicated PM10 inlet, which was heated to minimise the influence of humidity on the BC measurements (Kanaya et al., 2013). Samples were acquired at 5 s time intervals and have been averaged to a 10-minute time resolution.
An Aerodyne compact Time-of-Flight Aerosol Mass Spectrometer (AMS) provided the non-refractory chemical composition of submicron aerosol at AIRBOX (Drewnick et al., 2005). The AMS sampled through a membrane dryer (Nafion MD-700) 475 and a silica gel diffusion dryer which maintained sample relative humidities below 40%. Daily measurements through a highperformance particle filter were used to calculate detection limits and correct for concentrations of background air species (Allan et al., 2004). Samples were averaged to 10-minute intervals. At this time resolution, the campaign-average detection limit was 5 ng m −3 for SO 2 -4 . On board the RVI, an Aerodyne Time of Flight Aerosol Chemical Speciation Monitor (ACSM) was used to obtain chemical 480 composition of the non-refractory submicron aerosol. A full description of its design and operation is given in (Fröhlich et al., 2013). The ACSM inlet efficiency is at a maximum for vacuum aerodynamic diameters between 100-450 nm (Jayne et al., 2000;Liu et al., 2007) and therefore the composition measurements best represent accumulation mode aerosol. The AMS sampled through a membrane dryer (Nafion MD-700) which maintained sample relative humidities below 40 %. Samples were averaged to 10-minute intervals, and at this time resolution the campaign-average detection limit was 18 ng m −3 for SO 2 -4 .
In addition, atmospheric radon-222 concentrations were measured both at AIRBOX and on the RVI using dual-flow-loop two-filter atmospheric radon detectors. Radon concentrations have been shown to be an accurate, independent measure of residual terrestrial influence within an airmass. Radon can subsequently be used to determine if an airmass has a marine or terrestrial origin, and can be satisfactorily used to detect 'baseline' airmass at locations such as Cape Grim, Tasmania, Ausralia (Chambers et al., 2021). While at Cape Grim, the baseline radon concentration is considered to be 30 mBq cm −3 or below, 490 on board the RVI and at AIRBOX, terrestrial influence over the airmasses was much higher given their coastal location, prevailing wind directions and the fact that coral atolls, when exposed, are also a source of radon. For the RVI, a threshold of 300 mBq cm −3 was used to determine marine from terrestrial airmass. At AIRBOX no satisfactory threshold of radon could be determined to separate marine and terrestrial influences.

495
On board the RVI, meteorological observations were taken as part of the routine observations. Observations are available at 5 or 10 second, or 5 minute intervals. The five minute interval has been used in this study. Where port and starboard observations were available, an average over the two was taken. These observations have been processed by the Marine National Facility Chen, Z., Schofield, R., Rayner, P., Zhang, T., Liu, C., Vincent, C., Fiddes, S., Ryan, R. G., Alroe, J., Ristovski, Z. D., Humphries, R. S., Key-