The coupled climate–chemistry model ACCESS-UKCA (Australian Community Climate and Earth System Simulator–United Kingdom Chemistry and Aerosol) is used in this work to quantify the importance of global coral-reef-derived sulfur. The ACCESS-UKCA physical atmospheric model is based on the Global Atmosphere 4.0 configuration of the Unified Model at Version 8.4 , and the UKCA model includes the state-of-the-art GLObal Model of Aerosol Processes (GLOMAP)-mode scheme . GLOMAP-mode is a two-moment microphysical aerosol scheme that simulates aerosol mass and number distributions across four soluble modes (corresponding to nucleation, Aitken, accumulation and coarse modes) and, in this work, one insoluble mode (Aitken) . The size distributions of these modes are shown in Table .

GLOMAP-mode simulates coagulation , condensation of H2SO4 and secondary organics , new particle formation of sulfate aerosol in the free troposphere via binary homogeneous nucleation and organic-mediated boundary layer nucleation . Aqueous chemistry calculates the dissolution of SO2 and H2O2 into cloud droplets within the soluble accumulation and coarse modes , and aerosol dry deposition is calculated following , and . GLOMAP-mode includes particle compositions of sulfate, sea salt and elemental and organic carbon .

The DMS surface water concentration (DMSw) climatology is used in this work. and before that used a global database of observations to derive a gridded global DMSw climatology via a piecewise cubic Hermite interpolation technique over 54 well-defined biogeographic ocean provinces. This climatology is the standard for most global climate modelling studies and is discussed in more detail in .

In this work, as in , the fluxDMS parameterisation is used, and a full description of this scheme can be found in . Detailed discussion about fluxDMS parameterisation choices and sensitivities can also be found in . In short, the scheme calculates the fluxDMS under three wind-induced sea states representing smooth (10 m wind speeds less than 3.6 m s-1) and rough (10 m wind speeds between 3.6 and 13 m s-1) gas transfer as well as wave breaking and bubble bursting (10 m wind speeds greater than 13 m s-1). is considered a conservative parameterisation and was chosen for this work based on recommendations by and that suggest many parameterisation overestimate the fluxDMS.

With online chemistry, ACCESS-UKCA includes four key oxidative pathways to convert DMS into SO2, which are shown in Table . SO2 can then be further oxidised into H2SO4 (Table ), after which it can contribute to aerosol growth or new particle formation. Description of these processes can be found in .

Anthropogenic emissions, pre- and post-2000 respectively, are provided by and and biomass emissions by . Emissions of other species (biogenic, primary aerosol) are described in . SSTs and sea ice are prescribed following the Atmospheric Model Intercomparison Project (AMIP) method . The UKCA is coupled to the ACCESS model via the radiation scheme and the large-scale cloud and precipitation schemes; both the direct and indirect aerosol forcing are modelled. ACCESS-UKCA has a resolution of 1.25∘ latitude × 1.875∘ longitude with 85 vertical levels. Where the model is nudged, ERA-Interim is used at 6-hourly intervals (via horizontal wind components and potential temperature; see Sect.  for more details). Further model details and evaluation are available in .

We have developed a DMS surface water concentration (DMSw) climatology based on , in which additional DMSw over coral reef regions is included. To determine the amount of DMSw to be added to the climatology, we first needed to know where coral reefs were located globally. The global coral reef distribution database was used to calculate the fraction of each ACCESS-UKCA grid box covered by coral reefs. The database is the most comprehensive global database of warm water coral reefs, 85 % of which was drawn from the Millennium Coral Reef Mapping Project, a remote sensing project at spatial resolutions of up to 30 m.

The gridded areal distribution drawn from the coral reef database, shown in Fig. , was then used to weight a fixed concentration of DMSw to be added to the DMSw climatology. For example, the maximum fraction in any grid box found in Fig.  is 15.6 % in the northern GBR. In this study, we used 50 nM as the fixed concentration (a number of different DMSw concentrations were tested, from 10 to 500 nM). Therefore, at this grid box, with the highest density of coral reefs, the amount of DMSw added to the climatology as a coral reef source is 0.156×50=7.8 nM. The choice of 50 nM was somewhat subjective, in part due to the relatively few estimations of large-scale coral reef production of DMSw, as described in Sect. . Nevertheless, below we describe how this choice aligns with observations found in the literature. In addition, we made a conscious choice to create a climatology that represents a plausible maximum DMSw in an attempt to ensure a response to this perturbation. We note that at this initial stage, the DMSw climatology developed here does not vary in time beyond that of the monthly dataset.

The weighted addition of 50 nM of DMSw caused a global mean increase of 0.03 nM and an additional fluxDMS of 0.3 Tg yr-1 S. The fluxDMS of the climatology and the additional coral reef fluxDMS are shown in Fig. . These increases do not impact the global sulfur budget, contributing only 1.7 % of additional sulfur to the global fluxDMS. The 50 nM climatology adds a mean of 0.74 nM and a maximum of 7.8 nM to coral reef regions, values that are within those found in the literature. The additional daily fluxDMS simulated by ACCESS-UKCA over coral reefs, shown in Fig. b (maximum of 621.9 µg m-2 d-1 S), is similar to that of the estimations of fluxDMS due to three coral species in response to tidal stress (708.9–1547.6 µg m-2 d-1 S). Furthermore, suggest that the total fluxDMS from the GBR and surrounding lagoon is approximately 0.002 Tg yr-1 S. If this number is extrapolated to global coral reef regions, then an annual fluxDMS of 0.12 Tg yr-1 S is estimated. The values from the and estimations both suggest that the amount of DMSw attributed to coral reefs in this study is within the high end of what is currently observed.

However, it is noted that the values stated here are averages over large grid boxes and so likely overestimate the extent of coral reef influence. Nevertheless, the 50 nM perturbation was chosen in part to ensure that if no significant changes in the atmosphere were found, it would not be because the additional coral reef DMS was too small.

To study the impact of coral-reef-derived DMS in ACCESS-UKCA, two sets of simulations were performed: nudged and free-running. In both sets of simulations, a control (using the , climatology, referred to as L11) and experimental simulation (using the , climatology with additional coral and hereafter referred to as the L11C50 simulation) were completed. The nudged simulations follow the methods described by , in which the control and coral simulations were nudged to the ERA-Interim dataset in the free troposphere at 6-hourly intervals using horizontal winds and potential temperature. We note that by nudging the model, we limit the model's ability to respond to the DMS flux perturbations. However, nudged simulations, by restricting meteorological feedbacks, give us an indication as to which responses are purely dynamical as opposed to those that may in fact be caused directly by DMS perturbations.

In the second set-up, the model was allowed to freely run, with no nudging applied. In both sets of simulations, SSTs are prescribed. Without nudging, feedbacks from the meteorology, such as changes in wind fields, are able to manifest within the model. In the free-running simulations, much greater model variability manifested, and differentiating between a true signal from the perturbed DMS field and internal model variability was difficult. For this reason, seven simulations (for both the L11 and L11C50), each of 10 years, were performed with the free-running set-up to provide an ensemble. Each set of simulations used different atmospheric initial conditions from a previous nudged simulation, taken at 00:00 UTC on 1 January for 1996, 1998, 1999, 2000, 2001, 2002 and 2003 (noting that 1997 was excluded as it was a strong El Niño year). Whilst this exercise was computationally expensive, it was able to provide sufficient data to perform statistical analyses.

The following regions were defined for statistical analysis: the Maritime Continent–Australian (MC-Aus) region from 17.4∘ S to 10∘ N, 95.625 to 153.75∘ E; the Queensland (QLD) land-only region from 30 to 10∘ S, 138.75 to 153.75∘ E; and the Australian land-only region from 45 to 10∘ S, 112.5 to 153.75∘ E. Four grid points were selected for analysis of aerosol size distribution: a location off the coast of Papua New Guinea (PNG) at 10.5∘ S, 151∘ E; a point in the East Pacific (E.Pac) at 10.5∘ S, 165∘ E; a location over the GBR at 20∘ S, 151∘ E; and a point in inland QLD at 20∘ S, 140∘ E. These regions and points are shown in Fig. a. In addition the boundary of the South Pacific convergence zone (SPCZ) is defined as the area within the 6 mm d-1 or greater precipitation threshold .

To test the significance of differences in a given field, the two-tailed Student T test and the respective field significance are used. Both of these methods have been evaluated at confidence levels of the 95th percentile (p<0.05) unless otherwise stated. In addition to this, ensemble agreement, where at least five out of the seven ensemble members agree on the change in sign, has been shown as stippling as a further indicator of confidence.

This section describes the changes in atmospheric DMS (DMSa) when coral reefs are removed (L11 minus L11C50), noting that the results are presented this way in order to demonstrate what the impact of the loss of coral reefs may be on the climate system. The annual differences in DMSa (with respect to the L11C50 simulation, shown in Fig. a) for the nudged simulations (Fig. b) and the free-running ensemble (Fig. c) are spatially very similar to each other and to the respective change in fluxDMS (shown in Fig. b). Good agreement across the ensemble is found with the free-running simulations over reef regions. Figure d–g show, for both the free and nudged simulations, a strong seasonal signal over the MC-Aus region in terms of both mean value and the range of values. A seasonal signal is found in both the DMSa and fluxDMS (see Table ) and is in part due to the variation of the L11 DMSw climatology. In DJF, the range of results from the ensemble is larger than that of any other season, and it appears that the free-running meteorology is actually dampening the response of DMSa compared to the nudged runs. This damped response is not surprising as surface wind speeds were found to be stronger in this region at this time in the free ensemble (not shown). Increased wind speeds in the free ensemble cause an increased fluxDMS, leading to a smaller decrease in fluxDMS and DMSa than in the nudged run. Such examples of wind-driven DMS responses are also found in other locations (and in other seasons), including over the Southern Ocean at around 60∘ E and 60∘ W (see annual plots in Fig. c). These responses highlight the complexity of the DMS–climate system. In addition, wind-driven responses in other aerosol sources are found in regions around the globe (e.g. from sea salt and dust) but are not discussed further here.

Figure  shows changes in SO2 that are spatially similar to those of DMSa. However the reductions around smaller coral reefs, such as those in the central Pacific, Indian or Caribbean oceans, are of lesser magnitude than those of the coral reefs in the MC-Aus region. This is true for both the nudged (Fig. b) and the free simulations (Fig. c), which implies that these areas of lower reef density are unlikely to have a significant impact on regional climate. In terms of seasonality, Table  and Fig. d–g indicate less SO2 variability compared to DMSa throughout the year for both the free and nudged simulations. MAM (March, April, May) has the smallest, statistically insignificant change (-0.9 % and -1.3 % for the free and nudged simulations respectively), while DJF (December, January, February) and JJA (June, July, August) show larger changes over the MC-Aus region of -3.2 % and -3.8 % (p<0.05) in the free ensemble and -2.7 % and -2.2 % (p>0.1) for the nudged runs. The violin plots indicate there is a similar degree of variability for each of these seasons (compared to DMSa), and the influence of free-running meteorology is again noted where changes in SO2 are found outside of coral reef regions.

The removal of coral-reef-derived DMS leads to a significant (p<0.05) decline in nucleation-mode aerosol number concentration of -5.1 %, -6.0 % and -6.2 % for MAM, JJA and SON respectively in the free ensemble over the MC-Aus region (see Table ). In the nudged simulations, nucleation-mode number concentration decreases by 4.6 % (p<0.05) in SON and by 4.4 % (p<0.1) in MAM. Although not statistically significant, JJA continues to show the largest changes (-6.1 %) and DJF the smallest (-3.8 %) in the nudged runs. The changes in nucleation-mode number concentration are strongly reflected in the number concentrations of particles greater than 3 nm dry diameter (N3) which are shown annually in Fig. . The violin plots in Fig. d–g show a large range in the response in DJF while a much smaller range in SON, with many of the individual free-running pairs agreeing on the magnitude of the change.

These small aerosols (nucleation-mode-sized) show a greater response to changes in DMS than the larger aerosol sizes (see Table ) in both the free and nudged simulations, aligning with the results of . This result indicates that fewer homogeneous nucleation events are occurring in the free troposphere as a result of removing coral-reef-derived DMS. Hence, fewer particles are being entrained back into the boundary layer. This process is confirmed by the vertical profile of N3 showing a decrease in the upper levels of the atmosphere (not shown).

The size distributions shown in Fig.  at four grid points also show the larger decreases are occurring in the nucleation mode and the Aitken mode. The range of locations shown by the size distribution, including directly over coral reefs (GBR and PNG) and remotely from coral reefs (East Pacific and QLD), demonstrates that the effect of coral-reef-derived DMS loss is not restricted to directly over coral reef regions. This is shown notably at the QLD location for JJA in both the nucleation and Aitken modes and SON for the Aitken mode. Over the MC-Aus region, decreases in the soluble Aitken aerosol number occur in MAM and JJA of 1.9 % and 2.1 % respectively (p<0.05). The changes in both the nucleation- and Aitken-mode number concentrations are accompanied by similar changes in aerosol mass at this size (see Table ).

At the larger sizes (soluble accumulation and coarse modes) little change in aerosol number is found on average over the MC-Aus region. However, significant (p<0.05) decreases in aerosol mass are found in the coarse mode during DJF, JJA and SON and in the accumulation mode in DJF (p<0.1) and JJA (p<0.05) for the free-running ensemble (see Table ). Declines of smaller magnitude (p>0.1) are found for the nudged runs. For the free ensemble, the larger change in mass instead of number suggests a reduction in aerosol growth at these sizes as the larger aerosol sizes rely on condensational growth and cloud processing to interact with DMS-derived sulfate. One exception to the observed change in mass instead of number is the region east of PNG, which varies by season (not shown). In this region, a small increase in accumulation-mode aerosol number is found, accompanied by a larger increase in the aerosol mass. This increase is more evident in the nudged runs.

So far we show that the largest response in aerosol due to loss of coral-reef-derived DMS occurs over the MC-Aus region. For this reason, all subsequent plots focus on the MC-Aus region. It is noted that there are impacts on aerosol from the free-running meteorology outside of this domain, but these can mostly be explained by changes in surface wind speeds due to variability in the free-running ensemble (not shown). In addition, the influence of the free-running meteorology is found to have a larger impact on the results in subsequent analysis, with greater seasonal variation.

Despite the removal of coral-reef-derived aerosol having the largest impact on the nucleation-mode aerosol, these small aerosols do not interact with model radiation via direct or indirect aerosol effects. While the changes are small in the larger sized aerosol number and mass, a cumulative response to the loss of coral-reef-derived sulfur has some interesting impacts. Figure  shows the column-integrated seasonal cloud condensation nuclei with a dry diameter greater than 70 nm (CCN70) response to removal of coral-reef-derived DMS. For the nudged simulations in Fig. b, f, j and n, a consistent, yet small and insignificant reduction in CCN70 is found over the MC-Aus region (between 0.6 % and 0.9 %). Interestingly, over the SPCZ region, decreases of 1.0 %, 0.3 %, 1.9 % and 1.1 % are also found in the nudged simulations for each season (DJF, MAM, JJA and SON) respectively. The SPCZ is a relatively clean region, with few CCN70-sized aerosols (see Fig. a) and also few coral reefs. This lack of direct aerosol source suggests that changes in aerosol from other regions are affecting this region via transport along the SPCZ.

For the free ensemble, in Fig. c, g, k and o, the SPCZ region again stands out in MAM, JJA and SON, with decreases in CCN70 of 2.8 %, 2.2 % and 1.9 % for each season respectively (MAM and SON, p<0.1, and JJA, p<0.05). In this region, little change in surface wind speeds is found, suggesting that the changes found here are also likely due to loss of aerosol from coral reefs and their subsequent transport, as seen in the nudged runs.

Over the MC-Aus region in the free-running ensemble, JJA and SON have decreases in CCN70 of -1.4 % (p<0.1 and p>0.1 respectively). The violin plots in the far right column of Fig.  indicate large model spread in the CCN70 changes. SON (Fig. o) has the most ensemble agreement over the MC-Aus region, while little agreement is observed in other seasons. For this reason (and because the subsequent dynamical results to be discussed in Sect.  onwards were the largest in this season), the SON results will be shown and discussed from this point only (although the statistics for all seasons can be found in the continuation of Table ).

Whilst the changes in aerosol discussed in this section are small, it is worth noting that in , where all marine DMSw was removed, a global decrease of 8 % of all N3 (17 % for Australia) was found. For CCN70, a decrease of just 5 % was found globally (8 % for Australia). It is clear from the perturbation of total marine DMS that global DMS contributes only a small amount to the total aerosol number. Hence local differences from coral-reef-derived DMS found in this study could be considered to be relatively large.

Aerosols in the Aitken, accumulation and coarse modes are used by the ACCESS-UKCA radiation scheme to calculate aerosol direct effects. In Fig. , the SON aerosol optical depth (AOD), clear-sky outgoing shortwave radiation at the top of the atmosphere (abbreviated to SW↑TOA,CS) and clear-sky incoming shortwave radiation at the surface (SW↓Surf,CS) are shown. The clear-sky radiation fields are examined in this section to allow for the detection of direct effects from aerosol without the influence of clouds. Changes in the all sky fields are dominated by the convective cloud response, which will be discussed in Sect. .

The spatial response of AOD to changes in aerosol in the nudged simulations (Fig. b) is broadly similar to the CCN70 changes (Fig. n). Decreasing AOD is found over most of the MC-Aus region, while the area to the east of PNG experiences an increase. The seemingly amplified aerosol response in the PNG region is due to the increased accumulation-mode sulfate mass. This increase is not believed to be associated with coral-reef-derived DMS removal but a function of complex non-linearities and model sensitivities. This anomaly is likely to be dampening the area averages over the MC-Aus region.

In Fig. f, the change in nudged SW↑TOA,CS is consistent with the change seen in AOD, indicating that there is a weak reduction in the amount of shortwave radiation being reflected out to space at the top of the atmosphere over significant coral reef regions. This decrease in SW↑TOA,CS suggests more shortwave radiation is passing through the atmosphere and reaching the surface. For the nudged runs, this is broadly true, as shown by the SW↓Surf,CS in Fig. j. These results indicate that when no interaction with meteorology is allowed, a weak, statistically insignificant direct aerosol effect is associated with certain regions over the MC under clear-sky conditions in austral spring.

For the free-running ensemble, however, the result is far less clear, primarily due to interactions with meteorology. The AOD response (Fig. c) is again broadly consistent with the changes in CCN70 (Fig. o), where over the MC-Aus region a decrease of 1.7 % (p<0.1) is found. The SW↑TOA,CS responds as expected with a decrease of 0.11 W m-2 over the MC-Aus region (p<0.1, with considerable model agreement). However, the subsequent SW↓Surf,CS response (Fig. k) is not as clear. While the SON MC-Aus average suggests a weak, insignificant increase in shortwave radiation reaching the surface, the spatial patterns are inconsistent with the AOD, aerosol fields or the SW↑TOA,CS.

In ACCESS-UKCA, the shortwave (defined as wavelengths between 0.2 and 5 µm) radiative transfer scheme considers not just the scattering and absorption of energy by aerosol and cloud droplets, but also the absorption of energy by water vapour . Water vapour has an effect at wavelengths greater than 0.7 µm. In Fig. o, the mean change in water vapour throughout the column is shown, which is more spatially consistent with the free-running SW↓Surf,CS response than that of the AOD or SW↑TOA,CS. This consistency is particularly clear over the Australian region. An increase in water vapour in the column would suggest more absorption of radiation throughout the column, and hence less energy is received at the surface, as found in the SW↓Surf,CS results (Fig. k). Upper level warming over the tropics (not shown) is also found, and we note that changes in temperatures nearer the surface are limited due to the prescribed SSTs. It is emphasised that this interaction of energy with water vapour is only found in the free ensemble, when meteorology is allowed to vary.

The cause of the change in water vapour is difficult to determine as it is intricately linked to both local- and large-scale climate processes. For example, meteorological effects found in this study that could explain the water vapour increase include the following:

warming at upper levels (15–20 km) of the atmosphere over the tropics (not shown), indicating more water vapour can be held in the atmosphere, but could also be a result of increased water vapour;

It is hypothesised that the small reduction in aerosol resulting from decreased DMSa has caused an increase in shortwave radiation passing through the atmospheric column, resulting in warming, which may cause increased evaporation, vertical transport and convective activity and subsequently increased water vapour. Despite many of the responses described above having good ensemble agreement, low confidence is attributed to this hypothesis for three reasons:

The top-of-the atmosphere direct radiative effect that is proposed to initiate these responses is small. Furthermore, the changes in aerosol and AOD that have caused the radiative effect are also small and insignificant. Hence it is unreasonable to suggest such small changes in aerosol could be causing a direct radiative effect.

Therefore, despite a weak yet significant (p<0.1) decline in SW↑TOA,CS, it is concluded that over the MC-Aus region, no robust impact on climate via the direct radiative effect can be confidently detected. This result may be due to averaging over a large area, as the same processes are found over QLD with generally greater statistical significance and ensemble agreement. This is discussed further in Sect. .

Indirect aerosol effects, such as cloud brightening or lifetime effects, take place as CCN particles activate and become cloud droplets. In ACCESS-UKCA, aerosol activation depends on the size and composition of aerosol as well as the atmospheric supersaturation, which is influenced by the vertical velocity. Thus, the significant increase in smaller size aerosol found in Sect.  may have some influence over indirect aerosol effects in certain conditions, although it is noted that larger aerosols have a much greater ability to influence indirect effects. Figure  shows the responses of cloud droplet number (CDN), cloud liquid water path (LWP) and low cloud fraction (noting that only the cloud microphysics scheme can respond to change in CDN, not the convective scheme). For both the nudged and free simulations, the changes found in CDN are consistent with, though weaker than, the changes observed in the CCN70 fields (noting again the vertical integral of CDN through the column is shown). No statistical significance is attributed to the changes in CDN over the MC-Aus region, although reasonable ensemble agreement is found in Fig. c. These weak changes suggest that the reductions in CDN are unlikely to have an effect on cloud properties or large-scale precipitation (via the second indirect effect where fewer CDN, given the same availability of liquid water, would increase rainfall).

A small increase in LWP is found on average over the MC-Aus region in the free simulations, although Fig. g–h show a large amount of variability in the region and little model agreement. This change in CDN and LWP has had little impact on properties such as low cloud fraction (Fig. k) or large-scale precipitation (not shown), as shown in Table , over the MC-Aus region for each season. In addition there is very little model agreement in these responses. For the nudged simulations, constrained meteorology has meant little response of cloud properties is allowed. Subsequently, these results suggest that the loss of coral-reef-derived aerosol has little or no impact on climate via the indirect aerosol effects.