Introduction

DMS, as the predominant volatile sulfur compound residing in the upper ocean surface, assumes a pivotal role in the regulation of Earth’s climate1,2. Upon its release into the atmosphere, DMS undergoes oxidation, giving rise to the formation of SO42− particles. These particles directly reflect solar radiation and also act as cloud condensation nuclei (CCN), manipulating the radiative properties of clouds and thereby inducing a subsequent cooling effect on Earth’s surface temperatures3,4. The Intergovernmental Panel on Climate Change (IPCC) Sixth Assessment Report (AR6)5 reveals that future scenarios project a substantial absorption of 38–70% of CO2 emissions by oceans and land by the end of the 21st century. This absorption is anticipated to lead to a significant reduction in ocean pH, resulting in profound consequences for marine biological processes and potentially influencing DMS production. Therefore, it is crucial to incorporate the mechanisms underlying the response of DMS production to ocean acidification (OA) in order to furnish reliable projections of DMS emissions.

Numerous mesocosm studies6,7,8,9,10,11 have contributed to enhancing our understanding of the response of DMS production to ocean acidification (OA), indicating that sea-surface DMS concentrations tend to decrease in the presence of high CO2/low pH conditions. However, the relationship between pH (pCO2) and algal DMS production, as well as DMS decomposition, is intricate12. Some studies13,14 have indicated that a decrease in pH can lead to reduced DMS production by affecting the phytoplankton responsible for DMSP production (the primary precursor of DMS), but cases where DMSP production remained unaffected have been reported15. Six et al.16 utilized the correlation between sea-surface DMS concentrations and pH to project future global DMS emissions and associated radiative forcing. State-of-the-art Earth System Models (ESMs) have incorporated mechanisms to simulate the response of DMS production to OA, some17,18 directly employing parameterization schemes proposed by Six et al.16. Nevertheless, the predictive capacity of this relationship in ESMs has not been evaluated due to inadequate observational coverage19 of biogeochemical variables. Furthermore, the current generation of ESMs exhibits limitations in accurately constraining marine productivity20, which can restrict their ability to precisely predict the response of DMS production to OA.

This study aims to evaluate the future trajectory of global sea-surface DMS concentrations within the context of ocean acidification driven by elevated CO2 levels, as well as to investigate their impact on aerosol radiative forcing. To accomplish this, we established a robust relationship between sea-surface DMS and pCO2 (a critical indicator of ocean acidification) using comprehensive observational datasets spanning a global scale. This relationship was then employed to refine the sea-surface DMS concentration outputs derived from multiple models participating in the Coupled Model Intercomparison Project Phase 6 (CMIP6)21. Consequently, we generated projections of global sea-surface DMS concentrations from 2020 to 2099 under the socio-economic Pathway (SSP) 5-8.5 scenario22. Furthermore, we quantified the radiative forcing induced by DMS by integrating an advanced aerosol microphysics model with a global 3D-chemical transport model driven by meteorology data for the SSP5-8.5 future scenario.

Results

The relationship between pCO2 and DMS and the projection method

We observed a strong negative correlation between sea-surface pCO2 and DMS across the global ocean, with an R2 value of 0.89 (Fig. 1a). The matched data, as described in “Methods”, were subsequently partitioned into two groups based on seawater location in the Northern and Southern hemispheres, as illustrated in Supplementary Fig. 2a, b. Notably, the strength of the anti-correlation is relatively weaker for the Northern hemisphere (R2 = 0.40) compared to the Southern hemisphere (R2 = 0.87). In the Northern hemisphere, there are significantly higher nutrient inputs compared to the Southern hemisphere, and regions downstream of densely populated areas exhibit elevated deposition rates23,24. We postulated that these factors contribute to a broader range of fluctuations in DMS concentrations in the Northern Hemisphere due to increased human activities. The plot displaying the correlation between remotely sensed climatological chlorophyll (Chl) data and changes in DMS measurements (Supplementary Fig. 2c, d) reveals that the correlation is lower in the Northern Hemisphere (R2 = 0.23) compared to the Southern Hemisphere (R2 = 0.53). In addition, within the same range of DMS concentration changes (Log10(DMS): 0–0.7), corresponding Chl concentration fluctuations in the Northern Hemisphere are more significant compared to those in the Southern Hemisphere. However, due to discrepancies between satellite-derived Chl concentrations and in-situ Chl data, satellite-based Chl data may weaken the relationship between Chl and DMS25.

Fig. 1: Assessment of pCO2 and DMS relationships in ESMs.
figure 1

The scatter plot of pCO2 and DMS measurements across the global ocean (a), and four ESMs estimated monthly oceanic pCO2, DMS concentrations, and NPP (b). The red dashed lines indicate the linear regressions between pCO2 and DMS, as well as pCO2 and NPP, as estimated by the ESMs. Significant trends are marked by asterisks (∗ for P < 0.05 and ∗∗ for P < 0.01). The values of trends are evaluated using the Mann–Kendall test. R2, Slope, and N represent Pearson’s correlation coefficient, the slope of the regression line, and the number of data points.

Full size image

We devised a projection method based on the relationship derived from Fig. 1a and applied it to improve the estimation of sea-surface DMS concentrations provided by CMIP6 ESMs under the SSP5-8.5 future scenario. In this method, we calculated the changes in model-predicted oceanic Log10(pCO2) (ΔpCO2) and Log10(DMS) (ΔDMS) for each specific month (e.g., from January 1st of the current year to January 1st of the previous year). For each grid cell, if the ratio of ΔDMS to ΔpCO2 exceeded the regression slope defined in Supplementary Eq. (11), we refined ΔDMS by multiplying the slope from Supplementary Eq. (11) by ΔpCO2. Conversely, when this ratio fell below the specified threshold, we retained the ΔDMS estimates derived directly from the ESMs. For a detailed description of the projection equations (Supplementary Eqs. (2)–(5)), please refer to the accompanying Supplementary Methods.

Evaluation of CMIP6 ESMs against to observations

Concerning the four ESMs (Supplementary Table 1) utilized in this study, all of them took into account the influence of ocean acidification on marine chemistry26. Nevertheless, it’s worth noting that only CNRM-ESM2-1’s simulations account for the influence of pH changes on DMS production, as per the straightforward parameterization scheme initially proposed by Six et al.16 (refer to Supplementary Methods). To validate the application of the fitted relationship in refining future DMS changes, we assessed the ability of the four CMIP6 ESMs to capture the relationship between pCO2 and DMS changes depicted in Fig. 1a. Figure 1b displays the pCO2 and oceanic DMS concentrations estimated by these models, corresponding to the same temporal and spatial coordinates as the measurement data in Fig. 1a. Nonetheless, our study found that all four ESMs used encounter challenges when attempting to replicate the observed relationship between pCO2 and DMS concentrations.

CNRM-ESM2-127 incorporates prognostic DMS parameterizations, allowing it to simulate the biological processes associated with the release of DMS precursors into seawater and the subsequent DMS sinks. In our quest to dissect the role of biology in influencing DMS concentration changes, we turn our attention to marine net primary productivity (NPP). Figure 1b illustrates that the trend in NPP closely parallels the trend of DMS in response to pCO2 changes for CNRM-ESM2-1. This implies that the parameterization in CNRM-ESM2-1, which incorporates pH-dependent constraints, might not sufficiently constrain the response of DMS production to pH changes, indicating a need for further refinement to more accurately represent this process. NorESM2-LM18 also employs prognostic DMS parameterizations. However, DMS in NorESM2-LM is directly released into the water and is computed as a function of temperature and simulated detritus export production. Due to this approach, the trend in NPP does not explicitly align with the trend of DMS in response to pCO2 changes for NorESM2-LM. In the case of UKESM1-0-LL28, a simpler empirical parameterization is used to simulate oceanic DMS, which is based on chlorophyll, light, and a nutrient term. Chlorophyll and nutrient are considered the primary indicator of surface ocean primary production. Consequently, the trend in NPP closely mirrors the trend of DMS in response to pCO2 changes for UKESM1-0-LL. In contrast, MIROC-ES2L29 exhibits no clear correlation between DMS and NPP. This may be attributed to the parameterization of Aranami and Tsunogai30 used, as indicated in Bock et al.’s20 analysis.

Prospective trends in DMS concentration and flux

Figure 2a presents the trends in global annual mean sea-surface DMS concentration as estimated by four CMIP6 ESMs (BASE) and the refined projections from this study (PROJ) for the period 2020–2099 under the SSP5-8.5 scenario. The PROJ results demonstrate a consistent decrease in sea-surface DMS concentration across all four ESMs throughout the entire period from 2020 to 2099. Specifically, by the end of the century (2099), the global mean sea-surface DMS concentration in PROJ exhibits a substantial decline of 15.1% (with a range of 7.0–26.3%) compared to the Historical results (average of the historical period spanning from 1960 to 2014). In contrast, the BASE results exhibit diverse trends primarily influenced by variations in the DMS parameterization schemes employed by the four ESMs, leading to disparities in the projected direction of the trend. The global mean sea-surface DMS concentration in BASE shows a marginal increase of 1.2% (with a range of −9.5% to 9.5%) in 2099 relative to the Historical results (Supplementary Fig. 3).

Fig. 2: Comparisons of sea-surface DMS concentrations estimated by four CMIP6 ESMs (BASE) and those projected in this study (PROJ).
figure 2

The time series of global annual mean sea-surface DMS concentrations between 2020 and 2099 under SSP5-8.5 scenario (a). Spatial distributions of changes in the sea-surface DMS concentrations (1° × 1°) among the results of Historical, BASE, and PROJ (b). Historical represent the Averaged sea-surface DMS concentrations from CMIP6 ESMs Historical experiments, 1960–2014.

Full size image

We observed a widespread decline in oceanic DMS concentration across almost all ocean regions (PROJ-Historical), as shown by Fig. 2b. The most substantial reduction of 71.3% occurred in the southern hemisphere. In contrast, due to projected sea-ice changes, the Arctic region exhibited an increase of over 50% in DMS concentrations between BASE and Historical (BASE-Historical), as well as PROJ-Historical results. This could be attributed to the retreat of sea ice, which enhances DMS production by increasing light availability for photosynthesis31. Prior studies within the CMIP6 model evaluation20 have similarly demonstrated an increase in oceanic DMS concentrations with a decrease in the fraction of sea-ice cover in the Arctic region under the SSP5-8.5 scenario. Examining the global annual mean DMS flux, we observed an 8.7% increase in BASE but a 10.9% decrease in PROJ compared to the Historical results (Fig. 3). The changes in fluxes showed a shift toward positive values (−10.9% to 8.7%), unlike the changes in concentration (−15.1% to 1.2%). This disparity between concentration and emission changes can be mainly attributed to the positive relationship between air-sea exchange parameterization and sea-surface temperature (Eq. (S6)), which experienced significant increases in 2099 under the SSP5-8.5 scenario20. On a global scale, our projections indicate a decrease of −16.1% in sea-surface DMS concentration and an 18% decline in flux (PROJ-BASE) by 2099 compared to estimates provided by CMIP6 ESMs. For the spatial distribution of annual mean seawater DMS concentrations, please refer to Supplementary Fig. 4.

Fig. 3: The spatial distributions of changes in the oceanic DMS emissions fluxes (2° × 2.5°) among the results of Historical, BASE, and PROJ results.
figure 3

The sea–air DMS flux is calculated using the Nightingale et al.56 parameterization of gas transfer velocity.

Full size image

Prospective changes in DMS-induced radiative forcing

To explore the DMS radiative forcing for 2099 under the SSP5-8.5 scenario, we performed three different annual simulations (Supplementary Table 2). The REF simulation utilized DMS emissions calculated from averaged sea-surface DMS concentrations (BASE) obtained from four ESMs, the PRD simulation represented DMS emissions based on our projected estimates (PROJ), and the ND simulation had DMS emissions turned off. In 2099, the global mean top-of-atmosphere (TOA) direct radiative forcing (DRF) and cloud-albedo indirect radiative forcing (IRF) induced by DMS decreased 16.4% (from −0.13 to −0.11 W m−2) and 12.1% (from −0.24 to −0.22 W m−2), respectively, when comparing the PRD–ND simulations to the REF –ND simulations (Fig. 4). This decrease corresponds to a reduction of 0.02 W m−2 and 0.03 W m−2 for DMS-derived DRF and cloud-albedo IRF, representing ~9.5% and 11.1% of the total radiative forcings induced by aerosol radiation interactions and cloud interactions in 2100 under the SSP5-8.5 scenario as reported by AR6 (Table AIII.4e in Full Report5). Regions exhibiting higher reductions in atmospheric DMS concentrations (30–40%) from the PRD-REF simulation were primarily located in the high-latitude southern hemisphere, and large decreases in annual mean DRF (20–30%) and IRF (15–60%) were also observed over the high-latitude southern hemisphere.

Fig. 4: The simulated atmospheric DMS concentration and DMS-induced radiative forcing in 2099.
figure 4

The first and second row presents the spatial distributions of annual mean atmospheric DMS concentration and its derived all-sky DRF and cloud-albedo IRF in 2099. The third row presents the percent changes in DMS concentration and the magnitude of the all-sky DRF and cloud-albedo IRF between PRD and REF simulations.

Full size image

The average concentrations of SO42− and particles with diameters larger than 80 nm (N80, a proxy for CCN-sized particles32) decreased from 0.25 μg m−3 to 0.21 μg m−3 and 39.5% to 36.5% (averaging the percent differences of each grid cell) within the global domain in 2099 between RPD and REF simulations (Supplementary Fig. 5), respectively. The spatial distributions of DRF (Fig. 4) were generally consistent with those of SO42− concentrations enhanced by DMS emissions (Supplementary Fig. 5). However, some discrepancies were observed in the distributions of the effects of DMS on IRF and N80. For instance, higher increases in N80 from DMS emissions were observed around the equatorial belt in the Pacific Ocean, which did not directly correlate with stronger negative IRF in those areas. These discrepancies can be attributed to factors such as surface albedo and cloud fractions32,33,34.

Discussion

The substantial decline observed in global anthropogenic emissions over the past decade35 accentuates the growing significance of natural emissions, including those originating from DMS. Consequently, it becomes imperative to accurately quantify these forthcoming emissions, given their potential implications for climate. In our study, we found that the CMIP6 ESMs did not adequately capture the relationship between sea-surface pCO2 and DMS when compared with multi-year and globally distributed observational datasets established in this research. Therefore, further refinement of the future sea-surface DMS concentration estimated by CMIP6 ESMs is necessary. Our findings suggest that if the relationship depicted in Fig. 1a remains applicable under future oceanic conditions, the anticipated rise in anthropogenic CO2 storage within the ocean could have a substantial impact on DMS concentrations, resulting in a reduction in the magnitude of DMS-induced radiative forcing. Therefore, it is crucial to consider the potential climate implications of changes in DMS production resulting from relative biogeochemical drivers when projecting future climate change.

We acknowledge several limitations in our study. The utilization of two ratios (Supplementary Fig. 2a, b) for the calculation of DMS concentration might be seen as rather simplistic. To investigate the pCO2 and DMS ratios in different oceanic regions, we conducted a comparative analysis across the global ocean at latitudinal bands of 90°, 60°, 45°, and 30°, as depicted in Supplementary Fig. 6. The results in the 45°S–0°and 30°S–0° (Supplementary Fig. 6g, l) regions revealed a relatively higher regression slope (approximately −0.8) in these areas when compared to the high-latitude regions in the Southern Hemisphere (e.g., 90°S–45°S (f) or 90°S–60°S (j)), where the slopes ranged from −2.6 to −2.4. This suggests that the negative trend of DMS in response to pCO2 changes in the 45°S–0° and 30°S–0° regions (Supplementary Fig. 6g, l) is comparatively slower than in the high-latitude Southern Hemisphere regions (Supplementary Fig. 6f, j). Therefore, we re-evaluated the global annual mean sea-surface DMS concentration by applying Equations S2-S5, utilizing four ratios derived from the 45° latitudinal band interval (Supplementary Fig. 6f–i), as these ratios exhibit statistically significant trends (P < 0.05). The recalculated annual mean sea-surface DMS concentrations for the period 2020–2099 indicate a 0.02–0.06 μmol m−3 increase compared to the values computed using two ratios (Supplementary Fig. 7a). Notably, the global mean sea-surface DMS concentration in 2099, employing four ratios, reveals a decrease of 13.3% (ranging from 5.0 to 24.1%) relative to Historical results. This suggests a potential slight underestimation in estimates for DMS emissions and resultant radiative forcing, likely attributable to the use of the Southern Hemisphere’s regional average slope (Supplementary Fig. 6a) in the calculations.

However, from Supplementary Fig. 6g, k, l, it can be observed that the range of pCO2 measurements in this region has a relatively short interval. Therefore, using the results from this region may not fully reflect the response of DMS to significant pCO2 changes. Furthermore, for the Northern Hemisphere, the results in the 0°–45°N (h) and 45°N-90°N (h) regions, with slopes ranging from −0.8 to −0.6, are similar to the results in the 0°–90°N (b) region (−0.7). However, in the 0°–30°N (m) and 30°N–60°N (n) regions, the trend of DMS with pCO2 changes is not significant (P > 0.1), indicating that these results are not suitable for regional constraints. Future research demands a richer dataset of observations and an expanded range of parameters influencing the mechanisms behind pCO2 and DMS responses to explore variations at various spatial scales.

This study employed an offline approach to modify the simulated DMS concentration results from ESMs. Incorporating the pCO2-dependent constraint parameterization derived from our research into ESMs and conducting oceanic DMS simulations through online methods could potentially produce more accurate outcomes. In addition, further advancements in laboratory experiment results regarding the pCO2 (pH) and DMS relationship may also guide updates in parameterization schemes.

Methods

Description of Global Observational Datasets and ESMs

The global observational datasets utilized in this study were sourced from The Surface Ocean CO2 Atlas36 and the Global Surface Seawater DMS Database37, respectively. In total, there were 30,568,073 and 88,786 valid measurements for seawater pCO2 (1957–2020) and sea-surface DMS concentrations (1972–2017). We matched these two datasets since the longitude and latitude of pCO2 and DMS measurements are equal in decimal place, and simultaneously the sampling date should be equal in hour scale. For example, both of pCO2 and DMS sampling geographical coordinates and date are 55.1°N, 150.2°E and 1 October, 2010, 17:00, we assumed that the pCO2 and DMS were from the same sampling site at the same time. After removing duplicates, we got a total of 11,617 pairs for pCO2 and DMS (Supplementary Fig. 1), which cover a time span from 1987 to 2015 (28 years). The minimum and maximum values of pCO2 are 110.7 and 695.1 μatm, to eliminate the “noise” from low-frequency variation of measurements below a given space and timescale38, data from each 5μatm interval within 100 to 700 μatm (100–105, 106–110, …, 696–700) were averaged and log-transformed.

In this study, four state-of-the-art CMIP6 ESMs (CNRM-ESM2-127, MIROC-ES2L29, NorESM2-LM17, and UKESM1-0-LL28) were utilized to evaluate the future trends in global sea-surface concentrations. Previous study39 has indicated the capability of these ESMs to simulate the prominent large-scale characteristics of ocean circulation. The analysis focused on monthly sea-surface DMS concentration and the NPP outputs labeled “r1i1p1f1” and “r1i1p1f2” from CMIP6 historical and SSP585 scenario experiments. Notably, MIROC-ES2L and UKESM1-0-LL employed empirical parameterizations using chlorophyll and relevant variables to estimate DMS concentrations, while CNRM-ESM2-1 and NorESM2-LM incorporated prognostic models accounting for marine biota. All model datasets were obtained from Earth System Grid Federation (ESGF) nodes, with further details on key characteristics provided in Supplementary Table 1.

Simulation details

We conducted simulations of aerosol size, number, and mass concentrations using the GEOS-Chem version 13.2.0 coupled with the TwO-Moment Aerosol Sectional (TOMAS) model40. TOMAS was used to model aerosol microphysics processes, which tracks the total aerosol particle number and the mass of each aerosol species (sulfate, hydrophilic and hydrophobic organic carbon, externally and internally mixed elemental carbon, mineral dust, sea salt, and aerosol water) across 15 logarithmically size bins (from 3 nm to 10 μm)41,42. GEOS-Chem-TOMAS has detailed hydrocarbon—nitrogen oxide—ozone-volatile organic compounds - bromine oxides tropospheric chemistry43 and a wet and dry deposition scheme for aerosols and gas species44,45,46. Aerosol species were fully coupled to gas-phase chemistry47,48,49, with the ISORROPIA II scheme to calculate the thermodynamic equilibrium between inorganic aerosols and their gas-phase precursors50. To estimate the top-of-atmosphere (TOA) all-sky direct radiative forcing (DRF) and cloud-albedo indirect radiative forcing (IRF), we utilized the Rapid Radiative Transfer Model for Global Climate Models (RRTMG)51. For the DRF, we calculated aerosol optical properties using GEOS-Chem-TOMAS output aerosol mass concentrations. These aerosol optical properties and monthly mean albedo and cloud fraction as input to drive the offline RRTMG model to calculate DRF with an external mixing assumption of the aerosol species. For the cloud-albedo IRF, we calculated cloud droplet number concentration following the activation parameterization from ref. 52. Then we used RRTMG to calculate the changes of TOA radiative flux from the changes affect cloud drop radii based on the perturbation method used in previous studies53,54. More details about the implementation of RRTMG in GEOS-Chem-TOMAS are described in ref. 53.

To investigate the potential changes in atmospheric DMS concentrations and its oxidation products in the future, we employed the meteorology, emissions, and boundary conditions from the Global Change and Air Pollution (GCAP 2.0) model55 to drive GEOS-Chem-TOMAS. The GCAP 2.0 framework allows GEOS-Chem to utilize meteorology data archived from CMIP6 ESMs. Our simulations were conducted at a horizontal grid resolution of 2° × 2.5° and 40 vertical layers. The initial conditions were obtained by performing spin-up simulations for 19 years (2040–2049, 2090–2098) to achieve a chemistry-climate steady state for the year 2099.