Characteristics and trends of surface ocean dimethylsulfide (DMS) concentrations and fluxes into the atmosphere of four Earth system models (ESMs:
CNRM-ESM2-1, MIROC-ES2L, NorESM2-LM, and UKESM1-0-LL) are analysed over the recent past (1980–2009) and into the future, using Coupled Model
Intercomparison Project 6 (CMIP6) simulations. The DMS concentrations in historical simulations systematically underestimate the most widely used
observed climatology but compare more favourably against two recent observation-based datasets. The models better reproduce observations in mid to
high latitudes, as well as in polar and westerlies marine biomes. The resulting multi-model estimate of contemporary global ocean DMS emissions is
16–24
Despite several decades of investigations, the quantification of interactions between aerosols and climate remains poorly constrained and understood
Changes in climate variables, for instance surface wind, sea surface temperature (SST), or downwelling irradiance, can affect both the production of
DMS and its surface concentration, as well as its transfer rate from the ocean to the atmosphere, potentially driving a DMS–climate feedback
Recent observations and mesocosm experiments have improved our understanding of how changes in microalgae dominance and DMS production in response to
climate warming
In this paper, we use the most up-to-date generation of observational data products and long-term measurements to assess estimates of the surface
ocean DMS concentrations and emissions to the atmosphere from the latest generation of ESMs, using their contributions to the sixth phase of the
Coupled Model Intercomparison Project
In Sect.
Key characteristics of the ocean and marine biogeochemical components of the ESMs. Column 2: horizontal grid points (tripolar grids) and number of vertical levels. Column 4 in brackets: prognostic variables involved in the DMS parameterisations.
The present work draws on the results of four state-of-the-art ESMs that have contributed to CMIP6 (CNRM-ESM2-1, MIROC-ES2L, NorESM2-LM, and
UKESM1-0-LL), whose key characteristics are provided in Table
Table
Nonetheless, as documented in the reference papers of CNRM-ESM2-1, MIROC-ES2L, NorESM2-LM, and UKESM1-0-LL, these ESMs are able to simulate the main
large-scale features of the ocean circulation. Recent work has also suggested that these models have improved their performance in simulating the
ocean mixed-layer depth (MLD), which is an important driver for marine biogeochemistry and marine DMS emissions
As shown in Table
In CNRM-ESM2-1, DMS concentration is computed by the biogeochemical model PISCES
NorESM2-LM includes a fully interactive description of the DMS cycle across the ocean–atmosphere interface following
Compared to CNRM-ESM2-1 and NorESM2-LM, MIROC-ES2L and UKESM1-0-LL use a much simpler approach to simulate the marine DMS cycle. Indeed, DMS concentration is diagnosed using empirical parameterisations that relate the DMS concentration to other marine biogeochemical or ocean hydrodynamical variables such as chlorophyll (hereafter Chl) and MLD. Despite their relative simplicity, the two parameterisations remain quite different.
In MIROC-ES2L, the seawater concentration of DMS is computed according to the parameterisation of
A low
In UKESM1-0-LL, the seawater concentration of DMS is computed within the ocean biogeochemistry model MEDUSA
In the original parameterisation, the fitted parameter values were
In this paper, we use monthly outputs of the CMIP6 historical (1850–2014) and ssp585 scenario (2015–2100) experiments. All datasets were downloaded
from Earth System Grid Federation (ESGF) nodes. The number of realisations of each model for both experiments is reported in Table
NorESM2-LM has been run with two different grid resolutions of the atmospheric model: the version labelled LM has a low atmosphere (250
UKESM1-0-LL realisations of the historical experiment include two different forcings variants, f3 (runs 5, 6, and 7) and f2 (all other available
runs). The difference between both forcings is related to the stratospheric sulfate aerosol optical depths (AODs) that influence stratospheric sulfur chemistry. In f3 forcing,
the AODs were accidentally kept at 1850 values
Number of available members used in the ensemble means and (in brackets) reference for the dataset, for the historical and ssp585 CMIP6 simulations.
Our analysis focusses on surface ocean DMS concentration and marine emission of DMS to the atmosphere (variables dmsos and fgdms in CMIP6). Where
relevant, we use other variables such as the ocean surface Chl concentration (chlos), the vertically integrated marine primary production (intpp), the
10 m wind speed (sfcWind), the sea surface temperature (tos), or the sea-ice cover (siconc) for additional analysis. When compared to
observations or between each other, model outputs are interpolated on a regular Mercator grid of 1
In this study, we compare model outputs to three climatologies of surface ocean DMS concentration.
First, the widely used climatology of
Despite the large number of measurements included in this DMS climatology, the spatial and temporal data coverage is limited (Figs. 1 and S1 of
Another methodology using an artificial neural network (ANN) has very recently been developed by
Another strategy to produce a reliable climatology of sea surface DMS concentration has been proposed recently by
The flux of DMS to the atmosphere has been assessed by several authors, but the resulting datasets are not readily available. However, a recent study
from
Mean (1980–2009) surface ocean DMS concentration (nM) for the CMIP6 historical experiments of the four models: CNRM-ESM2-1, MIROC-ES2L, NorESM2-LM, UKESM1-0-LL, and the MMM. Annual means of L11, G18, and W20 are also plotted. Minimum, area-weighted median, and maximum from each model or data product are also displayed. Hatching on the model plots shows locations where the DMS concentration is outside the range covered by L11, G18, and W20 (see text for details).
The annual mean DMS concentration over the period 1980–2009 is plotted in Fig.
Area-weighted statistics: median, mean, and spatial standard deviation of annual DMS surface ocean concentration (nM) shown in Fig.
Overall, the striking observation that can be made is the lack of agreement between models, as well as between models and observational
products. Table
Spatial correlation coefficients between L11, G18, and W20 and the models (first three rows) and between the individual models (rows 4–7), derived from the data displayed in Fig.
Individual model patterns shown in Fig.
As expected, the output of MIROC-ES2L resembles those of the parameterisations of
NorESM2-LM results can be compared to those obtained with the rather similar ocean biogeochemistry model HAMOCC, within the MPI-ESM-LR model, as
presented by
UKESM1-0-LL features specific patterns that are distinct from the other three models, especially regarding the uniform low concentration over vast
areas. This reflects the threshold set to 1
The general findings outlined above are strengthened by the analysis of the Pearson spatial pattern correlation, which is presented in
Table
Mean annual differences of surface ocean DMS concentration (nM) between models and the climatologies of L11, G18, and W20 (model minus climatology). Model data are as in Fig.
The cross-correlation between models, presented in the second part of Table
The last row in Table
Monthly mean (1980–2009) zonal surface ocean DMS concentration (nM) for the same models and observations as in Fig.
The difference in annual mean concentration between each model and the three climatologies is shown in Fig.
Monthly mean statistics of surface ocean DMS concentration.
The monthly mean (1980–2009) zonal DMS concentration is shown in Fig.
The zonal mean seasonal cycle of MIROC-ES2L is very similar to that of both
NorESM2-LM shows a noticeable double maximum in both hemispheres, around 40
UKESM1-0-LL shows a northern maximum starting earlier than the other models, in agreement with L11. However, as pointed out by
We further extended the pattern correlation analysis presented above for annual mean to monthly data. Figure
Pattern correlations have low values (lower than 0.2) in February–March and again in September to November, which can be related to the smoother features and inter-hemisphere DMS gradients during these months. Conversely, during the summer months in both hemispheres, even if the models do not reach such elevated DMS values as in L11, the stronger concentration gradients contribute to a higher correlation. When compared to G18 and W20 (Fig. S4), the pattern correlations show smoother seasonal variations.
Overall, the MMM has the lowest RMSE in nearly all months, with significantly lower values than each model when compared to G18 and W20 (Fig. S4). The MMM and NorESM2-LM also have the best correlation coefficients with L11 in almost all months, with coefficients higher than 0.4 during 7 months of the year. They also have the best pattern correlations when compared to G18 and W20.
Further to the spatial analysis presented in the previous section, we now focus on the analysis of seasonal variations. For that purpose, we used the
same ocean partitioning into 54 biogeographical provinces
Monthly 1980–2009 area-weighted surface ocean DMS concentration (nM) in the 54 oceanic regions defined by
In Fig.
This figure shows the skills of each model in reproducing the seasonal cycle for each region. The seasonal cycle in the northern Atlantic
(regions 2–6 and 11) is relatively well reproduced regarding the timing and duration of seasonal maximum, but the amplitude is generally lower in the
models. The agreement is also satisfying in the northern Pacific (regions 30, 32, 33, 44) and in the regions located south of
Correlation coefficients of the linear regressions between the monthly time series of the models and L11. Time series and regions, with their numbering and colour code, are those of Fig.
In Table
Among the four models, none appear to have significantly better skills than the other regarding the seasonal cycle. The MMM presents slightly more uniform skills throughout the 54 provinces, with a correlation generally within the range of individual model correlations (35 out of 54 regions) or better than all models (17 out of 54 regions). There are only two exceptions (regions 8 and 17) where the correlation of MMM is worse than all individual models.
Several hypotheses can be proposed to explain the poorer correlation of the modelled seasonal cycles with the observations at low latitudes. First, it
was already shown that models present some features at low latitudes that do not agree with climatologies. For instance, the strong enhancement of DMS
concentration in the eastern equatorial Pacific was found in all models except CNRM-ESM2-1. While this model feature leaves an imprint on the annual mean
(see Fig.
The choice of the gas transfer parameterisation has an important impact on the calculated marine DMS emission; we thus briefly recall here their main
characteristics. To compute the transfer of DMS from the ocean into the atmosphere, UKESM1-0-LL uses the parameterisation of
Annual mean DMS flux (
Mean annual DMS flux (
DMS flux into the atmosphere is presented in Fig.
The flux of DMS features spatial patterns and seasonal cycles that stem from both the surface ocean concentration and the wind speed. To help understand how these main drivers impact the resulting flux, the maps of annual wind speed and the zonal monthly wind speed are shown in the Supplement (Fig. S5).
Overall, the maps of annual mean flux show a large spatial variation, which mirror in part the patterns of annual mean concentration (see
Fig.
The variety of patterns shown in Fig.
Monthly mean zonal DMS flux (
Figure
Global mean DMS emissions (
Time series of mean annual global area-weighted surface ocean DMS concentration (nM,
To conclude this section, Table
Because of the weakest dependence on wind speed in the parameterisation of
The case of CNRM-ESM2-1 and NorESM2-LM is interesting as these models have very close global mean DMS concentration, yet NorESM2-LM computes a total
annual DMS flux that is about 20 % lower than CNRM-ESM2-1 (see also Fig.
To conclude, the global DMS emission computed by the four CMIP6 models are well within the recent literature values, with the best estimate of
Figure
Comparison of the mean and relative difference of DMS concentration (conc. in nM) and flux (
Figure
Time series of mean annual area-weighted surface ocean DMS concentration (nM, left panels) and annual regional DMS fluxes (
To gain further insight into the respective trends of DMS concentration and flux, Fig.
A comparison of Figs.
Mean trends over 1980–2009: left column is surface ocean DMS concentration (
Same as Fig.
Figures
Ultimately, Figs.
Metrics of the scatter plots of Fig.
Several previous studies have investigated the evolution of DMS concentration and flux in the future. Although these studies are not necessarily
comparable to ours with respect to scenarios, model type, and methods, we summarise the results of these studies regarding global DMS trends. An early
study by
To conclude, there is no agreement in the literature with regards to the sign and amplitude of the trends of DMS concentration and flux in the future
Scatter plot of change of DMS concentration as compared to the change in NPP as simulated by CMIP6 models over the historical 1980–2009 period (left column) and ssp585 2071–2100 period (right column). For both variables, the reference (
The aim of this section is to answer the question of whether other variables could give further insight into the most likely trend of DMS concentration, especially in low-latitude regions which determine the global trend as found in the previous section.
The biological control of the change in DMS concentration has been highlighted in previous modelling studies
Figure
Although the limited current knowledge about the NPP–DMSP–DMS relationships hampers our ability to constrain this emergent property, several lines of
evidence tend to suggest that there is a positive correlation between NPP and DMS concentration. Firstly, noting that some studies observed no
correlation between DMS and Chl
Such a relationship between changes in NPP and changes in DMS has consequences for future projections, because it suggests that the overall model
response in DMS concentration and emission will mirror changes in NPP. A recent study by
In polar regions, the role of DMS concentration in governing the DMS emission is superseded by the dynamics of sea-ice cover, in line with the common model assumption of a linear relation between the DMS flux and the ice-free area fraction. To make it clear to the reader, we emphasise here that in all CMIP6 models, the DMS flux variable represents the actual flux over the entire grid cell and thus already accounts for the reduction due to the sea ice, if present.
From top to bottom in each column are time series (1950–2014) of May–August values of integrated DMS emission (
Figure
Several of the time series presented in Fig.
As compared to the DMS concentration values found by G19 between 1998 and 2016 (in the 2.7–3.0
Scatter plots with mean annual May to August values of DMS emissions (Gg S) vs. ice-free extent (%) over the same regions as those of
Metrics of the scatter plots of Fig.
Comparing the time series of DMS concentration and flux together demonstrates again the tight relationship between both variables, especially when considering only the means over ice-free pixels (dashed lines): the specific role of sea ice is then negligible, and the similar behaviours are
clearly visible. As compared to G19, who found a mean DMS flux in the
3–4
All together, these observations made for the sea-ice extent, as well as the DMS concentration and flux, explain well the resulting DMS-integrated emissions
from May to August (top row in Fig.
Scatter plots of DMS emissions vs. ice-free extent over the 1950–2014 period are shown in Fig.
We further extended this analysis to the ssp585 scenario, using the 65-year-long period from 2036 to 2100 so that both analyses rely on the same
time windows. Scatter plots of DMS emissions vs. ice-free extent (see right column of Fig.
In this study, we analyse surface ocean DMS concentration and flux into the atmosphere from four CMIP6 ESMs (CNRM-ESM2-1, MIROC-ES2L, NorESM2-LM, and
UKESM1-0-LL) over the historical and ssp585 simulations. The parameterisations of DMS in these ESMs have various degrees of complexity, and while they
may have already been evaluated, either in a previous
Our analysis of contemporary (1980–2009) climatologies of simulated surface DMS concentration shows that, overall, agreement is poor between models
and also between models and reference datasets, such as the L11 dataset. This is consistent with previous work
Analysis of the annual cycles in each of the 54 biogeographical provinces defined by
The multi-model ensemble mean (MMM) shows good skill in reproducing spatial patterns and seasonal variability and compares generally better with observational climatologies than individual models.
The comparison of marine DMS emissions confirms the importance of the air–sea flux parameterisation on the resulting flux. The estimates of DMS
emissions relying on observed surface ocean DMS concentrations and state-of-the-art air–sea flux parameterisations range between
16–28
The comparison of trends of DMS fluxes and of DMS concentrations over the whole simulation period (historical + scenario; 1850–2100) reveals that the
current generation of CMIP6 ESMs disagree on the sign of these trends. Two models (CNRM-ESM2-1 and MIROC-ES2L) simulate an increase in ocean DMS
concentrations and emissions, whereas two other models (NorESM2-LM and UKESM1-0-LL) predict a moderate decrease. As a consequence, our work shows that
the lead-order uncertainty in the future evolution of marine DMS emissions has not been reduced in the current generation of models compared to that
of previous modelling experiments
Our analyses using CMIP6 ESMs confirm the conclusions of
On the contrary, there is no consensus on how the current generation of models simulate the long-term trends in DMS concentrations and emissions in
low-latitude biomes. Further investigating the relationship between DMS concentration and biological productivity reveals that three models
(CNRM-ESM2-1, NorESM2-LM, and UKESM1-0-LL) predict a positive correlation between the trend in ocean surface DMS concentrations and the trend in marine
primary production, while the fourth model (MIROC-ES2L) displays no strong correlation between these variables. This raises questions regarding the
ability of empirical parameterisations of DMS concentration to predict future evolution, since they have been calibrated in present conditions.
Despite the qualitative agreement of the three models regarding the NPP–DMS relationship, the predictive ability is limited given the large
uncertainties in the future evolution of marine primary production. The modelling challenge here is particularly vast as
Although none of the marine biogeochemical models studied here are currently intended to represent specific taxa of marine phytoplankton, it is
interesting to connect the most likely behaviour of those taxa in response to climate change. For instance,
The four studied models use tripolar grids in the oceans, with none of the four being common to each other. Conversely, the climatologies are provided on
regular 1
The multi-model ensemble mean (MMM) is computed in pixels where at least three models have valid data. This criterion has been chosen so that no pixel in MMM is based on a single model (for instance, the Caspian Sea is only described in UKESM1-0-LL) but to retain areas which would be described in all but one model (the Red Sea and Persian Gulf are not described in the ocean model of MIROC-ES2L).
Readers may note differences in metrics provided in this article and metrics provided in other literature (e.g. mean and median of the L11
climatology displayed in Table
In the study of
A special case arose for NorESM2-LM, for which the ocean grid has one more row (
In Fig.
As presented in Sect.
We are aware of more recent work about ocean partitioning into biomes that provides both updated static partitioning and dynamic partitioning
Datasets from CMIP6 simulations are available from every ESGF node, such as
The climatology of
The climatology of
The climatology of
The scripts used to process and plot the data are available upon request to the author. All plots presented in this paper have been produced with
The supplement related to this article is available online at:
JB, MM, PN, and RS designed the study; JB performed research and analysed data; JB, MM, and RS wrote the paper with significant inputs from all co-authors.
The authors declare that they have no conflict of interest.
We gratefully acknowledge Martí Galí, Nadja Steiner, and an anonymous reviewer for their comments that helped to improve and clarify this paper. Josué Bock thanks Martí Galí for his help in using his climatology of DMS and for providing the map of Arctic regions. Josué Bock thanks Wei-Lei Wang for his help in using his climatology of DMS. Josué Bock thanks Yanchun He and Mats Bentsen for their advise in using NorESM2-LM datasets. The MIROC-team acknowledge JAMSTEC for use of the Earth Simulator supercomputer. Supercomputing time was provided by the Météo-France/DSI supercomputing center.
This work was supported by the European Union's Horizon 2020 research and innovation program with the CRESCENDO project under the grant agreement no. 641816, the TRIATLAS project under the grant agreement no. 817578. Jerry Tjiputra and Jörg Schwinger acknowledge Research Council of Norway funded projects KeyClim (295046) and Columbia (275268). Simulations of MIROC-ES2L are supported by the TOUGOU project “Integrated Research Program for Advancing Climate Models” (grant number: JPMXD0717935715) of the Ministry of Education, Culture, Sports, Science, and Technology of Japan (MEXT). JPM was supported by the Met Office Hadley Centre Climate Programme funded by BEIS and Defra (grant no. GA01101). RS acknowledges the European Union’s Horizon 2020 research and innovation programme under grant agreement no. 101003536 (ESM2025 – Earth System Models for the Future).
This paper was edited by Alexey V. Eliseev and reviewed by Martí Galí, Nadja Steiner, and one anonymous referee.