Introduction
Ocean-emitted gases and particles control the number, size distribution and
composition of aerosols in remote oceanic areas (Brooks and Thornton, 2018).
These aerosols scatter sunlight and can act as cloud condensation nuclei that
alter the radiative properties of clouds, both microscopic (cloud droplet
number concentration and effective radius) and macroscopic (cloud abundance,
albedo and lifetime). Interactions between natural aerosols and clouds are
a major source of uncertainty in climate projections, confounding the
calculation of natural and anthropogenic radiative forcing and the
attribution of anthropogenic climate change (Carslaw et al., 2013).
Therefore, there is an urgent need to better understand and model the oceanic
sources of aerosols, and to better resolve their variations at relevant
spatial and temporal scales, from weekly to seasonal to interannual.
The gas dimethylsulfide (DMS) is produced by marine microbial food webs in
the sunlit layer of the ocean. With its emission currently estimated at
28 TgSyr-1, it contributes about 70 % of natural sulfur
emissions to the global atmosphere and a major portion of the marine emission
of organic volatiles (Carpenter et al., 2012; Schlesinger and Bernhardt,
2013; Simó, 2011). The cloud-seeding activity of DMS and its potential
role in climate regulation were first postulated 3 decades ago (Charlson
et al., 1987; Shaw, 1983). The so-called CLAW hypothesis (Charlson et al.,
1987) proposed that a negative feedback could operate between marine
phytoplankton, DMS emission and cloud albedo, potentially regulating the
Earth's climate. Posterior research showed that the mechanisms behind the
potential loop are far more complex than initially envisaged. This, and the
estimated low sensitivity of each step of the feedback to changes in its
forcing factors, led Quinn and Bates (2011) to refute the CLAW hypothesis.
Nevertheless, atmospheric studies powered by new analytical techniques
(Kulmala et al., 2014) and modeling have shown instances where marine DMS
controls ultrafine aerosol particle formation in the Arctic (Leaitch et al.,
2013; Park et al., 2017), temperate North Atlantic (Sanchez et al., 2018),
Antarctica (Yu and Luo, 2010) and the tropical South Pacific atmospheres
(Modini et al., 2009). Moreover, Quinn et al. (2017) recently reported that
DMS-derived aerosols dominate cloud condensation nuclei populations over most
of the global ocean. Hence, the influence of DMS on marine stratiform cloud
albedo remains in the spotlight (Brooks and Thornton, 2018), and the
occurrence of a “seasonal CLAW” in remote marine atmospheres is becoming
increasingly conceivable (Levasseur, 2013; Vallina and Simó, 2007a).
DMS is produced by marine microbial food webs through a complex network of
biological interactions and chemical processes (Simó, 2004). Its primary
source is the enzymatic cleavage of dimethylsulfoniopropionate (DMSP),
a multifunctional osmolyte that accumulates at high (hundred
millimolar) intracellular
concentrations in some phytoplankton, especially haptophytes, dinoflagellates
and some picoeukaryotes (Stefels et al., 2007). DMSP cleavage is catalyzed by
a wide diversity of enzymes, called DMSP lyases, produced by some
phytoplankton (Alcolombri et al., 2015) and bacteria (Curson et al., 2011).
Breakage of phytoplankton cells through zooplankton grazing, viral attack and
autolysis releases DMSP to the algal boundary layer and the dissolved phase
and enhances DMS production (Simó, 2004; Stefels et al., 2007). Another
process that contributes to DMS production is the diffusive release of DMS
from phytoplankton cells, which proceeds almost instantaneously after
intracellular DMSP cleavage by DMSP lyases or by photochemically produced
radicals (Lavoie et al., 2015; Mopper et al., 2015). Once in seawater, DMS is
removed by biotic and abiotic processes. DMS budgets in the upper mixed layer
(UML) indicate that, on average, about 90 % of dissolved DMS is consumed
by bacterial oxidation and UV-driven photolysis, and only 10 % is emitted
to the atmosphere through turbulent diffusion (Galí and Simó, 2015).
Seawater DMS concentration controls the emission flux because the oceanic UML
is largely supersaturated with respect to the atmosphere. DMS concentration
in the UML is regulated by a subtle dynamic equilibrium between production
and consumption processes with a typical timescale of less than
4 days (Galí and Simó, 2015). Over the seasonal cycle, DMS
concentration varies mainly in response to the phenology and ecological
succession of microbial species and their interplay with physical forcing
factors, particularly solar exposure and nutrient supply, which are in turn
regulated by vertical mixing (Galí and Simó, 2015; Lizotte et al.,
2012; Simó and Pedrós-Alió, 1999). For instance, diatom blooms,
typical of nutrient replete conditions at high latitudes, are characterized
by low DMSP concentration per unit biomass and low DMSP-to-DMS conversion
yield (Lizotte et al., 2012). The opposite is true for microbial communities
typical of stratified, nutrient-depleted and highly irradiated surface
waters, both at low and high latitudes (Galí and Simó, 2010; Lizotte
et al., 2012). Under these conditions, two main factors act synergistically
to increase DMS concentration (Galí and Simó, 2015; Vallina et al.,
2008): the higher contribution of DMSP-rich species to total phytoplankton
biomass (Galí et al., 2015; Stefels et al., 2007); and the higher
DMSP-to-DMS conversion yield at the microbial community level, possibly
caused by the effects of nutrient and irradiance stress (Galí et al.,
2013; Stefels, 2000; Sunda et al., 2002, 2007; Vallina et al., 2008). As
a result, similar DMS concentrations may occur in waters that differ by 1 or
2 orders of magnitude in phytoplankton biomass (Lizotte et al., 2012), and
DMS tends to peak in summer across polar to tropical latitudes, lagging the
annual chlorophyll peak by some months in the subtropical gyres. The mismatch
between phytoplankton biomass, DMSP and DMS, termed the DMS summer paradox
(Simó and Pedrós-Alió, 1999), is an essential feature that
biogeochemical models strive to reproduce with mixed success (Le Clainche
et al., 2010).
With nearly 50 000 DMS measurements taken between 1972 and 2010, the global
sea-surface DMS database (https://saga.pmel.noaa.gov/dms/, last access:
13 April 2017) is a valuable resource for model development and validation.
Gridded monthly climatologies (Kettle et al., 1999; Lana et al., 2011)
calculated from this dataset are the standard DMS product used as input to
atmospheric chemistry and climate models, hence emphasizing the seasonal
climatological view (Mahajan et al., 2015; McCoy et al., 2015). At the other
end, the climatic role of DMS is often evaluated through extreme sensitivity
tests that examine the response of Earth system models to order-of-magnitude
perturbations of DMS emission (e.g., Grandey and Wang, 2015). In comparison,
contemporaneous decadal-scale DMS variability has received less attention.
This gap can be filled using empirical remote sensing algorithms, a handful
of which have been developed since the early 2000s (Tesdal et al., 2016; see
also the pioneering works of Jodwalis and Benner, 1995, and Thompson
et al., 1990). Interestingly, large discrepancies exist among global DMS
fields estimated with interpolated climatologies, empirical algorithms or
prognostic biogeochemical models (Tesdal et al., 2016). Although it is
tempting to attribute these discrepancies to the poor skill of the models,
they may also arise from issues in the calculation of the climatology.
Here we present DMSSAT, a new remote sensing algorithm for
DMS that proceeds in two steps: (i) estimation of the concentration of the
phytoplanktonic DMS precursor, total dimethylsulfoniopropionate (DMSPt), from
remotely sensed chlorophyll and light penetration, and from climatological
mixed layer depth (MLD); (ii) estimation of DMS concentration from DMSPt and
solar irradiance. This two-step empirical algorithm reflects, with
a simplified formulation, the mechanistic understanding of oceanic sulfur
cycling described in the previous paragraphs. The DMSPt sub-algorithm was
presented by Galí et al. (2015) and is briefly described in Appendix A.
Thus, here we focus on the second step, based on the nonlinear relationship
between DMS, DMSPt and photosynthetically available radiation (PAR) at the
sea surface. We implement our algorithm to produce a global DMS climatology,
which we compare to the latest version of the interpolated DMS climatology
(Lana et al., 2011; L11 hereafter) and to climatologies derived from other
remote sensing algorithms that follow similar rationales (Simó and Dachs,
2002; Vallina and Simó, 2007b). Finally, we implement our algorithm using
14 years of MODIS-Aqua satellite data in the subtropical and the subpolar
North Atlantic and in the northeast Pacific to illustrate and understand
interannual DMS variability.
Methods
Datasets used for algorithm development and validation
In situ concentrations of DMS and DMSPt (nM) and chlorophyll a (Chl,
mgm-3), accompanied by ancillary data (bottom depth, temperature,
salinity, wind speed), were downloaded from the global sea-surface DMS
database. The latter was complemented with additional in situ datasets
recently obtained by the authors' teams. After quality control, the database
had 41 304, 3700 and 9182 measurements for in situ DMS, DMSPt and Chl,
respectively, with 3637 DMS–DMSPt and 8141 DMS–Chl pairs.The in situ database
was extended with geophysical and biogeochemical parameters, including
satellite retrievals collocated in time and space (“matchups”) and gridded
climatological datasets, following Galí et al. (2015; see below).
Detailed information regarding data sources, quality control and processing
can be found in the Supplement and in Tables S1–S3.
We performed satellite matchups using SeaWiFS (1997–2010) and MODIS-Aqua
(2003–2012) retrievals of remotely sensed Chl (mgm-3), vertical
attenuation coefficient at 490 nm (Kd490, m-1), particulate
inorganic carbon (PIC, molm-3) and daily photosynthetically
available radiation at the sea surface (PAR,
molphotonsm-2d-1). To maximize the amount of available
matchups, and after verifying the consistency between the two sensors, we
produced merged variables by averaging SeaWiFS and MODIS-Aqua matchups. We
employed a hierarchical search procedure whereby the matchup criteria were
progressively relaxed from 1 to 8 days and from single-pixel to
5×5 pixel bins (Sect. S2 in the Supplement). These merged variables are hereafter
designated with the SAT subscript (e.g., ChlSAT). Daily sea-surface temperature (SSTSAT, ∘C) from the AVHRR
sensors was also matched to the database.
Relationship between DMS, DMSPt and PAR across oceanic biomes
and data availability. (a) DMS
vs. DMSPt. (b) DMS / DMSPt ratio vs. mean
daily irradiance (PAR) at the sea surface. (c) Longhurst
biogeochemical provinces and biomes. (d) In situ DMS data
counts in 5∘×5∘ latitude–longitude bins;
stippling indicates bins with measurements available for 3 or more
months. In (a, b) small grey dots represent
individual data points and large colored dots represent the median
in a given Longhurst biogeochemical province and month and the
corresponding interquartile ranges. Province–month medians are
colored by biome following the map in (c), which also shows
the amount of DMS–DMSPt–PAR measurements available in each biome. In
(a–c), the R2 and n (data counts) outside and
inside parentheses correspond to non-binned data and province–month
(MLongh) binned data, respectively. Regression lines in (a, b), calculated with MLongh binned data, are only
illustrative.
The database was further extended with monthly climatological data: daily PAR
from SeaWiFS (1997–2010 average); mixed layer depth (MLD, in meters) from
the monthly MIMOC climatology (Schmidtko et al., 2013); bottom depth from the
General Bathymetric Chart of the Oceans (GEBCO08); and sea-surface nitrate
and phosphate concentrations (µM) from the World Ocean Atlas 2009
(WOA09). Nutricline depths were calculated from WOA09 vertical profiles as
the depth where nitrate and phosphate first exceeded 1 and
0.4 µM, respectively. Nutricline depth estimations were robust to
changes of ±50 % in these concentration thresholds.
The mean daily PAR in the upper mixed layer (PARMLD) was
calculated as
PARMLD=[PARSAT/(Kd490SATMLD)][1-exp(Kd490SATMLD)].
When satellite matchups were not available (before September 1997), we used
climatological PAR from SeaWiFS (1997–2010 average) in order to increase the
temporal coverage of the PARSAT and
PARMLD variables. Statistical analyses done with
climatological or matchup PARSAT gave very similar
results. This procedure was not followed with other variables (Chl, PIC,
Kd490) that show wider interannual variations.
Statistical analyses and data binning schemes
All statistical analyses were conducted using (i) non-binned data; (ii) data
binned by month and 5∘×5∘ latitude–longitude bins
(M5×5); and (iii) data binned by month and the 56 Longhurst
biogeochemical provinces (designated MLongh; Longhurst, 2010). Data binning
eliminates low-frequency variation (“noise”) below a given space or
timescale. MLongh binned data were further aggregated into six biomes: two
Polar biomes (Arctic and Antarctic), two mid-latitude Westerlies biomes
(Northern and Southern hemispheres), one Trades biome (tropical latitudes)
and one global Coastal biome (Fig. 1c). Variables with a right-skewed
approximate lognormal distribution entered statistical analyses after
log10 transformation: DMS, DMSPt, DMS / DMSPt ratio, Chl, nitrate
and phosphate concentrations. To further account for non-normality, we
conducted statistical explorations using both bin means and bin medians.
To develop the DMS algorithm we analyzed the relationship between DMS, the
DMS / DMSPt ratio and environmental variables (listed in
Table 1). After an exploratory analysis based on the calculation of Pearson
correlation coefficients (Table 1), we built several regression models where
DMS was estimated as a function of in situ DMSPt concentration and additional
variables (Tables 2 and S4). We added one variable at a time in order of
decreasing data availability, and significant terms were selected using
stepwise regression with entrance and removal p values set at 0.001 and
0.005, respectively. The logic for adding one variable at a time, rather than
building a single initial model with all the predictor variables, is that the
size (N) of the data subset used for model fitting decreases rapidly when
variables with sparse coverage are combined. Each set of initial predictors
was tested across the three degrees of data binning described above and three
degrees of model complexity: linear without interactions, linear with
interactions and quadratic with interactions. This 3×3 nested
structure provided a stringent test for the robustness of a given regression
model. Improvements in model performance were assessed based on the increase
in adjusted r squared, Radj2, and the decrease in
root-mean-square error (RMSE) and the Akaike information criterion (AIC).
Correlation analysis (Pearson's r) for different data binning
levels. Pearson's r with p value <0.01 are not shown;
values in italic are 0.0001<p<0.01; n/a: not applicable. “Ratio”
refers to log10(DMS / DMSPt). DMS, DMSPt, Chl, MLD,
[NO3] and [PO4] were log10 transformed;
Pearson's r calculated on log10-transformed variables were
higher than those calculated on the same non-transformed variables,
and similar in magnitude to Spearman's rank correlations. Bottom and
nutricline depth values are shown as positive values (deeper is bigger). See the text
for other acronyms.
Non-binned
Monthly 5∘×5∘
Monthly Longhurst
data
binning (M5×5)
binning (MLongh)
DMS
N
Ratio
N
DMS
N
Ratio
N
DMS
N
Ratio
N
In situ data
DMSPt
0.65
3637
n/a
1442
0.58
308
n/a
157
0.46
122
n/a
87
Chl
0.45
to
-0.33
to
0.37
to
-0.28
to
0.34
to
-0.45
to
SST
-0.02
41 304
0.29
3637
1562
0.45
308
322
0.56
119
Salinity
-0.12
0.27
0.32
Wind speed
-0.12
-0.13
-0.12
-0.20
-0.27
Bottom depth
-0.19
0.10
-0.12
0.16
0.26
Day length
0.42
0.06
0.43
0.49
Climatological data
MLD
-0.37
35 505
-0.13
3433
-0.32
1474
298
-0.51
312
-0.24
119
[NO3]
0.06
to
-0.19
to
0.16
to
-0.31
to
to
-0.45
[PO4]
0.05
39 478
-0.15
3637
0.13
1535
-0.34
308
318
-0.32
N-cline
-0.14
0.30
-0.22
0.44
0.55
P-cline
-0.12
0.24
-0.14
0.45
0.37
Satellite match-up data
PARSAT
0.32
16 411
0.35
1123
0.30
498
0.46
124
0.52
171
0.67
86
PARMLD
0.12
to
0.37
to
0.15
to
0.49
to
0.36
to
0.66
ChlSAT
0.37
41 088
-0.42
3620
0.22
1539
-0.34
307
0.28
321
-0.39
119
PICSAT
0.24
-0.27
0.29
-0.30
0.33
Regression models were further optimized for global and regional domains
using the bootstrap method followed by nonlinear optimization as described in
Sect. S4. Selected models were then validated using an independent data
subset composed of in situ DMS measurements and their satellite matchups
(described in Sect. 3.1.3) and evaluated using a wide array of skill metrics
(following Galí et al., 2015): R2, RMSE, mean absolute percentage
error (MAPE), percentage bias and the slope of major axis (type II) linear
regression between observations and model estimates
(SlopeMA). All analyses were carried out using MATLAB
R2013b.
Algorithm implementation
The newly developed DMSSAT algorithm (Fig. 2) was
implemented to produce (i) a monthly global DMS climatology and (ii) several
regional time series with 8-day resolution for the period 2003–2016. Further
details and data sources can be found in Sect. S5 and Table S2.
Scheme of the DMSSAT algorithm. The
algorithm proceeds in two steps: the DMSPt sub-algorithm (described
by Galí et al., 2015 see Appendix A) and the DMS sub-algorithm
(this study). Dashed lines mark the PIC-based equation of the DMSPt
sub-algorithm, which in practice is not used when gap-free satellite
Chl fields are used as input.
Global DMSSAT fields were computed using ocean color data
from SeaWiFS (1997–2010 monthly climatology, 1/12∘ grid), SST from
AVHRR and the MIMOC monthly MLD climatology. We used SeaWiFS data to maximize
the temporal overlap between the satellite-based DMSSAT
climatology and the in situ data used to produce the L11 climatology, which
span the period 1972–2010. Note, however, that DMSPtSAT
climatologies derived from SeaWiFS and MODIS-Aqua are extremely similar
(Galí et al., 2015). We established a reference
DMSSAT run where ChlSAT was computed
with a band-ratio algorithm (OC4-OCI standard NASA algorithm) and the
euphotic layer depth (ZeuSAT) was computed as the 1 %
penetration depth of 490 nm radiation (ZeuSAT=4.6/Kd490). The impact of this choice was evaluated with sensitivity
tests where ChlSAT and ZeuSAT were
calculated with the semi-analytical algorithms of Garver–Siegel–Maritorena
(GSM; Maritorena et al., 2002) and Lee et al. (2007), respectively, which are
more appropriate in optically complex waters. Observation gaps caused by low
solar elevation at high latitudes in winter were left blank. Global monthly
DMSSAT fields were averaged onto 1 and 5∘ grids
for mapping and comparison to other DMS climatologies: the interpolated L11
climatology (Lana et al., 2011), and the climatologies derived with the
empirical algorithms of Simó and Dachs (2002; SD02) and Vallina and
Simó (2007b; VS07). The procedures used to calculate these datasets are
briefly described in Sect. 3.2.
Regional DMSSAT time series between 2003 and 2016 were
computed using daily MODIS-Aqua data (4.64 km) combined with the
MIMOC MLD climatology. As done for the global implementation, we produced
DMSSAT fields using both band-ratio and semi-analytical
bio-optical products. We also performed a test comparing
DMSPtSAT obtained with the MIMOC MLD climatology vs.
model-derived MLD time series, showing little DMSPtSAT
sensitivity (Fig. S1 in the Supplement). Since non-climatological satellite
data contain gaps caused by cloudiness, we applied a binning and gap-filling
procedure to obtain full coverage, such that the final regional time series
had a resolution of 8 days and 27.8 km. We produced
DMSSAT time series for the Bermuda Atlantic Time-series
Study (BATS) station (31∘40′ N,
64∘10′ W) and for the Northern Hemisphere at latitudes >45∘ N. The latter dataset was then sampled at selected North
Atlantic sites and at the Ocean Station P (OSP) in the NE Pacific
(50∘ N, 145∘ W). Satellite time series were compared to the
L11 climatology and to in situ DMS and DMSPt. These in situ data, kindly
provided by the BATS (Levine et al., 2016) and OSP
(https://www.waterproperties.ca/linep/, last access: 29 November 2017)
teams, were not used in algorithm development.
Discussion
The DMSSAT algorithm captures in situ variability
(Figs. 4–9 and S3) using a small set of predictor variables (Fig. 2).
Moreover, it reproduces the mismatch between DMS and Chl such that, at
a given ChlSAT concentration, determined DMS can vary by up
to 40-fold. This mismatch is larger than that produced by the SD02 or the
VS07 algorithms, but still smaller than in the database (Fig. 7d–h). The
progressive dissociation between ChlSAT and DMS imposed by
the two-step algorithm, and the nonlinear relationships embodied in Eq. (2)
(Fig. 3), allow DMSSAT to produce a DMS peak in the right
concentration range in summer across different latitudes, despite
order-of-magnitude variations in chlorophyll concentration (Le Clainche
et al., 2010).
Independent validation using satellite matchups suggests that
DMSSAT estimates are globally within ±10 % of in
situ measurements. However, this is at odds with the global mean bias of
-33 % suggested by comparison to the L11 gridded climatology. In some
regions, assessments of DMSSAT bias based on comparison to
matchups and to L11 are consistent (Table 3), indicating shortcomings of the
algorithm and/or in input satellite data. For instance, the globally
optimized coefficients produce too low DMS in northern high latitudes, which
we solved through regional tuning, building on the relative abundance of
DMS(P) measurements north of 45∘ N (Table 2, Fig. 8). In southern
high latitudes, too-low DMSSAT primarily results from the
negative ChlSAT bias (estimated at <-50 % by
Johnson et al. (2013) and -56 % in our matchup dataset). Improving
DMSSAT in this region would require, in the first place, the
improvement of bio-optical algorithms and an important sampling effort to
better document DMS(P) dynamics (Fig. 1; Jarníková and Tortell,
2016). In contrast to high latitudes, database matchups in the Westerlies and
Trades biomes suggest smaller DMSSAT bias than comparison
to L11 gridded data (Table 3). While some of this bias certainly arises from
too-low DMSSAT in late summer at specific locations
(Fig. 9), it is also plausible that divergences between
DMSSAT and DMSL11 arise from regional
biases in the interpolated climatology.
In what follows we pinpoint the strengths and weaknesses of our novel
approach focusing on two aspects. In Sect. 4.1 we examine how geo-statistical
shortcomings of the in situ DMS database cause bias in the L11 interpolated
climatology, highlighting the advantages of DMSSAT and
paving the way towards improving gridded DMS fields. In Sect. 4.2 we
speculate about potential causes behind high
DMS / DMSPt ratios in late summer and early fall. This exercise
shows our limited capacity to account for relevant biogeochemical processes
and explain their interannual variation using satellite data, and identifies
knowledge gaps that need to be tackled to improve diagnostic and prognostic
modeling of oceanic DMS(P). The rationale is that both kinds of issues
(“geo-statistical” and “biogeochemical”) are highlighted by discrepancies
among in situ data, L11, and the macroecological relationships embodied in
our algorithm.
Known sources of error and bias: interpolated climatology
vs. DMSSAT
Global DMS fields estimated by the L11 climatology and by the
DMSSAT algorithm show remarkable geographic differences
(Fig. 5), and their mean concentrations differ by a factor of ∼1.5.
Particularly, changes in the sign of the
DMSSAT–DMSL11 anomaly often follow the
somewhat artificial boundaries of the Longhurst biogeochemical provinces. As
explained below, regional and global biases in L11 arise from the application
of objective interpolation procedures to a global dataset characterized by
(1) the right-skewed distribution of DMS concentrations (Kettle et al., 1999;
Fig. 7), (2) the small amount of monthly data available in many
biogeochemical provinces (Fig. 1), (3) the absence of repeat measurements in
most oceanic regions (Halloran et al., 2010), (4) the low spatial resolution
of most DMS datasets and (5) the preferential sampling of DMS-productive
conditions.
A primary drawback for the calculation of interpolated climatologies is the
overrepresentation of biologically productive conditions in the sea-surface
DMS database. This sampling bias is clearly illustrated by comparing
SeaWiFS-retrieved Chl concentration in the global ocean and in the database
matchups (Fig. 7a). If we assume that SeaWiFS matchups (N=11600)
represent a random sample of the DMS database (N=41304), and considering
the global positive correlation between Chl and DMS (Fig. 7d), this implies
a sampling bias towards high DMS concentrations. The bias is largest when the
comparison is restricted to the spring–summer semester of each hemisphere,
when the median ChlSAT is 0.19 in the global SeaWiFS
climatology and 0.74 in the DMS database matchups. This is the period when
DMS peaks and has more influence on mean annual DMS concentration.
Sampling bias is intertwined with the right-skewed
distribution of
sea-surface DMS concentrations (Fig. 7b) and the poor spatial resolution of
most in situ DMS datasets. Spatial averaging, justified by data scarcity, is
appropriate when applied over small or sufficiently homogeneous regions.
However, when applied over a large and potentially heterogeneous Longhurst
province, it can propagate the sampling bias over the entire province and up
to the global scale. As illustrated in Fig. 6g, the mean DMS concentration in
M5×5 bins is systematically higher, by 40 % on average, than the
corresponding median. Province-level averaging converts the long tail of high
in situ DMS concentrations into too-large province means, such that the
global mode of DMSL11 is similar or even higher than that
of in situ DMS (Fig. 7b).
The influence of extreme in situ DMS concentrations is maximal in productive
regions, where mean/median ratios of around 4 are observed in
M5×5 bins (Fig. 6g). In these regions, sharp productivity gradients
and dynamic ecosystem processes complicate the task of sampling DMS through
all appropriate scales (Nemcek et al., 2008), suggesting the need to apply
finer-scale or dynamic regionalization (Devred et al., 2007) prior to
interpolation. Emerging biogeochemical relationships, like that between net
community production and DMS observed by Kameyama et al. (2013) in the
northeast Pacific, might assist DMS interpolation and determination, but
require validation across contrasting regions and relevant scales (Asher
et al., 2017). In low-latitude oligotrophic areas where DMS shows reduced
spatial variability (Royer et al., 2015), the method used to construct
interpolation-based climatologies (Kettle et al., 1999; Lana et al., 2011)
seems appropriate regarding spatial resolution.
In the temporal domain, the calculation of interpolated climatologies is
complicated by two factors: poor interannual coverage, i.e., the scarcity of
DMS measurements repeated in different years in a given region; and poor
seasonal coverage, i.e., the scarcity of fully resolved seasonal cycles.
Seasonal coverage is limited at the province level (see Table 1 in Lana
et al., 2011) and obviously worse in 5∘ bins (Fig. 1d). Regarding
interannual coverage at the MLongh level, 42 % of the province–month bins
contain measurements from a single year, 21 % from two years and
37 % from three or more years. Thus, data from one or two years are often
assumed representative of the mean ecosystem state in interpolated
climatologies, which is probably not the case in regions with wide
interannual variability or long-term trends (Vantrepotte and Mélin,
2011). While this does not necessarily bias global DMS fields, it can produce
artificial seasonal cycles. For example, L11 suggests the existence of early
spring and fall DMS peaks in the North Atlantic drift area, which result from
interpolation from neighboring regions (Fig. 8a). In contrast,
DMSSAT suggests these are improbable (spring) or
infrequent (fall) features. Another example is found at OSP, where
DMSL11, based on measurements made before 2003, is in poor
agreement with measurements made between 2005 and 2016. In February and June,
DMSSAT is in better accordance with in situ DMS data at
OSP (Fig. 9a and b).
In summary, caution has to be taken when comparing DMS measurements, their
derived climatological products, and independent model estimates that are not
collocated in time. This temporal mismatch may partly explain the poor
correlation between modeled DMS climatologies, on one hand, and the DMS
database and DMSL11 climatology, on the other (Tesdal
et al., 2016). Note that the latter study compared DMS fields binned into
monthly 5∘×5∘ boxes (M5×5), such that
82 % of the bins contained measurements from a single year.
Compared to interpolated climatologies, DMSSAT provides
a robust means to estimate DMS concentrations in sparsely sampled areas
because it relies on satellite observations and macroecological
relationships. The resulting DMS fields are in better accordance with natural
gradients in plankton abundance (biogeography, phenology) and environmental
forcing, as long as the models can account for the driving factors, as discussed below.
Unknown sources of error: how far can we go with remote
sensing algorithms?
By testing DMSSAT in challenging biogeochemical settings
(Fig. 9), we identified its main drawback: the failure to reproduce high DMS,
and more specifically DMS / DMSPt ratios higher than 0.3, at
intermediate PAR levels (Fig. 3), as observed between midsummer and early
fall at BATS and OSP in some years. This limitation can hardly be
fixed without identifying the underlying biogeochemical processes, which are
not necessarily the same in these contrasting biogeochemical regimes.
Although this feature is probably not widespread (see Fig. 2 in Lana et al.,
2011), its occurrence in emblematic time series stations warrants further
discussion.
A common explanation could be the underestimation of irradiance effects,
caused by the use of sea-surface PARSAT rather than
PARMLD as DMS predictor variable. For instance, a delay of
seasonal mixing, associated with deeper irradiance penetration, could enhance
stress-driven DMS production well into fall. Yet examination of the BATS and
OSP time series does not support this explanation. At both sites, the summer
MLD is stable at about ≤20 m and deepens slowly in late summer
(Levine et al., 2016; Steiner et al., 2012), such that
PARMLD declines faster than sea-surface
PARSAT (Eq. 1). Thus, using PARMLD
instead of surface PAR cannot delay the decline of modeled
DMS / DMSPt ratios through the summer, and other factors need
to be invoked.
In the oligotrophic BATS station, some modeling studies proposed
macronutrient limitation of bacteria (Polimene et al., 2011) and also
phytoplankton (Belviso et al., 2012) as drivers of the seasonal mismatch
between DMSPt and DMS, besides irradiance (Vallina et al., 2008). With this
in mind, we tried to factor phosphate and nitrate limitation into our
regression models using different variables: nutrient concentrations,
nutricline depths (Table S4) and limitation factors estimated according to
Michaelis–Menten kinetics (not shown). However, none of the tested variables
improved the regression models significantly. Moreover, macronutrient
availability (limitation) terms generally entered regression models with
positive (negative) coefficients, even when regressions were restricted to
oligotrophic low latitudes. This implies that macronutrient limitation of
phytoplankton growth globally acts to decrease DMS, offsetting nutrient
stress responses that increase DMS. Note also the irregular occurrence of
high DMS at the BATS station in late summer in different years (Fig. 9d; Levine et al.,
2016), and that a BATS-like seasonality is not observed at other sites with
late summer macronutrient limitation (Archer et al., 2009; Belviso et al.,
2012; Galí and Simó, 2015; Vila-Costa et al., 2008). Altogether,
these observations suggest that regional macronutrient stress responses are
difficult to generalize.
Analysis of the OSP time series also yields valuable information because
macronutrient concentrations remain at high concentrations in late summer in
this iron-limited regime (Harrison et al., 2004). While in situ DMS is
accurately estimated by DMSSAT in February and June, the
variable DMS peak occurring around August is strongly underestimated.
Previous studies emphasized the role of iron limitation at OSP, which
configures phytoplankton communities dominated by high DMSP taxa (Asher
et al., 2017; Levasseur et al., 2006; Royer et al., 2010; Steiner et al.,
2012). Interestingly, DMSPtSAT peaks in August at OSP, in
phase with the in situ DMS peak and in good accordance with the few available
DMSPt measurements (Fig. 9c), even though the DMSPtSAT
sub-algorithm does not explicitly resolve phytoplankton taxonomy (Galí
et al., 2015). Therefore, high DMS yields, possibly co-occurring with low DMS
removal rate constants (Asher et al., 2017), are required to explain
DMS / DMSPt ratios observed at OSP. High DMS yields probably
result from a combination of processes, including algal and bacterial DMSP
metabolism (Merzouk et al., 2006; Royer et al., 2010) and microzooplankton
grazing (Steiner et al., 2012). The striking late summer variability at OSP
is presently not captured by biogeochemical models (Steiner et al., 2012) or
empirical algorithms, and it remains unanswered whether it simply reflects
too low sampling frequency, or it is caused by processes that switch on/off
depending on environmental conditions in a given year, or by the variable
location of oceanic fronts in response to circulation patterns.
In summary, our analysis indicates that additional factors are needed to
better reproduce DMS seasonality in specific regions, where
DMS / DMSPt ratios are occasionally higher than the
“baseline” established by Eq. (2) (Figs. 3 and 9). Biotic interactions
involving phytoplankton, bacteria and microzooplankton, regionally
interacting with iron and macronutrient limitation in multiple ways, are good
candidates to explain strong deviations from the mean relationship between
DMS, DMSPt and irradiance. However, they can hardly be represented in
empirical algorithms with our current level of understanding, particularly
when interannual changes are considered (Figs. 8 and 9).
From a practical standpoint, tuning the Eq. (2) coefficients is a workable
alternative in certain regions (Figs. 8 and 9). If sufficient measurements
were available in all oceanic areas, Eq. (2) could perhaps be generalized in
a way that allows its coefficients to vary across different biogeochemical
regimes, while avoiding geographic discontinuities. The inclusion of
additional terms in Eq. (2) lacks strong statistical support when applied
globally, at least with the current dataset (Table S4). If posterior analyses
supported the addition of new satellite variables, their retrieval
uncertainty and its propagation to DMSSAT should be
considered. More obviously, climatological variables such as the WOA nutrient
concentrations are not appropriate to produce time series, and their use in
remote sensing algorithms should be minimized. The only climatological
variable used in our algorithm is MLD, which enters mainly as a categorical
variable (Galí et al., 2015), such that DMSPtSAT is
robust to MLD uncertainties (Fig. S1).
The question of the “optimal model complexity” is a pervasive one in
biogeochemistry, and the right answer may depend on the purpose of each
study. The algorithms tested here showed improved qualitative and
quantitative performance with increasing complexity (VS07 < SD02
<DMSSAT). VS07 failed to capture DMS patterns outside the
subtropical band, possibly due to its inability to modulate the
DMS–irradiance relationship depending on phytoplankton biomass. Inclusion of
phytoplankton biomass-dependent terms in SD02, and of implicit taxonomic
information through the embedded DMSPtSAT sub-algorithm in
DMSSAT, improved algorithm skill in productive regions,
where DMS shows wider seasonal cycles and sharper spatial gradients.
More sophisticated approaches may be needed to achieve significant
improvements in model skill, but they also suffer from major uncertainties.
For example, neural networks were successfully used to estimate DMS in the
Arctic (Humphries et al., 2012), but their robustness might be compromised by
the small training datasets, the use of climatological variables and the lack
of a mechanistic basis. Complex biogeochemical models with satellite data
assimilation have strong potential for resolving interannual DMS variations,
but reliance on several tens of poorly constrained parameters currently
limits their skill (Le Clainche et al., 2010; Galí and Simó, 2015;
Tesdal et al., 2016). An approach of intermediate complexity that deserves
further exploration is DMS diagnosis based on a simplified steady-state budget equation (Galí and
Simó, 2015). Applying empirical parameterizations for the main DMS
production and removal pathways, this approach could enhance the flexibility
of remote sensing algorithms across a wider range of biogeochemical settings.
Conclusions and outlook
Sensors on polar-orbiting satellites provide synoptic observations of the
global ocean surface every few days, and are thus well suited to resolve
spatial and temporal variations in DMS concentration. The
DMSSAT algorithm presented here, based on robust
macroecological relationships, reproduces the main spatial–temporal features
of sea-surface DMS(P) concentrations with remarkable skill using
satellite-retrieved Chl, euphotic layer depth and PAR, and climatological MLD. Other strengths of our approach are
its flexibility, allowing for regional tuning, and the minimal computing
cost.
When compared against the L11 interpolated DMS climatology (Lana et al.,
2011), the DMSSAT climatology shows similar latitudinal
profiles but disagrees in the basin-scale patterns. Examination of spatial
DMS statistics highlights possible shortcomings of the L11 climatology caused
by the combination of sparse and biased sampling, the right-skewed
distribution of DMS and the interpolation procedures used. High-resolution
measurements of DMS(P), if validated against traditional standard techniques
(Royer et al., 2014), will help improve interpolated climatologies and
models.
The global mean area-weighted DMSSAT concentration is
1.63 nM, 33 % lower than DMSL11
(2.43 nM). Global-scale DMSSAT fields are
insensitive to the choice of different Chl and euphotic depth satellite
products, but semi-analytical products should be used in optically complex
coastal waters to avoid DMS overestimation (after detailed examination at
regional scale). Globally, DMSSAT suffers from negative bias
for exogenous and endogenous reasons. In the Antarctic Ocean, it is affected
by the negative bias in satellite-retrieved chlorophyll. At the BATS and OSP
stations (at least), additional factors besides irradiance are needed to
enhance DMS concentration in late summer in some years. Excluding the Coastal
and Antarctic biomes, DMSSAT bias assessed using satellite
matchups ranges between -16 and -20 % (Table 3). This bias is
probably more realistic than the -33 % deduced from comparison to L11,
given the evidence for DMS overestimation in the L11 climatology.
Gauging global DMS emission is critical to understand gas-to-particle
conversion efficiency and the dynamics of cloud condensation nuclei (CCN) populations in the marine
boundary layer. Assuming a linear relationship between global mean DMS
concentration and emission (Fig. 8 in Tesdal et al., 2016),
DMSSAT suggests a global emission of
16–20 TgSyr-1 (depending on the assigned bias). These emission
values lie within the low range of current estimates (Lana et al., 2011;
Tesdal et al., 2016).
Unlike climatologies constructed from the database, the satellite-based
algorithm allows one to explore interannual change. Implementation of
DMSSAT in the subpolar Atlantic between 2003 and 2016
illustrates the wide interannual variability in the timing and magnitude of
the annual DMS peak(s) over large areas. This opens new avenues for studying
the imprint of oceanic aerosol precursors on cloud properties using
simultaneous ocean–atmosphere satellite observations (Krüger and
Grabßl, 2011; McCoy et al., 2015; Meskhidze and Nenes, 2006). If coupled
to atmospheric measurements and numerical models, DMSSAT
enables studying the effects of contemporaneous DMS variability on
atmospheric chemistry and clouds, which could lead to a better understanding
of intricate aerosol–cloud interactions. Further work is warranted to analyze
marine DMS emission variability patterns in regions where climate is
particularly sensitive to DMS, such as the Southern Ocean and the Arctic.