Introduction
Gas exchange across the ocean–atmosphere interface influences
the atmospheric abundance of many compounds of importance to climate and air
quality. Such compounds include greenhouse gases, aerosol precursors,
stratospheric ozone-depleting substances, and a wide range of photochemically
reactive volatile organic carbon compounds that influence tropospheric ozone.
Estimating the air–sea fluxes of all of these compounds requires knowledge
of their distributions in near-surface air and seawater and an understanding
of the transport processes controlling gas exchange across the air–sea
interface. The transport processes are not well understood, in large part
because of the paucity of direct air–sea gas flux observations. The
parameterization of gas exchange is a significant source of uncertainty in
ocean–atmosphere exchange in global models, particularly at high wind speeds
(Elliott, 2009).
Gas flux is typically calculated using the concentration gradient across the
air–sea interface (ΔC) and the gas transfer coefficient (K):
Flux=K⋅ΔC.
K represents the inverse of the resistance to gas transfer on both the
water and air sides of the interface (i.e., 1/K=rw+ra) and can be expressed in either waterside or airside units
(Liss and Slater, 1974). Equation (1) is a very simple expression that belies
the complex physical process involving diffusive and turbulent mixing at the
boundary between two mediums of very different densities. Wind stress is the
predominant forcing for gas transfer, but mixing at the interface is also
influenced by buoyancy, wind–wave interactions, wave breaking, surfactants,
and bubble generation. The interface is chemically complex owing to the
presence of organic films or particles, and, for some gases, the interface
may be biologically/photochemically reactive.
Most air–sea gas transfer calculations utilize wind speed-based
parameterizations derived from deliberate dual tracer observations (Ho et
al., 2011; Nightingale et al., 2000), sometimes scaled to agree with the
long-term global average oceanic uptake of 14CO2 (Sweeney et al.,
2007). The dual tracer technique is a waterside method that requires data
averaging over periods of hours to days, thus averaging over significant
changes in conditions. Eddy covariance is a direct flux measurement carried
out on the air side of the interface. In conjunction with measurements of the
air–sea concentration difference, eddy covariance studies can determine the
gas transfer coefficient, K, on short timescales (10 min–1 h). This
provides a capability to assess variability in K due to the influence of
rapid changes in near-surface processes (e.g., wind–wave interactions,
bubbles, surfactants). Eddy covariance requires high-frequency sensors, and
flux studies to date have been carried out on only a few compounds:
dimethylsulfide (DMS), CO2, methanol, acetaldehyde, acetone, ozone,
carbon monoxide, dinitrogen pentoxide, chloro(oxo)azane oxide and glyoxal
(Huebert et al., 2004; McGillis et al., 2001; Yang et al., 2013; Kim et al.,
2014; Blomquist et al., 2012; Bariteau et al., 2010; Marandino et al., 2005;
Coburn et al., 2014).
DMS air–sea transfer resistance is predominantly on the water side, a
characteristic it shares with CO2. DMS is moderately soluble and weakly
influenced by bubble-mediated gas transfer, in contrast to CO2, which is
sparingly soluble and strongly influenced by bubble-mediated gas transfer.
This makes DMS a useful tracer for waterside-controlled, interfacial gas
transfer. Measurements of gas exchange using insoluble gases have suggested
that the relationship between K and wind speed is non-linear (Nightingale
et al., 2000; Sweeney et al., 2007; Miller et al., 2010; Ho et al., 2011). In
contrast, the majority of DMS eddy covariance data suggests a linear
relationship between K and wind speed (Yang et al., 2011). Blomquist et
al. (2006) suggest that the differences in functional form of these
relationships may be due to the disproportionate influence of bubbles upon
the flux of insoluble gases (Woolf, 1997).
Physical process models have made significant progress in parameterizing gas
exchange with input terms that include but are not limited to wind speed.
However, these models are still in development and are capable of
substantially different estimates of K, depending on how non-wind-speed
terms such as wind–wave dynamics are applied in the model (Fairall et al.,
2011; Soloviev, 2007). Bell et al. (2013) recently demonstrated that some of
the scatter in eddy covariance measurements may be explained by
spatial/temporal differences in wind–wave interaction, although the role of
surfactants cannot be ruled out. Gas exchange measurements in an artificial
surfactant patch (Salter et al., 2011) and in laboratory studies using
natural surfactants (Frew et al., 1990) have demonstrated marked suppression
of gas transfer. Additional eddy covariance gas exchange observations are
required to improve these gas exchange models. Eddy covariance DMS flux
measurements have been made in the Atlantic Ocean (Bell et al., 2013;
Marandino et al., 2008; Salter et al., 2011; Blomquist et al., 2006) and
Pacific Ocean (Marandino et al., 2007, 2009; Yang et al., 2009), with three
of these studies at high northern latitudes. Only one previous study has been
performed in the Southern Ocean (Yang et al., 2011).
The Southern Ocean has a unique wind and wave environment: minimal land mass
in the Southern Hemisphere leads to strong, consistent winds and waves with a
long fetch. The duration of the wind speed event rather than the wind fetch
is the most important factor influencing the waves (Smith et al., 2011). This
region is very important in determining the global uptake of atmospheric
CO2 by the ocean (Sabine et al., 2004) and the supply of DMS as a source
of atmospheric sulfate aerosol (Lana et al., 2011). This paper presents data
collected in the Southern Ocean summer (February–March 2012) as part of the
New Zealand Surface Ocean Aerosol Production (SOAP) cruise (Fig. 1). During
the cruise, a variety of oceanic, atmospheric and flux measurements were
collected. The cruise targeted regions of extremely high biological activity
(blooms of dinoflagellates and coccolithophores) and encountered a number of
atmospheric frontal events leading to winds in excess of 11 m s-1.
Methods
Mast-mounted instrumentation and data acquisition setup
The eddy covariance setup was mounted on the bow mast of the R/V
Tangaroa, 12.6 m above the sea surface. Three-dimensional winds and
sonic temperature (Campbell CSAT3) and platform angular rates and
accelerations (Systron Donner Motion Pak II) were measured on the mast and
co-located with the air sampling inlets for DMS. Air was drawn through the
sampling inlets at 90 SLPM under fully turbulent flow conditions (Re
> 10 000). Analog signals from all of these instruments were
filtered at 15 and then logged at 50 Hz (National Instruments SCXI-1143).
The ship's compass and GPS systems were digitally logged at 1 Hz. The mast
configuration was similar to that used during the Knorr_11 North
Atlantic cruise (Bell et al., 2013), with the following two changes.
An air sampling inlet with integral ports for standard delivery was
fabricated from a solid block of PTFE. The design minimized regions of dead
space that might attenuate high-frequency fluctuations and result in loss of
flux signal.
A shorter length of 3/8′′ ID Teflon tubing was used between the mast and the
container van. A 19 m inlet was used during SOAP in contrast to the 28 m
inlet used during Knorr_11 (Bell et al., 2013).
Cruise track during the SOAP study, which began and finished in
Wellington, New Zealand. The phytoplankton blooms (B1–3) and waypoint 1
(WP1) locations are identified.
Atmospheric and seawater DMS
DMS was measured in air and in gas equilibrated with seawater using two
atmospheric pressure chemical ionization mass spectrometers
(Bell et al., 2013). In both instruments, a heated
(400 ∘C) radioactive nickel foil (Ni-63) generates protons that
associate with water molecule clusters in the sample stream. Protonated
water vapor (H3O+) undergoes a charge transfer reaction to form
protonated DMS ions (m/z=63) that are then quadrupole mass filtered and
counted. Tri-deuterated DMS (d3-DMS, m/z=66) was used as an internal
standard for both instruments.
Atmospheric measurements were made with the University of California, Irvine
(UCI) mesoCIMS instrument (Bell et al., 2013). A gaseous d3-DMS standard was
introduced to the atmospheric sample stream at the air inlet via a three-way
valve mounted at the base of the bow mast. The gas standard was diverted to
waste every 4 h and the response of the d3-DMS signal recorded as a measure
of the inlet tubing impact on signal delay and frequency loss. Air from the
bow mast was sub-sampled at approximately 1 L min-1 and DMS levels
were calculated as follows:
DMSa=S63S66⋅FStdFTotal⋅CTank,
where S63 and S66 represent blank-corrected signals from DMS and
d3-DMS, respectively (Hz), FStd and FTotal are the
gas flow rates of the d3-DMS standard and the inlet air (L min-1), and
CTank is the gas standard mixing ratio.
Seawater measurements were made with a smaller instrument (UCI miniCIMS),
which utilizes a modified residual gas analyzer as the mass filter and ion
detector (Stanford Research Systems RGA-200; Saltzman et al., 2009). Aqueous
d3-DMS standard was delivered by a syringe pump (New-Era NE300) to the ship's
underway seawater supply upstream of the equilibrator (see Bell et al., 2013,
for details). The natural DMS and the d3-DMS standard are both transported
across the membrane and the DMS concentration in seawater in the equilibrator
is then calculated as follows:
DMSSW=Sig63Sig66⋅FSyrFsw⋅CStd
Sig63 and Sig66 represent the average blank-corrected ion
currents (pA) of protonated DMS (m/z=63) and d3-DMS (m/z=66),
respectively, CStd is the concentration of d3-DMS liquid standard
(nM), FSyr is the syringe pump flow rate (L min-1), and
Fsw is the seawater flow rate (L min-1). Seawater
concentrations were averaged at 1 min intervals for the entire SOAP data
set. Lag correlation analysis between the ship surface seawater temperature
and equilibrator temperature records identified that a 3–4 min adjustment
in the DMSsw was required to account for the delay between water
entering the seawater intake beneath the hull of the ship and it reaching the
miniCIMS equilibrator.
We compared our seawater measurements with discrete samples collected and
analyzed by the NIWA team using sulfur chemiluminescence detection (SCD). The
NIWA discrete analyses were performed on water collected from both the
underway supply and from CTD Niskin bottles fired in the near surface
(< 10 m). The analytical techniques (SCD and miniCIMS) typically
agreed well and these results will be discussed elsewhere. Throughout the
cruise, data from the underway and CTD bottles were in good agreement
(Fig. 2), with the exception of day of year (DOY) 54–55, when the ship's
underway supply became significantly contaminated. The contamination was
biological and resulted in DMS levels at least twofold higher than from a
Niskin bottle fired at the same depth. Flushing and soaking the underway
lines in a biologically active cleaning solution
(Gamazyme™) and cleaning the equilibrator with
dilute (10 %) hydrochloric acid resolved the problem. The data from
DOY 54–55 have been excluded from our analysis.
Time series data (10 min averages) from the SOAP cruise. The dashed
black line in (a) indicates neutral atmospheric stability
(z/L=0). SOAP k660 data (e) are divided into on station
(squares, ship speed < 1.5 m s-1) and off station (circles,
ship speed ≥ 1.5 m s-1).
DMS flux calculation: eddy covariance data processing and quality control
Air–sea flux calculation involved the same procedure detailed in Bell et
al. (2013). Apparent winds were corrected for ship motion according to the
procedures of Edson et al. (1998) and Miller et al. (2008). Relative wind
speed was adjusted to correct for air-flow distortion according to the wind
direction-dependent correction presented by Smith et al. (2011), which uses
the computational fluid dynamics Gerris model (Popinet et al., 2004). 10 min
flux intervals with a mean relative wind direction within
±90 ∘(where winds onto the bow =0∘) were retained for
subsequent data analysis. The DMS signal was adjusted relative to the wind
signals to account for the timing delay due to the inlet tubing. The delay
was estimated to be 1.9 s from the periodic firing of a three-way valve on
the bow mast. An equivalent delay estimate was ascertained by optimization of
the cross-correlation between DMS and vertical wind. Flux intervals were
computed from the co-variation in fluctuations in vertical winds (w′) and
DMS (c′) flux. The internal d3-DMS standard exhibited negligible covariance
with vertical wind, confirming that no density correction due to water vapor
or temperature fluctuations (i.e., Webb correction) was required for our DMS
fluxes.
Cospectral analysis objectively removed intervals with large low-frequency
fluctuations, and the criteria for elimination are defined in Bell et
al. (2013). This process reduced scatter in the data without introducing an
obvious bias. High-frequency flux loss in the inlet tubing was estimated by
modeling a filter based on the d3-DMS signal attenuation when the bow mast
valve was switched. The inverse filter was then applied to wind speed binned
DMS cospectra. This enabled an estimate of the necessary wind-speed-dependent
high-frequency loss correction (flux gain =0.004U10n+1.012).
DMS gas transfer velocity calculation
Gas transfer velocities were calculated following
KDMS=FDMSΔC=FDMSDMSsw-DMSair/HDMS,
where FDMS is the measured DMS air–sea flux
(mol m-2 s-1), DMSsw is the seawater DMS level
(mol m-3), DMSair is the atmospheric DMS partial
pressure (atm), and HDMS is the temperature-dependent DMS
solubility in seawater (atm m3 mol-1; Dacey et al., 1984).
KDMS values were calculated from the cruise data using 10 min
averages.
The water side only gas transfer coefficient, kw, was obtained
from the expression
kw=1KDMS-1α⋅ka-1,
where KDMS is the total DMS gas transfer coefficient, α
is the dimensionless Henry's law constant for DMS, and ka is the
air-side gas transfer coefficient. In situ ka values were
obtained from NOAA COARE driven by in situ measurements of wind speed,
atmospheric pressure, humidity, irradiance and air and seawater temperature.
The relative influence of ka upon our estimates of kw
was greater when measured KDMS was high (Fig. A, in the
Supplement). This has little impact upon our data, as the average (mean)
difference between kw and KDMS was 7 % and showed
no wind speed dependence (Fig. B in the Supplement). In order to compare our
results with various other gas transfer parameterizations, kw was
then normalized to a Schmidt number of 660 (CO2 at 25 ∘C):
k660=kw⋅660ScDMS-1/2,
where ScDMS is calculated using the ship's seawater temperature
recorded at the bow and Eq. (15) in Saltzman et al. (1993).
Results
Cruise track, meteorological, and oceanographic setting
The SOAP cruise sampling strategy was to identify phytoplankton blooms using
ocean color imagery and then use underway sensors (e.g., chlorophyll a
fluorescence, DMS) to map out the in situ spatial distribution. Three blooms
were identified and sampled: B1, B2 and B3 (Fig. 1). B1 was an intense
dinoflagellate-dominated bloom at approx. 44.5∘ S, 174.7∘ E
(DOY 45.9–49.8) with extremely high levels of seawater DMS (16.8±1.5 nM). After B1, the ship headed southwest to a waypoint (WP1) at approx.
46.3∘ S, 172.5∘ E (DOY 50.5). The waters at WP1 contained
moderate DMS signals (3.8± 0.4 nM) and weakening fluorescence (0.83±0.38 mg m-3), so minimal time was spent at this location. The
return transect into B1 from WP1 is discussed in detail in Sect. 3.3. The
second bloom (B2) was a coccolithophore-dominated bloom at approx.
43.6∘ S, 180.2∘ E (DOY 52.9–56.1) that had stronger DMS
signals (9.1± 2.9 nM) and fluorescence (0.99±0.35 mg m-3).
After sampling B2, the B1 location was revisited and a new bloom (B3) was
identified with a mixed population of coccolithophores, flagellates and
dinoflagellates (DOY 57.9–60.5). B3 DMS levels (5.9±1.5 nM) were
substantially lower than in B1.
The time series plot in Fig. 2 describes the oceanographic and meteorological
variability throughout the cruise. Surface ocean temperatures (SSTs) were
consistent at 14.7±1.0 ∘C, while atmospheric temperature
fluctuated just above and below the SST. Weather systems from the north
brought relatively warm air and systems from the south brought cooler air.
For example, the atmospheric front on DOY 55 from the south caused air
temperatures to drop from approximately 18 to 12 ∘C (Fig. 2a).
Frontal systems passed over the ship regularly throughout the cruise and the
final system (DOY 61.6–64) brought intense winds from the north. During
SOAP, the horizontal wind speeds predominantly ranged from 1 to
15 m s-1. The atmospheric boundary layer was stable
(z/L > 0.05) for approximately 25 % of the cruise
(Fig. 2a). Yang et al. (2011) suggest that a stable boundary layer leads to
greater scatter and a potentially negative bias in k660 vs. wind speed
plots. Our data do not suggest increased scatter or any bias during stable
periods (Fig. C in the Supplement), and we have not filtered the SOAP
k660 data on this basis.
Oceanic and atmospheric DMS levels were extremely high during the first half
of the cruise (DOY 44–54; Fig. 2c). The majority of this period was spent
in and around B1 waters, with elevated seawater DMS (> 10 nM)
and atmospheric DMS (> 600 ppt). Oceanic DMS was always at least
an order of magnitude greater than atmospheric DMS, meaning that the air–sea
concentration gradient was effectively controlled by DMSsw. The
second half of the cruise (DOY 55–65) encountered less productive blooms
with lower seawater DMS levels. The reduction in oceanic DMS was mirrored by
lower atmospheric DMS levels (151±73 ppt, DOY 55–65).
(a) 10 min average DMS gas transfer coefficients vs. mean
horizontal wind speed during the SOAP cruise, expressed as k660 and
U10n (see Methods). For reference, the NOAA COARE model output for DMS
is plotted, calculated using average SOAP input parameters and the
turbulent/molecular coefficient, A=1.6, and the bubble-mediated
coefficient, B=1.8. The red dashed line is interfacial transfer velocity
only. The blue solid line includes the bubble contribution to gas transfer.
The green dashed line is the least squares linear regression fit to the SOAP
10 min averaged data (k660=2.31U10n-1.51). (b)
Residual values from the least squares linear regression fit in (a).
The green dashed line is exact agreement with linear regression model. Red
squares are the median residual within each 1 m s-1 wind speed bin.
Negative deviation of the median residuals from the linear regression
demonstrates the positive skew in k660.
10 min average DMS fluxes (FDMS) measured by eddy covariance are
plotted in Fig. 2. FDMS reflected the seawater DMS levels, with
three notable peaks while inside B1 waters
(> 60 µmol m-2 day-1, DOY 48–50).
FDMS was generally lower during the second half of the cruise
(13±10 µmol m-2 day-1, DOY 55–65), but elevated
fluxes were still observed due to increased horizontal wind speeds (e.g.,
approx. 45 µmol m-2 day-1 on DOY 61.6). SOAP gas
transfer coefficients were calculated at 10 min intervals (Fig. 2e)
following Eqs. (1)–(6) using measurements of FDMS, oceanic and
atmospheric DMS levels and SST. During some periods of constant wind speed,
the NOAA COARE (v3.1) estimates are close to the observed k660 values
(e.g., DOY 51). However, at various times during the cruise, the NOAA COARE
estimates exhibit significant divergence from the observed k660 values.
The difference was sometimes positive, as on DOY 48, and sometimes negative,
as on DOY 53. These divergences are not random scatter about the COARE
prediction and suggest that unaccounted-for processes are influencing our
measurements of gas transfer.
Wind speed dependence of gas transfer coefficients
The SOAP gas transfer coefficients exhibit a positive correlation with wind
speed (Spearman's ρ=0.57, p < 0.01, n=1327; Fig. 3a). A
linear least squares fit to the data gives k660=2.31± 0.11U10n-1.51± 0.97 with an adjusted R2=0.25. As with previous shipboard
eddy covariance DMS studies, using a second-order polynomial does not improve
the fit to the data (adjusted R2=0.25). The linear model is not well
suited for this data set, because the residuals are not normally distributed
(Fig. 3b). The frequency distribution of the SOAP k660 measurements
exhibits positive skewness at all wind speeds (Fig. D in the Supplement). The
skew in the SOAP k660 data appears to originate in the frequency
distribution of seawater DMS. Surface ocean DMS distributions are typically
characterized by positive skew, and this is evident in the global surface
ocean DMS database (Lana et al., 2011).
It is not surprising to see skewed distributions in the SOAP data, as the
cruise encountered strong, non-linear gradients in biological activity. There
is no skewness in the distribution of winds within each wind speed bin.
Skewness in the seawater DMS distribution should propagate into the DMS flux
distribution simply because air–sea flux is proportional to air–sea
concentration gradient, which is controlled in turn by seawater DMS levels
(Figs. E and F in the Supplement). If FDMS and ΔC are
highly correlated, then the variance in k660 should be considerably less
than that in either parameter and would exhibit less skew. This is not the
case: k660 exhibits a similar skew to FDMS and ΔC.
For example, the correlation coefficient between DMS flux and seawater
concentration in the 13–14 m s-1 wind speed bin (Spearman's ρ=0.45, p < 0.01, n=47) is considerably lower than expected.
Decorrelation of DMS flux and seawater concentration is likely due to
mismatches between seawater DMS levels measured aboard ship and those in the
actual footprint of the flux. Misalignment between seawater DMS levels and
the flux footprint is virtually unavoidable in a region of strong spatial
heterogeneity, where wind direction and ship track are never perfectly
aligned.
As a result of the frequency distribution observations in the SOAP data set,
we reexamined data from a recent North Atlantic cruise (Bell et al., 2013;
Figs. E–G in the Supplement). The frequency distributions of k660,
FDMS and DMSsw exhibit a similar positive skewness to
that in the SOAP data set. In order to represent the central tendency of the
k660 data better and to assess the relationship with wind speed,
geometric means were computed for 1 m s-1 wind speed bins (Fig. 4).
Geometric binned k660 data from both cruises are lower than the
arithmetic binned data. The binned k660 SOAP data demonstrate a
shallower slope using the geometric means.
The SOAP k660 bin average data (Fig. 5) exhibit a linear relationship
with wind speed for low and intermediate winds, as found in previous DMS flux
studies (e.g., Huebert et al., 2010; Yang et al., 2011; Marandino et al.,
2007, 2009). For wind speeds up to 14 m s-1, the binned geometric mean
SOAP data yield a linear regression equation of k660=2.07U10n-2.42, which is slightly shallower than that obtained from a compilation of
previously published DMS gas transfer measurements (k660=2.6U10n-5.7; Goddijn-Murphy et al., 2012). In the higher wind speed bins (above
10 m s-1), the relationship between k660 and wind appears to
weaken. A weaker relationship between k660 and wind speed at high wind
speeds was also observed in the North Atlantic (Bell et al., 2013). In both
cruises, there are limited data at wind speeds above 10 m s-1, so this
phenomenon should be viewed with caution. Bell et al. (2013) suggested that
the effect could be due to suppression of near-surface waterside turbulence
due to wind–wave interactions (Soloviev et al., 2007; Donelan et al., 2010).
The SOAP study did not include direct measurements of wave properties or
surfactants. Significant wave height was estimated using satellite reanalysis
products from ECMWF and NCEP, which agreed well (Spearman's ρ=0.91,
p < 0.01, n=2876). Significant wave height exceeded 4.5 m
during SOAP. There is no obvious relationship between significant wave height
and the scatter in the relationship between gas transfer and horizontal wind
speed during SOAP (Fig. J in the Supplement). In situ fluorescence was used
as an indicator of biological activity during SOAP. Fluorescence sensors were
located in seawater continuously pumped through the ship from the
near-surface intake beneath the hull. The variability in the gas transfer
velocity data is not explained by surface ocean fluorescence (Fig. K in the
Supplement). Note that fluorescence is not necessarily a reliable indicator
of surfactant concentrations. The relative importance of waves and/or
surfactants in air–sea gas exchange remains unclear and requires dedicated
measurements to be made concurrent with direct assessments of gas exchange by
eddy covariance.
Bin average gas transfer coefficients for this study (SOAP) and the
data collected in the North Atlantic (Knorr '11). Mean values were
calculated for 1 m s-1 U10n bins using arithmetic (solid
lines, filled symbols) and geometric (dashed lines, open symbols)
approaches.
Uncertainties in K introduced by flux footprint and seawater DMS
heterogeneity
As discussed above, spatial heterogeneity of seawater DMS can introduce
uncertainty in gas transfer coefficients derived from eddy covariance
studies. It is logistically challenging to quantify footprint effects from a
single ship, and it has not been done on prior studies. On the SOAP cruise,
the fortuitous alignment of winds and ship track downwind of the
dinoflagellate-dominated bloom (B1) provided a unique opportunity to quantify
the length scale associated with the flux footprint.
Bin average gas transfer coefficients from this study compared with
prior published DMS eddy covariance measurements: Wecoma (Marandino et al.,
2007), Knorr '06 (Marandino et al., 2009), SO-GasEx (Yang et al.,
2011), DOGEE (Huebert et al., 2010), BIO (Blomquist et al., 2006), TAO
(Huebert et al., 2004), VOCALS (Yang et al., 2011) and Knorr '11
(Bell et al., 2013). Geometric mean SOAP k660 values were calculated for
1 m s-1 U10n bins (error bars represent ±2 SE;
minimum data points per interval = 6).
The SOAP cruise spent approximately 5 days mapping out the spatial extent of
B1 waters, then transited out of the bloom to WP1 about 150 km to the
southwest. The ship then steamed back into and across B1 at a ship speed of
5.1±0.7 m s-1, over about 18 h (DOY 50.85–51.35; Fig. 6).
Meteorological and oceanographic conditions were relatively constant during
the B1 transect, with wind speeds ranging from 5.5 to 9.7 m s-1, wind
direction from 5 to 33∘, air temperature of 15.4±0.8, and SST of
14.4±0.5 (Fig. 7). Atmospheric stability was neutral to stable during
this period. A detailed picture of surface ocean DMS levels in and around B1
can be seen from the data collected between DOY 45.65 and DOY 51.35
(Fig. 6). DMS levels exhibit a sharp step change at approximately
44.6∘ S. DMS concentrations south of the bloom were less than 5 nM.
Near the bloom center, levels increased rapidly over a few kilometers from
below 10 nM to greater than 15 nM. Atmospheric DMS levels were quite stable
during the transect, with a mean of 489±58 ppt. The ship's heading
(approx. 27∘) meant that winds blew almost directly onto the bow,
with less than 10∘ difference for the final 60 km of the transect
back into B1.
Figure 7 depicts seawater DMS levels (green symbols) as the ship steamed into
B1 waters. DMS levels 120 km away from the bloom were below 5 nM and
consistently 5–10 nM until the southern perimeter of the bloom (0 km). DMS
levels increased rapidly to 15–20 nM as the ship moved into the bloom. DMS
flux divided by the horizontal wind speed is also presented. We assume a
relatively linear relationship between k660 and U10n and that
fluctuations in FDMS /U10n (Fig. 7, blue symbols) are driven
primarily by changes in ΔC (i.e., DMSsw). Spikes in
FDMS /U10n are evident in DMSsw after a
consistent distance/time lag. The gas transfer velocities are shown in
Fig. 7e during the transect into B1. COARE model output for DMS is plotted as
a reference line. Spikes in k660 are coincident with sharp changes in
FDMS/U10n prior to the lagged corresponding change in
DMSsw.
Latitude–longitude map of surface ocean DMS concentrations (nM) in
and around B1 waters between DOY 45.65 and 51.35. Start and end points of
the transect into B1 are indicated. The arrow indicates the prevailing wind
direction along the transect.
On this transect, the eddy covariance flux footprint was directly ahead of
the ship, so a lag would be expected between the FDMS and ΔC (i.e., DMSsw). The maximum correlation between
FDMS/U10n (using the midpoint of the flux interval) and
ΔC was obtained for a lag of 8 min. This lag represents a distance
of ∼2.5 km at 5.1 m s-1 ship speed. Applying this lag to the
calculation of gas transfer velocity reduced the scatter (Fig. 8). We
compared the flux footprint obtained from the lag calculation to a flux
footprint calculation using an online version of an analytic dispersion model
(http://www.geos.ed.ac.uk/abs/research/micromet/java/flux.html; Kormann
and Meixner, 2001). We ran the model with representative conditions for the
SOAP B1 transect: measurement height =12 m; wind speed =8 m s-1; roughness length =0.02 m (minimum value available);
zero-plane displacement =0.5 m (minimum value available); sensible heat
flux =-20 W m-2; air temperature = 15 ∘C. The footprint
model predicts a peak relative flux contribution (defined as 90 % of the
relative flux) 0.8 km ahead of the ship, less than half of the distance
inferred from the field observations. The calculated footprint is highly
sensitive to the input parameters. During the SOAP B1 transect, atmospheric
stability was slightly stable but close to neutral (z/L ∼+0.1).
Relatively small changes in wind speed (±1 m s-1), temperature
(±1 ∘C) or sensible heat flux (+10 W m-2) alter the
stability such that model predictions of the peak footprint contribution
range from 0.3 to 1.9 km. Model runs where measurement height was varied to
reflect the limits of ship motion (significant wave height from ECMWF
suggests the vertical displacement of the flux inlet was at least 2.5 m)
gave minimum and maximum peak flux footprint contributions of 0.4 and
2.0 km, respectively.
Shipboard measurements during the south–north transect into B1. The
data are plotted as a function of distance from the southern perimeter of the
bloom. Symbols represent 10 min averages, with the exception of 1 min
average seawater DMS concentrations (d). The red line in
(e) is the COARE model output for DMS, shown for reference.
Despite the sensitivity of the model to the input parameters, these estimates
are not as large as the footprint derived from the lag calculation. Flux
footprint models make the assumption that the surface source is spatially
homogeneous. This was not true during the SOAP B1 transect – the location of
the peak contribution to the flux was not the same as the peak in the
footprint model. Greater DMSsw concentrations at the farthest
extent of the flux footprint will cause the flux signal to be dominated by a
signal from farther afield than implied by the footprint model. This is the
likely explanation for the mismatch between our correlation analysis and the
flux footprint model output.
Huebert et al. (2010) addressed surface ocean spatial heterogeneity for their
estimates of DMS gas transfer velocity during the June 2007 Deep Ocean Gas
Exchange Experiment (DOGEE) in the North Atlantic. When the hourly
DMSsw relative standard error of the mean (RSEM) exceeded 0.25,
gas exchange data were not included in their analysis. Removing k660
data with high DMSsw variability during DOGEE improved the
correlation between k660 and wind speed. We assessed variability in our
high-frequency DMSsw data by calculating the forward-looking,
running standard deviation (SD) on a 1 h timescale. The relative standard
deviation (RSD) was then calculated by dividing the SD by DMSsw.
Using the RSD would not have been reliable for identifying the outlying
k660 data during the B1 transect (Fig. 8a). The scatter in k660 vs.
U10n in the entire SOAP data set cannot be reduced on the basis of the
associated RSD values (Fig. L in the Supplement).
10 min average DMS gas transfer coefficients (k660) vs. mean
horizontal wind speed (U10n) during the south–north transect into B1
waters. Data are colored by the relative standard deviation (RSD) for
corresponding DMSsw (see text). (a) Gas transfer
velocities calculated before adjustment of DMSsw to account for
decoupling from the flux footprint. (b) k660 calculated using
seawater DMS shifted by 8 min to account for the lag between measured flux
and ΔC (see text).
Conclusions
The SOAP k660 bin average values are in good agreement with
previous gas transfer studies using eddy covariance of DMS (Yang et al.,
2011; Bell et al., 2013; Marandino et al., 2007). As noted earlier, these
studies provide evidence that interfacial gas transfer is a relatively linear
function of wind speed for low to intermediate wind speeds. There is some
evidence that the dependence on wind speed weakens at higher wind speeds,
both in this study and in the Knorr_11 study (Bell et al., 2013).
There is no evidence in any of the DMS eddy covariance data sets that the
interfacial (non-bubble-mediated) component of gas transfer has a wind speed
dependence greater than linear. However, there are still very limited data
above 10 m s-1, and the high wind speed trends are uncertain.
The scatter in the SOAP data is typical of shipboard eddy covariance flux
measurements. This arises from fluctuations in near-surface turbulence and
vertical entrainment, vertical shear, ship motion, heterogeneity in seawater
DMS and variations in atmospheric DMS due to chemical losses (Blomquist et
al., 2010). We note the skewness of the gas transfer velocities in a given
wind speed range and use geometric statistics to characterize the central
tendency and variance of the data. This skewness is likely driven by the
inherent log-normal distribution of seawater DMS levels. We propose that
spatial heterogeneity in seawater DMS causes decorrelation between the
measured seawater DMS and the observed DMS flux, which results in skewness
propagating into the calculated transfer coefficients. The data from this
study may be particularly influenced by the large differences in seawater DMS
values inside and outside the phytoplankton blooms. Similar skewness was
observed in data from the North Atlantic Ocean (Bell et al., 2013), and this
phenomenon likely affects all DMS eddy covariance studies to some degree. If
so, then some transformation of the DMS gas transfer velocities is warranted.
The transect from WP1 into B1 provided a unique opportunity to quantitatively
estimate the spatial extent of the eddy covariance flux footprint. The data
suggest that the shipboard flux measurements were sensitive to changes in
seawater DMS approximately 2.5 km upwind of the ship, a surprisingly large
distance. This transect was conducted under neutral to stable conditions,
when one might expect the flux footprint to be relatively large. This result
is much greater (twofold or more) than that predicted using an analytic
dispersion model (Kormann and Meixner, 2001). The discrepancy between the
flux footprint model output and our correlation analysis is probably because
the model assumes spatial homogeneity in the DMSsw concentrations
within the flux footprint. A flux footprint model developed for marine
air–sea gas flux measurements would be an invaluable tool for the
ocean–atmosphere gas exchange research community.
During the SOAP cruise, we saw no obvious evidence of a first-order
biological effect on gas transfer coefficients. From this, it could be
inferred that surfactants in the dinoflagellate and coccolithophore blooms
did not exert a significant effect on water side turbulence. Any modification
of the gas transfer velocity vs. wind speed relationship by surfactants or
waves during SOAP was masked by other influences upon the variability in gas
flux measurements. Minimizing the scatter in gas transfer velocity is
critical in order to observe the influence of non-wind-speed processes and to
draw firm conclusions about their impact upon air–sea gas transfer. The
challenge for the gas exchange community is that heterogeneity in seawater
DMS concentrations is linked to phytoplankton growth, which likely also
determines surfactant effects upon the gas transfer velocity.