A high-resolution survey of DMS, CO2, and O2/Ar distributions in productive coastal waters



[1] We present continuous, high-resolution measurements of surface dimethylsulfide (DMS), pCO2, and O2/Ar obtained in coastal waters off British Columbia, Canada, using membrane inlet mass spectrometry (MIMS). Sampled underway at a frequency of twice per minute (every ∼160 m at 10 knots cruising speed), our data reveal fine-scale structure in gas variability and its covariance with a number of hydrographic parameters. All parameters exhibited large ranges (pCO2, 200–747 ppm; DMS, <1–29 nM; chl a, <0.1–33 μg L−1), highlighting the dynamic nature of the study area. A strong anticorrelation between pCO2 and O2/Ar was observed across the survey region, with the distributions of these gases influenced by biology and its interplay with physical processes. In contrast, DMS levels, which varied dramatically over short distances, showed no significant correlations with any single variable for the full, high-resolution data set. However, when measurements were binned to a much coarser spatial resolution, we found a linear relationship between surface DMS and the chlorophyll/mixed layer depth ratio. The slope of this relationship differed significantly from that previously derived from open ocean data. We used several statistical techniques to estimate the spatial variability of gases and hydrographic parameters and the inherent sampling errors associated with low-frequency sampling approaches. These analyses emphasize the importance of high-resolution sampling in coastal areas, particularly for DMS.

1. Introduction

[2] Rising concern about global climate change, driven by high anthropogenic CO2 emissions has led to increased efforts to quantify the ocean's role as a source or sink of climatologically active gases [Denman et al., 2007]. These include greenhouse gases such as CO2 that insulate the planet, as well as gases such as dimethylsulfide (DMS), which can potentially cool Earth's climate through the formation of cloud condensation nuclei that promote cloud formation and scatter incoming radiation [Charlson et al., 1987]. Decades of oceanographic gas surveys have culminated in the synthesis of thousands of measurements of pCO2 and DMS into monthly global climatologies for both gases (Takahashi et al. [2002] and Kettle et al. [1999], respectively). Although the number of measurements continues to steadily increase, observations in continental shelf waters remain sparse. These regions thus suffer from low spatial and temporal resolution and are poorly represented in global gas data sets [Takahashi et al., 2002; Kettle et al., 1999]. This is a significant limitation since coastal regions, despite their small areal extent, play a disproportionately large role in air-sea gas exchange because of their high productivity and dynamic physics. Coastal waters are particularly large sources of trace biogenic gases such as DMS with annual emissions from continental margins comparable to those from the much larger oligotrophic ocean regions [Andreae and Raemdonck, 1983; Andreae, 1990]. Neglecting to include these areas in global climatologies may thus impart significant errors on DMS flux estimates. Moreover, as DMS has a short atmospheric lifetime (on the order of days) [Chin and Jacob, 1996], it is especially important to identify its local origins, especially in coastal areas where biogenic sulfur sources affect the relative importance of anthropogenic ones [Jones et al., 2001].

[3] Compounding the problem of low measurement resolution is our poor understanding of the relative importance of the underlying processes driving the observed gas distributions. Although biological processes such as photosynthesis, respiration and calcification are believed to dominate the seasonal variations in pCO2 over most of the surface ocean [Takahashi et al., 2002], temperature changes likely play a larger role in driving air-sea CO2 fluxes at high latitudes [Murata and Takizawa, 2003]. In our study region, strong biological CO2 drawdown occurs in summer while outgassing dominates in winter, yet net annual CO2 fluxes are still poorly constrained [Ianson and Allen, 2002].

[4] The DMS cycle is more complex than that of CO2 and we are still far from a mechanistic understanding of the factors driving DMS production [see Simó, 2004, and references therein]. While it is clear that DMS originates from the algal metabolite dimethylsulfoniopropionate (DMSP), there are species-specific differences in DMSP production [Keller et al., 1989], which in turn are affected by the nutrient status of the cells [Sunda et al., 2002; Bucciarelli and Sunda, 2003]. In addition, the process of DMSP release and conversion to DMS depends on complex foodweb interactions which are affected by the physicochemical environment [see Simó, 2001]. As a result, attempts to relate DMS levels to single parameters such as chlorophyll, phytoplankton species composition, or even DMSP concentrations have proven largely unsuccessful [Leck et al., 1990; Kettle et al., 1999]. Recent efforts have thus focused on finding empirical relationships between DMS levels and combinations of relevant biological, physical, and chemical variables that would obviate the need for a full mechanistic understanding of the processes involved [i.e., Anderson et al., 2001; Simó and Dachs, 2002; Belviso et al., 2004a]. Perhaps the most promising of these is the algorithm of Simó and Dachs [2002] (hereafter referred to as SD02) which simulates oceanic DMS based solely on the ratio between SeaWiFS surface chlorophyll and a climatological mixed layer depth (CHL/MLD). However, this algorithm was developed and tested using mainly oceanic measurements, with data from highly productive waters (chl a > 15 μg L−1) specifically filtered out. The effectiveness of this algorithm at predicting DMS levels in coastal regions has thus far not been determined.

[5] In order to fully capture gas distributions in coastal areas and begin to understand the biological and physical processes driving these distributions, it is desirable to measure multiple gases at high resolution, simultaneously. Many current sampling methods measure individual gases in isolation at insufficient resolution to capture fine-scale variability. This is particularly problematic for DMS which is at best measured a few times per hour with purge-and-trap gas chromatography. Recent work has demonstrated that membrane inlet mass spectrometry (MIMS) provides the capability to resolve spatial variability in DMS and other gases [Tortell, 2005a, 2005b]. This method is ideally suited for underway surveys since a suite of both major and trace gases can be measured simultaneously at a rate of twice per minute, providing a dramatic increase in spatial resolution over current methods, particularly for DMS.

[6] Here we present the first dedicated application of MIMS to dynamic, productive coastal waters off British Columbia, Canada. The emphasis of the present study was to examine the covariance of the gas and hydrographic data to relate observed DMS levels to other environmental variables at high spatial resolution. Our second objective was to test the applicability of the predictive SD02 algorithm for use in coastal regions. Our findings suggest that, at coarse resolution, DMS concentrations in highly productive areas can be related to the CHL/MLD ratio, although by a different scaling factor than that observed in oceanic waters. However, despite the broad-scale relationship between DMS and CHL/MLD, we observed significant fine-scale heterogeneity in DMS distributions that could not be predicted empirically. Our results emphasize the utility of multigas measurements and the need to sample gases at appropriate spatial scales that are especially short in coastal waters.

2. Methods

2.1. Study Area and Sampling

[7] Dissolved gas, temperature, salinity and chlorophyll a measurements were made underway along several transects off the west coast of British Columbia, Canada, between 11–19 August 2004 onboard the CCGS John P. Tully (see Figure 1a for transect locations). The cruise track crossed many dynamic oceanographic features including nearshore straits influenced by strong tidal mixing (T1a, T8, Figure 1a), as well as open shelf areas that are sites of upwelling and eddy formation. The northern shelf transects (T1b-5, Figure 1a) were located in Queen Charlotte (QC) Sound, an area that serves as the source region for anticyclonic (downwelling) eddies that carry coastal waters rich in iron offshore to the Gulf of Alaska [Johnson et al., 2005]. During summer, this region is thought to experience relaxation from strong winter downwelling, with very little upwelling. Transect 5 crosses from QC Sound into an area where sporadic summer upwelling is expected off Vancouver Island. Near the southern end of Vancouver Island, transects 6–7 enter the Juan de Fuca (JdF) eddy, a persistent, localized summertime feature. This cyclonic (upwelling) eddy supports a highly productive, diatom-dominated community [Marchetti et al., 2004].

Figure 1.

Surface plots of (a) temperature (°C), (b) salinity (psu), (c) chlorophyll a (μg L−1), (d) pCO2 (ppm), (e) O2/Ar (ion current ratio), and (f) dimethylsulfide (DMS) (nM). Numbering of transects is shown in Figure 1a. See section 2 for explanation of gaps in salinity, chlorophyll, and DMS plots.

2.2. Gas Measurements

[8] Membrane inlet mass spectrometry (MIMS) was used to measure dissolved gases underway (DMS, O2, Ar, and CO2) as described in detail in Tortell [2005a]. Briefly, seawater from the ship's underway intake system (4.5 m depth) was pumped through polypropylene tubing into a sampling cuvette connected to the mass spectrometer. Flow rates through the cuvette were controlled with a gear pump and were kept constant at ∼200 ml min−1. After diffusion through a dimethylsilicone membrane, gases were measured in the vacuum chamber by single ion monitoring of the signal intensities at the relevant m/z (mass to charge) ratios (32, 40, 44 and 62 for O2, Ar, CO2, and DMS, respectively). We use a Hiden Analytical HAL20 triple filter quadrupole mass spectrometer with an electron impact ion source set at a 500 μA emission current. DMS was detected using a secondary electron multiplier (chaneltron) with an applied voltage of 900 V, while all other gases were detected using a Faraday cup. It was necessary to use both detectors to capture the wide dynamic range of analyte concentrations.

[9] During our survey, all gases were measured twice per minute, or approximately every 160 m at the typical vessel cruising speed of ∼10 knots. This measurement resolution was determined by sampling frequency, rather than instrument response time. Our sampling frequency was limited to a maximum of twice per minute by the time required to switch between the chaneltron and Faraday detectors (i.e., the time needed to charge and dissipate voltages on the electron multiplier). Much higher sampling frequencies can, however, be achieved if only a single detector is used (i.e., only DMS is measured). Laboratory experiments measuring instrument response to step function perturbations in gas concentrations have shown that the analytical response time of the MIMS is less than 30 s for all gases. For O2 and Ar, response times are on the order of several seconds, while response times for CO2 and DMS are ∼5–10 s because of the greater polarity of these molecules.

[10] The DMS signal output from the MIMS was calibrated using standards prepared by hydrolysis of sterile DMSP stock solutions (Research Plus Inc.) in 1 M NaOH. Aliquots of the liquid DMS standard were added to 500 ml volumes of DMS-free seawater (>1000 m depth) and were always diluted 105–106-fold to keep the pH of the final DMS standard constant, and prevent matrix effects on the membrane. Standards were analyzed on the MIMS by recirculating the liquid through the sampling cuvette using a gear pump connected to a manual sampling valve. The detection limit based on a 3:1 signal-to-noise ratio of the blanks was 1 nM. On the basis of previous method comparison exercises [Tortell, 2005a], the accuracy of our MIMS measurements is ±∼5% relative to standard purge and trap GC analysis. On the last day of sampling we experienced a power failure which caused the instrument to shut down. This resulted in a malfunction in the secondary electron multiplier (SEM) used to measure DMS. Thus, DMS data were not available along T8 (Figure 1f).

[11] Gas standards were not available for CO2, O2 or Ar. However, independent measurements of pCO2 obtained from an underway IR pCO2 equilibrator (LI-COR LI-6262) were used to calibrate the MIMS CO2 signal (ion current at m/z 44). Calibration coefficients were determined from a linear regression between the CO2 signal from the MIMS and the corresponding pCO2 measurement from the equilibrator [see Tortell, 2005a, Figure 4b]. Each day of sampling was calibrated separately to compensate for shifts in the m/z 44 baseline of the MIMS. Coefficients of determination for the pCO2 calibration were as follows: r2 = 0.98 for T1, r2 = 0.60 for T2, T3, and T4, r2 = 0.87 for T5, r2 = 0.96 for T6, T7 and T8. The poor correlation between the MIMS and equilibrator data for T2–4 is expected because the range of pCO2 values encountered along these transects was small (304–335 ppm), such that any noise in the MIMS CO2 signal and any minor time offsets between the two instruments would be amplified. Time offsets in the form of (1) small (<1 min) discrepancies in the time stamps, (2) slight differences in the time lags from the seawater intake, and (3) differences in equilibration time, could have accounted for the discrepancies between the two instruments. For example, the IR-pCO2 equilibrator has an equilibration time constant associated with the liquid-gas exchange, such that it would respond more slowly to a rapid change in CO2 concentration as the headspace needs to equilibrate with the inflowing sample. In contrast, the MIMS membrane equilibrates more quickly and thus has a faster response time. However, despite the low r2 values for T2–4, the average absolute difference between the two instruments was ∼5 ppm.

[12] Standards were not available to calibrate the oxygen and argon signals from the MIMS. However, the O2 signal was normalized to Ar to yield a biologically relevant parameter representing the balance between photosynthesis and respiration. This is possible because of the similar solubilities of these two gases (which negate temperature and salinity effects on saturation state), and because argon concentrations are unaffected by the biological processes that produce and consume oxygen [Craig and Hayward, 1987]. Although bubble injection will have a slight influence on the O2/Ar ratio, the effect is negligible compared to the biological signal [Craig and Hayward, 1987] and can be ignored (as by Kaiser et al. [2005]) for the purposes of this study.

[13] Although the O2/Ar ratios we present are uncalibrated, they do provide an indication of the degree of biological oxygen saturation in surface seawater samples [Craig and Hayward, 1987]. Previous laboratory and field studies [i.e., Tortell, 2005b] using air-equilibrated water samples as standards have shown that O2/Ar current ratios between 12 and 15 represent an oxygen saturation of 100% over a large range of temperature and salinity conditions (greater than those observed during the present study). The range in the measured saturation ratio represents changes in ion source performance over multiple cruises and varying operating parameters (i.e., filament emission current, electron multiplier voltage), as opposed to temperature and salinity effects on the ratio. The ion source performance and operating parameters (e.g., filament emission currents and electron multiplier voltages) used in this study were within the range of previous settings during times when O2/Ar signals were calibrated to determine saturation states. Thus, we believe that O2/Ar ratios in the 12–15 range roughly represent the boundary conditions (±1 unit) for 100% O2 saturation for all past and present operating scenarios. It is important to note that this range of uncertainty is much smaller than the observed variability (∼8–24) in the O2/Ar ratio during the present study. We are thus confident that measured O2/Ar ion current ratios >16 reflect biological oxygen supersaturation, while values <11 represent undersaturation.

2.3. Hydrographic Measurements

[14] Surface temperature, salinity and chlorophyll a fluorescence data were collected in conjunction with the gas measurements using a SeaBird SBE-25 CTD continuously sampling from the same seawater intake system as the MIMS. During parts of T1 and T3, the SBE-25 was not logging data because of loss of battery power, creating gaps in the salinity and chlorophyll data along these transects (Figures 1b and 1c). Fortunately, the temperature data were available for these areas from the underway pCO2 equilibrator. The fluorescence signal was converted to a chlorophyll a concentration by a calibration coefficient determined from a linear regression against discrete chlorophyll a measurements (n = 18, r2 = 0.97, p < 0.0001). The latter were determined fluorometrically following filtration of seawater onto 25 mm GF/F filters and extraction in 90% acetone for 24 h [Parsons et al., 1984]. All hydrographic data from the underway CTD were also acquired every 30 s and aligned with the corresponding gas measurements from the MIMS. Thus, all data used in the spatial analyses were sampled at a common frequency. At a number of locations along the cruise track, CTD depth profiles were obtained using a SeaBird SBE 911+ CTD attached to a rosette sampler.

2.4. Data Binning Procedures

[15] In order to test the SD02 algorithm, average values of DMS and chlorophyll a data were calculated for each 1/4° × 1/4° surface grid sampled. Only areas where both chlorophyll and DMS concentrations were measured simultaneously were used in the analysis, thus excluding parts of T1b, T3 and all of T8. The binning procedure reduced the data ∼100-fold to 29 pairs of CHL and DMS values. A corresponding mixed layer depth (MLD) was assigned to each data pair. In most cases, MLDs were determined from CTD density profiles and were defined as a 0.125 σt change from the surface value in accordance with the criteria of the SD02 algorithm. In QC Sound (T1b-5; Figure 1a) where we had good CTD coverage, linear interpolation was used where necessary to assign an MLD value to each grid on the basis of neighboring CTD profiles. Because of the narrow range of MLDs in this region (6–12 m), any error introduced by this interpolation was small.

[16] CTD data were not available during our cruise in the vicinity of T1a, T6 and T7. Strong tidal forcing and turbulent flows, rather than wind or buoyancy fluxes control MLDs at the head of QC Strait (T1a), such that the upper mixed layer is consistently deep (40–80 m, P. Cummins, personal communication, 2006). We used horizontal gradients in surface temperature and pCO2 to identify these tidally mixed regions, assuming that higher pCO2 and colder temperatures were indicative of deeper mixing. This estimate yielded a linear horizontal gradient in MLDs decreasing from 80 m at the head of QC Strait to 50 m toward the mouth. For T6 and T7, MLDs were estimated on the basis of 26 CTD profiles taken in the same area three weeks later (D. Mackas, unpublished data, 2004). Mixed layer depths during the latter survey ranged from 8 to 24 m, and fell within the 95% confidence interval of average summertime MLDs determined from several decades worth of CTD data collected in that area [Thomson and Fine, 2003]. Because of the dense spatial CTD coverage of the latter survey, the largest uncertainty in the MLD estimates for T6 and T7 arises from the time lag between the two cruises. However, given that winds were weak to moderate prior to, and following our occupation, MLD estimates for T6 and T7 are likely accurate, and we estimate a maximum error of 30%. Once a mixed layer depth was assigned to each 1/4° × 1/4° square, we calculated the expected DMS concentration using the CHL/MLD ratio and the appropriate equation from SD02 (see section 3.3).

2.5. Spatial and Statistical Analysis

[17] We examined the spatial variability of the gas and hydrographic data using two different statistical approaches. Lagged autocorrelation functions were computed to estimate decorrelation length scales (DLS) for the various parameters as described by Murphy et al. [2001]. The DLS gives an indication of the spatial length scale at which measurements become statistically independent of each other (i.e., uncorrelated). We also estimated the errors that could result from undersampling highly variable surface data, by calculating interpolation errors as described by Hales and Takahashi [2004]. Essentially, this approach calculates the average error obtained by resampling high-resolution data sets with increasingly lower resolution. As sampling resolution decreases (i.e., the distance between sampling points increases), the interpolation error increases up to an asymptotic value (see Hales and Takahashi [2004] for a full description of the approach).

3. Results

3.1. Surface Gas and Hydrographic Distributions

[18] The surface distributions of temperature, salinity and chlorophyll a are shown alongside pCO2, O2/Ar and DMS data from the MIMS in Figures 1a1f, respectively. All parameters exhibited large ranges highlighting the dynamic nature of the study region. Surface temperatures ranged from 10.0 to 18.6°C (avg. 16.2°C; Figure 1a), salinity ranged from 24.2–32.3 psu (avg. 31.4; Figure 1b), while chlorophyll a concentrations ranged from <0.1–33.2 μg L−1 (avg. 2.6 μg L−1; Figure 1c).

[19] Gas distributions over the study region were also highly variable. The partial pressure of CO2 during this survey ranged from undersaturated values as low as 201 ppm to highly supersaturated at 747 ppm with an average of 362 ppm. The surface maps show spatial associations between the distribution of pCO2 and that of several physical and biological variables (Figures 1a1d). Although CO2 concentrations were below atmospheric equilibrium values over most of the study region, they were strongly undersaturated in areas that had high chlorophyll concentrations, indicative of high phytoplankton biomass, and presumably a strong biological CO2 sink (Figures 1c and 1d). These low pCO2 areas are evident in Figure 1d along T1a-b in QC Strait, and along T6 and T7 in what is likely the JdF eddy. In contrast, regions of high pCO2 occurred in JdF Strait (T8) and at the head of QC Strait (T1a), areas influenced by strong tidal mixing which brings deep waters enriched in respired CO2 to the surface. Along T1a in QC Strait we observed a dramatic transition from supersaturated to undersaturated conditions characterized by an almost 500 ppm drop in pCO2 levels (747 ppm to 255 ppm) in the span of 26 min or a distance of 8 km (Figure 1d). Smaller regions of CO2 supersaturation also occurred to a lesser extent when the cruise track crossed the shelf break in areas of localized upwelling (e.g., T5, T7; Figure 1d). The physically induced changes in pCO2 are corroborated by the corresponding low temperatures in these regions which reflect deep waters mixing to the surface (Figure 1a). Our subsampling of the region yielded an average pCO2 value of 362 ppm suggesting that at the time of our survey there was little net flux of CO2 to or from the ocean, despite local areas of large pCO2 disequilibria.

[20] O2/Ar ratios expressed as a ratio of the m/z 32/40 ion currents from the MIMS ranged from 7.9 to 23.5 with an average of 15.1 (Figure 1e). High O2/Ar ratios in our survey region (>16) coincided with areas of elevated chlorophyll concentrations and low pCO2 suggesting apparent biologically induced oxygen supersaturation (T1, T5, T7; Figures 1c1e). A large area of low O2/Ar values occurred in JdF Strait (T8) concurrent with the elevated pCO2 levels of this tidally mixed zone (Figures 1d and 1e).

[21] DMS concentrations ranged from undetectable (<1 nM) to 28.7 nM with an average of 5.8 nM (Figure 1f). Areas of high DMS levels were confined to QC Sound and corresponded to regions of moderate chlorophyll a levels (Figures 1c and 1f). In contrast to the other gases, DMS concentrations showed no broad coherence with any single biological or physical parameter across the entire survey region.

[22] Although the maps of near-surface gas distributions shown in Figure 1 provide a general overview of the sampling area, they obscure much of the fine-scale structure revealed by the high-resolution MIMS data. Figure 2 shows an expanded view of T5 off the west coast of Vancouver Island that highlights the covariance of the gas and hydrographic data in greater detail. The highest DMS concentrations encountered during the survey occurred along this 8-h transect. Interesting physical features are also apparent in this high-resolution view, including a region that appears to be influenced by localized upwelling just south of 50.8°N. This area is illustrated by a large and sudden drop in temperature (ΔT > 4°C), a slight increase in salinity, a dramatic spike in pCO2pCO2 > 100 ppm) and an associated decrease in the O2/Ar ratio (2.5 unit decline). Moving northward from this region, both DMS and chlorophyll concentrations increased (in conjunction with increasing temperatures) until the two variables became uncoupled at 51.0°N (Figure 2a). From this point, DMS continued to increase despite a drop in chlorophyll levels. Another sharp temperature front was encountered at the northernmost section of the transect, where a sudden increase in temperature was associated with a rapid, large (>20 nM) drop in DMS. This decline in DMS levels occurred in under 20 min, over a short distance (∼5.5 km), but was represented by ∼40 individual data points (Figure 2a). This feature could not have been adequately resolved with traditional sampling methods. The high-resolution data reinforce the observation that DMS appears to vary independently of the other parameters over large spatial scales, while exhibiting synchronous change with various physical and biological parameters across smaller-scale features such as sharp mixing fronts. In contrast, pCO2 and O2/Ar showed a striking anticorrelation along the entire sampling transect at even the finest resolution, reflecting the coupling of these gases through photosynthesis and respiration (r2 = 0.89, p < 0.0001, Figure 2b).

Figure 2.

Detailed south–north view of all variables measured along T5 (see Figure 1a for location); (a) DMS (thick line), chl a (thin line); (b) O2/Ar (thick line), pCO2 (thin line); (c) salinity (thick line), temperature (thin line).

3.2. Covariance of Gas and Hydrographic Data

[23] Figure 3 illustrates the relationship between pairs of variables for the entire data set. The strong anticorrelation between pCO2 and O2/Ar that was obvious for T5 (Figure 2b) is also apparent for the pooled data set (Figure 3a). The high coefficient of determination for the linear regression between these two variables (slope (a) = −0.020 ± 0.001, intercept (yo) = 22.6 ± 0.042, r2 = 0.90, p < 0.0001; Figure 3a) indicates that despite a lack of calibration, the O2/Ar signal is providing meaningful information on biologically induced oxygen saturation. Overlaid on this scatterplot is the chlorophyll concentration. The highest chlorophyll is associated with the highest O2/Ar values and the lowest pCO2 values, while low chlorophyll occurs over a larger range of O2/Ar and pCO2 values (Figure 3a). A trend is also evident between O2/Ar and chlorophyll although there is considerably more scatter around this relationship (Figure 3b). From the temperature data overlaid on this plot it is clear that the main deviations from linearity occur at cold surface temperatures that represent water masses from the tidally mixed JdF and QC Straits (purple and blue regions, Figure 3b). These upwelled waters bring with them the low O2/Ar signatures of deep water that has been subject to oxygen loss due to respiration. They thus possess anomalously low O2/Ar ratios compared to waters with the same chlorophyll concentrations that have been at the surface longer and had time to equilibrate to the atmosphere and the surrounding biology (Figure 3b). Exclusion of all the data at temperatures below 13.0°C yields a linear correlation between O2/Ar and chlorophyll levels in these “aged” surface waters (a = 0.201 ± 0.002, yo = 15.5 ± 0.017, r2 = 0.73, p < 0.0001; see section 4). In contrast, DMS concentrations were not related to chlorophyll levels (Figure 3c), nor to any other individual variable across the entire survey region. Concentrations of this gas were low in regions of both very high and low chlorophyll biomass and in both warm and cold waters (Figure 3c).

Figure 3.

Scatterplots of pooled data from all transects showing the relationship between individual variables across the entire survey region: (a) pCO2 and O2/Ar with corresponding chl a concentrations overlaid (color bar), (b) chl a and O2/Ar with corresponding temperature overlaid (color bar), and (c) chl a and DMS with corresponding temperature overlaid (color bar).

3.3. A Test of a Predictive DMS Algorithm

[24] Not surprisingly, no simple correlative analysis was capable of explaining the fine-scale distribution of DMS in our survey. Of all the variables, chlorophyll covaried with DMS along parts of some transects (Figure 2a), but was a poor predictor of DMS in general. We also observed areas where DMS changed sharply with temperature and salinity across fronts (Figures 2a and 2c), suggesting a direct, or indirect, physical influence on DMS concentrations, but neither temperature nor salinity were good predictors of DMS levels in the full data set. SD02 successfully used a CHL/MLD ratio as a combined biological/physical predictor variable for DMS. Their study, along with previous ones aimed at defining a predictive DMS algorithm used coarsely sampled bottle measurements which effectively eliminate much of the spatial variability in DMS concentrations. Since MLD measurements were not available at the resolution of our underway data, and to compare our analysis with that of SD02, we greatly reduced the spatial resolution of our data set by averaging measurements into larger spatial units (∼25 km). With this coarser data set, we were able to test the applicability of the CHL/MLD ratio in the coastal waters of our survey.

[25] Following the binning procedure (see section 2), we applied the appropriate equation to calculate the predicted DMS concentration on the basis of the magnitude of the CHL/MLD ratio. In all cases, this ratio was greater than 0.02 and we thus used equation 2 of SD02 that linearly relates DMS to the CHL/MLD ratio according to

equation image

[26] In all but two cases, the SD02 equation overestimated the DMS concentration by a factor of at least 2, resulting in a poor fit to the data (Figure 4, dashed line). However, when the actual DMS concentrations were plotted against their corresponding CHL/MLD ratios, we observed a good linear fit (n = 27, r2 = 0.83, df = 26, p < 0.0001; Figure 4), after excluding two significant outliers from the regression. The resultant fit to our data is

equation image
Figure 4.

Average DMS concentrations plotted against CHL/MLD ratios for the 1/4° × 1/4° grids. The dashed line represents the predicted DMS concentration based on equation 2 of SD02 (DMS (nM) = 55.8 ± 10.8 × [CHL/MLD (mg m−4)] + 0.6 ± 0.9). The solid line is the linear regression of the actual data with a = 21.0 ± 1.87, yo = 0.11 ± 0.62, r2 = 0.83, n = 27, df = 26, p < 0.0001; Filled symbols represent data points with measured MLDs; open symbols represent data points with estimated MLDs (see section 2). The two outliers on the far right were excluded from the regression.

[27] This slope is less than half that of the SD02 formulation. The fit is possible despite the fact that many of our chlorophyll values exceeded the 15 μg L−1 cutoff of SD02 algorithm and mixed layer depths for some of the points were estimated using data from other cruises (open symbols, Figure 4; see section 2). However, we must add the caveat that our data do not meet the strict criteria for linear regression as they exhibit heteroskedacity. We used a linear fit to the data to make our results directly comparable to the SD02 relationship, but note that the fit to the latter data set is more robust. This fact, along with the presence of the two outliers (and the exclusion of high chlorophyll values by SD02) highlights the difficulty in using empirical relationships to predict DMS in regions of high productivity and large heterogeneity, such as coastal margins. This difficulty further emphasizes the importance of high-resolution sampling in these areas. Despite these challenges, however, it is clear that a different scaling factor exists between DMS and CHL/MLD in our coastal data set compared to open ocean waters.

3.4. Spatial Analyses

[28] In addition to examining the covariance of the measured parameters, we used our high-resolution data set to quantify the length scales of variability of the gas and hydrographic data. We computed lagged autocorrelation functions (ACF) along individual transects from which the decorrelation length scales (DLS) were defined as the first zero crossing of the function. Transects T1 and T3 were excluded because of gaps in the hydrographic data (see section 2). A summary of the DLS for each variable along each of the six major transects analyzed is presented in Table 1. DLS were only calculated where all six parameters were measured simultaneously ensuring an equivalent number of data points along each transect. Although on average the gases appeared to vary on shorter spatial scales than the hydrographic data, the differences were not statistically significant (see section 4).

Table 1. Decorrelation Length Scales for Each Parameter Along the Six Major Transects
TransectLength, kmTSChl aO2/ArpCO2DMS
  • a

    Dimethylsulfide (DMS) has the shortest mean decorrelation length scales (DLS: in kilometers) of all six parameters even when transect 8 is excluded from the analysis for all variables. The differences in the means are not, however, statistically significant because of the strong dependence of DLS on total transect length.
mean DLS 12.013.612.

[29] An additional analysis was used to estimate the errors that could result from low-resolution sampling. Following the work of Hales and Takahashi [2004], we resampled our high-frequency data at ever coarser resolution, and calculated the average error resulting from linear interpolations between the coarsely sampled data. As the sampling frequency decreases, the interpolation error increases to an asymptotic value. Beyond this point, coarser sampling has little to no effect on the magnitude of the interpolation error [see Hales and Takahashi, 2004, Figure 19].

[30] Table 2 summarizes the asymptotic interpolation errors calculated for each of the six parameters along the six major transects surveyed. Frontal regions evidently had significant effects on the magnitude of the asymptotic interpolation error for all parameters. Transect 5 crossed a local upwelling zone and was characterized by two sharp temperature fronts (Figure 2c). Consequently, along this transect the asymptotic interpolation errors for most parameters were large, particularly for DMS which changed dramatically at frontal zones (Figure 2a). The asymptotic interpolation error for DMS in this case was almost 100% of the 8.6 nM mean concentration along this transect (Table 1). In contrast, T7 and T8 were characterized by large, sharp gradients in chlorophyll concentrations (Figure 1c) which drove high variability in pCO2 and O2/Ar levels. Mixing of deep waters to the surface along other parts of these transects further increased the ranges of these gases (∼450 ppm change in pCO2 along T8). As a result, the absolute errors for chlorophyll, pCO2 and O2/Ar were largest along T7 and T8 (Table 2).

Table 2. Absolute and Relative Asymptotic Interpolation Errors Along the Six Major Transects
TransectAbsolute Interpolation ErroraRelative Interpolation Error, %b
T, °CS, psuChl a, μg L−1O2/ArpCO2, ppmDMS, nMT, °CS, psuChl a, μg L−1O2/ArpCO2, ppmDMS, nM
  • a

    Absolute asymptotic interpolation error represents the maximum error approached (in the respective units of each parameter), when the data are resampled at increasingly lower resolution.

  • b

    Relative asymptotic interpolation errors are normalized to the mean value for each parameter.

mean error0.740.142.61.0463.650.56671261

4. Discussion

4.1. Spatial Scales of Variability

[31] Biogeochemical variability in this coastal zone was observed both in the large range of values measured and in the short distances over which gas and surface water hydrography varied. The ACF shows that on average, the majority of the variability in the gas, temperature, salinity and chlorophyll distributions occurred on spatial scales of less than 20 km (Table 1), consistent with the expected Rossby radius (the correlation length for physical properties, [Cushman-Roisin, 1994]) in this region. Although the mean DLS seemed to be shorter for the gases (<10 km), than for the hydrographic parameters (12–14 km), the differences were not statistically significant owing to large variability in the DLS. However, much of this variability appears to result from an artifact of the analysis in which the DLS for any given parameter increases with increasing transect length (data not shown). Thus, part of the variability in the mean DLS for each parameter is due to variability in the transect length, making comparison of DLS for a given parameter difficult between transects and even between studies [i.e., Murphy et al., 2001; Hales and Takahashi, 2004; Tortell, 2005b]. Nonetheless, variability in DLS between parameters along any single transect is meaningful, and the overall mean DLS values are likely representative of inherent differences in the length scales of variability of these parameters (Table 1). Thus it appears that the gases do vary on shorter length scales than temperature, salinity and chlorophyll, with DMS possibly exhibiting the shortest DLS as suggested by previous data [Tortell, 2005b]. In any case, the results clearly indicate that gases exhibit variability over distances that are not sufficiently resolved in many field studies.

[32] The asymptotic interpolation errors show that for DMS and chlorophyll in particular, insufficient sampling resolution can introduce errors approaching 100% of the mean concentration. Large errors (∼20%) are also associated with pCO2 and O2/Ar measurements (Table 2). Although the relative error for pCO2 and O2/Ar is smaller than that for DMS or chlorophyll a, a 46 ppm average absolute error in the mean pCO2 concentration would be highly significant if accurate estimates of regional air-sea fluxes were required, and could even change the direction of the flux. The same is true in the case of O2/Ar, where a unit change in the ratio can represent a 5–10% change in the saturation state of O2 [Tortell, 2005b]. This uncertainty would have large implications for estimates of net community productivity from O2/Ar ratios [as in Kaiser et al., 2005]. While the asymptotic interpolation errors represent maximum uncertainty associated with low-resolution sampling, e.g., hydrostations with separations of 20–60 km, significant errors can also occur during underway sampling at insufficient resolution. For a more meaningful application of the analysis we calculated actual sampling errors for our underway pCO2 and DMS measurements resampled at the frequency of the pCO2 equilibrator (5 min), and a typical underway DMS sampling frequency of 30 min. For the pCO2 equilibrator, the interpolation errors along individual transects ranged from 1.4 to 8.3 ppm, equivalent to a maximum error of 1.5% relative to the mean. This relatively small error would not be particularly significant for most biogeochemical studies, and it thus appears that the pCO2 equilibrator is capable of capturing nearly all of the underlying variability. In contrast, a 30-min sampling regime for DMS introduced errors of between 0.7 and 2.8 nM, or as high as 41% of the mean along a given transect. This error in the mean DMS concentration translates to an equivalent error in the DMS flux, an estimate already prone to large errors because of the uncertainty in the gas transfer coefficient [Nightingale et al., 2000].

4.2. The pCO2 and O2/Ar Distributions

[33] The distributions of pCO2 and O2/Ar during this survey exhibited considerable patchiness with regions of strong undersaturation and supersaturation in close proximity to each other. The range of pCO2 concentrations encountered (201–747 ppm) was similar to other summertime values reported for the southwest coast of Vancouver Island [Ianson et al., 2003], and the Oregon coastal upwelling system [Hales et al., 2005]. In all three surveys, regions of intense CO2 supersaturation were confined to narrow nearshore strips with the majority of the outer shelf areas undersaturated as a result of biological drawdown. The large regions of CO2 undersaturation found along the shelf (T6, T7) and in QC Sound (T1b-5) may have compensated for the locally intense CO2 sources identified in the straits (T1a, T8). However, because of the large disparity in supersaturated and undersaturated waters coupled with incomplete spatial coverage, our measurements cannot yield a comprehensive CO2 budget for this region. It is important to recognize that these tidally influenced regions of persistently high CO2 must have a large impact on net annual carbon budgets, whereas localized upwelling along the shelf leads to transiently high CO2 concentrations that are usually quickly drawn down by biological production. Although there is still considerable debate as to whether coastal upwelling zones are net sources or sinks of CO2 over an annual cycle [Ianson and Allen, 2002; Hales et al., 2005], Hales et al. [2005] suggest that CO2 uptake in the upwelling margins along the west coast of North America is equivalent to 50% of the entire summertime, oceanic North Pacific CO2 sink. These findings clearly demonstrate the disproportionately large influence of coastal margins in the oceanic carbon cycle and the need to incorporate these areas into global CO2 climatologies.

[34] There is much current interest in identifying the relative importance of biological processes versus physical ones in driving the distribution (and hence air-sea flux) of CO2 [Sarmiento et al., 2000]. Our results demonstrate that the opposing biological processes of photosynthesis and respiration strongly influenced the large variability in surface pCO2 in this productive, coastal margin. This influence is evident from the tight anticorrelation between pCO2 and O2/Ar along all sampling transects even in the finest resolution (Figures 2b and 3a), and the occurrence of low pCO2 levels in regions of high biomass along certain transects (e.g., T1b, T7, Figures 1c and 1d). By comparison, such a tight relationship between pCO2 and O2/Ar would not be expected if surface water CO2 drawdown and/or efflux were largely attributable to a temperature and salinity-dependent solubility pump. Clearly, physical processes play a role in CO2 distributions through lateral and vertical water transport, but our results indicate that biological processes largely set the gas content of water masses in this region.

[35] Atmospheric exchange also affects the distribution of pCO2 in surface waters although its effects are not as pronounced as those of mixing and biological drawdown. This is due in part to the slow rate of CO2 exchange which occurs on timescales of days to weeks and is about 10 times slower than the rate of O2 exchange [Broecker and Peng, 1982]. These differential gas exchange rates may explain the divergent slopes of the pCO2 versus O2/Ar relationship on the left side of the plot in Figure 3a. Here, pCO2 concentrations of ∼200 ppm are associated with two very different levels of chlorophyll biomass and O2/Ar ratios. At the top of the plot, the highest chlorophyll levels (red area, Figure 3a) correspond to the highest O2/Ar ratios and lowest pCO2 values measured, yet lower down on the y axis, the same pCO2 levels are associated with much lower chlorophyll and O2/Ar levels (Figure 3a). Thus it appears that the O2 concentrations in these latter waters were able to reequilibrate to the present phytoplankton biomass much faster than CO2 levels. Because of the longer “history” of the pCO2 signature, a strong anticorrelation between pCO2 levels and chlorophyll concentrations was only evident in areas where biomass was high, presumably where productivity rates were at their peak.

[36] The O2/Ar ratio has been used as a proxy for the net community production of a water mass integrated over a timescale that depends on the gas transfer velocity and the depth of the mixed layer [Kaiser et al., 2005]. We observed a linear relationship between O2/Ar and chlorophyll levels in surface waters indicating increasing net production with increasing phytoplankton biomass (Figure 3b). However, several factors complicate the relationship between chlorophyll and O2/Ar. On the one hand, deep mixing brings up cold subsurface waters with the low O2/Ar values characteristic of oxygen loss due to respiration. These newly upwelled waters thus carry an oxygen deficit (compared to surrounding surface waters), that persists despite the growth of phytoplankton and the addition of photosynthetically derived oxygen. This scenario may explain the disparity in O2/Ar values for a given chlorophyll concentration, with much lower O2/Ar values found in cold surface waters (purple and blue regions in Figure 3b). Our data thus confirm the difficulty of estimating net community productivity from O2/Ar ratios in regions with significant upwelling or vertical mixing (as by Kaiser et al. [2005]). In contrast, high O2/Ar ratios may persist in the mixed layer despite the presence of low surface chlorophyll levels (red regions in Figure 3b). These waters may have warmed and stratified to the extent of cutting off the supply of nutrients from below the mixed layer and causing the phytoplankton to be removed from the surface before the O2/Ar productivity signature could be reset by atmospheric exchange. Thus knowledge of the relevant timescales of both gas exchange and phytoplankton turnover rates is critical when inferring production rates from gas distributions.

4.3. Factors Driving DMS Distributions

[37] The range of DMS concentrations encountered during our survey was large and variable, from undetectable (<1 nM) to almost 30 nM. The upper end of this range is at least an order of magnitude higher than current estimates of the global mean DMS concentration [Belviso et al., 2004b], and comparable to data obtained in various other coastal regions [Leck et al., 1990; Townsend and Keller, 1996; Locarnini et al., 1998; Tortell, 2005b]. However, our measurements provide much higher spatial resolution than previous surveys and thus offer new insight into the small-scale patchiness of DMS distributions in coastal waters (Figure 1f).

[38] We compared our DMS data to measurements taken in the same general area (coastal B.C. between 48 and 55°N), during the same week, the previous year by automated purge-and-trap gas chromatography sampling at a frequency of once every 30 min (J. E. Johnson, unpublished data, available at http://saga.pmel.noaa.gov/dms/). The range of DMS concentrations encountered in 2003 (0.7–17.6 nM) was smaller than that observed during our 2004 survey (<1–28.7 nM), but the mean DMS concentrations were remarkably similar (5.3 nM versus our value of 5.8 nM). It is possible that the range of concentrations during 2003 was lower because Johnson's survey was unable to resolve the true variability of the DMS distributions (as indicated by our 30-min interpolation error analysis). As observed previously [Hales and Takahashi, 2004; Tortell, 2005b], we note that the measured range of a given parameter is much more prone to sampling errors than the regional mean. It is thus not surprising that the overall mean of Johnson's survey was similar to that of the present survey, despite a much smaller range of reported DMS concentrations. However, many environmental factors could potentially account for the interannual variability in DMS concentrations. These include differences in wind speed which would reduce ventilation to the atmosphere (winds were much lighter during our survey in 2004, rarely exceeding 5 m s−1), and in the rates of photolysis and bacterial consumption both of which can be much larger sinks for DMS [Kiene and Bates, 1990; Kieber et al., 1996]. In addition, underlying differences in phytoplankton species composition and biomass, zooplankton grazing rates, nutrient supply and light intensity could all account for interannual variability in DMS concentrations. Despite these potential differences, we observed a striking similarity in the spatial distribution of DMS between the 2 years. Both the 2003 survey and our 2004 survey reported peak DMS levels in the region north of Vancouver Island between 51 and 52°N. The area of summer relaxation over the broad shelf in QC Sound may thus be a region of persistently high DMS during August.

[39] The high DMS concentrations (>10 nM) observed in QC Sound during our survey were associated with moderate levels of chlorophyll (T1b, T4, T5; Figures 1c and 1f). In contrast, DMS concentrations were low in areas where chlorophyll levels exceeded 15 μg L−1 (T1a, T7; Figures 1c and 1f). Since DMSP production by phytoplankton is known to be highly species-specific [Keller et al., 1989], variability in phytoplankton community composition may account for differences in DMS levels between regions. This taxonomic effect is potentially confounded, however, by differing environmental conditions across our sampling regions. Since DMSP and its byproducts are powerful antioxidants, cells increase their DMSP (and DMS) production when exposed to oxidative stressors such as low nutrients and high UV light [Sunda et al., 2002]. These conditions generally occur in later stage blooms when surface waters have stratified, cutting off the supply of nutrients from below the thermocline and exposing cells to higher levels of UV light. Stratified waters tend to favor flagellate groups such as haptophytes that have adapted to living under low-nutrient, high-light conditions [Margalef, 1978] and are perhaps noncoincidentally the same groups that are prominent DMSP producers [Keller et al., 1989]. Furthermore, nutrient limitation can induce DMSP production in groups such as diatoms [Sunda et al., 2002; Bucciarelli and Sunda, 2003], that have traditionally been considered low DMSP producers [Keller et al., 1989]. The high DMS concentrations in QC Sound occurred where mixed layer depths were less than 12 m and surface nitrate was depleted, suggesting an oxidative stress effect on DMSP/DMS production consistent with a later stage phytoplankton bloom. This scenario may explain the decoupling of DMS concentrations and surface chlorophyll levels north of 51°N along T5 (Figure 2a) that likely coincided with a shoaling mixed layer possibly due to freshwater input (as seen from the declining salinity trace, Figure 2c). Increased DMSP/DMS production coincident with the decline of the bloom and the sinking of the phytoplankton could explain the increase in DMS levels despite the drop in surface chlorophyll. Note that a decoupling between surface DMS and chlorophyll can also occur with the formation of a subsurface chlorophyll maximum that would not be measured during underway surface sampling.

[40] In contrast to the stratified waters of QC Sound, the waters in QC Strait and in the JdF eddy (T1a, T7) were receiving a steady nutrient supply through localized upwelling or deep mixing resulting in high (or measurable) surface nitrate and high (likely diatom-dominated) phytoplankton biomass, with little DMS production. The occurrence of low DMS levels in recently upwelled waters has been observed previously [Belviso et al., 2003]. Since changes in phytoplankton species composition generally occur in conjunction with varying environmental conditions, it is difficult to distinguish between species composition effects and nutrient/light effects in determining DMS levels. As in previous studies [Locarnini et al., 1998; Tortell, 2005b], DMS concentrations during this survey changed dramatically at fronts, regions where abrupt gradients in nutrient concentrations, light regimes, productivity and plankton community composition are common. This trend reflects the complex interplay of physics and biology that characterizes the oceanic DMS cycle [see Simó, 2004].

4.4. Toward Global DMS Prediction

[41] Intense efforts have been aimed at understanding the factors controlling DMS production in the oceans, and quantifying its atmospheric flux in order to evaluate the feasibility of the hypothesized biologically mediated homeostasis (CLAW hypothesis) [Charlson et al., 1987]. It has proven difficult to project future oceanic DMS emissions in a changing climate because of an incomplete mechanistic understanding of the DMS cycle and uncertainties in the global and seasonal distributions of this gas. These uncertainties coupled with uncertainties in the gas transfer coefficient [Nightingale et al., 2000] hamper the ability of atmospheric sulfur models to evaluate the modulating effect of this gas on global climate.

[42] Two complementary approaches are needed to better constrain the global distribution of DMS. Accurate, spatially resolved global DMS measurements with good seasonal coverage are the first step, and MIMS can greatly facilitate this endeavor. Even with an automated MIMS system, it would still be unfeasible to map the entire ocean at sufficient temporal resolution. Hence, the second approach involves developing predictive algorithms that simulate DMS levels on the basis of well-constrained biogeochemical parameters. A recent comparison of several such algorithms has shown that they have various strengths and weaknesses and differ in their ability to accurately replicate DMS levels both seasonally and regionally [Belviso et al., 2004b].

[43] We chose to evaluate the SD02 algorithm with our coastal data set because it is one of the few without a modeled term, and it relies on two commonly measured parameters (chlorophyll and MLD). Since DMS originates from phytoplankton-produced DMSP, some dependence on chlorophyll (an indicator of algal biomass) is expected. However, DMSP production by phytoplankton is both taxon-specific, as well as dependent on the physiological status of the cells, and thus not simply correlated to bulk chlorophyll. Moreover, the release and breakdown of DMSP to DMS involves the entire foodweb which itself is strongly influenced by the physicochemical context. The mixed layer depth determines the extent to which planktonic organisms in surface waters are exposed to turbulence, nutrients and light. By extension, it strongly influences both phytoplankton and bacterial growth rates and species composition, and thereby DMSP production and its subsequent breakdown [Simó and Pedros-Alio, 1999]. For example, shallow MLDs favor the growth of prolific DMSP producers, but also inhibit bacterial growth rates through high UV light exposure. This combination leads to high DMS production as the reduced sulfur supply exceeds the bacterial sulfur demand [Kiene et al., 2000]. In contrast, when bacterial growth is not inhibited by UV damage, much of the sulfur in DMSP is incorporated into bacterial biomass, thus lowering the DMS yield. Furthermore, independent of any biological considerations, surface DMS levels may be inversely related to the thickness of the mixed layer by a simple dilution model, such that DMS levels are high where MLDs are shallow and vice versa [Aranami and Tsunogai, 2004].

[44] Although the original SD02 algorithm failed to recreate the large-scale DMS levels found during our survey, with a different slope the CHL/MLD ratio may be useful as a predictor of coarse resolution surface DMS concentrations (Figure 4). Our slope is less than half that of the original algorithm which may result from the higher relative proportion of diatoms (with lower DMSP content per unit chlorophyll) [Keller et al., 1989] in coastal regions compared to oceanic waters. Our coastal data encompassed a greater than 10-fold larger range of CHL/MLD ratios (0.03–2.26) than the open ocean data originally used to formulate the algorithm (<0.20), and our observed DMS concentrations were ∼2-fold higher. Although we did not explicitly measure MLDs outside the area of QC Sound, we chose to incorporate the whole survey region to expand the ranges of MLD, DMS, and chlorophyll concentrations used to test the algorithm. This addition was particularly important since the linear regression was strongly influenced by the data points with high CHL/MLD ratios that originated from the region of high DMS in QC Sound where the range in MLD was small. While the relationship between chlorophyll and DMS for the full high-resolution data set was stronger in this area (r2 = 0.58) than for the entire survey region (Figure 3c, r2 = 0.06), it was nonetheless further improved by incorporating the MLD. With the modified slope, we observed a reasonable fit between the DMS data and the CHL/MLD estimates up to a CHL/MLD ratio of 1.0. The two prominent outliers that were excluded from the regression had CHL/MLD ratios greater than 1.0 and came from the region of high biomass along T7. Although there is potential error in the MLDs for the two outliers, we are confident that it is small. On the basis of the measured DMS and chlorophyll concentrations, MLDs at these sites would have had to have been unreasonably deep (75–100 m) in order to fit the curve. It is possible that the linear relationship simply does not hold beyond CHL/MLD ratios greater than 1.0 which would characterize rare regions of very high biomass and relatively shallow mixed layers. The area where these high CHL/MLD ratios were observed was likely part of the JdF eddy, where exceedingly high phytoplankton biomass is sustained throughout the summer through the constant injection of nutrients into the eddy core [Marchetti et al., 2004]. The SD02 algorithm uses either a linear or a logarithmic equation to model DMS concentrations depending on the magnitude of the CHL/MLD ratio. Our results indicate a modified linear equation works in the CHL/MLD range of 0.02–1.0 but not beyond. While the CHL/MLD ratio proved capable of simulating broad-scale, coarsely resolved patterns of DMS, it appears that this predictive power breaks down at very fine spatial scales particularly at frontal zones (i.e., T5, Figure 2a). This indicates that fine-scale measurements are indeed necessary to fully capture DMS variability in dynamic regions, and argues for a better understanding of the processes affecting oceanic DMS distributions across a wide range of spatial and temporal scales.

5. Conclusions

[45] Our results emphasize the utility of membrane inlet mass spectrometry for resolving spatial variability in gas distributions, and the need for high-frequency sampling in coastal waters if the aim is to accurately quantify fluxes of both major and trace gases in these important regions. The application of MIMS (or other comparable analytical tools) is thus likely to significantly increase our understanding of oceanic gas distributions over the coming decade. As more information becomes available on the concentration and variability of gases in dynamic coastal regions, new insight may be gained into the biological and physical controls on gas distributions. In the case of the DMS cycle, a better mechanistic understanding of the underlying production and consumption processes is required. The coupling of focused process studies with real-time underway surveys is needed to move toward this understanding. Ultimately, this information will aid in the development of better predictive algorithms which are needed to understand the potential impact of DMS emissions on future climate, and conversely, the impact of climate change on the oceanic cycle of DMS. As a first step, our findings extend the utility of the CHL/MLD ratio as a predictor of DMS levels into highly productive, temperate coastal waters, during the summer season. Similar analyses in various coastal waters over different seasons are of course necessary to test the general applicability of this derived algorithm.


[46] We would like to thank the Captain and crew of the CCGS John P. Tully. In addition we thank D. Tuele, M. Robert, and D. Anderson for technical assistance at sea. We are grateful to M. Buckley and R. Stannard at Hiden Analytical for their tireless efforts in the development of the MIMS. D. Mackas kindly provided unpublished CTD data. The comments from two anonymous reviewers greatly improved the final manuscript. This work was supported by NSERC grants to NN and PDT, and financial support from Fisheries and Oceans Canada to DI.