Annual cycle of air-sea CO2 exchange in an Arctic Polynya Region



[1] During the Canadian International Polar Year projects in the Cape Bathurst polynya region, we measured a near–complete annual cycle of sea surface CO2 (pCO2sw), atmospheric CO2 (pCO2atm), sea surface temperature (SST), salinity (S), and wind speed (U). In this paper, we combine these data with ancillary measurements of sea ice concentration (Ci) to estimate the mean annual (September 2007–September 2008) air–sea CO2 exchange for the region. For the non–freezing seasons the exchange was calculated using a standard bulk aerodynamic approach, whereas during the freezing seasons we extrapolated eddy covariance measurements of CO2 exchange. Our results show that in 2007–08 the region served as a net sink of atmospheric CO2 at a mean rate of -10.1 ± 6.5 mmol m− 2 d− 1. The strongest calculated uptake rate occurred in the fall when wind velocities were highest, pCO2sw was significantly lower than pCO2atm, and ice was beginning to form. Atmospheric CO2 uptake was calculated to occur (at lower rates) throughout the rest of the year, except for a brief period of outgassing during late July. Using archival U, Ci, and pCO2sw data for the region, we found that winds in 2007–08 were 25–35 % stronger than the decadal mean and were predominately easterly, which appears to have induced a relatively late freeze–up (by ∼ 3 weeks relative to mean conditions) and an early polynya opening (by ∼ 4 weeks). In turn, these conditions may have given rise to a higher CO2 uptake than normal. Estimated winter CO2 exchange through leads and small polynya openings made up more than 50% of the total CO2 uptake, consistent with recent observations of enhanced CO2 exchange associated with open water components of the winter icescape. Our calculations for the Cape Bathurst polynya region are consistent with past studies that estimated the total winter CO2 uptake in Arctic coastal polynyas to be on the order of 1012 g C yr− 1.

1 Introduction

[2] Polynyas (areas of recurring open water that exist when a complete ice cover would be expected) are wide spread across the Arctic and Antarctic but only account for a small portion of the total area of polar seas [Barber and Massom, 2007]. Despite their limited extent, at times they provide the only direct connection between the atmosphere and the underlying ocean, making them important sites for air–sea interactions. The air–sea exchange of CO2 during time periods when polynyas occur has received considerable interest (e.g. Miller and DiTullio [2007]), but significant exchange has also been observed during other seasons [Else et al., 2011]. By tracking air–sea CO2 transfer through an entire annual cycle in geographic areas that host polynyas (areas we term “polynya regions”) we can better quantify their contribution to global and regional atmospheric carbon budgets.

[3] The potential for CO2 exchange at the air–sea interface is determined by the CO2 gradient across the interface, ΔpCO2 = pCO2swpCO2atm, where pCO2sw and pCO2atm are the sea surface and atmospheric partial pressures of CO2, respectively. To date, almost all observations of ΔpCO2 in polynya regions have been made in spring, summer, and fall, forcing annual budgets of CO2 exchange to rely on assumptions about the winter season. For instance, in the Northeast Water (NEW) polyna, Yager et al. [1995] observed low mid–summer pCO2sw caused by strong primary production earlier in the spring. They hypothesized that the region acted as a CO2 sink during the open water season with strong fall winds exploiting the negative ΔpCO2 gradient. They also predicted that ΔpCO2 would become positive in winter due to excess respiration, but that the winter ice cover would prevent any outgassing. Support for this hypothesized cycle was provided by Miller et al. [2002] who observed a brief positive ΔpCO2 before the spring opening of the Northwater polynya, followed by strongly negative ΔpCO2 as the ice cleared. Likewise, modelling [Sweeney, 2003] and summer pCO2sw measurements [Sweeney et al., 2000] in the Ross Sea polynya regions (Antarctica) have revealed a similar cycle (see also the review by Miller and DiTullio [2007]).

[4] Although the cycle described in Yager et al. [1995] has played an important role in shaping our understandings of air–sea CO2 exchange in seasonal sea ice areas, recent observations have reavealed some important deviations. In winter, ice remains mobile in many of these regions, allowing the formation of open-water leads and winter polynyas. If such regions follow the pCO2sw cycles described above, winter outgassing should occur wherever open water is found, possibly at a rate that would balance the summer/fall uptake. Instead, Anderson et al. [2004] reported water column measurements of dissolved inorganic carbon (DIC) and chlorofluorocarbons from the Storfjorden polynya (Svalbard, Norway) indicating a strong winter uptake of atmospheric CO2. Using similar data, Omar et al. [2005] calculated the rate of winter CO2 uptake in that region to be about 10 times larger than the surrounding non–polynya areas. In the Cape Bathurst polynya region (southeastern Beaufort Sea, Canada), we recently observed episodes of strong winter uptake using micrometeorological techniques (i.e., eddy covariance) [Else et al., 2011], and we were able to show that ΔpCO2 remained negative through the entire winter [Else et al. 2012a].

[5] In this paper, we use the pCO2sw data set described in Else et al. [2012a], combined with our direct observations of winter fluxes [Else et al. 2011] to calculate a full–year (2007–08) air–sea CO2 flux budget for the Cape Bathurst polynya region. In doing so, we create one of the most complete estimates of annual air–sea CO2 exchange available for any polar sea, allowing greater insight into the role of polynya regions in atmospheric carbon cycling. Our approach updates a preliminary estimate of CO2 exchange created for the region by Shadwick et al. [2011]. Finally, we examine whether the conditions experienced in 2007–08 (ice, ocean, atmosphere) represent a “typical” annual cycle for this region, and we discuss the implications of our results at the circumpolar scale.

2 Methods

2.1 Data Sources

[6] The majority of the data required for this study were collected during the International Polar Year ArcticNet and Circumpolar Flaw Lead System Study projects. The projects were conducted onboard the research icebreaker CCGS Amundsen, and are summarized in ArcticNet Inc. [2010] and Barber et al. [2010]. Multidisciplinary sampling was carried out during both projects while the ship remained mobile in the Cape Bathurst polynya region for most of an annual cycle (Figure 1). In this paper, we used data that were collected from October 22, 2007 to August 2, 2008 in western Amundsen Gulf, the primary location of ship activities (Figure 1). To create CO2 flux estimates spanning a complete annual cycle, we defined a study period of September 15, 2007 to September 14, 2008, necessitating an extension of the data by ∼ 6 weeks on either side of the field program. Various ancillary data sets were required to facilitate this extension, and fill other gaps in the field data.

Figure 1.

The study area. The dashed line outlines the approximate location of the Cape Bathurst polynya (and also outlines Amundsen Gulf). The solid polygon denotes the area in which field data were collected.

[7] Wind speed was measured by a propeller anemometer installed on an open–lattice meteorological tower located at the bow of the research vessel (see Else et al. [2011] for instrument details, including data quality control). We used hourly averages, and when ship based data were not available (usually due to ship orientation) we used hourly data collected at the Cape Parry (preferably) or the Paulatuk Environment Canada weather stations (see Figure 1). All wind speed data were scaled to a height of 10 m above sea level, and the different data sources showed reasonable agreement when compared during periods of overlap (root mean square differences of velocity on the order of 4–5 m s − 1).

[8] Ice concentration was obtained at daily intervals from the University of Bremen AMSR-E passive microwave satellite imagery products [Spreen et al., 2008]. This data product reports ice concentration on a ratio scale from 0 (ice free) to 1 (complete ice cover) at a spatial resolution of 6.25 km.

[9] Sea surface temperature (SST) was measured during the field experiment using a ship–based CTD (SeaBird 911+). The CTD was deployed over the side of the ship during the open water seasons, providing SST samples at a water depth of ∼ 2 m. During the winter the CTD was deployed through a moon pool, and the SST sampling depth was ∼ 10 m (as per Else et al. [2012a], the error associated with these different depth measurements is small). To extend the SST data set beyond the duration of the field program, we used daily L3 MODIS-Aqua daytime 11 μ SST satellite imagery.

[10] Salinity was also measured using the ship-based CTD. No supplemental measurements were available beyond the dates of the sampling program, so we used a simple interpolation based on a polynomial fit to the field data.

[11] pCO2atm was measured on the meteorological tower at the bow of the ship, using a near–infrared gas analyzer (LiCor model LI-7000) system which drew sample air from an inlet located ∼ 14 m above sea level. The instrument was calibrated regularly [Else et al., 2011; Else et al., 2012b], and data acquired when the sample air may have been contaminated by ship emissions were removed in post–processing. For the time period when the ship was not in the area, we used data from the Scripps CO2 Program sampling station at Barrow, Alaska (approximately 1000 km west of our study area) [Keeling et al., 2001]. Those data were available approximately bi–weekly during the winter and weekly through the rest of the year, and showed good agreement with our ship–based samples (root mean square error 4.1 ppm, based on 14 dates when coincident measurements were made at Barrow and on the CCGS Amundsen).

[12] pCO2sw was measured using a shower–type equilibrator connected to a short (∼ 5 m) intake line with an inlet at ∼ 5 m depth. The equilibrator's headspace air was continuously cycled through an infrared gas analyzer (LI-7000) calibrated daily with ultra–pure nitrogen (to set the instrument's zero reading) and a CO2–air mixture traceable to WMO standards (to set the span reading). A detailed description of this instrument, its components, and the reproducibility of the measurements can be found in Else et al. [2012a] and Else et al. [2012b]. To extend these data beyond the field study time period, we assumed that changes in pCO2sw between August and October are modulated by water temperature. We used measurements of pCO2sw and SST from the start of the field season to define an initial state, and then calculated pCO2sw for the 6 weeks prior to sampling using the thermodynamic equations of Takahashi et al. [1993] and the MODIS SST data. We did likewise for the 6 weeks after the end of the field sampling program. This is of course a simplification, but the results of this approach (shown in Figure 2c & e, discussed below) yield a reconstruction very similar to what might be achieved by simple interpolation between summer 2008 and fall 2007 data (i.e., the approach used by Shadwick et al. [2011]).

Figure 2.

Observations of (a) wind speed, (b) ice concentration, (c) pCO2sw (x symbols are observed daily means, open diamonds are extrapolated data as per section 2.1), (d) pCO2atm (x symbols were observed on the research vessel, open diamonds are from Barrow, Alaska), (e) sea surface temperature (x symbols were observed on the research vessel, open diamonds were obtained from the MODIS instrument), and (f) salinity (measured onboard the research vessel). For panels b-f, the solid lines are the fitted polynomials that were used for resampling the data to one–hour intervals.

2.2 Flux Calculation

[13] For any ice–affected sea it is useful to partition the net surface CO2 flux (math formula) into a component that describes the exchange between open seawater and the atmosphere (which we denote math formula) and one that describes exchange between the ice surface and the atmosphere (math formula):

display math(1)

[14] Although not often considered, observations of significant math formula have been reported using micrometeorological techniques [Semiletov et al., 2004; Zemmelink et al., 2006; Miller et al., 2011; Papakyriakou and Miller, 2011], chamber techniques [Semiletov et al., 2004; Nomura et al., 2010; Geilfus et al., 2012], and laboratory experiments [Nomura et al., 2006]. While the community is approaching consensus on the expected direction of these exchanges (CO2 outgassing during winter, CO2 uptake in the spring/summer), the observed fluxes vary by 1–2 orders of magnitude. Therefore, in this paper we limit ourselves to deriving only math formula, with the understanding that math formula could be added to create a more complete budget as the state of the science evolves.

[15] To calculate math formula, we used a standard form of the bulk flux equation, scaled by the ice concentration (Ci, reported as a fraction of the areal coverage):

display math(2)

where k is the gas transfer velocity, and α is the solubility of CO2 in seawater (a function of SST and salinity). During the non–freezing season (i.e., when air temperature exceeds SST), we used the parameterization of Sweeney et al. [2007] to calculate k:

display math(3)

where U is the wind speed at 10 m height, and Sc is the Schmidt number (a function of SST). The time period that we applied this k parameterization to was September 15 – November 14, 2007 and May 16 – September 14, 2008 (see below for explanation), and we denote those fluxes as math formula.

[16] It is important to note that this approach makes the implicit assumption that the Sweeney et al. [2007] k–U relationship (derived for the open ocean) is applicable for the conditions encountered during this time period. However, when sea ice is present there is evidence to suggest that modification of surface water turbulence (and hence k) may occur [Loose and Schlosser, 2011]. This modification may be due in part to fetch limitations imposed on open water patches by the surrounding sea ice. At low wind speeds, short fetches create steep waves [Vickers and Mahrt, 1997] which are linked to higher gas transfer [Zappa et al., 2004], but at higher wind speed, whitecap formation and bubble–mediated gas exchange becomes restricted [Woolf, 2005]. Furthermore, sea ice itself can generate turbulence via convection and current shear [McPhee, 2005], but it also may attenuate wind-driven turbulence by reflection and scattering of wind waves [Masson and LeBlond, 1989]. Fortunately, the spatio–temporal extents of conditions where these interactions are most severe tends to be limited – for example, in the Cape Bathurst polynya the spring break–up and fall freeze–up periods (when Ci is most variable) only last 7 and 3 weeks (respectively) on average [Galley et al., 2008]. Therefore, open ocean k-U parameterizations have been widely used in Arctic seas (e.g. Bates and Mathis [2009] and references therein), with the understanding that calculated fluxes await future refinement.

[17] Of more relevance to our study is the fact that standard k–U parameterizations cannot account for observations of enhanced winter fluxes (e.g. Anderson et al. [2004]; Loose et al. [2009]; Else et al. [2011]). These enhanced fluxes (which we denote as math formula) appear to be associated with the formation of new sea ice in open water [Else et al., 2011], which in the Cape Bathurst polynya region typically occurs in flaw leads throughout the entire winter. To provide a rough estimate of how strong this enhanced winter flux might be, we calculated the mean winter transfer velocity from measurements made in the Cape Bathurst polynya (math formula = 458 ± 334 cm hr − 1, [Else et al., 2011]) and used that value in equation 2. The math formula value was derived from 32 half–hour eddy covariance samples, which span a large range of mean wind speed (4 – 18 m s − 1), air temperature (-5 – -23 C), sensible heat flux (-15 – 150 W m − 2) and ΔpCO2 (-80 – -38 μatm) conditions. For this study, we calculated math formula from November 15 2007 – May 15 2008, which corresponds to the time period when SST was at the freezing point (see section 3.1), surface air temperatures were consistently lower than SST, and conditions were typically within the bounds of those experienced when collecting the kenh data.

[18] It is worth noting that we did find significant non–linear relationships between observations of kenh and wind speed (ln(kenh) = 0.16U + 3.97, r2 =0.55) and kenh and air temperature (ln(kenh) = -0.09 Tair + 4.49, r2 =0.33). Both of these relationships are consistent with our hypothesis that the flux enhancement is partly driven by wind and high sensible heat fluxes that produce a very turbulent sea surface [Else et al., 2011]. However, we decided that it was imprudent to use parameterizations based on these variables at this time given how limited the data set is, and given that causal links between these variables and kenh have not yet been proven. Instead, using the mean kenh value (and incorporating the considerable variability around that mean) gives us a first–order estimate of how strong the flux may have been through the winter.

2.3 Data Integration

[19] A key challenge for the flux calculations was the discontinuous sampling intervals inherent in some of the variables. The field data set (SST, salinity, pCO2sw, pCO2atm) was discontinuous because the ship would often leave the study area for several days to sample in other locations. The Scripps CO2 data sampling interval was also discontinuous (sampling occurred weekly through most of the year, and was less frequent during the winter), and SST was only retrievable from MODIS imagery on cloud–free days. On the other had, the wind speed and AMSR–E ice concentration data were acquired continuously at regular intervals. To deal with this issue, we fit polynomials to the pCO2sw, pCO2atm, SST, and salinity data to interpolate over periods of missing data. This approach gives a generalized view of the annual cycle of these variables and eliminates high–frequency (e.g. diurnal) variability, which has been observed to play an important role in CO2 exchange at low latitudes (e.g. McGillis et al. [2004]). However, we did not observe significant high–frequency variability in pCO2sw in this region [Else et al., 2012a], and we do not expect high–frequency variability in pCO2atm, SST, or salinity to play a strong role in influencing CO2 flux calculations. The variability that did occur in these parameters was characterized by calculating standard deviations around the polynomial fits, and the standard deviations were used in error evaluation of the resulting CO2 flux calculations.

3 Results

3.1 Observed Conditions

[20] Time series of the variables required for the flux calculations are shown in Figure 2. Wind velocities (Figure 2a) were highest between October and November 2007, with a mean speed (and standard deviation) of 7.2 ± 4.8 m s− 1 and storm episodes with sustained winds in excess of 15 m s− 1. Over the winter (December to March) winds were somewhat weaker, with mean velocities of 6.2 ± 3.7 m s− 1 and storms that rarely sustained winds in excess of 15 m s− 1. Spring (April to June) saw a slight increase in mean winds (6.5 ± 4.0 m s− 1) and similar storm activity. Winds were weakest through the open water season (July to September 2008) with mean velocities of 5.4 ± 3.0 m s− 1 and limited storm activity.

[21] The time series of Ci is shown in Figure 2b, and an animation showing Ci over the annual cycle is available in supplemental Figure S1. Ice began forming in the region around October 25, and reached 0.99 on December 10. Throughout the winter season the ice cover remained mobile, resulting in leads and small polynyas that kept Ci fluctuating between 0.95 and 0.99. Ice concentration first dropped below 0.9 in mid–April, followed by a rapid decrease that started on May 10 and resulted in ice–free conditions by June 22.

[22] The observed annual pCO2sw cycle is displayed in Figure 2c. To briefly summarize the findings of Else et al. [2012a], pCO2sw was low (∼ 300 μatm) upon arrival in October 2007, likely due to SST cooling from a summer maximum and fall phytoplankton blooms. After ice formation, pCO2sw slowly increased due to brine rejection and associated convection, respiration, and gas exchange, reaching a maximum of ∼ 380 μatm in late March. This was followed by a gradual decrease in spring caused by an ice algae bloom, and then an accelerated decrease due to open water phytoplankton blooms that reduced pCO2sw to ∼ 300 μatm in early June. In mid-July pCO2sw increased rapidly to ∼ 410 μatm as a result of gas exchange and surface warming, and then began to decrease as the sea surface cooled.

[23] Atmospheric pCO2 (Figure 2d) varied from ∼ 375 μatm in the fall to ∼ 395 μatm in mid-winter, following the expected global seasonal CO2 pattern. These pCO2atm values kept ΔpCO2 negative for most of the annual cycle, except for a brief positive excursion in mid–July.

[24] Sea surface temperature through the study period is shown in Figure 2e. In the fall, SST reached the freezing point in mid–November, and then decreased slightly through the winter as the freezing point was lowered by increasing salinity (Figure 2f). A rapid increase in SST then started in mid–May as Ci decreased (Figure 2b), reaching a maximum near 8C in mid–July. This was followed by a more gradual decrease in SST as waters cooled in the fall.

[25] Salinity followed a distinct cycle, which has been described thoroughly in Shadwick et al. [2011] (Figure 2f). At the beginning of the study period salinity was low (28–29), and then gradually increased to a maximum near 32 in April due largely to brine rejection by the growing ice cover and associated convection. As the ice began to melt in May, salinity decreased back towards fall values.

3.2 Calculated Fluxes

[26] The time series of calculated fluxes during the non–freezing season (math formula) is shown in Figure 3. The strongest fluxes (up to -40 mmol m− 2 d− 1) occurred in October 2007, when winds were highest (Figure 2a) and ΔpCO2 was strongly negative (Figures 2c & d). Significant CO2 uptake by the ocean continued through most of November, but math formula was essentially cut off towards the end of the month when Ci began to consistently exceed 0.95 (Figure 2b). In early May, math formula resumed with small episodes of uptake, and then proceeded in earnest in late May and June as the ice cover was removed by wind forcing. Despite negative ΔpCO2 similar to the fall of 2007, maximum uptake rates during the spring were around -30 mmol m− 2 d− 1 because of weaker winds. In July, a short period of outgassing (maximum 5 mmol m− 2 d − 1) occurred due to the mid-summer pCO2sw peak (Figure 2c). This was followed by modest uptake in August and September (maximum ∼ -15 mmol m− 2 d− 1) due to the light wind conditions and moderately negative ΔpCO2. Overall, we calculated a mean math formula of -5.2 ± 2.0 mmol m− 2 d− 1 over the non–freezing period (184 days).

Figure 3.

Calculated math formula (black line) and math formula (red line) for the study area.

[27] Also shown in Figure 3 is our rough estimate of the winter enhanced flux (math formula). These results show two key features: high–magnitude fluxes in the fall and spring, and weaker but still significant fluxes throughout the winter. The high flux estimates in the fall were a result of the large amounts of open water present as the ice began to form while ΔpCO2 was strongly negative (Figures 2b, c & d). This initial pulse of rapid uptake is consistent with the findings of Anderson et al. [2004], but to our knowledge the potential for strong uptake as polynyas form in the spring has not been previously identified. The high spring flux estimates occurred because under–ice primary production began to reduce pCO2sw in March and April [Shadwick et al., 2011; Else et al., 2012a] and when dynamic processes began to break up the ice in early May, the atmospheric conditions were still conducive to new ice formation and thus enhanced flux.

[28] The mean math formula calculated for the freezing period (182 days) was -15.0 ± 10.0 mmol m− 2 d− 1. If this estimate is reasonable, the amount of carbon transferred to the mixed layer by this exchange should be commensurate with dissolved inorganic carbon (DIC) budgets for the region. In the same study area, Shadwick et al. [2011] observed a mixed layer DIC increase over the 2007-08 winter period (which they defined as November 15 – March 15) of approximately 200 μmol kg− 1. Assuming a 50 m mixed layer depth, the DIC increase resulting from math formula over that time period would be ∼ 37 ± 30 μmol kg− 1. Shadwick et al. [2011] constrained the winter increase in DIC due to brine rejection to be ∼ 97 μmol kg− 1, and Nguyen et al. [2012] reported water column community respiration rates of 11.0 ± 5 μgC L − 1 d − 1 over the same time period, which would have increased DIC by 109 ± 30 μmol kg− 1. Summed up, these components (gas exchange, brine rejection, respiration) exceed the observed 200 μmol kg− 1 DIC increase by 22%. However, we argue that the air–sea gas exchange rates can still be considered reasonable in light of this budget, given evidence that the respiration rates are likely overestimated [Nguyen et al., 2012], and given the large uncertainties in the budget terms.

[29] Since the freezing period (184 days) is essentially equal in length to the non–freezing period (182 days), this estimated flux enhancement greatly increases the annual math formula relative to simply computing math formula for the entire year. By combining the non–freezing and freezing season fluxes, we obtain a total annual average math formula of -10.1 ± 6.5 mmol m− 2 d− 1 for the Cape Bathurst polynya region.

4 Discussion

4.1 How “Typical” Was 2007–08?

[30] Mean annual fluxes for a given region are often compiled in papers that attempt to budget air–sea gas exchange on larger scales (e.g. Borges et al. [2005]; Cai et al. [2006]; Bates and Mathis [2009]; Chen & Borges [2009]). This approach is most effective when the compiled studies provide some analysis of interannual variability. We do not have the data set necessary to calculate a complete annual budget for our study area over several years, but in this section we use other data sets to discuss the long–term variability in many of the parameters that directly affect CO2 flux. The goal of this analysis is to determine if our mean annual flux estimate is typical of the mean state of the Cape Bathurst polynya, or if it represents a somewhat anomalous year.

4.1.1 pCO2sw Conditions

[31] To examine interannual variability in pCO2sw, we collected data from other cruises during which pCO2sw measurements were carried out in the study region between 2003–2009. The first available data are from Mucci et al. [2010], who calculated pCO2sw from surface DIC, pH, and total alkalinity (TA) measurements during fall 2003 and spring/summer 2004. In 2006, our research group made direct pCO2sw measurements on the CCGS Amundsen using an underway instrument composed of a membrane contactor (Liqui-Cel SuperPhobic x50) and an LI-7000 gas analyzer, similar to a system described in Else et al. [2008a]. In 2009, we returned to the area, this time with a commercially built underway pCO2sw system (General Oceanics model 8050), which has an expected accuracy of 2 μatm [Pierrot et al. 2009].

[32] The pCO2sw measurements made during these cruises are shown in Figure 4 in comparison to the generalized cycle and variability observed in 2007–08. Despite the diversity in measurement techniques, there is remarkable consistency in pCO2sw observations made between September and November, with most falling within ± 1 standard deviation of the 2007–08 measurements. This suggests that negative ΔpCO2 and a decreasing trend in pCO2sw as the sea surface cools is typical for the region in autumn, at least over the past decade.

Figure 4.

pCO2sw observations made in the Cape Bathurst polynya region during other cruises in comparison to the polynomial fit of the 2007–08 pCO2sw data. The dashed lines show ± one standard deviation around the polynomial fit.

[33] The only other time period where we can make interannual comparisons is in the summer open water season. As Figure 4 shows, 2004 appeared to follow a similar pattern to 2008, with pCO2sw increasing to a peak near 400 μatm in mid–July and then decreasing through August. It is worth noting that spring/summer ice cycles in 2004 followed a very similar pattern to 2008, with break–up occurring at essentially the same time. In 2009, pCO2sw appeared to experience a similar summer pattern to 2008, but perhaps with a delayed increase (starting at the end of July) and a smaller peak value (∼ 375 μatm in August). Ice break–up in 2009 occurred nearly one month later than in 2008, and the region did not become completely ice free until mid–August. This apparently delayed and reduced the summer pCO2sw increase, possibly by limiting air–sea gas exchange and keeping SST low (SST reached a maximum of ∼ 5 C in 2009 compared to ∼ 8 C in 2008). These results suggests that variations in the annual pCO2sw cycle may be linked to sea ice cycles, which raises the question of how typical the ice conditions were in 2007–08.

4.1.2 Ice Conditions

[34] We compared the 2007–08 ice conditions to a regional climatology based on the AMSR-E Ci data from 2002–2010 (Figure 5). In 2007, ice formation began about three weeks later than average, but reached a concentration of 0.80 only one week later than average. The winter leads and small polynyas generated by ice motion in 2007–08 are typical for the region, as evidenced by the long–term mean winter Ci below 1 and variability (standard deviation ∼ 0.02) around that mean. The significant decline in Ci in mid-May 2008 was also typical, although the region became ice–free more than 4 weeks earlier than average. Overall, there were very few instances when Ci in 2007–08 fell outside of one standard deviation of the 2002–2010 mean. Therefore, the 2007–08 ice cycle could be described as a “light” ice year with late freeze–up and rapid spring break–up, but not anomalously light within the context of the last decade.

Figure 5.

2007–08 ice concentration in the Cape Bathurst polynya region (red line) compared to the 2002–10 mean (black line). The grey shaded area represents ± one standard deviation around the 2002–10 mean.

4.1.3 Wind Conditions

[35] Wind speed and direction in 2007–08 were grouped by season and compared to wind data collected at the Cape Parry weather station between 2000–2010. Figure 6 shows that fall 2007 wind velocities were significantly higher than average, but adhered to the usual easterly direction. Winter and spring 2007–08 winds also displayed a typically easterly direction with stronger than average velocities. Summer 2008 wind velocities were substantially higher than the long–term mean and were even more easterly than usual. One–way analysis of variance (ANOVA) testing confirmed that mean wind velocities during all four seasons were significantly higher than the 2000-2010 seasonal averages. We can thus conclude that 2007–08 was an anomalously windy year, with winds predominantly from the east.

Figure 6.

Wind rose plots comparing the frequency of hourly wind velocities and directions by season observed in 2007–08 (left column) to the 2001–10 mean (right column). SON = September, October, November; DJF = December, January, February; MAM = March, April, May; JJA = June, July, August. Data are from the Cape Parry Environment Canada weather station.

4.1.4 Sensitivity of the Computed Flux to Interannual Variability in Wind and Ice Conditions

[36] Although we cannot perform a complete analysis of interannual variability, we can examine the sensitivity of the mean annual math formula to interannual variability in wind speed (U) and Ci. To do this, we held the 2007–08 pCO2sw, SST, and salinity data constant, and then solved equation 2, applying the historical Ci and U data both independently and in combination. The estimated enhanced gas transfer velocity was applied when Ci exceeded 0.80 in the fall and fell below 0.80 in the spring.

[37] As shown in Table 1, the calculated mean annual math formula using this approach ranges from -6.0 to -12.4 mmol m− 2 d− 1, with an average difference from 2007–08 of ± 1.7 mmol m− 2 d− 1. Allowing only wind speed to vary showed a relatively weak impact on mean annual math formula with a mean difference from 2007–08 of 0.6 mmol m− 2 d− 1. However, the analysis shows a stronger sink (by 0.2–0.8 mmol m− 2 d− 1) in 2007–08 than in any other year, suggesting that the abnormally strong winds enhanced CO2 uptake in our study year. In contrast, by allowing only Ci to vary, math formula ranges from a 3.3 mmol m− 2 d− 1 stronger sink to a 2.9 mmol m− 2 d− 1 weaker sink, indicating a high sensitivity to Ci. Much of this sensitivity arises from our estimate of math formula, both in terms of the length of the season to which the estimate is applied (which ranges from 179–233 days), and the strength of the enhancement (which ranges from a mean of -7.6 – -21.2 mmol m− 2 d− 1). Obviously, these numbers are somewhat speculative given the nature of our estimate of math formula. Nevertheless, given the multiple influences that variability in Ci has on math formula (including its potential role in contributing to variability in pCO2sw), it is safe to say that the mean annual air–sea exchange of CO2 in the Cape Bathurst polynya region is highly sensitive to Ci.

Table 1. Estimated Impacts of Interannual Variability in Sea Ice Concentration and Wind Speed on Air-Sea CO2 Flux
YearWind Velocity (m s-1)Ice Concentration 
SON mean (std dev)DJF mean (std dev)MAM mean (std dev)JJA mean (std dev)Freeze Up DateBreak Up DateWinter Ci mean (stdev)Mean FCO2as Both VaryMean FCO2as:enh Both VaryFCO2as vs '07-'08 Ice VaryFCO2as vs '07-'08 Wind Vary
  1. Columns 2-5 give seasonal (SON = September, October, November, DJF = December, January February, MAM = March, April May, JJA = June, July, August) mean/standard deviation wind velocities for each year. Columns 6-8 present sea ice concentration data for each year. Columns 9-10 show the total (including the enhanced fluxes) and winter CO2 flux (respectively). Columns 11–12 show the difference in total flux relative to 2007–08 if only ice concentration or wind speed is allowed to vary.
2000-015.2 (2.9)5.6 (3.7)5.2 (2.8)4.6 (2.8)------0.8
2001-025.6 (3.7)5.1 (3.4)4.9 (3.1)4.3 (2.6)------0.8
2002-036.1 (3.9)5.2 (3.3)4.5 (3.3)3.9 (2.6)15-Oct17-Apr0.962 (0.064)-12.2-21.2-3.20.6
2003-045.6 (3.5)6.1 (3.6)5.7 (3.8)4.2 (3.1)1-Nov25-May0.980 (0.038)-6.0-
2004-055.2 (3.6)6.4 (3.8)6.0 (3.7)4.6 (2.8)26-Oct10-May0.957 (0.060)-9.5-14.0-0.30.6
2005-066.4 (3.7)6.1 (3.6)5.1 (2.9)4.4 (2.4)2-Nov23-Jun0.974 (0.041)-8.9-
2006-075.9 (3.5)6.0 (3.8)5.4 (3.4)4.1 (2.5)12-Nov4-Jun0.970 (0.044)-9.2-13.4-0.50.7
2007-086.9 (4.6)6.4 (3.6)6.6 (3.9)5.8 (3.2)16-Nov13-May0.967 (0.043)-10.1-15.0 -
2008-095.5 (3.2)5.5 (2.8)5.1 (2.7)5.1 (3.1)8-Nov10-Jun0.961 (0.060)-12.4-18.1-3.30.4
2009-106.1 (3.9)5.8 (3.3)6.1 (3.5)5.4 (2.8)9-Nov17-May0.973 (0.032)-7.8-

4.2 Large Scale Atmosphere/Ocean Circulation and CO2 Fluxes in the Cape Bathurst Polynya

[38] Given the critical role of sea ice, it is interesting to consider the two remarkable events that bookended the 2007–08 year: the record low Arctic sea ice extent in September 2007 (since eclipsed by the September 2012 minimum), and the second–lowest recorded sea ice extent in September 2008. Those events raise the question of whether some of the conditions we observed in the Cape Bathurst polynya were influenced by the same factors that drove hemispheric–scale reductions in sea ice extent.

[39] In Figure 7, we show the seasonal sea level pressure in the Arctic during 2007–08, which reveals high pressure on the North American side of the Arctic for most of the year. Intense high–pressure was particularly obvious over the Beaufort Sea in fall 2007 and spring 2008. A strong Beaufort high drives anti–cyclonic circulation, explaining the strong easterly winds that we experienced during these seasons (Figures 2, 6). It also may explain some of the ice conditions we experienced, in particular the late fall freeze–up and the early spring break–up.

Figure 7.

Mean seasonal sea level pressure (in mbar) for the Arctic in 2007–08. SON = September, October, November; DJF = December, January, February; MAM = March, April, May; JJA = June, July, August. Data are from the NCEP/NAR reanalysis (Kalnay, 1996), plotted using the web-based plotting tools of the NOAA/ESRL Physical Sciences Division (Boulder, Colorado).

[40] Galley et al. [2009] hypothesized that the spring opening of the Cape Bathurst polynya is largely driven by easterly winds. We investigated this hypothesis using the ice–out dates between 1980–2004 reported by Galley et al. [2008], and found that easterly winds in March–April–May are indeed significantly correlated (p < 0.01) with earlier polynya opening dates (r = 0.59). It seems likely, therefore, that the early ice free conditions experienced in the spring of 2008 were driven to some extent by winds associated with the high pressure anomaly in the Beaufort Sea.

[41] Although easterly winds in June–July–August are likewise significantly correlated (r = -0.50, p < 0.01) with later freeze–up dates, the relationship between freeze–up date and large scale pressure variability is more complex. One important control which may drive earlier freeze–up in Amundsen Gulf is the intrusion of an “ice tongue” from the Beaufort pack ice. Galley et al. [2008] show that this ice tongue existed in about 40% of years between 1980–2004, and Galley et al. [2009] proposed that its presence seeds ice formation in Amundsen Gulf, potentially leading to earlier freeze–up. A strong Beaufort high (such as we observed in fall 2007) drives anticyclonic ice motion in the Beaufort Gyre, which moves ice away from the archipelago [Kwok, 2006; Lukovich and Barber, 2006], likely inhibiting the intrusion of this ice tongue. Certainly, archived data from the Canadian Ice Service show there was no ice tongue present in Amundsen Gulf in fall 2007, which may account for the late freeze–up.

[42] Based on this discussion, the timing of open water in the Cape Bathurst polynya was likely linked to pressure patterns in the Beaufort Sea, both through controls on wind forcing in Amundsen Gulf, and through behaviour of sea ice in the Beaufort Gyre. One of the major causes of the anomalous basin–wide conditions in 2007 and 2008 was the persistence of a strong “Dipole Anomaly” (DA), a statistical description of sea level pressure variability in the Arctic [Wang et al., 2009; Overland and Wang, 2010b]. The phase of the DA that persisted for all of 2007 and 2008 is associated with anomalously high pressure on the North American side of the Arctic [Wang et al., 2009], as is apparent in Figure 7. Given the important role that this pressure pattern played in influencing our observations in 2007–08, the DA may be useful in explaining conditions in the Cape Bathurst polynya.

[43] If this hypothesis of a link between strong easterly winds, light ice conditions, and a dominant DA is sound, our presence in the Cape Bathurst polynya in 2007–08 was particularly fortuitous. The Arctic Ocean appears to have transitioned from a period where the positive phase of the Arctic Oscillation dominates circulation patterns, to one in which the DA is dominant [Overland and Wang, 2010a; Maslanik et al., 2007]. Therefore, the 2007–08 CO2 flux budget that we have calculated for the Cape Bathurst polynya may be representative of the region under contemporary circulation patterns. Although the emergence of the DA is difficult to explain (and thus difficult to predict), it may in part be due to climate warming as a result of anthropogenic forcings [Zhang et al., 2008]. If the pattern were to intensify as those forcings intensify, we might expect higher easterly winds, earlier break–up, later freeze–up, and more winter ice motion in the Cape Bathurst polynya, all of which would increase atmospheric CO2 uptake.

4.3 Implications at the Circumpolar Scale

[44] Our study area is an example of the many coastal polynyas distributed around the circumpolar Arctic (e.g. Barber and Massom [2007]). If the rates of atmospheric CO2 uptake that we estimated are similar in other Arctic polynyas, then the Arctic–wide uptake of CO2 through these features may be significant. Of particular interest is the winter CO2 uptake. In the Storfjorden polynya, Omar et al. [2005] estimated a mean winter CO2 flux of -14.8 mmol m− 2 d− 1, and performed a simple extrapolation to calculate a total winter uptake for all Arctic coastal polynyas of 2.3 × 1012 gC yr− 1. This uptake rate is on the same order of magnitude as that observed in many mid– and high–latitude shelf seas [Cai et al., 2006], despite the total area of coastal polynyas being relatively small. The finding is particularly significant, because at least some of the carbon absorbed by this processes is expected to be removed from further exchange with the atmosphere via the formation of deep and intermediate water by brine rejection [Omar et al., 2005].

[45] Our estimated rate of winter uptake for the Cape Bathurst polynya region (-15.1 ± 10.0 mmol m− 2 d− 1) is remarkably similar to the rates found by [Omar et al. [2005], lending support to their theory that polynyas play a significant role in CO2 sequestration. If we extrapolate our mean winter flux to the same area that Omar et al. [2005] used for coastal polynyas (3.5 × 1010 m2) and assume a 184–day freezing season (section 3.2), we calculate a slightly smaller uptake for Arctic coastal polynyas of 1.2 × 1012 gC yr− 1. Even the lower rate of exchange that we estimated can be partly explained by differences between the two regions; the Storfjorden polynya opens periodically through the winter, keeping the mean open water fraction relatively high (7–13%, Skogseth et al. [2005]), while the Cape Bathurst polynya does not experience a large–scale opening until spring, with a lower open water fraction in winter (2–4%, Table 1). This illustrates that upscaling will result in an estimate biased by the peculiarities of the model polynya, but our results support Omar et al. [2005]’s conclusion that their estimate is of the right order of magnitude.

[46] We caution, however, that further investigations are required to confirm if negative ΔpCO2 throughout the winter is a typical state for Arctic polynya regions. Along the southern margins of Amundsen Gulf and on the Mackenzie Shelf, we observed episodes of strongly positive ΔpCO2 associated with wind–driven upwelling events in the late fall [Else et al., 2012a, 2012b]. We do not know if that upwelling persisted through the winter, but it is quite plausible that certain polynyas may be sustained by upwelling–favourable winds, or that ice motion in a polynya may drive winter upwelling (e.g. McPhee et al. [2005]). Furthermore, evidence of winter supersaturation has been presented for the Northwater polynya [Miller et al., 2002] and in landfast ice near the re–occurring flaw lead north of Point Barrow, Alaska [Semiletov et al., 2007]. If polynya or flaw lead regions with supersaturated winter pCO2sw are in fact widespread, then the CO2 uptake estimate of Omar et al. [2005] must be offset to some degree by such regions.

[47] During the non–freezing seasons, Arctic shelf seas typically absorb CO2 at rates between -1.0 to -15 mmol m− 2 d− 1 (see for example Table 3 in Else et al. [2008b]). The mean CO2 uptake rate that we observed (-5.2 ± 2.0 mmol m− 2 d− 1) is median to that range, although considerable variability in both the magnitude and direction occurred over the ice–free season (Figure 3). Our results thus reinforce the paradigm that, with a few notable exceptions (e.g. Anderson et al. [2009]), Arctic shelf seas generally absorb atmospheric CO2 during non–freezing seasons [Bates and Mathis, 2009].

5 Summary and Conclusions

[48] The potential role of polynya regions as atmospheric CO2 sinks has long been acknowledged [Yager et al., 1995], but opportunities for thorough investigation have been rare. This study shows that the Cape Bathurst polynya region does indeed act as a strong CO2 sink, but not entirely for the reasons that Yager et al. [1995] suggested. Yager et al. [1995] hypothesized that in spring–opening polynya regions, potential winter outgassing would be prevented by a sea ice cover. In contrast, we found that in the presences of a broken and mobile ice cover, a persistently negative ΔpCO2 coupled with an enhanced rate of gas exchange allowed CO2 uptake through the winter at a mean rate in excess of the open water seasons. When combined with the exchange that occurred through the other seasons, our calculated mean annual rate of CO2 uptake for this region was approximately 3 times higher than the preliminary estimate of Shadwick et al. [2011].

[49] A sensitivity analysis on the effect of variability in wind and ice conditions on CO2 exchange revealed that anomalously strong winds during 2007–08 likely enhanced CO2 uptake relative to the decadal mean. Further adding to the strong uptake were light ice conditions, characterized by a late freeze–up (by ∼ 3 weeks) and early breakup (by ∼ 4 weeks). The light ice conditions may have been caused in part by the strong easterly winds, and in turn appeared to influence the pCO2sw cycle. Both the wind and ice conditions that we observed in 2007-08 were likely controlled by a strong north–south pressure gradient, a characteristic of the Arctic Dipole phase that was particularly strong through the study period. If that phase of the Arctic Dipole pattern becomes the norm for the Arctic (as recent observations by Overland and Wang [2005] and Maslanik et al. [2007] suggest), then the mean annual CO2 flux we estimated for 2007–08 may be quite representative of current and future air–sea CO2 exchange in the region.

[50] Net annual uptake of atmospheric CO2 by the Cape Bathurst polynya region agrees with the paradigm that most Arctic shelf seas act as sinks for atmospheric CO2. Bates and Mathis [2009] estimated that the Arctic Ocean absorbs 0.6–1.9 × 1014 gC yr− 1 during the open water seasons, and our estimated open water CO2 fluxes are within the range of observations used to arrive at that conclusion. More interestingly, our study also supports an emerging school of thought that the Arctic Ocean may absorb a significant amount of CO2 during the winter. Although initial order–of–magnitude estimates indicate that the total winter flux is smaller than the open water flux (1012 gC yr− 1 vs. 1014 gC yr− 1), the area through which this exchange occurs is smaller by a similar magnitude. Thus, if the area of winter open water increases, absorption of CO2 via this pathway could become more significant. The potential for increased winter open water is a distinct possibility, given the recent decline of multi–year ice [Maslanik et al., 2011], drastic decreases in summer sea ice extent (e.g. Stroeve et al. [2007]), and observations of increased winter ice motion [Hakkinen et al., 2008]. However, testing of this hypothesis requires a better understanding and quantification of the flux enhancements that occur in winter open water features, and an extensive survey of winter ΔpCO2 throughout the Arctic.


[51] Thanks to the Captains and crew of the CCGS Amundsen and the many people who helped in the field: Bruce Johnson, Sarah Woods, Kyle Swystun, Gauthier Carnat, Nes Sutherland, Keith Johnson, Jens Ehn, Silvia Gremes-Cordero, Sylvain Blondeau, Luc Michaud and many others. CTD data were provided by Yves Gratton. This work is a contribution to the International Polar Year-Circumpolar Flaw Lead system study (IPY-CFL 2008), supported by the Canadian IPY Federal program office, the Natural Sciences and Engineering Research Council (NSERC), and many other contributors. The authors of this paper are members of ArcticNet, funded in part by the Networks of Centres of Excellence (NCE) Canada, NSERC, the Canadian Institute of Health Research and the Social Sciences and Humanities Research Council. B. Else is supported by a Vanier Canada Graduate Scholarship, and received funding for logistics from the Northern Scientific Training Program. We gratefully acknowledge the continued support of the Centre for Earth Observation Science, the Canada Excellence Research Chair (CERC) in Arctic Geomicrobiology and Climate Change, and the University of Manitoba.