The spatial and interannual dynamics of the surface water carbonate system and air-sea CO 2 fluxes in the outer shelf and slope of the Eurasian Arctic Ocean

The Arctic is undergoing dramatic changes which cover the entire range of natural processes, from extreme increases in the temperatures of air, soil, and water, to changes in the cryosphere, the biodiversity of Arctic waters, and land vegetation. Small changes in the largest marine carbon pool, the dissolved inorganic carbon pool, can have profound impact on the carbon dioxide (CO2) flux between the ocean and the atmosphere, and the feedback of this flux to climate. Knowledge of relevant processes in the Arctic seas improves the evaluation and projection of carbon cycle dynamics under current conditions of rapid 20 climate change. Investigation of the CO2 system in the outer shelf and continental slope waters of the Eurasian Arctic seas (the Barents, Kara, Laptev, and East Siberian seas) during 2006, 2007, and 2009 revealed a general trend in the surface water partial pressure of CO2 (pCO2) distribution, which manifested as an increase in pCO2 values eastward. The existence of this trend was defined by different oceanographic and biogeochemical regimes in the western and eastern parts of the study area; the trend is likely 25 increasing due to a combination of factors determined by contemporary change in the Arctic climate, each change in turn evoking a series of synergistic effects. A high-resolution in situ investigation of the carbonate system parameters of the four Arctic seas was carried out in the warm season of 2007; 2007 was characterized by the next-to-lowest historic sea ice extent in the Arctic Ocean, on satellite record, to that date. The study showed the different responses of the seawater carbonate system to the environment changes in the western vs. the eastern Eurasian Arctic seas. The large open, highly productive water area 30 in the northern Barents Sea enhances atmospheric CO2 uptake. In contrast, the uptake of CO2 was strongly weakened in the outer shelf and slope waters of the East Siberian Arctic seas under the 2007 environmental conditions. The surface seawater appears in equilibrium or slightly supersaturated by CO2 relative to atmosphere because of the increasing influence of river


Introduction
The Arctic is currently undergoing dramatic changes which cover the entire range of natural processes; from extreme increases in the temperatures of air, soil, and water, to changes in the biodiversity of Arctic waters and land vegetation (Serreze and Barry, 2011;Bhatt et al., 2010).In 1896, the Swedish scientist Svante Arrhenius hypothesized that changes in the atmospheric concentration of carbon dioxide (CO 2 ) could alter the Earth's surface temperature and that this temperature change would be especially large in polar latitudes.This likely is the first formal description of what today is known as the Arctic amplification, i.e., a higher temperature increase in Arctic regions than in other regions of the globe (Serreze and Barry, 2011;Jeffries et al., 2013).The changes being observed today will probably become more intense in the coming decades through positive feedback, causing further changes in atmospheric circulation, river discharge, the carbon cycle, vegetation, conditions of terrestrial and submarine permafrost, and many other natural processes; the consequences will be noticed within, as well as outside of, the Arctic region (Serreze and Barry, 2011;Anderson et al., 1998;Macdonald et al., 2008;Semiletov et al., 2000Semiletov et al., , 2016;;Shakhova et al., 2014).Currently, these changes refer to a "new condition" of the Arctic climate (Kattsov et al., 2010;Jeffries et al., 2013;Wood et al., 2015).
The most obvious indicator of Arctic climatic change is the change of sea-ice cover, with a persistent decline in areal extent during the last decades.Since the start of satellite observations in 1979, the sea-ice extent in March, the period of maximum coverage, has declined by 2.6 % per decade (Serreze et al., 2007;Stroeve et al., 2012;Jeffries et al., 2013).However, during the last decade, the September seaice extent has decreased by 13 % relative to the 1979-2000 average (Jeffries et al., 2013).The change in sea-ice coverage is most pronounced in the large eastern Arctic shelf seas (ftp://sidads.colorado.edu).Moreover, the melt season has lengthened by 1-2 weeks per decade, and with continued Arctic warming it will expand further (Stroeve et al., 2014).
The environmental conditions vary between the Eurasian shelf seas, of which the Barents Sea is one of the largest and deepest (Jakobsson, 2002).According to the classification of Carmack et al. (2006), the Barents Sea is considered an "inflow" shelf sea or an Atlantic-influenced shelf sea (Findlay et al., 2015).The general inflow of warm and salty water from the Atlantic keeps a large part of the Barents Sea ice-free all year round.In the northern area, a portion of the warm Spitsbergen Current returns around Svalbard to the Barents Sea as "cold Atlantic" water (Kaltin et al., 2002) or "Arctic" water (Loeng, 1991).This area is largely covered with ice, but at a variable extent over the year, and the presence of polynyas contributes to water salinization in the winter (Carmack et al., 2006).However, the cooling of the Atlantic water (AW) during its passage through the Barents Sea produces the largest volume of high-density water that ventilates the deep Eurasian Basin water (Schauer et al., 2002).The Barents Sea receives little river input compared to other Arctic shelf seas (Anderson et al., 1998;Schauer et al., 2002).Due to high primary productivity (PP) and cooling during transit to the north, the waters of the Barents Sea constitute the strongest all-season sink for atmospheric CO 2 in the Arctic (Fransson et al., 2001;Omar et al., 2007;Bates and Mathis, 2009;Årthun et al., 2012;Lauvset et al., 2013).
The other shelf seas (the Kara, Laptev, and East Siberian seas) are classified as "interior" shelf seas (Carmack et al., 2006) or as river-influenced shelf seas (Findlay et al., 2015).The river discharge as well as the seasonal formation and melting of sea ice greatly impact the hydrology and chemistry of these shelf seas.Water of Atlantic origin enters the Kara Sea from the Barents Sea, and the Kara Sea also receives more than a third of the volume of riverine discharge flowing into the Arctic Ocean, mainly via the Ob and Yenisei rivers.
The waters of the Arctic seas have become warmer and fresher than they were several decades ago (Wood et al., 2015); among other effects, this warming and freshening has caused increased OM input to the shelf seas, resulting in severe aragonite undersaturation (Semiletov et al., 2016).It has been shown that the acidifying effect of terrestrial OM decomposition at an erosion-dominated site was more than 5 times stronger than that of estuary freshening.
How climate change impacts the contemporary carbon cycle in the Eurasian Arctic seas, including its consequences for transformation and fluxes, has been the subject of intense interest during the last decade (Anderson et al., 2011;Bischoff et al., 2016;Bröder et al., 2016;Charkin et al., 2015;Gustafsson et al., 2011;Karlsson et al., 2016;Macdonald et al., 2008;Sánchez-García et al., 2014;Tesi et al., 2014;Vonk et al., 2014).Small changes in the largest marine carbon pool, the dissolved inorganic carbon (DIC) pool, can have profound impacts on the CO 2 flux between the ocean and the atmosphere and the feedback of this flux to climate.Knowledge of relevant processes in the Arctic seas improves the evaluation and projection of carbon cycle dynamics under conditions of rapid climate change.The East Siberian Arctic seas (ESAS), including the Laptev Sea, the East Siberian Sea, and the Russian sector of the Chukchi Sea, are especially relevant in this perspective because they have the broadest and the shallowest shelves among the Arctic seas, they receive large volumes of river discharge, they are characterized by high rates of coastal erosion, and their drainage basins are underlain by permafrost (Macdonald et al., 2008;Semiletov et al., 2000Semiletov et al., , 2012)).
Studies of the Barents Sea carbonate system (Årthun et al., 2012;Lauvset et al., 2013;Pipko et al., 2011b;Yakushev and Sørensen, 2013) as well as the ESAS waters (Anderson et al., 2009(Anderson et al., , 2011;;Pipko et al., 2011aPipko et al., , 2015Pipko et al., , 2016;;Semiletov et al., 2012Semiletov et al., , 2013Semiletov et al., , 2016) ) have been performed during the last decade.The Kara Sea remains less studied in the context of carbonate system dynamics and the main research was accomplished in the shallow part of this sea (Makkaveev et al., 2010(Makkaveev et al., , 2015)).Most of the ESAS studies were carried out in ice-free areas, i.e., at depths normally limited to 70 m isobaths, which correspond to the area of the inner and middle shelves.Meanwhile, the deep part of the seas where the changes in the ice cover are most pronounced is the least investigated.To date, only one paper dedicated to CO 2 system dynamics in the East Siberian Sea outer shelf is available (Anderson et al., 2017); it is based on the field campaign accomplished within the framework of the SWERUS-C3 (Swedish -Russian -US Arctic Ocean Investigation of Climate -Cryosphere -Carbon Interactions) program aboard the Swedish icebreaker Oden.This study was accomplished in ice conditions, with the sea-ice concentration ranging between 70 and 100 %.
The objective of this contribution is to evaluate the importance of the meteorological and oceanographic conditions and biogeochemical processes which determine the surface water partial pressure of CO 2 (pCO 2 ) and air-sea CO 2 fluxes in the outer regions of the Eurasian Arctic seas' shelf/slope system along the AW inflow path in different years.This will add to the knowledge of the regional sensitivity to current changes and thus project the response of the entire Arctic carbon cycle to global climate warming.

Study area
The study is based on observational data collected in the Eurasian sector of the Arctic Ocean during the summer-fall (late August-early October) seasons of 2006, 2007, and 2009 (Fig. 1).An extensive investigation of the outer shelf and continental slope of the Barents, Kara, and Laptev seas and the northwestern East Siberian Sea was performed during the 2007 expedition on the research vessel Viktor Buynitsky within the framework of the Nansen and Amundsen Basins Observational System (NABOS) program (Fig. 1).Data collected in 2006 and 2009 within the framework of the NABOS program aboard the icebreaker Kapitan Dranitsyn were used for comparative analysis (Fig. 1).

Hydrological data
During all cruises, water samples for chemical analysis were taken with a standard Rosette system equipped with the SBE19+ CTD (conductivity, temperature, depth) probe to record conductivity and temperature.In 2007, another SBE19+ probe equipped with the same sensors was deployed in a 150 L plastic barrel into which flowing seawater was pumped from the depth of about 4 m at the rate of about 80 L min −1 .

Total alkalinity
Water samples were poisoned with a mercuric chloride solution at the time of sampling to halt biological activity (Dickson et al., 2007) and were stored in the dark at room temperature until they were analyzed ashore.Samples for total alkalinity (A T ) were analyzed in the lab within 1 month using an indicator titration method in which 25 mL of seawater were titrated with 0.02 M HCl in an open cell according to Bruevich (1944) and Pavlova et al. (2008).Measurements were performed at 20 • C, with the temperature in the cell controlled to within 0.1 • C. In 2000, the working group on carbon dioxide in the north Pacific of the North Pacific Marine Science Organization (PICES) performed an intercalibration of A T in seawater using certified reference materials (CRMs).The results of the intercalibration showed that the alkalinity values obtained by the Bruevich method are in agreement with the standard within ±1 µmol kg −1 when state-of-the-art analytical practice is applied (Pavlova et al., 2008).A T measurements were performed with a precision of ∼ 2 µmol kg −1 with the accuracy set by calibration against CRMs supplied by A. Dickson, Scripps Institution of Oceanography (USA).Batch no.96 was used in 2009.During the 2006 and 2007 cruises, the HCl concentration was determined using a standard solution of Na 2 CO 3 made up by carefully weighing Na 2 CO 3 of 99.995 % purity (DOE, 1994;Pavlova et al., 2008).

pH
A potentiometric method was applied to determine pH on the Pitzer pH scale (Pitzer, 1991) using a closed cell held at constant 20 • C temperature with a sodium and hydrogen glass electrode pair without liquid junctions (Tishchenko et al., 2001(Tishchenko et al., , 2011)).A TRIS-TRIS-HCl-NaCl-H 2 O buffer solution (Tishchenko, 2000(Tishchenko, , 2011) ) was used for calibrations on the Pitzer pH scale.Both the hydrogen glass electrode and the sodium glass electrode were calibrated using this buffer.Together with thermodynamic data (Dickson, 1990a), the pH values were converted from the Pitzer pH scale to the total hydrogen ion concentration scale (pH T 20) (Tishchenko et al., 2001(Tishchenko et al., , 2011;;Dickson et al., 2007).The precision of pH measurements was about 0.004 pH units, with the accuracy set by calibration against buffer solution on the Pitzer pH scale.

Partial pressure of CO 2
Continuous measurements of pCO 2 were performed in the surface mixed layer using a submersible autonomous moored instrument for CO 2 (SAMI-CO 2 sensor) with a precision of ±1 µatm (DeGrandpre et al., 1995).The sensor was deployed in the same barrel as the SBE19+ probe.The calibration procedures are described in detail in DeGrandpre et al. (1995).The temperature in the barrel was 0.55 • C higher than the sea surface temperature, and the pCO 2 measurements were corrected to in situ temperature using the equation of Takahashi et al. (1993).All in situ surface data described in this paper were averaged over 15 min intervals.
At oceanographic stations, surface pCO 2 values were calculated, on discrete samples, from pH T 20, A T , and inorganic nutrient data using the CO2SYS program of Lewis and Wallace (1998) with equilibrium constants of Mehrbach et al. (1973) refit by Dickson and Millero (1987), using the sulfate and borate dissociation constants of Dickson (1990a, b).

Wind speed
The wind speed was measured using an automated meteorological station (gradient Automatic Weather Station AWS 2700) located at a height of 15-20 m above sea level.The true wind speed was calculated using navigation information and was extrapolated to the height of 10 m.

CO 2 flux calculation
Air-sea CO 2 fluxes were calculated using the diffusive boundary layer model: where F is gas flux (e.g., in mmol CO 2 m −2 day −1 ), K w is gas-transfer velocity, S is CO 2 solubility (Weiss, 1974), pCO 2 is the difference between the atmospheric and oceanic pCO 2 , and f ice is the fraction of sea-ice coverage.Two relationships were used for calculating gas-transfer velocity (Wanninkhof, 1992;Wanninkhof and MacGillis, 1999).Gas-transfer velocities were calculated using onboard measured wind speed.

Apportionment of freshwater fractions
In order to determine the composition of water samples, we used a three-component mass balance, using salinity and A T in this evaluation (e.g., Ekwurzel et al., 2001;Fransson et al., 2001Fransson et al., , 2009)).The major freshwater (FW) sources are river water (RW) and sea-ice meltwater (MW), both mainly originating from the Arctic shelf areas.It is assumed that each summer sample is a mixture of Atlantic-derived seawater (SW; f SW ), RW (f RW ), and sea-ice MW (f MW ).For riverine A T , the average value of 840 µmol kg −1 was applied, which is typical of the largest Siberian rivers (the Ob, Yenisei, and Lena rivers) during the warm season (Tank et al., 2012); up to 90 % of the total river discharge enters the Arctic seas during this season (Dittmar and Kattner, 2003).The sea-ice values of salinity and A T are taken from Fransson et al. (2009); the Atlantic-derived water values of salinity and A T are taken from Pipko et al. (2011b).

Statistical treatment and graphical representation of the data
Data were tested statistically using an empirical distribution function test in the Statistics 7.0 software package.Descriptive statistics were calculated for the 95 % confidence interval of the mean (P = 0.95, α = 0.05).Most of the plots and maps in this study were created with the Ocean Data View software (Schlitzer, 2011).
3 Results and discussion

Meteorological conditions
In the summer season of 2006, low sea level pressure (SLP) dominated over the Arctic Ocean (Fig. 2a), resulting in dominating westerly winds that hampered penetration of RW into the central Arctic Ocean, as well as northern to northwestern sea-ice drift.The area of sea-ice cover was maximal (5.9 million km 2 ) and the sea-ice edge in 2006 had the most southern position of the 3 years studied (Fig. 1a), which also impeded the transfer of surface water to the deeper part of the ocean.The negative sea-ice concentration anomaly was strongest to the east of the study area, while the total anomaly for the whole Arctic Ocean was −0.5 million km 2 compared to the mean coverage during the 1981-2010 time period (ftp://sidads.colorado.edu).The sea-ice conditions varied from light to ice-free in the central Laptev Sea to heavy north of the Novaya Zemlya islands where the concentration reached 90-100 %.
The Arctic dipole (AD), which is characterized by low SLP on the Eurasian side of the Arctic and high SLP on the American side, was present in the summer of 2007 (Fig. 2b) and contributed to a 2007 record minimum sea-ice extent (Overland et al., 2014).The AD pattern persisted for part of the summer during each year following 2007.The interplay between the two regional centers of atmospheric pressure controlled the wind pattern, especially over the ESAS.The maximum summer winds and the most intensive transfer of RW and sea ice to the north and northwest occurred in the warm season of 2007, forced by the extreme pressure gradient between the two centers of action (Fig. 2b).In fact, strong winds were experienced during the 2007 cruise, with wind speed reaching 22 m s −1 in the western study area and 13 m s −1 in the eastern.In 2007, the sea-ice anomaly reached −1.6 million km 2 for the entire Arctic Ocean.The negative anomaly in the sea-ice concentration shows how much the ice concentration for a particular month in one year differs from the mean calculated for that month from 1981 to 2010; that mean was maximal in the ESAS, exceeding 50 % (ftp://sidads.colorado.edu).Actually, the entire study region was largely ice-free in 2007.
A high-pressure area was also present over the American side of the Arctic Ocean in the summer of 2009, with comparable pressure in the center of the anticyclone but with an even larger extent than in 2007 (Fig. 2c).However, the low-pressure center over the Siberian Arctic was much weaker, resulting in a weaker SLP gradient and cor-respondingly weaker wind.A significant part of the study area was covered with ice in 2009, with the sea-ice concentration reaching 95-100 % (Fig. 1).The total Arctic Ocean anomaly of sea-ice concentration in 2009 was also negative at −1.0 million km 2 (ftp://sidads.colorado.edu).Maximal interannual deviations in the sea-ice extent were detected in the ESAS, while the position of the sea-ice edge in the Barents Sea was far to the north and did not differ much between the years (Fig. 1a).

Temperature
In 2006, the temperature was close to the freezing point in the regions where sea ice was present in the northern Bar- ents and Kara seas as well as in the deep basin north of the Laptev Sea; surface temperatures varied from −1.82 to 3.30 • C. The highest temperatures (maximum 3.90 • C) were found in the ice-free waters of the Laptev Sea (Fig. 3).In 2007, low sea surface temperatures were found in the northern Barents Sea east of Svalbard, in the northern Kara Sea, and in the Vilkitsky Strait (down to −1.13 • C; Fig. 4).These low temperatures were associated with the ice edge vicinity and the presence of sea ice in the Vilkitsky Strait (Fig. 1).The waters in the northwestern Barents Sea retained the original AW characteristics (2 < T < 5 • C and S > 34.8; Hopkins, 1991).The highest sea surface temperatures (up to 4.11 • C) were measured west and north of Svalbard (Fig. 4).High surface water temperatures were also measured in the Laptev Sea (∼ 3.70 • C) and over the Lomonosov Ridge (Fig. 4).In 2009, surface water temperatures varied from −1.73 to 2.85 • C (Fig. 5).Temperature remained < 0 • C over most of the study area with > 0 • C values in the ice-free areas of the Laptev Sea and the Kara Sea.

Salinity
In 2006, the sea surface salinity, as measured at the oceanographic stations, covered a range from 27.00 to 33.57 (Fig. 3).High salinity was observed along the AW inflow path, i.e., in the northern Barents and Kara seas.As it enters the northern Laptev Sea, the water's salinity decreases through mixing with river runoff; the lowest salinity was measured in the ice-free waters of the Laptev Sea (Fig. 3).The sea surface salinity within the study region in 2007 varied substantially, as was expected considering the differences in oceanographic regimes of the studied seas.In the West Spitsbergen Current, the AW salinity was close to 35; it slowly decreased along the cruise track in the northwestern Barents Sea, slightly varied in the central part of the sea, and decreased in the eastern part (Fig. 4).In the eastern Kara Sea, which is influenced by river runoff, the salinity decreased to well below 30, a salinity level that also was observed in the western Laptev Sea as well as at the Laptev Sea continental slope.The lowest salinities were observed in the eastern Laptev Sea and northwestern East Siberian Sea, reaching well below 25 (Fig. 4).In 2009, the surface water salinity varied from 28.79 to 33.88 with relatively constant salinity in the waters of the Barents and the Kara seas, while it dropped sharply when entering the Laptev Sea where the lowest values were observed (Fig. 5).

Surface water pCO 2 spatial distribution in 2007
Seawater pCO 2 is affected by several processes.Some are physical, such as temperature and vertical as well as horizontal advection; others are biological, such as production/mineralization of OM.The importance of these processes for the surface pCO 2 values observed in fall 2007, which was characterized by the next-to-lowest historic seaice extent in the Arctic Ocean to that date, is discussed in the following.

The Barents Sea
The Barents Sea is an inflow shelf where the supply of nutrients from the Atlantic forms the basis for high PP.This, together with the accompanying heat loss, results in year-round CO 2 undersaturation of the surface layer.Consequently, it is an annual sink for atmospheric CO 2 , though with large www.ocean-sci.net/13/997/2017/Ocean Sci., 13, 997-1016, 2017  spatial variability (Fransson et al., 2001;Kaltin et al., 2002;Nakaoka et al., 2006;Omar et al., 2007;Bates and Mathis, 2009;Lauvset et al., 2013).Of the Arctic shelf seas, the Barents Sea can only compare with the highly productive Chukchi Sea where the surface pCO 2 can drop down to 100 µatm during the warm productive season (Bates, 2006;Pipko et al., 2002).
Our investigation shows that the surface waters in the lessstudied northern Barents Sea were also undersaturated, and thus the northern Barents Sea is a sink of atmospheric CO 2 (Fig. 4).The waters to the west of Svalbard are undersaturated by about 50 µatm, which is typical for this region when the flux from the atmosphere cannot keep up with the decrease in pCO 2 caused by the cooling of the northwardflowing water.During its flow to the north/northeast, the surface water is cooled further, mainly by melting sea ice; thus, it also freshens.Therefore, the temperature of surface waters decreased by ∼ 4. at 9-24 • E longitude; this caused a thermodynamic decrease of pCO 2 , calculated according Takahashi et al. (1993), by ∼ 70 µatm.The correlation between temperature and pCO 2 in this region was strong (R = 0.85), further emphasizing the importance of temperature in determining pCO 2 here (Fig. 6).An additional pCO 2 decline resulted from the addition of sea-ice MW, normally undersaturated in CO 2 (Nedashkovsky and Shvetsova, 2010;Rysgaard et al., 2012); the calculated sea-ice MW fraction was up to 10 % in the northeastern Barents Sea (Fig. 7).PP could have also contributed to lowering pCO 2 even if this study had occurred at the end of the productive season (end of September -beginning of October) because air-sea exchange might not have compensated fully for the biological drawdown.
To the east, in the northern Barents Sea, surface temperatures varied slightly with longitude and remained negative; salinity slowly decreased from 24-53 • E longitude and then reduced to ∼ 31 near 65 • E longitude (Fig. 6).In the eastern Barents Sea, the relationship between pCO 2 values and temperature was weak; salinity demonstrated a significant spatial variability, but a pCO 2 -salinity correlation was practically absent (Fig. 6).Therefore, in the northern Barents Sea, pCO 2 variability is driven by different processes in northeastern and northwestern parts of the sea; the temperature impact predominates in the west, and the influence of MW predominates eastward.

The Kara Sea
The Kara Sea surface waters were undersaturated in pCO 2 , with the lowest values in the west and highest to the east (Fig. 4).The pCO 2 correlation with temperature was weak (R = 0.28) for the entire Kara Sea, but pCO 2 was strongly negatively correlated with salinity (R = −0.72).Generally, the Kara Sea has two oceanographic subregions with different regimes: the western part where Atlantic origin water dominates, slightly modified by sea-ice melt, and the eastern part where this modified water is further diluted by river runoff from two Great Siberian rivers, the Ob and the Yenisei.This pattern is set by the general eastward direction of water transport in the Eurasian Arctic seas (e.g., Olsson and Anderson, 1997), but wind is the main driving force of the north RW flow in the Kara Sea (Harms and Karcher, 1999).In early summer 2007, with intensive AD development (Fig. 2), a western RW transport developed; in late summer, a northern/northeastern type of RW distribution predominated (Zatsepin et al., 2010), and the deep part of the sea was partly influenced by these waters.Note that in 2007 the discharge of the Ob and Yenisei rivers was the largest of the 3 years studied (2006, 2007, and 2009) and exceeded (+23 and +4 %, respectively) the average multi-year value for the 1999-2009 period (426 and 663 km 3 , respectively) (PART-NERS and Arctic-GRO project data); this river discharge additionally affected the northeastern Kara Sea.
Hence, we examine separately the relationships of pCO 2 with the hydrographic characteristics of the western and eastern regions (Fig. 8).The lack of reliable correlation of pCO 2 with the hydrography in the western Kara Sea was analogous to the northeastern Barents Sea.This emphasizes the fact that a similar source of water as well as similar processes that determine the carbonate system dynamics occurred in this part of the Eurasian Arctic seas; sea surface temperature slightly changed and MW was the predominant source of FW.
In contrast, processes determining carbonate system dynamics in the eastern Kara Sea were clearly different.The examination revealed a high positive correlation between pCO 2 and temperature (R = 0.84) and a strong negative correlation of pCO 2 with salinity (R = −0.62) in the eastern part of the sea (Fig. 8).Together with the increase of pCO 2 toward the east, significant correlations of pCO 2 values with hydrological parameters pointed to the role of RW.Riverine discharge added to the seawater warming, increased the pCO 2 , and contributed water enriched with CO 2 ; pCO 2 was also increased via decay of terrestrial OM, transported with RW.The negative correlation found in the eastern Kara Sea is typical for regions influenced by warm RW, with high pCO 2 and labile OM as a substrate for further CO 2 production (Semiletov et al., 2013).Furthermore, together with salinity, pCO 2 is a useful tracer of the river plume distribution within the Kara Sea (Fig. 8).
Note that, despite the presence of RW in the eastern part of the sea, the surface pCO 2 values remained below atmospheric values.

The Laptev Sea
High pCO 2 was observed in the Laptev Sea surface waters, even to levels that exceeded atmospheric (Fig. 4).Supersaturation was observed in the southern Laptev Sea outside the Lena River delta, which is typical in surface waters of the eastern Laptev Sea inner and middle shelves (Anderson et al., 2009;Semiletov et al., 2012Semiletov et al., , 2013;;Pipko et al., 2016).Moreover, high pCO 2 was also observed in the surface waters of the outer Laptev Sea shelf over the Lomonosov Ridge and further to the east (Fig. 4).The computed RW fraction reached 30-35 % in the southern Laptev Sea and decreased to 5-7 % north of the latitudinal transect extending from the Lena Delta (Fig. 7).High RW content in the surface water (up to 33 %) was also found over the Lomonosov Ridge, which indicated a more efficient northeastern transport of RW in the ice-free conditions of 2007.Thus, under intensive AD development, the northeastern transfer of RW prevailed and maximal RW content was found in the surface slope water over the Lomonosov Ridge.The MW had a small effect on the middle and outer shelves; a strong sea-ice-related brine signal was found here (the brine fraction reached 10 %).In the western outer shelf of the Laptev Sea, the surface waters were undersaturated in CO 2 relative to atmosphere; the lowest values were observed furthest to the north over the deep basin.This was the region dominated by Atlantic origin water mixed with MW (the MW fraction increased by up to 15 %); the temperature was low, as was the fraction of RW (Figs. 4,7).Similar to the eastern Kara Sea, strong correlations were found between pCO 2 and temperature/salinity in the Laptev Sea, the recipient of the large Lena River inflow.The correlation was positive with temperature (R = 0.78) and negative with salinity (R = −0.59).The surface water salinity of the Laptev Sea was lower than that of the Kara Sea (Fig. 4), with the FW source dominated by riverine runoff (Fig. 7).Thus, higher pCO 2 values were found in the Laptev Sea than in the Kara Sea, including values as high as supersaturation.
Consequently, large areas of CO 2 outgassing to the atmosphere were identified in the Laptev Sea, areas that might increase their strength as the climate warms.This is because when the permafrost thaws, more terrestrial OM will be exposed to microbial mineralization, producing CO 2 both within the drainage basins and in the river plume within the shelf sea.Photochemical transformation of terrestrial dissolved OM and direct photomineralization of OM also has an effect on increasing concentrations of CO 2 in surface waters (Bélanger et al., 2006;Fichot and Benner, 2014).The effect is strengthened by increased river discharge and by coastal erosion as a result of increasing water temperature and intensified wind and wave activity when the sea-ice cover decreases (Serreze et al., 2007;Shakhova et al., 2014Shakhova et al., , 2015)).Increasing ice-free water area leads to lower albedo and increased surface layer heating.Moreover, the supply of large quantities of optically active dissolved OM and suspended material (Dudarev et al., 2006;Sánchez-García et al., 2014;Vonk et al., 2014;Pugach and Pipko, 2013;Pugach et al., 2015;Charkin et al., 2015) also promotes the accumulation of solar radiation in the surface layer, which increases the heat content, leading to further lengthening of the ice-free season (Granskog et al., 2007(Granskog et al., , 2015;;Hill, 2008;Logvinova et al., 2016).Thus, water heating leads to additional increasing pCO 2 in the surface water.We also suggest that progression of sub-sea permafrost thawing and decrease in ice extent could result in a significant increase in carbon discharge from the sea floor (Nicolsky and Shakhova, 2010;Shakhova et al., 2014Shakhova et al., , 2015Shakhova et al., , 2017;;Vonk et al., 2014) producing additional CO 2 .

The East Siberian Sea
The surface water pCO 2 was in equilibrium with the atmosphere or slightly higher in the well-stratified waters of the East Siberian Sea.Supersaturation was observed not only in the shallow shelf sea, as previously described in literature (Anderson et al., 2009(Anderson et al., , 2011;;Pipko et al., 2005Pipko et al., , 2011b;;Semiletov et al., 2007Semiletov et al., , 2012)), but also in the adjacent deepwater area.The likely causes of the detected pCO 2 distribution are the anomalous dynamics of atmospheric processes, in particular the deep low-pressure area over land and highpressure area over the ocean, as well as the sharp reduction in sea-ice coverage.This led to RW being transported far to the north and northeast of the eastern Arctic shelf and to intensive warming of the surface layer.These waters carried a large amount of terrestrial OM to the deep sea, which would be further mineralized to CO 2 (Alling et al., 2010;Semiletov et al., 2016).
Surface water pCO 2 was somewhat higher at the oceanographic transect near the Lomonosov Ridge (the New Siberian Islands slope) than at the East Siberian Sea slope transect further to the east (Fig. 4); the pCO 2 of the East Siberian Sea slope transect was determined by different FW sources than was the pCO 2 of the Lomonosov Ridge.River runoff dominated as the FW source to the western transect, while sea-ice MW contributed a considerable volume to the eastern transect (Fig. 7).In addition, the use of a fourcomponent mixing model reveals the possible presence of significant concentrations (up to 25 % or more) of Pacific- derived waters at the eastern transect (Abrahamsen et al., 2009;Bauch et al., 2011).The possibility of Pacific-derived waters penetrating to the eastern slope of the Lomonosov Ridge has been discussed before (Makhotin, 2010 and references therein).The climatological circulation during 1997-2006 that was reconstructed using the 4-D variational approach shows the trajectories of several groups of Lagrangian particles (Shakhova et al., 2015); passive particles (or elements of the surface water mass) which originated in the eastern part of the East Siberian Sea between 160 and 170 • E (the area impacted by the Pacific-derived waters) took less than 2 years to be transported to the Lomonosov Ridge area.

Dynamics of air-sea CO 2 fluxes in 2007
The air-sea CO 2 flux was computed with high spatial resolution along the cruise track from the Barents Sea to the East Siberian Sea (Fig. 9).Most of the studied areas of the icefree waters served as a sink for atmospheric CO 2 .Regions of intensive terrestrial impact in the Laptev Sea and the East Siberian Sea were the exception, and they acted as a weak source to the atmosphere (Fig. 9).The Barents Sea is the strongest CO 2 sink in the Arctic region, yet estimates of the air-sea CO 2 flux in this area show large variability depending on subregion, season, and type of data used in the calculations (Lauvset et al., 2013).Using hourly averaged wind and cubic gas-transfer velocity parameterization (Wanninkhof and McGillis, 1999), we estimated the air-sea CO 2 flux in the least-explored subregion of the sea.The highest rates of CO 2 uptake were found in the northeastern Barents Sea (160.9 mmol m −2 day −1 ), where the high pCO 2 (around −150 µatm) coincided with high daily wind speed (15 m s −1 ).It should be noted that air-sea CO 2 fluxes were low, on the order of 10 mmol m −2 day −1 , in the region between 35 and 45 • E where the difference of pCO 2 values between sea and air was maximal (more than 150 µatm; Fig. 9), a result of the very low wind speed experienced here.A second region of the highest CO 2 invasion was detected in the northwestern Kara Sea, despite the fact that in this region pCO 2 was almost 2 times lower (Fig. 9).Furthermore, the average CO 2 uptake rate was significantly greater in the Kara Sea than in the Barents Sea (50.7 mmol m −2 day −1 vs. 27.9mmol m −2 day −1 ; Table 1), despite the fact that average pCO 2 was about half as much in the Kara Sea as it was in the Barents Sea (−46 µatm vs. −94 µatm).The reason for this discrepancy is the wind speed, which averaged 9 m s −1 in the Barents Sea and 14 m s −1 in the Kara Sea (Fig. 9, Table 1).Hence, it is obvious that it is the wind speed that yields the large spatial flux patchiness, because this parameter is much more variable in both time and space than is pCO 2 .In the eastern part of the Kara Sea, the CO 2 flux into the ocean was lower when the surface water pCO 2 was higher due to river discharge influence, which coincided with the weakening of the wind (Fig. 9).
In order to compare our estimates with those calculated by Lauvset et al. (2013), which carefully assessed the seasonal cycle of air-sea CO 2 fluxes in the Barents Sea, daily wind speed and quadratic parameterization of gas-transfer velocity (Wanninkhof, 1992) were used for calculating CO 2 fluxes in the northern Barents Sea.The CO 2 flux into the seawater during fall 2007 reached an average of 126 g C m −2 year −1 and varied from 6 to 501 g C m −2 year −1 .Determined by low water temperature and high wind speed, the obtained values were close to the maximum average CO 2 uptake in the southern and central Barents Sea in highly productive spring months (April and May) (Lauvset et al., 2013).As the dataset by Lauvset et al. (2013) did not cover the north part of the  sea comprehensively, the data obtained during our cruise add information enabling a more accurate estimation of the absorption capacity of the whole Barents Sea in the fall season.
As noted before, there are both CO 2 sink and source regions in the Laptev and the East Siberian seas (Fig. 9).The southern Laptev Sea and the northwestern East Siberian Sea, where the terrestrial influence was significant and surface layer temperatures were the highest, served as a weak source of CO 2 to the atmosphere.However, even if there were regions of large pCO 2 , the fluxes were quite small (Fig. 9) which is a consequence of low winds; as a result, the investigated area of the Laptev Sea as a whole was a weak sink for atmospheric CO 2 while that of the East Siberian Sea was a weak source of CO 2 to the atmosphere (Table 1).

3.5
The interannual variability of pCO 2 and air-sea CO 2 fluxes 3.5.1The Barents and Kara seas The surface layer of the Barents Sea was permanently undersaturated with respect to CO 2 and the ice-free waters were a sink of atmospheric CO 2 , although the CO 2 flux was limited in the presence of sea-ice cover (Figs. 3,5,9, Table 1).
The lowest surface water pCO 2 was observed in 2009 under temperature conditions close to freezing; the mean value (170 µatm) was less than half that of the atmosphere.The mean pCO 2 value was higher in 2009 (189 µatm), with maximal magnitude in 2007 (280 µatm).Nevertheless, the CO 2 uptake was higher in 2006 and even higher in 2007 (Table 1).This is an effect of less sea-ice cover during investigations in 2006 relative to 2009 (sea-ice concentration in the study area was ∼ 50 and 80 %, respectively) and lack of sea-ice cover in 2007.The higher wind speed in 2007 was an additional driver which increased CO 2 uptake.The Kara Sea was also a sink of CO 2 in 2006, 2007, and 2009, but to a variable degree, from close to zero in sea-icecovered or RW-influenced areas to −170 mmol m −2 day −1 in western ice-free waters.The surface water pCO 2 at the two stations carried out in 2006 in the eastern part of the Kara Sea was slightly above 250 µatm (mean value of 259 µatm).Note that RW was not found in the eastern Kara Sea in 2006 (Fig. 7).That was determined by cyclonic atmospheric circulation (Fig. 2), which prevented spreading of the Ob and Yenisei river waters far to the east; sea-ice MW was the main source of the FW in this region.The higher pCO 2 in 2007, which increased from 284 to 372 µatm, was associated with higher sea surface temperature and the presence of RW in the eastern part of the sea (Figs. 4,7).Minimal surface pCO 2 (mean value of 206 µatm) was revealed in 2009 (Fig. 5).Nevertheless, the average CO 2 uptake rate was the highest in 2007 (Table 1).Once again, this is a result of the fact that the main part of the study area in 2009 was in ice-covered waters while most of the 2007 cruise was in open water (Fig. 1); the wind was also stronger in 2007.This finding stresses the importance of declining sea-ice coverage and strengthening wind for the ocean's ability to take up atmospheric CO 2 in northern parts of the Barents and Kara seas which are mainly remote from direct terrestrial discharge.

The Laptev and East Siberian seas
The most extensive study over the 3 years was conducted in the Laptev Sea and in the northwestern East Siberian Sea, the region where negative September ice concentration anomalies were most pronounced for the whole Arctic Ocean (ftp://sidads.colorado.edu).Surface waters of the outer shelf and slope in 2006 and 2009 were undersaturated in CO 2 relative to the atmosphere; 2007 was an exception (Table 1).
For comparative evaluation, we selected a transect north of the New Siberian Islands over the Lomonosov Ridge (Figs. 1,10).Note that pCO 2 data of the selected transect, reported in Fig. 10, were calculated for discrete samples (from A T , pH T 20, and inorganic nutrient data) by means of the CO2SYS program of Lewis and Wallace (1998).
The salinity distribution along the transect during the three cruises shows a similar general pattern but with some significant variations especially in the top 30-50 m (Fig. 10).Of the 3 years, 2007 had the lowest surface salinity and the most pronounced halocline (Fig. 10).However, the largest interannual differences were in the seawater temperature distribution.In late summer 2007, the surface layer was the warmest and was underlain by a sharp thermocline coinciding in depth with the halocline to form a strong pycnocline that restricted vertical exchange.A characteristic feature of the vertical distribution of pCO 2 along the transect in late summer 2007 was a pronounced subsurface maximum of pCO 2 (Fig. 10) and higher pCO 2 values in the surface waters.The subsurface maximum was found exactly at the slope, and coincided with a layer of brine-enriched southeastern Laptev Sea bottom waters (Bauch et al., 2011).During years with prevalent offshore wind, such brine-enriched waters are exported to the Arctic Ocean halocline at about 50 m water depth (Bauch et al., 2009(Bauch et al., , 2011)).
Westerly winds during the ice-free period in summer 2007 advected the Lena River plume to the northeast.Thus, the low salinity was mainly related to the effective transfer of RW into the deep ocean due to the wind field and icefree conditions, but salinity was also somewhat impacted by higher river discharge in 2007 (an average of 752, 822, and 738 km 3 for the Lena and Kolyma rivers in 2006, 2007, and 2009, respectively).The calculated RW content in the surface layer reached over 30 % in 2007, but did not exceed 20 % in 2006 or 10 % in 2009 (Fig. 11).Distribution of normalized A T also confirmed the presence of a large amount of RW in 2007 (Fig. 11).The content of brine (indicating negative seaice MW) in the surface water was also significantly higher during summer 2007 compared to the fall of 2006 and2009 (Figs. 7, 11).This part of the Laptev Sea is known as a large sea-ice production region, and thus the brine signature builds up during the winter season.One consequence is that little sea-ice MW was observed, up to a maximum of only 5 %, even though sea-ice melt decreases the brine signal in the summer.The maximum MW fraction was in the low salinity range of the surface water at the southern end of the section in 2006 (Figs. 7, 10, 11).
The substantial impact by river discharge in 2007 was characterized by high pCO 2 , resulting in weak oversaturation in the south of the transect.Thus, pCO 2 conditions (Table 1) favoring CO 2 outgassing into the atmosphere were observed even on the northern edge of the outer shelf (Fig. 10).This situation is not typical for the deep waters of the Laptev Sea because the outer shelf normally acts as a sink for atmospheric CO 2 unlike the middle and inner shelves (Semiletov et al., 2007;Anderson et al., 2009;Pipko et al., 2016).
The pCO 2 in the lowest-salinity surface water was around 270 and 300 µatm in 2006 and 2009, respectively, although the runoff source ranged between 10 and 20 %.The higher CO 2 content of river runoff in 2007 not only contributed to high pCO 2 but also enhanced surface water temperature due to its high concentration of CDOM, which adsorbs solar radiation (Pugach et al., 2015;Semiletov et al., 2013).Except for the lower runoff content compared to 2007, the main con-tribution to the pCO 2 interannual variability was seawater temperature that was about 4 • C higher in 2007 (Fig. 10).
In summary, the strength and direction of air-sea CO 2 fluxes on the outer shelf and continental slope of the Laptev and the East Siberian seas varied significantly among the years (Figs. 4,5,9, Table 1).The area was a sink of atmospheric CO 2 in 2006 and 2009, and the surface water of the East Siberian Sea was a weak source in 2007 (Table 1).That change from sink to source was related to the distribution of the RW plume on the shelf, the transport of terrestrial OM (and its oxidation to CO 2 ) by this plume, and the plume's impact on water temperature, as well as sea-ice extent and wind speed (Fig. 1, Table 1).It should be noted that pCO 2 was higher in the deep regions of the Laptev and East Siberian seas than in the deep regions of the Kara Sea, even if the river discharge is higher in the Kara Sea than in the ESAS.This is likely a combination of the dominating flow of the river plume and the source pCO 2 in the water that mixes with the runoff; the water in the Kara Sea comes from the Barents Sea with its low pCO 2 , while the Laptev Sea is dominated by inflow from the Kara Sea.

Conclusions
This 3-year study of the outer shelf and the continental slope waters of the Eurasian Arctic seas has revealed a general trend in the surface pCO 2 distribution, which manifested as an increase in pCO 2 values eastward, from the surface waters of the highly productive Barents Sea to the poorly productive eastern Laptev Sea and western East Siberian Sea, which are strongly influenced by terrestrial runoff and coastal/sub-sea permafrost erosion.It has been shown that the influence of terrestrial discharge on the carbonate system of ESAS surface waters is not limited to the shallow shelf.Furthermore, during certain meteorological conditions, the surface waters of the outer shelf, as well as those of the continental slope of the ESAS, can become supersaturated with respect to atmospheric CO 2 .
Contemporary climate change affects air temperature in the Arctic region leading to sea-ice reduction and permafrost thawing, both on land and offshore.Increasing air temperatures cause increased water temperature and strengthened wind activity, which intensify water mass dynamics and airsea exchange.It was shown that these changes in the Arctic climatology can affect the capacity of this region to serve as a source or a sink for atmospheric CO 2 in two opposite ways.On the one hand, larger areas of open water due to sea-ice reduction and longer ice-free periods can cause the outer shelf and slope of the west Siberian Eurasian Arctic seas (Barents and Kara seas) to develop a growing capability to absorb atmospheric CO 2 ; on the other hand, growing river discharge and degradation of permafrost, associated with thermal erosion of coasts and river banks, can increase the effectiveness of the ESAS to act as a CO 2 source due to increased terres-trial export of labile eroded carbon, a significant portion of which oxidizes to CO 2 .This was the situation in the deep regions of the Laptev and East Siberian seas in 2007, when sea-ice decline was especially pronounced, resulting in an increase of the area where seawater pCO 2 was in equilibrium or slightly supersaturated with respect to atmosphere and a reduction of CO 2 absorption in the ESAS.
This study has shown that contemporary climate change impacts the carbon cycle of the Eurasian Arctic Ocean and influences air-sea CO 2 flux.It also highlights the importance of considering small-scale variations in meteorological and hydrological parameters, varying both in time and space, for estimating the air-sea exchange of CO 2 .During these times of rapid environmental changes, results of this study stress the need for comprehensive multi-year investigations of dynamic deep-sea regions in order to estimate current and predict future capacity of the Arctic basin as a sink for atmospheric CO 2 based on high-resolution spatial coverage of the Arctic Ocean.

Figure 1 .
Figure 1.Ship routes and positions of oceanographic stations in the study area.(a) Positions of oceanographic stations performed in 2006, 2007, and 2009 are marked as colored circles: 2006 -blue; 2007 -red; 2009 -green.Positions of the sea-ice edge during the expeditions are marked as colored curved lines: 2006 -blue; 2007 -red; 2009 -green.Panel (b) indicates the ship's route, along which the high-frequency measurements were performed in 2007.

Figure 6 .
Figure 6.Relationship between pCO 2 (µatm) and temperature ( • C) (a), and between pCO 2 and salinity (b) in the northern Barents Sea during the 2007 study: western part -W, orange color; eastern part -E, blue color.

Figure 8 .
Figure 8. Relationship between pCO 2 (µatm) and temperature ( • C) (a), and between pCO 2 and salinity (b) for samples collected in the western (W, orange color) and eastern (E, blue color) Kara Sea during the 2007 study.
each parameter is shown in parentheses; n is the the number of measurements.b Flux calculated according toWanninkhof and McGillis (1999); negative values correspond to CO 2 flux into the ocean.

Table 1 .
Mean values and standard deviations for pCO