Seasonal Water Mass Evolution and Non‐Redfield Dynamics Enhance CO2 Uptake in the Chukchi Sea

Abstract The Chukchi Sea is an increasing CO2 sink driven by rapid climate changes. Understanding the seasonal variation of air‐sea CO2 exchange and the underlying mechanisms of biogeochemical dynamics is important for predicting impacts of climate change on and feedbacks by the ocean. Here, we present a unique data set of underway sea surface partial pressure of CO2 (pCO2) and discrete samples of biogeochemical properties collected in five consecutive cruises in 2014 and examine the seasonal variations in air‐sea CO2 flux and net community production (NCP). We found that thermal and non‐thermal effects have different impacts on sea surface pCO2 and thus the air‐sea CO2 flux in different water masses. The Bering summer water combined with meltwater has a significantly greater atmospheric CO2 uptake potential than that of the Alaskan Coastal Water in the southern Chukchi Sea in summer, due to stronger biological CO2 removal and a weaker thermal effect. By analyzing the seasonal drawdown of dissolved inorganic carbon (DIC) and nutrients, we found that DIC‐based NCP was higher than nitrate‐based NCP by 66%–84% and attributable to partially decoupled C and N uptake because of a variable phytoplankton stoichiometry. A box model with a non‐Redfield C:N uptake ratio can adequately reproduce observed pCO2 and DIC, which reveals that, during the intensive growing season (late spring to early summer), 30%–46% CO2 uptake in the Chukchi Sea was supported by a flexible stoichiometry of phytoplankton. These findings have important ramification for forecasting the responses of CO2 uptake of the Chukchi ecosystem to climate change.

In addition, the seasonal evolution of water masses associated with variations in biological processes plays a key role in controlling nutrients and carbon cycles in the Chukchi Sea. High primary production associated with Pacific Water inflow and sea ice retreats transforms the Chukchi Sea from a nutrient-rich system in spring to a nutrient-limited system in summer Zhuang et al., 2020). Such a substantial change in nutrient availability potentially changes the uptake ratio of nutrients and carbon by phytoplankton to form particulate organic matter (POM) and dissolved organic matter (DOM) during primary production and complicates our understanding of the seasonal net community production (NCP) and CO 2 uptake from the atmosphere in the Chukchi Sea (Tu et al., 2021;Zheng et al., 2021). Deviations of C:N uptake stoichiometry from the Redfield ratio have been widely observed over the world's oceans (Koeve, 2006;Martiny et al., 2013;Park et al., 2008). Several model studies have suggested that non-Redfield stoichiometry of phytoplankton should be taken into account to better project the net primary production, food quality and ocean carbon uptake in response to climate change (Buchanan et al., 2018;Kwiatkowski et al., 2018). However, a variable C:N uptake stoichiometry within a seasonal scale is rarely investigated and it is much less known how a non-Redfield C:N assimilation ratio of phytoplankton may affect NCP and CO 2 uptake in a climate-sensitive region such as the Chukchi Sea.
As the Chukchi Sea is one of the fastest changing regions in the world due to anthropogenic climate changes (Lewis et al., 2020;Meredith & Sommerkorn, 2019;Onarheim et al., 2018;Stroeve & Notz, 2018;Woodgate & Peralta-Ferriz, 2021), it is important to better understand the fundamental drivers, causes, and patterns in seasonal biogeochemical dynamics. Although many efforts have been made to demonstrate the seasonal variation in CO 2 fluxes and biogeochemistry in the Chukchi Sea (Ardyna & Arrigo, 2020;Lowry et al., 2015;Ouyang et al., 2021;Zheng et al., 2021), most of them were based on observations that were either within a single season or over multiple seasons across different years. Given that sea ice conditions in the Chukchi Sea vary from year to year, the timing and magnitude of physical and biological processes change correspondingly, perhaps causing a large uncertainty in deconvoluting the complicated relationship between seasonal air-sea CO 2 flux, NCP, and carbon and nutrient consumption using data collected across different years. Therefore, measurements of biogeochemical properties across multiple periods within a single year provide us a great opportunity to elucidate the links between seasonal evolution of water masses and the associated biogeochemical processes and ecosystem responses, which are essential for forecasting the responses of the Chukchi ecosystem to climate change.
Here, we present a data set of underway sea surface temperature (SST), salinity (SSS) and sea surface CO 2 partial pressure (pCO 2 ) measured in five consecutive cruises in the Chukchi Sea from spring to fall in 2014 (Table 1). We take advantage of this unique data set to examine both the spatial and seasonal variations in sea surface pCO 2 , and then explore the dominant drivers of seasonal pCO 2 change along water mass evolution. We also use discrete sample-based biogeochemical properties including DIC, total alkalinity (TA), nutrients (NO − 3 and PO 3− 4 ), dissolved oxygen (DO), and chlorophyll a (Chl a) from three cruises to identify how seasonal evolution of water masses affect biogeochemical dynamics. In addition, we calculated NCP based on seasonal changes in DIC and nutrients and discussed possible mechanisms for the observed inconsistency in the derived NCP values. Finally, we used a box model to evaluate the impacts of non-Redfield C:N uptake by phytoplankton on seasonal CO 2 uptake in the Chukchi Sea.

Study Area and Water Masses
The water mass properties and circulation in the Chukchi Sea have been extensively studied and presented in many previous works (Corlett & Pickart, 2017;Gong & Pickart, 2015;Li et al., 2019;Lin et al., 2019;Pacini et al., 2019;Pickart et al., 2019;Stabeno et al., 2018). Briefly, three main current branches of Pacific Water flow northward through Bering Strait onto the Chukchi Sea shelf (Figure 1). On the eastern side, the ACW branch flows along the Alaskan coast toward Barrow Canyon (known as the Alaskan Coastal Current (ACC) in summer and fall). The central branch flows through Central Channel between Herald Shoal and Hanna Shoal (Weingartner Cruise Period Vessel Data source of sea surface pCO 2 Data source of discrete samples  (Weingartner et al., 2005). However, there is increasing evidence that the flows are not as rigidly confined to these three separate pathways as was originally thought. For example, a portion of western branch flows eastward to the north of Herald Shoal, where it meets the central branch before joining the eastern branch and exiting the shelf via Barrow Canyon ( Figure 1). It is evident that topographic features such as canyons and shoals greatly affect the shelf circulation.
To focus on the Chukchi shelf, data used for this study were limited from 66°N to 74°N to a maximum of 120 m to exclude deep waters from the Canada Basin. We adopted the water mass classification used by several previous studies (Table 2; Corlett and Pickart et al., 2017;Gong & Pickart, 2015;Pacini et al., 2019;Pickart et al., 2019;Wang et al., 2022). The coldest water is known as Newly-Ventilated Winter Water (NVWW; T < −1.6°C), which is typically observed from spring to early-summer over the entire shelf. As summer progresses, this winter water can be converted to Remnant Winter Water (RWW; −1.6°C < T < 0°C) by solar heating or mixing with warmer summer water, which sometimes lasts into summer (Gong & Pickart, 2015). During summer and early fall, the Chukchi Sea mainly contains two Pacific summer waters. The first is ACW, which originates in the Gulf of Alaska. In the summer, the ACC carries predominantly ACW in the eastern Chukchi Sea, in which the water temperature can reach >10°C. The second is a combination of Anadyr Water and central Bering Shelf water, which is referred to as Bering Summer Water (BSW). The seasonal sea ice Melt Water (MW) forms a layer of fresh water on the Chukchi shelf. Sometimes it can also be divided into early-season melt water and late-season melt water (Table 2; Gong & Pickart, 2015;Pickart et al., 2019). This freshwater layer evidently occupies a large portion of the Chukchi/Beaufort shelfbreak regions in the mid-to late-summer. In these regions, the densest and saltiest water, the Atlantic Water (AW; T > −1.0°C and S > 33.6), can be observed at depths below ∼150 m, but it is rarely observed on the Chukchi shelf. However, AW can be upwelled occasionally through Barrow Canyon (Pisareva et al., 2019) and Herald Canyon (Pickart et al., 2010).

Underway Measurements
It is unusual that there are five survey cruises in 1 year (2014) with underway sea surface temperature (SST), salinity (SSS) and pCO 2 measured in the Chukchi Sea (Table 1). To our knowledge, this is the only year so far with such high-density observations. The underway data were obtained from the SOCAT database (v2020; Bakker et al., 2016) and Japan Agency for Marine-Earth Science and Technology (JAMSTEC) ( Table 1). Specifically, the underway SST and SSS were measured using an underway water monitoring system in an intake port near the bow of the ships (5-7 m below the waterline). The underway sea surface pCO 2 from USCGC Healy and RV Xuelong was measured using an underway pCO 2 system (General Oceanic, USA) described in Pierrot et al. (2009). The data from RV Mirai was analyzed by a Greenhouse Gas Analyzer (Los Gatos Research, USA). These systems were monitored and calibrated with four certified gas standards (NOAA) every 2-3 hr. The overall precision of the reported sea surface pCO 2 values was estimated to be ± 2 atm. The underway pCO 2 system and data reduction procedure were described by Pierrot et al. (2009).

Discrete Sample Analysis
Discrete biogeochemical samples were collected during three cruises (Table 1). The datasets of discrete samples included nutrients (NO 3 − and PO 4 3− ), dissolved oxygen (DO), Chl a, DIC and TA. The data, collected from USCGC Healy in spring, from RV Xuelong in summer, and from RV Mirai in fall, have been archived in publicly accessible data centers (Table 1).
In the spring cruise, nutrients, DO and Chl a were measured onboard. Nutrient samples were determined using a Seal Analytical continuous-flow AutoAnalyzer 3 following a modified procedure after Armstrong et al. (1967). DO was measured using a spectrophotometer following the standard Winkler titration method, with a precision of ±1 μmol kg −1 . Chl a measurements were made using a Turner 10-AU fluorometer (Turner Designs, Inc.) (Holm-Hansen et al., 1965) and were calibrated against a pure Chl a standard (Sigma). See more description of sampling and analysis of nutrients, DO, and Chl a in Arrigo et al. (2017). DIC and TA samples were collected into 250 mL borosilicate glass bottles poisoned with 100 μL saturated HgCl 2 solution to prevent biological activity following the procedure described in Dickson et al. (2007). DIC and TA samples were analyzed at the University of Southampton, UK via coulometric (DIC) and potentiometric (TA) titration methods using a VINDTA 3C (Marianda), with a precision of ±2 μmol kg −1 .
In the summer cruise, nutrient data were measured onboard using a San ++ automated continuous flow autoanalyzer with a precision of ±2% (SKALAR Inc., Netherlands). The detection limits were 0.1 μM for NO − 3 and 0.03 μM for PO 3− 4 . Chl a samples were measured following similar procedures to the spring expedition. DO was measured onboard using the spectrophotometric method based on Winkler titration with a precision of ±1 μmol kg −1 . See more descriptions of sampling and analysis of nutrients and Chl a samples in Zhuang et al. (2020). DIC and TA samples were collected following similar procedures described above, and then were analyzed at the University of Delaware, USA. DIC samples were analyzed using a LiCOR LI 7000 infrared CO 2 detector coupled to an automated DIC analyzer (Apollo SciTech, USA). TA samples were measured via open-cell potentiometric titration system (Apollo SciTech, USA). The precision of the DIC and TA measurements was ∼0.1% (∼± 2 μmol kg −1 ; Chen et al., 2015).
In the fall cruise, nutrients were analyzed using QuAAtro system. DO was measured via the spectrophotometric Winkler titration method (Kimoto Electric Co. LTD) following the World Ocean Circulation Experiment Hydrographic Program method (Dickson, 1996). The precision of DO measurement was 0.23 μmol kg −1 (Kimoto Electric Co. LTD). Chl a was measured using a fluorophotometer (Turner Designs) (Welschmeyer, 1994). Nutrient samples were analyzed following the GO-SHIP Repeat Hydrography Manual (Hydes et al., 2010) using reference materials for nutrients in seawater (Sato et al., 2010). The precision was 0.08% for NO − 3 and 0.14% for PO 3− 4 . See more descriptions of sampling and analysis of nutrient, DO, and Chl a in Nishino et al. (2020). DIC was measured using an automated coulometric analyzer (Nippon ANS, Inc.) with a precision of ±0.7 μmol kg −1 and TA samples were measured using a custom spectrophotometric system (Nippon ANS, Inc.) with a precision of ±0.57 μmol kg −1 . Note that DIC and TA samples collected in all three cruises were calibrated using certified reference materials (CRMs) from Scripps Institution of Oceanography.
In addition, discrete pCO 2 was calculated with surface TA and DIC (<10 m) to increase the data coverage over temporal and spatial scales. The pCO 2 was calculated by the seacarb package in R (Gattuso et al., 2018) with carbonate dissociation constants of Millero et al. (2006), as recommended by Evans et al. (2015). The uncertainty of pCO 2 values computed from TA and DIC is estimated to be ± 18 atm (RMSE) with a mean systematic difference from the observed pCO 2 of 0.5 atm (N = 111).

Decomposition of Seasonal pCO 2 Change
To explore the dominant drivers for seasonal pCO 2 change in each water mass, we decomposed the net change of pCO 2 into thermal and non-thermal effects, assuming that these two effects altered pCO 2 within a time frame much shorter than the residence time of the water mass on the Chukchi shelf. Therefore, the net change in pCO 2 6 of 28 (Δ (pCO 2 )) could be calculated as the difference between the initial pCO 2 (pCO 2ini ) under the ice in spring (538 ± 68 μatm) and the observed pCO 2 (pCO 2ob ) at a given sampling time: (1) The potential thermal effect on pCO 2 change (Δ(pCO 2 ) T ) was driven by the seasonal variation in SST and was quantified by adjusting the initial pCO 2 to the observed temperature (Takahashi et al., 1993): where T obs is the observed temperature and T spring is the mean temperature observed under the ice in spring (−1.60°C). The non-thermal effect on pCO 2 change (Δ(pCO 2 ) nT ) was mainly driven by the seasonal biological CO 2 drawdown and ice melt, and was calculated as the difference between the observed total net seasonal change and the thermal effect:

Seasonal CO 2 Flux Calculation
To avoid calculational bias induced by the overweighted impact of highly dense data point concentrated within a small area, we averaged the data points into daily-grids of 0.25° latitude × 0.25° longitude, and then we calculated seasonal air-sea CO 2 flux for each grid following: where K s is the solubility of CO 2 and k CO2 is the CO 2 gas transfer velocity. K s was calculated using underway SST and SSS (Weiss, 1974). The value of k CO2 (cm/hr) is estimated from the second moment of wind speed at 10 m above the sea surface, <U 10 2 > (Wanninkhof, 2014): To calculate <U 10 2 >, we used the wind product from the NCEP-DOE Reanalysis 2 data set (https://www.esrl. noaa.gov/psd/data/gridded/data.ncep.reanalysis2.html). A regional mean of <U 10 2 > over the study period was derived from the 6-hr wind speed squared. The term ice% is the sea ice concentration, which is averaged from daily ice% data obtained from the Scanning Multichannel Microwave Radiometer (SMMR) on the Nimbus-7 satellite and the Special Sensor Microwave/Imager (SSM/I) sensors on the Defense Meteorological Satellite Program's (DMSP)-F8, -F11, and -F13 satellites (https://nsidc.org/data/nsidc-0079; Comiso, 2017). For the area with heavy ice covered, we adopted Takahashi et al. (2009) approach that for sea ice where satellite observation is >90% ice cover, a 10% open water area is assumed. In this way, 10% of the open water persisted for CO 2 flux in the area with >90% ice cover where satellite data with resolution of 25 km × 25 km fails to resolve fine scale structure, such as leads.

Estimates of Net Community Production
The NCP can be estimated from the nutrient and DIC changes in the upper mixed layer. Since the residence time of water masses on the Chukchi Sea shelf (3-7.5 months) (Stabeno et al., 2018;Woodgate & Peralta-Ferriz, 2021) is longer than the timespan of a phytoplankton bloom (<1-2 months), we can assume that water masses move as a single parcel over the shelf, where physical (mostly warming and freshening) and biogeochemical changes occur over time. Therefore, at any particular location, the observed drawdown of nutrients and DIC in the surface mixed layer can be attributed to NCP. However, nutrient and DIC concentrations are not only influenced by NCP, but also by meltwater/river-runoff dilution, and in the case of DIC, by air-sea CO 2 exchange and CaCO 3 dissolution and formation. Thus, to quantify the biological-induced changes in nutrients, DIC, and TA, we normalized all nutrient (NO 3 − and PO 3− 4 ), DIC and TA concentrations to a reference salinity (S 0 , i.e., mean salinity under the ice in spring) to remove the impact of Arctic Ocean freshwater input into the Pacific winter water. As river discharge impact in spring and early summer is either limited to the estuarine and nearshore areas within the Arctic or treated as part of the Pacific water endmember (Text S1 and Figure S1 in Supporting Information S1), the salinity change in Chukchi Sea is mainly attributed to ice melt dilution for period of interest (spring to early summer); thus, non-zero meltwater endmembers are used (Friis et al., 2003;Jiang et al., 2008): The meltwater salinity, DIC and TA values are set as 5, 400 μmol kg −1 , 460 μmol kg −1 (Rysgaard et al., 2007), respectively, which are equivalent to DIC S=0 = 60 μmol kg −1 and TA S=0 = 106 μmol kg −1 (Cai et al., 2010;Ouyang et al., 2020). The contributions of meltwater to NO − 3 (μM) and PO 3− 4 (μM) concentrations are negligible (Clark et al., 2020).
The deficit of nNO − 3 (or nPO 3− 4 ) is the result of phytoplankton uptake between the time before the phytoplankton bloom begins (t 1 ) in spring and the time of sampling (t 2 ). Then, the vertical integral of the deficit of nNO − 3 (or nPO 3− 4 ) within surface mixed layer is assumed to be equal to NCP at any given location over the time interval between t 1 and t 2 (Δt ), Analogue equations of Equations 10 and 11 can be written for nPO 3− 4 . The average mixed layer depth (MLD) of 25 m was used in this study. While this is a highly simplified treatment, the MLD uncertainty would apply equally to C-, N-, and P-based NCP. The estimates of nutrient-based NCP were then converted to carbon units by multiplying by a M C:N:P uptake ratio of 106:16:1 (Arrigo et al., 2017;Codispoti et al., 2013;Hansell et al., 1993;Redfield, 1958).
The nDIC change reflects the influence of both phytoplankton uptake, air-sea CO 2 exchange, and CaCO 3 dissolution and formation. Therefore, we calculated biological-induced nDIC change by subtracting the DIC added by the CO 2 influx and possible CaCO 3 formation and dissolution (CaCO 3 + H 2 O + CO 2 ↔ Ca 2+ + 2HCO 3 − ): where is the water density and CO 2 flux term was calculated following Equation 4. The term of ΔDICCaCO3 can be quantified as half of the change in salinity-normalized TA (nTA) corrected for the removal of NO 3 − to form organic matter (Brewer & Goldman, 1976;Lee, 2001): Note that due to lack of information we are not able to differentiate biological calcification and the dissolution of CaCO 3 mineral (ikaite) trapped in the melting ice.

Box Model
The goal of the box model is to investigate how a flexible C:N uptake ratio by phytoplankton affects DIC fixation and air-sea CO 2 fluxes in the Chukchi Sea. Similar box model approaches have been described in previous studies (Ouyang et al., 2020. Briefly, the initial TA and DIC values were set to the mean value of observations in the surface mixed layer in late spring, and the initial sea surface pCO 2 was calculated from the initial TA and DIC using the seacarb package in R (Gattuso et al., 2018). TA in the subsequent season was linearly interpolated based on observations. For each simulation step, sea surface pCO 2 was calculated from TA and DIC at the corresponding step. The change in DIC inventory (∆DIC) in the surface mixed layer was calculated as where FCO 2t , NCP t , and ∆DIC (diluted)t indicate the changes in DIC inventory in the mixed layer induced by the CO 2 air-sea flux, NCP, and dilution by meltwater at simulation time step t, respectively. In this case, the simulation time interval ( ) is 1 day. The surface seawater density, , was calculated using SST and SSS. The FCO 2 term was calculated with Equation 4. The ∆DIC (diluted)t term was computed by simplifying the ice melt dilution and assuming that the ratio of TA/DIC in the thin first-year ice nearly equals that in the surface seawater (Rysgaard et al., 2007). Thus, Because the Chukchi Sea is N-limited during the growing season , NCP t was calculated based on nitrate; NCP N-based . Two scenarios with different C:N uptake ratios were examined in the box model. For the fixed-stoichiometry scenario, a C:N uptake ratio of 6.6 (Redfield ratio) was used. For the non-Redfield scenario, much higher C:N uptake ratios were used (i.e., 10.9 and 12.1 in Table 5). Therefore, And DIC at time step t + 1 was iteratively calculated as follows: With the new DIC and TA, a new pCO 2 was calculated for the next simulation step, and this simulation process repeated until the last day of the simulation.

Sea Surface Temperature, Salinity, and pCO 2
Underway SST, SSS, and pCO 2 were measured during five cruises in the Chukchi Sea, from May through September in 2014, providing a consistent way to examine the seasonal transition of the ocean from a CO 2 source to a sink and to explore the role of changing water masses.
Because sea ice melt began in the southern portion of the Chukchi Sea in early June, we divided the spring observations into two periods: early spring (15-31 May) and late spring (1-21 June 1). Clearly, SST distribution was greatly affected by sea ice (Figure 2a). SST was near freezing (−1.67 ± 0.07°C) under the ice in the north, while SST in the south was slightly warmer (−1.28 ± 0.28°C). The location of SST fronts corresponded well with the edge of heavy ice cover around 69.5°N (ice concentration ∼80%). However, the distribution of SSS was not as closely correlated with ice as SST was (Figure 2b). In the southern portion of Chukchi Sea, the surface water was relatively fresher due to mixing with newly melted water and river discharge. The saltier water appeared in the eastern and central Chukchi Sea under the ice. Interestingly, relatively warmer and fresher water was also found in the northeast part of Chukchi Sea toward the shelfbreak (Figures 2a and 2b), which was likely due to local ice melt events. The average pCO 2 in the south was 309 ± 50 μatm, which was much lower than that in the north (536 ± 74 μatm). Such a contrasting pattern in pCO 2 indicated that the southern ice-free Chukchi Sea had already become a CO 2 sink while the northern ice-covered Chukchi Sea still remained supersaturated with respect to atmospheric CO 2 (∼400 μatm) and was a potentially strong CO 2 source in early spring ( Figure 2c).
As sea ice retreated northwards in the late spring (Figures 2d-2f), more warm and fresh waters advected from south to north. However, their impacts were confined by sea ice. The SST, SSS, and pCO 2 values in the ice-covered OUYANG ET AL.
10.1029/2021JC018326 9 of 28 10 of 28 area were similar to those earlier in the spring. Only a few observations south of the ice edge showed signals of warmer and fresher Pacific summer waters. Although there were no observations in the southern Chukchi Sea during this period, we expect that water there might have become warmer and fresher with much lower pCO 2 compared to early spring, which was confirmed by observations on the last day (21 June) of the spring cruise in the south of Bering Strait (Figures 2d-2f).
In early summer (9 July-7 August), the ice edge retreated further northwards to around 71-72°N (Figures 3a-3f). The mean SST increased to 6.64 ± 1.22°C in the southern and central Chukchi Sea but remained low (−0.83 ± 1.31°C) under the ice in the northern Chukchi Sea (Figures 3a and 3d). The distribution of SSS was not only affected by the northward advection of water but also by dilution by meltwater. Compared to spring, the mean SSS (31.0 ± 0.7) over much of the southern and central Chukchi Sea had decreased by 0.5-1.0, although this decrease was more substantial (by 3-4) near the shelfbreak (Figures 3b and 3e). The most notable change in the carbonate system between spring and early summer was a widespread reduction in sea surface pCO 2 . The lowest pCO 2 values were found in the vicinity of the Bering Strait in the south and along the ice edge in the northern Chukchi Sea (Figures 3c and 3f).
The Chukchi Sea had become a completely ice-free region in late summer (12 August 12-9 September 9; Figures 3g-3l). SST continued to increase over the entire Chukchi shelf (Figures 3g and 3j). The highest SST appeared in the middle of August along the Alaska coast, reaching up to ∼11°C (Figure 3g). The SSS showed a similar distribution as in early summer, with saltier water in the southern and central Chukchi Sea and fresher water in the eastern and northern portion as a result of massive sea-ice melt and river discharge (Figures 3h  and 3k). Compared to early summer, the mean sea surface pCO 2 in late summer increased from 202 ± 63 μatm to 265 ± 69 μatm but remained much lower than the atmospheric pCO 2 (Figures 3i and 3l). It is noteworthy that the lowest pCO 2 was observed repeatedly north of Bering Strait and in the area surrounding Hanna Shoal (Figures 3i  and 3l).
Between late summer and fall, the mean SST over the entire shelf decreased from 5.42 ± 3.04°C to 3.49 ± 1.73°C, while the mean SSS remained nearly unchanged (Figures 4a and 4b). For most of the Chukchi Sea, surface water pCO 2 was lower than the atmospheric pCO 2 , averaging 283 ± 50 μatm, indicating that the CO 2 sink lasted into fall. However, extremely high pCO 2 started to appear in the northern Bering Sea along the western pathway of Pacific Summer Water (Figure 4c).

Water Column Structure
Discrete samples of biogeochemical parameters were collected during three cruises (Table 1). Since these three cruises conducted survey sections along the Central Channel, we are able to show the seasonal evolution of biogeochemical properties in the water column along a transect from the Bering Strait extending to the shelf break in the northern Chukchi Sea.
In early spring, water columns on the Chukchi shelf were well mixed, exhibiting little vertical structure for most biogeochemical properties ( Figure 5). The exceptions were a subsurface (20-30 m) Chl a maximum in the southern Chukchi Sea (Figure 5g) and the frontal structures of large gradient in DIC and NO − 3 across the shelf breaks (73°N; Figures 5f and 5h). However, latitudinal patterns were significantly different from the south to the north, which was due to different levels of primary production regulated by the timing of ice melting. Seawater temperature and salinity under the ice were within a narrow range of −1.78°C to −1.60°C and 32 to 33, respectively, indicating that during this period, almost all waters on the northern shelf were NVWW (Figure 5d). Low concentrations of Chl a (<1 μg l −1 ), high NO − 3 (>15 μM), and positive apparent oxygen utilization (AOU) beneath the ice suggested that the spring bloom had not yet started there (Figures 5g-5i) and the ambient high DIC was likely accumulated from remineralization in the previous winter water (Figure 5f). In contrast, primary production within RWW had begun in the southern Chukchi Sea, indicated by a much higher Chl a concentration (>4 μg l −1 ), lower NO − 3 , and negative AOU (Figures 5g-5i). As a result, mean DIC in the south was ∼100 μmol kg −1 lower than that in the north (Figure 5f and Table 3).
A few weeks later, as the ice conditions changed, the water at the ice edge became warmer and fresher as a result of local sea ice melt freshening together with atmospheric heating (70.5°N; Figures 6a and 6b). Although most water columns still consisted of NVWW and were well mixed, a two-layer structure appeared in the southernmost stations (Figure 6c), indicating that seasonal stratification of the water column had begun. In addition, the Chl a OUYANG ET AL.

10.1029/2021JC018326
15 of 28 concentration increased to 2.8 ± 2.2 μg l −1 and the previously positive AOU had turned negative, indicating net oxygen production by a phytoplankton bloom (Figure 6i). The late-spring ice retreat further strengthened primary production in the northern Chukchi Sea. In turn, the associated biological CO 2 drawdown resulted in lower DIC in the surface layer ( Figure 6f).
Conditions during the early summer transect were vastly different from the spring conditions ( Figure 7). The fully mixed water column representative of spring conditions had transformed into a clear two-layer structure with a shallow, warm and fresh surface mixed layer with ACW properties separated by a halocline from BSW and winter water near the bottom (Figures 7a and 7d). Strong primary production over the shelf led to a large reduction in DIC and NO − 3 in the surface mixed layer (Figures 7f and 7h; Table 3). Simultaneously, the surface extremely high O 2 (AOU < −100 μmol kg −1 ) coincided with the subsurface Chl a maximum in the north of Bering Strait and over the shelfbreak (Figures 7g and 7i). Interestingly, the surface extremely low NO − 3 (<1 μM) and supersaturated DO extended throughout the water column in the central Chukchi Sea (68°N −70°N Figures 7h and 7i), which were likely induced by local mixing events occasionally breaking the two-layer structure or a mechanical overturning induced by internal wave mixing Nishino et al., 2015).
Although we only have observations in the southern Chukchi Sea and in the Canada Basin (i.e., no data in the central and northern shelf) in late summer, it is likely that the strong two-layer stratification remained in the water column over most of the Chukchi Sea (Figure 8). The highest Chl a concentration (>25 μg l −1 ) repeatedly appeared at the similar location in the north of Bering Strait associated with BSW (68°N; Figures 8d and 8g) but AOU increased from <−100 to ∼ −30 μmol kg −1 (Figure 8i). The surface DIC in the southern Chukchi Sea (probably in the northern Chukchi Sea as well) became higher than in early summer by ∼100 μmol kg −1 (Figure 8f). As the water column stratification strengthened, the gradients in DIC, NO − 3 and AOU between surface and bottom layers became larger (Figures 8f, 8h and 8i; Table 3). The highest NO − 3 (>20 μM) was observed in the bottom waters in the southern Chukchi Sea. Three weeks later in fall, the stratified vertical structure persisted over most of the shelf area (Figures 9a-9d). However, a widespread reduction in Chl a concentrations suggested that primary production had become much weaker. High Chl a concentrations were only observed in the biological hot spot in the southern Chukchi Sea (68°N; Figure 9g), which was present throughout the entire growing season. NO − 3 became depleted in the surface waters (Figure 9h), limiting further phytoplankton growth, while replenishment of NO − 3 was observed in bottom waters in the southern Chukchi Sea. Consequently, surface O 2 approached equilibrium with the atmosphere (AOU was ∼0 μmol kg −1 ) and surface DIC remained the same as it was in late summer (Figures 9f and 9i).

Water Mass Evolution and CO 2 Flux
Seasonal evolution of water masses affects primary production and subsequently air-sea CO 2 exchange. To investigate the correlation between water mass change and ocean CO 2 uptake, we first illustrate the relationship between water mass properties and seasonal change in sea surface pCO 2 in T/S diagrams (Figure 10). In spring, most of the net change in pCO 2 was negative since a high pCO 2 value (i.e., 538 μatm) in the earlier winter water under the ice was used as the reference value (Figure 10a). The high pCO 2 values that exceeded atmospheric pCO 2 were concentrated within the cold and salty NVWW and RWW beneath the ice, which was resulted from remineralization in and vertical mixing of waters from the previous winter ( Figure 10a). As sea ice started to retreat in late spring and early summer, the dilution by meltwater and strong biological CO 2 uptake led to extremely low (<100 μatm) pCO 2 along the ice edge (Figures 3c and 3f) with the largest seasonal reduction in the early-season meltwater (Early-MW; Figure 10b). These two processes primarily triggered the transition from a CO 2 source to a sink within a few weeks. During this period, the average CO 2 flux changed from 1.1 ± 1.1 mmol m −2 d −1 (from sea to air, or a source) to −7.2 ± 3.4 mmol m −2 d −1 (from air to sea, or a sink) in the northern Chukchi Sea (Figures 11a and 11b; Table 4). As the season progressed, the surface NVWW and RWW were largely transformed or replaced by summer Pacific waters (i.e., ACW and BSW). Consequently, the nutrient-rich BSW sustained strong primary production and resulted in another extremely strong CO 2 uptake center in the southern Chukchi Sea north of the Bering Strait (∼68°N, Figure 11c), whereas the warmer (SST > 8°C), relatively nutrient-poor ACW had relatively weaker CO 2 uptake along the eastern coastal area (Figure 11c). We noticed a small area of CO 2 efflux in the eastern Chukchi Sea in late summer ( Figure 11c) and an area of CO 2 source water in the BSW regime in the southern Chukchi Sea in fall (∼69°N; Figure 11d). These high pCO 2 surface waters were likely due to local vertical mixing as energetic currents in this area of shallow depths (<50 m) could break down stratified vertical structure and bring high pCO 2 bottom water up to the surface . The regional mean CO 2 flux in late summer and fall were estimated to range from −18.3 ± 8.9 to −10.5 ± 8.4 mmol m −2 d −1 in the southern Chukchi Sea and from −13.3 ± 6.6 to −21.1 ± 2.9 in the northern Chukchi Sea (Table 4).
Decomposition of the seasonal change in pCO 2 provides more insight into the controlling drivers. Between spring and early summer, the largest reduction in pCO 2 was observed in the early-season MW and BSW, while the pCO 2 reduction in the ACW was relatively less (Figure 10b). Specifically, the warm ACW and river runoff have the largest thermal effect on changing pCO 2 ; a warming of more than 10°C in SST along could potentially increase pCO 2 by ∼300 μatm (Figure 10f). In contrast, thermal effects were the relatively weaker driver for pCO 2 change in both the BSW and MW (Figure 10f). In addition, the nutrient-rich BSW provided a continued nutrient supply for sustaining primary production and biological CO 2 drawdown along the Central Channel pathway. The strong reduction of pCO 2 within BSW and MW domains (Figure 10j) even exceeded the increase induced by warming (Figure 10f), resulting in a net decrease in pCO 2 and a strong CO 2 sink (Figures 10b and 11b). Similar patterns existed in the following seasons (Figures 10, 11c and 11d) As the CO 2 uptake mainly occurred within Pacific Summer Water and thermal and non-thermal effects had different impacts on CO 2 flux in different water masses, we estimated CO 2 fluxes separately in two types of water masses: the ACW (including RW) and non-ACW (including BSW and MW but excluding NVWW and RWW). We found that the CO 2 uptake potential in the nutrient-rich non-ACW was significantly larger than that in the  (Table 4). Such difference became less contrasting because these two water masses tended to have more similarities as they were modified by mixing, meltwater input, and other local processes in the northern Chukchi Sea (Table 4). This finding has important implications for CO 2 uptake in the Chukchi Sea. Woodgate (2018) found that annual inflow of Pacific waters has increased by more than 50% from 1990 to 2015, but there was no significant trend in annual mean flow in the ACC. Corlett and Pickart (2017) also confirmed that BSW (non-ACW) increased relative to ACW on the Chukchi slope from 2002 to 2004 to 2009-2014. Thus, non-ACW was likely responsible for the increase in total annual flow. If this is confirmed to be the case, we suggest that the Chukchi Sea will be a greater CO 2 sink in the future, as non-ACW will play a bigger role in CO 2 uptake.

Water Mass Evolution and Net Community Production
Spatial and temporal variations in water mass distributions substantially affect primary production in the Chukchi Sea during the growing season. In spring, both the southern and northern Chukchi Sea were largely occupied by the winter water (NVWW and RWW), which provided sufficient nutrients for phytoplankton to bloom (Figures 12a-12d; Arrigo et al., 2017). From late spring to early summer, winter waters were gradually transformed to or replaced by the warmer summer waters (BSW and ACW), particularly in the surface layer (Figure 12). At the same time, both normalized nitrate (nNO − 3 ), phosphate (nPO 3− 4 ), and DIC (nDIC) decreased rapidly as a result of strong consumption due to phytoplankton growth. In late summer and fall, only a small portion of RWW existed in the bottom layer in the northern Chukchi Sea, while BSW and ACW accounted for the majority of surface water over the entire Chukchi Sea (Figure 12). For the bottom layer, NO − 3 and PO 3− 4 were replenished by a combination of advection of nutrient-rich BSW and possible vertical mixing from the bottom layer. This input of NO − 3 and PO 3− 4 exceeded the biological drawdown in the southern Chukchi Sea, which resulted in a net increase in nNO − 3 and nPO 3− 4 concentrations in both the surface and bottom layers (Figures 12a and 12c). However, in the northern Chukchi Sea, nNO − 3 concentration remained depleted and nPO 3− 4 was the same in the surface layer but both increased in the bottom waters (Figures 12b and 12d). The high residual nNO − 3 and nPO 3− 4 in the bottom water were likely the remaining within winter waters rather than BSW delivery (Zheng et al., 2021).
As primary production is the main driver modifying water column biogeochemical properties during the growing season, we estimated NCP using the observed drawdown of nutrients (nNO − 3 and nPO 3− 4 ) and nDIC in the surface mixed layer. The spring concentrations of nNO − 3 , nPO 3− 4 and nDIC were used as an initial condition to determine the subsequent biological drawdown by NCP. As the largest changes were observed in the period between spring and early summer, which were also associated with the most evident water mass evolution, therefore, we will mainly focus on this period in the following discussion.
Although there were no observations in late spring in the southern Chukchi Sea (Figures 12a, 12c and 12e), a water mass analysis showing that the winter water still made up most of the southern Chukchi Sea during this period (Pacini et al., 2019). Based on that, we assume that water parcel in the southern Chukchi Sea between early spring and early summer was not greatly changed and it is still appropriate to estimate NCP based on the deficits of nutrients and nDIC. The decreases of nDIC in the surface mixed layer between spring and early summer were 79 ± 87 and 95 ± 51 μmol kg −1 , respectively, in the southern and northern Chukchi Sea (Figures 12e and 12f; Table 5). After correcting for air-sea gas exchange and CaCO 3 formation and dissolution, these nDIC deficits yield mean NCP estimates of 38.4 ± 26.2 and 61.4 ± 20.0 mmol C m −2 d −1 , respectively, in the southern and northern Chukchi Sea (Table 5). Our nDIC-based NCP estimates in 2014 were comparable with NCP estimates made using a similar approach in early summer in 2004 in the northern Chukchi Sea (Mathis et al., 2009). During the same period, nNO − 3 in the surface mixed layer was fully depleted over the entire Chukchi Sea (Figures 12a  and 12b). Such NO − 3 depletion was also observed in early summer in 2010 and 2011 Lowry et al., 2015). The mean nNO − 3 deficits in 2014 were 10.24 ± 0.20 and 9.92 ± 1.57 μM in the southern and northern Chukchi Sea, respectively (Table 5), giving an NCP of 3.51 ± 0.07 and 5.06 ± 0.80 mmol N m −2 d −1 in the southern and northern Chukchi Sea, respectively (Table 5). Because the mean nPO 3− 4 deficits were 1.19 ± 0.11  (Table 5), nPO 3− 4 -based NCP was estimated to be 0.41 ± 0.04 and 0.48 ± 0.11 mmol P m −2 d −1 in the southern and northern Chukchi Sea, respectively (Table 5). Given a M C:N:P uptake ratio of 106:16:1 (Redfield, 1958), nNO − 3 -based NCP was estimated to be 23.2 ± 0.5 and 33.4 ± 5.3 mmol C m −2 d −1 in the southern and northern Chukchi Sea. The analogous rates for nPO 3− 4 -based NCP were 43.3 ± 4.0 and 50.9 ± 11.3 mmol C m −2 d −1 , respectively. Thus, the nDIC-based NCP was significantly higher than the NCP derived from nNO − 3 consumption by 66%-84% and was close to or slightly higher than nPO 3− 4 -based NCP (Table 5). Although river runoff or precipitation may dominate endmembers in some locations and give different NCP estimates, it still cannot reconcile discrepancy in NCPs (Text S2 and Table  S2 in Supporting Information S1). Therefore, other processes must add nutrients or remove DIC, or a novel, unaccounted for mechanism is required to explain this disparity. N 2 fixation could be one possible process providing new nitrogen to the system and would reconcile the inconsistency between the NCP estimates (Park et al., 2008). Recent studies have shown that N 2 fixation can occur at low temperatures and high latitudes (Harding et al., 2018;Mills et al., 2018;Shiozaki et al., 2018;Sipler et al., 2017), which challenges the historical views that N 2 fixation is a warm-water process constrained to subtropical and tropical oligotrophic areas (von Friesen and Riemann, 2020;Zehr & Capone, 2020). Shiozaki et al. (2018) reported the N 2 fixation rates of 0.08-3.60 nmol N L −1 d −1 during late summer in 2015 in the Chukchi Sea, which is equivalent to 0.002-0.09 mmol N m −2 d −1 integrated over the top 25 m. Accounting for such a new source of nitrogen (using the upper limit), nNO − 3 -based NCP would slightly increase by only ∼2%, to 3.60 ± 0.07 and 5.15 ± 0.80 mmol N m −2 d −1 in the southern and northern Chukchi Sea, respectively (Table 5). Thus, N 2 fixation cannot explain the inconsistencies in NCP estimates based on the deficits of nDIC and nNO − 3 . Figure 10. T/S diagram for sea surface pCO 2 in the Chukchi Sea in different seasons. The gray dots represent all underway measurements and colored dots denote the gridded pCO 2 (0.25° latitude × 0.25° longitude) (a-d) pCO 2 difference between initial pCO 2 under the ice (i.e., 538 μatm) and observed pCO 2 (e-g) potential pCO 2 changes induced by thermal effects, and (i-l) pCO 2 changes induced by non-thermal effects. The approximate water mass boundaries are denoted by solid lines. NVWW is newly-ventilated winter water; RWW is remnant winter water; BSW is Bering summer water; ACW is Alaskan coastal water; MW is meltwater; RW is river water. The freezing line is indicated by the dashed line. Figure 11. Seasonal variation in air-sea CO 2 flux in the Chukchi Sea in 2014. CO 2 flux were derived from gridded pCO 2 data (0.25° × 0.25°). Negative values of CO 2 flux values indicate that CO 2 uptake from the atmosphere. Square, circle and triangle represent winter water (Newly-Ventilated Winter Water and remnant winter water), Alaska coastal water (ACW and river water), and non-Alaska coastal water (Bering summer water and meltwater), respectively.

Table 4
Seasonal Variations in CO 2 Flux (mmol m −2 d −1 ) in the Chukchi Sea    CaCO 3 formation or bio-calcification would also contribute to DIC removal in the water column. However, a net increase in nTA in both southern and northern Chukchi Sea (Table 5) suggests that CaCO 3 dissolution (e.g., mineral ikaite in ice) may dominate TA dynamics, which actually adds more DIC in the water column. By comparing the NCP results of with or without CaCO 3 dissolution correction, we further estimated that CaCO 3 dissolution accounted for ∼2.6% and 11.0% NCP-nDIC based in the southern and northern Chukchi Seas, respectively, during the period between spring and early summer. Thus, CaCO 3 dissolution cannot explain the inconsistencies in NCP estimates either.
An alternative explanation is that, during intensive primary production, phytoplankton carbon and nutrient uptake in the Chukchi Sea are non-Redfieldian; DIC uptake exceeds that expected for the observed N and P uptake. Based on deficits of nDIC and n PO 3− 4 , we found that C:P uptake ratio only slightly deviated from the Redfield ratio around 94.0-128.0 (Table 5). Since the remaining PO 3− 4 (∼0.5 μM) would not limit phytoplankton growth, the inconsistency in NCP must be attributable to a non-Redfield C:N uptake behavior. We found that phytoplankton growth in a N-limited surface water led to a C:N uptake ratio of about 10.9-12.1 (Table 5), which is close to the measurements of the particulate organic C:N ratio in the Chukchi Sea in the summer of 2002 (C:N ratio >9; Bates et al., 2005), but much higher than the Redfield ratio and the observed annual mean of particulate organic C:N ratio (∼6.4; Frigstad et al., 2014). We attribute this disagreement to a rapid seasonal change in the phytoplankton C:N uptake ratio or an ecological response. The high NCP during late spring throughout early summer makes the Chukchi Sea an extremely NO − 3 -limited ecosystem Codispoti et al., 2013;Mills et al., 2015;Zheng et al., 2021), resulting in a higher C:N uptake ratio and hence a higher particulate organic C:N ratio (Bates et al., 2005). In contrast, the C:N uptake ratio is relatively low (close to the Redfield ratio) in early spring (Bates et al., 2005) and fall (Frigstad et al., 2014) when NO − 3 is replenished.
Additional evidence for C:N variability comes from water column-integrated NCP estimates. As phytoplankton growth and nutrient consumption may erode the nitricline, NCP calculated from surface water depletion of NO − 3 and DIC probably underestimates the true NCP or overestimates the C:N uptake ratio, we computed depth-integrated NCP for the entire water column. Previously, phytoplankton growth was observed in bottom waters in the southern Chukchi Sea in summers of 2010 and 2011 . In 2014, we observed excessive consumption of NO − 3 in bottom waters not only in the southern Chukchi Sea, but also in the northern part (Figures 7h and 12b). The mean depth-integrated NCP in the southern and northern Chukchi Sea derived from nDIC changes were 65.4 ± 46.7 and 98.7 ± 55.8 mmol C m −2 d −1 , and NCP derived from n NO − 3 changes were 5.3 ± 1.5 and 8.4 ± 4.9 mmol N m −2 d −1 , respectively. These proportional increases in NCP estimates yield a C:N of 11.8-12.2, similar to that from the mixed layer analysis (Table S1 in Supporting Information S1), which confirms that C:N uptake ratio in the period between late spring and early summer was much higher than the Redfield ratio (C:N ∼ 6.6). Given that nNO − 3 was still depleted in late summer and fall in the northern Chukchi Sea (Figure 12b), we further suggest that the C:N uptake ratio would remain high in the northern Chukchi Sea, while the C:N uptake ratio may return to the canonical Redfield ratio in the southern Chukchi Sea in the later season due to NO − 3 replenishment exceeding consumption (Frigstad et al., 2014;Zheng et al., 2021).

Flexible Stoichiometric C:N Uptake Ratio Enhances Air-Sea CO 2 Uptake
The Chukchi Sea has been identified as a N-limited ecosystem during the growing season Codispoti et al., 2013;Mills et al., 2015;Zheng et al., 2021;Zhuang et al., 2020). Thus, DIC assimilation and associated NCP calculation is directly linked to the uptake of the most limited nutrient (NO − 3 ) and an assumption of a fixed C:N uptake ratio (Arrigo et al., 2017;Hansell et al., 1993). However, these assumptions are not always valid, especially when phytoplankton are experiencing N-limitation, because the internal C:N ratio of phytoplankton can deviate from the Redfield ratio (Finkel et al., 2010;Spilling et al., 2015) and the produced carbon-rich DOM can contribute a significant fraction of NCP (Baetge et al., 2020;Bif & Hansell, 2019). For example, when nutrients are sufficient, phytoplankton can assimilate and store extra nutrients for future use (i.e., luxury consumption; Elrifi & Turpin, 1985;Sommer, 1991). When nutrients in the internal pool or ambient water are not adequate, phytoplankton may release the fixed carbon as extracellular dissolved organic carbon (DOC) (Myklestad, 1995) or produce transparent exopolymeric particles (Mari et al., 2001;Vernet et al., 1998) and carbon-rich organic matter (Humphreys et al., 2019). From a biogeochemical perspective, this non-Redfield C:N uptake by phytoplankton may greatly impact seasonal variation in sea surface pCO 2 and air-sea CO 2 flux. For example, Fransner et al. (2018) found that CO 2 uptake from the atmosphere can be underestimated by 50% in OUYANG ET AL.

10.1029/2021JC018326
24 of 28 the northern Baltic Sea if a fixed Redfield ratio is used to determine carbon assimilation. However, there is little evidence for such a mechanism operating in the polar regions.
To demonstrate how a flexible C:N uptake stoichiometry by phytoplankton and possible CaCO 3 dissolution during active ice melting affect the air-sea CO 2 flux, we used a box model to reproduce sea surface pCO 2 in the Chukchi Sea from spring to early summer in 2014. Three simulations were performed; scenario 1 with a non-Redfield C:N uptake ratio corrected for CaCO 3 dissolution, scenario 2 with a non-Redfield C:N uptake ratio without correcting for CaCO 3 dissolution and scenario 3 with a fixed Redfield C:N ratio ( Figure 13). In the simulation with a non-Redfield C:N uptake stoichiometry (scenario 1 & 2), more efficient DIC drawdown led to a lower DIC concentration during early summer when compared to the Redfield ratio scenario, which results in better agreement with observations in both the southern and northern Chukchi Sea (Figures 13c and 13d). Furthermore, the non-Redfield stoichiometry approach better reproduces observations of pCO 2 during this intensive growing season (Figures 13e and 13f), whereas carbon fixation in the fixed-stoichiometry (Redfield ratio) simulation is not efficient enough to counteract the effects of warming and air-sea CO 2 exchange on sea surface pCO 2 . More importantly, the higher carbon fixation with a non-Redfield C:N uptake ratio, compared to the fixed Redfield uptake ratio, enhances air-sea CO 2 uptake in the Chukchi Sea. The net CO 2 uptake from the atmosphere in the non-Redfield stoichiometry scenario over the simulation period (day 136 to day 219 for the southern Chukchi Sea and day 167 to day 219 for the northern Chukchi Sea) is estimated to be 751 mmol C m −2 and 343 mmol C m −2 , respectively, in the southern and northern Chukchi Sea, which are 42% and 85% higher than CO 2 uptake in the fixed-stoichiometry scenario (528 mmol C m −2 and 185 mmol C m −2 ). Thus, we conclude that about 30%-46% of CO 2 uptake during this intensive growing season is supported by phytoplankton stoichiometric flexibility. By comparing the results of scenarios 1 and 2, we further quantified that the correction of CaCO 3 dissolution during ice melting from spring to early summer accounted for ∼1% and 9% CO 2 uptake in the southern and northern Chukchi Sea, respectively, which is a significant fraction of NCP estimate. The difference in the net uptake of atmospheric CO 2 between these simulations could increase as nutrient-limited conditions and high C:N uptake mechanisms extend to a longer growing season (e.g., late summer and fall).

Summary
Seasonality of biogeochemical properties determines how an ocean ecosystem responds to external and internal forcings. By synthesizing a rare data set of underway measurements and discrete samples collected in five consecutive cruises in 2014, we presented a complete seasonal cycle (covering spring through fall) of sea surface pCO 2 and biogeochemical properties in the Chukchi Sea.
We first explored the dominant drivers of seasonal pCO 2 change as water masses evolve, revealing that thermal and non-thermal effects have different impacts on sea surface pCO 2 and air-sea CO 2 fluxes in different water masses. The non-ACW with stronger biological CO 2 removal and a weaker warming effect has a stronger atmospheric CO 2 uptake potential in summer months than ACW does. We suggest that the Chukchi Sea will become a greater CO 2 sink in the future as the proportion of nutrient-rich non-ACW increases.
We then estimated NCP for the most intensive growing period (spring to early summer) using observed DIC and nutrient data, and found that carbon-based NCP was consistently higher than NO − 3 -based NCP by 66%-84%. We attributed this inconsistency in NCP estimates to a non-Redfield uptake of carbon and nutrients.
To investigate this hypothesis, we performed two model simulations to test how a flexible stoichiometry of C:N uptake ratio can affect seasonal biogeochemical dynamics and air-sea CO 2 exchange. Comparing modeled results and observations, we show that a variable phytoplankton C:N stoichiometry is needed in order to better simulate and understand the seasonal biogeochemical dynamics and air-sea CO 2 exchange. In particular, this stochiometric flexibility in phytoplankton enables more efficient DIC-fixation, which contributes about 30%-46% of CO 2 uptake from atmosphere in the Chukchi Sea. These model results also have important implications for biological pump estimates in the Chukchi ecosystem and parametrizing C and N cycles in regional biogeochemical models.