Effects of phytoplankton community composition and productivity on sea surface pCO2 variations in the Southern Ocean

Abstract The Southern Ocean is a vast net sink for atmospheric carbon dioxide (CO2), with marine phytoplankton playing a crucial role in CO2 fixation. We assessed how changes in the dominant phytoplankton community and net primary productivity (NPP) affected variations in the partial pressure of CO2 in surface water (pCO2sw) in the Indian sector of the Southern Ocean during austral summer. pCO2sw was negatively correlated with total phytoplankton and diatom abundances, as estimated from pigment signatures, in the zone south of the Antarctic Circumpolar Current; however, pCO2sw was not correlated with haptophyte abundance. Additionally, a stronger correlation was found between pCO2sw and total phytoplankton NPP than between chlorophyll a concentration and pCO2sw. We reconstructed pCO2sw at inter-annual scale using satellite data and assessed the inter-annual variability of air-sea CO2 flux. Over the period from 1997 to 2007, the integrated CO2 fluxes over the study region showed very large variations from a small source to a strong sink. Variations in the integrated CO2 fluxes were also correlated with changes in satellite-derived phytoplankton community in the Indian sector of the Southern Ocean and changes in the dominant phytoplankton community may control CO2 dynamics in the marginal ice zone.


Introduction
The Southern Ocean is one of the most important regions for the global carbon cycle, as it is a vast net sink for atmospheric carbon dioxide (CO 2 ) (e.g., Landschützer et al., 2015;Lenton et al., 2013;McNeil et al., 2007), and thus can absorb a large amount of anthropogenic CO 2 (Gruber et al., 2009;Khatiwala et al., 2009;Ito et al., 2010;Mikaloff Fletcher et al., 2006). During the austral summer in particular, phytoplankton productivity plays a crucial role in the carbon cycle of the Southern Ocean (e.g., Metzl et al., 2006;Takahashi et al., 2012). In the zone south of the Antarctic Circumpolar Current (SACCZ) in the Indian sector of the Southern Ocean, Takao et al. (2012) demonstrated that changes in net primary productivity (NPP) were strongly affected by the dominant members of the phytoplankton community using satellite remote sensing data. The transport efficiency of particulate organic carbon (POC) from the surface to deeper waters (i.e., the biological carbon pump) is strongly dependent on the composition of the phytoplankton community (e.g., Tr� eguer et al., 2018). Diatoms, which account for approximately 40% of annual NPP in the global ocean (Nelson et al., 1995), are one of the main contributors to the global carbon cycle. Large diatoms in particular appear to increase the efficiency of the biological carbon pump due to their heavy silica frustules (e.g., Tr� eguer and Pondaven, 2000). Coccolithophores, which belong to the division Haptophyta, also seem to play an important role in the regulation of atmospheric CO 2 . Precipitation of calcium carbonate (CaCO 3 ) by coccolithophores (Ca 2þ þ 2HCO 3 À ⇔ CaCO 3 þ CO 2 þ H 2 O) increases CO 2 concentrations in the surrounding seawater and is a potential source of CO 2 from the ocean to the atmosphere (Frankignoulle et al., 1994), whereas carbonate also increases sinking velocities and thus particle export to the deep ocean (e.g., Armstrong et al., 2002;Buitenhuis et al., 2001). On the other hand, the non-calcifying haptophyte Phaeocystis antarctica regularly forms huge colonies with high sinking rates in seasonal ice zones and coastal Antarctic waters, resulting in significant POC export from surface waters (DiTullio et al., 2000). The dominant phytoplankton community is affected by long-term climate change in the Southern Ocean. For example, in coastal waters along the western shelf of the Antarctic Peninsula (WAP), which has been the most rapidly warming region on earth over the past 50 years (e. g., Turner et al., 2005), greater dominance of cryptophytes over diatoms has occurred, which is hypothesized to be a response to regional warming (Moline et al., 2004). Montes-Hugo et al. (2009) found a significant change in chlorophyll a (chl a) concentrations over the past 30 years  along the WAP based on satellite and field data. This change could be accompanied by changes in the dominant phytoplankton community associated with rapid regional climate change along the WAP. More recently, Brown et al. (2019) found a nearly fivefold increase in summer oceanic CO 2 uptake over the past 25 years  along the WAP with greater upper ocean stability and changes in phytoplankton dynamics (i.e., biomass and community composition), which is affected by changes in wind, sea ice and meltwater dynamics, driven by regionally specific climatic variations, such as the Southern Annular Mode (SAM). The links between regionally specific climatic variations and satellite-derived dominant phytoplankton communities and NPP data were also examined. Alvain et al. (2013) observed large-scale shifts in the dominance of diatoms, with larger anomalies during extreme positive periods of the SAM in the Southern Ocean. In the Indian sector of the marginal ice zone (65-70 � S), Johnston and Gabric (2011) found a significant positive correlation between summer NPP and the SAM index over the previous decade (1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007).
Several studies have reported an influence of phytoplankton biomass, as measured through chl a concentration, on pCO 2 sw variations in the Southern Ocean (e.g., Carrillo et al., 2004;Jabaud-Jan et al., 2004;Metzl et al., 2006;Nomura et al., 2014;Stoll et al., 2002). In the Indian sector, Jabaud-Jan et al. (2004) reported a strong oceanic CO 2 sink associated with a significant decrease in surface silicic acid concentration, suggesting that a large diatom bloom had occurred. However, few studies have explored the relationship between phytoplankton community composition and pCO 2 sw in the Southern Ocean. Moreau et al. (2012) investigated the influence of phytoplankton community composition on variations in the difference between atmospheric and sea surface pCO 2 (ΔpCO 2 ) near the WAP. The authors found a significant negative correlation between chl a concentration and ΔpCO 2 at stations dominated by diatoms, but not at those dominated by phytoflagellates. The enhanced oceanic CO 2 uptake along the WAP over the past 25 years as mentioned above could be associated with shifts in phytoplankton community composition, as there has been a large significant increase in diatoms with higher ΔpCO 2 values relative to cryptophytes with lower ΔpCO 2 values at a similarly shallow mixed layer condition (Brown et al., 2019). In the Weddell Sea, Moreau et al. (2013) reported significant negative correlations between carbon biomass of both diatoms and phytoflagellates and ΔpCO 2 . These results suggest that phytoplankton community composition plays a crucial role in the relationship between chl a concentration or carbon biomass and CO 2 dynamics, although previous studies have been limited to waters along the WAP and in the Weddell Sea. Thus, assessing how changes in the abundance and productivity of the phytoplankton community affect changes in pCO 2 sw in other regions of the Southern Ocean is crucial. The objective of this study was to determine how changes in the phytoplankton community and NPP affect pCO 2 sw levels in the Indian sector of the SACCZ and how this impact inter-annual integrated air-sea CO 2 fluxes in this region.

Sampling procedures and hydrographic measurements
Samples were collected to measure NPP, phytoplankton community composition, macronutrients, and pCO 2 sw in the Indian Sector of the Southern Ocean during three cruises of the TR/V Umitaka-Maru (Tokyo University of Marine Science and Technology) during the austral summers of 2005/2006, 2007/2008, and 2008/2009. During these cruises, 11 stations in the SACCZ (Fig. 1) were found that exhibited a temperature minimum (T min ) layer, thought to be the remnant of the mixed layer from the previous late winter. Its properties were used as proxies for those of the winter mixed layer (e.g., Ishii et al., 1998Ishii et al., , 2002Jennings et al., 1984). In this study, same sampling procedures and sensors for hydrographic measurements were used throughout the three cruises. Vertical profiles of water temperature and salinity at each station were obtained with a CTD system (SBE 9plus, Sea-Bird Electronics, Inc.) that simultaneously measures seawater conductivity, temperature, and pressure (depth). Pre-cruise calibrations were performed for the conductivity, temperature, and pressure sensors at Sea-Bird Electronics, Inc. The accuracy of temperature and conductivity is �0.001 � C and �0.0003 S/m, respectively. Conductivity was converted to salinity using the SBE Data Processing software, and sensor salinity was corrected by the value of bottle salinity taken from several depths at some stations. To identify observational data in the SACCZ, the position of the Southern Boundary of the Antarctic Circumpolar Current during each cruise was located using vertical profiles of water temperature and salinity obtained with the CTD system, following the criteria of Orsi et al. (1995). The mixed layer depth (MLD) at each station was determined from vertical profiles of water temperature. In this study, the MLD was defined as the depth at which the change from a near-surface value at 10 m depth was 0.2 � C (de Boyer Mont� egut et al., 2004;Dong et al., 2008).
Samples for NPP and phytoplankton community composition in surface waters were collected from approximately 1-5 m depth using Teflon-coated Niskin bottles attached to a CTD system (ICTD, Falmouth Scientific Inc.). To obtain depth-integrated NPP, seawater samples were collected at several depths above the euphotic layer depth, which is defined as the depth at which photosynthetically available radiation (PAR) is 1% of its surface level. Samples for macronutrients, including nitrite plus nitrate (hereafter referred to as nitrate), phosphate, and silicic acid, were collected at several depths from the surface to bottom using the Niskin bottles attached to the SBE CTD system. Macronutrients were analyzed at an onshore laboratory (samples from 2005/2006) or onboard the ship (samples from 2007/2008 and 2008/2009) with a Bran þ Luebbe autoanalyzer (AACS-III) following the manufacturer's protocol. The analytical precision (coefficient of variation) was 1.1% at 20 mmol m À 3 nitrate, 2.1% at 2 mmol m À 3 phosphate, and 0.8% at 80 mmol m À 3 silicic acid.
The surface seawater temperature and salinity were measured continuously with a temperature sensor (DS103, Tateyama Kagaku High-Technologies Co.) and a conductivity sensor (Excell Thermosalinograph, Falmouth Scientific Inc.) along cruise tracks using seawater that drew from the bottom of the ship (approximately 5 m below the sea surface). During the three cruises, the accuracy of temperature and salinity is �0.02 � C and �0.03, respectively.
In vivo fluorescence at sea surface was measured continuously with a fluorescence sensor (WETStar, WETLabs Inc.). Sea surface chl a concentration was estimated from the fluorescence value based on the relationships among fluorescence, chl a concentration, and PAR measured in the same time. Total chl a concentration was determined with a Turner Designs fluorometer (Model 10-AU) using the non-acidification method of Welschmeyer (1994). Sea surface PAR was measured with an air quantum sensor (LI-190SB, LI. COR Inc.).

pCO 2 measurements
Atmospheric pCO 2 (pCO 2 air ) and pCO 2 sw were measured semicontinuously using a system similar to that described by Inoue et al. (1995Inoue et al. ( , 1999 for the 2005/2006 cruise and a system similar to that described by Murphy et al. (2001) for the 2007/2008 and 2008/2009 cruises. Uncontaminated seawater was collected from a sea chest onboard the ship and introduced into an equilibrator at approximately 10 L min À 1 . An aliquot of sampled air equilibrated with seawater was introduced into a non-dispersive infrared gas analyzer after removing water vapor. Ambient air was collected from an inlet on the bow side of the ship to measure pCO 2 air . The effect of increasing seawater temperature on the measured value of pCO 2 sw was corrected using the temperature difference between the sea surface and the equilibrator and the iso-chemical temperature dependence of pCO 2 sw reported by Copin--Mont� egut (1988,1989). The CO 2 measurement system was calibrated at 12-h intervals using four air-based standard gases with CO 2 concentrations of 200-400 ppm based on the World Meteorological Organization (WMO) mole fraction scale. Atmospheric pressure data recorded onboard the ship were used to calculate pCO 2 sw and pCO 2 air from the measured CO 2 concentrations. The overall uncertainty of pCO 2 sw measurements was estimated as approximately 3 μatm. The value of pCO 2 sw at each station was averaged over 10-min intervals, which included the time of water sampling.
In this study, we used the fugacity of CO 2 (fCO 2 ) in surface water obtained from the Surface Ocean CO 2 Atlas (SOCAT) version 2019 database (Bakker et al., 2016) as independent pCO 2 sw measurements.
Cruises of SOCAT data used in this study are listed in Table S1.

Net primary productivity and phytoplankton pigment analysis
Seawater samples (0.5-1 L) collected for NPP analysis were dispensed into acid-cleaned polycarbonate bottles and inoculated with a solution of NaH 13 CO 3 (99 atom% 13 C, Shoko Co., Ltd.) equivalent to approximately 10% of the dissolved inorganic carbon (DIC) in seawater. The bottles were incubated under natural light for 24 h in an on-deck incubator. The temperature in the incubator was maintained using surface seawater pumped from the bottom of the ship. Photosynthetic rates were calculated according to the method of Hama et al. (1983). A detailed description of the sample analysis procedure is provided in Hirawake et al. (2011).
Seawater samples (0.5-7 L) for phytoplankton pigment analyses were filtered onto 25-mm Whatman GF/F filters. The filters were blotted and stored in liquid nitrogen or a freezer at À 80 � C prior to analysis on land. The analytical procedure used for pigments samples is detailed in Hashihama et al. (2008). In this study, surface pigment samples taken from 1 to 5 m depth were interpreted using the matrix factorization program CHEMTAX (Mackey et al., 1996) to estimate the contributions of each algal class to total chl a, as determined using a high-performance liquid chromatography (HPLC) system. The CHEMTAX calculations were described in detail by Takao et al. (2012).

Seasonally integrated net community production
To assess the effects of biological activities on the CO 2 system in seawater from the beginning of spring until our observation period, seasonally integrated net community production (NCP), defined as NPP minus respiration by all heterotrophic organisms in the water column (Codispoti et al., 1986;Minas et al., 1986), was estimated from the deficit of macronutrients in the surface layer compared to the remnant of the winter mixed layer (e.g., Hoppema et al., 1999;Jennings et al., 1984;Shim et al., 2006). Sweeney et al. (2000) noted relatively constant carbon/nitrate (C/N) ratios among phytoplankton species. In the Indian sector of the Southern Ocean, Ishii et al. (1998Ishii et al. ( , 2002 reported that the C/N ratios of composite plots in the marginal ice zone were 6.2-6.3. Therefore, NCP in this study was estimated from the deficit of nitrate (ΔN) and the classical Redfield ratio (i.e., C/N ¼ 6.6; Redfield et al., 1963).

Hydrography and pCO 2 sw
Hydrographic conditions and CO 2 values at each station are listed in Table 1. Sea surface temperature (SST) ranged from À 0.9 � C at Stn. FG3 to 1.5 � C at Stn. II7, and sea surface salinity (SSS) ranged from 32.7 at Stn. II7 to 33.9 at Stn. FG3. pCO 2 sw at each station varied widely, from 205 μatm at Stn. II7 to 391 μatm at Stn. L33, and pCO 2 sw along the cruise tracks in the region south of 60 � S ranged from 191 to 424 μatm (Fig. 2a (Fig. 2a). Although this variation was related to changes in atmospheric pressure in our study region, the variation of pCO 2 air along the cruises tracks was much smaller than that of pCO 2 sw (Fig. 2a).   (1997/ 1998-2006/2007 during summer) using this study. The satellite image of sea ice concentration was derived by averaging 11 AMSR-E daily images from the water sampling date shown in Table 1. The black line represents the climatological location of the Southern Boundary of the Antarctic Circumpolar Current (SB) reported by Orsi et al. (1995). (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.) The values of SST, SSS, and chl a concentration along the cruise tracks varied widely in the region south of 60 � S (Fig. 2b-d). In areas west of 80 � E, relatively high (low) pCO 2 sw was observed with low (high) chl a concentrations and relatively constant SST, indicating that biological processes control the distribution of pCO 2 sw (e.g., Jabaud-Jan Stoll et al., 2002). High chl a concentrations were observed with relatively low SSS near the sea ice edge in the areas around 40 � E and 70 � E (Stns. I10 and II7). Values of pCO 2 sw , SST, and SSS in areas east of 140 � E changed gradually from north to south. The highest pCO 2 sw values were observed along the 140 � E transect, with relatively low chl a concentration and high SST, indicating that thermodynamic processes control the distribution of pCO 2 sw (e.g., Takahashi et al., 1997).
At 7 of 11 stations, potential temperature and salinity diagrams indicated clear T min layers (black dashed box in Fig. 3). At station FG3, near the coast of Antarctica where the MLD was deeper than 100 m, a T min layer was found near the bottom (i.e., 239 m). Compared to other stations, the sea ice minimum occurred more than 1 month later at station FG3 (Nomura et al., 2014). Thus, weaker thermal stratification at station FG3 may be caused by slow sea ice retreat. At three other stations (C10, I10, and II7), the water masses present included Antarctic surface water (ASW), modified circumpolar deep water (mCDW), and modified shelf water (mSW), which was formed from vertical mixing of shelf water with mCDW (e.g., Orsi and Wiederwohl, 2009;Williams et al., 2010). At these stations, T min layers were observed within mSW, and no clear T min layers were found with the water properties of the remnant mixed layer from the preceding late winter (winter water: WW), defined by potential temperature between À 1.9 � C and À 1.5 � C and salinity between 34.2 and 34.5 (Tomczak and Liefrink, 2005). In this study, the deficit ratio of silicic acid to nitrate (ΔSi/ΔN) in the water column from the preceding winter to the observation time and seasonally integrated NCP were estimated using values of macronutrients measured in the T min layers within the WW. The values of nitrate and silicic acid in T min layers within the WW were 30.3 � 1.2 mmol m À 3 and 62.2 � 7.0 mmol m À 3 (n ¼ 11), respectively. These values were comparable to climatological values of nitrate and silicic acid in August (30.7 � 1.5 mmol m À 3 and 60.5 � 2.6 mmol m À 3 ) reported previously for the same region (Garcia et al., 2014).

Phytoplankton community composition and pCO 2 sw
Based on pigment signatures, diatoms and haptophytes contributed strongly to the surface chl a biomass, making up more than 80% of the phytoplankton community at most stations (Fig. 4). The contributions of diatoms to chl a biomass varied from 19 to 92%, with a mean of 53%. Type 8 haptophytes (presumably Phaeocystis antarctica; Zapata et al., 2004) were the second greatest contributors to chl a biomass, ranging from 0 to 72% (mean 21%). For ΔSi/ΔN in the water column, Sweeney et al. (2000) noted that a value greater than 1.0 indicates the predominance of diatoms, whereas a value less than 0.5 indicates the predominance of P. antarctica. However, macronutrient uptake ratios of phytoplankton are thought to change not only among species but also with environmental and physiological conditions (e.g., Arrigo et al., 1999). At 10 of 11 stations, ΔSi/ΔN values in the water column were greater than 1.0 (Table 1). These results suggest that diatoms may dominate phytoplankton assemblages in the water column from austral spring to summer, as well as during our observation periods. Fig. 5 illustrates the relationships between pCO 2 sw and the surface chl a concentration for total phytoplankton, diatoms, and type 8 haptophytes. Surface chl a concentrations measured with HPLC ranged from 0.03 to 1.15 mg m À 3 (Fig. 5a). The chl a concentrations for diatoms and type 8 haptophytes ranged from 0.01 to 0.57 mg m À 3 (Fig. 5b) and 0-0.19 mg m À 3 (Fig. 5c), respectively. In the SACCZ during our cruises, pCO 2 sw was negatively correlated with chl a concentrations for total phytoplankton (r ¼ À 0.87, P < 0.001, n ¼ 11) and diatoms (r ¼ À 0.75, P < 0.01, n ¼ 11) but not correlated with that for type 8 haptophytes (Fig. 5). Along the cruise tracks in the region south of 60 � S, pCO 2 sw was also negatively correlated with the surface chl a concentration estimated from the fluorescence sensor, with a weaker linear relationship (i.e., gray dots in Fig. 5a). As noted above, Moreau et al. (2012) reported a significant negative correlation between chl a concentration and ΔpCO 2 at stations dominated by diatoms, but not at stations dominated by phytoflagellates during summer and fall in 3 consecutive years (i.e., 2002, 2003 and 2004) in coastal waters along the WAP. Using long-term records over the past 25 years (1993-2017), Brown et al. (2019) also reported a significant negative correlation between chl a concentration and ΔpCO 2 along the WAP. These findings are consistent with the results of the present study. We used pCO 2 sw rather than ΔpCO 2 , and no significant differences were found between the results for pCO 2 sw and those for ΔpCO 2 (data not shown). On the other hand, Moreau et al. (2013) reported a significant negative correlation between the carbon biomass of phytoflagellates estimated using microscopy and ΔpCO 2 in the Weddell Sea. We did not observe such a trend in this study, although we estimated the abundance of each phytoplankton group using pigment signatures rather than microscopy. These results suggest that the effects of phytoplankton community composition on CO 2 dynamics may differ spatiotemporally within the Southern Ocean. Thus, further studies examining the relationship between pCO 2 sw and phytoplankton community composition in the Southern Ocean are needed. Fig. 6 shows the relationships of pCO 2 sw with surface NPP and depthintegrated NPP within the euphotic layer (PP eu ). Surface NPP ranged from 4.4 to 82.8 mg C m À 3 day À 1 (Fig. 6a). Similar to the relationship with the surface chl a concentration, pCO 2 sw was negatively correlated with surface NPP (r ¼ À 0.90, P < 0.001, n ¼ 11). PP eu ranged from 158 Table 1 Summary of data from each station. SST, sea surface temperature ( � C); SSS, sea surface salinity; pCO 2 sw , sea surface pCO 2 (μatm); pCO 2 air , atmospheric pCO 2 (μatm); xCO 2 air , atmospheric CO 2 mole fraction (ppm); ΔSi/ΔN, deficit ratio of silicic acid to nitrate in the water column; PP eu , depth-integrated NPP (mg C m À 2 d À 1 ) within the euphotic layer; NCP, seasonally integrated net community production (g C m À 2 season À 1 ). to 1499 mg C m À 2 day À 1 (Table 1) and was negatively correlated with pCO 2 sw (r ¼ À 0.92, P < 0.001, n ¼ 11; Fig. 6b). Although NPP data for the Indian sector of the Southern Ocean are very limited, several studies have reported gross primary production (GPP) integrated throughout the water column in the area south of 60 � S during austral spring to summer. Depth-integrated GPP was 301-4570 mg C m À 2 day À 1 along 30-80 � E (Westwood et al., 2010), 125-890 mg C m À 2 day À 1 along 80-150 � E (Strutton et al., 2000), and 49-895 mg C m À 2 day À 1 at 140 � E (Yoshikawa et al., 2007). Goldman et al. (2015) reported a strong correlation between NPP and GPP values in the Southern Ocean based on 14 C incubations, with a slope of 0.61. Using this ratio with depth-integrated GPP values (i.e., 49-4570 mg C m À 2 day À 1 ) reported in previous studies, estimated PP eu varied from approx. 30 to 2790 mg C m À 2 day À 1 , and our PP eu values of 158-1499 mg C m À 2 day À 1 fell within this range.

Net primary productivity and pCO 2 sw
The relationship between pCO 2 sw and PP eu was consistent with that of surface NPP, and is expressed as follows: The relationships of pCO 2 sw with surface and depth-integrated NPP were slightly stronger than that between pCO 2 sw and surface chl a concentration (Tables 2 and 3). Values of pCO 2 sw during our observation periods were affected by pCO 2 sw changes in physical and biological processes, especially from the beginning of austral spring into summer. As discussed in several studies (e.g., Moreau et al., 2012Moreau et al., , 2013Schloss et al., 2007;Shadwick et al., 2015), air-sea exchange of CO 2 is a process that occurs over a longer time scale (e.g., the residence time of CO 2 in the mixed layer ranges from several weeks to months) than the changes in biological factors such as phytoplankton abundance, composition and productivity analyzed in this study (i.e., hours). Seasonally integrated NCP ranged from 2 to 63 g C m À 2 (Table 1); it was negatively correlated with pCO 2 sw (r ¼ À 0.64, P < 0.05, n ¼ 11; Fig. 6c) and positively correlated with PP eu (r ¼ 0.62, P < 0.05, n ¼ 11). These results suggest that CO 2 dynamics in the study region are driven mainly by biological activities, and the high phytoplankton abundance and productivity measured during our observation periods may have occurred continuously or intermittently from the beginning of austral spring to summer. Temporal changes in satellite chl a concentrations before the cruises would support this process. Relatively high satellite-derived chl a concentrations were observed before the cruises at stations with relatively high NCP (Table 1 and Fig. S1).

Estimation of pCO 2 sw distributions from satellite data
Previously, researchers estimated the distribution of pCO 2 sw using satellite images of chl a concentrations based on empirical relationships among in situ parameters, and reported that pCO 2 sw estimates incorporating chl a concentrations were more accurate in the North Pacific (e.g., Ono et al., 2004;Sarma et al., 2006) and the Southern Ocean (e.g., Xu et al., 2016). However, satellite-derived chl a concentrations are highly uncertain in the Southern Ocean (e.g., Dierssen and Smith, 2000;Hirawake et al., 2000;Johnson et al., 2013). Hirawake et al. (2011) proposed a new PP eu algorithm based on phytoplankton light absorption (an absorption-based algorithm) that does not use chl a concentration, and found that satellite PP eu values derived from the Sea-viewing Wide  Fig. 3. Potential temperature and salinity diagram for sampling stations in water depths shallower than 500 m. Solid curve represents the 28.27 kg m À 3 neutral density surface, horizontal line represents the freezing point of seawater, and black dashed box represents water properties of the remnant mixed layer from the preceding late winter (i.e., winter water).
Field-of-view Sensor (SeaWiFS) were in better agreement with in situ values than previous chl a-based algorithms in the Indian sector of the Southern Ocean (Hirawake et al., 2011;Takao et al., 2012). Our results suggest that the distribution of pCO 2 sw during austral summer in the study region can be estimated using chl a concentration and PP eu values (Figs. 5 and 6 and Tables 2 and 3). In this study, to determine the mean pCO 2 sw distribution in the Indian sector of the Southern Ocean, we employed satellite PP eu and chl a images with 9-km spatial resolution derived from the SeaWiFS, which was determined with the absorption-based algorithm by Hirawake et al. (2011) and the OCI algorithm by Hu et al. (2012), respectively. In this study, the reconstructions of pCO 2 distribution were restricted to the period 1997/1998-2006/2007 due to the use of satellite PP eu and chl a values derived from SeaWiFS. Fig. 7 shows the climatological satellite-derived PP eu distribution in our study region during the austral summer (December-February) from 1997/1998-2006/2007. The climatological satellite pCO 2 sw distribution during the austral summer was estimated using pCO 2 sw data corrected to the reference year 2000 ( Fig. 8a and S2), with an expected annual pCO 2 sw increase of about 2.0 μatm per year (Metzl, 2009), and the climatological map was compared with that of Takahashi et al. (2009, hereafter referred as to T09). Although the climatological map of T09 was constructed using pCO 2 sw values obtained from 1970 to 2007, data collected before 1996 in the Indian sector of the Southern Ocean is very limited (Takahashi et al., 1997(Takahashi et al., , 2002. The longitudinal distribution of climatological mean pCO 2 sw during the austral summer averaged over 55-65 � S Fig. 4. Contributions of each algal class to total chl a biomass in the surface waters of the SACCZ as estimated using CHEMTAX. Note: the common name "Green algae" includes Chlorophyceae, Prasinophyceae, and may include Euglenophyceae.  Table 2. agreed with that of T09, although the minimum pCO 2 sw value was observed in the area west of 90 � E in our results (Figs. 8 and 9, S2 and S3). These results suggest that the pCO 2 sw distribution could be estimated at high spatial resolution using the simplified relationship proposed here; however, the applicable period of ocean color data is limited to austral summer, when phytoplankton productivity is enhanced. For the period 1997/1998-2006/2007 during the austral summer (December-February), the root mean squared error (RMSE) between monthly mean pCO 2 sw estimated from satellite PP eu with 9-km spatial resolution and in situ pCO 2 sw along cruise tracks for SOCAT version 2019 (Table S1) was 29.3 μatm (n ¼ 5598; Fig. S4), which is relatively high to RMSE for a combined approach of a self-organizing map and feedforward neural network (ETH-SOMFFN) reported by Landschützer et al. (2015) for the entire Southern Ocean (i.e., the region south of 35 � S) and all seasons (apporox. 20 μatm for SOCAT and 25 μatm for LDEO).
This could be due to limitations of spaciotemporal scale for our pCO 2 sw estimates (i.e., the region south of 55 � S and austral summer only) and properties of our model based on biological process without consideration of other effects, such as changes in water temperature and salinity.

Interannual variations in pCO 2 sw and air-sea exchange of CO 2
Several pCO 2 mapping methods have been developed to estimate the spatiotemporal variations in pCO 2 sw and air-sea exchange of CO 2 using in situ pCO 2 databases, at both regional and global scales (e.g., Landschützer et al., 2015;R€ odenbeck et al., 2015). In this study, we propose a novel approach to estimate spatiotemporal variations in pCO 2 sw in the Southern Ocean using satellite PP eu images. Fig. 10 shows interannual  Table 3.

Table 2
Coefficients of regression for the relationships between pCO 2 sw and chl a concentrations derived from total phytoplankton, diatoms, or type 8 haptophytes. Bold numbers denote statistical significance at the 95% confidence level (P < 0.05). Bold numbers denote statistical significance at the 95% confidence level (P < 0.05).

Fig. 7.
Climatological satellite-derived PP eu distribution during the austral summer (December-February) from 1997/1998 to 2006/2007, which was reported by Takao et al. (2012). White areas represent no satellite data due to cloudiness or sea ice cover.
variations in the longitudinal distribution of reconstructed pCO 2 sw averaged over the region south of 55 � S during the austral summer (December-February). The longitudinal pCO 2 sw distribution based on satellite PP eu values was compared with that based on ETH-SOMFFN (Landschützer et al., 2017). Our longitudinal pCO 2 sw distribution was consistent with that from ETH-SOMFFN, despite being obtained using different mapping methods and datasets. This complementarity supports the robustness of our method for assessing air-sea exchange of CO 2 in the Southern Ocean. In some years, pCO 2 sw distributions for ETH-SOMFFN were higher than our results (i.e., the period [2006][2007]. For this period, SOCAT data had relatively low pCO 2 sw levels in the area from 80 � to 130 � E around 65 � S (Fig. 1), and our results better reproduced SOCAT data than ETH-SOMFFN (Fig. 10). For the period 2000-2001, in contrast, the ETH-SOMFFN model relatively well reproduced high pCO 2 sw levels in the area from 80 � to 100 � E, which was not reproduced in our result. In this region, lower PP eu values were observed in the period 2000-2001 (Fig. S5), and physical process such as upwelling/deep mixing, which enhances a vertical transport of DIC-rich subsurface waters from the deep ocean (e.g., Shim et al., 2006), would be more dominant than biological process. The pCO 2 sw for ETH-SOMFFN was developed using SOCAT version 4 (Landschützer et al., 2017), which did not include the relatively low pCO 2 sw data for the zone 30 � to 70 � E reported by Roden et al. (2016) as well as our data. Thus, relatively high pCO 2 sw levels for ETH-SOMFFN could be due to the pCO 2 sw dataset used for the model development.
Integrated air-sea CO 2 flux (FCO 2 ) over the study region was calculated using the following equation: where k is the CO 2 gas transfer coefficient (Edson et al., 2011); s is the solubility of CO 2 in seawater, which is a function of temperature and salinity (Weiss, 1974); A is the sea ice concentration; u is wind speed from NCEP (National Centers for Environmental Prediction) reanalysis monthly data; and Sc is the Schmidt number of CO 2 (Wanninkhof, 1992  , and accounted for the expected annual increase of 1.7 μatm per year (Metzl, 2009). Monthly mean pCO 2 sw distributions were estimated based on satellite PP eu and ETH-SOMFFN shown in Fig. 10. Estimated climatological FCO 2 integrated over the region south of 55 � S derived from satellite PP eu and chl a images during the austral summer was À 11.2 and À 11.8 Tg C per season, respectively, which is comparable to the climatological value (À 10.9 Tg C per season) reported previously for the same region . Interannual variations in integrated FCO 2 over the study region from 1997/1998-2006/2007 are shown in Fig. 11, and integrated FCO 2 ranged from À 23.5 Tg C per season in to 1.8 Tg C per season in 1997/1998. A strong sink of CO 2 in 1999 would be due to relatively high PP eu in the study region, mainly the eastern areas compared to other years ( Fig. 11 and S5). In this period, wind speeds were also relatively strong in the study region. This enhanced PP eu with increasing wind speeds could possibly be related to the regional-specific climatic variations such as SAM in the marginal ice zone (e.g., Johnston and Gabric, 2011;Lovenduski and Gruber, 2005). A significant positive correlation was found between the SAM index and the integrated PP eu (r ¼ 0.65, P < 0.05, n ¼ 10) and the averaged wind speed (r ¼ 0.87, P < 0.001, n ¼ 10) over the study region (Fig. S6). This enhanced PP eu with increasing wind speeds could be due to changes in sea ice and meltwater dynamics (i.e., changes in light and/or iron availability), driven by enhanced westerly winds in positive phases of the SAM (e.g., Doddridge and Marshall, 2017).
Although no significant trends were found in integrated FCO 2 from 1997/1998-2006/2007 due to a sharp decline in 1999/2000, a statistically significant reduction was found from 2000/2001-2006/2007, with an annual decrease rate of 2.6 Tg C per year (P < 0.05). A similar reduction was found using integrated FCO 2 based on ETH-SOMFFN (Landschützer et al., 2017). The increasing oceanic CO 2 uptake over 2000/2001-2006/2007 was consistent with other estimates in the Southern Ocean (e.g., Fay et al., 2018), although their study area was different from this study. These independent results support the conclusion that the Southern Ocean CO 2 sink was increasing in the 2000s .  To assess the impact of changes in phytoplankton community on this trend, we investigated the relationships between integrated FCO 2 and satellite-derived dominance ratios of diatoms and Phaeocystis-like (i.e., type 8 haptophytes) phytoplankton (Alvain et al., 2005(Alvain et al., , 2008Ben Mustapha et al., 2014) over the region south of 55 � S in the Indian sector of the Southern Ocean (Fig. 12). These ratios were calculated from the number of pixels where each phytoplankton community (e.g., diatoms) is dominant; the sum of the number of pixels for each dominant phytoplankton community was recorded, including nanoplankton-sized autotrophic flagellates (nanoflagellates), Prochlorococcus, Synechococcus, diatoms, and Phaeocystis-like phytoplankton (for more details, see Takao et al., 2012). Over the period from 1997 to 2007, integrated FCO 2 during austral summer based on satellite remote sensing and ETH-SOMFFN was negatively correlated with the dominance ratio of diatoms, but not correlated with the Phaeocystis-like group. The negative correlation between variations in FCO 2 and diatoms was consistent with the observations described in section 3.2 and in previous studies (Brown et al., 2019;Moreau et al., 2012Moreau et al., , 2013  Relationships between integrated FCO 2 and satellite-derived dominance ratio of (a) diatoms and (b) Phaeocystis-like (i.e., type 8 haptophytes) phytoplankton in the study region. The dashed line in each figure represents the regression line, which had statistical significance at the 95% confidence level.
Mixing process would enhance a vertical transport of CO 2 -rich waters from the deep ocean (e.g., Shim et al., 2006), and such physical processes may control CO 2 dynamics rather than biological processes in regions where Phaeocystis-like phytoplankton dominated. Our results suggest that a future shift in the dominant phytoplankton community from diatoms to another group due to climate change (e.g., Deppeler and Davidson, 2017) may have major implications for the carbon cycle of the Southern Ocean, causing changes in air-sea exchange of CO 2 in this region, as recently shown by the enhanced oceanic CO 2 uptake over the past 25 years with shifts in phytoplankton community composition (i.e., an increase in diatoms) along the WAP (i.e., Brown et al., 2019).

Conclusions
This study describes changes in the abundance of phytoplankton groups determined from pigment signatures and NPP and their effects on variations in pCO 2 sw in the Indian sector of the Southern Ocean. We found that pCO 2 sw in the SACCZ during our cruises was negatively correlated with chl a concentrations derived from total phytoplankton or diatoms, but not correlated with that of type 8 haptophytes (presumably P. antarctica). Furthermore, strong correlations were found between pCO 2 sw and NPP. These results suggest that changes in the dominant phytoplankton group and productivity may control CO 2 dynamics in the marginal ice zone of the Indian sector of the Southern Ocean, along with changes in physical processes such as thermodynamics, air-sea exchange of CO 2 , and horizontal and vertical advection of water masses. We also estimated the climatological mean pCO 2 sw during austral summer using satellite images and the empirical relationship between pCO 2 sw and PP eu (or chl a concentrations). Over the period from 1997 to 2007, the integrated FCO 2 over the study region showed very large variations from a small source in 1997/1998, a strong sink in 1999/2000 and 2006/2007, and a progressive increasing sink was found from 2000/2001-2006/ 2007. We found very coherent results with other methods, but the use of satellite PP eu to reconstruct pCO 2 sw and FCO 2 in the Southern Ocean is at present limited to the summer season. Previous studies suggest that the effects of phytoplankton community composition on CO 2 dynamics may differ spatiotemporally within the Southern Ocean (e.g., the WAP and the Weddell Sea). Thus, further studies examining the relationship between pCO 2 sw and phytoplankton community composition in the Southern Ocean are needed.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.