Decadal change of dissolved inorganic carbon in the subarctic western North Pacific Ocean

Dissolved inorganic carbon (DIC) was measured from 1992 to 2008 at two time-series sites in the subarctic western North Pacific; this region is a source of atmospheric CO2 in winter due to vertical mixing of deep waters rich in DIC. To estimate the decadal DIC increase resulting from CO2 uptake from the atmosphere, we corrected DIC for the contribution of biological activity below the temperature minimum (Tmin) layer (¡«100 m), which is the remnant of the mixed layer from the preceding winter. Decadal DIC increases in the Tmin layer and upper intermediate water (1.3¨C1.5 ¦Ìmol kg-1 yr-1; 100¨C200 m) were higher than those expected from oceanic equilibration with increasing atmospheric CO2 and those previously reported in the open North Pacific. The increase in water column CO2 was estimated to be 0.40 ¡À 0.08 mol m-2 yr-1. The decadal DIC change in the Tmin layer affects winter CO2 emission. The increase of atmospheric xCO2 in winter (2.1 ± 0.0 ppm yr-1) is higher than that of oceanic CO2 (0.7 ¡À 0.5 ppm yr-1) that calculated from DIC and total alkalinity in the Tmin layer. This difference suggests reduction of CO2 emission in winter is possibly controlled by the increase of total alkalinity.


Introduction
Anthropogenic CO 2 was taken up by the global ocean at a rate of 2.2 ± 0.4 PgC yr −1 during the 1990s (Denman et al., 2007). The most fundamental and reliable approach to the direct detection of the CO 2 uptake rate and its variation is to make more accurate and longer time-series observations at fixed stations, for example, station ALOHA in the subtropical North Pacific Ocean (e.g. Dore et al., 2003), BATS in the subtropical western North Atlantic Ocean (e.g. Bates et al., 2002), ESTOC in the subtropical eastern North Atlantic Ocean (e.g. Santana-Casiano et al., 2007), Ocean Station Papa (OSP) in the subarctic eastern North Pacific (e.g. Wong and Chan, 1991) and stations KNOT (44 • N, 155 • E) (e.g. Tsurushima et al., 2002) and K2 (47 • N, 160 • E) (e.g. Honda et al., 2006) in the subarctic western North Pacific Ocean (Fig. 1). Time-series observations are essential to understanding temporal variation of dissolved inorganic carbon (DIC) in the ocean.
Increases of DIC in the surface waters from 1997 to 2001 and in the subsurface waters from 1992 to 2001 were detected by time-series observations at KNOT (Tanaka et al., 2003;Wakita et al., 2005). DIC is positively correlated with AOU in the subsurface water due to the decomposition of organic matter. AOU was calculated by subtracting the observed concentration of dissolved oxygen (DO) from the saturated concentration calculated from temperature and salinity with the equation of Weiss (1970). A linear increase in AOU is superimposed on a bidecadal oscillation of AOU in the Oyashio region (near the subarctic western North Pacific) . Because the previously reported DIC increases in the subarctic western North Pacific are based on time-series data only from the 10-year period 1992 to 2001 (Wakita et al., 2005), more accurate and longer time-series data are required to confirm the increase of DIC.
At least annually since 1997, the Japan Agency for Marine-Earth Science and Technology (JAMSTEC) has conducted hydrographic observations at KNOT or K2 in the subarctic western North Pacific (Fig. 1). Hydrographic studies in this region were also conducted by other organizations between 1992 and 2003 (e.g. Tsurushima et al., 2002;Wakita et al., 2005). In this study, Fig. 1. Sampling stations and main ocean currents in the subarctic western North Pacific. The Kuroshio and Oyashio are the subtropical and subarctic western boundary currents in this region, respectively. This is modified from Yasuda et al. (1996). we compiled data obtained by JAMSTEC and others and investigated the temporal variation of DIC and related properties [AOU and total alkalinity (TA)] below the winter mixed layer of the subarctic western North Pacific for 1992-2008. We also discuss the decadal increase of DIC resulting from the uptake of CO 2 from the atmosphere through gas exchange at the air-sea interface.

Data sources and analyses
We prepared a set of data to investigate long-term climate change in the subarctic western North Pacific. All data were collected from the observations at the Japanese time-series stations KNOT and K2 from 1992 to 2008, which were made in 61 CTD casts on 44 cruises at KNOT and in 28 CTD casts on 15 cruises at K2; these cruises were made by seven research vessels: T/S Hokusei Maru (Hokkaido University), T/S Oshoro Maru (Hokkaido University), R/V Bosei Maru (Tokai University), R/V Hakuho Maru (Tokyo University), R/V Kaiyo Maru (Japan Fisheries Agency), R/V Hakurei No. 2 (Metal Mining Agency of Japan), R/V Natsushima (JAMSTEC) and R/V Mirai (JAMSTEC) (Tsurushima et al., 2002;Wakita et al., 2005;Watanabe et al., 2007). Approximately, 60% of all observations were done from the R/V Mirai. Data sets from K2 from 1999 to 2006  are published at the Carbon Dioxide Information and Analysis Center (http://cdiac.ornl.gov/) and the JAMSTEC data web site (http://www.godac.jamstec.go.jp/k2/index.html). DIC and TA were measured by using coulometric and potentiometric techniques, respectively. The precision of both DIC and TA was ±0.1% or better from duplicate determinations. TA from July 2000 to July 2001 during T/S Hokusei Maru cruise was determined by the improved single-point titration method, with precision of ±0.2% from duplicate determinations (Wakita et al., 2005). The DIC and TA values were determined with calibration against certified reference material provided by Prof. A. G. Dickson (Scripps Institution of Oceanography). The DO and nutrient concentrations from JAMSTEC observations were measured with an automatic titrator and a continuous flow analyser, respectively (e.g. Watanabe et al., 2007). We merged the KNOT data obtained by JAMSTEC cruises with those published by Wakita et al. (2005) and Tsurushima et al. (2002) for T/S Hokusei Maru, T/S Oshoro Maru, R/V Bosei Maru and R/V Hakuho Maru cruises. In addition, we added the oceanic physical and chemical data from KNOT and K2 collected as part of WOCE-P1 (1999) (http://whpo.ucsd. edu/data_access/show_cruise?ExpoCode=49KA199905_1) and WESTCOSMIC (1998WESTCOSMIC ( , 1999 (http://www.kanso.co.jp/ocean/ html-doc/english/top1.html).
The values of DIC, TA, DO and nutrients (silicate, phosphate and nitrate) had systematic errors among cruises, because the analytical methods used for these determinations and the precision and standards for analysis varied slightly from cruise to cruise. These systematic errors probably derive from the standardization because of the absence of a common reference material. DIC data for 1992-1997 during T/S Hokusei Maru cruise by Wakita et al. (2005) were not calibrated against certified reference material.
To investigate the temporal variation of chemical properties from the winter mixed layer to intermediate water, we corrected the systematic errors by assuming that ocean chemical properties in the North Pacific Deep Water (NPDW) were constant in our study area from 1992 to 2008 (Wakita et al., 2005;Watanabe et al., 2007). NPDW was defined as the water mass between 27.69σ θ (∼2000 db) and 27.77σ θ (∼3500 db), because chlorofluorocarbons were not detected below 27.69σ θ  and the bottom water deeper than the NPDW in the western North Pacific expanding from the Samoan Passage is characterized by potential temperature (θ ) lower than 1.2 • C (≈27.77σ θ ) (Johnson et al., 1994). Because the residence time of NPDW is ∼500 years (Stuiver et al., 1983), we can assume that the NPDW did not change significantly from 1992 to 2008 for the purpose of investigating the temporal variations in shallower waters.
The values of the systematic errors and factors were -6-19 μmol kg −1 for DIC, -15-8 μmol kg −1 for TA, 0.82-0.97 for DO, 0.95-1.09 for silicate, 0.98-1.12 for phosphate and 0.96-1.04 for nitrate. We corrected the values of these properties at the isopycnal surfaces of NPDW (σ θ = 27. 69, 27.70, 27.71, 27.72, 27.73, 27.74, 27.75, 27.76 and 27.77) to fit the mean observed values from 2006 to 2008 at the same isopycnal surfaces. The values at each of these isopycnal surfaces were obtained by linear interpolation. The minimum systematic errors were ±0.2% for DIC and TA, ±3% for DO and ±1% for nutrients. The standard deviations of DIC, TA, DO, silicate, phosphate and nitrate values in NPDW after the corrections were estimated to be ±1.9, ±1.3, ±1.5, ±1.0, ±0.01 and ±0.1 μmol kg −1 , respectively. These values agree well with the measurement precision for these properties and serve for use in evaluations of the temporal variation of chemical properties.
Because KNOT and K2 are both located in the western subarctic gyre (Fig. 1) and we lacked sufficient decadal time-series data for either station, we combined data from KNOT and K2. The typical water column structure in this region has a minimum temperature (T min ) layer at about 26.5σ θ (∼100 m), which is the remnant of the mixed layer from the preceding winter; the maximum temperature layer is at about 27.1σ θ (∼370 m) ( Fig. 2) (e.g. Osafune and Yasuda, 2006). Because KNOT is located just North of the Subarctic Front ( Fig. 1), data obtained there on some days (10 and 11 May 1999 (MR99-K02); 5 June 1999 (HO99-1); 22 July 1999 (HO99-3)) include no T min layer as a result of the migration of subtropical water (Tsurushima et al., 2002). Data from these days were excluded from the analysis of the T min layer in this study.
We here define the T min layer as the remnant of the winter mixed layer in early April, when the surface mixed layer was the coldest in the year (Fig. 3a). The depth, salinity and σ θ of the surface mixed layer reached annual maxima from middle March to early April (Figs 3b-d). Because the mean values of depth and σ θ in the winter mixed layer in early April (∼120 m, ∼26.5 kg m −3 ) were similar to those of the T min layer over all season from 1992 to 2008 (108 ± 18 m, 26.5 ± 0.1 kg m −3 ) (Figs 3b and c), in the T min layer there was no influence of the surface mixed layer from spring to fall. We can thus define the T min layer as the remnant of the winter mixed layer in early April.
The values of properties on each isopycnal surface (26.6σ θ -27.1σ θ ) were obtained by linear interpolation of discrete bottle sampling data. For the T min layer, after the values of T min , depth, salinity and σ θ were determined from continuous CTD data (1 db resolution), the contents of chemical parameters (DO, DIC, TA and nutrients) on the T min isopycnal surface were obtained by linear interpolation of discrete bottle sampling data. This was done because the values of depth, salinity and σ θ in the T min layer are not necessarily identical to those from discrete bottle sampling.

Seasonal variations of DIC and related properties in the surface water
In winter, the western subarctic gyre is a source of atmospheric CO 2 because of strong vertical mixing; from spring to fall it is a sink for CO 2 because of biological production (e.g. Tsurushima et al., 2002;Kawakami et al., 2007). At KNOT and K2, the maximum values of pCO 2 and net air-sea CO 2 flux occurred in early April (late winter); these maximum values were larger than the climatological mean values for this region reported by Takahashi et al., 2009 (Figs 3i and j). The values of pCO 2 in the ocean were calculated from DIC and TA by using the CO2SYS program (Pierrot et al., 2006), by using all data from 1992 to 2008 to examine the typical features of seasonal variation. In this calculation, we used the carbonate dissociation constants of Mehrbach et al. (1973) as refitted by Dickson and Millero (1987). The other properties also reached annual maxima in late winter ( Fig. 3): mixed layer depth (∼135 m), salinity (∼33.2), σ θ (∼26.5 kg m −3 ), DIC (∼2130 μmol kg −1 ), AOU (∼8 μmol kg −1 ), TA (∼2 240 μmol kg −1 ) and phosphate (∼1.9 μmol kg −1 ). The amplitude of seasonal variation of DIC in the surface mixed layer at KNOT (∼160 μmol kg −1 ) was larger than that at K2 (∼130 μmol kg −1 ) (Fig. 3e). These amplitudes in the western subarctic gyre are larger than those at other pelagic ocean time-series sites and are mainly attributed to biological production from spring to fall and strong vertical mixing in winter (e.g. Tsurushima et al., 2002). In summer, DIC and phosphate were lower at KNOT than at K2 (Figs 3e and h), which suggests that the difference of seasonal amplitude between KNOT and K2 is caused by differences in biological production (Kawakami et al., 2007). The values of DIC, AOU, TA, phosphate, pCO 2 and net air-sea CO 2 flux in winter (early April) at KNOT were similar to those in winter at K2 (Fig. 3). During winter, there were no differences in biological production and CO 2 emission between KNOT and K2. CO 2 emission during winter is controlled by strong vertical mixing of subsurface waters rich in DIC, TA and nutrients because of decomposition of organic matter and dissolution of CaCO 3 (e.g. Takahashi et al., 2006). These results indicate that the decadal DIC increase in the T min layer is affected by not only the increase of anthropogenic CO 2 , but also the temporal variation of CO 2 emission in winter due to strong vertical mixing.

Temporal variations of DIC and related properties in the subsurface water
We detected distinct trends in θ, salinity and isopycnal depths in the T min and 26.6-26.8σ θ layers (Fig. 4, Table 1). There was no apparent DIC increase observed in the subsurface water and AOU significantly decreased in the 26.6-26.8σ θ layer (P < 0.05) (Fig. 5, Table 1), even though DIC and AOU reportedly increased from 1992 to 2001 at KNOT (Wakita et al., 2005). These linear regressions model was fitted using the least squares approach. Long-term trends and bidecadal oscillations of values of θ, salinity, isopycnal depth, AOU and nutrients in the subsurface water in the northwestern subarctic Pacific Ocean are known Osafune and Yasuda, 2006;Watanabe et al., 2008). Recently, the bidecadal oscillations were linked to the nodal tidal cycle with an 18.6-year period (e.g. Osafune and Yasuda, 2006), in addition to the already known linkage to the bidecadal component of North Pacific Index (NPI) (Trenberth and Hurrell, 1994;Ono et al., 2001). The isopycnal depths, θ, salinity and AOU in the intermediate water are decreasing when the diurnal tide is strengthening, which occurred from the late 1990s to the late 2000s. It has also been reported that T min in the Oyashio upstream (near the subarctic western North Pacific) has a tendency to warm in the periods when the diurnal tide is both strong in the late 1990s and weak in the late 2000s (Osafune and Yasuda, 2006).
Decadal variations of DIC and AOU should be affected by this oscillation. However, we did not detect such DIC and AOU variations in our data collected during 1992-2008, because this period is shorter than a bidecadal cycle and because diurnal tide strengthening occurred from the late 1990s to the late 2000s (e.g. Osafune and Yasuda, 2006). A declining tendency of AOU was observed from the late 1990s (Fig. 5a), which is consistent with the AOU decrease observed from 1997 to 2006 in the North Pacific (Mecking et al., 2008). Because DIC is positively correlated with AOU in the subsurface water (r = 0.99) due to the decomposition of organic matter, decadal trends of DIC corrected for the contribution of organic matter decomposition should be detectable and larger than the decadal DIC trend we observed from 1992 to 2008.
TA significantly increased at rates of 0.5-1.6 μmol kg −1 yr −1 in the subsurface water (P < 0.05) (Fig. 5c, Table 1), despite the lack of a linear trend in TA at KNOT from 1997 to 2001 (Wakita et al., 2005). Also, the increase of potential alkalinity (PA), which is defined as the sum of TA and nitrate and is used to estimate the change in the dissolution of CaCO 3 , was in the range of 0.5-1.7 μmol kg −1 yr −1 (not shown). The saturation depths of aragonite and calcite were calculated at about 120 m (∼26.6σ θ ) and 170 m (∼26.8σ θ ) from TA and DIC data and the CO2SYS program (Pierrot et al., 2006), respectively. Those depths are consistent with previously published results . The observed TA and PA trends might have been caused by the increase of CaCO 3 dissolution resulting from the DIC increase, because anthropogenic CO 2 accumulates below the saturation depth of aragonite (∼120 m) in the western subarctic gyre; this saturation depth is shallower than that in the open North Pacific . This trend is comparable to the TA increases between GEOSECS (early 1970s) and WOCE (early 1990s) estimated by Sarma et al. (2002), which was 0.6 ± 0.4 μmol kg −1 yr −1 in the aragonite saturation horizons of the subarctic region in the North Pacific.

Estimate of DIC increase due to air-sea CO 2 gas exchange
The change in DIC in the subsurface water is controlled by the uptake of atmospheric CO 2 through gas exchange at the air-sea interface, the decomposition of organic matter and the dissolution of CaCO 3 (e.g. Gruber et al., 1996;Sabine et al., 2002). The decomposition of organic matter includes oxidation and denitrification of organic matter in the subsurface seawater (e.g. Sabine et al., 2002). Oxidation and denitrification of organic matter are related to AOU and N * [=(nitrate + nitrite) − 16 × (phosphate) + 2.90] as an index of nitrogen fixation-denitrification (Deutsch et al., 2001), respectively.    The change of DIC from the pre-industrial period caused by the oceanic uptake of CO 2 from the atmosphere through gas exchange ( C * ) is defined as follows: where DIC m , AOU m and TA m are the measured values. TA • is the preformed TA estimated using the equation of Sabine et al. (2002). The effect of decomposition from N * would be small, but not trivial: Watanabe et al. (2008) demonstrated a linear increase of N * on 26.8σ θ surface superimposed on a bidecadal oscillation. C eq280 is the theoretical DIC content of waters in equilibrium with the pre-industrial atmospheric CO 2 (280 μatm). Because C eq280 over time remains constant and its trend can be cancelled out, we calculated C * + C eq280 and its trend represents the DIC increase caused by the uptake of CO 2 from the atmosphere through the gas exchange at the air-sea interface.
C * + C eq280 has significantly increased at 0.7-1.5 (average 1.2 ± 0.3) μmol kg −1 yr −1 (P < 0.05) (Fig. 6, Table 1). These estimated increases in the winter mixed layer and upper intermediate water (T min and 26.6σ θ -26.8σ θ layers), which are between 100 and 200 m thick, are greater than those in deeper water (26.9σ θ -27.0σ θ ); C * + C eq280 has remained unchanged below 27.1σ θ . We estimated the increase of DIC accumulated in the water column to be 0.15 ± 0.04 mol m −2 yr −1 in the surface mixing layer and 0.25 ± 0.07 mol m −2 yr −1 in the intermediate water (100-400 m), using the C * + C eq280 increase and the thickness of the average depth on each isopycnal surface from T min to 27.1σ θ at increments of 0.1σ θ (  1973-1993(Ono et al., 2000] and in the global ocean [0.51 ± 0.09 mol m −2 yr −1 ; 1990s (Denman et al., 2007)]. This estimated inventory increase in the western subarctic gyre is higher than the anthropogenic CO 2 inventory increase in the Alaskan gyre (>45 • N along 152 • W) (<0.2 mol m −2 yr −1 ; maximum depth < ∼400 m; 1991/1992-2005/2006) estimated using an extended multiple linear regression technique . This different value is the reason why the decadal DIC increase caused by the uptake of CO 2 from the atmosphere in the western subarctic gyre is larger than that in the Alaskan gyre despite the same penetration depth in the two gyres.

Factors controlling the DIC increase due to air-sea CO 2 gas exchange
The C * increase reflects the increase in anthropogenic CO 2 and the temporal variation in the CO 2 difference between atmosphere and ocean in the surface mixed layer at the time of water mass formation (e.g. Gruber et al., 1996;Sabine et al., 2002). The C * + C eq280 increase in the T min layer in the subarctic western North Pacific would be affected by not only the increase of anthropogenic CO 2 but also the temporal variation of a strong CO 2 source during winter, because the T min layer is a remnant of the preceding winter mixed layer (e.g. Osafune and Yasuda, 2006). The western subarctic areas along the Kuril and western Aleutian arcs are strong CO 2 sources during winter owing to convective mixing of deep waters rich in respired CO 2 (e.g. Takahashi et al., 2006). This seasonal variation of pCO 2 has an opposite pattern to that of other open ocean time-series sites (ALOHA, BATS and OSP) (e.g. Tsurushima et al., 2002), because the primary causes of seasonal change in pCO 2 in the subarctic region (K2 and KNOT) and these other time-series sites are seasonal changes in DIC and temperature, respectively (e.g. Takahashi et al., 2006). The temporal variation of CO 2 emission in winter must affect that of DIC in the T min layer. If the winter increase rate of atmospheric CO 2 were higher than that of oceanic CO 2 from year-to-year, then the oceanic CO 2 uptake over time in this region would increase as a result of the reduction of CO 2 emission in winter due to decrease in the CO 2 difference between atmosphere and ocean.
The decadal increases of C * + C eq280 we observed in the T min layer (1.3 ± 0.3 μmol kg −1 yr −1 ) were higher than expected from oceanic equilibration with increased anthropogenic CO 2 in the atmosphere (0.7 μmol kg −1 yr −1 ), when calculated using the increase of atmospheric CO 2 (1.9 ppm yr −1 ) (Foster et al., 2007) and the Revelle factor in the T min layer (15.6 ± 0.4) at constant TA. This Revelle factor is consistent with previous results in winter (Takahashi et al., 2006). This C * + C eq280 increase was larger than the anthropogenic DIC increase since the 1990s at other open North Pacific stations [<1.1 μmol kg −1 yr −1 ; <30 • N along 149 • E (WHP P10) (Murata et al., 2009) ], and is comparable to higher anthropogenic CO 2 increases associated with the Kuroshio and California Currents on both edges of the basin (<1.5 μmol kg −1 yr −1 ; WHP P02) .
The CO 2 emission in winter is estimated from the difference between atmospheric and oceanic xCO 2 . We calculated oceanic xCO 2 by using the values of DIC and TA in the mixing layer (early April) (DIC win , TA win ). Phosphate and nitrate in the winter mixing layer also were calculated (PO 4win , NO 3win ). DIC, TA, phosphate and nitrate in the T min layer (DIC Tmin , TA Tmin , PO 4Tmin , NO 3Tmin ) are referred to as DIC win, TA win , PO 4win and NO 3win . Salinity normalized values of DIC, TA, phosphate and nitrate (nDIC win , nTA win , nPO 4win and nNO 3win ) were used to remove the influence of local evaporation and precipitation change, because salinity in the T min layer increased from 1997 to 2008 (Fig. 4c). A salinity of 33.1 was chosen as the constant to correct DIC, TA and nutrient data, as this represents the mean salinity observed in the T min layer over this period. For example, nDIC win was calculated by the following equation: where Sal win is salinity in the T min layer. The values of DIC, AOU, PO 4 and NO 3 in the T min layer (DIC Tmin ,AOU Tmin , PO 4Tmin and NO 3Tmin ) were annual minima in winter (Figs 3e, f and h), whereas TA Tmin had no a distinct seasonal variation (Fig. 3g). This increase of observed DIC Tmin , PO 4Tmin and NO 3Tmin are caused by the decomposition of organic matter after the previous winter (Figs 3e, f and h). We assumed that DO in the winter mixed layer is homogeneously saturated (i.e. AOU = 0) (e.g. Tsurushima et al., 2002), because the degree of saturation of DO in early April is ∼100% (98 ± 2%) because of the strong vertical mixing and air-sea exchange. DIC win , PO 4win and NO 3win were obtained from the observed DIC Tmin , PO 4Tmin and NO 3Tmin and decomposition with the following equations: where C/-O 2 , P/-O 2 and N/-O 2 are the stoichiometric ratios of carbon, phosphorus and nitrogen to oxygen during the decomposition of organic matter (Anderson and Sarmiento, 1994). We estimated the xCO 2 of surface seawater in winter from nDIC win and nTA win (Figs 7a and b). The contents of atmospheric xCO 2 in late winter (at the beginning of April) are cited by GLOBALVIEW-CO2 (2009) from 1997 The slope of the linear regression of oceanic xCO 2 on time was 0.7 ± 0.5 ppm yr −1 (n = 68), which was significant at the 90% confidence level (P ≤ 0.1) (Fig. 7c). This linear regression model was fitted using the weighted least squares approach, because calculated xCO 2 in the ocean was dispersing from yearto-year. This annual dispersion was not significantly uniform by Kruskal-Wallis test (P < 0.001), which indicates the necessity of weights. We used the weights as the inverse of variance of oceanic xCO 2 every year and obtained the linear regression of oceanic xCO 2 in winter. This oceanic CO 2 increase is consistent with the direct measurement values from 1995 to 2008 near KNOT and K2 of 1.2-1.5 ppm yr −1 for oceanic pCO 2, within standard errors (Dr. Nakaoka, NIES, personal communication).
The increase of atmospheric xCO 2 (2.1 ± 0.0 ppm yr −1 , n = 12) in winter is significantly higher than that of oceanic xCO 2 by the comparison of the two linear regressions (F < 0.001) (Fig. 7c). This significance was tested by using the F-distribution of the maximum likelihood estimate of variance, because the regression of oceanic xCO 2 was fitted using the weighted least squares approach. The difference of xCO 2 between atmosphere and ocean was calculated to be 1.4 ± 0.5 ppm yr −1 . These results suggest that the declining CO 2 emissions in winter were due to the declining xCO 2 gradient between atmosphere and ocean over time.
The temporal change in the CO 2 source in winter could be caused by increase of TA in the winter mixed layer. In the CO 2s system calculation from the DIC change at constant TA in the seawater, the oceanic xCO 2 increase in the seawater is proportional to the DIC increase. If TA in the seawater increases over time, the oceanic xCO 2 -increasing trend is smaller than that calculated by the increase of DIC under constant temperature and TA conditions. Applying this theoretical case to KNOT and K2, we estimated the oceanic xCO 2 increase to be 0.8 ppm yr −1 by using the increasing values computed from the regression lines of nDIC win (Fig. 7a, DIC = 1.2 × year − 368.4) and nTA win (Fig. 7b, TA = 1.1 × year − 46.6) at constant T min (1.63 • C, mean from 1997 to 2008) (Fig. 8). This estimated increase is similar to that by observed nDIC win and nTA win (0.7 ± 0.5 ppm yr −1 , Fig. 7c) and is considerably smaller than that using computed nDIC win increasing values and a constant nTA win value (2230 μmol kg −1 in 1997; 3.9 ppm yr −1 ) (Fig. 8). Thus, the increasing trend of nTA win could inhibit the decadal increase of oceanic xCO 2 against the nDIC win increase, which suggests a reduction of CO 2 emissions in winter. This indicates that TA increase favours the uptake of CO 2 in the ocean.
If increases of atmospheric and oceanic xCO 2 in winter continue in the future at the same rates as those from 1997 to 2008, the gradient between the atmosphere and ocean in winter might gradually decrease with time. Additional time-series data are required to confirm the factors controlling the temporal change of CO 2 emission in winter and to evaluate this speculation.
In addition to decadal increase of nDIC win and nTA win , the salinity, depth, nPO 4win and nNO 3win in the winter mixed layer (T min layer) also have significantly increased during the years 1997-2008 (P < 0.05) (Figs 4a and c and 7d and e), while σ θ remained constant (Fig. 7f). The increasing trend of nTA win , nPO 4win and nNO 3win may be caused by the enhanced winter convective mixing of deep waters rich in DIC, TA and nutrients. This increasing of nPO 4win (Fig. 7d) will follow the increasing trend of phosphate during the 1989-1993 period in the winter mixed layer of the subarctic western North Pacific . In contrast, multi-decadal decreasing trends of salinity, density and phosphate in the winter mixed layer from the 1970s to the 1990s in the Oyashio region and near the subarctic western North Pacific were thought to have been caused by the Temperature minimum is defined as the remnant of the winter mixed layer in early April. Salinity-normalized DIC (nDIC win ), TA (nTA win ), phosphate (nPO 4win ) and nitrate (nNO 3win ) data were used to remove the influence of local evaporation and precipitation. Values of oceanic xCO 2 in the winter mixed layer were calculated between nTA win and nDIC win . Regression lines for 1997 to 2008 are shown for nDIC win (dotted line, 1.4 ± 0.3 μmol kg −1 yr −1 , P < 0.0001); nTA win (dotted line, 1.1 ± 0.1 μmol kg −1 yr −1 , P < 0.05); atmospheric xCO 2 in winter (solid line; 2.1 ± 0.0 ppm yr −1 , P < 0.001); oceanic xCO 2 in winter (dashed line; 0.7 ± 0.5 ppm yr −1 , P ≤ 0.10); nPO 4win (dotted line, 0.012 ± 0.004 μmol kg −1 yr −1 , P < 0.05); and nNO 4win (dotted line, 0.20 ± 0.06 μmol kg −1 yr −1 , P < 0.05).
occurrence of surface stratification . Moreover, the temporal variability of depth and CO 2 emissions from the winter mixed layer must be relevant to atmospheric forcing (wind speed, etc.). The temporal variations of AOU on the isopycnal surface 26.7-27.2σ θ in the Oyashio region and wintertime wind stress curl anomaly around this region are negatively and positively correlated with the bidecadal component of NPI, respectively Ishi and Hanawa, 2005). Because DIC is positively correlated with AOU in the subsurface water (r = 0.99) due to the decomposition of organic matter, temporal variation of DIC also possibly has a bidecadal oscillation. Further investigation of the atmospheric effect is required to detect whether there is enhanced winter convective mixing of deep waters or not. The relationships among temporal variability of winter CO 2 emissions and atmospheric forcing or climate index must be investigated in the near future. In the future, more accurate and longer time-series data will be required to evaluate these speculations.

Summary
DIC and related chemical species have been measured from 1992 to 2008 at Stations KNOT (44 • N, 155 • E) and K2 (47 • N, 160 • E) in the subarctic western North Pacific in the western subarctic gyre. The surface mixed layer there in winter is a source of atmospheric CO 2 owing to strong vertical mixing (∼100 m) and from spring to fall it is a sink for CO 2 because of biological production (e.g. Tsurushima et al., 2002) (Fig. 3i). This seasonal variation of pCO 2 has an opposite pattern to that of other timeseries sites (ALOHA, BATS and OSP) (e.g. Tsurushima et al., 2002), because the primary cause for seasonal change in the subarctic region is seasonal DIC change (e.g. Takahashi et al.,  2006). The temporal change of CO 2 emission in winter must affect that of DIC in the T min layer. Decadal DIC increase below the T min layer is affected not only by the increase of anthropogenic CO 2 but also by temporal change of a strong CO 2 source during winter, because the typical water column structure here has the T min layer at ∼26.5σ θ (∼100 m) (Figs 2 and 3d), which is the remnant of the mixed layer from the preceding winter.
To estimate the decadal increase of DIC resulting from the uptake of CO 2 from gas exchange with the atmosphere, we corrected DIC for the contribution of the biological activity by calculating C * (e.g. Gruber et al., 1996;Sabine et al., 2002). Decadal changes of corrected DIC have significantly increased at 0.7-1.5 (avg. 1.2 ± 0.2) μmol kg −1 yr −1 in the subsurface water from 1997 to 2008 (Fig. 6, Table 1). These estimated increases in the winter mixed layer and upper intermediate water (T min layer and 26.6-26.8σ θ ; 100-200 m) are higher than those in the next deeper water layer (26.9-27.0σ θ ; 200-300 m) and have remained unchanged below 27.1σ θ (∼400 m). We estimated the water column inventory of CO 2 increase to be 0.40 ± 0.08 mol m −2 yr −1 , which is similar to those previously reported for the same region (Ono et al., 2000) and for the global ocean (Bindoff et al., 2007).
The decadal DIC increase by the uptake of CO 2 from the atmosphere into the winter mixed layer (T min layer) would be affected not only by the increase of anthropogenic CO 2 but also the reduction of CO 2 emission in winter. The estimated DIC increase in the T min layer and in the 26.6-26.8σ θ layer (1.3-1.5 μmol kg −1 yr −1 ) was higher than that expected from oceanic equilibration with increased anthropogenic CO 2 in the atmosphere (0.7 μmol kg −1 yr −1 ) and anthropogenic DIC increase since the 1990s in the other open North Pacific stations (<1.1 μmol kg −1 yr −1 ) Murata et al., 2009). The increase of atmospheric xCO 2 (2.1 ± 0.0 ppm yr −1 ) in winter (early April) is markedly higher than that of oceanic xCO 2 in winter calculated from observed DIC Tmin and TA Tmin (0.7 ± 0.5 ppm yr −1 ) (Fig. 7c), which suggests that the reduction of CO 2 emission in winter is due to a decreasing xCO 2 gradient between the atmosphere and ocean.
This temporal change of the CO 2 source in winter is possibly caused by an increase of nTA win in the winter mixed layer. The oceanic xCO 2 increase calculated from observed DIC Tmin and TA Tmin was considerably smaller than that calculated using computed nDIC min increasing values and constant T min (1.63 • C, mean from 1997 to 2008) and nTA win (2230 μmol kg −1 in 1997; 3.9 ppm yr −1 ) (Fig. 8). Thus, the increasing trend of nTA win would inhibit decadal increase of oceanic xCO 2 against nDIC win increasing, which suggests the reduction of CO 2 emission in winter. This indicates that TA increase favours the uptake of CO 2 in the ocean. Additional time-series data are required to confirm the factors controlling the temporal change of CO 2 emission in winter. Moreover, because CO 2 efflux from the winter mixed layer must be relevant to atmospheric forcing, the relationships between temporal variability of winter CO 2 emissions and atmospheric forcing must be investigated in the near future. Watanabe, H. Y. Inoue and S. Tsunogai of Hokkaido University for their valuable comments and discussion; Dr. H. Narita of Tokai University and Dr. K. Imai of Hokkaido University for their great support onboard; and the staff of Marine Works Japan Co., Ltd. who worked as marine technicians onboard. Finally, we also express our deep thanks to three anonymous reviewers who gave us many useful comments.