Millennial-scale variability in the behavior of North Pacific Intermediate Water during the last glacial and deglacial period, and its association with Dansgaard-Oeschger (D-O) cycles and Heinrich events, are examined based on benthic foraminiferal oxygen and carbon isotopes (δ18Obf and δ13Cbf) and %CaCO3 using a sediment core recovered from the northeastern slope of the Bering Sea. A suite of positive δ18Obf excursions at intermediate depths of the Bering Sea, which seem at least in part associated with increases in the δ18Obf gradients between the Bering and Okhotsk Seas, suggest the Bering Sea as a proximate source of intermediate water during several severe stadial episodes in the last glacial and deglacial period. Absence of such δ18Obf gradients during periods of high surface productivity in the Bering and Okhotsk Seas, which we correlate to D-O interstadials, suggests a reduction in intermediate water production in the Bering Sea and subsequent introduction of nutrient-rich deep waters from the North Pacific into intermediate depths of the Bering Sea. We argue that a reorganization of atmospheric circulation in the high-latitude North Pacific during severe cold episodes in the last glacial and deglacial period created favorable conditions for brine rejection in the northeastern Bering Sea. The resulting salinity increase in the cold surface waters could have initiated intermediate (and deep) water formation that spread out to the North Pacific.
 While variability of deep and intermediate water circulation in the North Atlantic is well documented for the last glacial and deglacial period [e.g., Sarnthein et al., 2000; Rahmstorf, 2002], it is still poorly understood in the North Pacific during that time due to the shallow lysocline that limits paleoceanographic investigations to bathymetric highs and the continental margins with water depths shallower than ca. 2000 m [e.g., Farrell and Prell, 1989]. In the modern North Pacific, deep water formation is absent because of low surface salinity that results from excess precipitation over evaporation in the area [e.g., Warren, 1983]. However, presence of intermediate water in the North Pacific is evident from a well-defined salinity minimum in the subtropical North Pacific at depths of approximately 300 to 800 m [e.g., Sverdrup et al., 1942; Reid, 1965; Talley, 1991, 1993; Yasuda, 1997], which is called North Pacific Intermediate Water (NPIW). One major modern source area of NPIW is the northern Okhotsk Sea [e.g., Yasuda, 1997; Shcherbina et al., 2003], where the formation of Okhotsk Sea Intermediate Water, a precursor of NPIW, is tightly coupled to northwesterly winds that dominate during winter and create polynyas (ice-free zones) in seasonal sea-ice [e.g., Shcherbina et al., 2003; Katsuki et al., 2010].
 Knowledge about NPIW formation and its extent is crucial to assess its CO2 storage capability and the meridional heat transport, which are important components of the global climate system [e.g., Levitus et al., 2000; Otsuki et al., 2003]. During the last glacial period, intermediate water ventilation in the North Pacific apparently reached as deep as ∼3000 m based on epifaunal benthic δ13C [Duplessy et al., 1988; Matsumoto et al., 2002] and NPIW is interpreted to have extended far south to the California margin during D-O stadial periods (DOS) of the last glacial and deglacial period, while it shrank or was absent during D-O interstadial periods (DOIS) [e.g., Keigwin and Jones, 1990; Behl and Kennett, 1996; Tada et al., 2000; Hendy and Pedersen, 2005; Okazaki et al., 2010]. However, because these sites are far from the source regions of NPIW, the observed changes in oxygenation level at intermediate depths [e.g., Behl and Kennett, 1996] may not necessarily represent variation in NPIW intensity [Hendy et al., 2004; Ortiz et al., 2004; McKay et al., 2005; Dean, 2007]. Kim and Park  advocated a source of NPIW in the northern marginal seas of the North Pacific during the last glacial period based on their coupled ocean-atmosphere model. Ohkushi et al.  hypothesized glacial NPIW production in the Bering Sea based on distribution patterns of radiolarian assemblages in the Bering and Okhotsk Seas. However, no quantitative data with sufficiently high time resolution has been presented so far on the possibility of intermediate water formation and the mode of its formation in the Bering Sea during the last glacial and deglacial period.
 In this study, we examine millennial-scale variability of NPIW and its association with abrupt climatic changes such as D-O cycles during the last glacial period, and Bølling/Allerød (B/A) and Younger Dryas (YD) during the deglacial period. We explore the possibility of the Bering Sea as a locus of NPIW formation by examining oxygen and carbon isotopes of benthic foraminifera (hereafter denoted δ18Obf and δ13Cbf) to infer changes in intermediate water temperature/salinity and oxygenation/ventilation, respectively, using a piston core obtained from the northeastern slope of the Bering Sea. Because δ18Obf are expected to be higher in the source area of NPIW, we also attempt to estimate the direction of lateral advection and the proximal source of NPIW by comparing δ18Obf at intermediate depths between the Bering and Okhotsk Seas. In addition, we use %CaCO3 to examine the relationship of surface productivity with intermediate water properties in association with abrupt climatic changes.
2. Oceanographic Setting
 The Bering Sea, surrounded by Eastern Siberia and Western Alaska to its north and the Aleutian and the Kommander Islands to its south, is the third largest marginal sea in the world with an area of ca. 2.3 × 106 km2 [Hood, 1983]. A more than 400 km wide and <200 m deep continental shelf (Bering Shelf) occupies the northeastern half of the Bering Sea, while a vast deep basin of ∼3900 m depth, called the Aleutian Basin, dominates the southwestern half. A NW-SE running continental slope separates the Bering Shelf and the Aleutian Basin. Two large topographic highs (<2000 m), the Shirshov and Bowers Ridges, extend ca. 500 km into the basin from the north and south, respectively (Figure 1).
 Surface waters in the Bering Sea largely originate from the westward flowing Alaskan Current entering through passes between the Aleutian Islands with the largest inflow occurring through the Near Strait [Favorite et al., 1976; Reed and Stabeno, 1999] (Figure 1). A cyclonic gyre dominates the surface current regime over the deep basin. The northeastern component of the gyre (along the shelf break), the so-called Bering Slope Current (BSC), feeds in part the southwestward directed Kamchatka Current, the most significant outflow to the North Pacific, and the northeastward directed Anadyr Current that contributes 1/3 of outflowing waters to the Arctic Ocean. The remaining 2/3 of outflowing waters originate from the dominantly northward flowing currents over the Bering Shelf [e.g., Springer et al., 1996]. Deep water exchange between the Bering Sea and the North Pacific occurs through the Kamchatka Strait (∼4000 m sill depth) and the Near Strait (∼2000 m sill depth). The present deep waters in the Aleutian Basin are most likely derived from the North Pacific Ocean [e.g., Roden, 1995; Emile-Geay et al., 2003], although weak recent ventilation in the eastern part of the deep Aleutian Basin has been suggested [Warner and Roden, 1995].
 Due to delivery of nutrients to the surface waters through upwelling of deep waters and melting of winter ice, the Bering Sea is also characterized by high primary productivity, in particular along the northeastern continental slope known as the Bering Sea Green Belt [Springer et al., 1996]. In the Aleutian Basin, summer insolation warms the upper 30 m of the water column, which is rich in oxygen and low in salinity and nutrients [Roden, 1995]. Below the surface layer, a 100 to 150 m thick temperature minimum layer is present, which is likely a remnant of winter convection [Roden, 1995]. The halo-, nutri- and oxyclines are found between 150 and 450 m depth. While salinity and density increase uniformly with depth, oxygen minimum and associated nutrient maximum layers are centered at approximately 900 m depth. Vertical profiles of temperature and salinity in the upper 500 m of the water column become almost constant during winter due to severe wind stress that mixes water to depths [Ohtani et al., 1972; Takahashi, 2005].
3. Sea Ice Conditions in the Bering Sea and Their Relation to Atmospheric Circulation
 At present, the winter sea ice edge in the Bering Sea is situated approximately at the shelf break. Polynyas are common features within the area covered by sea ice and can produce open water for days to weeks [Roden, 1995]. Polynya formation in the Bering Sea is related to the direction of storm tracks. For example, southerly winds can drive sea ice northward and close the polynya growth regions [Rodionov et al., 2007]. During anomalously cold winters, on the other hand, the Aleutian Low is displaced eastward of its normal position leading to more northerly winds and sea ice extending farther south [Rogers, 1981; Luchin et al., 2002], which is favorable for enhanced polynya formation. Based on the NCEP/NCAR reanalysis data for the period from 1950 to 2003, Rodionov et al.  showed that the coldest years in the Bering Sea are associated with anticyclonic circulation driven by intense high pressure cells dominating over northeastern Siberia and cyclone tracks typically running across the North Pacific just to the south of the Aleutian Islands. Vigorous air-sea exchange associated with cold air outbreaks and ice formation in areas of polynyas can locally increase surface water salinity up to 35‰ [Schumacher et al., 1983]. Consequently, dense waters may form, contributing to the cold halocline layer in the Aleutian Basin [Cavalieri and Martin, 1994].
4. Onboard Sample Preparation
 Piston core PC-23A was retrieved in summer 2006 from the upper slope of the northeastern part of the Bering Sea (60°09.52′N, 179°27.82′W) at a water depth of 1002 m during cruise MR06–04 Leg 2 by R/V Mirai (MR06–04 cruise report). The recovered core was immediately cut into 1 m long sections, which were left for 24 h, cut horizontally into halves and submitted to routine analyses. Sections were stored in a refrigerator until onboard sampling. Five to ten conventional smear slide analyses were conducted per meter of core during visual core description. Samples for X-ray guide tube micro-scanner (XGT) and soft X-ray radiography analyses were taken as thin slabs in plastic cases of 20 (length) × 2.5 (width) × 0.7 (depth) cm in size. Slab samples were taken continuously throughout the core and were immediately vacuum sealed and double packed to prevent air exchange. Samples used for isotope and 14C analyses of foraminifera and XRF analysis were sub-sampled at 2 cm intervals and double packed in bags. All samples were stored in a refrigerator onboard and then at the University of Tokyo at temperatures of 5°C for 5 to 11 months before analyses.
5. Analytical Methods
5.1. Soft X-ray Analysis
 Soft X-ray photographs were taken for all slab samples onboard using a SOFTEX PRO-TEST 150 to examine sedimentary structures and microstructures. Voltages ranging from 40–50 kVp, currents from 2 to 3 mA and irradiation time from 150 to 200 s were used to optimize photographic conditions.
5.2. XGT Analysis
 To characterize millennial-scale changes of sediment composition that may reflect changes in surface and/or intermediate water properties, a high resolution analysis of major elements was conducted using an X-ray micro-scanner (Horiba XGT-2700, hereafter designated as XGT), which is capable of continuous high resolution quantitative analysis of major elements [Koshikawa et al., 2003; Kido et al., 2006]. XGT is equipped with a Rh target and an X-ray guide tube of 100 μm in diameter. A tube voltage of 50 kV, tube current of 1 mA, scanning time of 100 s and scanning area of 1.25 × 1.25 cm were adopted following Koshikawa et al. . We followed the procedures of Kido et al.  that make corrections for absorption effects of Mylar film, thin water film and interstitial water to obtain X-ray fluorescence (XRF) intensities for Al, Si, K, Ca, Ti and Fe that are proportional to their contents and can be used for quantitative estimation. In this study, we use %Al2O3 and %CaO determined by XGT, which show a good correlation to %Al2O3 (r2 = 0.62) and %CaO (r2 = 0.8) determined by conventional quantitative XRF.
5.3. Conventional X-Ray Fluorescence Analysis
 Conventional XRF analysis was conducted for 102 selected samples to determine major element concentrations of ten elements including Al2O3, SiO2, K2O, CaO, TiO2 and Fe2O3, which are used to calibrate XRF peak intensities determined by XGT to major element contents. The preparation of glass beads is based on the method described in Yoshida and Takahashi . Measurements were carried out on the glass beads using an XRF spectrometer Phillips Model PW 1480 equipped with a Rh tube at an acceleration voltage of 50 kV and a current of 50 mA at the Department of Earth and Planetary Science, University of Tokyo. The analytical precisions (±sigma %: relative error) were within ±1.3%, ±0.7%, ±3.9%, ±2.9%, ±1.2%, and ±2.0% for Al2O3, SiO2, K2O, CaO, TiO2 and Fe2O3, respectively [Yoshida and Takahashi, 1997]. %CaCO3 was estimated based on the CaCO3/CaO atomic weight ratio of 100/56. The percentage of biogenic CaCO3 was estimated from total CaCO3 using the equation
%CaCO3 total measured by conventional XRF and biogenic %CaCO3 measured at a similar resolution using a coloumbmeter by Kim et al.  show a good correlation (r2 = 0.86) suggesting that detrital CaCO3 was relatively constant over time at ∼2.8% judging from the intercept of the regression line with the %CaCO3 total axis. By applying the average content of Al2O3 (∼14.4%) in PC-23A, detrital CaCO3/Al2O3 is ∼0.19. We accordingly estimate the contribution of biogenic CaCO3 in total CaCO3 measured by XGT using the equation
Hereafter, the term %CaCO3 refers to %CaCO3 biogenic (XGT).
5.4. Oxygen and Carbon Isotope Analyses of Benthic Foraminifera
 We conducted δ18Obf and δ13Cbf analyses of benthic foraminifera Uvigerina akitaensis and Buliminella tenuata from PC-23A in order to reconstruct temporal changes in intermediate water mass properties such as temperature, salinity, and the degree of organic matter decomposition within the intermediate water and surface sediment. Samples were selected at 4 to 14 cm intervals throughout the core (using ∼15 g of sediments per sample). Between 1 and 15 specimens of benthic foraminifera U. akitaensis and B. tenuata (if available a minimum of 7 specimens for U. akitaensis and 10 specimens for B. tenuata) were picked from the >150 μm fraction. B. tenuata was used only when U. akitaensis was absent (Table S1 in the auxiliary material), which was the case exclusively in laminated intervals.
 Benthic foraminifera in PC-23A are well preserved based on observation under the microscope. An exception occurs between 1380 and 1400 cm below seafloor (cmbsf), where they show signs of dissolution, and such specimens have not been employed in this study. Benthic foraminiferal samples were analyzed at the Woods Hole Oceanographic Institution using an isotope ratio mass spectrometer (IRMS) Finnigan MAT 253. Oxygen and carbon isotope data were calibrated with an NBS-19 standard and expressed relative to Pee Dee belemnite (PDB) using conventional δ notation in ‰. The instrumental precision was <0.09‰ for δ18Obf and <0.07‰ for δ13Cbf.
 The shallow infaunal U. akitaensis has been previously shown to calcify its test close to the calcite oxygen isotope equilibrium [Shackleton, 1974]. Accordingly, δ18Obf reflects the δ18O value and temperature of the local bottom water. The δ18O of the local bottom water, in turn, reflects δ18O of the average seawater, which is approximately represented by the benthic δ18O stack [e.g., Lisiecki and Raymo, 2005], and the salinity deviation of the local bottom water from the average seawater. In our data, δ18Obf values of the shallow infaunal B. tenuata are consistent with δ18Obf values measured on U. akitaensis (Table S1), and are also used in this study.
 On the other hand, δ13C of the shallow infaunal Uvigerina is offset from bottom water δ13C by as much as −0.8‰ due to influence of pore water δ13C, which largely depends on the organic carbon rain rate at shallow infaunal depths [McCorkle et al., 1990, 1997]. Since pore water δ13C decreases through decomposition of organic matter settling from above during the course of lateral advection of the water [Kroopnick, 1985; Matsumoto et al., 2002], δ13Cbf variations may be interpreted in terms of changes in the organic carbon rain rate and/or changes in intermediate water ventilation. Precision of the δ13C and δ18O estimate for Uvigerina is ±∼0.16‰ (1σ) due to intraspecific variabilities in the isotopic disequilibrium of δ13C and δ18O [Graham et al., 1981]. We use occurrences of laminations and/or the shallow infaunal B. tenuata as indicators of low oxygen on the sediment surface [e.g., Kennett et al., 2000].
 While in most samples of PC-23A foraminifera are exclusively of clear white color, foraminiferal shells with unusual yellow color are included at specific horizons, which are typically characterized by more negative δ13Cbf values as follows: −2.4‰ at 1601 cmbsf, −1.3 to −2.9‰ between 1235 and 1262 cmbsf, −2.0% at 1171 cmbsf, −1.5 to −1.7‰ between 998 and 1008 cmbsf, and −1.7‰ at 575 cmbsf (Table S1). Because such samples may have been influenced by authigenic carbonate overgrowth due to methane oxidation within the near surface sediments [Cook et al., 2011; Uchida et al., 2008], their δ13Cbf values are not used in this study.
5.5. Radiocarbon Measurements
 To estimate ages of sediments in PC-23A, 14C of planktonic foraminifera Neogloboquadrina pachyderma and Globigerina bulloides were measured at the NIES-TERRA Accelerator Mass Spectroscopy facility of the National Institute for Environmental Studies in Tsukuba, Japan. Samples were washed with tap water over a 63 μm screen, rinsed in distilled water and then dried at 50°C. Planktonic foraminifera were handpicked from samples, and then cleaned by soaking in 30% hydrogen peroxide solution to remove adhering contaminants. If abundance was high enough, single species (left-coiling N. pachyderma or G. bulloides) were used for dating. Otherwise, mixed species of N. pachyderma and G. bulloides were employed. Cleaning and graphitization of foraminifera at NIES-TERRA were carried out according to procedures after Uchida et al. [2004, 2005]. Resulting 14C ages, analytical precisions, calendar ages and their 2σ uncertainties are listed in Table 1.
Table 1. The 14C and Calendar Ages of Planktonic Foraminifera in Core PC-23Aa
Libby Age [14C ka]
Analytical Error [yr]
Cal. Age (−2σ) [ka BP]
Cal. Age (+2σ) [ka BP]
Cal. Age (Med) [ka BP]
Total Error (Lower Limit) [yr]
Total Error (Upper Limit) [yr]
The 14C analyses were performed at the AMS facility of the National Institute for Environmental Studies (NIES-TERRA). Calendar age conversions were performed using Intcal/Marine09 [Reimer et al., 2009] (ka BP = 1000 years before present). Planktonic species: G.b. = G. bulloides; N.p. = N. pachyderma. For 14C ages younger than 24 ka, total error includes the analytical error, the 2σ calibration error and the uncertainty in reservoir age. For 14C ages older than 24 ka, total error includes the analytical error, the uncertainty in reservoir age and an assumed uncertainty in radiocarbon calibration of ±2000 years as outlined in the text. Asterisk indicates radiocarbon dates used in this study.
G.b. + N.p.
G.b. + N.p.
G.b. + N.p.
G.b. + N.p.
G.b. + N.p.
G.b. + N.p.
G.b + N.p.
6. Lithology of PC-23A
 Sediments in core PC-23A were classified according to the ODP classification scheme [e.g., Ingle et al., 1990]. According to smear slide observation, the sediments are composed of siliceous and calcareous biogenic remains such as diatoms, radiolarians, nanofossils and foraminifera, and siliciclastic grains including quartz, feldspar, clay minerals and lithics with occasional contribution of volcanic glass shards. The degree of lamina preservation and/or bioturbation is commonly related to bottom water oxygenation levels, and classified as laminated, bioturbated (i.e., distinct burrows are identified) or strongly bioturbated (i.e., homogeneous in appearance) [e.g., Watanabe et al., 2007].
 A lithological column of PC-23A is shown in Figure 2. The core is dominantly composed of dark olive gray, gray or olive black silty clay, which is calcareous and/or siliceous to various extent and generally associated with pyrite-filled small burrows oblique to bedding (Figure 2a) or with patches of very fine sand sub-parallel to bedding (Figure 2b). Burrows generally have penetration depths of ∼1 cm and rarely exceed 4 cm after compaction. Laminated diatomaceous silty clay occurs at five intervals in the upper part and at one interval in the lowest part of PC-23A (Figure 2c). The lamination is characterized by alternation of olive black diatomaceous laminae with fromboidal pyrite and gray silty clay laminae with rare diatoms. Laminae, usually 1–2 mm thick, have distinct and continuous boundaries suggesting absence of bioturbation. Two to twenty-six cm thick layers of very fine sand are occasionally intercalated (Figure 2d). A sharp contact is observed in X-ray radiographs at ∼25 ± 1 cmbsf. One vitric ash layer was identified between 147 and 154 cmbsf.
7. Age Model for PC-23A
 Twenty-one samples were selected for 14C measurements from horizons with relatively high abundance of planktonic foraminifera. Raw 14C data were calculated to conventional 14C ages using the Libby half-life [Stuiver and Polach, 1977], and converted to calendar ages using the Intcal/Marine09 calibration curve [Reimer et al., 2009] and the Calib 6.0 software [Stuiver and Reimer, 1993]. For radiocarbon ages older than ∼30 thousand years (ka), radiocarbon age calibration becomes less well constrained due to larger uncertainties in the atmospheric Δ14C inventory and surface water reservoir ages, and limited availability of calibration data sets [Reimer et al., 2009]. Taking into account considerable variability in the different 14C calibration data sets (±∼1000 years) for 14C ages older than ∼30 ka [Reimer et al., 2009], the relatively large sparsity of calibration data points during the same period in Intcal/Marine09 [Reimer et al., 2009], and unknown reservoir ages during MIS 3, a calibration error of ±2000 years for 14C ages older than 30 ka is assumed in this study.
 Calendar ages were corrected assuming a 780 year reservoir age, which is the sum of the 400 year global surface water reservoir age [Stuiver and Braziunas, 1993] and a regional reservoir age (ΔR) of 380 years, assumed for the modern Bering Sea and subarctic North Pacific [e.g., Robinson and Thompson, 1981; Dumond and Griffin, 2002; Kuzmin et al., 2002]. Between 23 and 14 ka before present (hereafter “ka” denotes “ka before present”), surface water reservoir ages of 300 to 1150 years have been determined for the terminal region of the thermohaline circulation (subarctic North Pacific) [Sarnthein et al., 2007]. This uncertainty in reservoir ages (±∼400 years) is included in the total error assessment of radiocarbon dates (Table 1 and Figure 3).
 Results of 14C measurements, in part previously published in Itaki et al.  for calendar ages younger than 38 ka, are listed in updated form in Table 1. An age-depth diagram is presented in Figure 3. Calendar ages are shown as minimum, maximum and median values, which represent the range of analytical uncertainties. Total error of calendar ages older than 24 ka are the sum of uncertainties in the reservoir age, the calendar age calibration and the 14C analytical precision. The total error range is illustrated also in Figure 3.
 At the depth of 244.5 cmbsf (sample 3–34) both N. pachyderma and G. bulloides have been measured. The two ages disagree by 135 years, which is within the aggregate error of measurement precision, but also could be due to the error associated with bioturbation as this sample is from a non-laminated interval. In this study, we employed the age determined for N. pachyderma.
 Calibrated ages increase with depth with three exceptions. Sample 2–3 is younger than sample 1–25 by ∼20 years in calendar age, which is within the analytical error considering that the two samples are stratigraphically only 8.17 cm apart. As for sample 3–34, we chose the 14C age measured for N. pachyderma (sample 2–3) for our age model. Sample 7–40 is younger than sample 7–39 by ∼80 years in calendar age, which also is within the analytical error of these two neighboring samples that are 2.3 cm apart. We used the age of sample 7–40 for the age model, as it has a smaller analytical uncertainty than sample 7–39. The third exception is sample 14–32 at 1397.9 cmbsf with a calendar age of 36.17 ka, which is younger than the five overlying radiocarbon dates. Because PC-23A shows no sediment disturbances based on visual core description and X-ray radiographs, we find it unlikely that sample 14–32 was contaminated by younger foraminifera from depths of ∼950 cm or above, where calendar ages of less than ∼37 ka are expected, especially in view of reasonable calendar ages between 950 and 1300 cm depth. Further research is thus required to determine the reason for the obviously too young age. In this study, sample 14–32 is not employed for the age model.
 Radiocarbon values of the neighboring samples 13–19 (1262.0 cmbsf), 13–20 (1264.2 cmbsf) and 13–21 (1266.5 cmbsf) agree within ∼2 kyr, which are within the range of total error. From these three radiocarbon dates we used the calendar age of sample 13–21 (41.95 ± ∼2.7 ka), as it contained the largest amount of G. bulloides and corresponds to a %CaCO3 maximum (peak), where dating error due to bioturbation is minimized.
 In order to independently crosscheck our radiocarbon-based age model, we also show in Figure 3 the last occurrence (LO) of the radiolarian species Lychnocanoma nipponica sakaii in PC-23A [Itaki et al., 2009], which is widely recognized in the northern Pacific and in the Okhotsk Sea at ∼50 ka [Morley and Nigrini, 1995; Okazaki et al., 2003]. This event was dated as 48.6 ± 1.8 ka based on correlation of paleointensity records from several Bering Sea cores, in which LO of L. nipponica sakaii were identified, with the global paleointensity stack GLOPIS-75 of Laj et al.  [Okada et al., 2005]. Calendar age uncertainty in GLOPIS-75 for ages older than 40 ka is at least ±2 kyr [Laj et al., 2004; McMillan and Constable, 2006; Singer et al., 2009]. LO of L. nipponica sakaii occurred at a depth between 1330 and 1340 cmbsf in PC-23A, which implies an additional error of 0.23 kyr due to depth uncertainty of ±5 cm, when linearly interpolating between radiocarbon age control points. Thus estimated LO of L. nipponica sakaii at 48.6 ± 4 ka is consistent within error limits with the radiocarbon-based age estimation (Figure 3).
 We applied linear interpolation between the age control points. As shown in Figure 3, the sedimentation rate is ∼53 cm/kyr between core bottom (1758 cmbsf; ∼51.1 ka) and 1267 cmbsf (∼41.8 ka), ∼44 cm/kyr between 1267 and 1124 cmbsf (∼41.8 and ∼38.5 ka), ∼80 cm/kyr between 1124 and 1006 cmbsf (∼38.5 and ∼37.0 ka), slower at 23 to 29 cm/kyr in the interval between 1006 and 416 cmbsf (∼37.0 and ∼15.1 ka), and generally faster in the range of 57 to 171 cm/kyr between 416 and 28.1 cmbsf (∼15.1 and ∼11.1 ka). Based on a study of 210Pb on a multicore (MC-25) taken from the same site as PC-23A, the core top age is estimated as younger than 200 years (N. Harada, unpublished data, 2007). Hence, we consider the sharp contact at ∼25 ± 1 cmbsf as a non-depositional surface (hiatus) that spans the interval between ∼11.1 and ∼0.2 ka.
8.1. The δ18O Variation of Benthic Foraminifera
 The δ18Obf in PC-23A are presented in Figure 4f, jointly with their 2-point moving average, in order to filter decadal to centennial-scale noise and to emphasize millennial-scale variations. During MIS 3, the δ18Obf data range between 4.7 and 4.2‰ and are characterized by variations of 0.2 to 0.5‰ in amplitude with maxima observed at ∼49, ∼45, ∼43, ∼38, ∼35 and ∼30 ka. Between ∼27 and ∼18.7 ka (during MIS 2), δ18Obf are higher than during MIS 3, ranging from 4.6 to 4.9‰ with maxima at ∼27, ∼21 and ∼18.7 ka. The rapid deglacial decrease in δ18Obf begins at ∼18.7 ka, which agrees in timing with the initial melting of ice sheets and eustatic sea level rise [Yokoyama et al., 2000; Dyke et al., 2002]. Between ∼18.7 ka and the earliest Holocene, δ18Obf steeply decreases from 4.9‰ toward a minimum of 3.5‰ at ∼11 ka. The δ18Obf decrease is interrupted by two plateaus of rather constant δ18Obf values between ∼18.0 and ∼16.1 ka, and ∼13.0 and ∼12.1 ka, respectively, which are called “δ18Obf plateaus” in this study (Figure 5c).
8.2. The δ13C Variation of Benthic Foraminifera
 The δ13Cbf in PC-23A are presented in Figure 4e together with their 2-point moving average (averaging was not applied over intervals, where δ13Cbf data is missing due to absence of U. akitaensis or occurrences of yellowish foraminifera). At orbital scales, δ13Cbf slightly decreases from −1.1‰ at ∼50 ka to −1.9‰ at ∼47 ka, then increases to −0.7‰ at ∼38 ka, and shows an average value of −1.0‰ between 38 and 30 ka. After 30 ka, δ13Cbf decreases to −1.5 ‰ at ∼22.5 ka and increases again to −0.5‰ at ∼11 ka (Figure 4e). Superimposed on this orbital-scale trend, millennial-scale variations of 0.2 to 0.6‰ in amplitude are evident. During deglaciation δ13Cbf show two maxima between ∼17.3 and ∼15.9 ka, and ∼12.9 and ∼12.1 ka, which are interrupted by low δ13Cbf values between ∼15.3 and ∼13.0 ka that approximately correspond to a laminated interval (Figures 4 and 5d). It is worth to note that laminations start ∼500 years after the onset of the strong decrease in δ13Cbf at ∼15.6 ka. The low-oxygen benthic species B. tenuata is common all through the interval from ∼15.6 to ∼13.0 ka, suggesting that oxygen levels of intermediate water remained low during this period.
8.3. Variations in %CaCO3
 Microscopic observation revealed that the variations in %CaCO3 dominantly reflect variations in the absolute abundance of planktonic foraminifera (Figures 4c and 4d) and only to a minor degree benthic foraminifera abundances or the occasional presence of authigenic carbonate. %CaCO3 is highly variable prior to 36 ka with prominent peaks repeated every few thousand years. It is relatively low but still variable from 36 to 24 ka, whereas it is rather stable at moderately high values from 24 to 18 ka (Figure 4c). During deglaciation %CaCO3 is low in the bioturbated intervals (∼17.3 to ∼16.0 ka and ∼13.0 to ∼12.1 ka) and rather high in the dominantly laminated intervals between ∼15.6 and ∼13.0 ka (corresponding to the B/A period) as well as between ∼12.1 and ∼11.1 ka (earliest Holocene). A marked %CaCO3 peak occurs at ∼14.0 ka (Figures 4c and 5e). The temporal variation of %CaCO3 is consistent with %CaCO3 of PC-23A independently measured by a coloumbmeter [Kim et al., 2008; Itaki et al., 2009].
8.4. Relationship Between %CaCO3, δ18Obf and δ13Cbf, and Their Association With Greenland Ice δ18O
 The %CaCO3, δ18Obf and δ13Cbf profiles of PC-23A are compared to each other and to the NGRIP δ18O record [North Greenland Ice Core Project Members, 2004], in order to examine possible associations between D-O cycles and paleoceanographic changes recorded in the sediments at site MR06-04-23. Gorbarenko et al.  correlated intervals of increased %CaCO3 during the last glacial and deglacial period in the southern Bering Sea to DOIS, because these intervals were also associated with increased abundances of planktonic foraminifera (in particular thermophilic species) and warm water diatom species. Also in PC-23A, the prominent %CaCO3 peak during deglaciation coincides with the B/A warm period, which is associated with a laminated interval as was also noted by Cook et al. . Thus, if we assume that the intervals of high %CaCO3 are associated with warm periods during the last glacial period, as is observed during the well-dated deglaciation, %CaCO3 peaks between 51 and 37 ka in PC-23A can be correlated to a suite of DOIS evident in the NGRIP δ18O record, possibly to DOIS 14, 12, 11, 10, 9, and 8 (Figures 4b and 4c) within the error of our age model (<3 kyr). Between 37 and 25 ka, correlation of %CaCO3 peaks and NGRIP interstadials is more uncertain. Although a correlation of %CaCO3 peaks to DOIS 6, 5 and 4 is possible, it should be noted that not all %CaCO3 peaks can be matched to DOIS (Figures 4b and 4c).
 During deglaciation, the stepwise decreases in δ18Obf seem to lead similar decreases in δ18O of NGRIP by ∼0.5 to ∼1 kyr (Figures 5b and 5c). The δ18Obf plateaus are associated with distinct positive excursions of δ13Cbf (Figure 5d), low %CaCO3 (Figure 5e) and bioturbated intervals, and probably correlate to H1 and YD with apparent time leads of ∼1 kyr and ∼0.5 kyr, respectively (Figure 5). On the other hand, the interval between ∼15.5 and ∼13 ka, which is characterized by three consecutive δ18Obf minima, very low δ13Cbf values (−1.4 to −1.8‰ based on U. akitaensis), high %CaCO3, and dominance of lamination, probably correlates to the B/A period with an apparent time lead of ∼1 kyr. Finally, the low values of δ18Obf after 12 ka that are coeval with %CaCO3 peaks and low δ13Cbf, probably correlate to the earliest Holocene with an apparent time lead of ∼0.5 kyr (Figure 5).
 Similarly, during the last glacial period, positive excursions of δ18Obf tend to coincide with positive excursions of δ13Cbf and low %CaCO3 in PC-23A (Figure 4). If we assume that the association of higher δ18Obf, higher δ13Cbf and lower %CaCO3 with cold periods, as observed during deglaciation, is maintained during the last glacial period, positive excursions of δ18Obf and δ13Cbf, and low %CaCO3 correlate – within the error of our age model and under the constraints of our proposed correlation between %CaCO3 peaks and NGRIP interstadials – to Heinrich events 5, 4 and 3 (hereafter H5, H4, H3) and several DOS, tentatively to DOS 14, 12, 10, 8 and 6 (Figure 4). It should be noted that the positive δ13Cbf excursion at ∼45 ka (tentatively correlated to H5) corresponds in the earlier part to a minimum and in the later part to a maximum of δ18Obf. Positive δ13Cbf excursions at ∼40 and ∼28 ka are associated with weak temporal δ18Obf increases. The positive δ13Cbf excursion at ∼35 ka, possibly associated with DOS 8, is less pronounced. Also, the positive δ18Obf excursion at ∼27 ka, which might correlate to an episode of DOS 3, is associated with a generally weak δ13Cbf increase, although this increase is interrupted by a single negative δ13Cbf excursion. An exception to the general trend is the positive δ13Cbf excursion at ∼24 ka, possibly corresponding to H2, which is associated with rather low δ18Obf values. On the other hand, negative excursions of δ18Obf and δ13Cbf at ∼51, ∼44, ∼42, ∼38.5, ∼37 and ∼31 ka, which also correspond to %CaCO3 peaks, can be correlated to several DOIS, tentatively to DOIS 14, 12, 11, 9, 8 and 5 (Figure 4), although the decreases in δ18Obf are not as abrupt as the increases in %CaCO3.
 In summary, high δ18Obf and δ13Cbf events probably correlate with H1 and YD during deglaciation and, under the above outlined assumptions, to Heinrich events and several DOS during MIS 3.
9.1. Cold and/or Saline Intermediate Waters During the LGM in the Northeastern Aleutian Basin
 At site MR06-04-23 at ∼1000 m water depth, the water temperature is at present 2.8°C, salinity 34.37‰ (MR06–04 hydrological data), and δ18O of seawater −0.7‰ PDB (A. Ijiri, personal communication, 2009). Using the relationship between water temperature and δ18Obf of Shackleton  for Uvigerina species, the expected modern δ18Obf is ∼2.8‰, implying a δ18Obf difference of −1.8 to −2.1‰ between the LGM and modern settings. Accepting a change in δ18Obf of −1.0 ± 0.1‰ [Schrag et al., 2002] due to ice volume changes between the LGM and present, the remaining −0.7 to −1.1‰ of the LGM and modern δ18Obf difference would be due to temperature and/or local salinity changes. Consequently, under the assumption that local salinity anomalies did not change between the LGM and early Holocene, LGM intermediate water temperature was between −0.2 and −2.0°C using a 0.23‰ increase in δ18Obf per 1°C decrease in temperature [Epstein et al., 1953]. Alternatively, under the assumption of constant temperature, LGM salinity levels in the Bering Sea would have ranged between 36.1 and 37.1‰, if we use the equation
determined by isotopic analyses of seawater from 1000 to 10 m depth in proximity of site MR06-04-23 (Ijiri, personal communication, 2009). Although salinity of the deep LGM ocean could have reached 36.1 to 37.1‰ [Adkins et al., 2002], these extreme values were probably not reached at 1000 m depth in the Bering Sea, because an intermediate water mass of relatively low salinity extended to site MR06-04-23 during the LGM according to oxygen and hydrogen isotope analyses of pore water in PC-23A and two other neighboring cores from 853 and 1157 m water depth [Ijiri et al., 2009]. This implies colder than present intermediate water at site MR06-04-23, which is consistent with evidence from radiolarian assemblages that suggest the presence of an extended low-temperature intermediate water mass in the Bering Sea during the LGM [Tanaka and Takahashi, 2005].
9.2. Possible Causes of Millennial-Scale δ18Obf and δ13Cbf Variations at Site MR06-04-23
 The millennial-scale δ18Obf maxima during the last glacial period and the δ18Obf plateaus during the deglacial period in PC-23A, which we tentatively correlated to Heinrich events H5, H4, H3 and H1, as well as YD and several DOS (Figure 4), are considered to reflect periodic emergence of colder and/or more saline intermediate waters on the upper part of the continental slope. These intervals are also associated with lower %CaCO3, lower biogenic silica contents [Itaki et al., 2009] and higher δ13Cbf values. Combined, our results could be interpreted in terms of increased ventilation at intermediate water depths and low surface productivity during severe cold periods. Primary production would have been limited during Heinrich events and DOS due to lower nutrient levels in the surface water and a prolonged ice-covered period over the slope [Sancetta and Robinson, 1983] that limits availability of light and accordingly the length of the blooming season. In addition, new production of nutrient-poor intermediate water in the Bering Sea tends to dilute the nutrient-rich waters derived from the deep North Pacific that upwell at the terminus of the thermohaline circulation, decreasing nutrient availability and productivity in Bering Sea surface waters.
 Decreases in δ13Cbf, on the other hand, can be attributed to higher organic carbon accumulation due to increased surface productivity, and/or decreased ventilation and consequent aging of the water mass. Coeval maxima in %CaCO3 and minima in δ13Cbf in PC-23A at intervals, which we suggest to coincide with DOIS (Figure 4), are consistent with a scenario of enhanced surface productivity associated with poorer oxygenation within the surface sediment. An increase in oxygen demand within the sediments is also suggested by the occasional appearance of B. tenuata in those intervals. The %CaCO3 peaks and negative δ13Cbf excursions in PC-23A also commonly coincide with lower δ18Obf, which suggests a relationship between surface productivity and intermediate water temperature and/or salinity. Due to efficient delivery of nutrients through melting ice and upwelling from the deep Aleutian Basin, the Bering Sea is today characterized by high surface productivity, in particular along the Bering Sea Green Belt [Springer et al., 1996]. A similar scenario could be drawn for warm episodes of the last glacial period, when a reduction or cessation of intermediate water ventilation, accompanied by warming and/or a salinity decrease of the intermediate water (lower δ18Obf), could have facilitated upwelling of nutrient-rich North Pacific deep waters and, in conjunction with nutrients derived from melting ice, promoted higher surface productivity as suggested by increased %CaCO3 and lower δ13Cbf.
9.3. Possible Source of NPIW During the Last Glacial and Deglacial Period
 It is believed that the northwestern Pacific Ocean shallower than ∼2000 m water depth was well ventilated during the last glacial period [Keigwin, 1998]. The source of its ventilation, however, is still under debate [e.g., Matsumoto et al., 2002]. Ohkushi et al.  interpreted an opposite glacial-interglacial abundance pattern of Cycladophora davisiana, which dwells in well-oxygenated water, between the Okhotsk Sea (interglacial) and the Bering Sea (glacial) as a source switch of NPIW between the two seas. Decreased ventilation at intermediate depths in the Okhotsk Sea during the last glacial period has also been proposed based on radiolarian abundances [Okazaki et al., 2008].
 Today, seawater δ18O between 750 and 1250 m depth in the Okhotsk Sea and northeastern Bering Sea is −0.18 ± 0.04‰ (n = 73) and −0.35 ± 0.24‰ (n = 5), respectively, based on the world ocean δ18O database (G. A. Schmidt et al., Global Seawater Oxygen-18 Database v1.21, 1999, available at http://data.giss.nasa.gov/o18data/) and δ18O measured at site MR06-04-23 (Ijiri, personal communication, 2009). These data suggest an average negative δ18Obf gradient between the Okhotsk Sea, where NPIW primarily forms, and the northeastern Bering Sea (Δδ18Obf = δ18Obf Bering – δ18Obf Okhotsk). Such observation is consistent with our idea that the opposite situation, when NPIW formation in the Bering Sea surmounts that of the Okhotsk Sea, would be characterized by a positive Δδ18Obf.
 We accordingly compared our δ18Obf record of the northeastern Bering Sea slope with that of the central Okhotsk Sea core XP98-PC1 [Sakamoto et al., 2005] (Figure 4a), which is located at 51°00.9′N, 152°00.5′E (Figure 1), in order to explore the possibility of NPIW formation in the glacial Bering Sea. The water depth of XP98-PC1 (1107 m) is comparable with that of PC-23A. Site XP98–1 is located proximate to the modern source region of NPIW [e.g., Shcherbina et al., 2003]. Age control points of XP98-PC1, which are based on oxygen isotope stratigraphy and tephrachronology in MIS 3 and radiocarbon dating at the LGM and deglaciation [Sakamoto et al., 2005], are indicated in Figure 4a. Based on the proposed correlation between %CaCO3 peaks in PC-23A and DOIS in NGRIP for the interval older than 24 ka (Figures 3, 4b, and 4c), we construct a NGRIP-tuned age model for PC-23A, in an effort to minimize dating errors when comparing PC-23A and XP98-PC1. δ18Obf measurements on XP98-PC1 were carried out on Uvigerina proboscidea for the time period prior to 15.3 ka and U. akitaensis after 15.3 ka at Hokkaido University using an IRMS Finnigan MAT 251. Results were calibrated through a NBS-20 standard and expressed relative to PDB with an instrumental precision of <0.03‰ [Sakamoto et al., 2005]. Both species gave identical δ18Obf values within error and are therefore suitable for a direct comparison to the δ18Obf record of PC-23A.
 In orbital scale, the two δ18Obf records are similar in shape and values, although PC-23A shows higher amplitude variations in δ18Obf during MIS 3 (Figure 4a). In both cores, δ18Obf increase during MIS 2 (LGM) and rapidly decrease simultaneously during deglaciation. Aggregate uncertainties in age estimation between XP98-PC1 and the NGRIP-tuned PC-23A record prior to ∼25 ka allow us to discuss with confidence only gross features such as the δ18Obf increases to ∼4.7‰ at ∼46 and ∼39 ka, which we interpret to correspond to DOS 12 and H4, respectively. Even if considering the age uncertainties in the two age models and possible inter-machine and/or interlaboratory biases in δ18Obf analyses, which could be at the order of 0.1 to 0.2‰ [Zahn and Mix, 1991; Ostermann and Curry, 2000], δ18Obf of XP98-PC1 do not show such strong decreases as they are observed in PC-23A at ∼46 and ∼39 ka (Figure 4a). According to these data, therefore, a positive Δδ18Obf seems to have been established between the Bering and Okhotsk Seas at least during these two periods.
 We conclude from above considerations that intermediate water was colder and/or more saline in the Bering Sea than in the Okhotsk Sea during DOS 12 and H4 according to our correlation, and possibly during other cold periods such as DOS 14 and H3. At intervals, which we correlated to DOIS, on the other hand, the two cores show rather similar δ18Obf values suggesting absence of strong temperature and/or salinity gradients at intermediate water depth between the Bering and Okhotsk Seas. It should be noted that due to the limits of our age model, we cannot exclude the possibility that a positive Δδ18Obf could have started during the late part of DOIS 12 (Figures 4a and 4b). While further study is needed to better constrain the timing of Δδ18Obf increases, we suggest at this stage, in line with our discussion in chapter 8.4 on increases of δ18Obf during cold periods in the Bering Sea, that the establishment of positive Δδ18Obf would have been largely a feature of cold periods.
 Based on paired benthic-planktonic foraminiferal 14C age differences in the piston cores MR01-K03 PC4 and PC5 at 1366 m water depth off Shimokita peninsula in northeastern Japan, Ahagon et al.  proposed enhanced ventilation between ∼17 and ∼15 ka, and ∼12.5 and ∼11.4 ka corresponding to H1 and YD, respectively, and relatively sluggish ventilation between ∼15 and ∼12.5 ka corresponding to the B/A period (Figure 5a). In order to obtain information on the source region(s) and direction of lateral advection of intermediate water during these abrupt deglacial ventilation changes off Shimokita, we compared δ18Obf from PC-23A with those from piston core MD01–2412, which are also based on U. akitaensis [Sakamoto et al., 2006; M. Ikehara, unpublished data, 2009] (Figure 5c). δ18Obf of MD01–2412 were analyzed at Kochi University using an IRMS IsoPrime GV Instruments. Results were calibrated through a NBS-19 standard and expressed relative to the Vienna PDB with an instrumental precision of <0.03‰ [Sakamoto et al., 2006]. MD01–2412 is located geographically near PC4/PC5 at 44°31.65′N, 145°00.25′E in the southwestern Okhotsk Sea at a water depth of 1225 m (Figure 1).
 The δ18Obf plateaus of PC-23A that correspond to H1 and YD are ∼0.3‰ higher than respective δ18Obf values in MD01–2412 (Figure 5c), although it should be noted that Δδ18Obf may range between 0.1 and 0.5‰ during these periods due to possible inter-machine and/or interlaboratory biases in δ18Obf analyses [Zahn and Mix, 1991; Ostermann and Curry, 2000]. The observed positive Δδ18Obf may suggest that the source region of better ventilated intermediate waters during H1 and YD [Duplessy et al., 1989; Ahagon et al., 2003] was proximate to the northern Bering Sea rather than the Okhotsk Sea, supporting the view that lateral advection of intermediate water occurred from the Bering Sea to the southwestern Okhotsk Sea.
 Between ∼15.3 and ∼13.0 ka, a period approximately corresponding to the B/A period and mostly laminated in PC-23A, δ18Obf values in PC-23A and MD01–2412 are similar (Figure 5c), suggesting that the two regions were bathed by the same water mass, which could be linked to reduction or cessation of intermediate water formation in the Bering Sea. During the same period, δ13Cbf (based on U. akitaensis) are very low in PC-23A (Figure 5d), suggesting increased organic carbon flux and/or decreased ventilation. A reduction in intermediate water ventilation in the Bering Sea during the B/A period could be causally linked to the synchronous reduction in ventilation off Shimokita peninsula [Duplessy et al., 1989; Ahagon et al., 2003]. Ventilation was also reduced off Shimokita peninsula during the late LGM (Figure 5a), which is associated with absence of a significant lateral δ18Obf gradient between PC-23A and MD01–2412 (Figure 5c), suggesting a rather stagnant condition of the intermediate water.
 Hence, increases in intermediate water ventilation off Shimokita peninsula during H1 and YD [Duplessy et al., 1989; Ahagon et al., 2003] are consistent with colder and/or more saline intermediate water (NPIW) production in the Bering Sea, and its lateral advection to the region off northeastern Japan. Our results therefore support the Bering Sea as a source of NPIW during severe cold periods such as DOS 12, H4, H1 and YD and reduction or absence of NPIW production in the Bering and Okhotsk Seas during interstadial periods.
9.4. Northern Hemispheric Cooling and Associated Atmospheric Reorganization During Stadial Periods as Triggers of NPIW Formation in the Bering Sea
 The benthic isotopic records in PC-23A and their comparison with those in the Okhotsk Sea as well as with ventilation changes off northeastern Japan demonstrate the potential importance of the Bering Sea as a source of the ventilated intermediate water masses observed in the glacial North Pacific [Keigwin, 1998; Duplessy et al., 1989; Matsumoto et al., 2002]. Our results are also consistent with the hypothesis of Ohkushi et al. , who advocated the occurrence of better ventilated intermediate water in the Bering Sea during the last glacial period based on radiolarian assemblages. Using a coupled ocean-atmosphere climate model, Kim and Park  suggested significant increases in surface water densities in the Bering Sea during the last glacial period due to surface water cooling and salinity increases, which would primarily have resulted from a decrease in the precipitation/evaporation ratio over the northern North Pacific. The associated destabilization of the water column could have effectively contributed to NPIW formation in the northern marginal seas of the North Pacific [Kim and Park, 2008]. A positive salinity feedback arising from a compensatory northward transport of saline subtropical Pacific surface waters could have further enhanced NPIW formation in the Bering Sea. However, our results suggest that NPIW formation – at least in the Bering Sea – was not necessarily active all through the glacial period, but only during severe cold periods, such as Heinrich events and some DOS.
 An important mechanism that may modulate surface water densities and thereby constrain the timing and location of intermediate water formation is the rejection of brine during sea ice formation around persistent polynyas [e.g., Talley, 1991; Shcherbina et al., 2003; Haley et al., 2008]. Such ice-free zones are often created when dominant winds from the continent push sea ice away from the coast. The generated dense water sinks to depths until a new density equilibrium with the surrounding water masses is reached. Such process is believed to be a primary control on NPIW formation in the modern Okhotsk Sea [e.g., Talley, 1991, 1993; Yasuda, 1997].
 Today, cold northerly winds over the Okhotsk Sea are associated with the Aleutian Low over the Okhotsk Sea region and the Siberian High over central Siberia and northern China [e.g., Cavalieri and Parkinson, 1987]. The coldest winters in the Bering Sea, on the other hand, occur when the Aleutian Low is displaced to the east and anomalously high pressure dominates over northeastern Siberia favoring northerly winds over the Bering Sea [Luchin et al., 2002; Rodionov et al., 2007] (Figure 6). In accordance with these observations, Cavalieri and Parkinson  reported episodic out of phase relationships in sea ice extent at weekly to monthly scale between the Bering and Okhotsk Seas during the years 1972 to 1976, when northerly winds over the Bering Sea, which led to an increase in sea ice extent over the Bering Shelf and Aleutian Basin, were associated with easterly or southerly winds over the Okhotsk Sea, which effectively decreased sea ice extent there (Figure 6). During the last glacial period, the Bering Sea was a semi-enclosed basin similar to the present Okhotsk Sea, and should have experienced increased extents of seasonal sea ice [e.g., Sancetta and Robinson, 1983; Katsuki and Takahashi, 2005]. Moreover, a prevalence of northerly winds has been suggested for the LGM based on loess deposits in Alaska [Muhs et al., 2003]. Hence, if northerly winds were dominant over the Bering Sea during cold periods such as Heinrich events in analogy with modern cold winters and the LGM, intense brine rejection and associated intermediate or deep water ventilation would have occurred around potentially widespread areas of polynya over the northeastern part of the Aleutian Basin.
9.5. Calm Realm With High Productivity in the Bering and Okhotsk Seas During Interstadials and the Bølling/Allerød Period
 Primary productivity at the northeastern slope of the Bering Sea appears to have drastically increased during periods, which we suggest to correspond to DOIS between ∼50 and ∼35 ka, based on prominent peaks of %CaCO3, generally increased biogenic silica contents [Itaki et al., 2009], and relatively low δ13Cbf. Gorbarenko et al.  also found increased productivity during almost all DOIS of the last 80 kyr in the Okhotsk Sea based on bio-productivity indicators such as barium, organic carbon, carbonate, chlorine and the abundance of benthic foraminifera. These high productivity events in both the Bering and Okhotsk Seas were possibly associated with decreases in NPIW formation in the Bering Sea, which are suggested by the occurrence of warmer and/or less saline, and possibly less oxygenated (consequently more nutrient-enriched) intermediate waters on the northeastern Bering Sea slope. Also the general absence of lateral advection between the Bering and Okhotsk Seas based on similar values in δ18Obf at intermediate depth during those periods implies that calm and rather stratified intermediate to surface water conditions should have prevailed in the two seas.
 As has been described before, the reduction in NPIW in the Bering Sea and deposition of laminated sediments during deglaciation at site MR06-04-23 apparently preceded the abrupt Bølling and earliest Holocene warming in Greenland by ∼1 and ∼0.5 kyr, respectively (Figure 5). These time leads significantly exceed the analytical calendar age uncertainties of ∼70 to ∼130 years (Table 1). We suggest that the time leads primarily reflect a temporal increase in the local surface water reservoir age to ∼1700 years at the onset of the Bølling and ∼1200 years at the onset of the earliest Holocene as a result of upwelling of old deep water that accumulated in the deeper part of the North Pacific during the LGM [Galbraith et al., 2007; Marchitto et al., 2007; Sarnthein et al., 2007; Tada et al., 2009]. These results are consistent with subarctic Pacific deep water ages of ∼1900 years during H1 and ∼1250 years during YD [Galbraith et al., 2007], although these ages may be overestimated by a few hundred years due to temporarily increased reservoir ages (∼1150 years) in the subarctic Pacific during deglaciation [Sarnthein et al., 2007]. Although further research is needed to clarify the nature of these time leads, our results suggest that climate warming associated with the onset of the B/A period and the earliest Holocene would have temporarily interrupted formation of NPIW in the Bering Sea, giving way to nutrient-rich, and warmer and/or less saline North Pacific deep waters that upwelled at the Bering Sea continental slope [Brunelle et al., 2007]. The ascension of this old and oxygen-poor water mass in the Bering Sea could have well increased the reservoir age of surface waters by several hundred years and led to the deposition of laminated sediments.
 Such shut-down of intermediate water formation in the Bering Sea during the B/A period might have also decreased oxygen levels along the western and eastern margins of the North Pacific beyond thresholds and caused the occasional deposition of laminated sediments [Behl and Kennett, 1996; van Geen et al., 2003; Crusius et al., 2004; Hendy and Pedersen, 2005], directly through decreased lateral advection of oxygen-rich intermediate water and/or indirectly through increased consumption of dissolved oxygen in the intermediate water due to enhanced productivity caused by upwelling of nutrient-rich deep waters along the continental margin. Enhanced productivity along the North Pacific margins during the B/A period may have temporarily increased biological sequestration of CO2, which is a possible mechanism to counteract large scale pCO2 rise during deglaciation.
10. Summary and Conclusions
 We examined a possible link between D-O cycles and changes in intermediate water properties and surface productivity in the northeastern Bering Sea. Coeval positive excursions of δ18Obf and δ13Cbf, and low %CaCO3 in PC-23A suggest emergence of relatively cold and/or saline and possibly better oxygenated intermediate water at the northeastern Bering slope during periods of low surface productivity, which we tentatively correlated to H5, H4, H3, as well as several DOS during MIS 3, and to H1 and YD during the deglaciation. A comparison of δ18Obf between the central Okhotsk Sea and the Bering Sea indicates that intermediate water considerably cooled and/or became more saline in the Bering Sea during at least two cold periods of MIS 3, which we tentatively correlated to DOS 12 and H4, while such cooling and/or salinity increase was not evident in the central Okhotsk Sea during MIS 3.
 Furthermore, a comparison between ventilation changes off eastern Japan with gradients of δ18Obf between the Bering Sea and the southwestern Okhotsk Sea indicates that intermediate water was colder and/or more saline in the Bering Sea (higher δ18Obf) than in the southwestern Okhotsk Sea (lower δ18Obf) during H1 and YD, when the ventilation age off northeastern Japan significantly decreased. We accordingly propose that the Bering Sea was the probable source of NPIW during H1 and YD and possibly other severe cold periods of MIS 3, and that the cold and/or saline water mass may have occasionally spread southward at least to the area off Shimokita peninsula.
 During periods, which we correlated to DOIS and the B/A period, on the other hand, NPIW formation seems to have weakened (or ceased) both in the Bering and Okhotsk Seas, judging from the occurrence of warmer and/or less saline intermediate water at the Bering Sea slope and the absence of significant δ18Obf gradients between the Bering and Okhotsk Seas. These periods were also characterized by high surface productivity in both the Bering and Okhotsk Seas. Based on coeval occurrences of low δ18Obf, low δ13Cbf and high %CaCO3 in PC-23A, we suggest that a reduction or shutdown of ventilation in the Bering Sea during warm periods facilitated upward migration of aged, nutrient-rich waters from the deep North Pacific and thus promoted surface productivity in the northeastern Bering Sea. These processes could have also caused the temporal increase in the marine reservoir age observed during the B/A period and the earliest Holocene. A reduction of ventilation during the B/A period in the Bering Sea may have been causally linked to sluggish ventilation off eastern Japan and could have contributed to generally low oxygen conditions at intermediate depths in the North Pacific at that time.
 An ultimate trigger for enhanced intermediate water production in the Bering Sea could have been predominantly northerly winds over the Bering Sea arising from an eastward shift of the Aleutian Low during severe cold periods, which should have created favorable conditions for active polynya formation and brine rejection in the northeastern Bering Sea. In the generally colder and dryer glacial environment, owing to greater continentality of the Bering Sea area [Ager, 2003; Kurek et al., 2009], the resulting increase in sea surface salinity and density could have been sufficient for ventilating intermediate and/or deep waters. At the same time, such eastward shift of the Aleutian Low would have led to more easterly or southerly winds over the Okhotsk Sea that prevented polynya formation there and facilitated the introduction of relatively moist air from the North Pacific. These processes would have lowered sea surface salinity and effectively reduced intermediate water formation in the Okhotsk Sea during cold episodes of the last glacial period.
 We express our sincere thanks to the crew and onboard scientists of cruise MR06-04 for their assistance in sampling and sample preparation. We are thankful for assistance in picking planktonic foraminifera for 14C analysis by K. Khim and S. Kim (Pusan University, South Korea). We are grateful for provision of modern bottom water oxygen isotope data at the studied core site by A. Ijiri. The comments of two anonymous reviewers greatly helped to improve the manuscript. This work is partly supported by grants-in-aid for scientific research from the Japan Society for Promotion of Science (18101001) to R.T. and by scholarships of the Ministry of Education, Science, Sports and Culture of Japan and the University of Tokyo to S.F.R.