Determining the ultimate cause and effect of millennial-scale climate variability remains an outstanding problem in paleoceanography, partly due to the lack of high-resolution records predating the last glaciation. Recent cores from Santa Barbara Basin provide ~2500–5700 year “windows” of climate with ~10–50 year resolution. Ages for three cores, determined by seismic stratigraphic correlation, oxygen isotope stratigraphy, and biostratigraphy, date to ~293 ka (MIS 8), ~450 ka (MIS 12), and ~735 ka (MIS 18). These records sample the Late Pleistocene, during which the 100 kyr cycle strengthened and the magnitude of glacial-interglacial cyclicity increased. Thus, these records provide a test of the dependence of millennial-scale behavior on variations in glacial-interglacial cyclicity. The stable isotopic (δ18O) composition of planktonic foraminifera shows millennial-scale variability in all three intervals, with similar characteristics (duration, cyclicity) to those previously documented during MIS 3 at this site. Stadial G. bulloides δ18O values are 2.75–1.75‰ (average 2.25‰) and interstadial values are 1.75–0.5‰ (average 1‰), with rapid (decadal-scale) interstadial and stadial initiations of 1-2‰, as in MIS 3. Interstadials lasted ~250–1600 years and occurred every ~650–1900 years. Stadial paleotemperatures were 3.5–9.5°C and interstadial paleotemperatures were 7.5–13°C. Upwelling, evidenced by planktonic foraminiferal assemblages and δ13C, increased during interstadials, similar to MIS 3; high productivity during some stadials was reminiscent of the Last Glacial Maximum. This study builds upon previous records in showing that millennial-scale shifts were an inherent feature of Northern Hemisphere glacial climates since 735 ka, and they remained remarkably constant in the details of their amplitude, cyclicity, and temperature variability.
 To date, researchers have been able to elucidate the general structure of late Quaternary climate fluctuations, based on a variety of archives including ice cores, marine sediments, and terrestrial deposits [e.g., Behl and Kennett, 1996; Dansgaard et al., 1993; EPICA Community Members, 2004; Lisiecki and Raymo, 2005; Winograd et al., 1992]. These records show globally correlated climate oscillations on orbital, millennial, and decadal timescales, with abrupt shifts between cold and warm episodes (both glacial-interglacial and stadial-interstadial) [e.g., Alley et al., 1993; Bond and Lotti, 1995; Dansgaard et al., 1993; EPICA Community Members, 2004; Grootes et al., 1993; Hendy and Kennett, 1999]. The recurring nature of these fluctuations, and the common pattern of rapid warming followed by slower cooling, suggests common feedbacks acting on the Earth system. However, our understanding of the causes, effects, and temporal evolution of these shifts is limited by the shortage of high-resolution data beyond ~100 ka. This lack of resolution has particularly hampered the interpretation of higher-frequency stadial-interstadial shifts.
 Santa Barbara Basin (SBB, Figure 1a) provides a unique opportunity to examine well-preserved sediments spanning the past ~1.4 Ma. SBB's high productivity and bathymetric configuration yield sedimentation rates in excess of 100 cm/kyr [e.g., Behl, 1995]—an order of magnitude higher than most other marine sites—that remained remarkably high and constant at many points throughout the basin for up to 1 Myr, despite glacial-interglacial climate shifts [Behl, 1995; Marshall, 2012]. Previously published paleoceanographic archives to 145 ka based on ODP Site 893 and subsequent piston cores [Behl and Kennett, 1996; Cannariato and Kennett, 2005; Hendy and Kennett, 1999; Hill et al., 2006] show SBB to faithfully record hemispheric climate events, including stadial-interstadial shifts during MIS 3 [Behl and Kennett, 1996; Hendy et al., 2002; Hendy and Kennett, 1999]. Importantly, climate shifts in SBB were likely synchronous [Hendy et al., 2002] with stadial-interstadial shifts observed in the Greenland ice core records [e.g., Dansgaard et al., 1993; Grootes et al., 1993].
 To extend the SBB record, a suite of piston cores was acquired along a submarine anticline (Figures 1b and S1 in the auxiliary material), where older strata are accessible at the sediment surface [Nicholson et al., 2006]. Estimated core ages were based on stratigraphic correlation to and interpolation between dated reference horizons, isotopic stratigraphy, and biostratigraphy (see auxiliary material for additional details). Although this suite of cores does not provide a continuous record, it does provide discrete high resolution “windows” of Quaternary climate since ~750 ka. For this study, we focus on cores MV0508-11JPC, MV0508-16JPC, and MV0508-20JPC, each of which spans ~2500–5700 years and is dated to 293 ka ±5 kyr (the MIS 8.6-8.5 transition) [e.g., Stirling et al., 2001], 450 ka ±5 kyr (during MIS 12), and 735 ka ±5 kyr (during MIS 18), respectively [Marshall, 2012]. Stadial-interstadial shifts in these cores enable the study of millennial-scale variability at time periods bracketing several key Late Pleistocene climatic events. The cores postdate the Mid-Pleistocene Transition (~900–800 ka), during which glacial-interglacial cyclicity shifted from 41 to 100 kyr timescales and mean global ice volume increased [Lisiecki and Raymo, 2007; Mudelsee and Schulz, 1997; Ruddiman et al., 1989]; thus, our records are within the “100 kyr world.” However, the oldest core (~735 ka) predates the rapid ramp-up of 100 kyr power in global climate records, which occurred from ~700–600 ka [Mudelsee and Schulz, 1997; Ruddiman et al., 1989]. The two oldest records also predate the Mid-Brunhes Event [Jansen et al., 1986] at ~430 ka, after which glacial-interglacial cycles display higher amplitude than before [e.g., EPICA Community Members, 2004]. Records of millennial-scale variability from the inception of the 100 kyr world [Kleiven et al., 2011; Raymo et al., 1998] are scarce, and the impact of the aforementioned Pleistocene climate events on millennial-scale behavior is unclear. Existing lower resolution records [e.g., Barker et al., 2011; Loulergue et al., 2008; McIntyre et al., 2001; McManus et al., 1999; Raymo et al., 1998] exhibit millennial-scale variability back to the MPT and early Pleistocene; however, the time resolution of these records (200–2500 years) confounds efforts to determine the similarity of those millennial-scale shifts to the Dansgaard/Oeschger cycles of MIS 3. In this investigation, we use three high (10–50 year) resolution records from SBB to address the following questions:
 Is there evidence for millennial-scale variability in the Pacific dating back to 735 ka (i.e., inception of the 100 kyr world)?
 Has the amplitude, duration, or rapidity of warming versus cooling of observed millennial-scale variability changed since the 735 ka?
 Is millennial-scale variability reflected in surface currents and productivity throughout the past 735 kyr?
 To address these questions, we examined changes in planktonic and benthic foraminiferal δ18O and δ13C, planktonic foraminiferal assemblages, and preservation of sediment laminations; together, these proxies yield records of temperature, productivity, upwelling, and oxygenation. We compare these records to published MIS 3 data from SBB [Behl and Kennett, 1996; Hendy and Kennett, 1999; 2000; 2003].
1.1 Site Setting
 The SBB is a silled basin ~600 m deep, with a western sill at 470 m and an eastern sill at 230 m [Eichhubl et al., 2002]. Regional surface water masses include the cool, fresh, nutrient-rich, equatorward flowing California Current (CC); the warm, salty, nutrient-poor, poleward flowing Davidson Countercurrent; and upwelled waters [Lynn and Simpson, 1987]. The CC and upwelled waters dominate SBB in spring and early summer, when the North Pacific High and continental low intensify and northerly winds strengthen [Hendershott and Winant, 1996; Huyer, 1983]. Countercurrent waters dominate the rest of the year [Hendershott and Winant, 1996]. The California Undercurrent, with a core of flow at 250–300 m, transports warm, saline, low-oxygen waters from the Eastern Tropical North Pacific poleward through the Southern California Bight [Lynn and Simpson, 1990]. Deep SBB waters derive from cold, fresh North Pacific Intermediate Water (NPIW) that is variably mixed with Subtropical Gyre waters and California Undercurrent water [Bograd et al., 2008], and are further depleted of oxygen by degradation of organic matter [Behl, 1995]. The present-day lack of oxygen excludes bioturbation and preserves annual laminations deposited in the basin [Behl, 1995].
 Planktonic foraminiferal assemblages, δ18O and δ13C, and sediment laminations were used as climate indices in SBB during MIS 3 [Behl and Kennett, 1996; Hendy and Kennett, 1999; 2000; 2003]. Typical stadial-interstadial shifts in planktonic foraminiferal δ18O were 1–2‰, from stadial values of 2.75–2.25‰ to interstadial values of 1.75–1‰ [Hendy and Kennett, 1999]. Interstadials initiated rapidly (in decades), lasted 200–3000 years [Hendy and Kennett, 1999], and occurred every 850–5300 years. Interstadials in early MIS 3 (40–60 ka) were longer with more variable temperatures than later interstadials [Hendy and Kennett, 1999]. A crucial aspect of the SBB record is that stadial-interstadial shifts were interpreted to occur near-synchronously with Greenland interstadial (IS) events, and IS 1–17 can be simply and directly correlated between the two records [Hendy et al., 2002; Hendy and Kennett, 1999]. Shifts in preservation of sediment laminations were also synchronous with Greenland IS events, with bioturbated stadials and laminated interstadials [Behl and Kennett, 1996]. These shifts were primarily explained as switches in deep basin source waters [Behl and Kennett, 1996] rather than productivity driven, as evidenced by synchronous 1–2°C shifts in basin bottom water temperature [Hendy and Kennett, 2003].
 The modern planktonic foraminiferal assemblage in SBB is dominated by Globigerina bulloides, Neogloboquadrina pachyderma (d), and Turborotalita quinqueloba, with minor contributions from N. dutertrei, Orbulina universa, and Globigerinoides ruber [Kincaid et al., 2000]. Stadial assemblages during MIS 3 were dominated by N. pachyderma (s), T. quinqueloba, and Globigerinita glutinata, whereas interstadial assemblages were similar to the modern. See the auxiliary material for information on ecological/oceanographic associations of these species.
 The depth-specific marine environment of growth or calcification is key to interpreting isotopic or derived paleotemperature data and assemblages. For this study, we collected stable isotopic data from G. bulloides, N. pachyderma (s), and the benthic species U. peregrina. G. bulloides lives in the surface mixed layer in SBB [Pak and Kennett, 2002]. N. pachyderma (s) is a colder, higher latitude-adapted species. It appears to spend much of its life in the upper 50 m [Kohfeld et al., 1996; Ortiz et al., 1996] and responds to surface water temperature and stratification [Reynolds and Thunell, 1986], but it often adds a thick calcite layer to its shell at depths of up to 200 m [Kohfeld et al., 1996], so its δ18O signal may be biased toward colder temperatures. U. peregrina is infaunal, inhabiting the upper 2 cm of sediment [Corliss, 1991].
 The marine sediments used in this study were part of a suite of thirty-two 3–5 m jumbo piston cores collected in Santa Barbara Basin (34.21°N, 119.67°W) at water depths of 128–191 m along a submerged anticline (the “Mid-Channel Trend,” Figure 1b) aboard the R/V Melville in 2005. The Mid-Channel Trend is tectonically active, and submarine erosion along its crest provided access to pre-modern strata. Detailed analysis of high-resolution multi-channel seismic reflection and single-channel data, along with stratigraphic correlation with well logs from ODP 893 and industry wells, indicates that Quaternary sediments were deposited continuously across the deep paleobasin [Escobedo, 2009; Nicholson et al., 2006].
 A chronostratigraphic framework for deep SBB sediments was constructed by identifying intermediary seismic reflectors between two dated horizons (at ~120 ka and ~1 Ma) [Marshall, 2012]. These intermediary reflectors, inferred to have sequence stratigraphic significance, were correlated across the basin to seafloor outcrop or industry well locations, using multichannel seismic and towed chirp data. Cores were targeted and collected in relation to these outcrops (Figure S1). Up to six cores with known relative stratigraphic positions were recovered between each adjacent seismic reference horizon. Estimated core ages were refined within this stratigraphic framework by oxygen isotope stratigraphy (G. bulloides δ18O) and by the identification of three biostratigraphic datums and a dated ash layer within the cores themselves or in close stratigraphic proximity [Marshall, 2012] (see auxiliary material for detailed stratigraphic information and additional constraints on chronology). Based on the calculated sedimentation rate of 100 ± 20 cm/kyr (discussed below), each core spans approximately 2500–5700 years. The estimated ages of cores MV0508-11JPC, MV0508-16JPC, and MV0508-20JPC fall within MIS 8, MIS 12, and MIS 18, respectively. The presence or absence of sediment laminations was determined by visual observation at <1 cm resolution and recorded shipboard during core collection.
2.2 Stable Isotopic Analyses
 Fifty to seventy-five samples were collected from each of cores MV0508-11JPC, MV0508-16JPC, and MV0508-20JPC. Samples for isotopic analyses of G. bulloides were collected in 1 cm increments every 1–2 cm, and samples for N. pachyderma (s) and U. peregrina were collected in 2 cm increments every 5 cm. Analyses were split between UC Santa Barbara (UCSB) and UC Davis (UCD); see the auxiliary material for details.
 Samples for isotopic analysis consisted of 15–30 G. bulloides or N. pachyderma (s) specimens, or 3–12 U. peregrina specimens, picked from the >150 µm size fraction. Samples were cleaned ultrasonically in methanol for 3 s and roasted under vacuum (at 375°C for half an hour at UCD, and 350°C for 2 h at UCSB). Analyses were performed on a Fisons Optima isotope ratio mass spectrometer (IRMS) at UCD and a Finnigan MAT 251 IRMS at USCB. Both use a common acid bath at 90°C and calculate instrument precision (1σ) from repeated analyses of NBS-19. Precision at UCD is ±0.04‰ for C isotopes and ±0.06‰ for O isotopes; precision at UCSB is ±0.09‰ for both isotopes. All data are reported in standard delta (δ) notation as per mil relative to the Vienna Pee Dee Belmnite standard. See the auxiliary material for discussion of statistical interlab data comparison. For the purpose of discussion, we round δ18O and δ13C values to the nearest 0.05‰.
2.3 Faunal Analyses
 Samples for faunal assemblage analyses were collected in 2 cm increments every 5 cm and were performed on sample splits of the >150 µm size fraction containing >300 planktonic specimens, where possible (only samples with >100 planktonic specimens are plotted). See the auxiliary material for explanation of small gaps in the record. Because construction of the δ18O record only required ~30 G. bulloides and N. pachyderma (s) specimens, it is more complete than the planktonic assemblage record. Species identifications were based on the taxonomy of Kennett and Srinivasan  and Parker . We plot foraminiferal species abundances both in terms of relative contribution to the assemblage (%) and number of specimens per cubic centimeter of core sediment (abundance). N. pachyderma (d) is presented as both abundance and as % N. pachyderma (d) out of all N. pachyderma (d) and (s) specimens, hereafter referred to as “% N. pachyderma dextral.”
2.4 Paleotemperature and δ18O of Seawater
 Following the approach of Bemis et al. , we apply species-specific temperature calibrations to G. bulloides and U. peregrina δ18O data, accounting for ice-volume effects and modern local salinity (described below). We acknowledge that this approach does not quantitatively account for paleosalinity variability, although recent work [Pak et al., 2012] indicates higher salinity during interstadials, resulting in an underestimate of stadial-interstadial paleotemperature shifts by our methods. For G. bulloides, we use the relationship of Mulitza et al. :
where δ18Oc is the δ18O of foraminiferal calcite, and δ18Ow is the δ18O of seawater. This equation is calibrated from 2°C to 25°C with a 1σ error of ±1.24°C, and is statistically indistinguishable from the calibration of Bemis et al.  but has a broader calibration range [Mulitza et al., 2003]. The tight linkage between δ18O changes in G. bulloides and N. pachyderma (s) strengthens the veracity of the G. bulloides paleotemperature signal, although it is possible that salinity variability at the G. bulloides depth habitat (upper 20 m) would also be present in the N. pachyderma (s) habitat (~50 m).
 For U. peregrina, we use the relationship of Shackleton :
which was calibrated from 0.8°C to 7°C. Using Shackleton's original data, we calculated a 1σ error of ±0.8°C. We also applied these paleotemperature calibrations to MIS 1–3 G. bulloides and U. peregrina data from Hendy and Kennett [1999, 2003].
 To calculate δ18Ow during each study period, we referred to sea level records from Bard et al. , Bintanja et al. , and Siddall et al.  and assumed that a 10 m sea level change yields a 0.1‰ change in δ18Ow [Shackleton and Opdyke, 1973]. We approximated the effect of salinity on δ18Ow by applying regionally calibrated δ18Ow:salinity relationships [Bemis et al., 2002; Zahn and Mix, 1991] to modern salinity data from CALCOFI station 80.52 [Lynn et al., 1982]. For details, see the auxiliary material. Recent culturing work has shown that the G. bulloides δ18O:temperature relationship is likely quadratic (Spero, personal communication, 2012), not linear as in the calibration used in this study. This correction would yield warmer temperatures (up to 1°C) at high δ18O values.
 We use an average documented sedimentation rate of 100 ±20 cm/kyr to estimate the duration of events and transitions (see the auxiliary material for details). The terms “stadial” and “interstadial” denote groups of 10 or more data points bracketed by positive or negative shifts in average G. bulloides δ18O. Features common to all three δ18O records include the following: G. bulloides and N. pachyderma (s) δ18O vary in tandem, although G. bulloides values are more variable, especially during interstadials. G. bulloides δ18O is also negatively offset from N. pachyderma (s) δ18O, with a greater offset during interstadials. N. pachyderma (s) δ13C often tracks changes in G. bulloides δ13C and shows similar or slightly more positive values, with a greater offset when G. bulloides δ13C is more negative. Laminated sediments are observed during every interstadial in all three cores and appear/disappear synchronously with interstadial/stadial initiations. All of these features are also observed in the SBB MIS 3 data.
3.1 MIS 1–3 Paleotemperatures
 Paleotemperature calculations performed on previously published MIS 1–3 data [Hendy and Kennett, 1999; 2003] suggest G. bulloides stadial temperatures of 4–9°C (average ~6°C) and interstadial temperatures of 7.5–13.5°C (average ~10.5°C), ±1.25°C (Figure 2), with typical stadial-interstadial shifts of 6°C. Most interstadial initiations were rapid (decadal to centennial), with much of the temperature change occurring in a few decades. Stadial initiations were often more gradual, but occurred on similar timescales (decades to centuries), and were often comprised of an initial rapid cooling followed by more gradual cooling. A few interstadials, such as IS 12, have a more complex structure that is not directly reflected in Greenland ice core records. U. peregrina paleotemperatures (Figure S2 in the auxiliary material) ranged from 2.6°C to 6.1°C, ±0.8°C, with an average of 4.4°C (excluding an unrealistically warm (7°C) point at 45.47 ka). Last Glacial Maximum (LGM) calculations indicate G. bulloides and benthic temperatures of 4.9 ± 1.25°C and 3.7 ± 0.8°C, respectively. Termination 1 was marked by a warming in G. bulloides and U. peregrina of ~8°C and ~2°C, respectively. Mid- to late Holocene calculations indicate that G. bulloides temperatures averaged 13.1°C and ranged from 12.6°C to 13.6°C, ±1.25°C, and benthic temperatures averaged 5.7 ± 0.8°C and ranged from 5.2°C to 5.9°C, excluding two unrealistically cold points (<2°C) at 0.08 and 0.60 ka.
3.2 MIS 8.6-8.5, ~293 ka (MV0508-11JPC)
3.2.1 Stable Isotopic Record
 The δ18O data (Figure 3) exhibit a large, two-step negative shift in all three species over 110 cm in the record (1100 ± 275 years). During the first step at 425–355 cm below seafloor (cmbsf) (700 ± 175 years duration), G. bulloides, N. pachyderma (s), and U. peregrina δ18O decrease by 1.55‰, 0.95‰, and 1.2‰, respectively. From 355–330 cmbsf (250 ± 60 years), benthic δ18O rises by 0.45‰, and planktonic δ18O increases sharply (by 1.1‰ in 10 cm), then stabilizes. The second step comprises a rapid (330–315 cmbsf, 150 ± 40 years) negative shift of 2‰ in G. bulloides and 0.85‰ in N. pachyderma (s) δ18O, to peak negative values of −0.5‰ and 0.65‰. Benthic δ18O decreases by 0.6‰ to a peak negative value of 2.5‰. From 315 cmbsf to the top of the core (3150 ± 800 years), planktonic δ18O undergoes several rapid, high-amplitude stadial-interstadial shifts (up to 1.9‰ in 10 cm in G. bulloides), superimposed on a long-term rise in both planktonic and benthic δ18O (0.9‰ in U. peregrina). Benthic δ18O shows five oscillations of 0.2–0.65‰, which occur in tandem with shifts in planktonic δ18O.
 A large, negative δ13C excursion in all three species at 205–155 cmbsf (500 ± 125 years) is marked by a −4.7‰ shift in planktonic species (to −4.9‰) and a −2.7‰ shift in benthic species (to −3.4‰) (Figure 3). At the onset of the excursion, there is a 6 cm authigenic carbonate layer. There is no synchronous δ18O excursion. Aside from the negative excursion, planktonic δ13C values range from −2.75‰ to 0‰, and benthic values range from −2.5‰ to −0.65‰. G. bulloides δ13C and δ18O vary in tandem and shift in the same direction. U. peregrina δ13C often varies in step with planktonic values.
3.2.2 Paleotemperature Calculations
 Paleotemperature calculations (Figure 3) suggest that the total MIS 8.6-8.5 temperature increase is ~12°C in G. bulloides, from 3.5°C during MIS 8.6 to 15.5°C at the peak of MIS 8.5, ±1.25°C. U. peregrina paleotemperatures undergo a similarly dramatic increase of 4.5°C, from 2.5°C to 7°C, ±0.8°C. Stadial-interstadial shifts in G. bulloides are up to 7°C over 10 cm (100 ± 25 years).
3.2.3 Planktonic Foraminiferal Assemblages
 Total planktonic foraminiferal abundance peaks during MIS 8.6 and during the two warming steps of the MIS 8.6-8.5 transition (Figure 3). The MIS 8.6 peak comprises N. pachyderma (s), T. quinqueloba, and G. bulloides, whereas both interstadial peaks comprise primarily N. pachyderma (d) and G. bulloides. Peaks in abundances of warm forms are of the greatest magnitude observed in all three records during the two warming steps. G. glutinata (not plotted) abundance varies in tandem with that of G. bulloides, with a relative contribution of 3–25%. The upper portion of the core (0–280 cmbsf) exhibits a patchy faunal record in association with glauconitic sediments likely deposited through winnowing and hence of limited value in this study.
3.3 MIS 12, ~450 ka (MV0508-16JPC)
3.3.1 Stable Isotopic Record
 The δ18O data (Figure 4) show three stadial-interstadial shifts of decreasing amplitude, defined by rapid shifts in G. bulloides δ18O from 2.2‰ to 1.2‰ at 279–273 cmbsf (60 ± 15 years), from 2.3‰ to 1.2‰ at 195–192 cm bsf (30 ± 8 years), and from 2‰ to 1.1‰ at 93–87 cmbsf (60 ± 15 years). During the first two interstadial initiations, benthic δ18O values decrease substantially, with a 0.5‰ decrease at 295–255 cmbsf (400 ± 100 years) and a 0.7‰ decrease at 205–175 cmbsf (300 ± 75 years). These negative trends in benthic δ18O begin 15 cm (150 ± 40 years) before each interstadial initiation and continue for an additional 10–20 cm.
 The δ13C data (Figure 4) show a large negative δ13C excursion in all three species. G. bulloides and N. pachyderma (s) shift by −3.1‰ (to −4.1‰) and −1.7‰ (to −2.7‰), respectively, at 125–90 cmbsf. U. peregrina δ13C shifts by −2.25‰ (to −3.25‰) at 130–105 cmbsf. G. bulloides δ13C recovers to more positive values over the upper portion of the record via several rapid oscillations (of up to 2.6‰ in 10 cm). Aside from the large negative shift, G. bulloides δ13C ranges from −0.3‰ to −2.3‰, and correlates positively with G. bulloides δ18O. U. peregrina δ13C ranges from −0.4‰ to −2.2‰ and exhibits four negative shifts, the lower three of which correlate with negative planktonic δ13C shifts.
3.3.2 Paleotemperature Calculations
 Paleotemperature calculations (Figure 4) indicate G. bulloides stadial temperatures of 3.5–9.5°C, ±1.25°C, and interstadial temperatures of 11–13°C, ±1.25°C. The average benthic paleotemperature is ~4.5 ± 0.8°C, but striking episodes of bottom-water warming (by 2–3°C) occurred during the first two interstadial initiations. The apparent temperature inversion at the top of the core is likely an artifact of the linear G. bulloides calibration we used [Mulitza et al., 2003]; recent work (Spero, personal communication, 2012) indicates that such a linear δ18O:temperature relationship likely underestimates temperatures at high δ18O values.
3.3.3 Planktonic Foraminiferal Assemblages
 Abundance of N. pachyderma (s) peaks during every stadial, with greater amplitude peaks during colder stadials (Figure 4). Abundance of G. bulloides and T. quinqueloba follows each other closely, and also peaks during stadials. N. pachyderma (s) dominates every assemblage, comprising 45–85% during both stadials and interstadials. Accordingly, the total planktonic abundance closely follows that of N. pachyderma (s). The % N. pachyderma dextral is low throughout the record (≤25%) and appears not to correlate with stadial-interstadial shifts. Warm forms are scarce throughout the record, and like % N. pachyderma dextral, do not respond to stadial-interstadial shifts. G. glutinata (not plotted) and G. bulloides abundance varies in tandem, with a relative G. glutinata contribution of 2–20%.
3.4 MIS 18, ~735 ka (MV0508-20JPC)
3.4.1 Stable Isotopic Record
 The planktonic δ18O data (Figure 5) exhibit four stadial-interstadial shifts, starting at 355, 245, 125, and 45 cmbsf. G. bulloides δ18O shifts from stadial values of 2.5–1.5‰ to interstadial values of 1.5–0.5‰. U. peregrina δ18O values are relatively stable, with an average value of 3.6‰. However, there is a subtle negative shift of 0.4‰ that begins 5 cm (50 ± 10 years) before the second interstadial initiation.
 The G. bulloides δ13C data (Figure 5) show several high-frequency negative spikes during or close to the first two interstadials, to peak negative values of −4.4‰. Most of these excursions are matched at lower amplitude in the N. pachyderma (s) and U. peregrina data. Aside from the negative excursions, G. bulloides and U. peregrina δ13C vary from −0.3‰ to −2.4‰ and from −0.95‰ to −2.85‰, respectively.
3.4.2 Paleotemperature Calculations
 Calculated paleotemperatures for this interval (Figure 5) indicate stadial G. bulloides temperatures of 4–8°C, ±1.25°C, and interstadial temperatures of 8.5–11.5°C. ±1.25°C. Benthic temperatures remained relatively constant at ~3°C, except for an increase from 2.2°C to 3.7°C just before the second interstadial initiation.
3.4.3 Planktonic Foraminiferal Assemblages
N. pachyderma (s) abundance peaks during every stadial and comprises 60–85% of the stadial assemblage; during interstadials, the N. pachyderma (s) contribution is more varied, ranging from 25% to 80% (Figure 5). Total planktonic abundance closely follows that of N. pachyderma (s). Warm forms are rare throughout the record, and their abundances do not respond to stadial-interstadial shifts. G. bulloides and T. quinqueloba abundance varies in tandem with total abundance, but G. bulloides also peaks during the first two interstadials. The % N. pachyderma dextral varies in tandem with stadial-interstadial shifts, but remains <45% throughout the record. G. glutinata (not plotted) abundance varies in tandem with G. bulloides abundance, with a relative contribution of 0–24%.
4.1 Interpretation of Upwelling/Productivity Proxies and the Effect of Millennial-Scale Changes in Salinity, Sea Level, and [CO32−]
 We interpret peaks in abundance of G. bulloides and/or T. quinqueloba to reflect increased productivity [Kincaid et al., 2000; Ortiz and Mix, 1992; Sautter and Thunell, 1991]; see the auxiliary material for discussion of ecological/oceanographic associations of planktonic foraminifera. Peaks in productivity indicator species that coincide with negative excursions in planktonic δ13C are interpreted to reflect upwelling, due to the lower δ13C of DIC in upwelled waters. However, because increased productivity raises planktonic δ13C, positive excursions in planktonic δ13C do not necessarily imply a lack of upwelling. In general, higher productivity appears to occur under upwelling conditions during some interstadials, and during some stadials at times of high N. pachyderma (s) abundance. Thus, productivity and temperature appear somewhat decoupled.
 Changes in salinity on millennial timescales may have affected measured δ18O and paleotemperature calculations. A G. bulloides Mg/Ca record from the California Current (ODP Site 1017E) shows stadial-interstadial δ18Ow changes during MIS 3 of 0.5–1‰, with interstadials showing more positive δ18Ow values, likely due to increases in salinity [Pak et al., 2012]. This relationship implies that our paleotemperature calculations may actually underestimate planktonic stadial-interstadial temperature changes by 2–3.5°C.
 Changes in global ice volume on millennial timescales may also have affected measured δ18O and paleotemperature calculations. Studies spanning MIS 3 show that sea level may have varied by 20–40 m on millennial timescales [e.g., Siddall et al., 2008]. This would change δ18Ow by ≤0.4‰—much smaller than a typical stadial-interstadial δ18O shift. Additionally, the rate of sea level change would not have been fast enough to contribute appreciably to the very rapid (decadal) δ18O shifts observed in these records. The fastest documented rate of sea level rise is 5 m per century during Meltwater Pulse 1A [Stanford et al., 2006]. Using this maximum possible rate, it would still take 800 years for sea level to rise by 40 m. Consequently, dramatic δ18O decreases at interstadial initiations in this study do not significantly reflect changes in global δ18Ow.
 Changes in carbonate ion concentration ([CO32−]) also affect δ18O and δ13C [Spero and Lea, 1996; Spero et al., 1997]. An upwelling-induced decrease in [CO32-] of 60 µmol/kg (equal to the difference between modern surface waters in the Southern California Bight (Spero, personal communication, 2012) and upwelled waters described by Peeters et al. ) would raise G. bulloides δ18O by 0.3‰ [Spero et al., 1997]. This is equivalent to 1.4°C in apparent cooling, using the calibration of Mulitza et al. . Because our records indicate stronger upwelling during interstadials (see discussion below), our paleotemperature calculations may underestimate planktonic stadial-interstadial temperature changes. Similarly, an upwelling-induced decrease in [CO32−] would raise G. bulloides δ13C by 0.84‰ [Spero et al., 1997], thus partly counteracting lower δ13C of DIC in upwelled waters [as in Peeters et al., 2002]. This implies that negative excursions in G. bulloides δ13C observed in our cores during upwelling may be a lower limit on true shifts in δ13C of DIC.
4.2 MIS 1–3 Paleotemperatures
 Paleotemperature calculations (using raw data from Hendy and Kennett [1999; 2003]) suggest that, although average MIS 3 stadial and interstadial conditions were milder than the LGM or Holocene, the very coldest stadials were similar to the LGM temperature of 4.9°C, and peak interstadial warmth was similar to the average mid- to late Holocene temperature of 13.1°C (Figure S2 in the auxiliary material). The average Holocene temperature is within error of the April (upwelling) temperature of ~12.5°C [Thunell, 1998] and the annual average temperature at 20 m (the depth habitat of G. bulloides) of ~13.5°C [Bemis et al., 2002]. U. peregrina stadial-interstadial temperatures also correspond to the full range of LGM to average mid- to late Holocene temperatures (3.7–5.7°C, ±0.8°C). The average Holocene benthic temperature of 5.7 ± 0.8°C is within error of the modern basin bottom-water temperature of 6.3°C [Lynn et al., 1982].
4.3 MIS 8.6-8.5, ~293 ka (MV0508-11JPC)
 The large, two-step negative shift in planktonic and benthic δ18O (Figure 4) likely reflects the MIS 8.6-8.5 transition, which was marked by a 25 m sea level rise (from 65 to 40, ±12 m below sea level (bsl)) [Bintanja et al., 2005], as recorded in coral reefs [Stirling et al., 2001]. This climate transition has been noted in numerous benthic δ18O records and stacks [e.g., Bassinot et al., 1994; Martinson et al., 1987; Prell et al., 1986] and appears as a reversal to intermediately warm conditions that interrupt the gradual cooling through MIS 8. However, stage 8.5 corresponds to a deglacial-scale vegetation shift on the Iberian Peninsula [Desprat et al., 2009]. In the SBB record, the δ18O shift is also deglacial in scale, with a 3.25‰ and 1.2‰ decrease in G. bulloides and U. peregrina, respectively (Figure 3). The warming spanned 1100 ± 275 years (110 cm), but Holocene-like warmth only lasted a few hundred years. The following increase in both planktonic and benthic δ18O over the upper 300 cm (3000 ± 750 years) of the record (during MIS 8.5) likely indicates both decreasing temperature and increasing ice volume, and an ultimate return to near-glacial conditions. During this time, planktonic δ18O underwent several high-amplitude oscillations of up to 7°C, with interstadial and stadial initiations occurring in decades. MIS 8.5 interstadials were similar in length, character, and magnitude to shorter interstadials during MIS 3 (i.e., IS 3, 6, 9, 13, and 15; Figure 2). Paleotemperature calculations show a total MIS 8.6-8.5 temperature rise of ~12°C in G. bulloides and 4.5°C in U. peregrina. Both shifts were larger than calculated Termination 1 temperature increases of ~8.5°C in G. bulloides and 2°C in U. peregrina.
 A strong positive correlation between planktonic δ18O and δ13C (except during the sharp negative δ13C excursion at 155–205 cmbsf, discussed below), and peaks in G. bulloides abundance during the two warming steps of the MIS 8.6-8.5 transition (Figure 3), implies upwelling during interstadials. MIS 8.6 shows signs of high productivity, with peaks in G. bulloides and T. quinqueloba abundance but relatively higher G. bulloides δ13C. There is also a strong synchronous peak in N. pachyderma (s) abundance, indicating a stronger and/or colder CC. Compared with MIS 3, which showed a G. bulloides δ13C range of −0.25‰ to −1‰ [Kennett et al., 2000], G. bulloides δ13C during MIS 8.6-8.5 was more variable and more negative, with a range of −0.25‰ to −2.75‰. This may indicate more/stronger upwelling during MIS 8.6-8.5. We note that the total number of planktonic foraminifera is approximately four times greater than in any samples from MIS 3 [Hendy and Kennett, 2000], although this difference can be partially explained by lesser compaction of the MIS 3 sediment.
 The magnitude of the sharp negative δ13C excursion in all three species (−4.7‰ in planktonics, −2.7‰ in benthics) is interpreted to reflect the influence of isotopically depleted methane, either as a primary (ocean floor or subsurface) and/or secondary (post-depositional authigenic) signal. This excursion occurs within a laminated interval, reflecting basin anoxia. An authigenic carbonate layer in the middle of the excursion (173–179 cmbsf) provides evidence for anaerobic methane oxidation:
 This reaction raises alkalinity by two moles for every mole of methane oxidized, thus raising the calcite saturation state of ambient seawater and promoting inorganic calcite precipitation. Similar inorganic carbonate layers have been observed at modern methane hydrate vents such as Hydrate Ridge [Torres et al., 2003]. We note that the MIS 8.5 methane release episode, similar to the one in the MIS 12 record (discussed below), occurred when benthic temperatures were warmer and more variable than in the MIS 18 record, which shows no evidence for methane release. Negative δ13C excursions of similar magnitude in benthic and planktonic foraminifera were described in SBB records during MIS 3 interstadials [Kennett et al., 2000].
 The more precise age control of this core allowed for direct comparison of the MIS 8.6-8.5 δ18O data to the synthetic Greenland δ18O reconstruction of [Barker et al., 2011], which was modeled from the EPICA Dome C δD record assuming a bipolar seesaw mechanism (Figure S3). The remarkable correspondence between peaks and troughs of the foraminiferal and synthetic ice core records, despite inequalities in amplitude, argues for a linkage between millennial-scale climate shifts in Greenland and SBB.
4.5 MIS 12, ~450 ka (MV0508-16JPC)
 The MIS 12 interstadials were brief (10–35 cm, 100 ± 25 to 350 ± 90 years), initiated and ended rapidly (3–6 cm, 30 ± 8 to 60 ± 15 years), and were spaced 80–100 cm apart (800–1000, ±250 years), similar to shorter interstadials during MIS 3 (i.e., IS 3, 6, 9, 13, and 15; Figure 2). Calculated G. bulloides paleotemperatures (3.5–9.5°C during stadials and 11–13°C during interstadials, ±1.25°C) were similar to those during MIS 3, but stadial temperatures were more variable. U. peregrina temperatures increased by ~2–3°C just before interstadial initiations, similar to MIS 3 [Hendy and Kennett, 2003].
 Interstadial upwelling is implied by peaks in % G. bulloides that coincide with negative G. bulloides δ13C shifts. However, productivity may have been highest during the two colder stadials, when peak G. bulloides and T. quinqueloba abundance coincided with peak N. pachyderma (s) abundance, and G. bulloides δ13C was relatively higher than during interstadials. Compared to MIS 3 [Kennett et al., 2000], interstadial G. bulloides δ13C was more negative and variable, ranging from −0.25‰ to −2.25‰. Similar to the MIS 8.6-8.5 record, the sharp negative δ13C excursion (−3.1‰ in G. bulloides, −1.7‰ in N. pachyderma (s), and −2.25‰ in U. peregrina) in all three species is too large to be due to changes in carbon export, and thus may reflect the influence of methane.
4.6 MIS 18, ~735 ka (MV0508-20JPC)
 Planktonic δ18O (Figure 5) showed cyclic variations, with returns to similar values during each stadial and interstadial. Interstadials initiated rapidly (in 5–10 cm), lasted 30–40 cm (300–400, ±80 years), and occurred every 80–120 cm (800 ± 200 to1200 ± 300 years), similar to shorter interstadials during MIS 3 (i.e., IS 3, 6, 9, 13, and 15; Figure 2). Stadial initiations (except the first) also occurred in decades, but were of lower magnitude than interstadial initiations, similar to interglacial terminations. Typical stadial-interstadial G. bulloides δ18O shifts were ~1‰. Calculated G. bulloides paleotemperatures (4–8°C during stadials and 8.5–11.5°C during interstadials, ±1.25°C) are similar to those during MIS 3, but lack the peak intervals of warmth of MIS 3. U. peregrina temperatures of 2.5–5.3°C show similar variability as MIS 3, but colder average temperatures (3.3°C, compared to the MIS 3 average of 4.4°C).
 Greater upwelling during the first interstadial and second stadial is evidenced by peaks in G. bulloides and T. quinqueloba abundance and sharp negative excursions in G. bulloides δ13C. The second interstadial and third stadial indicate increased productivity but not upwelling, with peak G. bulloides and/or T. quinqueloba abundance coinciding with relatively positive G. bulloides δ13C. Subsequent stadials and interstadials do not indicate high productivity. G. bulloides δ13C values (−0.25‰ to −4.25‰) are more negative and have a larger range than G. bulloides δ13C values during MIS 3 [Kennett et al., 2000]. This may indicate more or stronger upwelling during MIS 18 than during MIS 3. We note that the total abundance of planktonic foraminifera, especially N. pachyderma (s), is up to 10 times greater than in any MIS 3 samples [Hendy and Kennett, 2000].
4.7 Surface Ocean Circulation During Millennial-Scale Shifts
 One likely mechanism of stadial-interstadial temperature change in SBB is shifts in the relative dominance of the CC versus the Countercurrent. SBB's position at the confluence of these currents makes it a very sensitive indicator of regional oceanographic changes. Planktonic assemblage data show that this mechanism was likely the dominant driver of temperature change during MIS 3 [Hendy and Kennett, 1999; 2000]: Peaks in abundance of N. pachyderma (s), which occurred only during stadials, imply a stronger/colder CC presence in SBB [Ortiz and Mix, 1992]; the opposite pattern was observed with N. pachyderma (d), suggesting a stronger influence from the Countercurrent during interstadials (see the auxiliary material for discussion of ecological/oceanographic associations of planktonic foraminifera). Planktonic assemblage data from this study show the same pattern.
 One notable difference between our records and the MIS 3 archive is that % N. pachyderma dextral is much more muted and less consistently responsive to stadial-interstadial shifts during MIS 18 and MIS 12 than during MIS 3 [Hendy and Kennett, 1999, 2000]. Even during interstadials, % N. pachyderma dextral was <45% during MIS 18 and <25% during MIS 12, compared to 80–90% during the MIS 8.6-8.5 transition and MIS 3 interstadials, even though calculated G. bulloides paleotemperatures were similar. Interestingly, % N. pachyderma (s) during interstadials (often >50%) was also far greater during MIS 18 and MIS 12 than during MIS 3, when interstadial % N. pachyderma (s) dropped to almost zero [Hendy and Kennett, 2000]. A possible explanation is that this species had not yet evolved to occupy its modern ecological niche (see the auxiliary material) [Kucera and Kennett, 2000]. Alternatively, if temperature shifts within the CC (rather than relative dominance of the countercurrent) were responsible for stadial-interstadial temperature change, this could also explain the muted response in N. pachyderma (d) during stadial-interstadial shifts and the lack of warm forms, especially O. universa, during interstadials.
 The common pattern of increased upwelling and productivity during interstadials is similar to that during MIS 3, both in SBB and offshore nearby Point Conception [Hendy and Kennett, 2000; Hendy et al., 2004]. Since upwelling appears more commonly during warm interstadials, the drivers for temperature and upwelling must have differed. If the apparent increase in upwelling stemmed from stronger/more persistent upwelling-favorable (northerly) winds, one would expect such a shift to be accompanied by a more dominant CC and colder temperatures, since northerly winds are the primary control of CC strength on seasonal timescales. Warmer temperatures during increased upwelling imply that nutrient supply may be the dominant driver of the upwelling signal. Nutrients are supplied to upwelled waters in the modern SBB by the California Undercurrent [Hendy et al., 2004], a nutrient-rich water mass at ~300 m depth [Lynn and Simpson, 1990]. Modern studies show that the undercurrent weakens when the CC is at its seasonal strongest [Lynn and Simpson, 1990], so it is reasonable that nutrient supply to upwelled waters may have been reduced during stadials, when the dominance of N. pachyderma (s) suggests a very strong CC influence. MIS 3 records targeting a nearby modern upwelling cell (ODP Hole 1017E, 50 km offshore Point Conception) also show increased undercurrent influence during interstadials relative to stadials, based on sedimentary nitrogen isotopes [Hendy et al., 2004].
 Instances of increased productivity during stadials (shown by peak abundances of G. bulloides, T. quinqueloba and N. pachyderma (s) in conjunction with relatively high G. bulloides δ13C) appear similar to paleoceanographic conditions near Point Conception during the LGM [Hendy et al., 2004]. Higher LGM productivity in the absence of clear upwelling evidence may have been caused by the greater nutrient content of the CC relative to the Countercurrent, or by deepening of the mixed layer by increased storm frequency [Hendy et al., 2004]; greater terrigenous supply of nutrients via increased runoff may also have contributed. The presence of bioturbated sediments even during stadial peaks in productivity implies that organic matter supply did not strongly influence benthic oxygenation.
 In all our records, stadial-interstadial shifts in presence/absence of sediment laminations support the hypothesis of Hendy and Kennett  that the age and/or source of intermediate water in SBB shifted on millennial timescales. Although changes in SBB surface water productivity may have played some role in controlling benthic oxygenation, benthic warmings and/or freshenings just prior to many interstadial initiations (for example, at 295–255 and 205–175 cm bsf in the MIS 12 record, Figure 4) also imply rapid intermediate water response to stadial-interstadial shifts.
 Millennial-scale ocean-climate oscillations are shown here to have occurred in SBB during glacial periods dating to 735 ka. The similarity of millennial-scale variability in all our records implies that it is seemingly unaffected by Late Quaternary shifts in glacial-interglacial dynamics (i.e., the ramp-up of 100 kyr power in global climate records at 700–600 ka [Mudelsee and Schulz, 1997; Ruddiman et al., 1989] and the Mid-Brunhes Event at 430 ka [Jansen et al., 1986]). This conclusion builds on interpretations of lower-resolution early Pleistocene records showing millennial-scale shifts during a time of significantly different glacial-interglacial dynamics [Raymo et al., 1998] and agrees with the findings of Barker et al.  and Loulergue et al,  that millennial-scale variability was present throughout the Late Pleistocene (Figure 2). In addition, our data from MIS 8.5, when sea level was only ~40 m bsl [Bintanja et al., 2005], agree with lower resolution records showing the repeated appearance of MIS 3-like variability at sea levels below 30 m bsl [McManus et al., 1999].
 Our data showcase the benefits of a high-resolution marine record, in that ocean circulation and productivity/upwelling can be interpreted in addition to features of temperature variability, and a detailed depiction of the full amplitude and duration of abrupt climate shifts is possible. The planktonic foraminiferal assemblage and δ13C data show that the mechanism of temperature shifts in SBB (via changes in the California Current or shifts in relative dominance of the countercurrent), as well as associations between productivity/upwelling and stadial-interstadial shifts, may have been similar in SBB throughout the Late Pleistocene. This study provides data from a midlatitude site with sufficient resolution to test and recalibrate proposed climate models, such as that of Barker et al. , that are primarily derived from polar or tropical records.
 We analyzed high-resolution “windows” from SBB sediment records dated to MIS 18 (735 ka, ±5 kyr), MIS 12 (450 ka, ±5 kyr), and the MIS 8.6-8.5 transition (293 ka, ±5 kyr). Our results indicate that millennial-scale variability was present back to MIS 18 and had similar characteristics as during MIS 3, in terms of rapidity of warming and cooling, duration of interstadials, and range of stadial and interstadial temperatures. Stadial G. bulloides δ18O values were 2.75–1.75‰ (average 2.25‰) and interstadial values were 1.5–0.5‰ (average 1.1‰), as in MIS 3. Also similar to MIS 3, interstadial initiations (0.9–1.9‰) occurred in decades; stadial initiations also occurred in decades, but were of slightly smaller magnitude (0.7–1.6‰), since they were often composed of a large initial cooling followed by more gradual cooling. Stadial-interstadial durations were similar to shorter events during MIS 3; our age model suggests that interstadials were 80–500 years in duration and occurred every 700–1500 years.
 Upwelling (indicated by negative shifts in G. bulloides δ13C concurrent with peaks in G. bulloides and T. quinqueloba abundance) and surface water temperature were somewhat decoupled, with greater upwelling during interstadials. Our data support two possible mechanisms for stadial-interstadial temperature change in SBB: (1) relative dominance of the CC versus the countercurrent or (2) temperature shifts within the CC. Our findings imply that millennial-scale shifts were an inherent feature of glacial climates in the Northern Hemisphere throughout the past 735 kyr, and they remained remarkably constant in the details of their amplitude, cyclicity, and temperature variability.
 We wish to thank K. Thompson and D. Winter for sample preparation and analytical assistance. Special thanks go to L. Beaufort, Aix-Marseille Universite, for his important contribution to the age scale by identifying the LO of Pseudoemiliania lacunosa in the sediment sequence. We are grateful for the excellent support provided by the dedicated crew of the R/V Melville, NCOR, USGS, and the scientific party in acquiring the necessary data and cores to conduct this study. Support was provided by the National Science Foundation (OCE-0825322 to TMH, JPK, and RJB, and OCE-0350573 to JPK and CN), with additional support to SMW from the Geological Society of America Graduate Student Grant, Evolving Earth Foundation, UC Davis Cordell Durrell Fund, and Bodega Marine Lab Fellowship. Data from this publication can be found online at the World Data Center for Paleoceanography.