Evolution of marine sedimentation in the Bering Sea since the Pliocene

Arctic and subarctic regions are shaped by processes involving sea ice and continental ice, the formation of dense oxygenated deep waters, and some of the most productive ecosystems in the world, all of which are very sensitive to climate change (Grebmeier et al., 2010; Comiso et al., 2008). Marine sediment, deposited on the seafloors of these regions, derived from riverine, glaciomarine, volcanic sources, and from surface-water biological productivity, records dramatic changes in climate on glacial-interglacial and other scales of climate variability. Based on past studies, it is expected that patterns of glacial-interglacial changes include lower biogenic sedimentation during glacials compared to interglacials (Levitan and Stein, 2008) and sustained perennial or nearly perennial sea ice cover during peak glacials (Takahashi, 2005; Tanaka and Takahashi, 2005; Katsuki and Takahashi, 2005). Because extensive areas of the shelf were subaerially exposed during glaciations, the deposition of sediment moved from the outer shelf edge to the deep basins in turbidite sequences (Ewing et al., 1968; Scholl and Creager, 1973) and to sediment drifts located outside the Bering Sea (e.g., Meiji Drift; VanLaningham et al., 2009). Turbidites in the deep Aleutian and Kamchatka Basins, at the end of the last glacial period, indicate increased input of siliciclastic material derived by melting of Bering Sea ice and the alpine glaciers of Alaska, the Aleutians, and eastern Siberia (Gardner et al., 1982).

Several important sites adjacent to the Bering Sea were explored during Ocean Drilling Program (ODP) Leg 145: Sites 881, 882, 883, and 884 (Rea et al., 1993, 1995). Maslin et al. (1996) observed a dramatic increase in ice-rafted debris (IRD), a decrease in sea-surface temperature (>7.5 °C) and opal mass accumulation rates (MARs) (fivefold decrease), and a decrease in both total organic carbon and CaCO₃ MARs at Site 882 at 2.75 Ma. The observed shift is coeval with the IRD change found in the Norwegian Sea and suggests that the Arctic and northeast Asia were significantly glaciated from 2.75 Ma onward. Provenance studies by McKelvey et al. (1995) and Krissek (1995) suggest that the origin of IRD in the northwest Pacific Ocean and the Gulf of Alaska is the Bering Sea off the Kamchatka Peninsula and southeastern Alaska, respectively.

The sites drilled during Integrated Ocean Drilling Program (IODP) Expedition 323 in the Bering Sea provide, for the first time, a detailed history of marine sedimentation in a subarctic Pacific basin from the earliest Pliocene to recent time (Expedition 323 Scientists, 2010; Fig. 1; Table 1). The Expedition 323 drill sites at Bowers Ridge provide sediment records that span the past 5 m.y., and the sites at the Bering slope provide records spanning the past 2 m.y. (Table 1). Prior to IODP Expedition 323, the only deeply drilled sites were Sites 188 and 185, drilled by Deep Sea Drilling Project (DSDP) Leg 19 (Scholl and Creager, 1973); old drilling technology resulted in poor recovery.

This study presents the result of quantitative sedimentologic analyses of five of the seven sites drilled during Expedition 323: Sites U1340 and U1341 on Bowers Ridge, Site U1339 on the Umnak Plateau, and Sites U1343 and U1344 on the Bering slope (Sites U1339, U13443, and U1344 are herein referred to as the Bering slope sites; Fig. 1; Table 1). The locations of the drill sites and the ages of the sediment recovered during Expedition 323 allow cross-basin comparisons of sedimentary processes and provide a framework to assess variations in the relative contributions of the main sediment sources during major reorganization of Earth’s climate. These include the early Pliocene warm period (ca. 5–3 Ma), with mean global temperature ∼3 °C warmer than today (Haywood and Valdes, 2004), the onset of significant Northern Hemisphere glaciation, which began gradually as early as 3.6 Ma (Mudelsee and Raymo, 2005) and was well developed by 2.6–2.7 Ma in the North Pacific (Krissek, 1995), the gradual cooling of upwelling zones (Brierley et al., 2009) and establishment of strong Walker Circulation in the tropical Pacific by ca. 1.8 Ma (Ravelo et al., 2004), and the transition from 41 k.y. to 100 k.y. glacial-interglacial cycles between 1.2 and 0.6 Ma, referred to as the mid-Pleistocene transition (Clark et al., 2006).

Multiple sediment sources together with an asymmetric geometry of the Bering Sea, characterized by a broad shallow continental shelf, a steep continental slope, and a deep abyssal basin dissected by submerged ridges, account for the general distribution of marine sediment on the modern seafloor of the region. This distribution is mainly terrigenous sediment on the Bering shelf, mixed biogenic and terrigenous sediment on the Bering slope, and mainly biogenic sediment in the abyssal basins and open sea submerged ridges (Fullam et al., 1973). Biogenic productivity in the Bering Sea is the highest in the world (Sigler et al., 2010), and it results in the deposition of vast quantities of siliceous microorganism remains. Although sponge spicules, silicoflagellates, and radiolarians are present, diatom valves compose almost all of the siliceous biogenic fraction (Lisitsyn, 1969; Fullam et al., 1973).

Sediment samples from the top cores of the Expedition 323 sites (Figs. 2A–2C) corroborate past studies and provide additional detail regarding the modern sediment distribution across the Bering Sea. Although the texture of all modern sediment is mainly silt size, and the most abundant sedimentary component is diatom valves, there are significant differences among the sites. Diatom valves are most abundant in the southernmost Bering slope site, U1339, and the least abundant in the northernmost, Site U1344. Clay minerals are generally <20% at all sites and Site U1344 is the only one that shows significant (>10%) amounts of silt- and sand-sized siliciclastic material. Volcaniclastic material is found at all sites, but is most common at Site U1339.

Before examining how sedimentation evolved through time in the Bering Sea, it is important to review the main physical and oceanographic characteristics of this basin and the sources and depositional mechanisms of modern marine sediment in the Bering Sea.

Approximately one-half of the Bering Sea is a shallow (0–200 m) neritic environment, the majority of the continental shelf spanning the eastern side of the basin (Fig. 1). Several bathymetric highs intersect the deepest parts of the marginal basin. The curvilinear Bowers Ridge extends 300 km north from the Aleutian island arc and forms the semienclosed Bowers Basin. The ridge is an extensionally faulted, arc-magmatic construct of early Tertiary age (Cooper et al., 1987a, 1987b). The eastern side, where Site U1340 is located, is steeper and offset by normal faults that create structural basins characterized by thick sedimentary wedges (>1.5 km). The western slope is more gentle, the thickness of the sedimentary sequence increases progressively to the west, and it is generally <1 km thick (Site 1341; Expedition 323 Scientists, 2010).

The Bering slope has the largest known submarine canyons in the world (Scholl et al., 1970; Carlson and Karl, 1984, 1988; Normark and Carlson, 2003). The southernmost of the Bering slope sites, U1339, is situated off Bristol Bay on the northwest flank of a section of the Umnak Plateau that is separated from the main shelf by the Bering Canyon, the longest canyon in the Bering Sea (442 km; Normark and Carlson, 2003). Site U1343 is located on a topographic high, isolated from the slope in front of the Zemchug Canyon, which is one of the largest canyons in the world and, according to Normark and Carlson (2003), was not an important source of sediment during the latest Quaternary. Site U1344, the deepest of all Expedition 323 sites, is on a small interfluve ∼10–15 km southeast of the Pervenets Canyon (Fig. 1). The canyons of the Bering slope are not in direct proximity to large rivers; however, sea-level reconstructions and the detection of buried channels located in the outer Bering shelf suggest that during sea-level lowstands, the Yukon and Kuskokwim Rivers must have meandered across the Bering shelf and, perhaps, influenced the development of the present-day heads of the Bering, Navarin, and Zhemchug Canyons (Carlson and Karl, 1984).

Water mass exchange with the Pacific through the Aleutian Islands, such as through the Kamchatka Strait, is significant, linking Bering Sea conditions to those of the Pacific (Fig. 1). The Alaskan Stream, an extension of the Alaskan Current, flows westward along the Aleutian Islands and enters the Bering Sea mainly through the Near Strait and to some extent through the Amchitka Strait (Stabeno et al., 1999). A part of the Subarctic Current also joins the Alaskan Stream, resulting in a combined volume transport of 11 Sv (Ohtani, 1965).

Currents generally flow northward, across the Bering shelf with a mean velocity of as much as 3 cm/s; the strongest of these, the Anadyr Current, flows toward the Anadyr Strait with velocities of 3–10 cm/s (Kinney et al., 2008). The area of high biological productivity, extending along the shelf break and southward over the Bering slope, is recognized as the Green Belt. The Green Belt is associated with the dynamics of the Bering Slope Current, which originates from the Alaskan Stream water that enters the Bering Sea through the western Aleutian Islands (Fig. 1), and flows eastward along the Aleutian Islands, consequently encountering the Bering shelf, where it is forced to turn to the northwest. Vertical mixing along the shelf break, particularly at slope canyon axes, results from eddies and instabilities in the Bering Slope Current and tidal mixing, and is responsible for the enhanced biological productivity of the Green Belt (Kinney et al., 2008) and consequently high organic carbon accumulation at the seafloor.

Nutrient-rich, oxygen-poor, bottom and intermediate depth water from the North Pacific fills the Bering Sea and is slightly modified by the mixing of relatively fresh, warm water; only very small amounts of bottom water are formed within the Bering Sea today (Warner and Roden, 1995). The oxygen and nutrient composition of Bering Sea subsurface water is further modified by respiration of organic matter in the water column (Nedashkovskiy and Sapozhnikov, 1999) and benthic denitrification (Lehmann et al., 2005). Respiration and the development of an oxygen minimum zone (OMZ) are particularly intense at water depths of ∼1000 m.

Much of the Pacific water entering the Bering Sea is matched by outflow through the Aleutian Islands and unidirectional northward transport of water mass through the Bering Strait, which is ∼50 m deep, to the Arctic Ocean (Stabeno et al., 1999). The northward-flowing water masses transport suspended silt-sized minerals originating from the northern Bering Shelf (McManus and Smyth, 1970) and nutrients into the Arctic Ocean. Restriction of flow through the Bering Strait in past times of lower sea level could have caused changes in global patterns of circulation or in nutrient and salinity distributions (Takahashi, 2005).

During winter months, the Bering Sea is currently partially covered by sea ice extending approximately to the position of the shelf edge (Fig. 1). Models predict nearly sea ice–free summers before the middle of the twenty-first century (Wang and Overland, 2009), and the Bering Sea marine ecosystem is responding to recent trends in summer sea ice retreats (Grebmeier et al., 2010). The coupling of changes in biological productivity and sea ice distribution can be understood by examining the biogenic sediment– and ice-delivered terrigenous grains. Strong winds and intense turbulence in shallow, open water and, most important, subfreezing temperatures, cause the formation of frazil and anchor ice (ice crystals attached to the bottom), and sediment entrainment, or suspension freezing, of particles from the sediment bed or water column to the sea ice (Reimnitz et al., 1998). IRD is generally much finer grained than the source sediment, with silt- and clay-size particles preferentially entrained by frazil ice. In contrast, anchor ice can locally incorporate very high percentages of sand and coarser clasts, which are less sorted. Melting of ice floes during summer also causes sorting and fractionation of grain sizes; melting exposes and concentrates previously dispersed sediment onto the surface of the sea ice, resulting in upward migration of particles as the thickness is maintained by new ice growth at the base in winter. As summer meltwater slowly trickles from low areas on floe surfaces, commonly at their edges, and into the sea, it carries with it the finest particles and leaves behind the coarsest (Reimnitz et al., 1998). In all, these mechanisms result into two levels of sediment sorting and fractionation by sea ice, one at the source during entrainment and one at the delivery during melting of the ice floes.

Three major rivers flow into the Bering Sea: the Kuskokwim and Yukon Rivers drain central Alaska and the Anadyr River drains eastern Siberia (Fig. 1). The Yukon is the longest of the three rivers and supplies the largest discharge into the Bering Sea, with a mean annual flow of 5 × 10³ m³/s, about two-thirds of that of the Columbia River (Hood, 1983). Lisitsyn (1969) recognized that the terrigenous component being deposited in the basin is generated in high-latitude areas largely underlain by permafrost. The shelf probably traps much of the coarse debris carried by these rivers; however, significant volumes of silt- and clay-sized debris are transported and eventually reach the deep basins (Fullam et al., 1973). Angular, silt-sized suspended minerals derived at least in part from the Yukon River have been found in water samples collected throughout the northern Bering Sea continental shelf, suggesting that turbid water could be an important transport mechanism of silt during the ice-free seasons, and even more significant during storms and ice breakup (McManus and Smyth, 1970). Bulk Ar-Nd isotopic measurements of sediment from the Meiji Drift, a large sediment body just outside the region southeast of the Kamchatka Strait (Fig. 1; VanLaningham et al., 2009), indicate that the terrigenous sediment input to the drift can be described by two end members, a Kamchatkan-Aleutian source during interglacial intervals and a Bering Sea source, heavily influenced by the Yukon River, during glacial times.

The Aleutian Ridge, a volcanic-plutonic arc, is the source for the majority of the abundant pyroclastic material in the sediment of the Bering Sea region (Fullam et al., 1973). Marine and subaerial erosion of the ridge crest provides much coarse clastic debris shed both to the north and the south.

The sediment data produced for this work are mainly of two types, sediment fabric and lithology. The main analytical methods used included laser particle size analysis, smear slide microscopy, and scanning electron microscopy (SEM). The combination of these three methods is a very effective way to both identify and quantify the sedimentary components in very fine sediments such as those that characterize the Bering Sea. While particle size analysis was performed on the entire suite of samples from the five sites (Table 1), smear slide and SEM were conducted only on a fraction of the total number of samples. To provide additional details regarding the components that make up each size class, a smaller subset of samples was selected and analyzed with both laser particle sizer and optical microscopy after removing the organic material, the biosilica, and after treating the sample with sodium metaphosphate (see following).

Because terms referring to sediment grain size such as clay, silt, and sand might be confused with similar lithologic terms, we differentiate clearly between composition and fabric: clay minerals and sand- or silt-size siliciclastics are lithologic terms, whereas clay, silt, and sand size refer only to the sediment fabric. For example, the majority of the sand-size particles in the Bering Sea sediment are diatom valves and not siliciclastic sands.

Samples were taken every ∼100 cm from the entire cored sequences at IODP Sites U1339, U1340, U1341, U1343, and U1344 (Tables 1 and 2). Particle size distributions in percent volume were analyzed using a Beckman-Coulter LS 13 320 laser particle size analyzer with an aqueous module equipped with a pump and a built-in ultrasound unit. This module analyzes small (∼1 g) amounts of sediment and the measured size distributions (0.04 μm to 2 mm) by combining conventional laser beam diffraction with polarized intensity differential scatter. Data interpolation and statistical analyses were obtained with the laser particle sizer proprietary software (Beckman Coulter Inc., 2003). Because all samples analyzed tend to log-normal grain-size distributions in the 0.04 μm to 2 mm spectrum, geometric rather than arithmetic statistics were applied to the values obtained by the logarithmically spaced size channels of the particle sizer using the method of moments (e.g., Folk, 1966). One potential bias of this method (as well as during smear slide preparation; see following) is that subsampling with a toothpick or other sampling devices might exclude very coarse particles (e.g., IRD). Repeatability of the grain size measurement was tested by subsampling and repeating the analysis multiple times (coefficient of variation <5%; for further details about the methods, see Aiello and Kellett, 2006).

To test potential flocculation of clay-size particles in the aqueous module, selected samples were treated with 4 g of sodium metaphosphate [Na(PO₃)⁶] and then analyzed. These deflocculation experiments show virtually no difference in the percent clay-size particle (<4 μm) or in the mean size between the untreated and treated samples (Fig. 3A), suggesting that flocculation of clay-size particles does not occur in the liquid module during laser particle size analysis.

Biosilica digestion was used to assess the contribution of the biogenic silica component (mainly diatom valves) in determining particle size characteristics such as mean size and standard deviation (see McCave et al., 1995). Wet alkaline leaching is the most common method used to extract biogenic silica in marine sediment samples, and relies on the fact that biogenic silica and aluminosilicates have different dissolution rates. However, it is difficult to ensure complete extraction of silica from the biogenic components without extraction from aluminosilicates. Furthermore, sediment varying in age, composition, and presence of resistant radiolarians and sponge spicules requires different strengths of leaching solution, making standardization of laboratory procedure difficult. The biogenic silica extraction method used in this study was modified from Mortlock and Froelich (1989) to allow a complete removal of the biosilica fraction, taking in account issues concerning the sample size and the strength of silica digestion. Aliquots of 2 g of sediment were placed into a 50 mL polypropylene centrifuge tube. Organic matter was removed with 5 mL 10% H₂O₂, and removal of carbonates was done with 1 molar HCL. Alkaline leaching was produced using 40 mL 1 molar NaOH in an 85 °C water bath and supernatant was removed after centrifuging at 2 and 4 h.

Smear slides were examined with a transmitted-light petrographic microscope equipped with a standard eyepiece micrometer. Biogenic and mineral components were identified and their percentage abundances were visually estimated under a petrographic microscope (methods in Rothwell, 1989). The potential errors associated with this visual method of analysis include biases during sampling with a toothpick for smear slide preparation and dishomogeneous distribution of the sediment particle on the smear slides. To improve reproducibility, 10 counts were done on different parts of a smear slide using a random walk and the average value of the 10 counts was used. Fields of view having too little (<10% cover) or too much (>30% cover) material were not measured, and different operators were alternating at the microscope. For linear regression analysis (Figs 4A, 4B, 4C) the original percent data were transformed to fit the assumption of the statistical model by calculating the arcsin of the square root.

SEM samples were prepared by placing <1 g of wet sediment on a pedestal and drying in a fume hood. The instrument used for the analysis is a Hitachi S-3400N-II variable pressure SEM.

Natural gamma ray (NGR) is a measurement of the gamma rays produced by the decay of uranium, thorium, and potassium isotopes in mineral particles in the cores. All core sections were analyzed by the natural gamma ray logger onboard (Expedition 323 Scientists, 2010). Age models for each site are derived from a combination of biostratigraphic information and paleomagnetic data used to identify magnetic reversals (Expedition 323 Scientists, 2010). For two of these sites, biostratigraphic age models have been refined by Teraishi at al. (2012) for Site U1341.

The cores recovered during Expedition 323 in the Bering Sea show prominent variability of sediment lithology and fabric with depth, suggesting that both the sediment sources and the mechanisms that controlled their delivery have changed during the past 5 m.y. in response to sea level and climate change. One of the most striking characteristics of the sediment in the Bering Sea is that the integrity of diatom valves is highly variable, occurring at scales ranging from millimeters (millennial), to tens of meters (glacial-interglacial cycles), to longer-term trends.

Two main sedimentary components combine to form the majority of the Bering Sea record, diatom valves and siliciclastic particles; the latter are mainly clay size and secondarily very fine silt size. Secondary components, which in some intervals make up the majority of the sediment, include volcaniclastic (Fig. 5A), large isolated clasts (dropstones) (Fig. 6B), silt- and sand-sized siliciclastic particles occurring both scattered and in discrete layers (Fig. 5C), and biogenic tests other than diatoms, including coccoliths, foraminifers, radiolarians, and sponge spicules. Authigenic minerals including carbonates and pyrite are also present, but are not discussed in this paper. Sedimentary structures (lamination, bedding, and trace fossils; Figs. 5A, 5B) together with variations in sediment characteristics (grain size, color reflectance) and physical properties (NGR, bulk density) are also helpful in reconstructing past depositional environments in the Bering Sea (Figs. 7 and 8; Table 2).

Smear slide counts indicate a significant negative correlation (R² = 0.67; n = 144) between diatoms and clay minerals (Fig. 4A; sponge spicules are included with diatoms because they are a significant component of the biosiliceous fractions at Sites U1340 and U1341). Silt- and sand-sized siliciclastic material, the third most common component in the Bering Sea sediment, is negatively correlated to diatoms (R² = 0.50; Fig. 4B) and shows no correlation with clay (R² = 0.08; Fig. 4C).

Classifications of deep-sea sediments range from those that are largely genetic to those that are largely descriptive, and there is no single classification that takes into account both genesis and descriptive properties for all kinds of deep-sea sediments (Boggs, 2000). For the Bering Sea sites we used a simple classification that accounts for the mainly bimodal (diatom valve and siliciclastic particle) composition of the sediment: diatom ooze, pelagic mud, diatom mud (e.g., see Figs. 9, 10, and 11). The three main sediment types are characterized as follows.

Diatom ooze is composed of generally preserved valves of centric and pennate diatoms (>∼50%) and secondarily by clay minerals (<∼20%), silt- and sand-sized siliciclastic and/or volcaniclastic particles (usually <10%), and minor (<10%) amounts of other biogenics, including nannofossils and foraminifers and trace (<5%) amounts of silicoflagellates, sponge spicules, and opaque minerals. Diatom ooze is the coarsest grained of the three main lithologies, having abundant sand- and silt-size particles predominantly composed of whole diatom valves and valve fragments (Figs. 9B, 9D, 9H, 9J). Diatom ooze is typically associated with interstadial and interglacial intervals characterized by relatively low NGR and low bulk density. Diatom oozes are generally dark green to dark gray and yellow and have relatively high values of the color reflectance (CR) parameter b* (Expedition 323 Scientists, 2010).

Pelagic mud is composed mainly of clay minerals (>∼30%), fine-grained, clay- to silt-size biogenic material (<∼20%), and silt- and sand-size siliciclastics (quartz, feldspar, and rock fragments) and/or volcaniclastics (<∼20%) (e.g., Figs. 9A, 9C, 9E, 9G, 9I). Depending on the proportion of clay minerals relative to the biogenic grains, the color varies from greenish-gray to dark greenish-gray. Pelagic mud is mainly found in the stadial and glacial intervals of the Bering slope sites; it has much finer grain size than the diatom ooze and higher values of NGR, bulk density, and lower CR parameter b* (Expedition 323 Scientists, 2010).

Diatom mud is an intermediate composition between diatom ooze and pelagic mud, composed of subequal proportions of diatom valves and clay (e.g., Fig. 9F). Other secondary components can also be a significant proportion of the sediment, in particular silt- and sand-size siliciclastic and volcaniclastic grains and other biogenic tests. This lithology is typical of sediment having intermediate NGR and sediment particles are mainly clay or silt size.

The correlation between percent biosiliceous tests (determined through smear slide analysis) and mean grain size shown in Figure 4D indicates that ∼40% of grain size variability can be explained by variation in abundance and preservation of diatom valves and secondarily, sponge spicules. To test whether diatom content is the main parameter controlling the mean size, particle size analysis was carried out on a subset of 30 samples from which organic matter, biocarbonate, and biosilica were removed (see Methods discussion). The results of this experiment support our interpretation (Fig. 3B): while the correlation between diatom counts and mean size for the bulk untreated samples is significant (R² = 0.45), the correlation between diatom counts and the mean size for bulk samples after removal of biosilica is not as strong (R² = 0.13).

Virtually all sediment samples of the Bering Sea record can be texturally defined as silt. The average mean size of the Bering slope sites is very fine silt, but the sediment at Bowers Ridge is generally coarser; the average mean size is fine silt and sand-size particles are more common (Figs. 7 and 8; Table 2). Clay-size particles include mainly clay minerals and secondarily very small fragments of biosilica tests; silt-size particles include mainly diatom valve fragments, pennate diatoms, small whole diatom valves, and secondarily siliciclastic and volcaniclastic particles and other biogenic tests. The sand fraction is mainly composed of whole centric diatom valves and secondarily siliciclastic, volcaniclastic particles and other biogenic material including foraminifers, sponge spicules, and fecal pellets. Therefore, Bowers Ridge sediments are generally coarser than the Bering slope sediment because they have more whole diatom valves and fewer clay minerals. Among the Bering slope sites, there is a northward trend of decreasing mean size, decreasing sand, and increasing amounts of clay-size particles (Figs. 7 and 3C).

The sedimentologic differences between the Bering sites are highlighted by the bivariate plot (Figs. 3C, 3D) of mean grain size versus standard deviation (S.D.; the higher the S.D. the less sorting). Bivariate plots and grain size statistics can be used to interpret depositional mechanisms and sediment sources. For example, the correlation between mean grain size and sorting is negative in sea ice–delivered IRD from the Beaufort Gyre and positive in adjacent shelf sediment (Reimnitz et al., 1998). This is because in reworked shelf sediment mean grain size and sorting are hydraulically controlled, leading to a positive correlation between mean particle size and sorting indicative of high-energy conditions (e.g., bottom currents). In the Bering Sea slope sites the grain size data of pelagic mud and diatom mud samples (samples having >30% clay-size content) show a weak negative correlation between mean size and sorting (linear regression line in Fig. 3C; R² = 0.35). The northernmost of the three sites, Site U1344, has the finest and best-sorted sediment, while the southernmost Site U1339 has the coarsest and least-sorted sediment. Unlike samples from the slope sites, samples from Bowers Ridge sites plot in a different part of the diagram and show no correlation between mean size and sorting (Fig. 3D).

Sites U1344, U1343, and U1339 provide records of sedimentation on the Bering slope that span the past 1.9 m.y., 2.4 m.y., and 0.8 m.y, respectively (Figs. 1 and 7; Table 1). The southernmost of the sites, U1339, influenced by the relatively warm Alaskan Stream that enters the Bering Sea through the Unimak and Amukta Passes and inhibits sea ice formation, has an average sedimentation rate of ∼30 cm/k.y. (Expedition 323 Scientists, 2010). Sites U1343 (sedimentation rate of ∼30 cm/k.y.) and U1344 (sedimentation rate of ∼45 cm/k.y.) are located farther north, closer to the modern seasonal sea ice limit and beneath the high-productivity Green Belt.

Prominent bedding alternations between greenish-gray pelagic mud and coarser-grained olive-gray diatom mud (or diatom ooze) beds characterize both Sites U1339 and U1343. The pelagic muds of the glacial intervals have poorly preserved diatom valves; pennate diatoms are rare and centric diatoms occur mainly in fragments of fine silt and smaller size (Figs. 9A, 9C, 9I). Conversely, in diatom muds and diatom oozes deposited during the Holocene and interglacials, centric diatoms are more abundant and better preserved, and pennate diatoms are more common (Figs. 2A–2C, 9B, 9D, and 9J). Well-preserved laminae occur in 6 distinct intervals representing the past ∼500 k.y. at both Sites U1339 and U1343. At Site U1339 the laminated intervals always occur in diatom ooze beds, suggesting a correlation between an increase in primary productivity and expansion of the OMZ (Fig. 5A). Only one laminated interval is present at Site U1344, which is located below the modern OMZ (Fig. 5B).

Volcaniclastics, a minor component of the sediment, decrease in abundance from the southern to the northern sites as a consequence of the increasing distance from the volcanic sources mainly located in the Aleutian Arc. Several prominent volcaniclastic layers, some of which are characterized by graded bedding, have either parallel or undulating sharp (unbioturbated) bottom boundaries, and gradual top boundaries, probably due to local bioturbation (Wetzel, 2009). Another overall minor component of the sediment, silt- and sand-sized siliciclastic material, is most abundant in the northernmost sites U1343 and U1344 and is found either scattered in the sediment or concentrated in discrete layers (Fig. 5C). Silt- and sand-sized grains are most commonly composed of quartz and feldspar, although mica and rock fragments (mainly polycrystalline quartz) are also present. Dropstones composed of gravel- to pebble-sized, rounded to angular isolated clasts are common throughout the three slope sites.

Of the three Bering slope sites, Site U1343 shows the most distinctive variations in grain size with depth (Fig. 7). The interval between ∼0.6 and 1.1 m.y., approximately the mid-Pleistocene transition, has the lowest content of sand-size particles and the highest of silt-size material (Fig. 9E). In the interval deposited prior to the mid-Pleistocene transition, grain size and NGR vary in cyles with periods of ∼14 m, equivalent to ∼40 k.y. Sediment deposited during peak interglacials is mainly diatom ooze (Fig. 9H) and has lower NGR, larger mean particle sizes, and higher content of sand-size particles (i.e., more whole centric diatom valves) compared to the pelagic muds deposited during glacials (Fig. 9G). Diatom mud with average NGR and mean size values (Fig. 9F) is the dominant sediment type deposited between glacial and interglacial peaks.

Site U1340, on the southeastern flank of the central Bowers Ridge, is at a water depth within the modern OMZ. Site U1341, on the western flank of the ridge, is at a water depth just below the OMZ (Fig. 1). The ages of the bottoms of Sites U1340 and U1341 are similar (ca. 5 Ma and ca. 4.3 Ma, respectively; Table 1). Diatom ooze is the dominant lithology at both Bowers Ridge sites, with lesser amounts of diatom mud and rare pelagic mud.

The early Pliocene sediment is dominantly coarse grained, and the coarsest intervals are characterized by having larger proportions of whole centric valves and spicules (Figs. 8, 10I, 11B, and 11C). Conversely, the fine-grained intervals are characterized by broken centric diatoms (Fig. 11D). At Site U1340, the oldest interval in the early Pliocene sediment includes a massive, 10-m-thick, incoherent black sand layer mainly made of moderately sorted angular volcanic clasts and sponge spicules, including reef-forming sponges suggesting resedimentation from shallower water seafloor areas, probably from the axis of Bowers Ridge (ca. 4.1 Ma; black arrow in Fig. 8; Figs. 6C and 10H).

In the late Pliocene record, the mean grain size becomes finer and the sediment is mainly highly fragmented diatom ooze (Figs. 8 and 10G). At Site U1341, the decrease in mean grain size reflects a gradual decrease in valve preservation (Fig. 11A) and in the abundance of sponge spicules between ca. 2.8 and ca. 2 Ma. NGR increases over the same interval, but the proportion of clay-size particles apparently does not (Fig. 8). Similarly, at Site U1340 mean grain size decreases ca. 2.6 Ma, but the transition seems more abrupt owing to a dramatic increase in fragmentation of diatom valves and a drop in abundance of sponge spicules over an apparently shorter stratigraphic interval; however the record is discontinuous due to the presence of numerous drilling gaps. At Site U1340, the fine-grained interval deposited after ca. 2.6 Ma includes sand-size volcaniclastic material, gravel, and isolated pebbles either scattered or concentrated in layers, including one layer ca. 2.3 Ma (black arrow in Fig. 8).

The overall increase in abundance of whole diatom valves throughout the early Pleistocene accounts for the increase in sediment mean grain size and the increase of sand-size particles (Fig. 8). The Pleistocene record includes alternations between (1) coarser sediment having more preserved centric diatom valves, fewer diatom fragments found in the clay-size fraction, and lower NGR, and (2) finer sediment having fewer preserved centric diatom valves, more diatom fragments found in the clay-size fraction, and higher NGR (cf. Figs. 10E, 10F). At Site U1340, the most prominent of these alternations has a period of ∼8 m, or ∼40 k.y.

Laminated diatom oozes are present mainly in the upper ∼1.8 m.y. of Site U1340 (Figs. 10A, 10D) and in the upper ∼0.2 m.y. of Site U1341. Faint centimeter-scale bands and millimeter-scale laminations have been observed in the oldest intervals of Bowers Ridge sites, where diatom oozes are also stiffer because of compaction. Individual laminae are differentiated mainly by changes in the type of biogenic material (e.g., monospecific diatom valves, sponge spicules, silicoflagellates), although laminae composed of volcanic ash are also present. Evidence of sediment delivery by icebergs including pebble or coarser sized dropstones is mainly concentrated in the last ∼1 m.y. at Site U1340 and in the last ∼2.7 m.y. at Site U1341 (Expedition 323 Scientists, 2010).

The most remarkable sedimentary structure in the Pleistocene record at Bowers Ridge is an ∼80-m-thick interval of soft sediment deformation in the upper part of Site U1340 that produces tilting and chevron folding of laminated diatom ooze and diatom mud beds (Fig. 6A). Above the deformed intervals there are prominent alternations between well-laminated green to yellow diatom oozes and homogeneous gray pelagic muds. Several laminated intervals (e.g., Figs. 10A–10D) show sharp bottoms and gradational and/or bioturbated tops. The laminated intervals have coarser grain sizes (sand content is >50%; Fig. 10C), while clay is >30% in the nonlaminated intervals (Figs. 10A, 10D). Laminations contain abundant whole diatom valves (Fig. 10C), and fragmentation of valves and clay content increase at the top of the intervals (Fig. 10B).

Lithologic variations in sediment recovered by IODP Expedition 323 can be interpreted within the context of the climatic history of the Bering Sea region over the past 5 m.y. The dominance of three sediment types, diatom ooze, diatom mud, and pelagic mud, is a result of a persistent supply of biogenic opal (diatom valves) and fine siliciclastic material from land. Variations in relative proportions of diatoms and siliciclastics and in grain size distributions provide insight into the climatic and oceanic evolution and variability of the Bering Sea region. Specifically, the history of productivity can be inferred by combining observations of grain size and diatom valve abundance and preservation.

The grain size of the Bering Sea sediment is strongly related to diatom abundance and diatom valve preservation: ∼40% of grain size variability is correlated to diatom content (Fig. 4D). The most important mineralogic component of diatom valves is biosilica, composed of opal-A, a very unstable silica compound that during diagenesis transforms into more stable opal-CT and quartz (Kastner et al., 1977; Hein et al., 1978). Siliceous biota such as diatoms, radiolarians, silicoflagellates, and siliceous sponges use dissolved silica in the formation of skeletal tests. Dissolved silica enters the ocean through riverine flux, glacial weathering (DeMaster, 1981), and sediment pore-water dissolution (Fanning and Pilson, 1974). Biogenic silica is unstable due to undersaturation in the water column (Gardner et. al 1997). Siliceous biota remove nearly all dissolved silica from oceanic surface waters; following death, the siliceous tests are dissolved primarily through the water column and in sediment (Heath, 1974), enriching deeper waters with dissolved silica (DeMaster, 1981). Only 2%–4% of biogenic silica produced globally is preserved in ocean sediment (Heath, 1974): earlier studies (e.g., Calvert, 1974) determined that the highest concentrations of biogenic silica in surface sediment across the world’s ocean are found where surface waters have the highest rates of dissolved silicon uptake; that is, diatom oozes are preserved on the seafloor only when there is high productivity at the surface, mainly in upwelling regions. Furthermore, changes in accumulation rates do not always coincide with paleorecords of other biogenic constituents (Lyle et al., 1988). However, other studies have shown a close link between biogenic silica in surface sediment and biosiliceous productivity patterns in overlying waters (e.g., Mortlock and Froelich, 1989; Koning et al., 2002). Further studies have linked downcore traces of biogenic silica to local paleoproductivity records with Milankovitch-like cycles (Mortlock and Froelich, 1989; Pisias, 1976; Pisias and Leinen, 1984).

High diatom productivity, which promotes biosilica preservation, is typical of the modern Bering Sea, particularly along the Green Belt through the summer, and at the edge of the sea ice in early spring during melting. This high productivity is reflected in sediment from core tops, Holocene sediment, and samples from the last interglacial, which are characterized by well-preserved diatom valves (e.g., diatom ooze; Figs. 2, 9B, 9D, and 9J). The mean grain size of these samples is >∼10 μm because of the relatively high abundances of well-preserved whole medium-silt-sized diatom valves. Conversely, sediment from the Last Glacial Maximum has poorly preserved (fragmented) diatoms and abundant clay-size particles, suggesting conditions of low biosilica flux due to reduced primary productivity during glacials (e.g., pelagic mud; Figs. 9A, 9C, 9I). The relationship between mean grain size and diatom abundance and/or preservation is observed throughout the Bering Sea record, including in sediment deposited prior to Northern Hemisphere glaciation (Figs. 10I and 11B–11D) and during Northern Hemisphere glaciation (Figs. 10G and 11A). Thus, the record of sediment grain size in the Bering Sea is a rough indication of the trends and variability of diatom productivity over the past 5 m.y.

The linear regression analysis obtained for a subset of samples rich in clay-size sediment (>30%) from the Bering slope sites shows that there is a negative correlation between sorting (S.D.) and mean grain size (Fig. 3C). This relationship is similar to what has been observed for siliciclastic sediment delivered by modern sea ice (e.g., Reimnitz et al., 1998). Figure 3C also shows a latitudinal distribution of sorting and grain size: of the three sites, the northernmost Site U1344 (most affected by modern sea ice) has the finest and best-sorted sediment, while the southernmost Site U1339 (most affected by the warm Alaskan Stream) has the coarsest and least-sorted sediment. Together, these observations suggest that clay mineral delivery in the Bering slope could have been originated at least in part from melting of sea ice. Moreover, the latitudinal trend of increasing clay and decreasing diatom content found at the three Bering slope sites suggests that, as in the modern Bering Sea, sea ice was present more extensively in the northern sites. At Site U1344, the deposition of pelagic mud could reflect more extended periods of sea ice cover. At Site U1339, the alternations between diatom mud and diatom ooze beds could record the activity occurring near the edge of sea ice, with primary productivity fueled during sea ice retreat and melting.

Smear slide counts indicate that there is no obvious correlation between clay mineral content and coarser (silt and sand size) siliciclastic content (Fig. 4C) and that the negative correlation between diatoms and coarser siliciclastic material is significant (Fig. 4B). These relationships indicate that the delivery of coarser siliciclastic particles occurred mainly during glacials (when diatom fluxes to the sediment were lower), and suggested a decoupling between the sources of clay minerals and silt- to sand-size siliciclastics; the latter could have derived mainly from resedimentation processes from the continental margin, including possibly iceberg rafting, rather than from sea ice.

The Bering Sea sediment record can be explained in the context of regional and global climate changes through the Pliocene and Pleistocene Epochs. In the early Pliocene warm period, prior to the ca. 3.6 Ma increase in global ice volume documented in the benthic δ¹⁸O record (Mudelsee and Raymo, 2005), the sediments at Bowers Ridge were characterized by coarser grain sizes and higher sand percent (Figs. 8 and 12). While some of this can be attributed to sand-sized particles of volcanic material and fragments of sponge spicules, particularly at U1340, the dominance of silt- and sand-sized particles in the sediment is primarily due to excellent preservation of whole diatom valves (e.g., Figs. 10I and 11C). During the early Pliocene warm period, the diatom assemblages included 20%–30% heavily silicified forms compared to <5% in sediment of late Pleistocene age (Expedition 323 Scientists, 2010). These heavily silicified forms are indicative of a stronger supply of silica to surface waters in the early Pliocene compared to today. In all, the diatom assemblages, grain size data, and preservational state of the diatoms found in early Pliocene sediment at Bowers Ridge indicate that there was a plentiful supply of nutrients, and relatively high diatom productivity and biogenic opal flux to the sediment, in the warm period prior to the widespread expansion of Northern Hemisphere ice sheets.

High productivity in the early Pliocene warm period at Bowers Ridge cannot be explained with a canonial understanding of the modern-day Green Belt, which is related to the sea ice margin retreat in the spring, and tidal forces and eddies that enhance vertical mixing of nutrients at the shelf break through the entire summer (Springer et al., 1996, Mizobata and Saitoh, 2004). There is no evidence for sea ice diatom or dinoflagellate forms in the early Pliocene (Expedition 323 Scientists, 2010), and Bowers Ridge is well south of the Bering shelf break. One possibility is that vertical mixing of nutrients in the central Bering Sea was enhanced due to stronger upwelling-favorable wind stress curl and/or to less stratified surface conditions, so that the buoyancy forcing was easily overcome by wind forcing. Another possibility, supported by the evidence of resedimentation at the bottom of Site U1340, is that Bowers Ridge was shallow enough that wind and tidal forces, acting on the topography of the ridge, or a so-called island effect, could have forced vertical mixing. Notably, high nutrient availability and productivity during the early Pliocene warm period has also been documented in the northwest Pacific at Ocean Drilling Program (ODP) Site 882, just south of the Bering Sea, and attributed to the weaker stratification compared to the modern ocean (Haug et al., 2005). Thus, in the context of results from ODP Site 882, the high nutrient supply to surface waters in the Bering Sea was likely a result of decreased stratification, rather than local interactions between Bowers Ridge and wind and/or tide forcing. However, future work should include a reconstruction of the subsidence history of Bowers Ridge and the analyses of paleoceanographic proxies to assess surface-water stratification.

The transition from the early warm Pliocene period to the ice ages is marked by an increase in global ice volume between 3.6 and 2.5 Ma (Mudelsee and Raymo, 2005), with some high-latitude regions showing dramatic changes in oceanographic conditions ca. 2.7 Ma (e.g., Haug et al., 2005), marking the expansion of ice sheets to temperate latitudes during the peaks of glacial cycles. In Bowers Ridge sites, an interval of fine-grained, fragmented valves and increasing NGR is found approximately between 2.6 and 2.0 Ma (Fig. 8). The decrease in mean grain size is due to poor preservation and fragmentation of diatom valves (Figs. 10G and 11A) and a slight increase in the volume of the clay fraction indicative of a decrease in diatom production and biogenic opal flux to the sediment. The increase in NGR at Site U1341 (indicative of clay mineral content; Fig. 8) combined with an increase in sea ice diatom and dinoflagellate forms (Expedition 323 Scientists, 2010) suggest that sea ice drift appeared, at least intermittently, for the first time in the open Bering Sea when global climate cooled during Northern Hemisphere glaciation.

Although diatom preservation, and presumably biogenic opal flux, appears to have decreased during Northern Hemisphere glaciation, the Bering Sea records are decidedly different than the ODP Site 882 northwest Pacific record. Specifically, at ODP Site 882, the opal MAR dramatically decreased ca. 2.7 Ma; diatoms became a minor component of the sediment after 2.7 Ma and until recent times (Haug et al., 2005). In the Bering Sea, although there was a decrease in diatom preservation, as indicated by mean grain size, diatoms continue to be the dominant component of the sediment; diatom ooze and diatom mud was deposited before, during, and after Northern Hemisphere glaciation. This suggests that the diatom MAR “crash” in the North Pacific, interpreted as indicative of an increase in high-latitude stratification (Haug et al., 2005), may have been limited to certain regions within the North Pacific. Increased stratification in the Bering Sea may explain a reduction in diatom flux and preservation state, but diatoms are a major component of the entire sedimentary sequence at Bower Ridge. This is evidence of continued, albeit variable, biogenic opal productivity throughout the past 5 m.y.

In the Bering Sea, interglacials were characterized by higher levels of primary productivity, which lead to the deposition of relatively coarse grained diatom oozes with well-preserved mainly centric diatoms and low NGR (e.g., Fig. 10E). Glacial sediment is composed of high-NGR diatom mud with poorly preserved diatom valves indicating low productivity (e.g., Fig. 10F). It may be that sea ice distribution was extensive during glacials, as is known for the Last Glacial Maximum (Tanaka and Takahashi, 2005), resulting in relatively low rates of productivity and biogenic opal flux to the sediment. There is also obvious bedding, such as that found in the upper sections of U1340 on Bowers Ridge, in some cases alternating between laminated and massive sediment (Figs. 10A–10D). Many of the laminated diatom ooze beds have a sharp bottom, while the top shows a gradual increase in clay content, deterioration in diatom valve preservation, and progressive increase in bioturbation. It is possible that these alternations could record millennial-scale variability. The relatively high clay content in this section of the record also suggests that the diatom mud could record colder periods of sea ice cover while the laminated diatom ooze was deposited during warmer, ice-free periods during conditions of enhanced primary productivity and expanded OMZ.

The results of the quantitative geologic analyses of five of the sites drilled during Expedition 323 in the Bering Sea (Figs. 1 and 12), the Bering slope (Sites U1339, U1343, and U1344; Fig. 7) and Bowers Ridge (Sites U1340 and U1341; Fig. 8), can be summarized as follows.

Virtually all sediment recovered in the Bering Sea can be classified as silt. Sediment is finer in the Bering slope (very fine silt) than in Bowers Ridge (fine silt) (Table 2); at Bowers Ridge diatom valves are more abundant and clay minerals are less common. The majority of the sediment is composed of diatom valves and clay minerals, mixed in various proportions. Pelagic mud is more common in the Bering slope than in Bowers Ridge, where diatom ooze mainly alternates with diatom mud.

Approximately 40% of grain size variability can be explained by variations in abundance and/or preservation of diatom valves (Fig. 4D). One of the most striking characteristics of the Bering Sea sediment is that the integrity of the diatom valves is highly variable and that the variability occurs at different scales, ranging from millimeter-scale laminae (Figs. 10A–10D) to tens of meters–scale glacial-interglacial cycles. Sediment deposited during the Holocene and the interstadials and interglacials have low NGR, and are relatively coarse grained with well-preserved diatom valves, suggesting high diatom productivity (e.g., Fig. 10E). Conversely, the sediment from the stadials and glacials has high NGR, poorly preserved diatoms, and abundant clay, suggesting low biosilica flux due to reduced primary productivity and the presence of sea ice (e.g., Fig. 10F).

Depth variations of the main biogenic and siliciclastic sedimentary components can be interpreted within the context of the climatic history of the Bering Sea during the past 5 m.y. In the early Pliocene warm period the sediment at Bowers Ridge is characterized by relatively coarse and well-preserved diatom valves of heavily silicified forms (e.g., Coscinodiscus marginatus) indicating high nutrient supply (possibly as a result of decreased stratification), relatively high productivity, and high biogenic opal flux to the sediment. The transition from the early warm Pliocene and the ice ages (between 3.6 and 2.5 Ma) is characterized by a decrease in diatom productivity and opal flux (indicated by a dramatic increase in diatom fragmentation), slightly higher clay mineral content, and the first appearance of sea ice diatom and dinoflagellate forms. Despite these changes, diatoms were the dominant component of the sediment both during and after Northern Hemisphere glaciation. This suggests that the diatom “crash” observed at ODP Site 882 in the North Pacific was not a pervasive feature of the entire North Pacific and its marginal seas.

We thank the scientists of Integrated Ocean Drilling Program Expedition 323 for the constructive discussions both during and after the cruise. We also thank Michelle Drake for her help with the smear slide analysis and Justin Peglow for his assistance with the particle size analysis. This work has been supported by grant 0962974 from the National Science Foundation Division of Ocean Sciences.