The Salinity, Temperature, and δ18O of the Glacial Deep Ocean

See allHide authors and affiliations

Science  29 Nov 2002:
Vol. 298, Issue 5599, pp. 1769-1773
DOI: 10.1126/science.1076252


We use pore fluid measurements of the chloride concentration and the oxygen isotopic composition from Ocean Drilling Program cores to reconstruct salinity and temperature of the deep ocean during the Last Glacial Maximum (LGM). Our data show that the temperatures of the deep Pacific, Southern, and Atlantic oceans during the LGM were relatively homogeneous and within error of the freezing point of seawater at the ocean's surface. Our chloride data show that the glacial stratification was dominated by salinity variations, in contrast with the modern ocean, for which temperature plays a primary role. During the LGM the Southern Ocean contained the saltiest water in the deep ocean. This reversal of the modern salinity contrast between the North and South Atlantic implies that the freshwater budget at the poles must have been quite different. A strict conversion of mean salinity at the LGM to equivalent sea-level change yields a value in excess of 140 meters. However, the storage of fresh water in ice shelves and/or groundwater reserves implies that glacial salinity is a poor predictor of mean sea level.

The general circulation of the modern deep ocean is dominated by two distinct water masses: In the Atlantic, warm, salty North Atlantic Deep Water (NADW) is formed north of Iceland and in the Labrador Sea, whereas in the Southern Ocean, cold, fresh Antarctic Bottom Water (AABW) forms on continental shelves. Because of its greater density, AABW underlies NADW in the Atlantic at all latitudes south of ∼40°N. No deep waters are formed in the North Indian or North Pacific Oceans today. The mean deep-ocean tracer properties are strongly weighted toward those of the Indo-Pacific abyss, where deep waters are roughly a 50/50 mixture of AABW and NADW (1) and the average overturning time is about 800 years (2).

Measurements of passive tracer fields (3–5) and the radiocarbon age difference between surface and deep waters (6–9) suggest that during the LGM the Atlantic was flooded by waters of a Southern Ocean origin, the LGM analog to NADW shoaled but still exited the Atlantic, and the mean overturning rate was somewhat slower than today. These changing patterns of deep tracer distributions are consistent over many glacial cycles (10).

Although these tracer distributions for the LGM have provided significant insight into the mechanisms of climate change, a better understanding of the deep circulation of the past requires constraints on the seawater density, a function of temperature, salinity, and pressure. Oxygen isotope ratios (δ18O) within the carbonate shells of benthic foraminifera are a function of both temperature and seawater isotopic composition, but distinguishing between these two effects is a long-standing problem in paleoceanography (11–13). A series of recent studies using pore fluids (14–16) and Mg/Ca ratios (17–19) have attempted to separate the temperature effect, but relatively little is known about the glacial salinity. Following a preliminary study in the North Atlantic (14), we used the pore-fluid approach (20, 21) to independently constrain the seawater salinity by using profiles of chlorinity from four Ocean Drilling Program (ODP) sites (Table 1). By combining these data with δ18O measurements from the same locations (16), we have determined the spatial differences in salinity and temperature for the glacial deep ocean (22).

Table 1

Model-derived salinity, temperature, and δ18O water for the four ODP sites at the LGM. Modern values were taken from GEOSECS data and checked against the Levitus database. LGM theta values were calculated by subtracting the reconstructed Δδ18O water from the measured benthic δ18O change and dividing this residual by 0.26‰/°C (51). This in situ temperature was converted to potential temperature with the use of standard conversion algorithms. LGM salinities are the product of the modern bottom water salinity and the model derived % change from glacial to modern. [Cl] values are converted to salinity with the use of the standard factor of 1.80655. LGM δ18O of seawater is the sum of the modern bottom-water value and the model-derived glacial-to-modern change. Error estimates for all modern values account for the local vertical gradient and the proximity of the site to nearby Geochemical Ocean Sections (GEOSECS) stations. LGM error bars are the propagated values of these modern estimates and the uncertainty from each model reconstruction.

View this table:

Sediment-pore water samples for these measurements were squeezed at sea and sealed in glass ampoules for transfer to our shore-based labs (23). Sample resolution varied between and within sites but ranged from every 1.5 m at site 981 to every 9 m at the bottom of site 1123 (Figs. 1 and2). At all sites there is a pore-fluid maximum in δ18O and [Cl] corresponding to the LGM ice-volume peak. If these systems were exactly analogous to diffusion in a pipe, then we would expect the pore-fluid peaks in [Cl] and in δ18O to be about 25 m below the sea floor. However, sediment accumulation over the last 20 thousand years (ky), compaction-driven advection, and the duration of the last glacial period all work to alter the actual depth. At our sites, the amplitudes from bottom-water minima to glacial maxima range from ∼0.25 to 0.5 g/kg in chlorinity and 0.3 to 0.4 per mil (‰) in δ18O (24). We measure [Cl] rather than salinity because the ratios of the major ions in seawater are not constant with depth in the sediment column. Specifically, sulfate reduction and cation-clay interactions change the chemical composition of salinity. Chloride, outside of methane clathrate formation, is conservative in pore fluids and faithfully tracks the diffusively altered history of bottom-water salinity variations. Because of the separate bottom boundary conditions for the two species, slopes of [Cl] below the LGM peak tend to be less steep than those for δ18O. In several cases, the [Cl] values begin to increase again below 100 m. Each of these sites is influenced by local volcanism, where the conversion of fresh ash to clay minerals removes water from the sediments, but this process does not affect significantly the upper part of the profile where porosity is larger and where there is more recent exchange with bottom waters (25).

Figure 1

The top 100 m of pore-fluid [Cl] for four ODP sites and our model fits to the data. Solid lines are model results using the coral sea-level curve as the bottom-water [Cl] history. Dashed gray lines are a small alteration to this curve during the Holocene. There is more scatter in the [Cl] than the δ18O because evaporation and fresh tap-water addition during squeezing have a much larger effect on the [Cl] values than on the δ18O values. At site 1093, several of the points in the upper 20 m were rejected because they evaporated between sampling and measurement at Caltech. We are confident in ignoring these points because the shipboard [Cl] profile agrees with all of our points except these. Site 1123 has a wide scatter at the LGM peak and could be fit by a variety of other curves. The top boundary is constrained to match the local bottom-water value. This is a much stronger constraint for [Cl] than for δ18O because the modern database of salinity values is so much larger than for oxygen isotopes. Our pore fluids are standardized to the same reference (IAPSO) as modern conductivity temperature-depth sensors. mcd, meters composite depth.

Figure 2

The top 100 m of pore-fluid δ18O for four ODP sites and our model fits to the data. Except for site 1123, these data have been reported previously. Model reconstructions are about 0.1‰ heavier than those in our earlier work due to the change in the bottom-water boundary condition's shape. Errors in our reconstruction represent 2σ estimates based on individual measurement errors and the scatter in the profile.

The pore-fluid model (26) is forced with a time series of bottom-water [Cl] (or δ18O) for the last 125 ky. To construct this time series, we used a benthic foraminifera δ18O record from Pacific core V19-30 (27) and the coral sea-level record for the last 30 ky (28–30). Because the benthic record contains effects due to both global ice volume and bottom-water temperature, we corrected the δ18O record for temperature following Chappell and Shackleton (31). The results yield slight differences (∼0.3% in salinity and ∼0.1‰ in δ18O) from earlier work (14) for which such corrections were not made. The cumulative uncertainty in our pore-fluid reconstructions is smaller than the uncertainty in the glacial-interglacial change in δ18O of benthic foraminifera used for the calculation of deep-ocean temperature.

Model fits for the chloride data are largely constrained by two parts of the profile, the LGM peak and the modern bottom-water value. Between these two points the model follows the history of sea-level change from glacial to modern (28–30) and the physics of diffusion. However, the fits in Fig. 1 consistently overestimate the [Cl] at a given depth for every site, similar to a problem identified for δ18O profiles (16). Because we are calibrated to the same salinity scale as all of the modern hydrography measurements, our misfit must be due either to the imposed shape of the modern eustatic sea-level curve or to deep-water reorganizations during the Holocene, and not due to the wrong modern [Cl]. By gradually altering the sea-level curve over the last 12,000 years to overshoot the modern sea level by several meters and then returning it to modern values in the last 1000 to 500 years, we can better fit the data (Fig. 1, gray lines). The choice of a sea-level curve to fit the upper portion of the data is somewhat arbitrary and requires further work. The required changes in the forcing function through the Holocene can be thought of as either a change in sea level over the mid-Holocene or a shift in the position of deep-ocean water masses. The latter explanation requires a shift in the opposite direction (i.e., toward fresher values) at other yet-undiscovered locations. In several sites, the same misfits are apparent for δ18O (Fig. 2). However, the modern database for deep δ18O is too sparse to rule out small misfits between our assumed bottom-water values and the true oceanic isotopic composition.

The modern and LGM temperature-versus-salinity diagram (Fig. 3) confirms that all deep waters are within error of the freezing point of seawater. The deep Pacific (site 1123), the Southern Ocean (site 1093), and the deep Atlantic (sites 981 and 1063) are all bathed in very cold waters. In contrast to the modern circulation, the bulk of the deep LGM ocean was relatively homogeneous in temperature. Temperature contrasts between separate source waters seen in the modern ocean were largely eliminated during the LGM. Site 1063 is colder and fresher than the other areas. The LGM freezing point of surface seawater is –1.98°C, about 0.2°C warmer than our reconstructed value for the Bermuda Rise. Within the propagated error, our estimate for site 1063 does not violate the freezing point constraint, but the result may indicate that the low-resolution sampling of core 1063A has underestimated the glacial seawater δ18O change.

Figure 3

The modern and LGM temperature-versus-salinity relationships. The inset shows selected GEOSECS stations for all waters above a density of 48.0 σ4 and below a depth of 2500 m. Southern Ocean stations are shown in gray. The trends from the Southern Ocean into the three major basins are shown in black. Modern theta and salinity for each ODP site are shown as large circles. The main figure shows our estimates for LGM theta and salinity. The deep North Atlantic, the deep Pacific, and the deep Southern Ocean temperatures are all within error of the freezing point of seawater. The black arrow indicates the modern mean salinity. The LGM mean salinity for 125 m of eustatic sea-level change is shown by the bold arrow. Contour lines for the LGM data are isopychnals of σ4.

Our salinity data also indicate that the modern deep Atlantic gradient of salty NADW and fresh AABW was reversed. At the LGM, the Southern Ocean was the saltiest water in the deep ocean by a large margin. This result places several important constraints on the mechanisms of deep-water formation. Today NADW is salty because of the influence of the Gulf Stream and North Atlantic Drift current system, which brings salty tropical waters from the Gulf of Mexico to the NADW formation region in the Greenland, Norwegian, and Iceland Seas (32). There is recent evidence that the Gulf Stream's flow through the Florida Straight was about two-thirds lower during the LGM as compared to the present (33). Coupled with about 3° to 5°C of tropical cooling (34, 35) and the associated decrease in evaporation, this decreased transport is an important part of the observed relative freshness of the glacial North Atlantic deep-circulation end member.

However, what is most notable about the northern salinity is its relative freshness compared to the Southern Ocean. Site 1093 is so salty that a large imbalance in the freshwater budget must dominate the surface waters where this water mass is formed. The most likely mechanism for this imbalance is increased sea-ice formation and export (36). This mechanism is supported by our LGM δ18O-versus-salinity reconstruction (Fig. 4). Modern hydrography shows that the Southern Ocean is anomalously salty relative to the surface-water trend for its δ18O value (37). Sea-ice formation at and export from the sites of deep-water formation in the Southern Ocean preferentially fractionates salt by brine rejection but does not substantially change the water δ18O (38). However, Toggweiller (39) has demonstrated that sea ice alone cannot explain the change in salinity from east to west across shelf waters before they slide off their western edges to feed new deep-water formation. He attributes half of the anomaly in the δ18O signal to mixing with melt waters from the land-based Antarctic ice sheet, which are very fresh and extremely depleted in δ18O. Without surface data to constrain the LGM δ18O-versus-salinity relationship, we cannot make the same sort of quantitative partitioning between sea ice and glacial melting. However, the data in Fig. 4strongly support a large amount of sea-ice formation in the Southern Ocean. Although site 1093 has the highest salinity of the whole LGM deep ocean, it has similar δ18O water value to most of the rest of the ocean. This signature of brine rejection without oxygen isotopic fraction is characteristic of sea-ice formation.

Figure 4

The modern and LGM δ18O-versus-salinity relationships. The mean modern δ18O is by definition 0‰. The modern Southern Ocean deviates from the surface-water trend due to the combined effects of sea-ice formation and melting of the Antarctic Ice Sheet. The LGM data shows this same trend where site 1093 is much saltier but not much different in δ18O from the other sites. These data support the interpretation that there was significant sea-ice formation in the Southern Ocean at the LGM.

The presence of waters near the freezing point alters the balance of salinity and temperature's effect on seawater density (ρ), governed by the equation Δρ/ρ = αΔT – βΔS (where ρ is density, T is temperature, and S is salinity). In the modern ocean, where deep waters are on average 1° to 2°C potential temperature, the ratio of β, the haline contraction coefficient, to α, the thermal expansion coefficient, is about 10. Given the large temperature range of modern deep waters as compared to the modern salinity range, the modern Δρ is largely thermally controlled. For seawater near the freezing point, the β/α ratio rises to about 25. Density variations in the LGM deep were over two times more sensitive to salinity changes than they are today. Thus, the large expansion of the deep salinity gradient and the contraction of the thermal gradient at the LGM in our data are consistent with a much colder deep ocean. A first-order conclusion from our data is that the LGM thermohaline circulation was really more “haline” than “thermal,” with temperature setting the sensitivity of the system to salinity variations.

The salinity reconstructions (Fig. 3) at all sites are higher than 35.85 practical salinity unit (psu), the mean glacial value based on a 125-m eustatic sea-level change from Barbados corals (28) [this is 3.33% in the notation of Fig. 1(40)]. Recent results from northern Australia (41), coupled with a global isostatic adjustment model, argue for 135 m of global eustatic sea-level change (42). In addition, the modern minimum Red Sea sill depth is 137 m, and this marginal basin seems to have remained in contact with the open ocean during the LGM (43). Therefore, 135 m of global sea-level change, corresponding to a mean ocean salinity of 35.95 psu, is a maximum estimate. With this global salinity constraint and our new data, we calculate a salt mass balance for the glacial ocean. On the basis of the data in Table 1, the glacial end members for northern- and southern-source deep waters are 35.83 and 36.15 psu, respectively. The southern end member is not the same value as site 1093 because this high-latitude source water must have mixed with fresher waters as it spread northward. With the use of Cd/Ca and δ13C data to determine mixing ratios for the glacial Indian Ocean [50/50 northern/southern source mix (44)] and the glacial Atlantic Ocean [40/60 mix (45)], we calculated a mean Pacific salinity of 35.90 psu as a residual. This value is still lower than our measured salinity of 36.15 psu at site 1123 in the South Pacific. One possibility is that there is a mass of fresh water in the glacial ocean that we have not identified, possibly in the North Pacific.

Another explanation for our measurements is that they are consistent with 135 m of sea-level change and that 5 × 106km3 of freshwater equivalent were sequestered without affecting sea level. Increases in both groundwater reserves and floating ice shelves at the LGM have the potential to account for this large amount of fresh water. Lowering sea level by 135 m exposes about 5% of the current ocean area to the atmosphere. This new area and the depth of connected pore space below it represent a significant volume of water under the coastal ocean and continents that today contains salt but at the LGM contained fresh water. The net result is an increase in the glacial mean salinity that does not change sea level. We calculate a freshwater equivalent of 4.5 × 106 km3 for this fresh-salt water exchange (46). In the case of floating ice, adding the equivalent of about seven times the modern volume of Antarctic ice shelves (46) would balance the salinity budget. This is comparable to the increase proposed by Denton and Hughes (47). Although each of these freshwater reservoirs come close to solving the imbalance on their own, together the two make the uncertainty in glacial salinity distributions so large that the mean LGM salt concentration is not a good predictor of eustatic sea-level change.

  • * To whom correspondence should be addressed. E-mail: jess{at}


Stay Connected to Science

Navigate This Article