Simulation of Oxygen Isotopes in a Global Ocean Model A. Paul*, S. Mulitza, J. Pätzold and T. Wolff

Simulation of Oxygen Isotopes in a Global Ocean Model

A. Paul*, S. Mulitza, J. Pätzold and T. Wolff

Universität Bremen, Fachbereich Geowissenschaften, Postfach 33 04 40, D-28334 Bremen, Germany *corresponding author (e-mail): [email protected]

δ18 Abstract: We incorporate the oxygen isotope composition of seawater Ow into a global ocean model that is based on the Modular Ocean Model (MOM, version 2) of the Geophysical Fluid Dynamics Laboratory (GFDL). In a first experiment, this model is run to equilibrium to simulate the present-day ocean; in a second experiment, the oxygen isotope composition of Antarctic Surface Water (AASW) is set to a constant value to indirectly account for the effect of sea-ice. We check the δ18 depth distribution of Ow against observations. Furthermore, we computed the equilibrium δ18 fractionation of the oxygen isotope composition of calcite Oc from a paleotemperature equation δ18 δ18 and compared it with benthic foraminiferal O. The simulated Ow distribution compares fairly δ18 well with the GEOSECS data. We show that the Ow values can be used to characterize different δ18 water masses. However, a warm bias of the global ocean model yields Oc values that are too light by about 0.5 ‰ above 2 km depth and exhibit a false vertical gradient below 2 km depth. Our ultimate goal is to interpret the wealth of foraminiferal δ18O data in terms of water mass changes in the paleocean, e.g. at the Last Glacial Maximum (LGM). This requires the warm bias of the global ocean model to be corrected. Furthermore the model must probably be coupled to simple atmosphere and δ18 sea-ice models such that neither sea-surface salinity (SSS) nor surface Ow need to be prescribed δ18 and the use of present-day Ow-salinity relationships can be avoided.

Introduction

natural assembly of all isotopic species. Therefore, Isotopes in the Hydrological Cycle water vapor evaporating from the sea surface is Measurements of the isotopic composition of wa- depleted of heavy isotopes relative to ocean water, ter at the various stages of the hydrological cycle while rain precipitating from a cloud is enriched has enabled the identification of different water relative to the cloud moisture. Hence whenever a masses and the investigation of their interrelation- water sample undergoes a phase transition (e.g. ships; measurements of the variations of the iso- evaporation or condensation), temperature-depend- topic composition of proxy materials in climate ar- ent isotope fractionation occurs (Fig. 1). In the case chives such as ice and sediment cores has become of tritium, this effect is generally masked by mixing a most useful tool in paleoclimate reconstructions water from different sources (Jouzel 1986). But in (Gat 1996; Schotterer et al. 1996). the case of deuterium and oxygen-18, the isotopic The isotopic species of the water molecule that composition of precipitation shows a large variation are of most interest to the geosciences are with latitude, height and continentality. The deple- 1 2 16 1 18 1 3 16 H H O, H2 O and H H O. A water molecule tion of precipitation in deuterium and oxygen-18 is containing the radioactive isotope 3H (tritium) or most pronounced in the polar regions. The variation either of the stable isotopes 2H (deuterium) and 18O in the isotopic composition of seawater is compara- (oxygen-18) is heavier than the water molecule tively small and mainly determined by freshwater 1H2H16O that is by far the most abundant in the input and mixing between water masses.

From FISCHER G, WEFER G (eds), 1999, Use of Proxies in Paleoceanography: Examples from the South Atlantic. Springer-Verlag Berlin Heidelberg, pp 655-686 656 Paul et al.

The stable isotopes deuterium and oxygen-18 The δ value that is commonly used to express have been incorporated into a number of atmos- the isotopic composition of a water sample is de- pheric models, e.g. the LMD model (Joussaume et fined by al. 1984; Joussaume and Jouzel 1993), the NASA/ GISS model (Jouzel et al. 1987), the ECHAM δ = (R / R −1)⋅1000‰, (1) model (Hoffmann and Heimann 1993; Hoffmann sample standard 1995) and the GENESIS model (Mathieu 1996). As these four models account for the various where Rsample and Rstandard refer to the isotope 2 1 18 16 fractionation processes that occur at the sea sur- ratio H/ H or O/ O in the water sample and in δ face and all succeeding stages of the hydrological the standard. Thus positive values indicate an cycle, they provide a tool for calculating the isotopic enrichment of the heavy isotopic species relative δ content of precipitation δP as well as of evapora- to the standard, and negative values indicate their tion δE (Juillet-Leclerc et al. 1997). Prior to their depletion. The generally used standard is the Stand- development there was no method to estimate δE ard Mean Ocean Water (SMOW, Craig 1961) or, independently (Craig and Gordon 1965). more recently, the Vienna Standard Mean Ocean Water (V-SMOW, Gonfiantini 1978).

Fig. 1. Sketch of the global δ18O cycle. In our simulations, only the oceanic part has been considered so far. Simulation of Oxygen Isotopes in a Global Ocean Model 657

The focus of this paper is on oxygen isotopes, of interest (Weiss et al. 1979; Jacobs et al. 1985; because oxygen-18 variations are directly related Bauch 1995; Toggweiler and Samuels 1995). to paleotemperature studies. It is in fact for this To give an explicit example, we reproduced the reason that most of the oceanographic work to date detailed freshwater budget for the Ross Sea conti- has concentrated on oxygen-18 and only few deu- nental shelf constructed by Jacobs et al. (1985). terium measurements have been performed (Craig From the difference between the δ18O content and Gordon 1965). of shelf water and its source they estimated that 36 cm a-1 of glacial meltwater is introduced to the shelf. They found that surface and deep waters Applications in Present-Day Oceanography move onto the shelf with an average δ18O content The distributions of both salinity S and oxygen iso- of -0.22 ‰. On the shelf, its δ18O content will be δ18 δ tope composition of seawater Ow or simply w lowered by marine precipitation, glacial meltwater are mainly controlled by the same two processes and the net freezing of sea ice, such that the aver- δ δ18 precipitation P and evaporation E. Locally S and w age O content of shelf water becomes -0.42 ‰: of surface waters are linearly related. However, the (1) Marine precipitation at a rate of 15 cm a-1 with δ δ18 δ18 slope of the w-S relationship varies between 0.1 an O content of -16.5 ‰ will lower the O for tropical surface waters and 1 for high-latitude content of shelf water by 0.005 ‰ a-1. (2) Further surface waters. This reflects the so-called tempera- depletion of 0.025 ‰ a-1 will result from the addi- ture effect: in high latitudes, precipitation occurs at tion of glacial meltwater at a rate of 36 cm a-1 with lower temperatures and is more enriched in oxy- an δ18O content of -36 ‰. (3) The decrease in δ gen-18 than in low latitudes, which leads to a strong shelf water w from sea ice freezing will be only depletion of the remaining cloud moisture. A slope 0.002 ‰ a-1, due to the near-zero slope of the freez- of 1 is observed for East Greenland surface wa- ing line. The net change in δ18O by -0.20 ‰ a-1 and ters, which are diluted by adding meltwater from the glacial addition of meltwater at a rate of 36 cm the permanent ice cap. Freshwater input from riv- a-1 require a shelf water residence time of 6.2 a. δ ers can also modify the w-S relationship. Hence S An important application of a detailed freshwa- δ and w label the surface waters in different ways. ter budget is the study of the role of sea ice in the Mixing of water masses that originate from the sea formation of Antarctic Bottom Water (AABW) surface creates the characteristic vertical sections (Toggweiler and Samuels 1995; Stössel et al. 1996). known from GEOSECS (Birchfield 1987; Östlund In the example of Jacobs et al. (1985), the onshelf et al. 1987). flow has an average salinity of 34.455 and the The isotopic composition of sea water is also offshelf flow has an average salinity of 34.589, affected by the formation of sea ice, but the which leaves a salinity enrichment of 0.134 units. fractionation effect is so small that the freezing This salinity enrichment corresponds to the brine process leads to an increase in salinity, with essen- rejected during the formation of sea ice at a rate of δ -1 tially no observable influence on w. This gives a 37 cm a , with a residual salinity of 4.1, from Ant- δ near-zero slope in a w-S diagram (Craig and arctic Surface Water (AASW) with a salinity Gordon 1965). As a result, the surface waters of of 34.1. Furthermore salt must be drained from the Southern Ocean show a wide range of salinities, 58 cm a-1 of sea ice to balance 15 cm a-1 of marine δ -1 although w is nearly constant. For Antarctic Sur- precipitation and 36 cm a of glacial meltwater. face Water (AASW), Jacobs et al. (1985) report a Toggweiler and Samuels (1995) pointed out that on δ w value of -0.31 ‰ with a standard deviation of the one hand brine rejection causes a net salinity 0.08 ‰. enrichment of only 0.134 units, which is much δ In summary, in present-day oceanography w smaller than the value proposed by Broecker (1986) δ allows to distinguish between different freshwater on the basis of open ocean GEOSECS w meas- sources. Thus it is possible to investigate the influ- urements, but on the other hand brine rejection is ence of local P - E, river-runoff and glacial melt- essential in compensating the effects of marine pre- water, sea ice formation and decay on a water mass cipitation and glacial meltwater on the shelf. 658 Paul et al.

Applications in Paleoceanography temperature equation and compared with benthic foraminiferal δ18O. In contrast to present-day oceanography, applica- Here we show the results of a simulation of the tions in paleoceanography do not rely on the iso- present-day ocean and discuss to what degree δ tope content of seawater, but on that preserved in w and δ can be used to characterize different water benthic and planktic foraminifera: the equilibrium c masses. We find that δ can indeed serve to trace fractionation of the oxygen isotope composition of w the formation and spreading of deep and bottom calcite δ18O or simply δ . For example, Broecker c c waters. (1986, 1989) used benthic and planktic foraminifera With respect to δ we face a fundamental prob- to constrain tropical SSTs and to determine the c lem: The decrease of temperature with depth acts salinity contrast between the tropical Atlantic and against the decrease of δ with depth. Hence be- Pacific Oceans. Duplessy et al. (1991) present a w low 2 km depth in the present-day Atlantic Ocean, reconstruction of Last Glacial Maximum (LGM) the δ values attain a nearly constant level and do sea surface salinity (SSS) derived from planktic c not allow to separate deep and bottom waters (cf. foraminifera. A number of studies address the ver- Zahn and Mix 1991, Figs. 4 and 5; however, in the tical structure of the ice-age ocean (e.g., Birchfield glacial Atlantic Ocean there might be a small off- 1987; Zahn and Mix 1991; Labeyrie et al. 1992). set of 0.2 ‰ between deep and bottom waters). A Lehman et al. (1993) (in a two-dimensional ocean technical problem that can probably be solved by model) and Mikolajewicz (1998) (in a three-dimen- more realistic heat flux and mixing sional ocean model) employ an idealized δ tracer w parameterizations is the warm bias of our ocean in order to compare meltwater-induced changes in model. This warm bias yields δ values that are too the thermohaline circulation with the geologic c light by about 0.5 ‰ above 2 km depth and exhibit record. a false vertical gradient below 2 km depth. The calculation of paleosalinity from δ requires w Our ultimate goal is a simulation of the linear δ -salinity relationships that are temporally w paleocean, e.g., at the Last Glacial Maximum invariant on short (monthly to seasonal) as well as (LGM). This probably requires the ocean model to on long (geological) timescales. Rohling and Bigg be coupled to simple atmosphere and sea-ice mod- (1998) critically review this basic assumption and els such that neither sea surface salinity (SSS) nor find that it is unwarranted for many regions of the surface δ need to be prescribed and the use of global ocean because of local changes in the fresh- w present-day δ -salinity relationships can be water budget and sea-ice coverage that are w avoided. advected into the ocean interior. This conclusion δ is supported by the estimation of surface w from an atmospheric model (Juillet-Leclerc et al. 1997) Ocean Model δ and the simulation of surface w in a global ocean model using full flux boundary conditions at the sea Governing Equations surface (Schmidt 1998). We feel attracted to the idea of Zahn and Mix The ocean model we employ is based on the Modu- (1991) to interpret the wealth of oxygen isotope data lar Ocean Model (MOM, version 2) of the Geo- in terms of water mass changes. To this end, we physical Fluid Dynamics Laboratory (GFDL) δ also incorporated w into an ocean model, but we (Pacanowski 1996). It is governed by the primitive δ prescribe w at the surface and employ restoring equations. These equations are derived from the boundary conditions, with the advantage that the basic laws of physics, particularly the conservation ocean model can be run to equilibrium. The depth equations for momentum, mass and energy. They δ distribution of w can then be compaired with the are formally written in spherical polar coordinates: δ λ φ observations. Since w and T are simulated simul- longitude increasing to the east, latitude increas- δ taneously, c can be determined from a paleo- ing to the north and height z increasing upward. Simulation of Oxygen Isotopes in a Global Ocean Model 659

The coordinate z is measured from the sea-level The first term on the left hand side of Eq. (2) is geo-potential surface a = 6367.456 km, parallel to the time rate of change of the longitudinal velocity the local direction of gravity (Gill 1982). Seven vari- component, the second term is the advection of the ables specify the physical state of the ocean: the longitudinal velocity component with the flow field, longitudinal, latitudinal and vertical velocity compo- the third term is a metric term that is due to the nents u, v and w, pressure p, density ρ, potential curvature of the Earth and the fourth term is the temperature θ and salinity S. Coriolis acceleration resulting from the Earth’s ro- Discussions of the governing equations may be tation; the first term on the right hand side is the found in the books by Washington and Parkinson longitudinal component of the pressure gradient (1986) and McGuffie and Henderson-Sellers (1996) force and the second and third term derive from at an introductory level, and in the book by Landau the vertical and horizontal frictional forces. The and Lifschitz (1987) and the articles by Veronis terms in Eq. (3) have an analogous meaning. (1973), Cane (1986) and Semtner (1986) at an The hydrostatic balance and continuity equations advanced level. Several approximations are used become: to simplify the conservation equations for momentum (Navier-Stokes equation) and mass (continu- ∂p ity equation): =−ρg (4) ∂z •the thin shell approximation (on account of the shallowness of the ocean z << a) ∂w 1 ∂u ∂  = −  + ()cosφv  (5) •the hydrostatic approximation (because ocean ∂z a cosφ ∂λ ∂φ  motions are slow, the gravitational force and the vertical gradient of pressure are balanced very Eq. (4) simply relates the vertical pressure gra- well) dient (on the left hand side) to the gravitational force (on the right hand side). Eq. (5) expresses the ver- •the Boussinesq approximation (since seawater tical gradient of the vertical velocity (on the left is nearly incompressible, variations in density are hand side) in terms of the longitudinal gradient of neglected, except where buoyancy effects are the longitudinal velocity and the latitudinal gradient concerned) of the latitudinal velocity (on the right hand side).

In Eqs. (2)-(5) t is time, r0 is the reference den- •the neglect of terms involving w in the horizon- sity, Av is the vertical eddy viscosity coeffcient and tal momentum equations (on the basis of scale g is the gravitational acceleration. Furthermore, L analysis) denotes the advection operator,

With respect to these approximations, the hori- 1 ∂ 1 ∂ ∂ L()α = ()uα + ()cosφvα + ()wα zontal momentum equations are: a cosφ ∂λ a cosφ ∂φ ∂z , (6) ∂ u uv tanφ 1 ∂ p ∂  ∂ u  + L()u − − fv = − +  A  + F u , ∂ t a ρ a cosφ ∂λ ∂ z  v ∂ z  0   where α is a generic scalar variable, and f is the (2) Coriolis parameter,

∂ v u 2 tanφ 1 ∂ p ∂  ∂ v  , (7)   ν f = 2Ω sinφ + L()v − + fu = − +  Av  + F . ∂t a ρ0 a ∂ φ ∂ z  ∂ z  (3) with Ω being the angular velocity of the Earth. Finally, Fu and Fv represent the horizontal eddy stress divergence, 660 Paul et al.

The equation of state for seawater relates ρ , seawater density to in situ temperature T , salinity S and pressure p: (8) ρρ= ()TSp,, . (14) 2 v (1− tan φ )v 2sinφ∂u / ∂λ  h h F = ∇ ⋅ ()ν ∇v +ν  2 + 2 2 ,  a a cos φ  In our ocean model we use a polynomial ap- (9) proximation to the equation of state developed by the Joint Panel on Oceanographic Tables and Stand- ards (UNESCO 1981). Here Ah is the horizontal eddy viscosity coeffi- cient, and In order to satisfy the viscous stability criterion, the horizontal viscosity and diffusivity coefficients are tapered towards the poles (Weaver and Hughes 1 ∂sλ ∂  1996; NCAR Oceanography Section 1996). Con- ∇⋅s = + ()sφ cosφ (10) αφcos  ∂λ ∂φ  sequently, extra metric terms are included in the   momentum equations, such that pure rotation does not generate a viscous stress (Wajsowicz 1993).  1 ∂α 1 ∂α  These extra metric terms are ∇α =  ,  (11)  acosφ ∂λ a ∂φ   u tanφ 1 ∂ v  ∂ A  1 ∂ v vsinφ  ∂ A  +  h −  +  h  2 2   2 2 2   a a cosφ ∂ λ  ∂ φ  a cosφ ∂ φ a cos φ  ∂ λ are the horizontal divergence and gradient opera- (15) tors, where s is a generic vector variable. The con- in the longitudinal momentum equation (2) and servation of heat is expressed in the conservation equation for potential temperature θ: F u = ∇  v tanφ 1 ∂ u  ∂ A  usinφ 1 ∂ u  ∂ A  −  h +  +  h  2 2   2 2 2   a a cosφ ∂ λ  ∂ φ  a cos φ a cosφ ∂ φ  ∂ λ ∂θ ∂  ∂θ  (16) +=L()θ  Dvh +∇⋅()D ∇θ , (12) ∂t ∂z  ∂z  in the latitudinal momentum equation (3). where the first term on the left hand side is the Model Resolution, Land-Sea Mask and time rate of potential temperature change and the Bathymetry second term represents the advection of potential The model resolution, land-sea mask and temperature with the flow field; the first term and bathymetry are the same as in the coarse-resolu- second term on the right hand side denote the ver- tion ocean model of Large et al. (1997). The only tical and horizontal diffusion of potential tempera- exception is that two deep vertical levels are added, ture. A similar conservation equation holds for sa- such that the maximum model depth is now 5900 linity S: m, and that there are 27 vertical levels in total, monotonically increasing in thickness from 12 m ∂S ∂  ∂S near the surface to 450 m near the bottom. The +=LS()  Dvh+∇⋅()DS ∇ . (13) ∂t ∂z  ∂z  longitudinal resolution is 3.6°. The latitudinal resolution is 1.8° near the equator (to better resolve the equatorial currents), increases away from the equa-

Here Dv and Dh are the vertical and horizontal tor to a maximum of 3.4°, then decreases in the mid- eddy diffusivity coefficients. latitudes as the cosine of latitude (to maintain Simulation of Oxygen Isotopes in a Global Ocean Model 661 horizontally isotropic grid boxes) and is finally kept Model Viscosity and Diffusivity, Filtering constant at 1.8° poleward of 60° (to prevent any and Additional Tracer further restrictions on the model time step). The continental outlines of the model are shown Since the horizontal resolution is coarser than the in Fig. 2. While the 9.4°- wide Drake Passage Rossby radius of deformation at all latitudes, the closely matches its actual width, the Indonesian effects of mesoscale eddies are only parameterized Channel, the Florida Straits and the connections to in terms of enhanced viscosity and diffusion. The horizontal viscosity is Α = 2.5 x 105 m2s-1. The the Sea of Japan are much wider than in reality. h vertical viscosity is A = 16.7 x 10-4 m2 s -1. Follow- Furthermore the Bering and Gibraltar Straits and v the Red Sea outflow are closed. ing Bryan and Lewis (1979), depth-dependent hori- The bathymetry used in the present study is zontal and vertical diffusivities are employed. taken from the ocean model of Large et al. (1997) Hence and is fairly realistic, e.g. the maximum depths of the Denmark Strait and Faroe-Iceland ridge are Dzhbhshbh()=+ D ( D − D )exp ( − z / 500 m ), (17) 734 m and 903 m. Regions deeper than 5000 m are filled in from the ETOPO5 topography data such that the horizontal eddy diffusion Dh decreases (NCAR Data Support Section 1986). 6 2 -1 from Dsh= 0.8 x 10 m s in the top layer to Dbh = 0.4 x 106 m2s-1 in the bottom layer. Furthermore

Fig. 2. Continental outlines and bathymetry of the model. Contour interval is 0.5 km. 662 Paul et al.

The set of boundary conditions at the ocean Cr surface consists of mid-month fields of the zonal Dzv()=+ D**arctan λ()zz − , (18) π [] and meridional wind stress components, the sea surface temperature (SST), the sea surface salin- -4 2 -1 -4 2 ity (SSS) and the sea surface oxygen isotope com- where D* = 0.61 x 10 m s , Cr = 1.0 x 10 m -1 λ -3 -1 position. The surface wind stress fields are obtained s , = 1.5 x 10 m and z* = 1000 m (Weaver and Hughes 1996), such that the vertical viscosity from the NCEP reanalysis data covering the four -4 2 -1 years 1985 through 1988 (Kalnay et al. 1996), and Dv is about 0.3 x 10 m s at the surface and 1.1 x 10-4 m2 s-1 in the deep ocean. To increase the per- the surface temperature and salinity fields are de- missible time step, the flow variables are Fourier- rived from the climatologies of Shea et al. (1990) filtered north of 75° N, and the horizontal viscosity and Levitus (1982), as described by Large et al. (1997). Dh is tapered north of 81.9° N. The oxygen iso- δ The surface δ forcing field is computed from tope composition of seawater w is incorporated into w the ocean model as a passive tracer that does not the mid-month fields of the sea surface salinity using δ influence the ocean circulation. a set of seven w -salinity relationships, in a man- δ ner similar to Fairbanks et al. (1992). The w and Experimental set-up salinity data are taken from GEOSECS (Östlund et al. 1987), Mackensen et al. (1996), Bauch (1995) In the control experiment (Experiment A), the ac- δ and Veshteyn et al. (1974). The set of w -salinity celeration technique of Bryan (1984) is used to relationships is depicted in Fig. 3 and summarized accelerate the approach to equilibrium in the deep in Table 1. ocean. Thus, in the first stage, the tracer time step δ The surface temperature, salt and w fluxes that is ten times larger in the deep ocean than at the enter the tracer equations are calculated from the surface. Furthermore the surface tracer time step usual restoring boundary conditions given by is ten times larger than the time step of 3504 s for the momentum and barotropic streamfunction equations. In the second stage, the integration is H * H * F = − ()T − T , F = − ()S − S , continued with equal time steps of 3504 s for all T τ s τ equations at all depths. The length of the accelerated phase is 96 momentum years, 960 surface H * Fδwww=−()δδ − . (19) tracer years and 9600 deep tracer years, as in the τ experiments of Large et al. (1997). The length of the synchronous phase is 15 years, which is the Here τ = 50 d is a relaxation time scale relative minimum length recommended by Danabasoglu et to the upper ocean depth H = 50 m, and T*, S* and δ al. (1996). In the sensitivity experiment (Experiment *w denote the prescribed SST, SSS and surface δ B), the length of the accelerated phase is 20 mo- w values. The fraction of a grid cell covered by mentum years, 200 surface tracer years and 2000 sea ice is diagnosed from the climatologic SST and deep tracer years, and the length of the synchro- subject to a strong temperature restoring, with a τ nous phase is 15 years. time constant ice= 6 d (cf. Large et al. 1997). Two Experiment A is started from a set of initial experiments are performed: In Experiment A, a conditions that corresponds to a state of rest with slope of a = 1.030 and an abscissa of b = -35.80 ‰ January-mean distributions of potential temperature is used to compute the isotopic composition of and salinity from Levitus (1982). The converged AASW. In Experiment B, a constant value of - state at the end of the accelerated phase of Ex- 0.31‰ is used. This corresponds to a zero slope δ periment A is then used as the set of initial condi- of the w -S relationship and indirectly accounts for tions for Experiment B. the sea-ice effect. Simulation of Oxygen Isotopes in a Global Ocean Model 663

δ Fig. 3. w -salinity relationships for various regions of the global ocean, based on the GEOSECS data. The Trop- ics extend from 25°S to 25°N. The Arctic Ocean includes the North Atlantic Ocean north of 60°N. The term Antarctic Zone is used to describe the Southern Ocean δ surface waters south of the 2°C-isotherme. The w -salinity relationship for the mid-latitudes is applied to the remaining regions of the global ocean. 664 Paul et al.

Region a b Note Source Arctic Ocean 1.010 -34.75 Veshteyn et al. (1974), Bauch (1995) Tropical Atlantic 0.180 -5.63 Östlund et al. (1987) Tropical Pacific 0.285 -9.50 Östlund et al. (1987) Tropical Indian 0.180 -5.74 Östlund et al. (1987) Midlatitudes 0.497 -17.05 Östlund et al. (1987) Antarctic Zone 1.030 -35.80 Experiment A Mackensen et al. (1996) 0.000 -0.31 Experiment B Jacobs et al. (1985)

δ Table 1. Slopes a and abscissas b of the w -S relationships. In the last column, the sources of the data used for calibration are given.

This is in good agreement with the estimates of Results Gordon et al. (1987) and Stramma and Lutjeharms (1996). The 25 Sv-flow through the Mozambique Channel is significantly stronger than the 5 Sv ob- Barotropic Streamfunction served by Stramma and Lutjeharms (1996). The The vertically integrated volume transport is pre- flow in the Benguela Current is 12 Sv. According sented in Fig. 4. Here we use a streamfunction map to Garzoli and Gordon (1996), the observed trans- rather than a vector map because a streamfunction port at 30°S in the upper kilometer of the ocean is map is easier to visualize on a global scale. The flow 13 Sv, with considerable seasonal variations. The is directed along the streamlines, with the sense of maximum Brazil Current transport amounts to 30 rotation clockwise for positive values, and anti- Sv, on the high side of the estimate of Peterson and clockwise for negative values of the stream- Stramma (1991). function. The flow rate through any line inter- A distinct feature of the vertically integrated secting two adjacent streamlines is given by the mass transport is the Antarctic Circumpolar contour interval. Current (ACC). The flow through Drake Passage The spatial features of the vertically integrated is 225 Sv. An observational estimate for this flow mass transport are determined by the wind forc- is 130±20 Sv (Whitworth and Peterson 1985). Thus ing and the bathymetry. Because of the identical it is clearly overestimated in our model. As dem- winds and the almost identical bathymetry, all gyre onstrated by Danabasoglu and McWilliams (1995), patterns and transport magnitudes are very similar the ACC flow through the Drake Passage depends to the solutions of Large et al. (1997). The major on the horizontal or isopycnal diffusion coefficients, currents and passage throughflows are reproduced showing an enhancement in transport with de- reasonably well. creased diffusivity. Furthermore, the ACC trans- In the Northern Hemisphere, the maximum port is larger in the case of horizontal diffusion than transport in the Gulf Stream is 21 Sv. This is lower in the case of isopycnal diffusion. In our model than that observed, which is about 40 Sv at 30°N, the vertically averaged horizontal diffusivity is upstream of the recirculation regime (Knauss 0.43x103 m2s-1. Extrapolating from the two solu- 1969), or 32.2±3.2 Sv for the Florida Current alone tions of Danabasoglu and McWilliams (1995) with (Larsen 1992). The maximum transport in the horizontal diffusion we find a flow of about 250 Sv. Kuroshio is low, too. The Atlantic inflow into the From the studies of Danabasoglu and McWilliams Nordic Seas is 4 Sv, about half the observational (1995) and Large et al. (1997) we conclude that estimate by Gould et al. (1985). we can obtain a lower and more realistic ACC In the Southern Hemisphere, we obtain for the transport by turning to an isopycnal diffusion Agulhas Current a maximum transport of 62 Sv. parameterization. Simulation of Oxygen Isotopes in a Global Ocean Model 665

Meridional Overturning Streamfunction of as a manifestation of the thermohaline circulation, although in the upper ocean the upwelling and The zonally integrated meridional overturning downwelling induced by the surface wind stress are streamfunction is presented in Figs. 5a-c. Figure the dominant processes. In Fig. 5a two shallow cells 5a shows the total transport of all the ocean basins are depicted that straddle the equator. These cells in one section, whereas Figs. 5b and c show the are driven by the divergence of the Ekman trans- total transport broken down among the Atlantic and port under the easterly trade winds. Figures 5a and Indo-Pacific basins, restricted to those latitudes b also indicate downwelling in mid-latitudes that where the ocean basins are limited zonally by con- produce another set of overturning cells, which is tinental boundaries. The sections that represent the linked to wind-induced divergences in the subpolar meridional overturning streamfunction are read in regions. The overturning cell in the Northern Hemi- the same way as the barotropic streamfunction map sphere is much weaker and shallower than the so- (Fig. 4). called Deacon cell in the Southern Hemisphere. The mode of ocean circulation most important This is due to a partial cancellation between the for ventilating the deep sea is the vertical overturn- northern cell and the thermohaline circulation ing (Toggweiler et al. 1989). It is usually thought (Toggweiler et al. 1989).

Fig. 4. Annual-mean barotropic streamfunction. Contour interval is 10 Sv below a vertically integrated volume transport of 60 Sv and 20 Sv above. The dashed lines represent negative contour levels and indicate anti-clockwise circulation. 666 Paul et al.

Between 1500 m and 3500 m the meridional Potential Temperature and Salinity Dis- overturning is dominated by the southward flow of tributions North Atlantic Deep Water (NADW), which is produced at a rate of 21 Sv. As much as 7 Sv of The annual and zonal mean distributions of poten- this transport upwells north of the Equator, only 14 tial temperature and salinity for the Atlantic Ocean Sv of it is exported into the Southern Ocean. An are presented in Fig. 6. Figure 7a depicts the cor- estimate for the total southward transport of responding deviation of potential temperature from NADW at 24°N is 17±4 Sv (Roemmich and the climatology of Levitus et al. (1994). All water Wunsch 1985), which is somewhat larger than the masses are up to 4°C warmer than the observa- 15-Sv flow in the model. tions, except for the AABW that is colder than the There is a northward flow of Antarctic Bottom observations by about 0.5°C. The warm bias is Water (AABW) in all three ocean basins, compen- largest in the thermocline, which is also more dif- sated by a return flow above 4000 m depth. In the fuse than observed. A warm bias and a diffuse Atlantic, the deep inflow at the bottom amounts to thermocline are two widespread problems in ocean 3 Sv. modelling (Danabasoglu and McWilliams 1995).

Fig. 5. Annual-mean meridional overturning streamfunction. Contour interval is ~2Sv except for the Atlantic where it is ~1Sv for negative levels. The negative contour levels are represented by dashed lines that indicate anti-clockwise circulation. (a) For the present global ocean. (b, right) For the present Atlantic Ocean. (c, right) For the present Indo-Pacific Ocean. Simulation of Oxygen Isotopes in a Global Ocean Model 667 668 Paul et al.

Fig. 6. Annual and zonal mean tracer distributions for the present Atlantic Ocean. (a) Potential temperature. Contour interval is 2°C. (b) Salinity. Contour interval is 0.25. Simulation of Oxygen Isotopes in a Global Ocean Model 669

Fig. 7. Annual and zonal mean tracer distributions for the present Atlantic Ocean. Difference between simulation and climatology. (a) Potential temperature. Contour interval is 0.5°C. (b) Salinity. Contour interval is 0.05. 670 Paul et al.

Figure 7b illustrates the deviation of salinity from Combined T-S-δ 18O Diagrams the climatology of Levitus and Boyer (1994) for the Figure 12 shows combined T-S-δ18O diagrams for Atlantic Ocean. It reveals that the tongue of Ant- the GEOSECS data and Experiments A and B. The arctic Intermediate Water (AAIW) is rather broad GEOSECS data were taken from three stations in and much too salty in the model ocean. The AABW the North Atlantic and five stations in the South At- is too fresh by up to 0.30 units. lantic. The ocean model output was sampled at the same geographic positions and depths. Fig. 12 also Seawater Oxygen Isotope Composition and depicts the temperature-salinity ranges of Mediter- ranean Water (MW), AAIW, NADW and AABW, Benthic Foraminiferal δ 18O as defined by Emery and Meincke (1986) (see δ In Fig. 8 the GEOSECS w along a cross section Table 2). in the Atlantic Ocean is compared with the annual The GEOSECS serial points all plot on curves δ and zonal mean w from Experiments A and B. Ex- that hook into the cold AABW (Fig. 12a). The two δ periment A yields a heavy w value of AABW, other end members are the relatively fresh AAIW δ whereas in Experiment B the w of AABW is close (stations 56 and 67) and the relatively salty NADW to the GEOSECS data. Both experiments lead to (stations 29, 37 and 115). Finally the influence of a broad tongue of AAIW that is also seen in the MW is reflected in the T-S curve of station 115. δ δ18 salinity distribution. The w of AAIW is heavier than These four water masses are also seen in the O observed by 0.1-0.2‰. Figure 9 extends the com- values, which proves that they can indeed be used δ parison of the GEOSECS and the simulated w to to characterize different water masses. However, the Pacific Ocean. In the upper 3000 m the agree- there can be no one-to-one correspondence, e.g. δ ment is reasonably good. The w of AABW is too there can be water types that are similar in tem- light by about 0.1‰. Unfortunately, there are no perature T and isotopic composition δ18O, but dif- δ GEOSECS w data between 30°S and 15°N. fer in salinity S, conceivably because of a sea ice Anticipating an application of the global ocean effect (stations 85 and 91 in the Southern Ocean). δ model in a paleoceanographic setting, Fig. 10 Table 2 also gives the w statistics for the serial presents core top benthic foraminiferal δ18O com- points from all GEOSECS stations in the Atlantic piled from 169 Atlantic core sites. To facilitate a that fall into the temperature-salinity ranges of the comparison with the model ocean, the equilibrium four water masses MW, AAIW, NADW and δ δ fractionation of c is computed for Experiment B. AABW: MW has the highest w mean value δ Two cases are studied: In the first case (Fig. 11a) (0.36‰), AABW the lowest (-0.13‰). The w the model temperature is inserted into the mean value of NADW (0.20‰) is between that paleotemperature equation of Erez and Luz (1983). of AAIW (0.00‰) and AABW. The 1σ intervals In the second case (Fig. 11b) the climatologic tem- do not overlap except for AAIW and AABW. This perature of Levitus et al. (1994) is used (cf. Fig. 4 reflects that AAIW and AABW both have sources δ of Zahn and Mix (1991) who employed the in the Southern Ocean, the lower w mean value δ GEOSECS temperature and w data to estimate of AABW being due to the addition of glacial melt- δ δ c). In the real ocean the decrease of w with depth water. is offset by a corresponding decrease of tempera- The influence of MW is missing from the ocean ture. Below about 2000 m water depth, a roughly model output simply because the Mediterranean is constant level of about 3‰ is reached, such that not included in the ocean model domain (Figs. 12b δ NADW and AABW can be seen in w , but not in and c). The other water masses are found in the δ c (Fig. 11b). This is also suggested by the Atlan- ocean model, but the deficiencies in their represen- tic core top benthic foraminiferal δ18O shown in Fig. tation are manifest in the temperature-salinity 10. The warm bias in the model temperature intro- ranges that differ significantly from those given by duces a false vertical gradient into the simulated Emery and Meincke (1986): AAIW and NADW δ c distribution (Fig. 11a). are too warm, and AABW is too cold and fresh Simulation of Oxygen Isotopes in a Global Ocean Model 671

Fig. 8 (a, b). Seawater oxygen isotope composition for the present Atlantic Ocean in units of parts per mil versus δ SMOW. Contour interval is 0.1 ‰. (a) GEOSECS w along a cross-section in the Atlantic Ocean. Crosses indicate δ data points. (b) Annual and zonal mean w from Experiment A. 672 Paul et al.

Fig. 8 (c, d). Seawater oxygen isotope composition for the present Atlantic Ocean in units of parts per mil versus δ SMOW. Contour interval is 0.1 ‰. (c) Annual and zonal mean w from Experiment B. (d) Difference between Experiment B and Experiment A. Simulation of Oxygen Isotopes in a Global Ocean Model 673

Fig. 9. (a, b) Seawater oxygen isotope composition for the present Pacific Ocean in units of parts per mil versus δ SMOW. Contour interval is 0.1 ‰. (a) GEOSECS w along a cross section in the Pacific Ocean. Crosses indicate δ data points. (b) Annual and zonal mean w from Experiment A. 674 Paul et al.

Fig. 9. (c, d) Seawater oxygen isotope composition for the present Pacific Ocean in units of parts per mil versus δ SMOW. Contour interval is 0.1 ‰. (c) Annual and zonal mean w from Experiment B. (d) Difference between Experiment B and Experiment A. Simulation of Oxygen Isotopes in a Global Ocean Model 675

MW AAIW NADW AABW Temperature range (°C) 2.6-11.0 2.0-6.0 1.5-4.0 -0.9-1.7 Salinity range (‰) 35.0-36.2 33.8-34.8 34.8-35.0 34.64-34.72 Oxygen-18 range (‰ SMOW) 0.26-0.54 -0.29-0.18 0.05-0.31 -0.27-0.04 Oxygen-18 mean value (‰ SMOW) 0.36 0.00 0.20 -0.12 Oxygen-18 standard deviation (‰ SMOW) 0.08 0.09 0.06 0.07 Oxygen-18 number of samples 19 34 64 77

Table 2. Temperature, salinity and oxygen-18 characteristics of Mediterranean Water (MW), Antarctic Intermedi- ate Water (AAIW), North Atlantic Deep Water (NADW) and Antarctic Bottom Water (AABW). The temperature and salinity ranges are as defined by Emery and Meincke (1986). The oxygen-18 statistics is computed for the serial points from all GEOSECS stations in the Atlantic that fall into these temperature-salinity ranges.

Fig. 10. Benthic foraminiferal oxygen isotope composition for the present Atlantic Ocean in units of parts per mil δ versus PDB (normalized to Uvigerina). Contour interval is 0.25 ‰. The c data are compiled from 169 core sites. Some data are unpublished, but most of them are taken from the literature (Broecker 1986; Birchfield 1987; Curry et al. 1988; Zahn and Mix 1991; Labeyrie et al. 1992; Bickert 1992; McCorkle and Keigwin 1994). 676 Paul et al.

δ Fig. 11. Equilibrium fractionation c computed from the paleotemperature equation of Erez and Luz (1983). (a) Tem- perature taken from Experiment B. (b) Temperature taken from climatology. Contour interval is 0.25 ‰. Simulation of Oxygen Isotopes in a Global Ocean Model 677

(cf. Figs. 6-8). Experiments A and B differ mainly 1 in the δ values of AAIW and AABW. In com- Fδw = − δ P−E ()P − E , (21) w ρ δ 0 parison to the GEOSECS data (Fig. 12a) the w of AABW is too heavy in Experiment A, but turns out δ to be too light in Experiment B, at least at station where S0 = 34.8 is the reference salinity and P-E 85 in the Southern Ocean. is the isotope content of the net surface freshwater flux. From Eqs. (20) and (21) we can derive the following expression for δ : Isotope Content of the Net Surface P-E Freshwater Flux F The surface salt and δ fluxes that are calculated δw w δ P−E = −S0 (22) from Eq. (19) can be related to the net surface fresh FS water flux P - E through δ Thus a slope of 0.5 (i.e. F w:FS = 1) would cor- respond to a δ value of the net surface freshwa- S0 w Fs = − ()P − E , (20) ter flux of -17.4 ‰, which is the value assumed by ρ0 δ Mikolajewicz (1998). The P-E distributions shown in Fig. 13 reveal a rich meridional structure with

Fig. 12. (a) Combined T-S-δ18O diagrams. (a) GEOSECS data. 678 Paul et al.

Fig. 12. (b, c) Combined T-S-δ18O diagrams. (b) Experiment A. (c) Experiment B. Simulation of Oxygen Isotopes in a Global Ocean Model 679 high positive values in the subtropics, where evapo- the resulting hydrography of the present-day wa- ration dominates over precipitation, small negative ter masses and compares it to the observations. The values in the equatorial trough region and high nega- regional means display the same problem as the tive values in the mid- and high latitudes, where zonal mean distributions shown in Figs. 6a, Fig. 7a: precipitation dominates over evaporation. Excess The entire ocean is too warm, except for the bot- evaporation concentrates the less volatile isotopic tom North Atlantic Ocean and Pacific Oceans. In 1 18 molecule H2 O along with salinity, in agreement the North Atlantic Ocean in particular the AABW δ with the positive P-E values (Craig and Gordon influence is too strong. The ocean as a whole is also 1965). too fresh. The prescribed surface δ18O field shows me- The present ocean model is best compared with ridional gradients across the boundaries of the dif- the case H1 ocean model of Danabasoglu and ferent regions given in Table 1, in particular in the McWilliams (1995), which exhibits a very similar North Atlantic at 25°N and in the Southern Ocean. depth distribution of the deviation from climatology δ These meridonal gradients are reflected in the w (note that in Fig. 7 we show the errors for the At- surface fluxes. But since the restoring is not very lantic Ocean that are somewhat larger than those stiff (t = 50 d), they are smeared out in the upper- for the global ocean). The only exception is the most model layer by advection and diffusion proc- AABW that is slightly too cold in the present ocean δ esses, such that the simulated surface w fields look model and slightly too warm in the case H1 ocean rather smooth (Fig. 14). In comparison to Experi- model. The reason for this is probably that ments A, Experiment B shows a less depleted net Danabasoglu and McWilliams (1995) did not em- surface fresh water flux in the Southern Ocean, ploy a strong temperature restoring under sea ice. δ along with a much smaller meridional gradient. We find that a surface w forcing field can be generated such that a global ocean model reproduces the GEOSECS δ data fairly well, both along Discussion w cross sections in the Atlantic and Pacific oceans The advantage of restoring boundary conditions and in terms of regional means. The too light δ over full flux boundary conditions is that the ocean bottom water w values in Experiment A partly model can be run to equilibrium without flux adjust- reflect the too strong AABW influence and partly δ ments. In a sense our approach is complementary result from the too light surface w values in the to the one of Schmidt (1998). Table 3 summarizes Southern Ocean. Experiment B in comparison to

Water depth (km) Temperature (°C) Salinity (‰) δw (‰ SMOW) Model Data Model Data Model Data North Atlantic > 1 4.95 34.87 34.93 0.28 0.24 2-4 4.66 2.9 34.86 34.92 0.29 0.26 > 4 1.99 2.4 34.60 34.89 0.07 0.21 Pacific 2-4 2.75 1.7 34.44 34.67 -0.04 0.00 > 4 1.08 1.1 34.41 34.70 -0.07 -0.01 Globe Whole-basin mean 5.34 34.59 34.74 0.08 0.08

δ Table 3. Hydrography of present-day water masses. The observed temperature, salinity and w are taken from GEOSECS as given in Zahn and Mix (1991) and Labeyrie et al. (1992). 680 Paul et al.

δ Fig. 13. Isotope content P-E of the annual mean net surface fresh water flux. (a) As diagnosed from Experiment A. (b) As diagnosed from Experiment B. Simulation of Oxygen Isotopes in a Global Ocean Model 681

δ Fig. 14. Annual mean surface w. (a) As simulated in Experiment A. (b) As simulated in Experiment B. 682 Paul et al.

δ Experiment A shows the effect of a homogenous of c below 2 km depth in the present-day Atlantic δ surface w distribution in the Southern Ocean on Ocean is an important feature in itself that must be δ δ the surface w fluxes and the zonal mean w dis- simulated correctly before the wealth of isotope δ tribution. The zero slope of the w -S relationship data can be interpreted in terms of water mass indirectly accounts for the sea-ice effect. As a changes. δ result, the bottom water w values are relatively In the present-day North Atlantic, water masses δ more enriched, which illustrates that w can be deeper than 1 km have considerably higher salin- δ used to constrain the role of sea-ice in AABW ity and w than in the global ocean. This salinity and δ formation (cf. Section 1). The apparantly better fit c excess persisted during glacial times, although δ of Experiment B to the GEOSECS w data results NADW is probaly produced at a lower rate (Boyle from a partial cancellation of errors in the repre- and Keigwin 1987, Duplessy et al. 1988). Zahn and δ sentation of AABW. Mix (1991) find that the LGM c values from deep δ18 The effect of the warm bias on Oc is depicted water Atlantic sites (between 2 and 4 km depth) in Fig. 11. In the upper 2 km where this bias is larg- are on average 0.2 ‰higher than those from δ est the c values are too light by about 0.5 ‰. Atlantic Bottom Water sites (below 4 km depth) - δ Below 2 km depth the simulated c values exhibit see the depth distributions of benthic foraminiferal a vertical gradient that is seen neither in the δ18O of Zahn and Mix (1991) (their Fig. 6) and of GEOSECS nor in the benthic foraminiferal δ18O (Labeyrie et al. 1992) (their Fig. 3-b). A meridi- δ data. In order to simulate this data successfully, the onal cross section shows that the deep water c bias towards warm temperatures must be reduced maximum is most pronounced north of 10°N (Zahn significantly. and Mix 1991, Fig. 7). It can be traced through the A warm bias and a diffuse thermocline are two north-east Atlantic up to Rockall Plateau where it δ widespread problems in ocean modelling, which can is reflected in high c values between 1.4 and 2.3 probably be overcome by using diffusion of trac- km depth (Jung 1996). δ ers along isopycnals rather than the physically un- The LGM c values from Pacific Deep Water justifiable horizontal diffusion of tracers sites are similar to those from Atlantic Bottom (Danabasoglu and McWilliams 1995). The formu- Water sites. Zahn and Mix (1991) conclude that if lation by Gent and McWilliams (1990) includes an the measured gradients could be taken at face δ additional tracer transport by mesoscale eddies, value, and if the slope of the w -S relationship were which also eliminates the effect of the Deacon cell assumed to be larger at the LGM than at present, on the zonal mean tracer distributions. Finally, the then Atlantic bottom waters and Pacific deep wa- ACC transport through Drake Passage can prob- ters would be allowed to have a common source, ably take a more realistic value due to the isopycnal again in the Southern Ocean. δ form stress that supports more drag with weaker A clear c signal can only be seen in the coldest flow. water at the very bottom of the glacial Atlantic The isopycnal transport parameterization taken Ocean. If the zone of mixing between NADW and by itself still leads to the formation of too cold and AABW rises by a few hundred meters, this can fresh AABW which can possibly be corrected for probaly not be detected in the depth distribution of δ by a more realistic heat flux parameterization c. Therefore it would be desirable to incorporate (Large et al. 1997) or coupling to an atmospheric a tracer such as δ13C into the ocean model that energy balance model and a sea ice ocean model could reveal additional information on the vertical (Fanning and Weaver 1996; Mikolajewicz 1998). structure of the deep ocean. The equilibrium fractionation of the oxygen iso- In computing the equilibrium fractionation δ δ18 tope composition of calcite c is a good water mass of Oc we used the paleotemperature equation δ tracer above 2 km depth. Below 2 km depth the c of Erez and Luz (1983), which is calibrated values attain a nearly constant level in the present- between 14°C and 30°C with cultured planktic day Atlantic Ocean and do not allow the separa- foraminifera. Zahn and Mix (1991) found that this tion of NADW and AABW. The near constancy paleotemperature equation gives the best fit to the Simulation of Oxygen Isotopes in a Global Ocean Model 683 benthic foraminferal δ18O values from northeast restoring boundary condition on salinity. For this Atlantic core tops at water depths >2 km. Many reason, and for paleostudies of more different pe- empirically derived paleotemperature equations riods than the LGM, the ocean model must prob- have been published during the past several dec- ably be coupled to a prognostic sea-ice model, or ades (Bemis et al. 1998). In particular, Shackleton to both a simple atmosphere and a prognostic sea- (1974) used core top data to calibrate a δ18O-tem- ice model. perature relationship for the benthic foraminifer A promising application beyond the glacial equi- Uvigerina. At low temperatures it yields larger iso- librium state is the deglaciation phase with rapidly δ18 δ topic differences between calcite and water Oc changing c values. Lehman et al. (1993) (in a two- δ18 - Ow than the one by Erez and Luz (1983). In dimensional ocean model) and Mikolajewicz (1998) contrast, Bemis et al. (1998) derived a new (in a three-dimensional ocean model) indeed em- δ paleotemperature equation that is also based on ployed an idealized w tracer in order to compare cultured planktic foraminifera and which at low meltwater-induced changes in the thermohaline temperatures yields smaller isotopic differences circulation with the geologic record. (cf. their Fig. 5). A smaller isotope fractionation δ effect corresponds to a saturation of c at a larger Conclusions depth and vice versa. However, our conclusions would not change if we had chosen any of the other In conclusion, δ18 paleotemperature equations. In particular, the strik- · a surface Ow forcing field can be gener- ing difference between Fig. 11a and b would not ated such that a global ocean model reproduces the δ18 be eliminated. With our choice of the GEOSECS Ow data fairly well, both along cross paleotemperature equation, these figures are di- sections in the Atlantic and Pacific oceans and in rectly comparable to Figs. 4 and 7 of Zahn and Mix terms of regional means, (1991). · a successful simulation of benthic fora- δ δ18 Figure 13 shows the P-E distribution, which miniferal O, however, requires a correction of reveals a rich meridional structure, quite different the warm bias of the present global ocean model, from the constant value assumed by Lehman et al. possibly by using an isopycnal transport para- (1993) and Mikolajewicz (1998). meterization, Our ultimate goal remains a simulation of the δ18 · Ow can be used to characterize different paleocean and an interpretation of the wealth of water masses, isotope data in terms of water mass changes. In · but since δ18O does not resolve the zone of the case of the LGM the proxy data coverage may c be dense enough to generate the surface forcing mixing between NADW and AABW, it would be δ13 fields from a combination of SST reconstructions, desirable to employ another tracer such as C planktic foraminiferal δ18O data and atmospheric that could reveal additional information on the ver- model output (e.g. Herterich et al. this volume). In tical structure of the deep paleocean, any case, the simulated equilibrium fractionation of · for paleostudies in general, the global ocean δ model must probably be coupled to simple atmos- the oxygen isotope composition of calcite c can be directly compared with the benthic foraminiferal phere and sea-ice models, such that neither SSS δ18 δ18 O data. Thus the simulation of oxygen isotopes nor surface Ow need to be prescribed and the δ18 in an ocean model has the potential to investigate use of present-day Ow-salinity relationships can and validate circulation regimes very different from be avoided. the present one. There is of course a problem in the calculation δ δ Acknowledgments of paleosalinity from c. It requires linear w -salinity relationships that are temporally invariant over We thank Dr. G. Edward Birchfield and Dr. long (geological) timescales (see Section 1). Hence Christopher D. Charles for their help during the upon going to the past it is desirable to avoid the early stages of this work. We are grateful to Dr. 684 Paul et al.

Harmon Craig for providing us with the GEOSECS Paleoceanography 4: 207-212 δ18 Bryan K (1984) Accelerating the convergence to equi- Ow data. We also thank Dr. Martina Pätzold for δ18 librium of ocean climate models. J Phys Oceanogr supplying their unpublished Ow data and Sven Stregel for preparing Fig.1. Figs. 2, 4-11 and 13-14 14: 666-673 were produced with the help of FERRET program, Bryan K, Lewis LJ (1979) A water mass model of the world ocean circulation. J Geophys Res 84: 2503-2517 which is a product of NOAA/PMEL. Fig. 12 was Cane MA (1986) Introduction to ocean modeling. In: prepared using GMT, i.e. the Generic Mapping Tool O’Brien JJ (ed) Advanced Physical Oceanographic developed by Dr. Paul Wessel and Dr. Walter Modelling. NATO ASI Series C 186, Reidel, Smith. A. P. is very grateful to Ron Pacanowski Dordrecht, pp 5-21 who shared with him the latest version of the MOM Craig H (1961) Standard for reporting concentrations of code and helped him to compile and run it at the deuterium and oxygen-18 in natural waters. Science University of Bremen. Thanks also to the NCAR 133: 1833-1834 oceanography section and to Nancy Norton in par- Craig H, Gordon LI (1965) Isotopic oceanography: Deu- ticular who provided the NCAR ocean model terium and oxygen-18 variations in the ocean and the (NCOM) bathymetry and surface forcing fields. marine atmosphere. In: Tongiorgi E (ed) Proceedings of the Third Spoleto Conference. Consiglio Finally he thanks Dr. Andreas Manschke for his Nazionale Delle Ricerche, pp 9-130 continued technical support. We acknowledge the Curry WB, Duplessy JC, Labeyrie LD, Shackleton NJ thorough reviews by Dr. Uwe Mikolajewicz and (1988) Changes in the distribution of δ13C of deep Dr. Thomas Stocker that improved the original water CO2 between the last glaciation and the manuscript very much. This research was holocene. Paleoceanography 3: 317-341 funded by the Deutsche Forschungsgemeinschaft Danabasoglu G, McWilliams JC (1995) Sensitivity of the (Sonderforschungsbereich 261 at Bremen Univer- global ocean circulation to parameterizations of sity, Contribution No. 198). Data are available un- mesoscale tracer transports. J Climate 8: 2967-2987 der www.pangaea.de/Projects/SFB261. Danabasoglu G, McWilliams JC, Large WG (1996). Approach to equilibrium in accelerated global oceanic models. J Climate 9: 1092-1110 References Duplessy JC, Shackleton NJ, Fairbanks RG, Labeyrie LD, Bauch D (1995) The Distribution of δ18O in the Arctic Oppo D, Kallel N (1988) Deepwater source variations Ocean: Implications for the Freshwater Balance of during the last climatic cycle and their impact on the the Halocline and the Sources of Deep and Bottom global deepwater circulation. Paleoceanography 3: Waters. Ber Polarforsch Bremerhaven 159, pp 1-144 343-360 Bemis BE, Spero HJ, Bijma J, Lea DW (1998) Duplessy JC, Labeyrie LD, Juillet-Leclerc A, Maitre F, Reevaluation of the oxygen isotopic composition of Duprat J, Sarnthein M (1991) Surface salinity recon- planktonic foraminifera: Experimental results and struction of the North Atlantic Ocean during the last revised paleotemperature equations. Paleocea- glacial maximum. Oceanol Acta 14: 311-324 nography 13: 150- 160 Emery WJ, Meincke J (1986) Global water masses: sum- Bickert T (1992) Rekonstruktion der spätquartären mary and review. Oceanol Acta 9: 383-391 Bodenwasserzirkulation im Östlichen Südatlantik Erez J, Luz B (1983) Experimental paleotemperature equa- über stabile Isotope benthischer Foraminiferen. Ber tion for planktonic foraminifera. Geochim Cosmochim Fachber Geowiss Univ Bremen 27, pp 1-205 Acta 47: 1025-1031 Birchfield GE (1987) Changes in deep-ocean water δ18O Fairbanks RG, Charles CD, Wright JD (1992) Origin of and temperature from the last glacial maximum to the global meltwater pulses. In: Taylor RE (ed) Radio- present. Paleoceanography 2: 431-442 carbon after four decades. Springer, Berlin Boyle EA, Keigwin L (1987) North Atlantic thermohaline Heidelberg New York, pp 473-500 circulation during the last 20,000 years: Link to high- Fanning AF, Weaver AJ (1996) An atmospheric energy- latitude surface temperature. Nature 330: 35-40 moisture balance model: climatology, interpentadal Broecker WS (1986) Oxygen isotope constraints on climate change, and coupling to an ocean general surface ocean temperatures. Quat Res 26: 121-134 circulation model. J Geophys Res 101 (D10): Broecker WS (1989) The salinity contrast between the 15,111-15,128 Atlantic and Pacfic oceans during glacial time. Simulation of Oxygen Isotopes in a Global Ocean Model 685

Garzoli SL, Gordon AL (1996) Origins and variability of Juillet-Leclerc A, Jouzel J, Labeyrie LD, Joussaume S the Benguela Current. J Geophys Res 101: 897-906 (1997) Modern and last glacial maximum sea surface Gat JR (1996) Oxygen and Hydrogen isotopes in the δ18O derived from an Atmospheric General Circula- hydrologic cycle. Ann Rev Earth Planet Sci 24: tion Model. Earth Planet Sci Lett 146: 591- 605 225-262 Jung SJA (1996) Wassermassenaustausch zwischen Gill AE (1982) Atmosphere-Ocean Dynamics. Interna- NE-Atlantik und dem Nordmeer während der letzten tional Geophys Series 20, Academic Press, San 300000/80000 Jahre im Abbild stabiler O- und C-Iso- Diego tope. PhD Thesis, Univ Kiel, pp 1-104 Gent PR, McWilliams JC (1990) Isopycnal mixing in Kalnay E, Kanamitsu M, Kistler R, Collins W, Deaven ocean circulation models. J Phys Oceanogr 20: D, Gandin L, Iredell M, Saha S, White G, Woolen J, 150-155 Zhu Y, Leetma A, Reynolds R, Chelliah M, Ebisuzaki Gonfiantini R (1978) Standards for stable isotope meas- W, Higgins W, Janowiak J, Mo KC, Ropelewski C, urements in natural compounds. Nature 271: 534-536 Wang J, Jenne R, Joseph D (1996) The NCEP/NCAR Gordon AL, Lutjeharms JRE, Gründlingh ML (1987) 40-year reanalysis project. Bull Amer Meteor Soc 77: Stratification and circulation at the Agulhas Retro- 437-471 flection. Deep-Sea Res 34: 565-599 Knauss J (1969) A note on the transport of the Gulf Gould WJ, Loynes J, Backhaus J (1985) Seasonality in Stream. Deep-Sea Res 16: 117-124 slope current transports northwest of Shetland. Labeyrie LD, Duplessy JC, Duprat J, Juillet-Leclerc A, Technical Report 7, ICES, Hydrography Commitee, Moyes J, Michel E, Kallel N, Shackleton NJ (1992) pp 1-13 Changes in the vertical structure of the North Hoffmann G (1995) Stabile Wasserisotope im allgemeinen Atlantic ocean between glacial and modern times. Zirkulationsmodell ECHAM. Dissertation, Univ Quat Sci Rev 11: 401-413 Hamburg, pp 1-99 Landau LD, Lifschitz EM (1987) Fluid Mechanics (2nd Hoffmann G, Heimann M (1993) Water tracers in the ed). Pergamon, Oxford, pp 1-539 ECHAM general circulation model. In: International Large WG., Danabasoglu G, Doney SC, McWilliams JC Atomic Energy Agency (ed) Isotope techniques in (1997) Sensitivity to surface forcing and boundary the study of past and current environmental layer mixing in a global ocean model: Annual-mean changes in the hydrosphere and the atmosphere climatology. J Phys Oceanogr 27: 2418-2447 Jacobs SS, Fairbanks RG, Horibe Y (1985) Origin Larsen JC (1992) Transport and heat flux of the Florida and evolution of water masses near the antarctic Current at 27° N derived from cross-stream voltages continental margin. In: Oceanology of the Antarctic and profiling data: Theory and observations. Philos Continental Shelf, Ant Res Ser 43: 59-85 Trans R Soc London A338: 169-236 Joussaume S, Jouzel J (1993) Paleoclimatic tracers: an Lehman SJ, Wright DG, Stocker TF(1993) Transport of investigaton using an Atmospheric General Circu- freshwater into the deep ocean by the conveyor. In: lation Model under Ice Age conditions. 2. Water Peltier WR (ed) Ice in the Climate System. NATO ASI isotopes. J Geophys Res 98: 2807-2830 Series I: Global Environmental Change, Vol 12, Joussaume S, Sadourny R, Jouzel J (1984) A general cir- Springer, Berlin Heidelberg New York, pp 187-209 culation model of water isotope cycles in the atmos- Levitus S (1982) Climatological Atlas of the World phere. Nature 311: 24-29 Ocean. NOAA Prof Pap 13, pp 1-173 Jouzel J (1986) Isotopes in cloud physics: Multiphase Levitus S, Boyer TP (1994) World Ocean Atlas. Volume and multistage condensation processes. In: Fritz P, 4: Temperature. NOAA Atlas NESDIS 4, pp 1-117 Fontes JC (eds) The Terrestrial Environmont, Hand- Levitus S, Russell B, Boyer TP (1994) World Ocean book of Environmental Isotope Geochemistry. Atlas. Volume 3: Salinity. NOAA Atlas NESDIS 3, Elsevier, New York, 2, pp 61-112 pp 1-99 Jouzel J, Russel GL, Suozzo RJ, Koster RD, White JWC, Mackensen A, Hubberten HW, Scheele N, Schlitzer R δ Σ Broecker WS (1987) Simulations of the HDO and (1996) Decoupling of 13C CO2 and phosphate 18 H2 O atmospheric cycles using the NASA GISS in Recent Weddel Sea deep and bottom water: General Circulation Model: the seasonal cycle Implications for glacial Southern Ocean for present-day conditions. J Geophys Res 92: paleoceanography. Paleoceanography 11: 203-215 14,739-14,760 686 Paul et al.

Mathieu R (1996) GENESIS Σ. A Water Molecule Stable Shackleton NJ (1974) Attainment of isotopic equilibrium Isotopes and Tracers Version of the GENESIS Glo- between ocean water and the benthonic foraminifera bal Climate Model Version 2.0. Model Description. genus Uvigerina: Isotopic changes in the ocean Interdisciplinary Climate System Section, CGD, during the last glacial. Colloq Int Centr Nat Rech Sci NCAR, Boulder, Colorado. Institute of Alpine and 219: 203-209 Arctic Research, Univ Colorado, Boulder, pp 1-19 Shea DJ, Trenberth KE, Reynolds RW (1990) A global McCorkle DC, Keigwin LD (1994) Depth profiles of δ13C monthly sea surface temperature climatology. NCAR in bottom water and core top C. wuellerstorfi on the Technical Note NCAR/TN- 345, NCAR, Boulder, Ontong Java Plateau and Emperor Seamounts. Colorado, pp 1-167 Paleoceanography 9: 197-208 Stössel A, Oberhuber JM, Maier-Reimer E (1996) On the McGuffie K, Henderson-Sellers A (1996) A Climate Mod- representation of sea ice in global ocean general elling Primer (2nd ed). J Wiley & Sons, Chichester, circulation models. J Geophys Res 101 (C8): pp 1-253 18,193-18,212 Mikolajewicz U (1998) A meltwater induced collapse of Stramma L, Lutjeharms JRE (1996) The flow field of the ‘conveyor belt’ thermohaline circulation and its the subtropical gyre of the South Indian Ocean. influence on the distribution of 14C and δ18O in the J Geophys Res 102: 5513 oceans. J Geophys Res, in press Toggweiler JR, Samuels B (1995) Effect of sea ice on the NCAR Data Support Section (1986) NGDC ETOPO5 glo- salinity of antarctic bottom waters. J Phys Oceanogr bal ocean depth & land elevation, 5-min. Data Set 25: 1980-1997 759.1, NCAR, Boulder, Colorado, pp Toggweiler JR, Dixon K, Bryan K (1989). Simulations of NCAR Oceanography Section (1996) The NCAR CSM radiocarbon in a coarse-resolution World Ocean ocean model. NCAR Technical Note NCAR/TN- model. 1. Steady state prebomb distributions. 423+STR, NCAR, Boulder, Colorado J Geophys Res 94: 8217-8242 Östlund HG, Craig H, Broecker WS, Spencer D (1987) UNESCO (1981) Tenth report of the joint panel of ocea- GEOSECS Atlantic, Pacific, and Indian ocean expe- nographic tables and standards. UNESCO Techni- ditions. Shorebased data and graphics. GEOSECS cal Papers in Marine Sciences 36, UNESCO, Paris Atlas Series, US Govt Printing Office, Washington Veronis G (1973) Large scale ocean circulation. Adv Appl DC, 7, pp 1-200 Mech 13 Pacanowski RCE (1996) MOM 2. Documentation, Us- Veshteyn VY, Malyuk GA, Rusanov VP (1974) Oxygen- er’s Guide and Reference Manual. Technical Report 18 distribution in the Central Arctic Basin. Oceanol- 3.2, GFDL Ocean Group, GFDL, Princeton, New ogy 14: 514-519 Jersey, pp 1-329 Wajsowicz RC (1993) A consistent formulation of the Peterson RG, Stramma L (1991) Upper-level circulation anisotropic stress tensor for use in models of the in the South Atlantic Ocean. Progr Oceanogr 26: large scale ocean circulation. J Comp Phys 105: 1-73 333-338 Roemmich D, Wunsch C (1985) Two transatlantic sec- Washington WM, CL Parkinson (1986) An Introduction tions: Meridional circulation and heat flux in the to Three- Dimensional Climate Modeling. University subtropical North Atlantic Ocean. Deep-Sea Res 32: Science Books, Mill Valley, California 619-664 Weaver AJ, Hughes TMC (1996) On the incompatibility Rohling EJ, Bigg GR (1998) Paleosalinity and δ18O: A of ocean and atmosphere models and the need for critical assessment. J Geophys Res 103: 1307-1318 flux adjustments. Clim Dyn 12: 141-170 Schmidt GA (1998) Oxygen-18 variations in a global Weiss RF, Östlund HG, Craig H (1979) Geochemical stud- ocean model. Geophys Res Lett 25: 1201-1204 ies of the Weddell Sea. Deep-Sea Res 26A: 1093-1120 Schotterer U, Oldfield F, Frölich K (1996) Global network Whitworth T III, Peterson RG (1985) Volume transport for isotopes in precipitation. IAEA, Vienna, pp 1-48 of the Antarctic Circumpolar Current from bottom Semtner A (1986) Finite-difference formulation of a pressure measurements. J Phys Oceanogr 15: World Ocean model. In: O’Brien JJ (ed) Advanced 810-816 Physical Oceanographic Modelling. NATO ASI Zahn R, Mix AC (1991) Benthic foraminiferal δ18O in the Series C 186, Reidel, Dordrecht, pp 187-202 ocean’s temperature-salinity-density field: constraints on the ice age thermohaline circulation. Paleoceanography 6: 1-20