In a modeling sensitivity study we investigate the evolution of the ocean circulation and of marine carbon isotope (δ13C) records during the Miocene (about 23–5 million years ago). For this purpose we ran an ocean-circulation carbon cycle model of intermediate complexity (Large Scale Geostrophic– Hamburg Ocean Carbon Cycle Model, version 2s) exploring various seaway configurations. Our investigations confirm that the Central American Seaway played a decisive role in the history of the Miocene ocean circulation. In simulations with a deep Central American Seaway (depth range 1–3 km), typical for the early to middle Miocene, deep water production in the North Atlantic is absent or weak, while the meridional overturning circulation is dominated by water mass formation in the Southern Ocean. Deep water formation in the North Atlantic begins when the Central American Seaway shoals to a few hundreds of meters, which is typical for the late Miocene. Our results do not support ideas that the mid-Miocene closing of the Eastern Tethys contributed to Antarctic glaciation. On the other hand, we find some water exchange between the Indian Ocean and the Atlantic via the Eastern Tethys during the early Miocene. Our model results for the Atlantic meridional overturning circulation and for Atlantic δ13C during the late Miocene are largely independent from depth variations of the Greenland-Scotland Ridge. To a large extent, the evolution of Miocene deep-sea δ13C records can be explained with large-scale ocean circulation changes. Our model-data comparison for the middle and early Miocene suggests that during the early Neogene the seaway effect on benthic δ13C may have been superimposed by further factors such as climate regime shifts and/or terrestrial carbon cycle changes.
 The Miocene epoch is characterized by a complex evolution of Earth's climate with respect to ocean circulation. Although the continents were already close to their modern position, lithosphere plate movements led to changes in the configuration of interoceanic passages which significantly influenced the ocean circulation. The Central American Seaway narrowed from 17 Ma on until its final closure around 2.7 Ma [Duque-Caro, 1990; Coates et al., 2003; Bartoli et al., 2005; Steph et al., 2006]. Decisive for the interoceanic deep water exchange was the emergence of the volcanic arc south of Central America and its collision with northern Columbia between 12.8 and 7.1 Ma [Coates et al., 2003], which hindered a deep water exchange between the Pacific and the Atlantic. The Eastern Tethys narrowed between 18 and 16 Ma but reopened afterward north of the Arabian Plate and finally closed around 14 Ma [Rögl, 1999; Meulenkamp and Sissingh, 2003]. The Fram Strait widened since 17.5 Ma to a passage deeper than 2000 m at 13.7 Ma [Jakobsson et al., 2007, Jokat et al., 2007]. The Greenland-Scotland Ridge deepened between 18 and 15.5 Ma, and again from 12.5 Ma on [Wright and Miller, 1996], which perhaps initiated major deep water flow from the Nordic Seas into the North Atlantic at 11.5 Ma [Wei and Peleo-Alampay, 1997].
 Erosional features south of the Denmark Strait like hardgrounds and glaucony rip-up clasts have been cited to indicate intensified bottom water currents related to such an overflow. However, high sediment accumulation rates on the Eirik and Garda Drifts south of Greenland suggest a major overflow of bottom water through the Denmark Strait not prior to 8 or 7 Ma [Wold, 1994]. On the other hand, drift sediments from the deep Northeast Atlantic indicate a major dense water flow over the Iceland-Scotland Ridge already at about 17 Ma (Björn and Garda Drifts) and a compensating northward flow at 14 Ma (Hatton and Snorr Drifts) [Wold, 1994].
 Major patterns of the Miocene deep water circulation have first been drawn by Woodruff and Savin  based on a global set of benthic carbon (δ13C) and oxygen (δ18O) isotope records as well as on foraminifer faunal counts from a set of deep-sea drilling cores. A major finding was that in the early Miocene intermediate and deep water masses in the Atlantic and Pacific apparently aged in a northward direction with a common source somewhere around Antarctica. There was no evidence for the formation of North Atlantic Deep Water (NADW) prior to about 14.5 Ma. Increasing δ13C values indicated some proto-NADW production during the middle Miocene which intensified during the late Miocene. With a more detailed investigation, this study was refined by Wright et al.  who could show that there was evidence for some NADW production between 20 and 16 Ma, and then from 12.5 Ma on.
 Opposite to the results from benthic δ13C data, Frank et al.  concluded from neodymium (ɛNd) and lead isotope time series in the northwest and southeast Atlantic that a continuous and strong NADW export persisted throughout the interval 14 to 3 Ma until the onset of Northern Hemisphere glaciation. However, the sampling by Frank et al.  is probably insufficient to properly trace NADW export across the equator. A more recent study interpreting ɛNd data from a depth transect on Walvis Ridge suggests that deep water stratification in the Atlantic changed between 10.6 and 7.3 Ma [Thomas and Via, 2007]. A potential cause is seen in the onset of deep convection in the Labrador Sea and the subsequent export of ɛNd-depleted water to the South Atlantic. The characteristic low-ɛNd signature of Labrador Sea Water results from the delivery of nonradiogenic neodymium which increases with glacial mechanical weathering. Deep convection and glacial weathering point to significant cooling of South Greenland during late Miocene but a permanent ice cover on Greenland is not recognized before 7.3 Ma [St. John and Krissek, 2002]. The inconsistencies indicate that the kinematic information which can be inferred from Neogene ɛNd records is equivocal.
 Two modeling studies have considered the sedimentary response to Central American Seaway changes [Heinze and Crowley, 1997; Heinze and Dittert, 2005]. Heinze and Crowley  focus on carbonate dissolution for a deep Central American Seaway and find significant lysocline shoaling in the North Atlantic while calcite saturation levels in the Pacific deepen. Studying opal deposition, Heinze and Dittert  show that Central American Seaway closing may be the reason for a major redistribution of biogenic silica in the world ocean.
 Here, we present results of a combined ocean-circulation carbon cycle modeling study in which we mimicked the evolution of environmental conditions during the Miocene in a series of sensitivity experiments with characteristic seaway configurations and sea ice distributions. The model is described in section 2, and the Miocene scenarios are introduced in section 3. In section 4 we explore the response of the meridional overturning circulation (MOC). The results are then employed to model marine δ13C (section 5). In this way our circulation fields can be readily evaluated by comparison with sediment records, which is an advantage compared to previous circulation modeling studies. The synthesis is a conceptual view on Miocene ocean circulation changes (section 6).
2. Model Description
 We study the ocean circulation using an updated version of the Hamburg Large Scale Geostrophic (LSG) circulation model (developed by Maier-Reimer et al. ). Major improvements are a new advection scheme for tracers [Schäfer-Neth and Paul, 2001; Prange et al., 2003] as well as an overflow parametrization for the bottom boundary layer [Lohmann, 1998; Lohmann and Schulz, 2000]. The effective horizontal resolution is 3.5° on an Arakawa-E grid while the vertical resolution is 22 levels (centered at 25, 75, 125, 175, 225, 275, 350, 450, 550, 650, 750, 850, 950, 1100, 1300, 1500, 1800, 2250, 2750, 3500, 4500, and 5500 m). We calibrated our circulation model by simulations of anthropogenic Δ14C [Butzin et al., 2005]. The ocean is forced by 10 year averaged monthly fields of wind stress, surface air temperature, and freshwater flux. These quantities serve as background climatology and originate from present-day climate simulations with the atmosphere general circulation model ECHAM3/T42 [Roeckner et al., 1992]. Based on atmospheric energy balance model considerations we apply a surface heat flux formulation which permits that sea surface temperatures (SST) can freely adjust to ocean circulation changes [Rahmstorf and Willebrand, 1995; see also Prange et al., 2003; Butzin et al., 2005]. The hydrological cycle is closed by a scheme which calculates catchment areas for continental runoff and allows for variable land-sea distributions, which permits that sea surface salinities (SSS) can freely evolve. An implicit method for the integration of the momentum equations permits a time step of one month, with a total integration time of 10,000 years.
 Marine carbon isotopes are modeled using the Hamburg Ocean Carbon Cycle Model, version 2s (HAMOCC2s) [Heinze and Maier-Reimer, 1999; Heinze et al., 1999], which considers geochemical tracers and biogenic particulate matter in the water column and in the bioturbated sediment. The model treats the dissociation of carbonic acid and the borate buffer [cf. Maier-Reimer and Hasselmann, 1987] as well as particulate organic carbon, calcium carbonate, and opal. Sedimentation is treated following Archer et al. . A ten-layer sediment module accounts for chemical reactions of biogenic particulate matter with pore water, diffusive processes in pore and bottom waters, vertical sediment advection as well as sediment accumulation, and bioturbation [Heinze et al., 1999]. Input of terrigenous matter is prescribed by global-mean present-day weathering fluxes at the sea surface (see Table 1 for a list of numerical values), which are asymptotically approached by the integrated sediment accumulation during the model run. The model is able to diagnose atmospheric carbon dioxide (CO2) as affected by processes in the ocean and predicts a present-day concentration of about 282 ppm for 12CO2. However, to separate the effect of ocean gateway changes on the marine carbon cycle during the Miocene from other influences such as changes in continental weathering (which was different from today) [e.g., Fantle and DePaolo, 2005], the model is restored to this present-day value (which is also in the range of atmospheric 12CO2 concentrations for the Miocene according to marine proxy records [e.g., Pagani et al., 2005; Pearson and Palmer, 2000]) using a time constant of 100 years. HAMOCC2s is driven by annual-mean temperatures, salinities, and velocities provided by the LSG circulation model. The velocity fields for tracer advection are used “off-line,” that is, they are taken as model input without further dynamic computations. The carbon cycle model adopts the spatial resolution of the circulation model but uses a different time step of 1 year. A parametrization for convective mixing retains seasonality effects which would get lost otherwise [Heinze and Maier-Reimer, 1999]. The total integration time for each experiment is 100,000 years.
Table 1. Overview of the Paleogeographical Setup and Some Carbon Cycle Model Parameter Values for the Different Simulationsa
DIC is total dissolved inorganic carbon (value for each DI12C, DI13C, DI14C), TAlk is total alkalinity, and POC is particulate organic carbon (value for each PO12C, PO13C, PO14C). PD is the control run, LM250 and LM500 are late Miocene, MM is middle Miocene, and EM and EMW are early Miocene.
Threshold value for the onset of CaCO3 productionb
Maximum possible export production rain ratio C(CaCO3):C(POC)c
282 ppm, diagnosed
as PD, fixed
as PD, fixed
as PD, fixed
3. Survey of Model Scenarios
 To track the evolution of the Miocene ocean circulation, we consider various seaway configurations which are listed in Table 1. In all simulations there is a three grid boxes wide Central American Seaway separating North and South America between about 11°N and 19°N, and Bering Strait and Hudson Bay are closed [e.g., Gladenkov et al., 2002]. The scenarios are as follows.
 Two experiments, LM250 and LM500, mimic the late Miocene stage of Central American Seaway closure starting from 8 Ma until about 2.7 Ma BP [Coates et al., 2003]. Central American Seaway depths are 250 m in LM250 and 500 m in LM500.
 Experiment MM aims to capture the middle to late Miocene transition of global climate cooling (12–10 Ma BP) and assumes a Central American Seaway depth of 1000 m according to evidence from benthic foraminifer assemblage studies by Duque-Caro . This scenario also comprises a narrow Eastern Tethys (with a width of two grid points in zonal direction) which is reconstructed to be still open during that time interval [Rögl, 1999; Meulenkamp and Sissingh, 2003].
 Two scenarios, EM and EMW, consider the early to middle Miocene (17–15 Ma) situation of warm Neogene climate. Both scenarios include a 3000 m deep Central American Seaway (following Duque-Caro ) and an Eastern Tethys which is wider than in experiment MM (four grid points in zonal direction compared to two in MM; see Figures 7, 8, and 9). Scenario EM employs present-day climate background conditions and sea ice margins, such as in experiments LM250, LM500 and MM. Sensitivity experiment EMW mimics a polar amplification of early Neogene climate warming with ice-free oceans [Harwood and Bohaty, 2000; Moran et al., 2006] by prescribing that surface air background temperatures are everywhere above the freezing point of seawater and that sea ice vanishes. This approach leads to lower bound estimates of true early Neogene surface air temperatures.
4. Evolution of the Large-Scale Ocean Circulation During the Miocene
 Our investigations start from a control integration for present-day conditions (PD) which reflects the modern MOC. Maximum PD volume transports in the Atlantic are 16 Sv for NADW and 4 Sv for Antarctic Bottom Water (1 Sv = 1 × 106 m3/s; see Figures 2a and 3). The control run captures the observed distribution of sea-surface salinities (Figure 4; the root-mean-square difference to observations [e.g., Levitus et al., 1994] is about 0.7 PSU for ice-free areas). The PD equilibrium state is then perturbed by instantaneous application of late Miocene tectonic forcing. Subsequently, the steady state results for the late Miocene serve in analogous fashion as the starting point to the middle Miocene experiments, which in turn provide the initial values for the early Miocene simulations.
 Our Miocene experiments yield significant changes of water mass properties and circulation patterns compared to PD. All simulations lead to distinctive freshening of the upper kilometer of the North Atlantic while salinities increase in the other oceans (compare Figures 5a, 6a, 7a, 8a, and 9a). In experiment EMW the Arctic Ocean and the subantarctic Southern Ocean do not freshen as much as in experiment EM. The amplitude of salinity changes corresponds to MOC shifts. Deep water formation in the North Atlantic slightly weakens in LM250, becomes considerably reduced in LM500, MM and EM, and ceases in EMW under ice-free conditions (Figures 2b–2f). Conversely, formation of deep and bottom waters in the Southern Ocean is substantially enhanced in all Miocene experiments (especially in the Pacific sector; see Figures 3c–3f), except for LM250 which approaches PD values. The MOC changes cause upper-level cooling in the North Atlantic and warming in the South Atlantic (not shown), except for experiment EMW in which high-latitude warming in the upper-level water column and global warming in the deep water layer reflects the absence of sea ice (not shown).
 All simulations show net export of upper-level water from the Pacific to the Atlantic via the Central American Seaway, which occurs at the expense of the Indonesian throughflow into the Indian Ocean (Figures 5–9). Barotropic (integrated) water fluxes through the Central American Seaway into the Atlantic amount 6–12 Sv and reduce the present-day upper-level salinity difference between the North Atlantic and North Pacific. In all experiments we find westward flow of surface water and eastward flow in the thermocline and at intermediate depths (Figures 5–9 and 10). Westward surface fluxes are between 2 Sv (LM250) and 6 Sv (in EMW) across 90°W. Eastward subsurface transports add up to 10–12 Sv, except for LM250 with an eastward subsurface flux of about 5 Sv. In simulations EM and EMW there is also westward deep water flow below 2.2 km of about 1–4 Sv in total (Figure 10).
 Most of the steady state Central American Seaway throughflow into the Atlantic turns southward, leading to reversal of the North Brazil current and to intensification of the Brazil current (Figures 6a, 7a, 8a, and 9a). This contributes to the upper-level maxima of South Atlantic volume transports seen in LM500, MM, and EM (Figures 2c–2e). The exception to these findings is LM250 in which the circulation bears similarity to the PD situation (Figure 5a).
 Interocean water fluxes via the Eastern Tethys Seaway are small. Simulations MM, EM and EMW show northward transport of upper-level water through the Eastern Tethys and southward flux at deeper layers (Figure 11). In MM northward and southward transports amount to about 3 Sv each, and the meridional water flux through the Eastern Tethys is largely balanced. The effect of the Eastern Tethys is then limited to the Indian Ocean where the southward outflow at deeper levels can be found as positive salinity anomaly until about 30°S (Figure 7b, 8b, and 9b). In experiments EM and EMW the meridional flow through the Eastern Tethys is not balanced. About 5 Sv (EM) to 6 Sv (EMW) of upper-level water pass across 26°N in northward direction while southward fluxes at deeper levels add up to about 2 Sv. This implies that 3 Sv (EM) to 4 Sv (EMW) of upper-level water are exported from the Indian Ocean water to the Atlantic. As a consequence, upper-level salinities in the subtropical North Atlantic do not decrease as much as in experiment MM (Figures 7a and 8a), while, conversely, upper-level salinities in the northern Indian Ocean are lower in EM and EMW than in MM.
5. A Model-Data Comparison Using Carbon Isotopes
 The 13C/12C isotope ratio of dissolved inorganic carbon (DIC) decreases along the pathway of a water parcel through the interior of the ocean. This is due to progressive oxidation of organic matter (formed near the sea surface) which is depleted in 13C. As the isotopic signature of DIC of ambient water is also recorded in the calcitic shells of benthic foraminifers, foraminiferal δ13C can be interpreted as a tracer of past ocean ventilation [e.g., Duplessy et al., 1984]. Our preindustrial control simulation employing PD thermohaline and circulation fields agrees well with δ13C observations from the 1990s (measurements by P. Quay, University of Washington, collected in the GLODAP v1.1 bottle data set [Key et al., 2004; Sabine et al., 2005]) and captures the characteristic deep water δ13C differences between North Atlantic and North Pacific (Figures 12a and 12b). In shallow water the field data mirror decreasing atmospheric δ13C values caused by the combustion of fossil fuels [Sonnerup et al., 1999], which is not considered in our preindustrial simulation. Model values in the upper 2 km of the North Pacific are high compared with observations which will be further discussed in section 6.
 For the Miocene, we consider model results along 30°W and 160°E. These meridians intersect Miocene sediment core locations in the Atlantic and in the Pacific (see below). We compare the model results with sediment data [Woodruff and Savin, 1989; Wright et al., 1992; Shackleton and Hall, 1997; Roth et al., 2000; Paulsen, 2005; Poore et al., 2006] (see Table S1 in the auxiliary material). We consider the time slice 17.0–14.8 Ma to represent the early Miocene to early middle Miocene warm interval (scenario EM), the period 12.0–10.5 Ma to represent the middle Miocene transition of global climate cooling (MM), and 7.0–5.3 Ma to display the early establishment of the modern deep ocean circulation (LM500 and LM250). The last time slice is placed between the end of the globally recognized carbon isotope shift [e.g., Bickert et al., 2004; Poore et al., 2006] and the end of the Miocene. In all data sets, ages were converted to the common, astronomically tuned Neogene time scale by Lourens et al. . To be unaffected by changing east-west gradients in the different ocean basins, we extracted records of the western Atlantic and western Pacific sites, only, and calculated the δ13C mean values of the data within the given time interval of each record (Table S1). Short-term variability in the δ13C records like the response to Milankovich forcing is therefore eliminated, but this is the only way to combine high- and low-resolution records in a common data set and to obtain a sufficient number of data points for the chosen time slices. The observations are plotted as dots above the model results shown in Figures 12c–12l.
 The δ13C distribution according to LM250 is qualitatively similar to the PD situation but deep water results are up to 0.7‰ elevated above modern values. Late Miocene sensitivity study LM500 (Figures 12e and 12f) exhibits significant carbon isotope shifts compared to LM250. In the Atlantic, the δ13C of deep water drops by about 0.8‰, and the meridional δ13C gradient reverses. Conversely, deep water δ13C values in the North Pacific increase by up to 0.5‰. Moreover, in LM500 the interocean δ13C gradient between Pacific and Atlantic reverses, and the deep Pacific becomes by up to 0.9‰ enriched above the Atlantic.
 The outcomes of experiment LM250 are largely in line with observations from the late Miocene (Figures 12c and 12d). In the Atlantic, there is a slight mismatch between benthic δ13C observations and both simulations but experiment LM250 better fits with the observations than LM500. In the Pacific, simulation LM250 matches the δ13C observations while the results of LM500 are too high.
 Middle Miocene case study MM shows a further drop of deep water δ13C relative to the late Miocene simulations (Figures 12g and 12h). Compared to LM500, bulk water δ13C values decrease by up to 0.2‰ in the western Atlantic and by up to 0.1‰ in large parts of the western Pacific. An exception to this overall isotopic enrichment is found in the upper 1500 m of the North Pacific where marine δ13C values increase by up 0.4‰. In the tropical Atlantic δ13C decreases to about 0.2‰ which is the lowest value in the Atlantic in all simulations. Apart from the overall isotopic depletion in deep water, the modeled meridional δ13C gradients in the Atlantic and in the Pacific are similar to LM500 in that δ13C decreases in northward direction in the Atlantic while δ13C decreases in southward direction in the east Pacific. A further look at interoceanic δ13C differences reveals that deep water δ13C values in the east Pacific are elevated by about 0.4‰ above the δ13C values in the west Atlantic.
 The results of MM are significantly lower than the observations from the middle Miocene (by up to 1.2‰ in the west Atlantic and up to 0.7‰ in the east Pacific; see Figures 12g and 12h). However, the model qualitatively agrees with sediment data in the Atlantic showing that δ13C values decrease in northward direction, and that δ13C values slightly increase with depth below ∼3.5 km. There are no observations to check the model prediction of δ13C minima at intermediate depths in the Atlantic and Pacific. The model fails to reproduce the sign of the meridional δ13C gradient indicated by sediment records in the east Pacific.
 Simulation EM (Figures 12i and 12j) exhibits a further decrease of deep-sea δ13C by 0.1–0.2‰ compared to MM. Exceptions are the North Atlantic (depth range 0.5–2 km) and the North Pacific (in the upper 1.5 km) where the δ13C values according to EM are up to 0.4‰ higher than in MM. We find minimum δ13C values in the equatorial Atlantic at intermediate depths. Deep and bottom water δ13C model values in the Atlantic tend to decrease in northward direction but the meridional gradient is much less pronounced than in MM and in LM500. In the east Pacific the modeled meridional δ13C gradient is similar to MM and LM500. The interoceanic isotopic differences are higher than in MM, with subsurface δ13C values in the east Pacific being up to 1‰ higher than in the west Atlantic.
 Sensitivity experiment EMW results in the highest δ13C values of all model runs (Figures 12k and 12l). In both oceans δ13C decreases from the south to the north. The meridional, interhemispheric gradient is less pronounced in the Pacific than in the western Atlantic where minimum isotopic ratios are found in the tropics at about 1 km depth. Overall, δ13C values in the North Pacific are about 0.5‰ higher than in the North Atlantic.
 The δ13C distributions obtained in EM are inconsistent with observations from the early Miocene. The model results are about 1‰ lower than the benthic measurements, and the model does not reproduce the meridional δ13C gradient shown by east Pacific sediment data. Moreover, there is no observational evidence for a large carbon-isotopic enrichment of the east Pacific above west Atlantic values. Simulation EMW agrees better with the observations. The absolute model values come close to the benthic measurements which are in the range 1–2‰, and the sign of the meridional δ13C gradient in the east Pacific is also in line with benthic records. There are no observations which could support or disprove the isotopic depletion of the northern west Atlantic compared to the east Pacific.
 Our simulations corroborate the widely accepted idea that the closing of the Central American Seaway had a significant impact on the global ocean circulation, and that the late Miocene to Pliocene strengthening of the Atlantic MOC seems to be closely related to the restriction of water exchange between the Atlantic and the Pacific. Experiments EM and EMW indicate that water exchange via a wide and deep Eastern Tethys may mitigate the effect of the Central American Seaway on the Atlantic MOC and on benthic isotopic patterns. On the other hand, the model-proxy data comparison reveals differences increasing with age, which points to the influence of further factors on Neogene deep-sea carbon isotope records.
 The δ13C paleodata as compiled from literature indicate that the modern ocean circulation started to evolve sometime between ca. 9.0 and 6.5 Ma, which can be seen in the distinct divergence of benthic records from the North Atlantic, the Southern Ocean, and the Pacific (Figure 1). Reconstructed late Miocene deep water temperatures from the western equatorial Atlantic show a colder and fresher early NADW becoming warmer and saltier along with the restriction of the Central American Seaway [Lear et al., 2003], which is in line with our modeling results. The late Miocene evolution of the deep ocean circulation is associated with major climate changes, like the beginning of the glaciation of Southern Greenland, and with a perturbation of the carbon cycle (the so-called late Miocene carbon isotope shift [e.g., Billups, 2002; Bickert et al., 2004]). However, it appears that the modern MOC started much earlier than the northern-hemispheric glaciation, and it is therefore questionable whether the closing of the Central American Seaway played a decisive role in this important cooling phase (see the recent review by Molnar ). Our experiments LM250 and LM500 can be interpreted to characterize different stages of the late Miocene ocean circulation. In simulation LM500 the export of proto-NADW to the South Atlantic is still weak and to some extend reinforced by the southward return flow of Pacific water which entered the Atlantic via the Central American Seaway. A similar intensification of southward flow is also seen in the modeling study by Prange and Schulz . Further shoaling of the Central American Seaway to a sill depth of 250 m leads to quasi modern Atlantic MOC and δ13C patterns.
 The Atlantic MOC according to our middle and early Miocene scenarios resembles that obtained in most previous sensitivity studies featuring a Central American Seaway deeper than 1 km [Maier-Reimer et al., 1990; Mikolajewicz et al., 1993; Murdock et al., 1997]. In contrast, Nisancioglu et al.  arrived at significant NADW formation rates even for a deep Central American Seaway but did not find any significant outflow of deep and bottom waters from the Southern Ocean. The reasons for these MOC differences are not clear, but our early Miocene circulation field EMW leads to a δ13C distribution which is in line with benthic foraminifer isotope records.
 Simulations MM, EM and EMW support the conclusions by Woodruff and Savin , Wright et al.  and later by Poore et al.  that deep water formation in the North Atlantic was absent or weak prior to 11 Ma, and challenge a recent analysis of neodymium records which suggests a much earlier onset of NADW production, already at the Eocene/Oligocene transition some 35 Ma ago [Via and Thomas, 2006]. On the other hand, a quantitative comparison reveals significant differences between the absolute values of observed proxy data and the results of EM and MM, which may be to due to the climatological and biogeochemical boundary conditions applied to these experiments.
 To study the effect of seaway forcing, most experiments employed present-day climate forcing including modern sea-ice margins (shown in Figure 4), which is considered as a first-order approximation of the Miocene background climate. While this is a reasonable assumption for the late Miocene, the approach is probably not adequate during warm periods of the early Neogene when sea ice was absent [Harwood and Bohaty, 2000; Moran et al., 2006]. This is further confirmed by the agreement between early Miocene δ13C observations and the results of simulation EMW. In this sensitivity experiment the absence of sea ice favors the production of 13C-depleted organic matter and hence the increase of δ13C in ambient polar surface waters. The elevated δ13C signal then propagates to other oceans and into the deep sea via deep water formation in the Southern Ocean.
 As stated in section 2, all simulations were carried out using present-day biogeochemical input fields. However, during the middle Miocene there were probably global carbon budget changes leading to systematically increased δ13C values in benthic sediment records (e.g., Vincent and Berger ; for a more recent study, see Diester-Haass et al. ). Such carbon budget changes were not considered in our simulations, which may explain the apparent low bias of our model results for MM.
 Simulations LM500, MM, and EM do not capture the sign of the meridional δ13C gradient observed in the east Pacific. In the same experiments the North Pacific was on average isotopically heavier than the North Atlantic, which appears to be unlikely in the light of the scattered observations. A closer inspection of the results of the control run points to a model bias in the North Pacific. As stated previously, the δ13C control run results in the eastern North Pacific are high compared to measurements (Figure 12b). This is probably due to the fact that the model produces too much North Pacific Intermediate Water (NPIW), with an integrated volume transport of more than 4 Sv across 30°N (see Figure 3a) while observations suggest about 2 Sv [e.g., You et al., 2003]. We suspect that too vigorous NPIW formation causing excessive ocean ventilation is also the reason for the mismatch with east Pacific observations in LM500, MM, and EM.
 The wide Eastern Tethys in simulations EM and EMW is a shortcut for saline and isotopically enriched upper-level water from the Indian Ocean entering the North Atlantic. For this reason NADW formation in these experiments does not completely cease, and δ13C values in the North Atlantic are higher than in MM where the Eastern Tethys is narrow and the throughflow is small. In the early Miocene simulations the carbon isotopic signature of the Atlantic is then communicated to the Pacific with the flux of residual NADW through the Central American Seaway.
 The final closure of the Eastern Tethys coincides with the middle Miocene climate transition at about 14 Ma which was a major cooling step in Cenozoic climate history. It has been conjectured that the closure of the Eastern Tethys triggered changes of the Indian Ocean heat transport which in turn catalyzed the Antarctic glaciation [e.g., Woodruff and Savin, 1989; Flower and Kennett, 1994; Shevenell et al., 2004]. We find that the closure of the Eastern Tethys is associated with upper level warming of the northern Indian by up to 6°C while in the depth range 500–2000 m the Indian Ocean cools by up to 4°C (not shown). The cooling occurs northward of 40°S. The thermal reorganization does not affect Antarctic Circumpolar Current transports. Overall, our results suggest that the contribution of the Eastern Tethys closure to the mid-Miocene Antarctic glaciation should have been small.
Wright and Miller  and Poore et al.  have proposed that vertical motions of the Greenland-Scotland Ridge, caused by variations of the Iceland mantle plume activity, may have modulated Neogene deep water exports from the Nordic Seas to the North Atlantic. They correlated Greenland-Scotland Ridge depth variations with an index for NADW formation and suggested that NADW fluxes decreased or even vanished during several periods with highstands of the Greenland-Scotland Ridge (between 14.5 and 11.0 Ma, 9.8–8.5 Ma, and 7.3–6.0 Ma). Here, we focus on the late Miocene. In simulation LM500 (tentatively associated with 9.8–8.5 Ma) the modeled δ13C values in the North Atlantic are lower than the observations, which would demand intensified instead of reduced ocean ventilation to match the observations. The Atlantic δ13C distribution according to LM250 (tentatively applied to 7.3–6.0 Ma) comes close to the observations, which suggests that depth variations of the Greenland-Scotland Ridge are not necessary to explain the late Miocene carbon isotopic record in the North Atlantic. A reason for this insensitivity may be that in model experiment LM250 (as well as in the PD run) the strongest convective activities in the high-latitude North Atlantic are located southward of the Greenland-Scotland Ridge in the Labrador Sea.
 For Quaternary climate background states it has been shown that the MOC may possess multiple equilibria characterized by different modes of deep water formation (e.g., see Maier-Reimer et al. , Prange et al. , and Romanova et al.  for sensitivity studies using the LSG model). We did not carry out a detailed stability analysis of the Miocene experiments, but we run the sequence of simulations also in the reverse direction stepping from EM forward in time and obtained virtually the same results (e.g., the global-mean differences are less than 0.1°C for temperature and smaller than 0.01 PSU for salinity; not shown). For this reason we conclude that our resulting circulation modes are not artifacts of multiple equilibria in the system.
 This study investigated the effects of various seaway configurations characteristic for the Miocene on the ocean circulation and on marine δ13C. For this purpose we carried out a series of modeling sensitivity experiments in which we explored various stages in the closing of the Central American Seaway and the Eastern Tethys.
 Our simulations provide a new framework to interpret Neogene geochemical tracer data and make several predictions which are in line with observational evidence. In all simulations we find net export of water from the Pacific to the Atlantic via the Central American Seaway. The vertical structure of the Central American throughflow is characterized by westward flow of surface water, eastward flow in the thermocline and at intermediate depths, and, in scenarios featuring a 3 km deep seaway, some westward deep water flow. Integrated volume transports through the Central American Seaway amount 6–12 Sv, thus reducing the present-day upper-level salinity difference between the North Atlantic and North Pacific and affecting NADW formation. In periods when the Central American Seaway was about 0.5 km deep or deeper (typical for the early to middle Miocene) deep water formation in the North Atlantic was absent or weak, while the MOC was dominated by water mass production in the Southern Ocean. NADW formation began when the Central American Seaway had shoaled to a few hundreds of meters (typical for the late Miocene), before the final closure of Central American Seaway and before the onset of the Northern Hemisphere glaciation. This is in line with benthic δ13C records and temperature reconstructions challenging the role of the closing of the Central American Seaway in this cooling step.
 Within our modeling framework we cannot support various hypotheses regarding the roles of the Eastern Tethys and of the Greenland-Scotland Ridge. Our results suggest a shortcut for Indian Ocean water export to the Atlantic via a wide Eastern Tethys during the early Neogene, but they do not corroborate conjectures that the mid-Miocene closing of the Eastern Tethys contributed to the Antarctic glaciation. Moreover, our findings regarding NADW formation and δ13C records during the late Miocene are largely independent from Greenland-Scotland Ridge depth variations.
 To a large extent, the late Neogene benthic δ13C record can be explained with large-scale ocean circulation changes. During the early Neogene, the effect of tectonic forcing may have interfered with the isotopic imprints of substantial climate regime shifts and of terrestrial carbon cycle changes which were beyond the scope of this study. Future three-dimensional modeling studies should address these issues.
 Special thanks go to Christoph Heinze for providing the model code of HAMOCC2s and to Heather Poore for providing Neogene isotope data. The paper benefited from discussions with Klaus Grosfeld, Gregor Knorr, Matthias Prange, and Michael Schulz and from reviews by Jim Wright and another anonymous referee. Andreas Manschke is acknowledged for technical assistance. Funded through Deutsche Forschungsgemeinschaft (Bi657).