Modeling variations of marine reservoir ages during the last 45 000 years

. When dating marine samples with 14 C, the reservoir-age effect is usually assumed to be constant, although atmospheric 14 C production rate and ocean circulation changes cause temporal and spatial reservoir-age variations. These lead to dating errors, which can limit the interpretation of cause and effect in paleoclimate data. We used a global ocean circulation model forced by transient atmospheric 1 14 C variations to calculate reservoir ages for the last 45 000 years for a present day-like and a last glacial maximum-like ocean circulation. A (cid:24) 30% reduced Atlantic meridonal overturning circulation leads to increased reservoir ages by up to (cid:24) 500 years in high latitudes. Temporal variations are proportional to the absolute value of the reservoir age; regions with large reservoir age also show large variation. Temporal variations range between (cid:24) 300 years in parts of the subtropics and (cid:24) 1000 years in the Southern Ocean. For tropical regions, which are generally assumed to have nearly stable reservoir ages, the model suggests variations of several hundred years.


Introduction
Late Quaternary sediments are frequently dated by means of their radiocarbon ( 14 C) content. 14 C originates in the atmosphere, where cosmic rays generate free neutrons that can react with nitrogen to produce 14 C (Masarik and Beer, 1999). After exchange with the other carbon reservoirs most of the radiocarbon is stored in the ocean, where it decays. The finite exchange flux between the reservoirs causes the radiocarbon age of marine sample always to be higher than that of a coeval atmospheric sample (Bard, 1988;Stuiver and Braziunas, 1993). This age difference is known as "reservoir age" and Correspondence to: J. Franke (joerg.franke@palmod.uni-bremen.de) has to be taken into account in the conversion from radiocarbon age to calendar age.
Present-day (PD) reservoir ages average globally at about 400 years (Bard, 1988;Hughen et al., 2004a). Regional reservoir-age anomalies for the time before nuclear weapon tests are mainly known from sites along coastlines (Reimer and Reimer, 2001). Accordingly 14 C dates are mostly corrected for a local but constant PD reservoir age instead of the global mean. Temporal reservoir-age variations in contrast are hardly considered when marine samples are dated, because they could only be scarcely reconstructed for limited time periods and at a few locations (Southon et al., 1990;Bard et al., 1994;Austin et al., 1995;Burr et al., 1998;Sikes et al., 2000;Siani et al., 2001;Waelbroeck et al., 2001;Keigwin and Schlegel, 2002;Kovanen and Easterbrook, 2002;Eiriksson et al., 2004;Bard and Rostek, 2005;Fairbanks et al., 2005;Bondevik et al., 2006;Schimmelmann et al., 2006;Hughen et al., 2006). These reconstructions suggest that reservoir-age changes of several hundred years occurred in the late Quaternary. Errors of such a magnitude might lead to a misinterpretations of cause and effect in paleoclimate time series.
Changes in the geomagnetic field, which directly influences the atmospheric 14 C production rate are considered to be the main reason for reservoir-age variations (Laj et al., 1996). Model experiments suggest that the strength Atlantic meridional overturning circulation (AMOC) also significantly influences atmospheric 14 C ( 14 C atm ) and subsequently reservoir ages, too (Delaygue et al., 2003;Muscheler et al., 2004). Running a spatially explicit ocean circulation model forced by changes in atmospheric 14 C offers the opportunity to assess a major part of reservoirs age variability, that is induced by 14 C production rate changes. LGM-like boundary conditions the AMOC is reduced by approximately one third and shallower such that AABW can reach further north at the bottom of the ocean.

Model setup
The reservoir-age calculation was done using a global model of intermediate complexity, the University of Victoria Earth System Climate Model (UVic ESCM) in version 2.7 . It consists of a three-dimensional ocean general circulation model (Modular Ocean Model, version 2, Pacanowski (1995)), coupled to a two-dimensional energymoisture balance model of the atmosphere (Fanning and Weaver, 1996) and a dynamic-thermodynamic sea-ice model . The horizontal resolution of all components is 3.6 • in longitude and 1.8 • in latitude. The ocean has 19 levels of irregular depth, increasing from 50 m at the surface to 500 m at the deepest levels . It is driven by variations in solar insolation over a year at the top of the atmosphere. The wind stress at the ocean surface is prescribed from a monthly climatology (Kalnay et al., 1996). We used the option of a rotated grid to avoid convergence of the meridians towards the North Pole. Sub-gridscale mixing is included following the Gent and McWilliams (1990) parametrization for mixing associated with mesoscale eddies. Vertical diffusion is increasing from 0.3 cm 2 s −1 in the thermocline to 1.3 cm 2 s −1 in the deep ocean (Bryan and Lewis, 1979).
To evaluate if the model simulates PD ocean circulation right, a control experiment was set up, using PD parameters as a solar radiation and land-ice distribution of the year 1950 Common Era (CE), monthly mean winds from reanalysis data of the 20th century (Kalnay et al., 1996) and a preindustrial atmospheric CO 2 content of 280 ppmv. In this configuration the model shows a maximum North Atlantic overturning of 20 Sv and a southward export of North Atlantic Deep Water (NADW) at 30 • S of 14 Sv (Fig. 1a). Antartic Bottom Water (AABW) reaches up to 30 • N. All these values agree fairly well with calculations based on observational data (Talley et al., 2003). Finally a circumpolar current of around 100 Sv is comparable with observations compiled by Orsi et al. (1995).
In the ocean part of the model, radiocarbon was included as a passive tracer following the guidelines of the Ocean Carbon Modeling Intercomparison Project (OCMIP-2, Orr et al., 2000): and Sc=2073.1−125.62 · SST + 3.63 · SST 2 −0.043 · SST 3 (4) where F air-sea is the flux of 14 C from the atmosphere to the ocean, K w is CO 2 gas transfer velocity, 14 C sat and 14 C surf are the 14 C concentrations in the atmosphere and surface ocean respectively, αC is the carbon solubility for water-vapor saturated air mol m 3 ·µatm , pCO 2 is partial pressure of CO 2 in the atmosphere, P is local sea level air pressure, P 0 is the mean sea-level air pressure of 1013.25 hPa, R std is the normalized ratio of 14 C 12 C , f ice is the modeled fraction of sea-ice coverage (height>1 cm), a is a constant to adjust the global flux, u 2 is windspeed in m s and Sc is sea-surface temperature (SST[K]) dependend Schmidt number.
The gas exchange with the atmosphere depends on the atmosphere to surface-ocean 14 C gradient, windspeed, sea-ice cover and sea-surface temperature. In the ocean the radiocarbon tracer is transported via diffusion and advection like all the other tracer (e.g. temperature, salinity). A sink has been added to account for the radiocarbon decay with the true half-life of 5730 years.
The atmosphere is treated as one well-mixed box with respect to 14 C because the atmospheric mixing time for 14 C is on the order of some years, which is much shorter than the timescale of interest. Splitting the atmosphere into troposphere and stratosphere is not necessary, because this would only have an influence on variations at timescales shorter than 20 years (Siegenthaler et al., 1980). The terrestrial biosphere has an effect if forcing variations are on timescales from a few decades to some centuries (Siegenthaler et al., 1980). As we concentrated on even longer variations, the terrestrial biosphere is also not taken into account, to make the model more efficient.

Control run and model evaluation
In the control experiment 14 C atm is held constant at 0‰, which is defined as the pre-industrial 14 C/ 12 C-ratio of the year 1850 CE. To evaluate the model, we compared the oceanic 14 C distribution with the global carbon climatology (Key et al., 2004). This dataset includes the radiocarbon measurements at the time of sampling as well as calculated estimates for natural background and bomb-produced 14 C. The gas exchange of the model was reduced by ∼20% compared to OCMIP recommendations (Orr et al., 2000), which is in agreement with recent calculations (Sweeney et al., 2007).

14 C variations in a bomb 14 C experiment
Since our main interest is to simulate temporal reservoirsage variations, the model response to 14 C production rate changes needs to be verified. This can be achieved in an experiment, which is forced by the well-known radiocarbon production due to the test of nuclear weapons during the second half of the 20th century. Estimates for the nuclear bomb strength were taken from Hesshaimer et al. (1994). The model was started from the the Suess-effect corrected PD equilibrium state described in the last section and was run for 40 years. 14 C measurements used for comparison are in the atmosphere spatially weighted global means based on regional data of Hua and Barbetti (2004) and in the ocean shallow-water coral data (Druffel, 1989).

Influence of different ocean circulation states
To study the influence of different ocean circulation states on the 14 C distribution, the model was forced by LGM-like boundary conditions: Insolation and land ice distribution were set to 21 kyr BP and the atmospheric CO 2 concentration was reduced to 200 ppmv (Table 1). Using these parameters the Atlantic Meridional Overturning Circulation (AMOC)  Kalnay et al. (1996) became weaker by roughly one third and it became shallower, such that AABW could penetrate further northward in the deep Atlantic ( Fig. 1). This weaker and shallower overturning cell is consistent with the glacial nutrient distribution, Pa/Th and most other circulation tracers (Schmittner et al., 2002;Meissner et al., 2003;McManus et al., 2004;Lynch-Stieglitz et al., 2007). Reconstructions of dust and pollen transport (e.g. Stuut et al., 2002;Shi et al., 2001) as well as climate models (e.g. Braconnot et al., 2007) suggest an increased wind speed during the last glacial maximum. In the UVic ESCM a higher wind speed leads to an intensification of NADW formation because the model cannot resolve the increased sea-ice/fresh-water transport with the East Greenland and Labrador Current to the North Atlantic that can be observed in models of higher spatial resolution such as the Community Climate System Model (CCSM) in the PMIP-2 project (Braconnot et al., 2007). To assess the influence of a reduced circulation which is in agreement with most reconstructions on the reservoir ages the PD wind fields are also used for the LGM simulations following Meissner et al. (2003).

Atmospheric 14 C forcing
To study past changes in oceanic 14 C content, we prescribed the temporal 14 C evolution based on reconstructions. For this we used the INTCAL04 dataset (Reimer et al., 2004) up to 25 kyr BP. Between 25 and 50 kyr BP we used the reconstructions by Fairbanks et al. (2005) and Hughen et al. (2006) because they could remove some of the uncertainties that caused disagreement in earlier reconstructions (e.g. Voelker et al., 1998;Bard et al., 1998;Goslar et al., 2000;Kitagawa and Plicht, 2000;Beck et al., 2001;Hughen et al., 2004b). Remaining uncertainties in the reconstructions such as the assumption of a constant low latitude reservoir age to estimate 14 C atm from marine reconstructions between 12.4 and 50 kyr BP are discussed in Sect. 5.1. The 14 C atm model-forcing dataset was constructed by interpolating an error weighted spline (Williams and Kelley, 2007) through the selected reconstructions which leads to a slightly smoothed 14 C production rate (Fig. 2).  (Reimer et al., 2004, orange), coral data (Fairbanks et al., 2005, red) and Cariaco Basin sediment data (Hughen et al., 2006, blue). A spline function (black) was interpolated through the data weighted by the 1·σ error of all the reconstructed values.
The long time period needed to reach an equilibrium between atmospheric and oceanic 14 C concentrations requires a model spin-up time of several thousand years (Siegenthaler et al., 1980). As only very few 14 C atm reconstructions exist prior to 50 kyr BP (Hughen et al., 2006;van Kreveld et al., 2000;Voelker et al., 2000), we spun up the model from 75 kyr BP using a 14 C production rate, calculated after (after Masarik and Beer, 1999) from a global paleomagnetic intensity stack (GLOPIS, Laj et al., 2004). At 50 kyr BP the forcing was switched to the interpolated 14 C spline. This change of forcings adds some uncertainty to the initial oceanic 14 C level, beside the possibility of a different ocean circulation state.

Control run and bomb experiment
The modeled 14 C distribution agrees with the pre-nuclear GLobal Ocean Data Analysis Project (GLODAP, Key et al., 2004) estimate mostly within ±10‰ (Fig. 3). Only in upwelling areas and in the deep northern Pacific Ocean the model predicts too negative 14 C values (up to 30‰ difference). This is the result of too strong tropical upwelling, which is a prevalent feature of most coarse resolution ocean models with high vertical diffusivity (Toggweiler et al., 1989;Orr, 2002). In the equilibrium experiment the global mean surface ocean 14 C is −61‰ close to the bomb 14 C corrected GLODAP estimate of −65‰. Therewith, the UVic ESCM is able to simulate the oceanic 14 C distribution more accurately than most OCMIP-2 models, many of which have problems simulating NADW reaching depths below 3000 m and which showed 14 C deviations in comparison with mea-surements up to ±50‰ over large regions of the deep northern Pacific and deviations ranging from +80 to −40‰ in the Southern Ocean (Orr, 2002).
Forced by the 14 C production rate changes due to nuclear weapon testing, the model is able to predict temporal 14 C atm variations in good agreement with observations (Fig. 4).
The uptake of bomb radiocarbon by the ocean can be analyzed by comparing the model's temporal response with a time series of reconstructed surface-ocean 14 C based on corals (Druffel, 1989). Near Bermuda the model slightly underestimates the 14 C measured in corals (Fig. 5. One reason could be that the UVic ESCM is too diffusive. Thus, the radiocarbon is transported into the subsurface ocean faster than in reality. An alternative explanation is that the wellstratified region, where the corals were actually sampled, is not resolved by the too coarse resolution of the model or that this region is the spatially shifted by a few degrees. This is supported by the fact that at a location 10 • south and 4 • east the observed 14 C is also predicted by the model (Fig. 5).

Time slices
We ran two simulations for the different circulation states of the ocean, both with the same time-dependent 14 C atm forcing. Three time slices of this simulation are plotted in Fig. 6. One for the "Laschamp" event 41 kyr BP, which is important in relationship to the radiocarbon history because the geomagnetic field collapsed almost completely, resulting in a high 14 C production (Laj et al., 2000). The LGM was chosen as a second time slice because of its paleoclimatic importance, and finally PD for comparison.
The PD reservoir ages, reached at the end of the transient experiment for a modern ocean circulation (Fig. 6a), are slightly smaller than in the equilibrium experiment because 14 C atm had a decreasing trend over the last centuries before this snapshot was taken. Nevertheless the reservoir-age estimates agree with present day observation of 400-500 years in the northern North Atlantic, 300-400 years in the subtropical oceans and up to 1000 years close to Antarctica. For an AMOC reduced by ∼30%, the model suggests larger reservoir ages than with a modern circulation, reaching values of 400-1400 years (Fig. 6b).
The LGM falls in a time period of decreasing 14 C atm . If the circulation would have been like today the model predicts reservoir ages to be below present day values, ranging from 200 years in the subtropical ocean up to 900 years close to Antarctica (Fig. 6c). The reduced AMOC leads to increased reservoir ages reaching from 300 to 1200 years (Fig. 6d).
During the Laschamp event and its high 14 C production rate the model suggests globally increased reservoir ages of 400-1400 years if the circulation would have been like today (Fig. 6e) and of 500-1800 years in case of the reduced AMOC (Fig. 6f).

Temporal reservoir-age variations
The amplitude of the modeled temporal reservoir-age variations over the last 45 kyr varies spatially. This is shown for the simulation with PD boundary conditions (Fig. 7). The amplitude of temporal variations with LGM boundary conditions are very similar, only shifted to higher reservoir-ages. The smallest changes occur in some subtropical regions but even there the range of temporal variation is rarely below 300 years. In the northern North Atlantic these variation are larger and reach up to 700 years. Largest reservoir-age varia-tions were modeled for the Southern Ocean where they could exceed 1000 years. Reservoir-age increases coincide with 14 C atm increases and the other way around.

Reduced Atlantic meridional overturning circulation
To analyze the influence of circulation changes induced reservoir-age differences separated from the temporal 14 C atm variations, the control experiment with the PD circulation is compared with an experiment of constant 14 C atm and LGM-like boundary conditions. In the LGM- Fig. 4. Observed and modeled 14 C atm due to nuclear weapon tests. The blue bars represent the 14 C production which was caused by nuclear weapon tests (Hesshaimer et al., 1994). The red curve shows the observed and globally averaged 14 C atm (Hua and Barbetti, 2004), while the black curve is the response of the model to the production rate forcing.
like simulation 14 C atm was set to zero instead of the reconstructed LGM value to allow for a direct comparison of the results.
In an equilibrium state, reservoir ages increase globally with the LGM forcing (Fig. 8). The circulation induced differences stay small at around 100-200 years in the subtropical and tropical regions with strong stratification and slow diffusive mixing of deep water into the surface. Largest anomalies of 250-400 years could be seen in the Southern Ocean and even up to 500 years in the Arctic Ocean, probably a result of increased sea-ice coverage that limits the gas exchange.

Atmospheric 14 C forcing
In principal there exist two possibilities to model atmospheric 14 C variations. One approach would be the forcing of the model with a 14 C production rate, the other option is 14 C atm reconstructions. The 14 C production rate variations can be calculated based on reconstructions of geomagnetic intensity (e.g. Valet et al., 2005;Laj et al., 2004) or 10 Be (Muscheler et al., 2004) using the method of Masarik and Beer (1999).
So far, no model has been able to reproduce the reconstructed 14 C values of far above 500‰ in the atmosphere during the last glacial, using 14 C production rates alone Laj et al., 2000Laj et al., , 2002Hughen et al., 2004b;Muscheler et al., 2004). More complex models showed even lower 14 C values in the atmosphere than simple models as it could be seen in the difference between a 17box and a 4-box model (Laj et al., 2002). Forced with the 14 C production rate, based on the global paleointensity stack , the UVic ESCM confirms the results of the box  clim-past. net/4/125/2008/cp-4-125-2008-supplement.pdf). To reach the observed 14 C atm values, which are approximately twice as large, major changes in the carbon cycle are required. One possible explanation would be a glacial deep-ocean carbon reservoir that is well isolated from the atmosphere and stores radiocarbon depleted waters .
Generally the absolute 14 C atm value is irrelevant for the reservoir age, as it can be seen in Fig. 7, in which glacial reservoir ages vary around the same level as ages in the Holocene, although 14 C atm was a few hundred permil higher. Instead, the rate of 14 C atm change is the essential factor. Hence, modeled reservoir age variation do not have to differ in magnitude between the 14 C production rate and the 14 C atm forcing, as long as production-rate increase and AMOC reduction do not occur at the same time. Still major reservoir age changes would stay unconsidered because the production rate can only explain half of the overall variability.
The 14 C forcing has the advantage that the variations, which were not caused by atmospheric 14 C production rate changes also appear in the simulated reservoir ages. Measurement uncertainties in the 14 C reconstructions could be reduced, so the large scatter between different datasets decreased over the last years. Nevertheless the reconstructions before 12.4 kyr BP still do not agree with each other completely (Fairbanks et al., 2005;Hughen et al., 2006), e.g. because the calibration to an age scale is often associated with some uncertainties or because an unknown reservoir age has to be assumed for marine samples. To convert the marine 14 C values from corals and sediments to an atmospheric value, constant reservoir ages have been applied, as no reservoir-age variation estimates exist for the low latitudes in the last glacial or the deglaciation. If we assume reservoir-age variation in (sub)tropics, a temporally increased reservoir age implies that the reconstructed 14 C atm is underestimated compared to the value calculated with a constant reservoir age. If the real reservoir age is smaller than the constant one, 14 C atm is overestimated. This together might cause modeled reservoir-age variations to be underestimated because reservoir-age increases coincide with 14 C increases and the other way around (see Sect. 4.3).
Based on the bomb experiment we think that the 14 C atm forcing can be treated as the 14 C production rate forcing in the model. Simulating long-term variations, there will always be enough time for the ocean to equilibrate with the atmosphere within the gas exchange rate limitations.
The disadvantage of the 14 C atm forcing is that any 14 C atm increase acts as if it was caused by a production rate increase. This may lead to artefacts in deep water formation areas during times of reduced deep water production, since the coupling of the surface layer to the deep ocean remains www.clim-past.net/4/125/2008/ Fig. 7. Modeled range of reservoir-age variations over the time period from 45 kyr BP to PD in (a). In (b) the regional mean reservoir-age variations are plotted for the areas indicated by equally colored rectangles in (a). unchanged. In these cases our modeled reservoir-age variations will be slightly underestimated again.

Reservoir-age variations
The comparably large oceanic carbon reservoir responds to atmospheric 14 C changes with a time lag due to limited gas exchange. A 14 C atm increase leads to a larger atmosphereocean 14 C difference and accordingly to a reservoir age increase, because the ocean cannot react fast enough, e.g. around 41 kyr BP (Fig. 7). As soon as 14 C atm stops rising or is reduced, the reservoir ages decrease again. The opposite is true when 14 C atm declines, e.g. around 15 kyr BP.
Changes in reservoir ages occur globally nearly simultaneously because the fast varying and well mixed atmosphere is the key driver (Fig. 7). In contrast, the amplitude of the reservoir-age variations differs at any location. Regions of large reservoir ages are as well as areas of large reservoir-age variations, like the Southern Ocean with more than 1000 years of PD surface 14 C age and also variations of more than 1000 years ( Fig. 6 and 7). In a period of a 14 C atm increase the radiocarbon content of the atmosphere increases while radiocarbon depleted water wells up from the deep ocean, which was once at the sea surface, when atmospheric 14 C was much lower. This causes the 14 C gradient between atmosphere and ocean to be larger than caused by the limited gas exchange alone. In the opposite case of a 14 C atm decrease, upwelling water was in contact with an atmosphere of higher 14 C which leads to smaller reservoir ages.
In case of reservoir age changes that are not caused by 14 C production rate variation, the simulated reservoir age change is correct, but not the level, at which the reservoir ages remain after that first change. Box models suggested that more deep water production will transport more radiocarbon into the deep ocean and finally decreases atmospheric 14 C. In contrast, a reduced AMOC will lead to a lower 14 C transport into the deep ocean and to increased atmospheric 14 C Laj et al., 2002;Hughen et al., 2004b). The observed global reservoir-age increase in our simulation with a reduced AMOC agrees with this finding (Fig. 8). Temporally stable reservoir-age shifts can be initiated by changes in the carbon reservoirs. A 14 C atm increase in the model forcing, increases the atmosphere-ocean 14 C difference in the first moment as well, but because of the constant deep water formation, the ocean starts to take up more 14 C, too. This decreases the reservoir ages again in the model, while they would remain larger in reality as long as the carbon reservoirs stay in a different state, like in the simulation with LGM boundary conditions.
Reservoir ages of 700-800 years were reconstructed in the northern North Atlantic (Bard et al., 1994;Sarnthein et al., 2001;Waelbroeck et al., 2001) and even 2000 years close to New Zealand (Sikes et al., 2000). Further evidence for large reservoir-age variations comes from 14 C-plateau matching, which suggest North Atlantic reservoir ages of 2000 years and more during Heinrich event 1 (Sarnthein et al., 2007). If we add up the modeled temporal variations of up to 1000 years, variations subsequent to the reduced AMOC of up to 500 years or even more in case of a complete deepwater formation shutdown and including the underestimated effects mentioned in section 5.1, reservoir-ages variations of above 2000 years, appear to be reasonable in some regions. From the model simulation we would expect such large variations only in the Southern Ocean but not in the northern North Atlantic.
Modeled reservoir-age variation are not limited to high latitudes, they reach up to a few hundred years in tropical oceans, which were believed to be nearly stable (Hughen et al., 2004a). This has implications for the dating of atmospheric samples, because it adds some uncertainty to all 14 C calibration curves, which assume a constant reservoir age prior to 12.4 kyr BP.

Potential of modeled reservoir ages
In constrast to reservoir-age reconstructions, estimates from an ocean circulation model are available at every location, time and also at different depth levels. The depth is important factor because reconstructions are often based on foraminifera that calcified between sea surface and 250 m depth (Simstich et al., 2003;Schiebel and Hemleben, 2005). The reservoir age of a species living in 250 m depth can severly differ from the surface reservoir age. This occurs especially in the North Pacific where reservoir ages in 250 m depth are up to 500 years larger than at the ocean surface. Model results suggest that it is also important to consider the living depth of a species before correcting for the reservoir age in other regions.
Nevertheless the modeled reservoir ages should not be directly applied to correct marine 14 C ages because only two constant ocean circulation states have been studied and because the UVic ESCM is not able to resolve all circulation features on a regional scale accurately. Thus, the model results indicate the possible reservoir-age variability for the specific region and time period of interest. The reliability of the model at the site and differences to the simulated ocean circulation need to be interpreted, before the modeled reservoir ages can be used for the 14 C age correction or to assess the error in the apparent age that is caused by reservoir-age variations.

Comparison of modeled and reconstructed reservoir ages for the Younger Dryas
Finally the reliability of modeled reservoir ages should be checked by a comparison with reconstructions. For this purpose the North Atlantic is the best covered region. For the time period from the Bølling to the Preboreal reservoir ages were reconstructed from co-existing marine and terrestrial Fig. 9. Reconstructed and modeled reservoir ages from the Bølling/Allerød over the Younger Dryas to the Preboreal: The symbols represent the reconstructed reservoir ages for three regions in the northern North Atlantic, Sweden (Björck et al., 1998), Norway (Bondevik et al., 1999(Bondevik et al., , 2006 and Orphans Knoll (Cao et al., 2007). The modeled mean reservoir ages at the sample locations are plotted for the PD circulation (straight curve) and for the reduced AMOC simulation (dashed curve).
material (Björck et al., 1998;Bondevik et al., 1999Bondevik et al., , 2006, from volcanic ash layers (Bard et al., 1994;Austin et al., 1995) and from corals (Cao et al., 2007). The presented reconstructions of Bondevik et al. (2006) and Cao et al. (2007) are based on the tree-ring 14 C series of Kromer et al. (2004). It should be noted that a new positioning of the tree-ring data has recently been suggested (Muscheler et al., 2008) which would decrease the reconstructed reservoir ages in the early Younger Dryas. The reconstructions show large scatter and have large error bars, which nearly cover the whole range of variations, e.g. reservoir ages from Norway at nearly the same time between 13.7 and 13.8 kyr BP show ∼400 years difference (Fig. 9). Nevertheless there is a trend from PD-like reservoir ages around 400 years in the Bølling, over an increase in the Allerød to the Younger Dryas reservoir ages of circa 600 years and finally a decrease towards a PD value in the Preboreal again. The model predicts the Bølling reservoir age, the increase in the Allerød and the PD-like values in the Preboreal very well in the run with the PD circulation. Only during the Younger Dryas (∼12.9-11.6 kyr BP) the modeled reservoir ages remain below the reconstructed values and they start to decrease too quickly after reaching a maximum at the beginning of the Younger Dryas. It is thought that the cause of this reservoir age increase was a slowdown of the AMOC during the Younger Dryas (e.g. McManus et al., 2004). In the model simulation with reduced AMOC, the predictions for Younger Dryas reach or even exceed the reconstructions. The fact that reservoir-ages decrease too early with the PD forcing demonstrates that a 14 C atm model forcing changes reservoir ages temporally like an atmospheric 14 C production rate variations, but it can only generate the initial peak of an carbon reservoir change induced reservoirage variation. The correct interpretation and consideration of www.clim-past.net/4/125/2008/ ocean circulation changes is therefore essential, when simulated reservoir ages should be applied for an age correction of marine samples.
A compilation of reconstructions from different location might create the wrong impression that reservoir ages should be the same everywhere. Indeed, the existence of local differences can hardly be seen in the scatter of the data but simulation results clearly show a ∼50 year reservoir-age difference between Norway and Orphans Knoll (Fig. 9). The modeled reservoir ages for Sweden differ largely between the simulated AMOC states. That highlights that a reduction of the AMOC always increases the reservoir ages but by spatially different amounts.
The comparison of modeling results and reconstructions shows that our model can simulate the reservoir-age variation induced by changes in 14 C atm in the correct order of magnitude. It is difficult to determine the quality of the model results due to the large scatter in the reconstructions.

Conclusions
Our ocean general circulation model confirms the results of previous box-model experiments that geomagnetic variations alone appear to be insufficient to explain reconstructed atmospheric 14 C variations in the last glacial and the deglaciation.
Simulations of past reservoir-age variations, using a 14 C atm forcing, emphasize the need to make a temporal, spatial and depth depended reservoir-age correction when marine samples are dated with the radiocarbon method. The model suggests reservoir-age variations of several hundred years within some centuries due to 14 C production rate and ocean circulation changes. The modeled reservoir-age variations are not limited to high latitudes and can reach up to a few hundred years in tropical oceans. This has implications for 14 C calibration curves, which are mainly based on coral data and a constant reservoir age.
For regions and time periods, where no reservoir-age variation can be reconstructed, the model results can indicate potential reservoir-age variations. Modeled reservoir ages are available online (www.reservoirage.uni-bremen.de).