O water isotope in the i LOVECLIM model ( version 1 . 0 ) – Part 3 : a paleoperspective based on present-day data-model comparison for oxygen stable isotopes in carbonates .

Bunker Cave 51,37 7,66 184 -7,90 -2,84 -5,70 -2,38 10,50 13,62 Riechelmann, 2010; Riechelmann et al., 2011 Cango -33,38 22,22 -5,44 -2,88 -5,40 -3,03 17,50 16,62 Talma and Vogel, 1992 Cascayunga Cave -6,07 -77,18 900 -7,30 0,74 -5,97 Reuter et al., 2009 Ceremosjna cave 45,00 21,00 530 -6,93 -1,95 Kacanski et al., 2001 Chauvet cave 44,23 4,26 240 -6,80 -1,53 13,20 15,00 Genty et al., 2006 Clamouse 44,00 5,00 800 -6,20 -1,53 -4,90 -2,35 14,50 18,77 Plagnes et al., 2002


Introduction
Water isotopes are commonly used as important tracers of the hydrological cycle.Because of small chemical and physical differences between the main isotopic forms of the water molecule (H 2 16 O, HDO, H 2 18 O), an isotopic fractionation occurs, principally during phase transitions of water (evaporation and condensation processes).Stable water isotopes have therefore been measured in a large variety of archives to reconstruct regional climate variations.In polar ice cores, the isotopic composition of the ice has long been used to reconstruct past air temperatures (Dansgaard, 1953;Jouzel, 2003;Jouzel et al., 2007).For tropical ice cores (Thompson et al., 2000), temperature but also precipitation intensity upstream of the air mass trajectories seems to play an important role to explain water isotope signals (Hoffmann et al., 2003;Vimeux et al., 2005).Over the continent, water isotope concentration is preserved in carbonate from speleothems or lacustrine organism such as ostracodes, reflecting the isotopic composition of meteoric water and the temperature dependent water-calcite oxygen isotope fractionation (Urey, 1947;Kim and O'Neil, 1997).For speleothems in tropical regions, T. Caley and D. M. Roche: δ 18 O water isotope in the iLOVECLIM model (version 1.0) -Part 3 some studies have suggested that the amount of precipitation, called amount effect, (Dansgaard, 1964;Rozanski et al., 1993) could be the dominant driver of water isotope concentration (Burns et al., 2002;Fleitmann et al., 2003;Wang et al., 2001).In lakes, water isotope composition can be dependant on temperature, air mass source area and/or a precipitation / evaporation ratio (Stuiver, 1970;Leng and Marshall, 2004).
Similarly to palaeocontinental archives, the seawater oxygen isotope concentration (δ 18 O sw ) is preserved in carbonates from various organisms such as foraminifers and corals.The carbonate isotopic concentration is mainly controlled by temperature and by the isotopic composition of seawater during their formation (Urey, 1947;Shackleton, 1974).
If water isotopes are one of the most widely used proxies in climate archives from the different realms, processes that control their compositions are various and complex.Models allow testing how the past variability of isotopes in water can be interpreted.For example, the importance of temperature versus precipitation changes for the climatic interpretations of tropical isotopic records (Risi et al., 2010).Models can also help to determine the validity of using modern spatial gradients to derive temporal gradients, i.e. the covariability of climate and water isotopes in a specific location over time (Schmidt et al., 2007;Legrande and Schmidt, 2009).By modelling the proxy (water isotopes) directly in the model, the results can also be directly compared with the data.In addition, simulation of climate and its associated isotopic signal can provide a "transfer function" between isotopic signal and the considered climate variable (Sturm et al., 2010).
General circulation models (GCMs) have been frequently used for isotopic studies (Jouzel et al., 1987;Joussame and Jouzel, 1993;Hoffmann et al., 1998;Schmidt, 1998;Delaygue et al., 2000;Paul et al., 1999;Noone and Simmonds, 2002;Mathieu et al., 2002;Lee et al., 2007;Schmidt et al., 2007;Yoshimura et al., 2008;Zhou et al., 2008;Tindall et al., 2009;Risi et al., 2010;LeGrande and Schmidt, 2011;Werner et al., 2011;Xu et al., 2012).Among them, the majority has been used exclusively to simulate water isotopes separately in the atmosphere and ocean components.Oxygen isotopes have been implemented in the three-dimensional model of intermediate complexity called iLOVECLIM (version 1.0), allowing fully coupled atmosphere-ocean simulations.The detailed implementations of oxygen isotopes in iLOVECLIM can be found in Roche (2013).The present work aims at showing preindustrial experiment results in comparison to different global datasets from marine and continental realms.We mainly focus our discussion on carbonate archives that are extensively used in climate studies.The validation of the model results against water isotope observations can be found in Roche and Caley (2013).The benefits, caveats, agreements and problems from this data-model comparison for oxygen stable isotopes in carbonates will be discussed to determine the potential and the interest of using iLOVECLIM for palaeoclimatic studies.

Water isotopes definition, notation and fractionation
Isotopic values are measured by a mass spectrometer and expressed relative to a standard as (δ) notations (Sharp, 2007): R is the abundance ratio of the heavy and light isotopes (e.g.N( 18 O) p /N( 16 O) P ) for substance P and δ is commonly reported in units of parts per thousand (‰).The standard for carbonate is Pee Dee Belemnite (PDB) (Craig, 1957) and that for waters is standard mean ocean water (SMOW) (Baertschi, 1976).More recently, standards have been defined with the prefix "Vienna" or VSMOW and VPDB to account for the exhaustion of the original standard materials (Coplen, 1996).
The magnitude of fractionation can be expressed on the δ scale through the logarithmic transformation 10 3 ln(α A−B ) where α A−B = R A /R B , α is the equilibrium fractionation and R is the partitioning of isotopes between two substances or two phases of the same substance A and B.

Oxygen isotopes in iLOVECLIM
The iLOVECLIM (version 1.0) model is a derivative of the LOVECLIM-1.2 climate model extensively described in Goosse et al. (2010).With regards to water isotopes, the main development lies in the atmospheric component in which evaporation, condensation and existence of different phases (liquid and solid) all affect the isotopic conditions of the water isotopes.In the ocean, the water isotopes are acting as passive tracers ignoring the small fractionation implied by the presence of sea ice (Craig and Gordon, 1965).For the land surface model, the implementation in the bucket follows the same procedure as for the water except that equilibrium fractionation is assumed during phase changes.A detailed description of the method used to compute the water isotopes in iLOVECLIM can be found in Roche (2013).

Global datasets
Global oxygen isotopic datasets for the atmosphere and ocean components have been compiled from published late Holocene measurements to allow comparison and discussion with iLOVECLIM results.Table S1 is a compilation of 74 δ 18 O calcite measurements and drip water data of speleothems from the literature.The δ 18 O of the calcite are reported for the last 1000 years of each record with the associated 2σ error when it was possible to calculate it.Table S1 also includes a compilation of 19 δ 18 O data of ice  S1).
cores from the literature.The δ 18 O of the ice are reported for the last 1000 yr of each record with the associated 2σ error when it was possible to calculate it.For the EPICA Dome C, Vostok and Illimani site, the δD values were converted in δ 18 O values using the global meteoric water line δD = 8.2×δ 18 O + 11.27 (Rozanski et al., 1993).To compare global oceanic isotopic datasets with iLOVECLIM results, we used the late Holocene planktonic foraminifer's δ 18 O database of Waelbroeck et al. (2005) and extend it with 62 data averaging over the last 3000 yr (Table S2).Similarly, for the intermediate and deep ocean, we used the published Atlantic benthic foraminifer's stable isotopic database of Marchal and Curry (2008) and extend it with 73 data covering the global ocean and averaged over the last 3000 yr (Table S2).

Oxygen isotopes: from the atmosphere to continental records
The water evaporated from the surface ocean undergoes a poleward transport over landmasses that results in gradual rainout and a depletion of the remaining moisture in 18 O.The δ 18 O in precipitation shows therefore systematic variations with latitude, altitude and distance from vapour source.The δ 18 O in precipitation can then be conserved in continental carbonate archives.Under equilibrium conditions, the δ 18 O of carbonates (speleothem) depends on both the temperature through its control on equilibrium fractionation between water and calcite (Hendy, 1971;Kim and O'Neil, 1997) and the isotopic composition of the drip water from the cave site in which the speleothem grew.Using the Kim and O'Neil (1997) equation for synthetic calcite and making the approximation that 1000 × lnα(calcite-water)≈ δcalcite-δwater, it  S1).
Figure 1 shows the results obtained with iLOVECLIM for the present-day δ 18 O in precipitation in comparison to drip water data from spelothems and ice core data (Table 1).The observed relationship is good (R 2 = 0.77).This confirms that the model is capable of representing the global distribution of δ 18 O in precipitation as demonstrated in Roche and Caley (2013), although an underestimation of the fractionation towards lower temperatures/latitudes is observed (Roche and Caley, 2013).
Combining the δ 18 O of precipitation and atmospheric temperature and using the equation of Kim and O'Neil (1997), we are able to reconstruct the calcite δ 18 O in the model and to compare it with late Holocene calcite-δ 18 O data (Fig. 2 and Table S1).
We conducted this analysis separately for the European region and for the rest of the world.Indeed, the interpretation of the δ 18 O in precipitation signal over Europe has been found to be complex in iLOVECLIM, probably linked to the representation of the Mediterranean region in the model (Roche and Caley, 2013).Nonetheless, a discrepancy between model results and data exist also when the Europe area is excluded (Fig. 2).An insignificant correlation (R 2 = 0.1) can be observed between measured calcite-δ 18 O and model results (Fig. 2).This weak relationship can be explained by a combination of factors.First, the representation of the δ 18 O in precipitation with the model.In particular the low resolution and simple representation of the atmospheric component Table 1.Data-model comparison for the calcite-δ 18 O signal of individual foraminiferal species using various palaeotemperature equations (Shackleton, 1974 andMulitza et al., 2003).Note that for the depth habitat, they have been estimated based on calcification depth found in the literature (Ottens, 1992;Kohfeld et al., 1996;Bauch, 1997;Bauch et al., 2002;Ganssen and Kroon, 2000;Ostermann et al., 2001;Pak and Kennett, 2002;Anand et al., 2003;Simstich, 2003;Pak et al., 2004;Kuroyangi and Kawahata, 2004;von Langen et al., 2005;Came et al., 2007;Farmer et al., 2007).However, foraminifera are not restricted to a certain depth level and can change their habitat depth with time or at specific locations (Waelbroeck et al., 2005).Another factor is the use of atmospheric temperatures rather than cave temperatures in the calcite-δ 18 O calculation with the modelling results.Indeed, the relationship between measured temperature in the cave and atmospheric temperature in the model is significant but weak (R 2 = 0.63, Fig. 3).Taken together, all these limitations lead to a bias of the calcite δ 18 O reconstructed when using the model results.Another source of uncertainties of this data-model comparison correspond to all potential biases which occur in the natural environment, processes operating in the atmosphere, soil zone, epikarst and cave system (Lachniet, 2009) and affecting the measured calcite-δ 18 O signal (Fig. 2).For example, recent studies suggest the precipitation of the calcite in different caves does not occur under equilibrium conditions indicating significant kinetic isotope effects during speleothem calcite growth (Mickler et al., 2006;Daëron et al., 2011).This is an important limitation for a global quantitative comparison between model and data.
For a better representation at the local scale of the calciteδ 18 O signal in tropical and mid-latitude region, models with higher complexity and resolution are necessary (Wackerbarth et al., 2012).Furthermore, more processes operating between the atmosphere and the cave system need to be integrated, in addition to the knowledge of the isotopic composition of precipitation and temperature, to allow a good quantitative reconstruction of the calcite-δ 18 O signal.It remains to be tested if qualitative data -model comparison for the calciteδ 18 O signal (i.e.anomaly between the past and present day) would give better results with our model.S1).
Another possibility is to use the method recently developed to extract the fossil water of speleothem record (Vonhof et al., 2006;van Breukelen et al., 2008).With this water, it becomes possible to reconstruct hydrogen and oxygen stable isotope analyses of fossil precipitation.Based on our work, it appears that reconstructed fossil water isotopic signals may allow a better quantitative comparison with modelling results (Fig. 1).The signal would also be more easily related to tropical precipitation (Roche and Caley, 2013) than the complex calcite-δ 18 O signal.

Oxygen isotopes: from the ocean to the foraminifer shells
The seawater oxygen isotope concentration (δ 18 O sw ) tracks regional freshwater sources (evaporation, precipitation and river input) and water mass exchange through surface ocean fluxes and ocean circulation (Jacobs et al., 1985;Skinner et al., 2003).It yields information on changes in the hydrological cycle in the past.The δ 18 O sw is preserved in ocean carbonate archives as foraminifer shells.δ 18 O of foraminifer calcite is mainly controlled by temperature and by the isotopic composition of seawater in which the shell grew (Urey, 1947;Shackleton, 1974).The temperature dependence of the equilibrium fractionation of inorganic calcite precipitation around 16.9 • C is given in Shackleton (1974) as Mathematically we can write the inverse equation for the δ 18 O of the water: + 0.27 − 21.9 + (310.6 + 10 T ).

Surface ocean
Because the calcite δ 18 O depends on temperature and δ 18 O sw , the modelled two parameters must be assessed.The δ 18 O sw results of the model were evaluated against δ 18 O in water samples in Roche and Caley (2013), and the agreement is found to be very good.We compare the Levitus and Boyer (1994) data and model results to assess the modelled temperature.From this comparison it appears that sea surface temperature (SST) in the model are in good agreement with data (overall, less than 1 • C of differences is observed) (Fig. 4) and therefore cannot lead to important bias on the calcite-δ 18 O signal.We focus on regions where SSTs are significantly warmer or colder in the model in comparison to data and where some notable discrepancies between the modelled distribution and data for the δ 18 O sw have been observed (Roche and Caley, 2013).
The main differences for the δ 18 O sw are in the Atlantic and Indian subtropical Ocean and the region offshore from  S2).Note that the legend is non-linear.
California (Roche and Caley, 2013).These regions are all marked by warmer SST in the model (Fig. 4).The δ 18 O sw in the model is slightly depleted in the Atlantic Ocean and even more depleted in the North Indian Ocean in comparison to data (Roche and Caley, 2013).These two effects (warmer SST and more depleted δ 18 O sw ) produce both a weak supplementary decrease of the δ 18 O calcite signal but no compensation effect.Concerning the region offshore from California, the warmer SST and more enriched δ 18 O sw could compensate each other but the modelled signal is compared to only two calcite-δ 18 O points of our dataset (Fig. 5).Therefore, the δ 18 O calcite signal of the model can be compared with data with a good accuracy because slight errors in the modelled temperature and water isotope distributions do not compensate each other.Also note that the problem of biased absolute SST values may not apply when anomalies with respect to past climates are calculated (for example the last glacial maximum).
dextral (Ottens, 1992;Ostermann et al., 2001;Pak and Kennett, 2002;Pak et al., 2004;Kuroyangi and Kawahata, 2004;von Langen et al., 2005;Came et al., 2007) and 0-150 m for N. Pachyderma sinistral (Kohfeld et al., 1996;Bauch, 1997;Bauch et al., 2002;Simstich, 2003) (Table 1).In first approximation, a depth habitat of 0-50 m seems to be suitable to compare the model results with a global and varied dataset.Indeed, for this depth habitat, the relationship between model results and data exhibit a very strong correlation (R 2 = 0.85, with a slope of 0.92 and an intercept of 0.15) (Fig. 6).For a depth habitat of 0-20 m, the slope of the relation is 0.89 and the intercept is 0.27, suggesting that the depth habitat is not the main control of the calcite-δ 18 O signal.We obtain a very good quantitative agreement for calcite δ 18 O with highdepleted values in the tropical regions (around −3 ‰) and more enriched values in high latitude regions (around 4 ‰), reflecting the combination of both δ 18 O sw and temperature changes with latitudes (Fig. 5) (Roche and Caley, 2013).
To take into account the seasonal effect we calculate the amplitude in calcite δ 18 O for the two extreme months in comparison to the mean annual value (Fig. 7).These two end-members represent the extreme configurations in which blooms lead to a 100 % of the carbonate deposit.The range between these two end-members corresponds to all the different possibilities for carbonate deposit.The entire range is from 0-2 ‰ with a weak or null seasonal effect at low and high latitudes, whereas we observe the more pronounced seasonal effect at mid-latitudes.This seasonal range in calciteδ 18 O values could explain part of the shift and dispersion observed with the equilibrium 1 : 1 line but correspond to a factor of secondary importance (Fig. 6).
Overall, the model is able to well reproduce the data.This implies that temperature and the isotopic composition of the seawater are the main control on the fossil-δ 18 O signal recorded in foraminifer's shells at the global scale, when all species are grouped together.
We then realised a data-model comparison for the calciteδ 18 O signal for individual species (Fig. 8).For this comparison, the use of palaeotemperature equations derived for living planktonic foraminifera (Mulitza et al., 2003) could be more appropriate than the palaeotemperature equation derived for inorganic precipitates (Shackleton, 1974).We test this possibility but the results indicate no differences for the different equations (Table 1), confirming that over the oceanic temperature range, the slopes of the equations derived for living species agree with the slopes obtained from inorganic precipitates (Mulitza et al., 2003).A significant relationship between data and model results is found for each species except N. Pachyderma sinistral (Table 1, Fig. 8).This foraminifer lives at high latitudes, in polar waters.Some caveats in the model for the polar regions as well as complex environmental factors driven by the δ 18 O signal of this specie are probably responsible for the weak correlation observed (Fig. 8).For example, N. Pachyderma has been found to have a secondary calcification phase at depth which can be as much as 75 % of the individual mass (Kohfeld et al., 1996;Schmidt and Mulitza, 2002).Also, the depth habitat of this species has been found to be very variable (Ottens, 1992;Kohfeld et al., 1996;Bauch, 1997;Ostermann et al., Fig. 8. Relationship between the ocean calcite δ 18 O calculated in the model (0-25 m for G. ruber pink; 0-50 m for G. ruber white, G. sacculifer and G. bulloides; 0-75 m for N. pachyderma dextral; 0-150 m for N. pachyderma sinistral) and calcite-δ 18 O measurements for each of the shallowest dwelling foraminifer species (Waelbroeck et al., 2005, extended by 62 points: Table S2).
We also note that when individual species are considered, the correlation is weaker compared to when species as grouped together (Table 1, Figs. 6 and 8).This probably reflects the stronger influence of species-specific habitat depth, seasonality and other ecological effects (Bemis et al., 1998;Schmidt and Mulitza, 2002;Fraile et al., 2007Fraile et al., , 2008)).In addition, biases linked to sedimentation and post-deposit effects such as bioturbation or dissolution (Waelbroeck et al., 2005) can also play a role.
To test the ability of the model to represent the data at a regional scale we generated enlargement of the North Atlantic Ocean (Fig. 9a).We investigate the potential of the calcite-δ 18 O signal to track the main hydrographic conditions in the region.To determine the position of the fronts, we use the actual temperature and salinity changes that characterise these fronts and convert it into calcite-δ 18 O values.The sub-Arctic Front (SAF) has a high range of variability for temperature and salinity values with a annual mean of around 13 • C Fig. 9. Regional comparison between simulated ocean calcite δ 18 O (0-50 m) for the present day with iLOVECLIM and calciteδ 18 O measurements on the various shallowest dwelling foraminifer species (Waelbroeck et al., 2005, extended by 62 points: Table S2) for (A) the North Atlantic Ocean: the North Atlantic Drift (NAD), major boundary return currents (PC, Portugal Current; AC, Azores Current) (Sverdrup et al., 1942;Crowley, 1981;Dickson et al., 1988) and major hydrographic fronts are indicated.Black lines indicate the position of hydrographic fronts as described in the text with the calcite-δ 18 O signal: PF, polar front (3.94 ‰); AF, Arctic front (2.93 ‰); SAF, sub-Arctic front (1.13 ‰) and blue lines indicate the position described in Dickson et al. (1988).(B) The South Atlantic Ocean: major hydrographic fronts are indicated: PF, polar front; SAF, South Antarctic front; STF, sub-tropical front (Peterson and Stramma, 1991).Note that the legend is non-linear.and 35.4 salinity units (s.u), respectively (Belkin and Levitus, 1996 and references therein).Using the actual relationship of Legrande and Schmidt (2006) for the North Atlantic Ocean (δ 18 O = 0.55 × sea surface salinity (SSS)−18.98)and the equation of Shackleton (1974), we obtain a value of 1.13 ‰ for the δ 18 O calcite.The polar front (PF) is characterised by surface water temperature and salinity lower than 0 • C and 34.4 s.u, respectively (Bauch et al., 1999).Using the actual relationship of Legrande and Schmidt (2006) for the GIN seas (δ 18 O = 0.6× SSS−20.71),we obtain a value of 3.94 ‰ for the δ 18 O calcite.The transition between the Atlantic waters and the Arctic domain is through the Arctic front (AF).In this area, the Atlantic inflow of water has salinity above 35 s.u and a temperature of 5 • C (Skagseth et al., 2008) leading to a δ 18 O calcite value of 2.93‰.These different values of δ 18 O calcite track faithfully the actual position of the front as described in Dickson et al. (1988) and are well documented by our data-model comparison results (Fig. 9a).
Similarly, the main hydrographic conditions in the South Atlantic Ocean with the hydrographic fronts are well documented by the calcite-δ 18 O signal (Fig. 9b).The STF is best defined by salinity (the 34.81 s.u halocline in this area) (Edwards and Emery, 1982) or the 11 • C isotherm at 100 m depth.Using the actual relationship of Legrande and Schmidt (2006) for the South Atlantic Ocean (δ 18 O = 0.51 × SSS−17.4) and the equation of Shackleton (1974), we obtain a value of 0.77‰ for the δ 18 O calcite.For the SAF, we use the definition of the isotherm 4 • C near 200 m depth, and a salinity of 34.5 s.u south of the front (Peterson and Stramma, 1991).Using the actual relationship of Legrande and Schmidt (2006) for the Austral Ocean (δ 18 O = 0.24 × SSS−8.45),we can calculate a δ 18 O calcite value of 2.65‰.For the polar front (PF), the definition relies only on the isotherm and not to salinity.Nowlin et al. (1977) used the northern extent of the 0 • C isotherm.We also use a mean value of −0.4‰ for the δ 18 O sw of the Austral Ocean (Legrande and Schmidt, 2006) leading to a value of 3.6 ‰ for δ 18 O calcite.These different values of δ 18 O calcite track well the actual position of the front as described in Peterson and Stramma (1991) (Fig. 9b).
Applying this data-model comparison for calcite δ 18 O in past climates could constitute an interesting tool for mapping the potential shifts of the frontal systems and circulation changes through time, assuming that changes in foraminiferal habitat depth (Telford et al., 2013) and salinitytemperature relationship in oceanic fronts stay relatively stable with time.Previous data studies on the amplitude of the calcite δ 18 O have documented hydrographic changes during the 8.2 kyr event, the Younger Dryas event, Heinrich events and the Last Glacial Maximum (LGM) (Cortijo et al., 2005;Eynaud et al., 2009).S2).The North Atlantic drift (red arrow in A) is well represented as the penetration of the Antarctic intermediate waters moving northward (black arrow in A).Note that the legend is nonlinear.

Intermediate and deep ocean
Since we obtain very good results at the surface of the ocean it is interesting to investigate intermediate and deep ocean results in comparison to available data of benthic foraminifers.To document the intermediate depth circulation, we present in Fig. 10a the ocean calcite δ 18 O for a mean depth of 500-1500 m.A good agreement is observed between model results and data.The North Atlantic drift is well represented and the penetration of the Antarctic intermediate waters moving northward is also visible.Any potential changes in these oceanic circulation systems in the past could be documented by a data-model comparison exercise.Outside of the Atlantic Ocean region, the data are too sparse to validate the model results.Supplementary data in the Indian Ocean, Pacific Ocean and Southern Ocean would be a major advance in order to validate the observed distribution in calcite δ 18 O at intermediate depth.
Concerning the mean deep ocean (2000-4000 m), Fig. 10b indicates a relatively good agreement between model results and data for the Atlantic Ocean contrary to the other parts of the world ocean.Nonetheless, the calcite δ 18 O simulated with iLOVECLIM for the deep ocean is globally too high in comparison to data.The values in the model are actually close to values reached between interglacial and glacial periods (3.8-4.2 ‰).After investigation, it appears that two factors can explain this abnormally high δ 18 O signal.First the modelled distribution of δ 18 O sw is 0.25 per mil lower than data between 20 • S and 30 • N (Roche and Caley, 2013).Second, the modelled temperature of deep water is around 2 • C lower than data, a pattern particularly marked in the Austral Ocean.The caveat for the deep ocean temperature with the model is unclear so far and would be the matter of future investigation.We hypothesise that it is linked to the deep water formation in the Southern Ocean since it shows a marked underestimation in that region.Together, these two biases explain the high δ 18 O calcite signal in the model.These results also elegantly suggest that changes in the deep water temperature by only few degrees can have an important effect of the benthic δ 18 O calcite signal, a pattern that requires further investigation during past climate changes.

Conclusions
Concerning the atmospheric component and contrary to the δ 18 O in precipitation, the limitation of the model together with the processes operating in the atmosphere, soil zone, epikarst and cave system hamper a good quantitative datamodel comparison for the calcite-δ 18 O signal.It remains to be tested whether the use of more complex isotope-enabled models at higher resolution added to the inclusion of supplementary processes operating between the atmosphere and the cave system would allow to obtain better results.It is necessary to assess such improvements in parallel with improvements regarding the potential biases occurring in the natural environment and affecting the measured calcite-δ 18 O signal.It also remains to be tested if qualitative data -model comparison for the calcite-δ 18 O signal (i.e.anomaly between the past and present day) would give better results.Based on our work and for simplicity reasons, it appears that reconstructed fossil water isotopic signals by fluid inclusions in speleothems is a more promising way to compare quantitative simulation results with climate data, a comparison which is today restricted to ice core records.The relationship between the δ 18 O in precipitation and climate variables such as temperature and precipitation rates could be investigated for past time periods.
Concerning the oceanic component, the simulated calcite δ 18 O compare quantitatively well with global late Holocene planktonic foraminifer δ 18 O data.Our results indicate that temperature and the isotopic composition of the seawater are the main control on the fossil-δ 18 O signal recorded in foraminifer's shells at the global and regional scale.Nonetheless, depth life, seasonality and other ecological effects play also a role and are more expressed when individual species are considered.Further works with more sophisticated ecological models are needed to refine these conclusions and increase the quantitative match of the modelled calcite-δ 18 O results with data.We argue that a data-model comparison exercise for calcite δ 18 O in past climate could constitute an interesting tool for mapping the potential shifts of the frontal systems and circulation changes throughout time.Of particular interest is the ocean circulation at the LGM.By simulating both the calcite δ 18 O and δ 18 O sw in the model and by comparison with foraminifer's calcite-δ 18 O data it would be possible to address the LGM tropical sea surface temperature controversy (large east-west gradients) (MARGO Project Members, 2009).
Similarly to surface ocean, the potential changes in intermediate oceanic circulation systems in the past could be documented by a data-model comparison exercise but an extension of the spatial data coverage is necessary.For deep ocean calcite-δ 18 O signal, the abnormally low-modelled deep ocean temperature would be the matter of future investigation in order to quantitatively compare the results with data.

Fig. 1 .
Fig. 1.Simulated δ 18 O in precipitation with iLOVECLIM in comparison to measured δ 18 O in precipitation from climate records (speleothems drip water and ice cores) (TableS1).

Fig. 4 .
Fig. 4. Anomaly of mean sea surface temperature (modelled in iLOVECLIM -Levitus and Boyer, 1994).The hatching denotes areas where the difference is less than 1 • C.

Fig. 10 .
Fig. 10.Simulated ocean calcite δ 18 O (500-1500 m in (A) and 2000-4000 m in (B)) for the present day with iLOVECLIM.Superimposed dots represent calcite-δ 18 O data measured on benthic foraminifer species (Marchal and Curry, 2008, extended by 73 points: TableS2).The North Atlantic drift (red arrow in A) is well represented as the penetration of the Antarctic intermediate waters moving northward (black arrow in A).Note that the legend is nonlinear.