Water isotopes during the Last Glacial Maximum: New general circulation model calculations

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 113,, doi: /2008jd009859, 2008 Water isotopes during the Last Glacial Maximum: New general circulation model calculations Jung-Eun Lee, 1 Inez Fung, 1 Donald J. DePaolo, 1 and Bette Otto-Bliesner 2 Received 23 January 2008; revised 21 April 2008; accepted 27 June 2008; published 8 October [1] The application of water isotopes to estimate the glacial-interglacial cycle of temperature (T) assumes the validity of the present-day spatial relationship between T a and d 18 O in precipitation (d 18 O p ) to estimate temporal changes of the temperature at a fixed location. We explored how and why the spatial relationship between annual mean T a d 18 O p is different from the temporal relationship at one location. Our general circulation-isotope model exhibits a spatial slope of 1.22%/ C between annual mean temperature at the top of the inversion layer (T i ) and d 18 O p over Antarctica, comparable to the observed value of 1.25%/ C from Dahe et al. (1999) and using the Phillpot and Zillman (1970) relationship between surface temperature and the temperature of the inversion layer. Over the Southern Ocean (45 60 S), local evaporation accounts for 50% of precipitation, and this evaporative flux (mean d 18 O e of 1%) increases the d 18 O of vapor (mean d 18 O v of 16%). During the Last Glacial Maximum (LGM; 21,000 years ago), the isotopic composition of the vapor near the ice edge (60 S) is calculated to be similar to the present values because evaporative recharge also accounts for 50% of the precipitation over the Southern Ocean. As a result, the isotopic composition of vapor during the LGM is close to the present values at the ice edge. The apparent temporal slope over eastern Antarctica is half of the observed spatial slope. Our LGM experiment estimates an Antarctic mean annual temperature decrease of 13 C at Vostok, much larger than previous estimates. Our experiments with two specifications of LGM sea surface temperatures suggest that the value of the temporal slope is related to the temperature decrease over the Southern Ocean. Citation: Lee, J.-E., I. Fung, D. J. DePaolo, and B. Otto-Bliesner (2008), Water isotopes during the Last Glacial Maximum: New general circulation model calculations, J. Geophys. Res., 113,, doi: /2008jd Introduction [2] One of the most important applications of water isotopes is the estimation of the glacial-interglacial cycle of temperature from ice cores [e.g., Lorius et al., 1985]. On the basis of the pioneering work by Dansgaard [1964], the relationship between the mean annual surface air temperature (T a ) and the d 18 O of precipitation (d 18 O p ) has been used to estimate paleotemperature, especially over the polar regions. Other proxy materials that include oxygen or deuterium isotopic composition in climatic archives such as in cave deposit, lake and ocean sediments, and of plant material, have proven useful for paleoclimate reconstructions [Gat, 1996]. For example, speleothem d 18 O indicates weaker Asian monsoon [Wang et al., 2001] and stronger northeastern Brazil precipitation during the glacial periods [Cruz et al., 2005]. 1 Department of Earth and Planetary Science, University of California, Berkeley, California, USA. 2 National Center for Atmospheric Research, Boulder, Colorado, USA. Copyright 2008 by the American Geophysical Union /08/2008JD [3] For temperatures lower than 15 C, the global empirical annual mean relationship is d 18 O p = 0.69 T a 13.6 [Dansgaard, 1964], where T a is temperature in Celsius and d 18 O p is in %. This empirical relationship is derived from present-day surface conditions around the globe, and is thus a spatial (mainly latitudinal) T a d 18 O p relationship. This should be distinguished from a temporal relationship, which describes the variation of d 18 O p with T a in a given geographical location through different time-varying climate regimes. Dansgaard [1964] chose 33 locations (mostly Greenland, Antarctica, and coastal regions of Atlantic ocean) to explain the observed relationship between temperature and the isotopic composition of the surface water with Rayleigh distillation model. For temperatures ranging between 0 and 20 C, the slope between temperature and d 18 O is 0.58%/ C [Rozanski et al., 1993]. [4] The T a d 18 O p relationship has been explained by the Rayleigh distillation model [Dansgaard, 1964]. The Rayleigh distillation model links the isotopic depletion with the temperature gradient between source and site. As an air parcel moves poleward from the tropical/subtropical ocean, temperature drops and condensation occurs. Since heavier isotopes are rained out preferentially, the remaining vapor and ensuing precipitation, in turn, has lower ratio of heavy 1of15

2 isotopes as the air parcel moves toward colder regions. In the Rayleigh distillation model, d 18 O p is determined only by the initial isotopic composition of the vapor (d 18 O v ) and the local temperature. The simplest Rayleigh model does not include the recycling of water via rainout and evaporative recharge between the subtropics and the poles. Evaporative recharge is expected where evaporation (E) exceeds precipitation (P) at the surface and because the turnover time of water vapor in the atmosphere is of order 10 days, much less than the advective time from the subtropics to the poles. Hendricks et al. [2000] points out the importance of the recharge of water vapor through the local evaporation in determining the isotope content of precipitation in oceanic regions. Thus, the d 18 O p should depend not only on local temperature, but also on the path history of the parcel and the relative time spent over the ocean. [5] In Greenland, the use of the spatial isotope temperature slope has been challenged by alternative paleothermometry methods such as the inversion of the borehole temperature profile [Cuffey et al., 1994; Johnsen et al., 1995], the thermal and gravitational diffusion of air in the firn arising during abrupt climate changes [e.g., Severinghaus et al., 1998]. White et al. [1997] compare contemporaneous observations of temperature and d 18 O p from the snow over Greenland and their temporal slope was about 0.29%/ C, smaller than 0.69%/ C of the present-day spatial slope. Cuffey et al. [1994] estimate past temperatures by measuring borehole temperature profiles and taking into account heat diffusion within fern and ice using an ice flow model. For Greenland, the estimated temporal slope is about half of the present-day spatial slope [Cuffey et al., 1994]. Temperature estimation from the thermal fractionation of gases in the fern in Greenland, Severinghaus et al. [1998] and Severinghaus and Brook [1999] obtain a temporal slope that is about half of the Dansgaard spatial slope. Two general circulation model studies have shown that changes in the seasonal precipitation timing in central Greenland might have caused a warm bias in the LGM water isotope proxy temperatures, and that this bias could explain the difference in the estimated paleotemperatures [Krinner et al., 1997; Werner et al., 2000]. [6] Boyle [1997] proposes that the slope of the T a d 18 O p relationship for polar snow remains constant at all times, but that the intercept varies in a simple fashion determined by changes in tropical sea surface temperature (SST). To obtain 0.35%/ C [Cuffey et al., 1994], he concluded that tropical ocean cooled about 5 C during the LGM. Hendricks et al. [2000], on the other hand, argues that distillation effect is not important until midlatitude oceanic regions because of evaporative recharge in the subtropics. They concluded that SST at midlatitudes around 45 S, not the tropics, is important initial temperature of distillation for Antarctica. [7] The validity of the LGM isotope thermometry for the Vostok core in central East Antarctica has been also challenged. Salamatin et al. [1998] applies borehole thermometry to Antarctica, and obtains a temperature change about twice as large temperature drop as the estimate by spatial slope. Gas fractionation method shows up to 35% larger temperature drop than the spatial slope for Antarctica [Caillon et al., 2001]. Both borehole and gas diffusion methods have been cri for the use over Antarctica [Jouzel et al., 2003]. Borehole methods, in particular, have been criticized because the low accumulation that prevails at East Antarctic inland sites such as Vostok could erase the glacial-interglacial surface temperature signal at the depth of the Last Glacial Maximum (LGM) and prevent accurate estimation of the glacial-interglacial temperature change at Vostok [Jouzel et al., 2003, and references therein]. Jouzel et al. [2003] conclude that the present-day spatial slope would correctly interpret glacial-interglacial temperature change within 20 30% accuracy at sites such as Vostok and European Project for Ice Coring in Antarctica (EPICA) Dome C. A recent paper by Masson-Delmotte et al. [2008] gives an excellent review on usages and limitations of water isotopes over Antarctica. [8] In this study, we re-examine the relationship between temperature and d 18 O p, in particular for Antarctica using a water isotope-enabled atmospheric general circulation model, for both the climatic conditions of the present day and Last Glacial Maximum (LGM; 21,000 years ago). We have developed an isotope module for the National Center for Atmospheric Research (NCAR) Community Atmospheric Model Version 2 (CAM2), and have shown the importance of the evaporative recharge in determining the d 18 O p over the oceanic regions [Lee et al., 2007] (hereinafter referred to as L2007). 2. Model Description [9] We incorporated HDO and H 18 2 O into the NCAR CAM2. Our incorporation is similar to the previous isotope-gcm studies [Joussaume et al., 1984; Hoffmann et al., 1998; Noone and Simmonds, 2002; Schmidt et al., 2005], with fractionation associated with phase changes. The details of the model can be found in the work of L2007. The global distribution of water isotopes in precipitation is reasonably simulated, and discrepancies between observed and modeled d 18 O p can be traced to errors in the precipitation simulation (L2007). The limitations of the atmospheric model and isotopic scheme are described by L2007. The mass fixer in CAM guarantees the conservation of global mass of water vapor and other tracers. The application of the mass fixer does not alter the conclusion of our paper, as the same algorithm is applied to all isotopologues of water. [10] We ran the isotope-enabled NCAR CAM2 with fixed SST and sea ice distributions for the present day and the LGM. Present-day SST is the climatological monthly mean derived from observations from 1949 to In the LGM run, we used monthly SST and sea ice distribution simulated by the fully coupled atmosphere-land-ocean-ice Community Climate System Model [Otto-Bliesner et al., 2006] with atmospheric carbon dioxide (CO 2 ), methane (CH 4 ), and nitrous oxide (N 2 O) at 185 ppm, 350 ppb, and 200 ppb respectively, and with the continental ice sheet extent and topography prescribed from the LGM ICE-5G reconstruction [Peltier, 2004]. There are large uncertainties in the LGM conditions including the topography. Krinner and Genthon [1998] illustrated the importance of the choice of the LGM ice sheet topography. They show that two different topographies with 1 km or more altitude difference over central Greenland yield quite different results. This means that the paleotopography is a significant source of uncertainty for the modeled paleoclimate. The LGM coastline is 2of15

3 Figure 1. The extent of sea ice for (a-c) the LGM, (d-f) present-day, and (g-i) difference for summer (December-February; DJF), winter (June-August; JJA), and annual (ANN) means. also specified by the ICE-5G reconstruction and corresponds to a globally averaged lowering of sea level of 120 m. More details in the LGM configuration are given by of Otto-Bliesner et al. [2006]. Surface ocean d 18 Os for the present day and LGM are 0.5 [Hoffmann et al., 1998] and 1.7% [Schrag et al., 1996], respectively. [11] The LGM simulation is reasonable and has validated with the available proxy data [Otto-Bliesner et al., 2006]. Figures 1 and 2 show the sea ice fraction and SST, respectively of the Southern Hemisphere for the present day and LGM. During the LGM, Southern Hemisphere sea ice area is about twice as large as the present [Otto-Bliesner et al., 2006]. The LGM CCSM simulation has a global mean annual sea surface cooling of 6.3 C compared to the present day. The tropical mean annual SST cooling is 2.6 C and mean SST difference increases poleward reaching 10 C at the sea ice edge. Our SST in Figure 2 is SST over sea ice and ocean water. Over open water, the SST decrease during the LGM is 3 C at45 S. Consistent with paleoproxy data, ITCZ shifted southward, and the Asian monsoon is weakened in the LGM simulation. As a result, d 18 O p decreases in the southern part of the current ITCZ and increases over Asian monsoon regions. [12] To illustrate the effect of changes in the SST gradient, we conducted an additional experimental run wherein present-day SST was lowered by 2 C everywhere. This corresponds to the evaporation-weighted sea surface temperature difference between the present day and LGM (1.9 C). This coincides with the LGM source temperature change inferred from the deuterium excess record from the Vostok ice core [Vimeux et al., 2002] Sea ice is prescribed where SST is lower than 1.8 C. The result of this experiment (PRS-2) is discussed in section 4. [13] The isotope-cam LGM simulation is initialized using the atmospheric state from the equilibrium simulation of the CCSM LGM run, and is integrated forward for 20 years using the CCSM SST s and glacial and sea ice extents as boundary conditions. The present-day simulation was integrated for 15 years, while the PRS-2 was integrated for 20 years from the CCSM LGM run. In all three cases, averages of the last 10-year integrations were used for our analysis. 3of15

4 Figure 2. (a-c) SST of the LGM (blue) and present-day (red), as well as (d-f) the difference. [14] In examining T a d 18 O p relationships in a model, there is an ambiguity about the appropriate temperature to use. Temperature inversions occur where there is strong radiative cooling at the surface, e.g., over Antarctica and Greenland. In these regions, a model s ability to diagnose surface air temperature from the prognosticated temperature in the free troposphere is tied to the thickness of the inversion layer. Krinner and Werner [2003] note that annual mean surface temperatures T a are overestimated in most of the GCM simulations, both for Vostok and for Summit, Greenland, implying difficulty with modeling the thickness of the inversion layer. This is the case with the NCAR climate model we are using [Collins et al., 2006]. Our GCM overestimates mean annual temperature at the surface of the atmosphere over Antarctica by 10 C, but inversion T is only 3 C warmer than the observation [cf. Guo et al., 2003, Figure 5], implying the mismatch is local effect. Isotope concentrations capture the temperature of condensation and interaction with the vapor in the atmospheric column rather than of surface air temperature. We calculated condensation temperatures as the column-averaged temperature weighted by the condensation rate at each level. Figure 3 shows the relationship between the simulated annual-mean condensation temperature, surface temperature and inversion temperature (T i ) for the present day and the LGM. T i is defined as the maximum temperature below 500 hpa. Poleward of 60 S, the difference between inversion and condensation temperatures remains relatively uniform for both the present and the LGM, suggesting that the latitudinal gradient in inversion temperature may capture the latitudinal gradient of the condensation temperature. In our model, the spatial slope between T i and d 18 O p compares well with that observed (a slope of 1.20 versus 1.25 of the observation from Dahe et al. [1994] and using the Phillpot and Zillman [1970] relationship between surface T a and T i ). Compared to this spatial slope between T i and d 18 O p,the slope between T a and d 18 O p is not as good (a slope of 1.04 compared to 0.84 of the observation from Dahe et al. [1994]) probably because the inversion layer is not well simulated in our model. Nevertheless, we used surface temperature for our analysis below because we do not have enough knowledge on the inversion thickness during the LGM. We also analyzed our results with the inversion layer, and we confirmed that it does not change our conclusion. We shall use the notation Df to indicate the difference (LGM minus present day) in variable f between the LGM and the present. 3. Results 3.1. Role of Evaporative Recharge of Water Vapor [15] Evaporation is driven by the gradient in vapor pressure between a thin molecular layer in equilibrium with Figure 3. Inversion temperature (T i ), condensation temperature (T c ), and atmospheric surface temperature (T a )for the present day. 4of15

5 Figure 4. The zonal mean isotopic composition of the evaporative flux, d 18 O e (red solid line), and the mean surface vapor, d 18 O v (blue dashed line), over the oceanic regions. From equation (1), R e increases with the increase of relative humidity and also with the increase of the 18 O gradient in between the interfaces. Since R* is relatively constant (it changes owing to the temperature dependence of the fractionation coefficient a), R e increases when R v decreases. We shall refer to d 18 R O e = ( e R SMOW 1) as the isotopic composition of the evaporative flux (R SMOW is the 18 Oto 16 O ratio in the standard mean ocean water). In high-latitude oceanic regions with heavy precipitation, d 18 O v is greatly depleted via isotopic exchange with raindrops, resulting in high 18 O v gradient between the molecular layers and the atmospheric boundary layer. The evaporative flux is thus highly enriched in 18 O. Craig and Gordon [1965] point out that d 18 O e can be 4% for a d 18 O v of 20%. As far as we the ocean surface and the atmosphere immediately above it. Water vapor near the ocean surface is depleted in 18 O relative to vapor in equilibrium with the ocean surface [Craig and Gordon, 1965; Lawrence et al., 2004]. This relative depletion could result from isotopic exchange with raindrops that condensed in the upper troposphere where there is very low d 18 O v [Araguas-Araguas et al., 2000], and could be significant even during a brief light rain [Lee et al., 2007]. Because the Southern Ocean is a stormy region, atmospheric vapor over the Southern Ocean is greatly depleted in 18 O compared to the equilibrium values with the ocean (Figure 4). [16] We employ d 18 O e (and corresponding R e, the ratio of 18 Oto 16 O) to characterize the isotopic signature of evaporation. Following Hoffmann et al. [1998], we can write E 16 and E 18, the evaporative flux of 16 O and 18 O respectively, as the following: E 16 ¼ C 16 ð1 hþ E 18 ¼ C 18 ð 2 R* h R v Þ ¼ C 18 4 R* {z} vap in molecular layer h ffl{zffl} R* vap in eqm with ocn ð ÞŠ ¼ C 18 2 ½R* h R* þ h R* R v ¼ C 18 4ð1 hþr* þ h ðr* R fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} v Þ5 fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} iso eqm iso diseqm þh ðr* R v 3 Þ5 where E is the evaporation flux, C is the drag coefficient, h is the relative humidity, R* is the isotope ratio in equilibrium with ocean water (R* =a 1 R o where a is temperature-dependent fractionation coefficient and R o is the isotope ratio of the ocean surface water), and R v is the isotope ratio in atmospheric vapor. If we define E 18 = R e E 16 and C 18 = C 16 (1 k 18 )[Hoffmann et al., 1998], the isotope ratio of evaporative flux, R e, can be written as R e ¼ ð1 k 18 ÞR* þ h h ð 1 k 18ÞðR* R v Þ ð1þ Figure 5. (a) Isotopic composition of precipitation over the land regions (gray) and the ocean regions (light blue). Red and blue squares are the values at 40 and 63 N over Europe, and the location is plotted in Figure 5c. The lines are regression lines at 40 and 63 N values. (b) Modeled surface T a d 18 O p relationship for the same stations in Dansgaard (red triangles) with regression line (red for the model output and black for the Dansgaard relationship), and two distillation pathways over Europe (blue crosses; 52 N and 57 N from left to right). Gray and light blue crosses are the same as in Figure 5a. (c) Locations of the regions where Dansgaard included in the relationship d 18 O p = 0.69T a 13.6 (red crosses) and of regions that shows different distillation pattern in Figure 5a. 5of15

6 Figure 6. Evaporation and precipitation. are aware, there is no measurement on the isotopic composition of the evaporative flux over the ocean. Over the land, ecologists have been measuring the isotopic composition of the evaporative flux [e.g., Lai et al., 2006], and the conclusion is that Craig-Gordon model (which is the basis of our evaporative flux equation) is a good approximation. [17] Figure 4 shows the isotopic composition of the evaporative flux and vapor above the ocean surface for the present-day simulation. When d 18 O v is low, the isotopic composition of the evaporative flux is high. As about half of precipitation over the Southern Ocean is derived from evaporation within the region, a large part of the d 18 O variation in precipitation over the ocean can be explained by the balance between local precipitation and evaporation (L2007). [18] The influence of high d 18 O e is shown in the presentday difference between the land and ocean T d 18 O p relationship (Figure 5). While d 18 O p decreases with temperature for regions with T a <15 C, vapor over the oceans is isotopically enriched relative to that over land at the same surface temperature. For T a >15 C, there is no apparent relationship between T a and d 18 O p because precipitation amount and other factors have greater influence than temperature on d 18 O p. The T a d 18 O v slope for ocean points is 0.24%/ C, smaller than 0.57%/ C for land points at T a > 15 C within 60 S 60 N. The correlation coefficients at T a >15 C within 60 S 60 N for ocean and land points are 0.74 and 0.82 respectively. Observation also shows increasing slope with T a [Masson-Delmotte et al., 2008]. Antarctica T a d 18 O p slopes vary from 0.60 to 0.91 % per C. [19] Superposed in Figure 5a are the values for 40 N (red squares) and 63 N in Europe (cf. Figure 5c). At these latitudes, the winds are westerly and relatively zonal, and temperature decreases eastward away from the ocean. The initial d 18 O v values (at high temperature and near the oceans) are more depleted at 63 N than at 40 N because there is less evaporation at the higher latitude, so that d 18 O p near the coast is more depleted at 63 N than 40 N. Nevertheless, at the same temperature, say 5 C, the d 18 O p at 63N is more enriched than d 18 O p at 40 N because the distillation path since last evaporative recharge is shorter. The slopes are 0.59 and 0.86 for 40 N and 63 N. These slopes can be interpreted as the influence of varying degrees of recharge. [20] Figure 5b shows the relationship for the Dansgaard stations: 6 island stations, 17 continental stations from the North Atlantic and polar regions including Greenland and Antarctica (cf. Figure 5c). A linear relationship exists among the Dansgaard stations T a d 18 O p in our model (red crosses in Figure 5c). The regression line (d 18 O p = 0.68T a 12.7; red line in Figure 5b) is quite similar to the Dansgaard relationship (d 18 O p = 0.69T a 13.6; black line in Figure 5b). When T a > 0 the Dansgaard slopes are less steep (0.35%/ C) than the distillation slopes inland at 40 N or 63 N discussed above because the stations are located all on the coast, further suggesting the different air mass origins and importance of evaporative recharge along the different air mass trajectories. Figure 7. The proportion of snow comprising precipitation. 6of15

7 Figure 8. Zonal mean of vertical distribution of d 18 O v. Figure 9. Zonal mean of (a-c) d 18 O e and (d-f) d 18 O p. 7of15

8 Figure 10. T a d 18 O v for the LGM (blue) and present-day (red) over the ocean. Latitudinal mean temperatures for the present day and LGM are added at the bottom Comparing the LGM and Present-Day Southern Ocean Condition [21] During the LGM, evaporation and precipitation are reduced compared to the present day (Figure 6). Particularly noteworthy is the significant reduction of winter (June- August; JJA) evaporation over S, which is icecovered during the LGM (cf. Figure 1). Evaporation during the LGM summer (December-February; DJF) remains similar to the present values despite the SST drop (5 7 C colder during the LGM; cf. Figure 2) because of the greater decrease in air temperature (thus larger vapor pressure gradient) and stronger winds (not shown). On average, mean annual (ANN) evaporation during the LGM is 20% lower (3.0 versus 2.4 mm/day) over S. [22] Overall, DP over the Southern Ocean follows DE. The LGM atmosphere is significantly drier with a 26% decrease in precipitable water and 11% decrease in global precipitation. Globally, DP sensitivity is similar to the IPCC AR4 scenario (2.2%/ C; Held and Soden [2006]). The regions where P > E is shifted toward the equator during the LGM, suggesting an equatorward shift of storm tracks. [23] The degree of isotopic interaction between precipitation and water vapor changes with temperature and precipitation rate [Lee and Fung, 2008]. In high-latitude oceanic regions where temperature is low and precipitation rate is high, precipitation tends to be more depleted in heavy isotopes (1%) than the precipitation in equilibrium with the surface water vapor [Lee and Fung, 2008]. This means that vapor is relatively less depleted in 18 O compared with lower latitudes. In addition, falling snow has very little isotopic interaction with atmospheric vapor that snow does not deplete surface vapor. Because of the colder LGM temperatures and hence a greater fraction of snow compared to rain (Figure 7), there is reduced depletion of vapor isotopes owing to interaction between the vapor and raindrops. Figure 8 shows the vertical distribution of mean d 18 O v over S, and the difference between the LGM and present-day simul. In the upper troposphere, d 18 O v is lower during the LGM than the present because of the colder temperatures. Near the surface, however, d 18 O v is higher during the LGM than the present because of the reduced interaction between the precipitate and the vapor. Overall, the zonal mean LGM d 18 O v distribution is within 1% of the present-day values outside of the ice-covered regions (60 S 40 N) where significant evaporation occurs over open water (not shown). [24] Mean d 18 O e of the evaporative flux is a function of vapor gradient and relative humidity (cf. equation (1)). Over S, d 18 O e during the LGM is lower than the present day Figure 11. Mean monthly (a) precipitation the present day (red) and LGM (blue) and (b) T a difference (LGM-PRS) at Vostok. 8of15

9 Figure 12. Antarctica. DT a (LGM-PRS), Dd 18 O p (LGM-PRS), and DdD p (LGM-PRS) for regions surrounding because of a smaller isotopic gradient resulting from the more enriched LGM atmospheric surface vapor (Figures 9a 9c). At low latitudes, mean d 18 O e is higher during the LGM because ocean water d 18 Ois1.2% higher than the present. d 18 O p, on the other hand, shows relatively large differences between the LGM and present day compared with d 18 O v (Figures 9d 9f). As discussed above, surface vapor is more enriched in 18 O during the LGM because snow does not interact with surface vapor. In the same way, the isotopic composition for snow is very low because it does not see the more enriched atmospheric surface vapor. [25] The d 18 O p of Antarctic precipitation varies mainly with the isotopic composition of vapor, and our modeled LGM d 18 O v values are close to those today in spite of the large SST difference, as d 18 O v depletion owing to precipitation is in near-balance with enrichment through evaporation. As long as the vapor is continually recharged over open water, the relationship between T a and d 18 O must deviate from the Rayleigh distillation model. [26] Because LGM temperature is 12 C lower than the present day around 60 S, but d 18 O v has very small variation, the LGM T d 18 ationship has 4% offset for 5 C< T a <0 C compared with the present day (Figure 10). The offset is smaller at higher temperature because T a decrease is small at low latitude. The offset around 20 C is 1.2%, most of which is explained by the higher d 18 O o during the LGM. The offset is 2% around 10 C T a d 18 O p Relationship Over Antarctica [27] In the previous two sections, we elaborated on the importance of evaporation from midlatitude to high-latitude oceanic regions (45 60 S) and how the evaporation keeps LGM d 18 O v over open waters similar to the present-day values. In this section, we focus on the T a d 18 O p relationship over Antarctica, and examine the difference between the temporal and spatial relationships. [28] Inferring annual temperature during the LGM from precipitation isotopes requires information about changes in the seasonality of precipitation as well as changes in the seasonality of temperature. Many GCM studies of the LGM attribute the differences between spatial and temporal slope between T a and d 18 O p to changes in the seasonality of precipitation [Krinner et al., 1997; Werner et al., 2000]. Contemporary observations show that annual precipitation 9of15

10 Figure 13. (a) T a d 18 O p relationship for the present day (orange) and LGM (light blue) over Antarctica. Symbols represent the values at four locations plotted in Figure 13b. (b) Temporal slope (DT a /Dd 18 O p ) between the present day and LGM. Blue line in Figure 13b is where the LGM ice fraction is 90%. over Antarctica is episodic, and determined by several heavy snowfall events [Bromwich, 1988]. The observations show slightly greater (25% of total annual precipitation during JJA compared to 10% during DJF) precipitation during winter, but Bromwich [1988] emphasizes that measuring precipitation is very difficult because of snow drifts in strong winds. Ekaykin et al. [2002] shows relatively small changes in seasonality of precipitation around Vostok station. In our model, the change in precipitation seasonality is small over Antarctica. At Vostok, present-day summer precipitation (DJF: 14.7 mm and 47% of annual precipitation) is higher than winter precipitation (JJA, 4.6 mm and 15% of annual precipitation) (Figure 11a). The annual-mean LGM precipitation is only 1/4 of the present day, but summer precipitation accounts for 66% of annual precipitation, and winter precipitation contributes only 9%. The small difference (6%) between the LGM and present-day winter precipitation contributions is not large enough to explain the difference in the T a d 18 O p relationship we obtain. Rather, a large part of the difference in the relationship comes from much greater temperature decrease in winter (19 C duringjulycomparedto10 C during January; Figure 11b), about double that in summer (Figures 11b and 12a 12c). Combined with the changes in precipitation seasonality, the greater winter temperature decrease leads to a significant decrease temporal T a d 18 O p slope compared with the present-day spatial T a d 18 O p slope. The problem may be not only seasonality but also intermittency of precipitation. The days of snowfall can be associated with higher air temperatures than the rest of the days. There have been some efforts to monitor the daily precipitation isotopic composition in Antarctica [Fujita and Abe, 2006], suggesting the difference between isotopic temperature and mean surface temperature could be as large as 8 C. [29] Figures 12d 12f show the seasonal and mean annual d 18 O p differences between the present day and LGM for regions around Antarctica. Because of the increase in sea ice over the Ross Sea and Weddell Sea, the DT a difference between eastern and western Antarctica becomes smaller during the LGM, and DT a is greatest over western Antarctica. The d 18 O difference around the coastal region is small because the vapor entering Antarctica has similar d 18 O (section 3.4). The Dd 18 O increases toward western Antarctica with the increase in DT a. [30] Figure 13 shows two T a d 18 O p relationships for Antarctica for the present day (red) and LGM (blue). The symbols are mean annual values for each grid point, and the temperature plotted is that at the surface. The LGM spatial slope (0.88%/ C) is close to, but somewhat smaller than, the present-day slope (1.04%/ C). With T i, the slopes are similar (1.20 versus 1.15 for the present day and LGM respectively; not shown), showing the similar trends in the condensation temperature (cf. Figure 3) and the lack of evaporative recharge over the Antarctic continent. The slopes using surface temperature are different because the difference between T i and T a increases more with temperature decrease during the LGM. The outstanding feature in Figure 13 is the displacement of the intercepts at the two different climate regimes as is expected from Figure 11. [31] Figure 13b shows the temporal slope in T a d 18 O p estimated at each location as Dd 18 O p /DT a. At the ice edge (ca. 60 S; blue line in Figure 13), Dd 18 O v is close to 0 (Figure 8). Since DT becomes larger in the inland direction, the temporal slope becomes larger inland. For example, the measured jdd 18 O p j at Taylor Dome is only 3% (Table 1) since Taylor Dome is near the ice edge and d 18 O v coming to Taylor Dome during the LGM is close to the present value. Table 1. Modeled Mean Temporal Slope S mdl From Dd 18 O p and DT a Between the LGM and Present Day, Measured LGM-Present- Day d 18 O p Differences, and Temperature Differences Estimated From Modeled Slopes and Measured d 18 O p Differences for Seven Ice Core Sites in Antarctica and Greenland a Station S mdl (%/ C) jdd 18 O p j obs (%) jdt a j est ( C) Vostok b 13 Dome C c 14 Byrd d 14 Taylor Dome d 11 Camp Century e 24 NGRIP f 22 GRIP g 21 a The location of the sites are plotted in Figures 14b and 16b. b Lorius et al. [1985]. c Lorius et al. [1979]. d Grootes et al. [2001]. e Dansgaard et al. [1982]. f NGRIP Members [2004]. g GRIP Members [1993]. 10 of 15

11 Figure 14. Greenland. DT a (LGM-PRS), Dd 18 O p (LGM-PRS), and DdD p (LGM-PRS) for regions surrounding As a result, the temporal slope at Taylor Dome is very small, and the jdt a j is greater than the values predicted by the present-day spatial relationship. At Vostok, however, the slope becomes a lot larger because jdd 18 O p j at Vostok is larger than jdd 18 O p j at Taylor Dome. Even if the jdd 18 O p j at Vostok is 70% greater than that of the Taylor Dome, jdt a j is only 2 C (13 versus 11 C). The slope shows higher values over the regions where jdt a j is the highest. Measurements from ice cores confirm that coastal regions like Taylor dome have smaller differences in 18 O between the LGM and present day [Grootes et al., 2001]. Table 1 shows the temporal slopes from our model results, d 18 O p differences between the present day and LGM from the ice core measurements, and the temperature difference between the present day and LGM estimated from our model slope and observed Dd 18 O p. [32] Other coastal ice cores such as the Law Dome, Siple Dome, and EPICA Dronning Maud Land show significantly higher temperature decreases that the slopes are greater than the slope for the Taylor Dome, consistent with the similar range of d 18 Op decreases over these ice cores [Morgan et al., 2002; Brook et al., 2005; EPICA Community Members, 2006] Greenland [33] Figure 14 shows DT a and D d 18 O p, for Greenland and Figure 15 shows the temporal slope (DT a /D d 18 O p )for the present day and LGM. Consistent with our explanation in the previous section, the Greenland slope is smaller than Antarctica because Greenland is closer to the ice edge. At the ice edge of the LGM, the d 18 O v is similar to the presentday values, and the slope is closer to 0. Since the temperature drop becomes larger northward, the slope becomes larger away from the ice edge. North Greenland Ice Core Project (NGRIP) Members [2004] showed that the amplitude for NGRIP (76.1N, 42.3W) is generally higher than that of the GRIP (72.3N, 38.8W) record. Whereas NGRIP and GRIP have very similar d 18 O p levels during the Holocene, glacial isotopic levels in the NGRIP record are systematically depleted by 1% to 2%. The difference between two sites should not be interpreted as 3 6 C cooling because NGRIP has a higher temporal slope. For the LGM case, the 1% difference is translated into only 11 of 15

12 1 C ofjdt a j (Table 1). The difference curve also compares relatively well to the global sea level curve [NGRIP Members, 2004], and it is explained to be related to the extent of the glacial continental ice sheet. We believe that this is related to the extent of sea ice (which is also related to land ice extent), and the difference in d 18 O p becomes higher when GRIP or NGRIP is farther from the open ocean. Because the temporal slope is relatively large for the Camp Century site, the temperature drop during the LGM is only slightly higher (2 3 C) than GRIP or NGRIP (Table 1) even if the d 18 O p difference is greater by 2 3%. [34] Our GRIP estimation (21 C) is close to the borehole thermometry by Johnsen et al. [1995] (24 C) and other model simulation results. Using ECHAM, Werner et al. [2001] also shows similar temperature decrease (22 C) to ours. SST in the work of CLIMAP Project Members [1976] (climate long-range investigation, mapping, and prediction) is more similar to our input SST over the Northern Hemisphere. Figure 15. (a) T a d 18 O p relationship for the present day (orange) and LGM (light blue) over Greenland. Symbols represent the values at four locations plotted in Figure 15b. (b) Temporal slope (DT a /Dd 18 O p ) between the present day and LGM. Blue line in Figure 15b is where the LGM ice fraction is 90%. 4. Effect of Nonuniform SST Changes [35] Crucial to the Rayleigh distillation model is the assumption of a single, or mean subtropical source vapor for polar precipitation. Previous GCM studies [Delaygue et al., 2000; Armengaud et al., 1998; Werner et al., 2001] have shown that polar precipitation originates from a mixture of water vapor originating from tropical and local evaporation. [36] When all source region temperature is lowered by 2 C (PRS-2 run), the offset between PRS and PRS-2 in the T a -d 18 O v relationship is smaller than the offset between PRS and LGM runs (Figure 16). We show summer results because the T a drop is more uniform during summer as a result of the similar distribution of sea ice between PRS and PRS-2 runs. Figure 17 confirms that the offset is very small in Antarctic precipitation. The mean apparent slope Figure 16. T a d 18 O v for the PRS-2 (orange) and present-day (red) runs. LGM values (blue) are also plotted as a reference. Latitudinal mean temperatures for the present day and PRS-2 runs are added at the bottom. 12 of 15

13 Figure 17. T a d 18 O p for the PRS-2 (blue) and presentday (red) runs over Antarctica PRS-2 during summer (December-February). over Antarctica is 0.9%/ C, closer to the spatial slope (1.0%/ C) compared to the mean temporal slope of 0.49%/ C between the present day and LGM. 5. Conclusion [37] To illustrate whether we can use the present-day spatial slope to estimate T a over Antarctica during the LGM, we performed isotope-enabled AGCM runs for the present day and LGM. In the above sections, we showed that the oceanic vapor is by far more enriched than that predicted by the distillation model because of the evaporative recharge from the open ocean. At the same time, changes in the precipitation type toward higher proportion of snow during the LGM also leaves vapor relatively enriched. As a result, d 18 O p of the vapor transported southward from 60 S during the LGM is similar to the present values (Figure 18). Thus, the temporal slope near the ice edge is small and much different from the global T a d 18 O p slope. The slope becomes larger inland because jdt a j is larger inland as Hendricks et al. [2000] shows. As a result, T a d 18 O p relationship in one ice core should be different from that of another ice core and should be related to its distance from the coast (Table 1). In addition, the fact that the winter jdt a j is larger than the summer jdt a j, combined with changes in seasonal precipitation, leads to more significant decrease in the temporal T a d 18 O p slope than the present-day spatial T a d 18 O p slope. [38] In our study, the magnitude of the mean temporal slope on Antarctica is dependent on jdsstj, the cooling in the Southern Ocean poleward of 40 S. In our LGM case, jdsstj is 8.1 C, and the temporal slope for Antarctica is 0.42%/ C at Vostok after correcting higher d 18 O in seawater during the LGM. The jdsstj and temporal slope values for our PRS-2 case are 4.8 C and 0.69%/ C (at Vostok), respectively. The temporal slope for the PRS-2 case at Dome C is close to the spatial slope of 0.84%/ C [Dahe et al., 1994]. Jouzel et al. [2007] argues that the spatial and temporal slopes are very similar over a wide range of climatic conditions. We posit here that the spatial and temporal slopes may be similar as long as the temperature departures are small. Jouzel et al. [2007], using CLIMAP Project Members [1976] boundary conditions with ECHAM (European Centre Hamburg Model), estimated 8 C cooling, while our model, using modeled SST Figure 18. Illustration of the fluxes in hydrological cycle and the isotopic composition. Fluxes are in the unit of 10 9 kg/m 2 /s. Black line in bottom panel is mean annual temperature ( C). 13 of 15

14 boundary conditions, estimated a 13 C cooling at Vostok. CLIMAP SST is 2 C warmer at 40 S than our input LGM SST. The temporal slope is 0.75%/ C for Jouzel et al. [2003] for 18 O and 0.42%/ C for our model at Vostok. A few studies have looked at the seasonal isotope-temperature slopes over Antarctica [e.g., Van Ommen and Morgan, 1997], and concluded that the slope is 0.44%/ C, consistent with our study. [39] The dependence of the temporal slope on temperature decrease itself challenges the application of T a d 18 O p relationships to derive temperature. Over Greenland where jdt a js for the ECHAM and out model are similar, Werner et al. [2001] also shows similar temporal slope to ours. SST in CLIMAP is more similar to our input SST over the Northern Hemisphere. The temporal slope we derived in Greenland is similar to that derived other temperature estimates [e.g., Cuffey et al., 1994]. This suggests that the temporal slope depends on the temperature decrease itself. New results from Schmidt et al. [2007] with the GISS isotope-enabled GCM also show that the temporal gradients are significantly smaller than the estimates from previous models and data, more in accord with our results. [40] Our study is limited by the difficulties in simulating LGM climate condition using GCMs. The uncertainties in prescribed albedo, topography, and radiative budget could affect the temperature decrease during the LGM. The comparison between LGM simulations and ice core data is often limited by the problem of a prescribed Antarctic topography. Masson-Delmotte et al. [2005] shows a significant part of the Greenland and Antarctic cooling of the GCM simulations is caused by the prescribed local elevation increase at the LGM. The simulated precipitation during the LGM is 1/4 of the present value, about half of that estimated for Vostok [Siegert, 2003]. However, precipitation rate does not account for the accumulation difference owing to the existence of blowing snow [Bromwich, 1988]. In addition to the model simulation uncertainties, there are great uncertainties in dating ice cores. Vostok ice core has been dated by an ice flow model aided by isotopic stratigraphic control points under an assumption that the rate of accumulation of ice varies proportionally to the derivative of the water vapor saturation pressure, which is itself dependent on temperature [Jouzel et al., 1993; Petit et al., 1999]. [41] Acknowledgments. This work was supported by the NOAA Office of Global Programs grant NA05OAR and NSF grant ATM D.J.D. acknowledges support from the Director, Office of Science, and Office of Basic Energy Sciences, of the U.S. Department of Energy (DOE) under contract DE-AC02-05CH The CAM2-CLM runs were carried out at the DOE s NERSC. The National Center for Atmospheric Research is sponsored by the National Science Foundation. References Araguas-Araguas, L., K. Froehlich, and K. Rozanski (2000), Deuterium and oxygen-18 isotope composition of precipitation and atmospheric moisture, Hydrol. Processes, 14, , doi: / ( )14:8<1341::aid-hyp983>3.0.co;2-z. Armengaud, A., R. D. Koster, J. Jouzel, and P. Ciais (1998), Deuterium excess in Greenland snow: Analysis with simple and complex models, J. Geophys. Res., 103, , doi: /98jd Boyle, E. A. (1997), Cool tropical temperatures shift the global d 18 O-T relationship: An explanation for the ice core d 18 O borehole thermometry conflict?, Geophys. Res. Lett., 24, , doi: /97gl Bromwich, D. H. (1988), Snowfall in high southern latitudes, Rev. Geophys., 26, , doi: /rg026i001p Brook, E. J., J. W. C. White, A. S. M. Schilla, M. L. Bender, B. Barnett, J. P. Severinghaus, K. C. Taylor, R. B. Alley, and E. J. Steig (2005), Timing of millennial-scale climate change at Siple Dome, west Antarctica, during the last glacial period, Quat. Sci. Rev., 24, , doi: / j.quascirev Caillon, N., J. P. Severinghaus, J.-M. Barnola, J. Chappellaz, J. Jouzel, and F. Parrenin (2001), Estimation of temperature change and of gas age -ice age difference, 108 kyr B. P., at Vostok, Antarctica, J. Geophys. Res., 106, 31,893 31,901, doi: /2001jd CLIMAP Project Members (1976), The surface of ice-age earth, Science, 191, , doi: /science Collins, W. D., P. J. Rasch, B. A. Boville, J. J. Hack, J. R. McCaa, D. L. Williamson, and B. Briegleb (2006), The formulation and atmospheric simulation of the Community Atmospheric Model version 3 (CAM3), J. Clim., 19, , doi: /jcli Craig, H., and L. I. Gordon (1965), Deuterium and oxygen 18 variations in the ocean and the marine atmosphere, in Stable Isotopes in Oceanographic Studies and Paleotemperatures, edited by E. Tongiorgi, pp , Cons. Naz. Del Ric., Pisa. Cruz, F. W., S. J. Burns, I. Karmann, W. D. Sharp, M. Vuille, A. O. Cardoso, J. A. Ferrari, P. L. S. Dias, and O. Viana (2005), Insolationdriven changes in atmospheric circulation over the past 116,000 years in subtropical Brazil, Nature, 434, Cuffey, K. M., R. B. Alley, P. M. Grootes, J. M. Bolzan, and S. Anandakrishnan (1994), Calibration of the d 18 O isotopic paleothermometer for central Greenland using borehole temperatures, J. Glaciol., 40, Dahe, Q., J. R. Petit, J. Jouzel, and M. Stievenard (1994), Distribution of stable isotopes in surface snow along the route of the 1990 International Trans-Antarctica Expedition, J. Glaciol., 40, Dansgaard, W. (1964), Stable isotopes in precipitation, Tellus, 16, Dansgaard, W., H. B. Clausen, N. Gundestrup, C. U. Hammer, S. J. Johnsen, P. M. Kristinsdottir, and N. Reeh (1982), A new Greenland deep ice core, Science, 218, , doi: /science Delaygue, G., V. Masson, V. J. Jouzel, R. D. Koster, and R. J. Healy (2000), The origin of Antarctic precipitation: A modeling approach, Tellus, Ser. B, 52, Ekaykin, A. A., V. Y. Lipenkov, N. I. Barkov, J. R. Petit, and V. Masson- Delmotte (2002), Spatial and temporal variability in isotope composition of recent snow in the vicinity of Vostok station, Antarctica: Implications for ice-core record interpretation, Ann. Glaciol., 35, , doi: / EPICA Community Members (2006), One-to-one coupling of glacial climate variability in Greenland and Antarctica, Nature, 444, , doi: /nature Fujita, K., and O. Abe (2006), Stable isotopes in daily precipitation at Dome Fuji, East Antarctica, Geophys. Res. Lett., 33, L18503, doi: /2006gl Gat, J. R. (1996), Oxygen and hydrogen isotopes in the hydrologic cycle, Annu. Rev. Earth Planet. Sci., 24, Greenland Ice-Core Project (GRIP) Members (1993), Climate instability during the last interglacial period recorded in the GRIP ice core, Nature, 364, , doi: /364203a0. GRIP members (1993), Climate instability during the last interglacial period recorded in the grip ice core, Nature, 364, Grootes, P. M., E. J. Steig, M. Stuiver, E. D. Waddington, and D. L. Morse (2001), The Taylor Dome Antarctic 18 O record and globally synchronous change in climate, Quat. Res., 56, , doi: / qres Guo, Z., D. H. Bromwich, and J. J. Cassano (2003), Evaluation of Polar MM5 simulations of Antarctic atmospheric circulation, Mon. Weather Rev., 131, , doi: / (2003)131<0384:eopmso>2.0. CO;2. Held, I. M., and B. J. Soden (2006), Robust response of the hydrological cycle to global warming, J. Clim., 19, , doi: / JCLI Hendricks, M. B., D. J. DePaolo, and R. C. Cohen (2000), Space and time variation of d 18 OanddD in precipitation: Can paleotemperature be estimated from ice cores? Global Biogeochem. Cycles, 14, , doi: /1999gb Hoffmann, G., M. Werner, and M. Heimann (1998), Water isotope module of the ECHAM atmospheric general circulation model: A study on timescales from days to several years, J. Geophys. Res., 103, 16,871 16,896, doi: /98jd Johnsen, S. J., D. Dahl-Jensen, W. Dansgaard, and N. Gundestrup (1995), Greenland palaeotemperatures derived from GRIP bore hole temperature and ice core isotope profiles, Tellus, Ser. B, 47, of 15