Annular variations in moisture transport mechanisms and the abundance of δ18O in Antarctic snow

Authors

  • David Noone,

    1. School of Earth Sciences, University of Melbourne, Parkville, Victoria, Australia
    2. Now at Division of Geological and Planetary Sciences, California Institute of Technology, Pasadena, California, USA.
    Search for more papers by this author
  • Ian Simmonds

    1. School of Earth Sciences, University of Melbourne, Parkville, Victoria, Australia
    Search for more papers by this author

Abstract

[1] Water isotopes are commonly used as indicators of climate state even though many biases and variations in processes affecting the polar signal have not been quantified. Results from the Melbourne University General Circulation Model suggest the annual cycle explains half of the monthly δ18O variance, and a semiannual variation contributes more than 15 in places. Eddy moisture convergence drives gross accumulation, while stationary flux allows sublimation of 25–30% of the precipitation. Part of the monthly anomaly variance is associated with a dominant annular disturbance in the circulation. This oscillatory mode alters the character of the transport processes through changes to the preferred location and strength of baroclinic cyclones. A Rayleigh model indicates that a third of the continental δ18O anomaly can be explained by temperature-dependent fractionation, while changes to the condensation give 3 times too much depletion. The residual is explained by the migration of the zone from which midlatitude air is entrained into the polar environment by cyclonic storms. The positive phase of the annular mode is associated with an increased contribution from the near-coastal region, which enriches the continental precipitation. Such vacillation introduces bias in reconstruction using modern analogues because the spatial temperature-isotope slope is modified.

1. Introduction

[2] Although examinations of water isotope measurements made on ice samples from cores drilled in the polar ice caps have led to an appreciation of climate variability on timescales from decades to hundreds of thousands of years, it remains unclear precisely what aspects of the climate system are being sampled. Associations between the isotopic “delta” values (a normalized anomaly, δ = (R/Rs − 1) × 1000, where R is the absolute ratio of heavy to light molecules and s denotes the “Vienna standard mean ocean water” standard) and both temperature and precipitation amount are observable in a variety of spatial and temporal settings [e.g., Craig and Gordon, 1965; Jones et al., 1993; Araguás-Araguás et al., 2000]. Classically, this robust empirical association is considered sufficiently compelling to explain the depletion through variations in the two control parameters of a simple Rayleigh distillation process for an isolated Lagrangian air parcel [Epstein et al., 1963; Dansgaard, 1964]. However, more rigorous analysis shows that the external conditions (such as temporal biases in the seasonal deposition of precipitation [e.g., Krinner et al., 1997; Noone and Simmonds, 1998; Schlosser, 1999; Delaygue et al., 2000a; Werner et al., 2001], the condition of the source region [e.g., Charles et al., 1994; Delaygue et al., 2000b; Werner et al., 2001], and mixing of ambient air parcels [e.g., Cole et al., 1999; Werner and Heimann, 2002]) are crucial to the quantitative description of the ice core record.

[3] The isotope content of accumulated snow reflects the temporal integral of a distribution of many individual snowfall events. Each event reflects conditions along the pathway traversed by the air masses. As such, the integral describes a stochastic sample of the conditions over a broad spatial region near the deposition site. As details are dictated by the large-scale circulation and state of the climate system, it is possible, at least in principle, to conjecture the inverse problem of reconstruction. This problem is, of course, underdetermined unless the numerous “climate parameters” that one might like to reconstruct are linked uniquely by an underlying physical paradigm. In this study the isotope record in Antarctica is of foremost concern, and the provision of a detailed quantification of the physical associations between factors that determine the isotopic depletion and the climate state is a primary aim. In this endeavor, an atmospheric general circulation model (GCM) is used to provide a mechanistic description of some variations in the isotope concentrations of snow. Initially, a description of seasonal variations in the modeled climatology of Antarctic precipitation and δ18O is given, and thereafter focus is directed to the importance of vacillation of a zonally symmetric disturbance about the mean state.

2. Isotope GCM Description

[4] The Melbourne University atmospheric GCM is a spectral primitive equation model and has been used extensively for a wide range of atmospheric investigations. For this study it is configured to have a horizontal resolution denoted by rhomboidal truncation of the harmonic series at wave number 21, with a Gaussian transform grid of 5.625° × 3.25° used to exactly resolve quadratic terms. There are nine levels in the “sigma” vertical coordinate. A more detailed overview of the model and the modifications to predict the isotopic state of all water reservoirs is described by Noone and Simmonds [2002]. It is sufficient to mention here that the treatment of water isotopes includes both equilibrium and kinetic fractionation processes during surface evaporation, condensation associated with large-scale and convective precipitation, and equilibration with ambient air as precipitation falls. Advective transport of mass mixing ratios of total water (H2O) and the isotopic forms (H218O and HDO) utilizes the monotonically interpolating semi-Lagrangian scheme of Williamson and Rasch [1994]. Vertical mixing due to moist convection is parameterized as a diffusive process because the mass flux and entrainment are not explicitly represented by the Manabe et al. [1965] convective adjustment used in the model. Noone and Simmonds [2002] argue that the global simulation of isotopic depletion is credible over the range of time and space scales resolved by the GCM. Results presented here are from a 17-year simulation, reported by these authors, that includes forcing by monthly varying sea surface temperature and sea ice concentrations representative of the period 1979–1995 and is consistent with the Atmospheric Model Intercomparison Project 2 specification [Gates et al., 1999].

3. Simulated Precipitation and δ18O

[5] The simulated annual mean high southern latitude precipitation is displayed in Figure 1a. The main features include reduced Antarctic precipitation with increasing distance from the coast, lower totals in the outflow basins, and a maximum immediately to the west of the Antarctic Peninsula. North of the coast, the precipitation increases with latitude and the distribution shows more spatial variability. The spatial scale of this variability is not dissimilar to that of the smallest waves resolved by the model. The inland values of less than 5 cm yr−1 are similar to those obtained from other climate models [e.g., Genthon, 1994; van den Broeke, 1997] and reanalysis products [e.g., Cullather et al., 1998; Noone et al., 1999]. They are also consistent with accumulation estimates based on glaciological survey [Giovinetto and Bentley, 1985; Vaughan et al., 1999], once the sublimation is taken into account (simulated to be about one third of the precipitation inland, not shown). The inland summer maximum is around 50% larger than the winter total. Fourier analysis reveals that only on the high-altitude plateau and isolated coastal regions does the annual harmonic explain more than 35% of the monthly mean variance. A semiannual cycle explains more than 10% of the variance inland and is also evident at around 60°S.

Figure 1.

Modeled annual mean (a) precipitation (contour interval of 20 cm yr−1 with extra contours at 5 and 10 cm yr−1) and (b) δ18O of precipitation (contour interval of 5‰). Stippling shows where the annual cycle explains more than 35% and hatching shows where the semiannual cycle explains more than 10% of the monthly variance. (c) Spectrally resolved topography showing regions where the height is underestimated (stippled) and overestimated (hatched) by 500 m. Contour interval is 500 m.

[6] The annual mean δ18O (Figure 1b) may be compared to the glaciologically based estimate of Zwally et al. [1998]. The greatest depletion is underestimated by around 8‰ at highest elevations. This underestimation is related to the fact that, owing to the steep Antarctic slopes, the elevation and details of the Antarctic topography are not particularly well captured in the model (see below), and there is an associated overestimate of the Antarctic surface temperature (of about 10–15 K). The isotopic depletion is spatially coincident with increasing topography, decreasing annual mean temperature and precipitation amount toward the pole. Using all model points over the continent, the slope of a regression between annual mean δ18O and temperature at the lowest model level is 0.90 K/‰. The annual cycle explains typically 60% of the monthly variance with an amplitude of around 8‰. Although there are some limitations, the simulation is not inconsistent with glaciological estimates and not unlike results from other isotope GCMs [Joussaume and Jouzel, 1993; Hoffmann et al., 2000; Werner et al., 2001].

[7] The precipitation and the δ18O both decrease with altitude. The simulated continental gradients are less than those observed and consistent with the poorly resolved slope of the coastal escarpment (Figure 1c). Similarly, the poor representation of the topography of the Antarctic Peninsula is evident in the precipitation and, most likely, mean δ18O, when compared with firn measurements [Jones et al., 1993; Turner et al., 1995].

4. Moisture Budget

[8] The horizontal moisture transport on pressure surfaces is deconvolved into stationary and eddy components for each month in the simulation by applying Reynolds averaging in the usual manner [e.g., Bromwich et al., 1995; Boer et al., 2001]. The zonally integrated meridional moisture flux convergence is shown in Figure 2 as a vertical section for summer (December–January–February (DJF)) and winter (June–July–August (JJA)). This indicates the loss of water vapor due to condensation. Also in the figure, the height above which the air mass is capable of ascending to an altitude of h = 2000 m is indicated. This is the height for which the Froude number (Fr = v/Nh, where v is the meridional wind speed and N is the buoyancy frequency) assumes its critical value of unity. The Froude number can be considered a quantifying expression for the reduction of a parcel's kinetic energy to zero during ascent. Again, this nonlinear quantity is decomposed into eddy and stationary components. The importance of this height is that an air mass beneath this height will be unable to ascend to the continental interior. As such, a clear separation between the precipitation regime and δ18O and of coastal and inland regions could be expected. Some observational evidence for this assertion has been given by Bromwich [1988] and Le Treut et al. [1988].

Figure 2.

Vertical section of zonally integrated (top) transient and (bottom) stationary moisture flux convergence for (left) December–January–February and (right) June–July–August expressed in units of kilograms per second. Negative contours are dashed, and the zero contour is bold. Contour interval is 20 with additional contours every 5 for magnitudes of less than 20. Stippling indicates magnitudes greater than 20. Data are masked where all zonal points are below the surface. A heavy long-dashed line indicates the zonal mean surface pressure, a heavy short-dashed line shows the limit of the boundary layer, and a heavy solid line shows the height of a subcritical surface layer, as described in the text.

[9] The model moisture convergence associated with the transient flow extends through the depth of the troposphere and is always positive. This dominates the inland accumulation and is largest in JJA, even though the vertical extent is greatest in DJF. In DJF the maximum convergence is above the boundary layer, while the winter maximum is near the surface. This is associated with colder surface temperatures under wintertime inversion conditions. The depth of the subcritical flow is much greater for the stationary component and indicates that the associated moisture convergence is confined to low elevations and, thus, is only positive over the coastal zone. Inland (and above the critical height) the stationary component is negative, indicating a net mass loss from the surface with outflow of dry air. The magnitude of this loss is typically 20–50% of the eddy convergence. The largest mass loss is found below the mean surface height and associated with northward flow over the Weddell and Ross Seas.

[10] The isotopic signal of the eddy and stationary moisture convergence differ owing to their flow characteristics. In particular, the stationary convergence is typically coincident with the boundary layer, and a stronger influence from local moisture sources is likely for the coastal precipitation. In contrast, the existence of the subcritical layer provides some separation from the surface for the moisture transported inland by the transient flow. Should the climate regime shift and alter (say, reduce) the critical height, one could propose that an ice drilling station initially below the critical height would become more strongly influenced by the supercritical flow. Should the isotopic condition of these flows differ, a shift in the isotope signature would be observed.

[11] Figure 3 shows explicitly, for DJF and JJA, the moisture source distribution for precipitation falling above 1500 m (refer to Figure 1c). This distribution is constructed with an extension of the “water tagging” method employed by Joussaume and Sadourny [1986] and Koster et al. [1986] and more recently by Delaygue et al. [2000b] and Werner et al. [2001]. Evaporative flux from predefined basis (source) regions is tracked through the GCM hydrologic cycle to precipitation over some deposition region of interest (here, inland Antarctica). This is achieved through incorporation of additional tracers specific to each source region. In this study, 42 regions are chosen and cover different geographic areas of the Earth's entire surface. Of these, 29 are placed in the Southern Hemisphere. As the sum of all moisture components equals the total water vapor mixing ratio (exactly analytically, and enforced to be so in the simulation at each time step), the precipitation falling over the deposition region can be quantitatively partitioned into the 42 contributions. The source distribution is thus found as a synthesis of these components onto the basis regions. Similarly, as the isotopic flux from each source region is known, the δ18O of each component can be found from the isotopic precipitation and reflects both the of δ18O the evaporation and the depletion of the vapor en route.

Figure 3.

Antarctic moisture sources for precipitation falling above 1500 m in (a) December–January–February and (b) June–July–August. Shading indicates source in units of percentage of precipitation per unit area (light stippling, >0.25; heavy stippling, >0.5; light hatching, >0.75; heavy hatching, >1 (×10−6) km−2). Source regions are outlined. Bold numbers give relative contribution from each region (global sum is 100%), and lighter numbers give the δ18O of the water from each region when it reaches the continent (global mass weighted mean is the δ18O of the precipitation).

[12] Of the 16 cm yr−1 precipitation in DJF in the deposition region, over 50% of the moisture comes from the continent itself with around 42% from inland sublimation. This is consistent with values of 35–40% from the European Center/Hamburg GCM [Werner et al., 2001] but greater than those from the Goddard Institute for Space Studies GCM of around 20–25%, which simulates lower sublimation rates [Delaygue et al., 2000b], although slightly different precipitation regions were used in these studies. At the early summer sea ice edge, the δ18O of the source water is large and indeed enriched in the Indian Ocean sector. This is driven by cold surface conditions and very depleted water vapor. It is important to note that the magnitude of these contributions is small (less than 1%). The maritime region south of 60°S is an important source (8%) with a greater contribution from the eastern hemisphere. This geographic bias is explained partially by the zonal asymmetry of the high-elevation deposition region itself. The oceanic sources are typically in the midlatitudes, where the magnitude of the mass transport by both transient baroclinic cyclones and mean flow is largest. The Indian Ocean sector provides the largest contribution, particularly in JJA. This result is similar to that of Delaygue et al. [1999], although they attributed a greater contribution to the Pacific as they include lower-elevation parts of West Antarctica in their precipitation region. The small contribution from the Southern Ocean around 60°S marks clearly the separation between the more local surface source in the seasonally ice-free sea ice zone and contributions from air masses transported to the high latitudes across the baroclinic zone. The annual mean δ18O of the inland precipitation is −32%. The δ18O of the source components typically decreases with latitude and distance from the deposition. This is the opposite of the reduction in depletion with proximity to the equator that is modeled in the evaporation (not shown). With the exception of the continental recycling, source regions south of 50°–55°S therefore provide an enrichment relative to δ18O the of the inland precipitation. This is seen also in the results of Werner et al. [2001].

[13] A most notable feature of the JJA sources is the reduction of the continental recycling concurrent with the reduction in the mean precipitation rate to 9 cm yr−1, suggesting little seasonal variation in the mass flux across the continental boundary. Given that they are mostly covered by sea ice, it may appear surprising that the coastal and high-latitude Southern Ocean zones are significant sources in winter. However, it should be borne in mind that even in that season the Antarctic sea ice has significant areas of open water [Watkins and Simmonds, 2000], and very high rates of evaporation occur in these “leads”. Unlike most models used in similar studies, the Melbourne University GCM represents the hydrologic, dynamic, and thermodynamic effects of open water in the ice pack [Simmonds and Budd, 1991]. Relative to DJF, the JJA δ18O of the noncoastal sources is typically more depleted by around 10–15‰ except for regions north of the midlatitudes and in the western South Pacific, where they are around 20‰ more depleted. In contrast, the depletion of the continental precipitation increases by only about 7‰ (to −38‰) and reinforces the assertion of greater contribution from the coastal regions where, again, the influence on the isotopic state is one of enrichment relative to the mean. Indeed, this is particularly evident for the contribution from the recycled component where the winter is less δ18O depleted that that in summer. The largest, nonrecycled contributions are found to originate quite uniformly between 30°S and 60°S, with a bias toward the Indian Ocean.

5. Dominant Variation in Circulation

[14] The most influential mode of variation is defined as that which explains the largest fraction of the variance in monthly anomalies of 500-hPa height (denoted equation image500) south of 30°S and is identified through principal component (PC) analysis. This analysis proceeds for some anomaly field, fij(t), as the solution of the eigenvalue problem of the area weighted spatial correlation matrix. A normalization is imposed such that empirical orthogonal functions are the correlation between the initial data and the PC, a(t). Regression maps, gij, are calculated as gij = equation image and, thus, show a characteristic (1 standard deviation) variation in f associated with the PC of interest.

[15] Figure 4a shows the equation image500 regression map for the first PC of simulated equation image500 (ZPC1). The figure displays the positive phase of the familiar annular pattern described by numerous authors [e.g., Karoly, 1990; Thompson and Wallace, 1998; Kidson and Watterson, 1999; Thompson and Wallace, 2000]. To be concise, the discussion that follows describes only the influence of this positive phase, although the arguments can be applied equally in reverse to the negative phase. The salient features include an intensification of the polar vortex, an increase in the strength of the westerly jet stream, and a meridional shift in the location of the circumpolar trough associated with the equatorward redistribution of air mass about a nodal point near 50°S. A temporal spectral analysis of the PC reveals a “red” spectrum without any substantial peaks. This is in qualitative agreement with other studies [e.g., Boer et al., 2001], even though the series here is relatively short and there is neither an interactive ocean coupling mechanism nor resolved stratosphere dynamics in the present model configuration. Owing to serial correlation in the monthly data, the number of degrees of freedom in the regression is reduced from 202 to 171 using the estimator suggested by Von Storch and Zwiers [1999]. Therefore a correlation ∣r∣ > 0.16 is required at the α = 0.05 level to indicate nontrivial regression. The key motivation for including all months in the analysis, rather than just winter months [e.g., Thompson and Wallace, 2000; Boer et al., 2001], is to retain the largest possible sample size. The analysis, however, was performed by using data for only the period JJA and also for an extended “winter” May–October. The results are qualitatively similar to those presented here. It is important to note that ZPC1 is uncorrelated with respect to season.

Figure 4.

Regression of the first principal component of 500-hPa height monthly anomalies south of 30°S (ZPC1) and monthly anomaly (a) 500-hPa height, (b) precipitation, and (c) δ18O of precipitation. Contour intervals are 10 m, 5 cm yr−1, and 0.2‰. Negative contours are dashed, and the zero contour is bold. Stippling and hatching indicates where the regression is significant at the α = 0.05 level.

[16] ZPC1 is regressed against the monthly anomalies of precipitation and δ18O (Figures 4b and 4c). While the precipitation regression is negative in the midlatitudes, an increase is found in the Southern Ocean that is discernible above the intrinsic variability of the region. Over the continent, the regression is weak and only limited areas of East Antarctica exhibit significantly reduced precipitation. This finding is in agreement with results from two 201-year GCM simulations and the National Centers for Environmental Prediction (NCEP) reanalysis [Boer et al., 2001].

[17] In contrast, δ18O the changes are more obvious over much of the inland plateau and provide a variation of typically 0.5‰, of the order of 10% of the magnitude of the annual cycle. Although the variations in the continental precipitation are barely significant, the isotopic anomalies are more robust. The use of ZPC1 shows that it is the spatial organization of the circulation that is of importance in comparison with the small correlations when equation image500 is considered point wise [Werner and Heimann, 2002]. The annular variations are of negligible influence over the Antarctic Peninsula, which provides a notable deviation from otherwise zonal symmetry. In the midlatitudes the depletion is reduced (positive anomaly) and explained heuristically as a reduction in the isotopic distillation in the baroclinic zone as indicated by the reduction in the precipitation. Assuming a constant source, a simple mass conservation argument for a poleward moving air mass suggests that enriched precipitation at midlatitudes requires that precipitation from the residual moisture at high latitudes be more depleted.

[18] Figure 5a shows that there is a dipole change in the temperature above the boundary layer. The warmer temperature in the midlatitudes coincides with the reduction in the precipitation and would suggest, from simple thermodynamics, that with no change in the source water, more water vapor can be present upon poleward transit. Similarly, an increase in the temperature gradient across the Southern Ocean would induce an increase in the moisture convergence, evident in the precipitation regression in Figure 4b. Poleward the gradient in temperature anomaly is the change in the δ18O of the water vapor (Figure 5b). This follows by considering a Rayleigh process with enhanced removal of heavy isotopes with increased rate of condensation. This change in the “rain out” is shown in Figure 5c, where the regression of the monthly mean mixing ratio tendency associated with condensation shows a greater retention of moisture (positive anomaly) in the midlatitudes and increased condensation over the Southern Ocean, with which the precipitation data corroborate. This result demonstrates also that the coastal precipitation changes are dominated by conditions near the surface and are associated with the subcritical flow. Again, only small changes over the continent are evident.

Figure 5.

Vertical section of zonal mean monthly anomaly (a) temperature, (b) δ18O and (c) mixing ratio condensation tendency regressed against ZPC1. Contour intervals are 0.1 K, 0.2‰, and 10 g kg−1 yr−1. Stippling and hatching indicate where the regression is significant at the α = 0.05 level. Surface pressure is shown as a heavy dashed curve.

[19] There are two (physically interdependent) factors which may be understood in terms of a Rayleigh process. Specifically, the depletion from some initial state is written as Δδ = Fα−1 − 1, where F is the fraction of moisture remaining and α represents the fractionation. Choosing values from the zonal mean climatology for the depletion, an effective fractionation can be found upon assuming that F is approximately equal to the change in mean mixing ratio (i.e., for parcels arriving at 75°S and 600 hPa from a source at 45°S and 850 hPa, ΔδGCM = −25‰ and F ≈ 0.14, yielding αe = 1.0132 and Te ≈ 255 K). A cooling of 0.2 K, typical in Figure 5a, increases the fractionation and leads to a anomaly of −0.07‰. This value is less than a third of that seen in the vapor (Figure 5b) and precipitation (Figure 4c). On the other hand, should the fractionation remain constant, the importance of the rain out (1 − F) can be determined. The change in F estimated from the regression of the zonal mean mixing ratio (not shown) is ΔF = −0.04 (i.e., proportionally more water lost during transit). This is also suggested by the gradient imposed by the tendency anomaly in Figure 5c. Applying the Rayleigh model gives a δ18O anomaly of −1.7‰ and about twice that seen in the regression data. However, these estimates implicitly assume that both the transport pathways and source moisture remain unchanged. This is clearly an incorrect postulation, as the combined effect gives an anomaly at least 3 times more depleted than full GCM calculation. Indeed, it is the need for fundamental insight into this discrepancy that is central to improving the interpretation of ice core records.

6. Variations in Transport and Sources

[20] As identified earlier, the moisture transport associated with Antarctic precipitation is dominated by the eddy component of the flow, which in turn is associated with the character of the baroclinic disturbances of the midlatitudes. In 6-hourly sea level pressure data from the model simulation, cyclonic depressions were identified and tracked by using the objective scheme developed at the University of Melbourne [Murray and Simmonds, 1991; Simmonds et al., 1999]. This scheme provides a range of diagnostics to describe the statistical character of the Lagrangian entities. In this report, only three such quantities are enlisted: the cyclone system density (proportional to the probability that a cyclone will be close to some location), the translational velocity (here, only the eastward component), and the cyclone depth (the pressure difference between the pressure at the “edge” of a cyclone and the central pressure). The quantities are aggregated into monthly statistics and the seasonal cycle is removed to obtain anomalies. These are regressed against ZPC1 and shown as zonal means in Figure 6, with the annual mean included for reference.

Figure 6.

Regression of ZPC1 and (a) cyclone system density (1 × 10−3 systems°latitude−2), (b) westerly translation speed (meters per second), and (c) cyclone depth (hectopascals). Regression anomalies are shown as the solid curve on the left scale, and the annual mean is shown as a light dashed curve on the right scale. (d) Moisture flux convergence (centimeters per year) and (e) critical flow height (hectopascals) decomposed into total (heavy solid line), stationary (light solid line), and eddy (dashed line) components.

[21] The system density shows a clear southward migration of cyclones associated with the positive phase of ZPC1 with the magnitude of the anomaly around 10% of the total. Notice that the total number of cyclones in the hemisphere changes little. There is an increase in the mean eastward translation, as a consequence of the enhanced steering level flow (i.e., Figure 4a). The greater number of cyclones centered south of around 55°S (coastal) are deeper than the mean, while those further north are less deep. Simmonds and Keay [2000] suggest that, for an ideal cyclone (circular with a parabolic cross section), the depth is proportional to the integrated poleward mass flux. Regression of the moisture flux convergence shows that north of 55°S the reduced precipitation is dominated by the changes in the stationary flux, although there is also a reduction in the eddy component. South of 60°S and over the continental escarpment, it is the changes in the eddy flux that are more important. The responses of the stationary and transient components are of opposite sign over the continent. ZPC1 describes an intensification of the polar vortex and a reduction in subsidence. Therefore the mass loss over the continent associated with the stationary outflow is reduced. As there is less ambient moisture near the coast, the eddy moisture convergence over the continent is reduced even though the mass flux increases with cyclone depth.

[22] Figure 6e shows the change in the critical height of flow reaching 2000 m based on the Froude number as before and now regressed against ZPC1 for the eddy, stationary, and total components of the flow. Recalling that it is the eddy component of the transport that most strongly influences the inland precipitation, it is of importance that near the coast the critical height has decreased by up to 10 hPa. The model climatology suggests the δ18O of vapor decreases with height at a rate of 0.042‰ hPa−1. As such, inclusion of air from lower in the troposphere could provide an enriching influence of the order of the observed inland signal. Another consequence of the critical height change is due to a greater proximity to the surface and the increased potential for stronger influence by local evaporation. As such, a larger contribution from the near-coastal region could be expected. Again this would be an enrichment.

[23] Figure 7 shows the regression of the source distribution and the isotopic content of Antarctic precipitation falling above 1500 m, in a manner similar to that displayed in Figure 3. This provides an integrated assessment of the changes in the flow and changes to the source region, at the expense of knowledge about the mechanisms for the transport. As in Figure 3, the change in both the fractional contribution to the continental precipitation and the isotopic content of the water from each source is marked. Both the variations in the magnitude of the source contribution and its isotopic influence are associated with a concomitant variation in the magnitude and δ18O of the evaporation (not shown). However, this accounts for typically only 20–30% of the δ18O anomaly because the signal is modified once the effects of changed circulation and condensation regime are incorporated.

Figure 7.

Regression of moisture source anomalies and ZPC1. Shading indicates source anomaly in units of fraction of precipitation per unit area (heavy stippling, <−0.03; light stippling, <0.01; light hatching, >0.01; heavy hatching >0.03 (×10−6) km−2). Source regions are outlined. Bold numbers give regression of relative contribution from each region, and lighter numbers give the isotopic ratio of the water from each region that falls inland.

[24] The large influence from the midlatitude Indian and western Pacific Oceans is reduced. Instead, the Southern Ocean plays a more significant role. There is also evidence for a larger contribution of less depleted water from the Atlantic Ocean. The magnitude of the anomalies is typically of the order of 10% in most regions (cf. Figure 3). There is little change in the continental recycling. The importance of the region of the Amundsen and Bellingshausen Seas is also reduced, which may be suggestive of systematic changes to the sea ice state and surface heating in phase with ZPC1 and cyclone activity in this sector. A southward shift in the source region is consistent with the southward shift in the location of cyclones and thus the mechanism for poleward transport. Furthermore, a reduction of the number and depth of cyclones to the north reduces the possibility of entrainment of air from midlatitudes to the polar environment.

[25] The change in the δ18O values shows that where the source region has increased in magnitude, the water is less depleted when it reaches deposition. Assuming only small changes to the δ18O of the evaporation itself, this suggests that water from these regions has a more direct transport route along which less depletion occurs. On the other hand, moisture from a more distant source now takes a less direct path and passes through a zone of increased moisture convergence en route to the Antarctic (Figure 5c) and thus becomes more depleted. One should recognize also that the changes in the δ18O of the evaporation are affected by the δ18O of the local vapor (i.e., the “isotopic humidity”). Should the surface layer temperature remain constant (Figure 5a), more depleted vapor will yield increased isotopic flux and contribute to the enrichment.

7. Discussion and Conclusion

[26] It has been shown that nonseasonal changes in the continental moisture flux are associated with changes in the cyclonic storms as dictated by variations in the baroclinic activity in the mean state. The ability of moist air mass to penetrate the continent interior is dependent on both the kinetic energy of the eddy component of the flow and the work required to overcome the stable stratification (as measured by the Froude number). This is dependent on not only the wind speed, but also the stability of the coastal zone which is, in turn, an intrinsic consequence of the quasi-equilibrium state of baroclinic storm activity at middle to high latitudes and surface heating. While the local variations in temperature-dependent fractionation are important, changes in the circulation affect the condensation history of parcels and the location of the source region. Moreover, circulation changes alter the isotopic state of water evaporated at the source. These provide stronger influence on the isotopic signal. With anomalous southerly migration of the baroclinic zone associated with the positive phase of the annual mode, the water source region is increasingly influenced by the enrichment from near-coastal regions. This is a consequence of the reduction in the entrainment of midlatitude air and increased depletion from rain out of residual midlatitude moisture. These factors combine to provide a net depleting influence on the continental precipitation. Although changes in the coastal storm activity facilitate a greater contribution from the eddy mass flux, the moisture convergence is reduced. The modulation of ZPC1 describes also a relaxation of the continental anticyclone and associated subsidence. This also allows a reduction in drainage of cold continental air and surface mass loss by the mean flow.

[27] One assumption that is made implicitly in reconstruction of temperature (or other climate parameters) from ice core isotope records is that the types of associations examined in this study remain constant during various epochs. One might expect, for instance, during glacial times, the thermodynamic forcing of the atmosphere by different (presumably cooler) ocean surface conditions would be sufficient to change the circulation patterns at high latitudes. If this were the case, the findings presented here would suggest there is the possibility for significant uncertainty in the reconstruction from simple calculations. Specifically, the regression between the changes in δ18O and temperature associated with the annular variation is 0.49 K/‰ (about half that from annual mean data), with a correlation coefficient of 0.62. Should there be no mechanism acting to compensate this effect, reconstructions utilizing the spatial slope will be in error by about 25%, when the basic state is systematically biased toward a particular phase of the annular mode of the amplitude considered here.

[28] Evaluation of the height of the subcritical flow is of particular interest as it leads to an interesting prediction. Should the mean critical flow height change with time (such as during the Last Glacial Maximum, global warming, or particular phases of the Antarctic mode), the isotopic signal at some location could change as it becomes on the opposite side of the critical location. From this study it is not possible to suggest if this mechanism alone would provide a substantial deviation in the isotopic records, but it is of worthy consideration when the locations of, for instance, Vostok and Dome Concordia are examined.

[29] The analysis here has focused on a prominent mode of variability in the geopotential height field. It should be recalled that this explains only 37% of the variance. By construction, identification of this mode from PC analysis means the results are orthogonal to other linear modes. As such, should some climate regime include a systematic shift in the mean state of ZPC1, a shift in the hydrologic and isotopic anomalies could be expected. This type of situation has been proposed in association with twentieth century warming as a consequence of a positive feedback that excites preferentially such modes [e.g., Shindell et al., 1999; Corti et al., 1999; Feldstein, 2002]. Thompson and Solomon [2002] suggest that an observable summertime trend toward a climate state in which the annular mode is biased toward the positive phase is associated with decreasing Antarctic stratospheric ozone concentrations. They show that the signature of the annular mode produces warming at midlatitudes and over the Antarctic Peninsula, while the continent itself cools. The present results predict a concurrent shift in the inland δ18O of 0.5‰ over the last 20 years. Over the Antarctic Peninsula, however, there would be little change in the of the snow. This can be verified when the high-resolution firn core data become available.

[30] Systematic change in the annular mode has clear implications for reconstruction of mean climate conditions in the past, when it is unlikely that the characteristics of the annular structure are the same as those of today. Alternatively, should this bias be separable from the background signal, it may be possible to deduce from the ice core record a history of the annular mode.

Ancillary