Antarctic Ice Sheet Elevation Impacts on Water Isotope Records During the Last Interglacial

Changes of the topography of the Antarctic Ice Sheet (AIS) can complicate the interpretation of ice core water stable isotope measurements in terms of temperature. Here, we use a set of idealized AIS elevation change scenarios to investigate this for the warm Last Interglacial (LIG). We show that LIG δ18O against elevation relationships is not uniform across Antarctica and that the LIG response to elevation is lower than the preindustrial response. The effect of LIG elevation‐induced sea ice changes on δ18O is small, allowing us to isolate the effect of elevation change alone. Our results help to define the effect of AIS changes on the LIG δ18O signals and should be invaluable to those seeking to use AIS ice core measurements for these purposes. Especially, our simulations strengthen the conclusion that ice core measurements from the Talos Dome core exclude the loss of the Wilkes Basin at around 128 kyr.

, it is necessary to understand how temperature, atmospheric circulation, spatially variable lapse rates, and sea ice feedbacks can all affect the recorded accumulation and isotopic signals when attempting to use these data to help us to deduce past AIS changes. Werner et al. (2018) and Sutter et al. (2020) explored the use of δ 18 O (and temperature) versus elevation relationships to help to evaluate possible AIS reconstructions. Werner et al. (2018) focused on the LGM using the isotope-enabled atmospheric general circulation model ECHAM5-wiso to produce a set of LGM simulations with different AIS reconstructions used in the framework of the Paleoclimate Modeling Intercomparison Project (Otto-Bliesner et al., 2017). A model-data (ice core) δ 18 O comparison allowed insight into the most likely LGM AIS configuration. More recently, Sutter et al. (2020) derived the most probable Wilkes configuration for the LIG by comparing δ 18 O anomalies from the Talos Dome Ice Core with a suite of ice sheet model simulations using the Parallel Ice Sheet Model (Golledge et al., 2015). Sutter et al. (2016) inferred the LIG δ 18 O signal for each of their model configurations using the present-day surface air temperature (SAT) versus elevation relationship from Frezzotti et al. (2007) to obtain temperature, and then to apply the SAT versus temperature derived from Werner et al. (2018). More generally, obtaining quantified information on LIG AIS loss from ice core measurements is still needed for the LIG community (Sime et al., 2019). The LIG AIS loss scenario is directly relevant to calculating future AIS loss probabilities (e.g., DeConto & Pollard, 2016;Edwards et al., 2019).
Here, we investigate the stable water isotope (δ 18 O) response to changes in AIS elevation at 128 kyr, using an ensemble of isotope-enabled climate model experiments from the HadCM3 model. We describe the patterns of SAT, precipitation, and δ 18 O in response to elevation changes and compare isotope-elevation relationships at the continental scale as well as at the location of ice cores spanning the LIG. Finally, we briefly discuss how our results might be used to help to interpret LIG isotope signatures.

Materials and Methods
The isotopic response to idealized changes in AIS elevation is simulated using the isotope-enabled coupled ocean-atmosphere-sea ice General Circulation Model, HadCM3 . Fractional isotopic content is expressed for oxygen-18 using standard δ 18 O notation (Supporting information Text S1). Two control simulations were used: a preindustrial (PI) simulation, and a 128 ka simulation centered on the LIG Antarctic isotope maximum using a modern day AIS configuration (Holloway et al., 2016). Then, a suite of eight idealized AIS elevation change simulations were performed also including 128 ka orbital and greenhouse-gas forcing. Each elevation change experiment includes a simple scaling of the AIS to isolate the impact of ice sheet elevation on temperature, precipitation, and δ 18 O. The change in elevation across the AIS is scaled relative to the prescribed change at the EPICA Dome C (EDC) ice core site using a scaling coefficient β, where and Z EDC is the EDC ice core site elevation in the modern day AIS configuration, Δz is the prescribed elevation change at EDC, which extends to ±1,000 m. Elevations across the Antarctic continent are then increased or decreased proportional to β: where Z A is the two-dimensional array of modern AIS elevations and Z A ′ is a new array of altered AIS elevations. Since this approach maintains the modern shape of the AIS, it reduces the influence of changing ice sheet configuration on circulation and climate and helps in isolating the effect of elevation changes alone. Experiments are performed with Δz equal to (±) 100, 200, 500, and 1,000 m (Supporting information Table S1). Each of the above elevation change scenarios is integrated for a total of 500 years to ensure that surface and middepth climate fields are sufficiently spun up with the imposed elevation changes. The last 50 years of each simulation are analyzed. We also include a simulation with the WAIS reduced to a uniform elevation of 200 m and remains ice covered, as published in Holloway et al. (2016), and following the approach of Holden et al. (2010).
For all our statistical analyses, averages are given with their associated standard deviation (average ± standard deviation). Linear relationships are considered significant when the p-value is lower than 0.05 (Supporting information Text S2).

Changes in Antarctic SAT and Precipitation
The LIG forcing, with no AIS elevation change, induces a mean annual Antarctic warming of 0.9 ± 0.6°C compared to PI (Supporting information Table S2) and precipitation increases of 0.6 ± 1.3 mm/month. Changes are larger in the coastal regions and show wider regional differences: for example, precipitation increases on the coast of the Bellingshausen Sea but decreases on the coast of the Amundsen Sea (cf., Otto-Bliesner et al., 2020).
Changes in precipitation with the elevation tend to follow SAT changes, that is, it decreases as the elevation of AIS is increased. Mean Antarctic precipitation anomalies compared to the LIG control simulation are 3.0 ± 4.7 mm/month for the DC -1 km experiment and -2.4 ± 4.2 mm/month for the DC + 1 km experiment. The largest changes in precipitation occur along coasts facing the Indian Ocean, the Weddell Sea, and along the Ronne Ice Shelf, where the orographic slopes are the steepest (Supporting information Figure S1; Krinner & Genthon, 1999). Deviations from the SAT-precipitation relationships are also the largest in coastal areas ( Figure 1). In particular, the Eastern part of the Peninsula and the WAIS coast display opposite elevation-precipitation relationships compared to the rest of the AIS. This may be due to changes in the localized Peninsula foehn-related drying and/or heat fluxes associated with a more stationary Amundsen Sea low when AIS topography is higher (Krinner & Genthon, 1999). These factors are liable to cause complications when interpreting accumulation change data from coastal and Peninsula ice cores during AIS changes (e.g., Medley & Thomas, 2019).

Antarctic Ice Core δ 18 O Anomalies
Mean Antarctic δ 18 O increases during the LIG by 0.6 ± 0.8‰ compared to PI. At the continental scale, when changing the entire AIS elevation, δ 18 O changes closely follow both SAT and elevation (Figures 1 and 2). This result is consistent with Holloway et al. (2016) and Steig et al. (2015), who report strong positive anomalies over the WAIS when WAIS elevations are reduced to 200 m ("Flat WAIS" experiment hereafter). Our results indicate that δ 18 O anomalies against PI are stronger when the elevation is decreased than when the elevation is increased, a feature also observed for SAT. For the DC -1 km simulation, mean Antarctic δ 18 O changes is +6.5 ± 2.9‰, compared to the PI simulation and −2.3 ± 2.4‰ for the DC + 1 km simulation. However, there are heterogenous patterns in δ 18 O anomalies-mainly in East Antarctica-in response to our idealized and linear elevation changes. Table S3) includes the δ 18 O changes at existing and planned ice core drilling sites. Since these are idealized topographies, and there are other influences on δ 18 O, it is not surprising that none of the simulated elevation changes provide a match to the PI to LIG δ 18 O differences observed in ice cores (Supporting information Table S4). The results of Holloway et al. (2016) show that a reduction in winter sea ice area of 65 ± 7% provides a closer match to the ice core data than any of the idealized AIS elevation change simulations presented here; it is thus of interest to understand how changes in ice sheet elevation and sea ice interact, which will be discussed below.

The Impact of AIS-Sea Ice Feedbacks on δ 18 O, Temperature, and Precipitation
Antarctic sea ice extent increases by 7.6% for the DC -1 km experiment and decreases by 10.8% for the DC + 1 km experiment (Figure 1). This confirms the AIS feedback on sea ice identified by Singh et al. (2016) (for the case of a 90% flattening of AIS compared to PI). Changes in surface wind stress affect the westerly momentum transfer to the ocean (Schmittner et al., 2011), modulating Northward Ekman transport and the associated Ekman drift of sea ice (Singh et al., 2016). In our simulations, a decrease in AIS elevation results in a noticeable reduction of the easterlies around 72°S and westerlies around 52°S (of approximately 8% and 5%, respectively, for DC -1 km), but with little shift in the maximum latitudes of wind speed (Supporting information Figure S2). These changes are likely driven via katabatic-easterlies-westerlies interactions (Sime GOURSAUD ET AL. 10.1029/2020GL091412 4 of 10 Map of Antarctic elevation change in response to elevation scaling of -1 km (first row), -500 m (second row), no scaling (third row), + 500 m (fourth), and +1 km (last row), relative to the height at EDC. (g) The orography of the reference Antarctic configuration ("Z," in km). The different panels (with the exception of (g)) display anomalies relative to a preindustrial control experiment using the reference Antarctic configuration, of (i) the orography ("ΔZ," in m, first column) with the September sea ice extent (≥15%, gray contours), (ii) precipitation ("ΔP," in mm/month, second column), and (iii) the surface air temperature ("ΔSAT," in °C, third column). September sea ice anomalies are given in the top right of the figures giving the orography and the September sea ice extent. Ice core locations with available data are indicated by black points, whereas ice core locations with no available data are indicated by green points. EDC, EPICA Dome C. However, it is noteworthy that the sea ice changes can be modified if WAIS and EAIS are adjusted independently; Steig et al. (2015) found a decrease in sea ice extent with a decrease in WAIS elevation. Thus, the sign of sea ice change depends on the details of the topographic change.
Even for our simple linearly scaled AIS scenarios, sea ice changes are spatially nonuniform around Antarctica. Sea ice extent changes are particularly variable with respect to AIS elevation in the Bellingshausen sector: a 50% increase occurs for the DC -1 km experiment (Supporting information Table S4 and Figure S3). This is likely also related to differing wind forcing, and thus sea ice export, associated with a more stationary and stronger Amundsen Sea low when AIS topography is lower (Krinner & Genthon, 1999;Steig et al., 2015). The Weddell sector shows particularly small changes (±5%). Variability in other sectors remains within a ±15% range. The Bellingshausen and Weddell sectors also stand out in that they present nonlinear AIS-sea ice relationships. Considering the other sectors separately, the mean rate of sea ice area change is -1% per 100 m of elevation change at Dome C (with a mean correlation coefficient of 0.93 and a p-value < 0.05).

(f)
In terms of their control on temperature, precipitation, and δ 18 O, these sea ice changes are small compared with the changes in sea ice explored in Holloway et al. (2016). Removing the AIS-sea ice feedbacks on δ 18 O using a linear relationship (Supporting information Figure S4) has a very small effect on precipitation (-3.0 ± 1.7% and 4.4 ± 2.4% changes compared to the LIG control simulation for the DC + 1 km and DC -1 km simulations, respectively), SAT (0.4 ± 0.5% and -0.5 ± 0.7% changes compared to the LIG control simulation for the DC + 1 km and DC -1 km simulations, respectively) and δ 18 O anomalies (0.9 ± 0.4% and -1.4 ± 0.6% changes compared to the LIG control simulation for the DC + 1 km and DC -1 km simulations, respectively).
The small size of the these indirect AIS-sea ice mediated impacts on temperature, precipitation, and δ 18 O lends confidence to the strategy of treating AIS and sea ice change impacts on δ 18 O as effectively independent of each other (Chadwick et al., 2020;Holloway et al., 2016Holloway et al., , 2017. In the following, we thus consider we can quantify the δ 18 O versus elevation relationship independently from other effects.

Linear SAT-Elevation and δ 18 O-Elevation Relationships
Werner et al. (2018) and Sutter et al. (2020) explored the use of δ 18 O (and SAT) versus elevation relationships to help to evaluate possible AIS reconstructions for the LGM and LIG, respectively. In each case, they used a linear relationship between climate variables and elevation to ascertain past AIS changes. Here, we can use our simulations to assess whether the SAT and δ 18 O versus elevation relationships used in these studies are supported by our suite of LIG simulations. To do this, we calculate slopes for these relationships, using all simulations on a grid point-by-grid point basis (Figures 3 and 4, and Supporting information Text S2).
The Ross Sea, Amundsen Sea, and the coastal regions (≤1,000 m a.s.l.) show no significant linear relationships, likely because the intersimulation noise in these quantities is larger than the signal, due to the small elevation changes prescribed in these regions. Outside these regions, where elevation changes are larger, slopes increase from the coast to the plateau. Mean slopes for ΔSAT versus elevation are -0.34 ± 0.24°C/100 m for regions currently between 1,000 and 2,000 m a.s.l. This rises considerably to -0.92 ± 0.11°C/100 m for regions above 3,000 m a.s.l. (Supporting information  Changes in precipitation (ΔP) and Δδ 18 O versus elevation have lower correlation coefficients compared to SAT-elevation relationships, especially on the plateau. Unlike for the SAT-elevation relationships, δ 18 O-elevation slopes are higher in coastal regions compared to the plateau, likely due to source-distance effects on ΔP and, subsequently, Δδ 18 O (Figure 3). This feature is also notable at the ice core locations (Figure 4). The GOURSAUD ET AL.
10.1029/2020GL091412 7 of 10 variability of Δδ 18 O versus elevation slopes is also spatially larger than for ΔSAT (and ΔP); they vary from -1.28 ± 1.38‰/100 m for regions between 1,000 and 2,000 m a.s.l. to -0.53 ± 0.22‰/100 m for regions above 3,000 m a.s.l. This high variability is also reflected in the Δδ 18 O versus elevation calculated at ice core locations (Supporting information Table S6), with the largest slope at the coastal Skytrain location (-3.52‰/100) and smallest slopes on the EAIS plateau, for example, -0.48‰/100 m at EDML.
Comparing our simulated relationships to those used by Sutter et al. (2020) to interpret the TALDICE δ 18 O ice core measurements, our simulations would suggest that the relationship used in Sutter et al. (2020) would underestimate the sensitivity of Δδ 18 O to elevation change by 43% in this region (Supporting information Text S3): they use a SAT versus elevation slope of -0.8°C/100 m (which seems to be an overestimate, see One of the reasons for the mismatch between our and the Sutter et al. (2020) calculations is their use of a relationship between δ 18 O-temperature not specific to LIG, in the use of calculating past AIS change. Indeed, as for Werner et al. (2018), we find different relationships for different times. If we use all grid points above 100 m a.s.l., a continent-wide average of the slope yields -0.83 ± 0.71‰/100 m (r = -0.9). This LIG-PI Δδ 18 O-elevation slope is similar to, but slightly higher than the LGM-PI slope obtained by Werner et al. (2018) (slope of -0.71 ± 0.3‰/100 m). Similarly to Werner et al. (2018), we thus obtain a continent-wide temporal Δδ 18 O-elevation slope, which is 30% lower than the observational present-day spatial δ 18 O-elevation slope (slope of -1.0‰/100 m, r = -0.9, Masson-Delmotte et al., 2008) and the HadCM3 simulated one (slope of -1.07 ± 0.02‰/100 m, r = -0.89). This, alongside the above, confirms that the use of a present-day spatial elevation gradient as a surrogate for temporal changes for LIG-PI changes must be done with a great deal of care, as it may be incorrect for a variety of locations, AIS changes, and changes through time. Finally we note that, as suggested by Sime et al., 2009), Noone (2009, Figure 4 clearly shows that for a variety of locations, Δδ 18 O does not vary in a linear way, so the use of any single gradient, even for a given ice core site, may vary with time and elevation.

Conclusions
Overall, we see that elevation-induced changes in δ 18 O follow those in SAT. Larger changes in SAT with elevation occur in coastal regions compared to the plateau. While both δ 18 O and precipitation tend to follow SAT changes when site elevation changes, differences do occur in East Antarctic coastal areas where the orographic slope is high. Compared to the eastern part, the Peninsula and WAIS coastal regions display opposite trends, that is, increasing (decreasing) precipitation with increasing (decreasing) AIS elevation. This suggests the need to (i) employ caution, (ii) model δ 18 O, and (iii) drill other ice core species according to accurate WAIS change scenarios to understand how WAIS change will imprint on WAIS ice cores. We note that Antarctic sea ice extent has a relatively modest response to our elevation change experiments. This leads to a small feedback of elevation on climate parameters through sea ice and tends to support the approach that we can look at the controls of sea ice and AIS change on ice core measurements independently (Holloway et al., 2016(Holloway et al., , 2017. We find a continental-wide average of the Δδ 18 O versus elevation relationship of -0.83 ± 0.71‰/100 m (r = -0.9 ± 0.29), thus 20% lower than the PI spatial slope, confirming that the spatial PI δ 18 O versus elevation relationship cannot be a surrogate for temporal relationships (Werner et al., 2018). We find that relationships vary significantly between different ice core locations, ranging from -3.52‰/100 m at Skytrain to -0.48‰/100 m at EDML.
Confidently dated ice core measurements covering the LIG are currently only available from East Antarctic ice core sites. Given the widespread expectation of major changes in WAIS elevation during the LIG, there is a need for new well dated ice cores covering the LIG from sites outside the EAIS, alongside further δ 18 O modeling. New ice cores drilled on the WAIS, particularly at Skytrain or Hercules Dome, will provide important insights for future AIS LIG reconstructions. The results above enable ice core δ 18 O measurements to be interpreted from an elevation point-of-view with more certainty.
Finally, we note that this study is limited by the model resolution of HadCM3 and our particular simulation set-up: prescribing small absolute changes in elevation in coastal regions. The Skytrain site would thus benefit from high-resolution modeling, ideally using a regional isotope-enabled climate model.

Data Availability Statement
The orography, surface air temperature, precipitation, and water stable isotope responses to idealized changes in AIS elevation simulated by the isotope-enabled coupled ocean-atmosphere-sea ice General Circulation Model HadCM3 are available on the data system managed by the UK Polar Data Centre (Goursaud et al., 2020)