Sea-ice feedbacks influence the isotopic signature of Greenland Ice Sheet elevation changes: Last Interglacial HadCM3 simulations

Changes in the Greenland ice sheet (GIS) affect global sea level. Greenland stable water isotope (δO) records from ice cores offer information on past changes in the surface of the GIS. Here, we use the isotope-enabled HadCM3 climate model to simulate a set of Last Interglacial (LIG) idealised GIS surface elevation change scenarios focusing on GIS ice core sites. We investigate how δO depends on the magnitude and sign of GIS elevation change and evaluate how the response is altered by 5 sea ice changes. We find that modifying GIS elevation induces changes in Northern Hemisphere atmospheric circulation, sea ice and precipitation patterns. These climate feedbacks lead to ice core-averaged isotopic lapse rates of 0.49‰ per 100 m for the lowered GIS states and 0.29‰ per 100 m for the enlarged GIS states. This is lower than the spatially derived Greenland lapse rates of 0.62-0.72 ‰ per 100 m. These results thus suggest non-linearities in the isotope-elevation relationship, and have consequences for the interpretation of past elevation and climate changes across Greenland. In particular, our results 10 suggest that winter sea ice changes may significantly influence isotopic-elevation gradients: winter sea ice effect can decrease (increase) modelled core-averaged isotopic lapse rate values by about -19% (and +28%) for the lowered (enlarged) GIS states respectively. The largest influence of sea ice on δO changes is found in coastal regions like the Camp Century site.


Experimental setup
We use the isotope-enabled General Circulation Model (GCM) HadCM3 to simulate the isotopic response to idealised variations in the elevation of the GIS. This GCM has been widely used to examine present, past and future climates (Stocker et al., 2013;Solomon et al., 2007) and consists of a coupled ocean, atmosphere and sea ice model. Tindall et al. (2009) presents the 5 implementation of the water isotope code in HadCM3.
We run a first ensemble of 16 idealised elevation change HadCM3 simulations with greenhouse-gas and orbital forcing centred at 125,000 years BP (125ka) (See Table A1). A 125ka control experiment (hereafter, 125Control) is performed including a present-day GIS configuration (IceBridge BedMachine Greenland, Version 3 - Morlighem et al. (2017a, b)). To generate the idealised elevation changes, we scale up and down the GIS height from ± 50 m up to ±1300 m; in particular, we scale elevations 10 relative to the elevation at the NEEM ice core site, following where ∆z is the elevation change prescribed; Z N EEM is the elevation at the NEEM ice core site in the present-day GIS configuration and β is the scaling percentage. GIS elevations are then decreased/increased by β; (2) For the rest of the results section, for clarity, we focus particularly on two example scenarios which depict medium-high GIS elevation changes (experiments marked in blue in Table A1).

Surface air temperatures
The orbital forcing dominates the climate in the 125Control simulation. In Greenland, summer local temperature increases 25 exceed 3.5°C due to the large increase in summertime insolation (Fig. 2c).
The local surface climate over Greenland is noticeably affected by local changes in GIS surface elevation. Decreases in GIS elevation act to increase surface air temperatures (SATs) across Greenland, and vice versa ( Fig. 1 and Fig. 2). In Greenland, the scenario with decreased elevation (m900) simulates positive SAT anomalies all year round compared with 125Control ( Fig. 2d-f). Annual local temperature increases exceed 4.5°C in m900 relative to 125Control. As expected, the increased elevation 30 scenario p900 shows negative SAT anomalies throughout the year relative to 125Control experiment ( Fig. 2g- Averaging across six ice core sites (Camp Century, NEEM, NGRIP, GRIP, GISP2 and DYE3), temperature lapse rates vary from 0.47°C per 100 m for the lowered GIS states to 0.44°C per 100 m for the enlarged GIS states (Fig. 1).

Atmospheric circulation
To better understand the variations in atmospheric circulation that occur in response to changes in surface elevation we show changes in the low-level wind pattern (at 850 hPa) and mean sea level pressure (MSLP) field. The 125Control simulation Over the Norwegian Sea, there is a increase in winter MSLP (local increases exceed +50 Pa) in m900 relative to 125Control 10 ( Fig. 5e). This increase is coincident with a sea ice increase ( Fig. 3d) and cooler SATs (Fig. 2e) over the same region compared to 125Control. Around northern Greenland, the scenario m900 shows a decrease in summer MSLP relative to 125Control (local decreases exceed -50 Pa; Fig. 5f); this is coincident with an decline in sea ice concentration (Fig. 3a). The scenario p900 shows and increase in annual MSLP over central Arctic Ocean (Fig. 5g).
Over Greenland, the surface winds respond to variations in the GIS surface elevation. Strong anticyclonic flow centred over

Changes in precipitation pattern
During summer, the 125Control shows an enhanced precipitation rate compared to PI mainly across southwestern and central Greenland (Fig. 6c). This is in line with results from other climate models (e.g., Otto-Bliesner et al., 2006;Merz et al., 2014).
There is a rise in precipitation rate over much of Greenland throughout the year in m900 compared to 125Control ( Fig. 6d to f). This is expected as the lowering of the orography leads to a wider spread of precipitation across Greenland from the east and 25 west which is blocked by the higher and steeper elevation of the present-day GIS. Local increases over south-east Greenland exceed 0.8 mm/day in m900 during winter (Fig. 6e). This increase in precipitation accords with a reduction in winter sea ice concentration along the east coast of Greenland relative to 125Control (Fig. 3d).
For the increased elevation scenarios, local changes in precipitation rate relative to 125Control are less widespread and smaller than for the decreased elevation scenarios during both seasons ( Fig. 6 d-i). Over south-east Greenland, p900 is up to 30 0.6 mm/day drier than the 125Control simulation during winter ( Fig. 6 h).
Precipitation increases linked to elevation decreases are much larger than the drying linked to elevation increases, implying non-linearities in the climate response to GIS elevation change ( Fig. 1 and Fig. 6). The core-average precipitation lapse rate

10
For the PI simulation, the September mean sea ice extent is 5.8×10 6 km 2 . The 125Control simulation shows a reduced September mean of 4.4 × 10 6 km 2 relative to PI; larger seasonal and latitudinal insolation variations (linked to the orbital forcing) lead to Arctic sea ice loss during summer/spring (e.g., Otto-Bliesner et al., 2006).
GIS elevation reductions lead to an increase in winter sea ice extent, whereas increases in the GIS elevation result in winter sea ice retreat (Fig. 1). In contrast to δ 18 O and SAT, variations in winter sea ice extent are smaller for decreases in GIS elevation 15 compared to increase elevation scenarios. For example, the March sea ice extent is reduced by −4.2% in p900 and increased by +1.7% in m900 compared to the 125Control simulation.
The decreased elevation scenario (m900) displays an increase of winter sea ice concentration on the Norwegian Seas and on the southern-eastern coast of Greenland compared to 125Control simulation (Fig. 3 d). The reduced cyclogenesis off the south-east coast of Greenland (Fig. D1), results in growth of winter sea ice over these regions (Fig. 3 d). This is probably 20 associated with a decrease in wind-driven ocean heat transport (e.g., Pausata et al., 2011;Stone and Lunt, 2013;Davini et al., 2015). The increased elevation scenario (p900) experience the same forcing but in opposite direction (Fig. 3 f and Fig. D1).
We also find some local changes in summer sea ice concentration; while p900 shows decreases of summer sea ice over the Beaufort Sea, it shows increases over the Fram Strait area. Similar patterns are found in m900 but in opposite direction and of lower magnitude (Fig. 3 a,c). 25 We ascribe these changes in summer sea ice to variations in ocean salinity caused by anomalous downwelling or upwelling, induced by anomalously low or high sea level pressure over the Arctic (Jackson and Vellinga, 2012). In HadCM3, the geostrophic balance of the Beaufort gyre can be altered ageostrophically by wind stresses linked to low-frequency sea level pressure variability (Jackson and Vellinga, 2012). Our increased elevation scenario (p900) show high sea level pressure anomalies over the Arctic basin (Fig. 5) which lead to downwelling in the center of the Arctic basin and upwelling along the coasts 30 respectively (Fig. C1). Since the surface water is fresher and colder than the subsurface water, this results in salinification near the coasts and freshening in the center of basin. The same mechanisms apply to the decreased elevation scenario (m900) but in opposite direction (Fig. C1).
The increase in wind speed along the Fram Strait in p900 compared to 125Control and vice versa for m900 (Fig. D1) also affects the advection of sea ice from the Arctic to the Atlantic ocean (Davini et al. (2015)).

The response of the isotopic lapse rate to changes in the background climate state
Malmierca-Vallet et al. (2018) demonstrate the importance of Arctic sea ice changes as a control on LIG Greenland ice core δ 18 O because of its impact on both the regional temperature increase and the moisture source. Thus, we also study 32 simula-5 tions that examine the joint impact of modified Arctic sea ice retreat and modified GIS morphology (considering both changes in the extent and elevation of the GIS -see Table A1 and section 2.1).
We use the sea ice retreat simulations of Malmierca-Vallet et al. (2018) to isolate the impacts of δ 18 O due to sea ice variation.
This allows to test to which extent Arctic sea ice changes may influence isotopic lapse rate values. Fig E1 shows  When considering total δ 18 Op anomalies (relative to 125Control) not corrected for sea ice changes, a non-linear δ 18 O lapse rate is observed over Greenland (Fig 7 -second column); The core-average δ 18 O lapse rate varies from 0.29‰ per 100 m for 20 the enlarged GIS states to 0.49‰ per 100 m for the lowered GIS states (Fig 7 -second column). These results thus strongly suggest a non-linearity in the isotope-elevation relationship, with higher δ 18 O-elevation gradients for lowered GIS states and vice versa.
When further deducting the winter sea ice effect, we find an almost stationary core-average δ 18 O lapse rate, slightly varying from 0.38‰ per 100m for the enlarged GIS scenarios to 0.39‰ per 100m for the lowered GIS scenarios (Fig 7 -third column).

25
The sea ice effect increases δ 18 O-elevation gradients by 28% in the enlarged GIS states and decreases δ 18 O-elevation gradients by −19% in the lowered GIS states. Indeed, this suggests that sea ice changes may strongly influence linearity in the isotopeelevation relationship.
The dependence of the δ 18 O variable on elevation variations occurs in response to variations in winter sea ice extent. GIS elevation reductions lead to an increase in winter sea ice extent, whereas increases in the GIS elevation result in winter sea ice 30 retreat (Fig. 1). Thus, the loss/increase of winter sea ice extent act as a positive/negative feedback on δ 18 O.
Our results are also in agreement with Erokhina et al. (2017), who point to a non-stationarity response of the climate to GIS elevation changes during the Holocene and Last Glacial Maximum (LGM). Erokhina et al. (2017) propose that following the transition from the LGM to the Holocene, mean annual temperature lapse rates over the GIS decreased by almost 20%.

4.3
The response of the isotopic lapse rate to the background climate state Isotope-elevation gradients have tended to be calculated from modern surface data (e.g., Dansgaard, 1973): a present-day spatial 5 relationship is presumed to apply to temporal changes. This disregards any impact that variations in the ice sheet elevation may have on the atmospheric circulation, precipitation patterns and eventually the isotopic composition.
Our idealized elevation change simulations with HadCM3 allow a fuller calculation. We find a smaller core-average δ 18 O lapse rate for enlarged GIS states (0.29‰ per 100 m) than for the lowered GIS states (0.49‰ per 100 m) (Fig 7). Hence, δ 18 O-elevation gradients do not remain constant across the parameter space of elevation changes. This strongly suggest non-10 linearities in the isotopic response to Greenland elevation change.
We also find that winter sea ice variations can increase/decrease modelled core-averaged isotopic lapse rate values by about +28% and −19% for the enlarged/lowered GIS states respectively (Fig. 7). These results thus suggest that sea ice variations may have a strong influence on δ 18 O-elevation gradients, especially at coastal areas such as the Camp century ice core site ( Fig. F1). In particular, at this location, we find that the winter sea ice effect decreases the δ 18 O-elevation gradient by −24% 15 in the lowered GIS states and increases the δ 18 O-elevation gradient by as much as 92% in the enlarged GIS states (Fig. F1 e).
While the largest influence of sea ice on δ 18 O changes is found at Camp Century site, DYE3 site shows the smallest (Fig. F1 e-f). These results point to elevation changes as a likely driver (together with GHGs and orbital forcing) on LIG δ 18 O changes at DYE3 ice core site. This is in agreement with previous LIG GIS modelling studies which propose a significant LIG lowering around the DYE3 area, even the total loss of ice (Robinson et al., 2011;Helsen et al., 2013).
Furthermore, our core-average δ 18 O lapse rates are also somewhat lower than the lapse rate of 0.56‰ per 100 m modelled (with the isotope enabled version of the European Centre Hamburg Model version 4) over Greenland for the LIG period by Sjolte et al. (2014). Note our modelled isotopic lapse rates contemplate the dynamical response of atmospheric circulation to 25 GIS elevation changes and Arctic sea ice variations, whereas previous studies disregard these effects.
These elevation change simulation results thus suggest possible non-linearity in isotope-elevation gradients. It would be useful if this was checked with other models to assess model-dependence in the results.

Conclusions
The results of this study are relevant for the interpretation of past climates from Greenland ice core records. Changing GIS 30 elevation in HadCM3 alters the NH atmospheric circulation circulation, sea ice and precipitation patterns over Greenland and further afield. These climate feedbacks result in lower isotopic-elevation gradients for enlarged GIS states, and vice versa. Our