Greenland climate simulations show high Eemian surface melt

This study presents simulations of Greenland surface melt for the Eemian interglacial period (~130000 to 115000 years ago) derived from regional climate simulations with a coupled surface energy balance model. Surface melt is of high relevance for ice core records because it can influence observations, e.g., lower the preserved total air content (TAC) used to infer past surface elevation. An investigation of surface melt is particularly interesting for warm periods, such as the Eemian interglacial period, with enhanced surface melt. Furthermore, Eemian ice is the deepest and most compressed ice preserved on Greenland, 5 which means that melt layers can not be identified visually. Therefore, a knowledge of potential melt layers would be advantageous. The simulations presented here show Eemian surface melt at all deep Greenland ice core locations. Estimated TAC, based on simulated melt during the Eemian, could explain the lower TAC observations: at the summit of Greenland (GRIP) a refreezing ratio of more than 25 % of the annual accumulation is simulated. As a consequence, elevated levels of surface melt during warm periods should be considered when interpreting Greenland TAC measurements as surface elevation changes. 10 Additionally to estimating the influence of melt on past TAC in ice cores, the simulated surface melt could also be used to identify potential coring locations where Greenland ice might be best preserved.


Introduction
The Eemian interglacial period (~130000 to 115000 years ago; thereafter~130 to 115 ka) was the last period with a warmerthan-present summer climate on Greenland (CAPE Last Interglacial Project Members, 2006;Otto-Bliesner et al., 2013;Capron  In this study, the SEB-derived SMB simulations are analyzed at six deep Greenland ice core locations -Camp Century, Dye-3, EGRIP, NEEM, NGRIP, GRIP -and an adjacent ice cap -the Agassiz ice cap (Fig. 1). Due to model topography misrepresentation at the ice sheet margins, i.e., the model topography is lower than in reality at the Agassiz ice cap location 65 (model resolution 25 km), a substitute location (Agassiz_sub) in the vicinity of the ice cap, with a model elevation similar to the observed elevations, is chosen (Tab. 1).

Observed surface melt
The pre-industrial regional climate and SMB simulations are validated against satellite and temperature observations at the locations of interest. The two observational melt day data sets are both derived from satellite-borne passive microwave ra- is identified by comparing 37 GHz, horizontally polarized (37 GHz H-Pol) brightness temperatures with dynamic thresholds associated with a melting snowpack (Mote, 2014). Unfortunately, the Agassiz ice cap is not covered by this data set. The sec-75 ond data set, T19H melt , covers May through September for most years between 1979 to 2010 on the 25 km MAR grid. It uses data collected at K-band horizontal polarization (T19H) with a constant brightness temperature threshold of 227.5 K (Fettweis et al., 2011). Both satellite data sets are discussed to show their different sensitivities and to illustrate the uncertainty of these satellite-based melt observations.
The seasonal temperature observations at weather stations and 10 m borehole temperatures (representing annual mean tem-80 peratures) are taken from a collection of shallow ice core records and weather station data (Faber, 2016). Finally, the bore hole temperatures from the Agassiz ice cap are taken from Vinther et al. (2008).

Observed total air content (TAC)
Firstly, the Dye-3 TAC for the ice core depth range of~240 to 1920 m was extracted from Herron and Langway (1987, Fig. 4 therein). Since Souchez et al. (1998) indicate that ice from warmer periods (higher δO 18 values), likely Eemian, is located 85 below 2000 m at Dye-3, the presented Dye-3 TAC record does not represent Eemian conditions. Secondly, the NEEM TAC observations (NEEM community members, 2013) cover the deepest section of the NEEM ice core from~2200 to 2500 m depth (corresponding to an age of~75 to 128 ka; not continuous) and an example for Holocene conditions from depths be-tween~100 to 1400 m (no age provided). Thirdly, the NGRIP TAC record (Eicher et al., 2016) includes the entire core from 130 to 3080 m, however the sampling resolution varies. An age model is provided for the entire data set with an oldest age of 90~1 20 ka. Finally, the GRIP TAC data set (Raynaud, 1999) covers depths from~120 to 2300 m and~2780 to 2909 m, while an age mode is only provided for the upper part (oldest ice 41 ka). For the deeper sections of the core, a published unfolding of the GRIP core (Landais et al., 2003, age bands in Fig. 3 therein) is used to assign an age to the observations. Note that only the Eemian sections for NEEM, GRIP, and NGRIP are shown in Fig. 7.
The Eemian ranges in Fig. 6  Holocene range for NEEM is calculated from the entire Holocene example provided in the NEEM community members (2013) data (nine data points; no age provided).

100
Calculation of the model-derived total air content (TAC) The model-derived TAC is calculated with the annual mean surface pressure and the annual mean near-surface temperature from the MAR regional climate simulations at every location of interest (Martinerie et al., 1992;Raynaud et al., 1997): where V c is the pore volume at close-off in cm 3 /g of ice, P c the mean atmospheric pressure at the elevation of the close-105 off depth interval in mbar, T c the firn temperature prevailing at the same depth interval in K, P 0 the standard pressure (1013 mbar), and T 0 the standard temperature (273 K). V c is calculated as a function of T c following an empirical relation (Martinerie et al., 1994;Raynaud et al., 1997): (RZ per ): where T AC ref rozen is calculated using Henry's solubility law (Sander, 2015) for N 2 and O 2 (neglecting other atmospheric gases) to account for air that is dissolved in the meltwater before refreezing: 115 with C a,N 2 , and C a,O2 being the aqueous-phase concentration of N 2 and O 2 , respectively: and C a,N 2 = P c * C atm,O2 * H cp,O2 where C atm,N 2 and C atm,O2 are the atmospheric concentration ratio (0.79 and 0.21) and H cp,N 2 , H cp,O2 are Henry's 120 solubility constants (10.49 × 10 −6 and 2.1982 × 10 −5 ) for N 2 and O 2 , respectively.

Temperatures
The simulated pre-industrial annual mean (near-surface) temperatures at the seven locations of interest (   The precipitation-weighted temperatures (Fig. A1), which is arguably closer to what is recorded in an ice core, show a similar pattern as the JJA temperatures (Fig. 3). However, the precipitation-weighted temperatures show a less pronounced warming for mid Eemian conditions (125 ka: orange), i.e., maximum 3°C warmer compared to pre-industrial (black). where refreezing percentage reaches 90% and more.

Total air content (TAC)
Theoretical TAC derived from simulated surface pressure and annual mean temperature (Raynaud et al., 1997) and reduced according to the amount of simulated refreezing ( Fig. 6 and Sec. 2) shows significantly lower values for the 125 ka simula-     The y-axes are reversed.

Discussion
The enhanced Eemian seasonality (Yin and Berger, 2010) and warmer Eemian summers (CAPE Last Interglacial Project the entire Greenland ice sheet, even at the summit of Greenland, is possible under recent climate conditions. Even though these extreme Greenland-wide melt events were caused by a rare large-scale atmospheric pattern (Neff et al., 2014) and were further enhanced by an externally caused albedo lowering (ash deposition from forest fires; Keegan et al., 2014), it is likely that such 185 events are more frequent in a warmer climate such as the Eemian interglacial period.
The simulations discussed in this study (regional climate plus a full surface energy balance) indicate surface melt and refreezing ( Fig. 4 and 5) at all deep Greenland ice core locations. Even central Greenland locations close to Summit show a melt of 100 mm year −1 (Fig. A2). Due to this surface melt, TAC derived from these simulations are between~25 % (GRIP) and~80 % (Dye-3, EGRIP) lower than modern (pre-industrial) values (Fig. 6). Even though the presented climate simulations Furthermore, the absence of a simulated annual warming, and proxy data showing Eemian peak temperatures as high as +7.5 ± 1.8°C (NEEM community members, 2013, without altitude corrections) and +8.5 ± 2.5°C (Landais et al., 2016) for NEEM (the North Greenland Eemian Ice Drilling project in northwest Greenland), and +5.2 ± 2.3°C (Landais et al., 2016, lower bound as the record only starts after the peak Eemian warming) for NGRIP (North Greenland Ice Core Project) 200 indicate that the climate simulations might be cold. The simulated JJA temperatures (Fig. 3) and the simulated precipitationweighted temperatures (Fig. A1) show a peak warming of only~3-4°C and~3°C, respectively. However, the fact that NEEM community members (2013) infer an elevation (at the deposition site) of several hundred metres higher than at NEEM today complicates the interpretation of how well the simulated temperatures fit the proxy-derived observations. If the estimated NEEM deposition site was lower than inferred, e.g., stronger influenced by melt, then the warming of~8°C at constant 205 elevation would be too high and the simulations would fit better.
Focusing again on the comparison of melt observations and simulations (Fig. 4), a strong underestimation of melt at the Agassiz site in the pre-industrial simulations becomes apparent. This strong underestimation is likely related to the fact that a substitute location for the ice cap is used, i.e., a model location with similar model and observed elevation is chosen. The selection of the substitute location is necessary, because the low model topography at the original core locations causes un-210 realistically high melt simulations. Furthermore, the Agassiz site is only covered by the satellite data which appears to be less sensitive to melt (T19H melt ), i.e., T19H melt shows less melt than MEaSUREs at all investigated locations. And although Eemian ice is absent at the Agassiz site, the simulated Eemian refreezing percentage (Fig. 5) of approximately 80% is consistent with the Agassiz melt record which indicates 100% melt during the Holocene optimum~10-11 ka (Fisher et al., 2012;Lecavalier et al., 2017).

215
Another important aspect for the interpretation of the simulated melt is the formation of melt layers and the amount of meltwater needed to form a (visible) melt layer. While the presented TAC calculations assume Henry's solubility law (Sander, on prevailing snow properties. A higher number of melt layers is not just the result of uniformly higher summer temperatures, but the result of an increased contrast between the pre-melt snow pack temperatures (strongly influenced by winter temperature) 220 and the summer melt rate (a function of summer temperature) (Pfeffer and Humphrey, 1998). Therefore, the enhanced Eemian seasonality might have been favourable for the formation of melt layers.
The simulated TACs for 125 ka conditions are mostly lower than the observations (Fig. 6 and 7) particularly at GRIP, NGRIP, and Dye-3. At NEEM on the other hand -the ice core with the most complete Eemian record (likely including peak warming) -the simulated 125 ka TAC matches well with the lowest observations, indicating that the high amount of simulated melt 225 could explain these observations. The variability of the observed NEEM TAC in the suggested melt zone between 127 and 118.3 ka (gray shading; NEEM community members, 2013) is large, likely due to the varying influence (i.e., number) of melt layers.
The Eemian TAC measurements at GRIP and NGRIP also show reduced values (not as low as at NEEM), which can be interpreted in a similar way as at NEEM -GRIP and NGRIP might have been influenced by Eemian melt as well. The 230 simulated 125 ka TAC for both locations are strongly reduced (relative to pre-industrial levels), but do not reach levels as low as at NEEM. However, these reduced TAC levels could still indicate significant surface melt.
Overall the lack of a better agreement between observed and simulated Eemian TAC (i.e., few TAC observations as low as the simulations) could be related to the sparse number of Eemian peak warming observations (most ice core records only start after the peak warming; particularly at GRIP, NGRIP, and Dye-3). However, another possible explanation could be a shift of 235 the precipitation rates in central Greenland towards much higher values during the Eemian interglacial period. Unfortunately, accumulation rates are unconstrained for the Eemian sections of Greenland ice cores.
Furthermore, another uncertainty to the interpretation of the simulations is the effect of the higher Eemian summer insolation on the TAC. An anti-correlation between local summer insolation and TAC is known in ice core records from East Antarctica during the last 400000 years (Raynaud et al., 2007) and the insolation signal is also found in Greenlandic TAC (NGRIP,  (Koerner, 1989). However, this scenario was rejected by later ice core studies showing evidence of Eemian ice (especially NGRIP and NEEM;North Greenland Ice Core Project members et al., 2004;NEEM community members, 2013). Furthermore, GRIP TAC measurements (Raynaud, 1999) have been interpreted as evidence for the elevation of the summit sites having remained above 3000 m of altitude during the Eemian and GRIP deuterium excess measurements remain in the normal range during the Eemian (Landais et al., 2003 surface air temperature and local wind patterns, and the simulated melt could be analyzed and then be used in order to identify specific weather patterns leading to the simulated high surface melt on Greenland.

Conclusions
The regional climate simulations (including a full surface energy balance) discussed in this study show surface melt at all Greenland ice core locations during the Eemian interglacial period. The amount of simulated refreezing exceeds 25 % of the 270 annual accumulation at the summit of Greenland (GRIP) and reaches values as high as 90 % at less central locations like Dye-3 and EGRIP. The simulated air pressure, temperature, and refreezing are used to calculate estimates of Eemian total air content which could explain the lowest corresponding ice core observations. This is true even though the discussed simulations could show conservative melt estimates, i.e., the simulations neglect processes which would likely increase the melt. Therefore, the possibility of widespread surface melt should be considered for the interpretation of Greenlandic total air content records from 275 warm periods such as the Eemian interglacial period. In the future, a robust map of Eemian melt estimates in Greenland could be used in combination with patterns of accumulation to identify potential sites for future Greenland ice cores. This procedure would increase the chances of finding well preserved Eemian ice without or with less influence by melt layers. These sites will have relatively high accumulation combined with low surface melt.