The reconstructed Eemian sequence (128.5–114 kyr bp, Fig. 2) allows for initial climate interpretations of this period. As mentioned above, the regular occurrence of melt features at 127–118.3 kyr bp is an indication of warmer temperatures at the depositional surface locations of the ice than the mean of the recent millennium at NEEM. This is independently confirmed by the decrease of δ15N in this zone, which is indicative of ∼5 °C warmer mean annual firn temperatures at the depositional site22,23 (Supplementary Fig. 7). Between 128.5 kyr bp and 126.0 kyr bp, δ18O ice increases from −35‰ to −31.4‰ while EDML δ18O ice values slowly drop from those of the warm early Antarctic period24. This bipolar see-saw behaviour suggests that inter-hemispheric heat redistribution by the Atlantic meridional overturning circulation25 was taking place within the Eemian interglacial period; this has also been observed during the last Northern Hemisphere deglaciation 19–11 kyr bp24,25,26,27. Before surface melt began between 128.5 and 126.7 kyr bp, the air content at the depositional site had a stable level of 85 ml kg−1 compared to the present level of 97.5 ml kg−1. When corrected for changing local summer insolation28,29,30 (Supplementary Fig. 10), the air content difference suggests a surface elevation at the depositional site 540 ± 300 m higher at the onset of the Eemian (128 kyr bp) than the surface elevation at NEEM today28,31,32,33. The locations of the depositional sites of the Eemian ice found in the NEEM ice core are modelled using a nested three-dimensional flow model34 (Fig. 1a and Supplementary Fig. 1). A second model used to date the NEEM ice core reaches similar locations35 within 20 km. At present, the surface elevation at the depositional site of the 128-kyr-BP ice (205 ± 20 km upstream from NEEM) is 330 ± 50 m higher than the present at NEEM33,34. The surface elevation increase of 210 ± 350 m at the 128-kyr-BP depositional site (Fig. 4c, blue) is the difference between the elevation at 128 kyr bp (540 ± 300 m) and the present elevation (330 ± 50 m), both related to the present elevation at NEEM. This surface elevation increase is expected at the onset of a warm climatic period due to increased precipitation and mass balance changes that occur before the central part of the ice sheet adjusts to the warmer climate by increasing the ice flow. This is also established at the onset of the present interglacial at 11.7 kyr bp32.

Figure 4: Reconstruction of the temperature and elevation history. Reconstruction of the temperature and elevation history through the Eemian based on the stable water isotopes (δ18O ice ) and the air content records. The zone with surface melt (127–118.3 kyr bp) is shaded in light grey. a, The measured δ18O ice record (black) on the constructed timescale. The average of the recent millennium (−33.60‰) is marked with a thin black line. It is seen that the δ18O ice values at the depositional locations in the melt zone are above −33.0‰ (grey horizontal line). The fixed-elevation change of temperature—constructed from the observed δ18O ice , the elevation changes determined from the air content and the upstream corrections (curves below)—is shown as a red curve using the red axis. The standard error range (orange shading) is a sum of the error of the δ18O ice and the elevation change correction (Supplementary Information 1.1, equations (2) and (5) in Supplementary Information, Supplementary Table 2). b, Air content (black) is plotted and guided by the two stable levels on each side of the melt zone. A dot-dashed line connecting these levels has been suggested with an error range as the dark grey shaded area. The standard error range is a sum of the error assumed in the zone with surface melt (127–118.3 kyr bp) and the 1% error on the air content measurements (Supplementary Information section 1.4). In addition, the average level 107–105 kyr bp is marked with a horizontal black bar. The changes in the air content are caused by pressure changes due to changing surface elevation at the depositional sites and changes to the air trapping processes in the firn assumed to be controlled by the changing summer insolation28,32,49,50. c, When corrected for upstream flow (cyan) and summer insolation changes (green), the air content curve can be ‘translated’ to elevation changes (blue, dashed) with the shaded zone indicating the uncertainty range introduced by this translation. Blue bars mark the air content of the ‘translated’ air content black bars. The standard error range is based on the error range of the air content (dark grey shaded area) and the additional standard errors from calculation of the elevation changes (equation (5) in Supplementary Information, and Supplementary Table 2). Full size image Download PowerPoint slide

In the period 127–118.3 kyr bp, the air content in the ice where surface melt occurred was highly variable and cannot directly be used for ice elevation reconstructions (Fig. 2, shaded zone). We can tentatively estimate elevation changes through the Eemian climate period by connecting the two air content levels before and after the melt zone (Fig. 4b) after correcting for summer insolation, which accounts for 50% of the observed change (Fig. 4c, Supplementary Information). At 126 kyr bp the surface elevation was 45 ± 350 m higher than at present. The δ18O ice increased to −31.4‰ at 126 kyr bp, exceeding the current mean value of the recent millennium of −33.6‰ (at the NEEM site) and the current mean value of −35.0‰ at the depositional site32,33 (Supplementary Information, section 2). Using the temperature–isotope relation of 2.1 ± 0.5 K ‰−1 (calibrated using data from the present interglacial32), the 3.6‰ anomaly at 126 kyr bp implies that precipitation-weighted surface temperatures were 7.5 ± 1.8 °C warmer at the depositional site compared to the last millennium. Note that the modelled location of the depositional site is the only modelled parameter required to compare the 126-kyr-BP data to the present-day data at the depositional site. When further correcting for the more uncertain elevation change of 45 ± 350 m at the 126-kyr-BP depositional site using a lapse rate of 7.5 ± 0.5 K km−1, the fixed-elevation temperature increase here is 8 ± 4 °C (Fig. 4a, red). Our data depict a gradual cooling until 110 kyr bp (Fig. 4a, red curve).

The reconstructed precipitation-weighted annual temperature changes are remarkably high. In general, warmer summer temperatures are reported from palaeorecords36,37, and a few find temperatures at 126 kyr bp on high Arctic latitudes as high as those reported from NEEM38,39,40. Climate models equipped with water stable isotopes point to a limited (1 °C) seasonality bias caused by a stronger enhancement of temperature and precipitation in summer than in winter41,42,43. A large spread in temperature has been reported among simulations of the last interglacial climate, which appear to systematically underestimate North Atlantic/Arctic warming, possibly due to missing vegetation and ice-sheet feedback37,42,43.

Within 6,000 yr, from 128 to 122 kyr bp, the surface elevation is estimated to have decreased from 210 ± 350 m above to 130 ± 300 m below the present surface elevation, which translates to a moderate ice thickness change of 400 ± 350 m after accounting for isostatic rebound. Based on this estimate, the ice thickness at NEEM decreased by an average of 7 ± 4 cm per year between 128 and 122 kyr bp and stayed at this level until 117–114 kyr bp, long after surface melt stopped and temperatures fell below modern levels.

Even with minimum ice thickness of only about 10% less than the present ice thickness at the NEEM site, as reported here, substantial melting can cause significant reduction of ice thickness near the margins; this in turn reduces the volume of the Greenland ice sheet. Although the documentation of ice thickness at one location on the Greenland ice sheet cannot constrain the overall ice-sheet changes during the last interglacial period, the NEEM data can only be reconciled with Greenland ice-sheet simulations30 that point to a modest contribution (2 m) to the observed 4–8 m Eemian sea level high stand44,45. For comparison, no continuing elevation change has so far been detected in areas with elevations above 2,000 m in north Greenland during the past few decades46. These findings strongly imply that Antarctica must have contributed significantly to the Eemian sea level rise47.

Despite the complex ice flow, the disturbed record of the deep ice in the NEEM ice core can be unambiguously reconstructed. The anatomy of the last interglacial shows that Greenland temperatures peaked after the onset of the Eemian, 126 kyr bp, with temperatures (at fixed elevations) 8 ± 4 °C warmer than the average of the recent millennium and multiple indications of summer melt. Temperatures gradually decreased during the interglacial, very probably owing to the strong local summer insolation decreasing trend. The surface elevation first increased due to increased mass balance to 210 ± 350 m above the present at 128 kyr bp, then decreased to 130 ± 300 m below the present elevation around 122 kyr bp. Our results provide multiple new targets to constrain coupled climate/ice-sheet models. Our record, together with recent observations of rainfall and strong surface melting in July 2012 at NEEM, show that conditions are conducive to the start of melt layer formation at NEEM, with the 2010–12 mean annual surface temperatures 1–2 °C above the 1950–80 average.

Our results have implications for both ice deformation near the bedrock and the response of the Greenland ice sheet to climate change. The combination of high-resolution RES data and NEEM glacial–interglacial ice layers brings new knowledge of the near-bed deformation of ice. We believe that the folding and disturbances we observe near the bed are strongly related to the rigid deformation properties of the interglacial ice. This offers an alternative explanation for the large anomalies in RES profiles recently observed under both the Antarctic and Greenland ice sheets, which were previously attributed to refrozen basal water48.