The research leading to these results has received funding from the UK-NERC consortium iGlass (NE/I009906/1) and is also a contribution to the European Union’s Seventh Framework programme (FP7/2007-2013) under grant agreement 243908, “Past4Future. Climate change - Learning from the past climate”. E. C. is funded by the European Union’s Seventh Framework Programme for research and innovation under the Marie Skłodowska-Curie grant agreement no 600207. A.G. has been supported by the Deutsche Forschungsgemeinschaft (DFG) under the special Priority Program INTERDYNAMIC (EndLIG project, grant GO 2122/1-1). This is LSCE contribution number 6010.

1 - Introduction

1The Last Interglacial (LIG, ~129-116 thousand years Before Present, hereafter ka) is commonly used as a test bed to investigate the response of the Earth System to global warming. Indeed, this period is characterized by a global climate a few degrees warmer than present-day, in particular in high latitudes (e.g. Capron et al., 2014), and a global mean sea level 6 to 9 m above present due to reduced Antarctic and Greenland ice sheets (Dutton et al., 2015) (fig. 1). The warmer-than-present LIG climate derives from high boreal summer insolation values, and not from enhanced greenhouse gas concentrations, which are relatively similar to pre-industrial values (Loulergue et al., 2008; Lüthi et al., 2008; fig. 1). However, the high-latitude warming reconstructed for the LIG may be reached before 2100 (Otto-Bliesner et al., 2013), making this time period at the forefront of paleoclimatic investigations.

2A dynamical representation of the LIG climate evolution is indispensable for two main reasons: (1) to determine the climate dynamics and feedbacks at play during a warmer-than-present climate, and (2) to evaluate, through robust model-data comparisons, the capability of General Circulation Models (GCM), which are used for future climate projections, to reproduce the spatio-temporal evolution of the LIG climate (e.g. Lunt et al., 2013; Bakker & Renssen, 2014). However, providing a climatic sequence of events across the LIG at a global scale is currently highly challenging, mostly because the comparison of paleoclimatic records from different types of archives and from different parts of the world is strongly limited by chronological issues.

3Here, we synthetize recent published efforts on archive’s chronologies and model-data comparisons that improve our spatio-temporal representation of climatic patterns and changes, and our understanding of the mechanisms at play in high latitude regions during the LIG. In section 2, we draw the reader’s attention to the complexities of chronologies in climatic archives (Govin et al., 2015). We use one example to illustrate how different chronological approaches can lead to large age discrepancies between the resulting timescales, and thus to highlight the importance of harmonizing chronologies in paleo-studies comparing multiple records. In section 3, we present a new data synthesis for the high-latitude regions which provides robust evidence for asynchronous LIG surface temperature evolutions across the world (Capron et al., 2014). In section 4, we use model-data comparisons at 130 ka to provide new insights on the climatic mechanisms that shaped the inter-hemispheric temperature asynchrony observed during the early LIG (Capron et al., 2014; Stone et al., 2016).

2 - Chronological limitations during the LIG : example of climatic alignments in marine sediments

4Because of an overall lack of absolute age constraints, it is often necessary to use indirect approaches to infer LIG age-depth models. Thus, chronologies of climatic archives are usually based on record alignment strategies, i.e. one record on a depth scale is aligned onto a dated reference record assuming simultaneous regional changes for a given climate variable (e.g. temperature, Govin et al., 2015). As a result, most LIG records are not displayed on chronologies independent from each other and from the assumption of climatic synchronicity. Here we use the case study of core MD95-2042 from the Portuguese margin to compare five different climatic alignment strategies that are commonly used to develop chronologies in marine sediments (fig. 2). A full evaluation of the underlying hypotheses, limitations and age uncertainties associated with these alignment methods can be found in Govin et al., (2015). We briefly summarize them below.

5(1) The original age model for core MD95-2042 relies on the assumption that benthic foraminifera δ¹⁸O variations mostly reflect global ice volume changes, and age markers were based on the alignment of identified high stand sea levels from the benthic foraminifera δ¹⁸O record with those identified in radiometrically-dated coral terraces (Shackleton et al., 2003). This age model presents severe limitations. First, the radiometric dating of corals at the time was associated with age uncertainties of around 2 ka (Shackleton et al., 2003). More importantly, benthic δ¹⁸O is strongly affected by seawater δ¹⁸O and temperature variations (Skinner & Shackleton, 2006), and cannot be directly related to global sea level changes, in particular on millennial time scales. Last but not least, local coral-based sea level reconstructions were not corrected for glacio-isostatic influences which, since then, have been revealed to be crucial to interpret coral records in term of global eustatic changes (Dutton & Lambeck, 2012). Reported age uncertainties are thus likely underestimated.

6(2) The chronology of marine sediments is often defined by aligning the benthic δ¹⁸O record onto the benthic δ¹⁸O reference LR04 stack (Lisiecki & Raymo, 2005). This alignment strategy assumes that benthic δ¹⁸O variations are synchronous around the world, and at first order reflect global ice volume changes. However, it has been shown that benthic δ¹⁸O variations may not occur at the same time and with the same amplitude around the globe, and thus cannot be used as a global marker (e.g. Skinner & Shackleton, 2005; Waelbroeck et al., 2011). Also, the LR04 stack is associated with large dating errors (i.e. 4 ka) during the LIG.

7(3) The recent improvements of ice core chronologies led to the use of ice core records as targets to define LIG age models in high-latitude sediment cores. Govin et al., (2012) proposed a new age model for core MD95-2042 based on the assumption that glacial-interglacial surface water temperature changes in the subpolar North Atlantic occur in phase with air temperature changes over inland Greenland, which is supported by ¹⁴C-dated North Atlantic records of the last glacial period (e.g. Bond et al., 1993). The planktonic G. bulloides δ¹⁸O record, which is thought to largely reflect local SST changes at site MD95-2042 (Shackleton et al., 2000), is thus aligned to ice core indicators of Greenland temperature (Govin et al., 2012).

8(4-5) The good dating accuracy of speleothems makes them ideal alignment targets to derive age models in marine sediments in their vicinity (e.g. Bar-Matthews et al., 2003; Drysdale et al., 2009). Drysdale et al., (2009) defined another age model (#4) to core MD95-2042 by aligning its SST record to calcite δ¹⁸O from the Italian Corchia cave. To facilitate the comparison with the method #3, we also aligned MD95-2042 G. bulloides δ¹⁸O record to Corchia calcite δ¹⁸O (#5, Govin et al., 2015). Both methods (#4-5) rely on two assumptions:
(i) variations in the strength of the North Atlantic meridional overturning circulation induce synchronous changes in regional SST, air temperature, evaporation and moisture advection to the Italian peninsula, and therefore Italian rainfall, and (ii) the amount of rainfall reaching Corchia cave is the dominant driver of the calcite δ¹⁸O signal (Drysdale et al., 2009). However, the uncertainty related to the interpretation of speleothem tracers, and our understanding of factors controlling them, remains difficult to quantify (Govin et al., 2015), and reported age uncertainties are minimal estimates.

9The comparison of these multiple chronologies (#1, 2, 3, 5; fig. 2) for one marine core shows age discrepancies that can be higher than 4 ka across Termination II, the LIG and the glacial inception. These age offsets are larger than or within the range of 1^st age uncertainties (from 0.4 to 4.2 ka, depending on the alignment methods; Govin et al., 2015). Therefore, this exercise shows that different chronological methodologies can produce very high age differences across the LIG, and calls for caution when comparing LIG marine records using various alignment targets or different reference time scales.

10Also, the comparison of alignment methods #4 and 5 reveals age differences up to 5 ka during the deglaciation, depending on whether the alkenone-derived SST or planktonic δ¹⁸O record is aligned to Corchia calcite δ¹⁸O (fig. 2). Thus, this example shows that, not only the type of alignment method, but also the type of paleo-tracers (e.g. reconstructing SST) used within a same alignment method, can lead to very high age differences (Govin et al., 2015).

11Therefore, this case study highlights (i) the importance of reporting precisely the alignment method for a given studied site, (ii) the importance of checking the coherence of age models when integrating multiple records, in particular from different regions, and (iii) the need to systematically estimate relative and absolute chronological uncertainties and integrate them in paleoclimatic investigations (Govin et al., 2015). As further illustration of these issues, we show in the next sections how harmonizing chronologies of records gathered in LIG data syntheses leads to our improved understanding of the LIG climate evolution and related processes.

3 - A new LIG climate data synthesis with harmonized chronologies

12Existing LIG surface temperature data syntheses (e.g. Turney & Jones, 2010; McKay et al., 2011) present two main limitations. First, they use the original chronologies of paleoclimatic records, whose coherence has not been checked, possibly leading to high age discrepancies between records (see section 2). Second, these data syntheses consist of a single time slice focused on the peak of LIG warmth. This approach is problematic, because there are multiple evidences that LIG surface temperatures peaked asynchronously around the globe (e.g. Bauch & Erlenkeuser, 2008; Govin et al., 2012).

13Here we present a new LIG surface temperature data synthesis for high-latitude regions (Capron et al., 2014), the first one with harmonized chronologies. We compiled 5 surface air temperature records above Greenland and Antarctica, and 43 sea surface temperature records from the North Atlantic and the Southern Ocean poleward of 40°N and 40°S across the LIG. Surface air temperature records are deduced from water stable isotopic profiles of ice cores. Sea surface temperatures (SST) are reconstructed from foraminiferal Mg/Ca ratios, alkenone unsaturation ratios, and faunal assemblages in marine sediment cores. The common temporal framework is based on the alignment of SST records to ice core records (Capron et al., 2014) onto the recent Antarctic ice core chronology AICC2012 (Bazin et al., 2013; Veres et al., 2013). The rationales for using this strategy are threefold:

14(1) It allows integration of both marine and ice core records.

15(2) It better reproduces known high-latitude climate features of the penultimate deglaciation and the LIG than benthic δ¹⁸O alignments to the LR04 stack (Govin et al., 2015).

16(3) AICC2012 is the first integrated ice core timescale over the LIG, based on a multi-site approach including five Greenland and Antarctic ice cores. The numerous new stratigraphic links significantly reduce the absolute dating uncertainty down to ± 1.6 ka (1δ) across the LIG (Bazin et al., 2013), making it a particularly well-constrained reference time scale. Finally, we quantified the surface temperature uncertainties using a Monte-Carlo analysis (i.e. reconstruction of 1000 different age-models per site) that integrates temperature reconstruction errors, and propagates chronological uncertainties (see Capron et al., 2014 for details). These results in a final 2δ uncertainty estimate of ± 2.6 °C on average for SST records and ± 1.5 °C for surface air temperature Antarctic records.

17We focus now on the four time slices built at 1++30, 125, 120, and 115 ka, for which we calculated temperature anomalies relative to present-day conditions and associated 2δ uncertainties over a 2 ka time windows centred on each date (fig. 3; see Capron et al., 2014 for details). We observe warmer-than-present-day climatic conditions at 130, 125, and 120 ka in the southern hemisphere, and at 125 and 120 ka only in the northern hemisphere. Thus, warmer than present-day conditions lasted longer in the southern high latitudes than in the northern ones. The magnitude of temperature changes is also larger in the northern than in the southern high latitudes during the LIG onset and demise (Capron et al., 2014). We take a particular look at the 130 ka time slice. It reveals significantly cooler-than-present-day conditions in the northern high latitudes i.e. on average 5.0 ± 0.7 °C cooler), and slightly warmer than present-day conditions in most southern hemisphere sites, i.e. 1.4 ± 0.4 °C and 1.7 ± 0.3 °C warmer than present-day on average in the Southern Ocean and above Antarctica, respectively. It thus highlights non-synchronous temperature maximum in the two hemispheres. We attribute this interhemispheric asynchrony (e.g. Govin et al., +2012; Capron et al., 2014) to the “bipolar seesaw” mechanism, induced by changes in the intensity of the Atlantic Meridional Overturning Circulation (AMOC; Stocker & Johnsen 2003). The melting of northern ice sheets extended beyond the penultimate deglaciation into the early LIG, and the associated release of meltwater into the North Atlantic may have delayed the full establishment of a vigorous AMOC, resulting in peak Antarctic temperatures while maintaining cold the North Atlantic (Govin et al., 2012).

18Despite its limited spatial extent, this new data synthesis gives so far the sole dynamical representation of LIG high-latitude temperature changes, revealing asynchronous LIG peak temperatures around the globe. In addition, while previous data syntheses (e.g. Turney & Jones, 2010; McKay et al., 2011) led to limited model-data comparison outcomes (Otto-Bliesner et al., 2013; Lunt et al., 2013) because they mix the timing of LIG peak warmth in a single time slice (fig. 4a), our new data synthesis provides improved benchmark for climate model snapshot simulations performed across different time intervals of the LIG (e.g. 130 ka, 125 ka, 120 ka), which we illustrate in the next section.

4 - Toward improved LIG model-data comparison exercises

19Model-data comparison exercises are useful to help disentangling the forcing and feedbacks shaping the LIG spatio-temporal climate evolution. Here we focus on the time slice at 130 ka, and we first compare it with a snapshot simulation performed at 130 ka (see Lunt et al., 2012 for details) with the GCM HadCM3 (Gordon et al., 2000) in the framework of the Paleoclimate Model Intercomparison Project (Phase 3, PMIP3; details on the experimental design of the model simulations can be found in Lunt et al., 2012). Our specific time slice at 130 ka shows that the HadCM3 model reproduces neither the colder-than-present-day North Atlantic conditions nor the early southern hemisphere warming reconstructed at 130 ka (fig. 4b; Capron et al., 2014). This experiment only simulates the climate response to the static orbital and greenhouse gas forcing at 130 ka, and thus, it does not take into account the potential dynamic AMOC response to a persistent melting of northern ice sheets (e.g. Govin et al., 2012, Sanchez-Goñi et al., 2012), that, we think, is responsible for the bipolar seesaw pattern identified from the data synthesis (section 3). However, other simulations suggest that reproducing the high-latitude climate features at 130 ka is a more complex issue. Indeed, a recent study (Langebroek & Nisancioglu, 2014) using an alternative GCM shows cooling conditions in the North Atlantic region at 130 ka without the need to invoke a freshwater forcing, while considering a feedback linked to the disintegration of the West Antarctic ice sheet resulted in an additional warming over Antarctica with the HadCM3 model (Holden et al., 2010). These various results call for further investigations.

20In order to further examine the role of northern fresh-water input on the asynchronous pattern observed at 130 ka, we perform sensitivity tests with the HadCM3 model with a freshwater forcing into the North Atlantic of varying intensity (from 0.1 to 1 Sv; see details in Stone et al., 2016). The best fit between model and data results is obtained when using a freshwater forcing of 0.2 Sv, a realistic value considering coral data-based constraints (Carlson, 2008). It is now possible to produce a modelled climate response similar to the colder-than-present North Atlantic temperatures and warmer-than-present Southern Ocean temperatures observed at 130 ka (fig. 4c). Note that it also leads to simulated surface air temperatures over Europe that are consistent with existing continental surface air temperature reconstructions (e.g. Sanchez-Goñi et al., 2012; see fig. 7 from Stone et al., 2016). These new results confirm that the impact of meltwater from the remnant northern ice sheets on global climate via the bipolar seesaw mechanism is a plausible mechanism to explain the early LIG climate as illustrated in the 130 ka time slice. An additional snapshot simulation in which the West Antarctic Ice Sheet is also removed, leads to warming in East Antarctica and the Southern Ocean, although it does not appreciably improve the model-data comparison (fig. 4).

21Beyond these results, the time slices and time-series included in our data synthesis have been used to evaluate both LIG transient and snapshot experiments from models of various complexity (Loutre et al., 2014; Pfeiffer & Lohmann ,2016). It was also recently used as a climate input for ice sheet modelling exercises investigating the contribution of Greenland and Antarctic to LIG sea level changes (DeConto & Pollard, 2016).

5 - Final remarks

22In this paper, we summarize three recent and complementary publications on the LIG climate. These new results highlight how the coupling of robust data synthesis with harmonized chronologies and of appropriate climate model experiments significantly improved our spatio-temporal representation of the LIG high-latitude climate evolution, and our understanding of climatic mechanisms that shaped the inter-hemispheric temperature asynchrony during the LIG. We stress how crucial it is to systematically associate climate alignments used to derive archive’s chronologies with (1) a clear statement of the underlying climatic hypothesis, (2) a detailed understanding of environmental parameters controlling selected tracers, (3) a careful evaluation of the synchronicity of aligned paleoclimatic records, and (4) the integration of carefully estimated chronological uncertainties in paleoclimatic investigations.

23Our new synthesis with harmonized chronologies provides the first dynamical representation of the LIG climate evolution across the high latitude regions. It illustrates the asynchronous establishment of LIG peak temperature between the two hemispheres, with surface temperatures peaking at 129.3 ± 0.9 ka in the southern hemisphere, compared to 126.4 ± 1.9 ka in the North Atlantic and 126.9 ka over Greenland (Capron et al., 2014). The 130 ka time slice provides a very clear illustration of this early LIG climate asynchrony: North Atlantic surface waters were 5.0 ± 0.7 °C cooler than present-day, while surface conditions were 1.4 ± 0.4 °C and 1.7 ± 0.3 °C warmer than present in the Southern Ocean and above Antarctica, respectively. At 125 ka, all high-latitude regions experienced warmer-than-present conditions, reaching 1.6 ± 0.5 °C, 0.8 ± 0.5 °C and 1.5 ± 0.2 °C, in the North Atlantic, the Southern Ocean and Antarctica, respectively.

24The existence of multiple data-based time slices enables one to evidence important missing processes in state-of-the-art model climate simulations to explain and reproduce the LIG climate evolution. In particular, our integrated model-data approach shows that the input of northern freshwater is an important player in the evolution of the early LIG climate. Our studies also highlight that it is key to compare model simulation with data from the appropriate time period. For this reason, we recently built a 127 ka time slice based on our data synthesis to be used as a benchmark to evaluate the upcoming LIG core simulation at 127 ka that will be performed in the framework of Phase 4 of PMIP (PMIP4) and Phase 6 of the Coupled Model Intercomparison Project (CMIP6) (Capron et al., in revision).

25An on-going initiative aims to produce a more comprehensive LIG synthesis of global spatial extent. It will also be associated with a coherent dating, quantitative uncertainty estimates, and it will extend to other parameters than temperature (e.g. stable isotopes) to test the representation of additional climatic tracers in the models included in the CMIP6/PMIP4 exercises.