The Ross Sea Dipole &ndash; Temperature, Snow Accumulation and Sea Ice Variability in the Ross Sea Region, Antarctica, over the Past 2,700 Years

. High-resolution, well-dated climate archives provide an opportunity to investigate the dynamic interactions of climate patterns relevant for future projections. Here, we present data from a new, annually-dated ice core record from the 20 eastern Ross Sea. Comparison of the Roosevelt Island Climate Evolution (RICE) ice core records with climate reanalysis data for the 1979-2012 calibration period shows that RICE records reliably capture temperature and snow precipitation variability of the region. RICE is compared with data from West Antarctica (West Antarctic Ice Sheet Divide Ice Core) and the western (Talos Dome) and eastern (Siple Dome) Ross Sea. For most of the past 2,700 years, the eastern Ross Sea was warming with increased snow accumulation and perhaps decreased sea ice extent. However, West Antarctica cooled whereas the western 25 Ross Sea showed no significant temperature trend. From the 17 th Century onwards, this relationship changes. All three regions now show signs of warming, with snow accumulation declining in West Antarctica and the eastern Ross Sea, but increasing in the western Ross Sea. Analysis of decadal to centennial-scale climate variability superimposed on the longer term trend reveal that periods characterised by opposing temperature trends between the Eastern and Western Ross Sea have occurred since the 3 rd Century but are masked by longer-term trends. This pattern here is referred to as the Ross Sea Dipole, caused by 30 a sensitive response of the region to dynamic interactions of the Southern Annual Mode and tropical forcings. 70 year periodicity from 0-1000 CE. The high coherency suggests that RICE and Siple Dome respond to similar forcings. The coherence analysis between RICE and WDC shows an enduring in-phase correlations from ~1000 CE to today for the bandwidth of 200 - 700 years. An anti-phase coherence is found from 0 - 500 CE. The coherence analysis between RICE and TALDICE identifies strong relationships predominantly for the early part of the records, from 660 BCE to ~500 CE and a weak coherence from about 1100-1400 CE, when RICE leads by ~75-100 years. The analysis suggests that for the past 2.7 ka the eastern Ross Sea (RICE, Siple Dome) and western West Antarctica (WDC) are climatologically closely linked in their response to forcings on decadal to centennial time scales. The relationship between the western (TALDICE) and eastern (RICE, Siple Dome) Ross Sea experienced a marked until ~500 we to represent a to and negative SAM forcing of region. (WDC). However, when longer-term trends are removed from the correlations, we find earlier periods of a strong Ross Sea dipole which we interpret to reflect previous time periods dominated by strongly positive SAM conditions from the 3 rd - 6 th Century and 9-12 th Century. Our observations are consistent with the reconstruction of the strongly negative SAM from the 14-18 th Century and the positive SAM from the 19 th Century to today. The continued improvements of array reconstructions (Stenni et al., 2017; Thomas et al., 2017) are an exciting development to further our knowledge of the drivers and effects of 5 past change and their implications for future projections.


Introduction
With carbon dioxide (CO2) and global temperatures predicted to continue to rise, model simulations of the Antarctic / Southern Ocean region show for the coming decades an increase in surface warming resulting in reduced sea ice extent, weakened Antarctic 35 Bottom Water formation, intensified zonal winds that reduce CO2 uptake by the Southern Ocean, a slowing of the southern limb of the meridional overturning circulation (MOC) and associated changes in global heat transport, and rapid ice sheet grounding line retreat that contributes to global sea level rise (Russell et al., 2006;Toggweiler and Russell, 2008;Anderson et al., 2009;Sen Gupta et al., 2009;Downes et al., 2010;Joughin and Alley, 2011;Marshall and Speer, 2012;Spence et al., 2012;Kusahara and Hasumi, 2013;Golledge et al., 2015;DeConto and Pollard, 2016;DeVries et al., 2017). Observations confirm an ozone-depletion -40 induced strengthening and poleward contraction of zonal winds (Thompson and Solomon, 2002b;Arblaster et al., 2011), increased upwelling of warm, modified Circumpolar Deep Water (Jacobs et al., 2011), a warmer Southern Ocean (Gille, 2002;Böning et al., 2008;Abraham et al., 2013), meltwater-driven freshening of the Ross Sea (Jacobs et al., 2002), ice shelf and mass balance loss, grounding line retreat (Joughin et al., 2014;Rignot et al., 2014;Paolo et al., 2015;Pollard et al., 2015), reduced formation of Antarctic Bottom Water (Rintoul, 2007) and Antarctic Intermediate Water (Wong et al., 1999), changes in sea ice (wind driven, regional increase and decrease in the Amundsen and Ross Seas, respectively) (Holland and Kwok, 2012;Stammerjohn et al., 2012;Sinclair et al., 2014), and dynamic changes of Southern Ocean CO2 uptake driven by atmospheric circulation pattern (Landschützer et al., 2015). Yet, these observational time series are short (Gille, 2002;Böning et al., 2008;Toggweiler and 5 Russell, 2008) and inter-model variability indicates physical processes and their consequences are not well captured or understood (Sen Gupta et al., 2009;Braconnot et al., 2012). While the skill of equilibrium simulations steadily improves, the accuracy of transient model projections for the coming decades critically depends on an improved knowledge of climate variability, forcings, and dynamic feedbacks (Bakker et al., 2017;Stouffer et al., 2017).
Here we present data from a new, highly-resolved and accurately-dated ice-core record, spanning the past 2.7 ka, from the Ross 10 Sea region. The Roosevelt Island Climate Evolution (RICE) ice core is compared with existing records in the region to investigate the characteristics and drivers of spatial and temporal climate variability in the Ross Sea region.

Site characteristics and relevant climate drivers
In this section, a brief overview is provided of the climatological and glaciological characteristics of the study site. 15

Dynamic interaction between tropical and mid-latitudinal climate drivers and South Pacific climate variability
Environmental conditions in the Pacific Sector of the Southern Ocean and Antarctica are dominated by four major atmospheric circulation patterns: the Southern Annular Mode (SAM), the Pacific-South American pattern (PSA1 and PSA2) that are related to El Niño Southern Oscillation (ENSO) variability, and the Inter-decadal Pacific Oscillation (IPO). The SAM, the leading empirical orthogonal function (EOF) of the Southern Hemisphere extratropical geopotential height fields on monthly and 20 longer time scales, describes the strength and position of the Southern Hemisphere westerly winds via the relative pressure anomalies over Antarctica (~65°S) and the mid-latitudes (~45°S) (Thompson and Wallace, 2000;Thompson and Solomon, 2002a). The persistent positive, summer trend of the SAM (decreasing pressure over Antarctica) has been linked to stratospheric ozone depletion and increase in atmospheric greenhouse gas concentration Thompson et al., 2011). The positive SAM is associated with above average warming of the Antarctic Peninsula, and cooler conditions over 25 East Antarctica due to a reduced poleward gradient and thus diminished transport of heat and moisture along with a reduction in katabatic flow (Thompson and Solomon, 2002a;Marshall et al., 2013;Marshall and Thompson, 2016). While the positive summer SAM trend (also weakly expressed during autumn) along the Antarctic margin is generally associated with an equatorward heat flux, the western Ross Sea is one of two regions (the Weddell Sea being the other) to experience an anomalous poleward heat flux (Marshall and Thompson, 2016), that transports heat and moisture across the Ross Ice Shelf. The positive 30 SAM has been shown to contribute at least partially to an Antarctic SIE increase, while a negative SAM has been associated with a reduced SIE (Bintanja et al., 2013;Ferreira et al., 2015;Kohyama and Hartmann, 2015;Holland et al., 2016;Turner et al., 2017). The future behaviour of the SAM over the next decades is a topic of active research due to the competing, and seasonally biased influences of projected stratospheric ozone recovery and greenhouse gas emissions (Thompson et al., 2011;Gillett and Fyfe, 2013;Bracegirdle et al., 2014). 35 PSA patterns represent atmospheric Rossby wave trains initiated by anomalously deep tropical convection during ENSO events, in particular during austral spring, which originate in the western (PSA1) and the central (PSA2) tropical Pacific (Mo and Higgins, 1998). PSA1 and PSA2 are defined as the 2 nd and 3 rd EOF respectively of monthly-mean extratropical geopotential height fields, with the negative (positive) phase resembling El Niño (La Niña)-like conditions (Mo, 2000).
Changes in SAT over West Antarctica have been linked to PSA1 variability (Schneider and Steig, 2008;Schneider et al., 40 3 Clim. Past Discuss., https://doi.org /10.5194/cp-2017-95 Manuscript under review for journal Clim. Past Discussion started: 1 August 2017 c Author(s) 2017. CC BY 4.0 License. 2012), while the warming of West Antarctica's winter temperatures has been linked to PSA2 (Ding et al., 2011). The positive polarity of PSA1 is associated with anticyclonic wind anomalies in the South Pacific centered at ~120°W, which have been linked to increased onshore flow and increased eddy activity (Marshall and Thompson, 2016;Emanuelsson et al., in review).
In contrast, during the positive phase of the PSA2, the anticyclonic centre shifts to ~150°W in the Ross Sea, creating a dipole across the Ross Ice Shelf, with increased transport of marine air masses along the western Ross Ice Shelf and enhanced 5 katabatic flow along the eastern Ross Ice Shelf (Marshall and Thompson, 2016). Sea ice feedbacks to the SAM and ENSO forcing in the western Ross Sea (as well as the Bellingshausen Sea) were found to be particularly strong when a negative SAM coincided with El Niño events (increased poleward heat flux, less sea ice) or a positive SAM concurred with La Niña events (decreased poleward heat flux, more sea ice) (Stammerjohn et al., 2008). The authors found that this teleconnection is less pronounced in the eastern Ross Sea. 10 The Inter-decadal Pacific Oscillation (IPO), an ENSO-like climate variation on decadal time scales (Power et al., 1999), is closely related to the Pacific Decadal Oscillation (PDO) (Mantua and Hare, 2002). While the PDO is defined as the first EOF of sea surface temperature (SST) variability in the Northern Pacific, the IPO is defined by a tripole index of decadal scale SST anomalies across the Pacific (Henley et al., 2015). A warm tropical Pacific and weakened trade winds are associated with a positive IPO, while a cooler tropical Pacific and strengthened trade winds are characteristic for a negative IPO. The phasing 15 of the IPO and PDO have been shown to influence the strength of regional and global teleconnections with ENSO (Henley et al., 2015). An in-phase IPO amplification of ENSO events has been linked to a strengthening of global dry / wet anomalies, in contrast to periods when the IPO and ENSO are out of phase, causing these anomalies to weaken or disappear entirely (Wang et al., 2014). In addition, a negative IPO leads to cooler SSTs in the Ross, Amundsen and Bellingshausen Seas, while a positive IPO is associated with warmer SSTs (Henley et al., 2015). The centre of anticyclonic circulation linked to precipitation at 20 Roosevelt Island (Emanuelsson et al., in review) moves eastward during the negative IPO from ~120°W during the positive IPO to ~100°W (Henley et al., 2015). It has been suggested that the negative IPO, at least in part, is responsible for the hiatus of global surface warming during 1940-1975Kosaka and Xie, 2013;England et al., 2014).
The Amundsen Sea Low (ASL), a semi-permanent low pressure centre in the Ross / Amundsen Sea, is the most prominent and persistent of three low pressure centres around Antarctica, associated with the wave number 3 circulation (Raphael, 2004). 25 The ASL is sensitive to ENSO (especially during winter and spring) and SAM (in particular during autumn) and to influence through meridional wind anomalies environmental conditions in the Ross, Amundsen and Bellingshausen Seas and across West Antarctica and the Antarctic Peninsula (Bertler et al., 2004;Steig et al., 2012;Ding and Steig, 2013;Turner et al., 2013;Marshall and Thompson, 2016;Raphael et al., 2016). Seasonally, the ASL centre moves from ~110ºW in during austral summer to ~150ºW austral winter . A positive SAM and / or La Niña event leads to a 30 deepening of the ASL, while a negative SAM and/or El Niño event causes a weakening . The IPO, through its effect upon ENSO and SAM variability, also influences the ASL and sea ice extent in the Ross and Amundsen seas (Meehl et al., 2016). Blocking events in the Amundsen Sea (Renwick, 2005), are sensitive to the position of the ASL and are dominant drivers of marine air mass intrusions and associated precipitation and temperature anomalies at Roosevelt Island (Emanuelsson et al., in review). 35

RICE site characteristics
Roosevelt Island is an approximately 120 km-long by 60 km-wide grounded ice rise located near the north-eastern edge of the Ross Ice Shelf (Figure 1). Ice accumulates locally on the ice rise, while the floating Ross Ice Shelf flows around Roosevelt Island. The ice surrounding Roosevelt Island originates from the West Antarctic Ice Sheet (WAIS), via the Bindschadler, 40 MacAyeal and Echelmeyer ice streams. Bedmap2 data (Fretwell et al., 2013)  Island, the ice is 764 m thick and grounded 214 m below sea level. Radar surveys across the Roosevelt Island ice divide show a well-developed "Raymond Bump" (Raymond, 1984) arching of isochrones suggesting a stable ice divide (Conway et al., 1999). The vertical velocity, constrained by phase-sensitive radio echo sounder (pRES) measurements, is approximately 20 cm a -1 at the surface relative to the velocity of 0 cm a -1 at the bed (Kingslake et al., 2014). with some design modification (Mandeno et al., 2013). The upper 60 m of the borehole was cased with plastic pipe and the remainder of the drill hole filled with a mixture of Estisol-240 and Coasol to prevent closure. The part of the RICE ice core record used in this study covers the past 2.7 ka and consists of data combined from the RICE-2012/13-B firn core (0.56-12.30 m depth) and the RICE Deep ice core (12.30 m -344 m depth). An overview of the core quality and processing procedures are summarised by Pyne et al. (in review). In this manuscript, we present new water stable-isotope (deuterium, δD) and snow 15 accumulation records and compare them with existing records from West and East Antarctica (Table 1).

RICE age model: RICE17
The RICE17 age model for the past 2.7 ka is based on an annually-dated ice core chronology from 0-344 m which is described in detail by Winstrup et al. (in preparation). The cumulated age uncertainty for the past 100 years is ≤ ±2 years, for the past 20 1,000 years ≤ ±19 years and for the past 2,000 years ≤ ±38 years, reaching a maximum uncertainty of ±45 years at 344 m depth (2.7 ka). The RICE17 timescale is in good agreement with the WD2014 annual-layer counted timescale from the WAIS Divide ice core dating to 200 Common Era (CE, 280 m depth). For the deeper parts of the core, there is likely a small bias (2-3%) towards undercounting the annual layers, resulting in a small age offset compared to the WD2014 timescale (Winstrup et al., in preparation). 25

Snow accumulation reconstruction
Ice core annual layer thicknesses provide a record of past snow accumulation once the amount of vertical strain has been accounted for. At Roosevelt Island, repeat pRES measurements were performed across the divide, providing a direct measurement of the vertical velocity profile (Kingslake et al., 2014). This has a key advantage over most previous ice-core 30 inferences of accumulation rate because vertical strain thinning through the ice sheet is measured directly, rather than needing to use an approximation for ice-flow near ice divides (e.g. Dansgaard and Johnsen, 1969;Lliboutry, 1979). Uncertainty in the accumulation-rate reconstruction increases from zero at the surface (no strain thinning) to a maximum of ± 8 % at 170 m true depth. Below 170 m, the uncertainty remains constant at ± 8 %. A detailed description of the accumulation-rate reconstruction is provided by Winstrup et al. (in preparation). 35

Water stable-isotope data
The water stable-isotope record was measured using a continuous-flow laser spectroscopy system with an off-axis integrated cavity output spectroscopy (OA-ICOS) analyser, manufactured by Los Gatos Research (LGR). The water for these 5 Clim. Past Discuss., https://doi.org /10.5194/cp-2017-95 Manuscript under review for journal Clim. Past Discussion started: 1 August 2017 c Author(s) 2017. CC BY 4.0 License. measurements was derived from the inner section of the continuous flow analysis (CFA) melt head, while water from the outside section was collected for discrete samples. A detailed description and quality assessment of this system is provided by Emanuelsson et al. (2015). The combined uncertainty for deuterium (δD) at 2 cm resolution is ±0.85‰. A detailed description of the isotope calibration, the calculation of cumulative uncertainties, and the assignment of depth is provided by Keller et al. (in preparation). 5

4
RICE data correlation with reanalysis data -modern temperature, snow accumulation, and sea ice extent trends ERAi reanalysis, occurs at 13.42 m depth in the firn. For this reason, the period 1979-2012 is predominantly captured in the RICE 12/13 firn core. Data from the RICE AWS suggest that precipitation at RICE can be irregular, with large snow precipitation events dominating the accumulation pattern (Emanuelsson, 2016). Therefore, we limit the analysis in this study to annual averages and longer-term trends. 15

Isotope-temperature correlation
In Figure  However, the correlation between RICE δD data and ERAi SAT is limited (Figure 2b), retaining a positive correlation across 25 the Ross Ice Shelf and northern Ross Sea. The time series correlation between the ERAi SAT record and the RICE δD data ( Figure 2d) is r=0.45 (p<0.01). We test the robustness of this relationship by applying the minimum and maximum age solutions within the age uncertainty (≤ 2 year during this time period, Winstrup et al., in preparation ) to identify the age model solution within the age uncertainty that renders the highest correlation. This optimised solution RICE δDo is shown in Figure   2e with a correlation coefficient of r=0.75 (p<0.001). The correlation of the RICE δDo record with the ERAi SAT data ( Figure  30 2b) produces a pattern that closely resembles the correlation pattern using the ERAi data itself (Figure 2a), suggesting that the δD record provides useful information about the regional temperature history.
From the comparison between RICE δDo and ERAi SAT records, we obtain a temporal slope of 5.50 ‰ ºC -1 (Figure 2f), which falls within the lower limit of previously reported values from Antarctica of ~5.56 ± 0.51 ‰ ºC -1 to ~6.80 ± 0.57 ‰ ºC -1 (Schneider et al., 2005;Masson-Delmotte et al., 2008;Fegyveresi et al., 2011). We use this relationship to calculate 35 temperature variations for the RICE δD record. The average annual temperature calculated for 1979-2012 from ERAi for the RICE site is -27.4 ± 2.4 ºC and for the δDo data: -27.5 ± 3.1 ºC. Although the year to year SAT variability appears to be well captured in the RICE δDo record, there are discrepancies in observed trends. While RICE δDo data suggest an increase in SAT from 1996 onwards, ERAi SAT data do not show a trend ( Figure 2e). It remains a challenge to determine how well reanalysis products, including ERAi data, and other observations capture temperature trends in Antarctica (Pages 2k Consortium, 2013;40 6 Clim. Past Discuss., https://doi.org /10.5194/cp-2017-95 Manuscript under review for journal Clim. Past Discussion started: 1 August 2017 c Author(s) 2017. CC BY 4.0 License. Stenni et al., 2017) and thus whether the observed difference in trend between ice core δDo and ERAi SAT is significant or meaningful.
Furthermore, we test the correlation with the near surface Antarctic temperature reconstruction (NB2014, Nicolas and Bromwich, 2014) , which uses three reanalysis products and takes advantage of the revised Byrd Station temperature record (Bromwich et al., 2013) to provide an improved reanalysis product for Antarctica for the time period 1958-2012 CE. We find 5 no correlation between the NB2014 records and the ERAi data at the RICE site, nor the RICE δD data for the 1979-2012 time period, perhaps suggesting some regional challenges.

Regional snow accumulation variability
Temporal and spatial variability of snow accumulation are assessed using 144 snow stakes covering a 200 km 2 array. The 3 m 10 long, stainless steel poles were set and surveyed in November 2010, re-measured in January 2011, and revisited and extended in January 2012 and November 2013. The measurements indicate a strong accumulation gradient with up to 32 cm water equivalent per year (weq a-1) on the north-eastern flank decreasing to 9 cm weq a-1 on the south-western flank. Near the drill site, annual average snow accumulation rates range from 20-30 cm weq a -1 from 2010 to 2013. Investigation of snow precipitation events as captured by ERAi data shows that large snow precipitation events are associated with north-easterly 15 flow (Emanuelsson et al., in review).
Additionally, the RICE AWS included a Judd Ultrasonic Depth Sensor for snow accumulation measurements. The sensor was positioned 140-160 cm above the ground and reset during each season. The 3-year record shows gaps ( Figure 3) which represent times when rime precipitation on the sensor caused snow height to be erroneously recorded at the same height as the sensor. This process was particularly prevalent during winter. Over the three years, the site received an average annual snow 20 pack of ~75 cm. These data represent height measurements and are not corrected for snow density. The ERAi data have been shown to best capture snow precipitation variability in Antarctica, but to perhaps underestimate the total amount (Sinclair et al., 2010;Bromwich et al., 2011;Wang et al., 2016). The ERAi data are not directly comparable to the AWS data presented here, as the ERAi data are reported in cm weq and do not capture periods of snow removal through wind scouring. Yet, there is a good agreement between the two data sets with respect to the relative rate of precipitation, which suggests neither winters 25 nor summers have been times of significant snow removal. However, the ERAi data suggest an average annual snow accumulation of only 11 cm weq a -1 . The average annual snow accumulation rate derived from the RICE ice core (Winstrup et al., in preparation) is 21 ± 0.06 cm weq a -1 for the same time period . Assuming an average density of 0.37 g/cm 3 (average density in two snow pits, 0-75 cm), the AWS data suggest 20 cm weq a -1 snow accumulation, which is comparable to our ice core snow accumulation rate. We attribute this difference to the spatial differences between 30 measurements of the snow stake array, the interpolation field of the nearest ERAi data point, and the actual drill site location.
The regional representativeness of RICE snow accumulation data is assessed by correlating the ERAi precipitation time series, extracted for the RICE location, with the ERAi precipitation grid data (Figure 4a). The correlation suggests that precipitation variability at Roosevelt Island is representative of the observed variability across the Ross Ice Shelf, the southern Ross Sea, and western West Antarctica. We note from Figure 4a that the sites of the Siple Dome ice core (green circle) and the West 35 Antarctic Ice Sheet Divide ice core (WDC, red circle) are situated within the positive correlation field, while Talos Dome (TALDICE, purple circle) shows no correlation to ERAi precipitation at RICE. A negative correlation is found with the region of the Amundsen Sea Low (ASL) Raphael et al., 2016). In Figure 4b, the RICE snow accumulation record is correlated with ERAi precipitation data and shows similar pattern. The resemblance of the correlation patterns suggests that the variability of the RICE snow accumulation data (Figure 4b) reflects regional precipitation variability ( Figure 4a) and thus 40 likely can elucidate regional snow accumulation variability in the past, in particular for array reconstructions such as i.e. Thomas et al. (2017). We also test the correlation with the optimised age scale derived for the δDo record and find that for 7 Clim. Past Discuss., https://doi.org /10.5194/cp-2017-95 Manuscript under review for journal Clim. Past Discussion started: 1 August 2017 c Author(s) 2017. CC BY 4.0 License. snow accumulation data this adjusted age scale (Acco) reduces the correlation but remains statistically significant (Table 2).
We note that the sensitivity of the correlation to those minor adjustments is founded in the brevity of the common time period.
However, none of the overall pattern and relationships changes significantly between the two age scale solutions.

4.3
Influence of regional sea ice variability on RICE isotope and snow accumulation 5 Sea ice extent (SIE) variability has been shown to influence isotope values in precipitated snow, particularly in coastal locations (Noone and Simmonds, 2004;Küttel et al., 2012) through the increased contribution of enriched water vapour during times of reduced sea ice and increased sensible heat flux due to a higher degree of atmospheric stratification leading to more vigorous moisture transport. Tuohy et al. (2015) demonstrated that for the period 2006-2012 ~40-60% of precipitation arriving at Roosevelt Island came from local sources in the southern Ross Sea. In addition, Emanuelsson 10 et al. (in review) demonstrated the important role of blocking events, that are associated with over 88% of large precipitation events at RICE, on sea ice variability via meridional wind field anomalies.
Snow accumulation at RICE is negatively correlated with SIE in the Ross Sea region (Figure 5a), with years of increased (decreased) SIE leading to reduced (increased) accumulation at RICE, confirming the sensitivity of moisture-bearing marine air mass intrusions to local ocean moisture sources and hence regional SIE. The correlation between ERAi SIE and the 15 and less snow accumulation. The correlation between SIEJ and ERAi precipitation at RICE is r= -0.67, and for SIEJ and RICE snow accumulation r= -0.56 (Table 2). Moreover, the correlation between SIEJ and ERAi SAT and SIEJ and RICE δDo is also 25 statistically significant with r= -0.38 and r= -0.53, respectively. The higher correlation with RICE δDo perhaps suggests that the influence of SIE in the Ross Sea region affects the RICE δD record both through direct temperature changes in the region as well as fractionation processes that are independent of temperature, such as the lengthened distillation pathway to RICE during periods of more extensive SIE.
The ERAi SAT and ERAi precipitation data at RICE (Table 2) reveal a positive correlation over large areas of Antarctica with 30 higher correlation coefficients over the eastern Ross Sea and eastern Weddell Sea (spatial fields not shown). At the RICE site, the correlation reaches r=0.66 (p<0.001). Moreover, the correlation between RICE δD and RICE Acc [or RICE δDo and RICE Acco] data yield a statistically significant correlation of r=0. 40 (p<0.05) [or ro=0.45, po<0.01], respectively. This suggests that years with positive isotope anomalies are frequently characterised by higher snow accumulation rates. In contrast, precipitation during low snow accumulation years might be dominated by precipitation from air masses that have travelled further and 35 perhaps across West Antarctica (Emanuelsson et al., in review) leading to more depleted isotope values and lower snow accumulation rates than local air masses from the Ross Sea region. drivers in the South Pacific challenging. We use the SAMA index developed by Abram and colleagues (2014) to test the fidelity of the SAM relationship with the climatic conditions in the Ross Sea over the past millennium (Table 2). In addition, the Southern Oscillation Index (SOI, Trenberth and Stepaniak, 2001) and Niño 3.4 index (Rayner et al., 2003) for PSA1, the Niño 4 Index (Trenberth and Stepaniak, 2001) for PSA2, along with the IPO Index (Henley et al., 2015) are used to investigate the influence of SST variability in the eastern (PSA1) and central (PSA2) tropical Pacific on annual and decadal time scales (IPO). 5

Influence of climate drivers on prevailing conditions in the Ross Sea region
In addition, we take advantage of a 850-year reconstruction of the Niño 3.4 index  to investigate any long term influence of the eastern Pacific SST on environmental conditions in the Ross Sea.
The correlation of ERAi data and modern ice core records covering the 1979-2012 CE period with indices of relevant climate drivers (Table 2) suggests that SAMA has an enduring statistically significant relationship with temperature, snow accumulation and SIE in the Ross Sea, with the positive SAM being associated with cooler temperatures, lower snow 10 accumulation / precipitation and more extensive SIE. The correlations remain robust and at comparable levels using a detrended SAMA record. In contrast, ENSO (SOI, Niño 3.4 and 4) and ENSO-like variability (IPO) have only linear statistically significant relationships with ERAi precipitation (but not with RICE snow accumulation) and SIEJ. The dynamic relationship between the phasing of SAM, PSA 1 and 2, and IPO maybe masking aspects of the interactions (Fogt and Bromwich, 2006;Emanuelsson, 2016;Marshall and Thompson, 2016). We note that the influence of SAM and PSA2 lead to a climate dipole 15 within the Ross Ice Shelf / Ross Sea Region with opposing trends of meridional heat flux and near surface winds between the eastern and western Ross Ice Shelf / Ross Sea region. In contrast, the teleconnections with the PSA1 and IPO cause changes that affect uniformly the entire Ross Ice Shelf / Ross Sea Region but with opposing effects in West Antarctica (Marshall and Thompson, 2016).

Temperature Variability 25
We find that the RICE and Siple Dome (Brook et al., 2005; WAIS Divide Project Members, 2013) isotope records share a long-term warming trend in the Ross Sea Region. In contrast, WDC isotope  and borehole temperature data (Orsi et al., 2012) exhibit a long-term cooling trend for West Antarctica, while TALDICE recorded stable conditions for coastal East Antarctica in the western Ross Sea.
Elevation changes influence water isotope values (Vinther et al., 2009). Thinning of Roosevelt Island, inferred from the 30 amplitude of arched isochrones beneath the crest of the divide and the depth-age relationship from the ice core (Conway et al., 1999, H. Conway, personal communication) is less than 2 cm a -1 for the past 3.5 ka. That is, the surface elevation has decreased less than 50 m over the past 2.7 ka. Assuming that these elevation changes are sufficient to influence vertical movement of the precipitating air mass, such an elevation change could account for an isotopic enrichment of ~2 ‰ (Vinther et al., 2009), which is insufficient to explain the total observed increase of 8 ‰. Furthermore, the RICE δD trend is characterised by two step-35 changes at 579 CE ± 27 years and 1492 CE ± 10 years (grey dotted lines in Figure 6), when decadal isotope values increase by 3 ‰ and 5 ‰, respectively, which suggests that elevation changes were not a principal driver. Using the temporal slope of 5.5‰ per ºC, these abrupt temperature transitions represent an increase of the average decadal temperature (Figure 7a The Siple Dome ice core δ 18 O record exhibits a similar isotope history to the RICE δD record. Siple Dome isotope data reveal an abrupt warming at 605 CE, some 27 years later than in RICE, but within the cumulative age uncertainty of the two records. After 605 CE, Siple Dome temperatures remain stable, although recording somewhat warmer temperatures from about 1875 5 CE. Late-Holocene elevation changes (thinning) at Siple Dome have been reported to be negligible (Price et al., 2007) and are unlikely to have caused the observed abrupt warming at 605 CE. In contrast to records from the western Ross Sea (Stenni et al., 2002;Bertler et al., 2011;Rhodes et al., 2012) and West Antarctica (Orsi et al., 2012), RICE and Siple Dome do not show a warming or cooling associated with the Medieval Warm Period (MWP) or the Little Ice Age (LIA), respectively.
The WDC δ 18 O record suggests a long-term isotope cooling of West Antarctica, confirmed by borehole temperature 10 reconstructions (Orsi et al., 2012). This trend is consistent with warmer-than-average temperatures during the MWP and cooler conditions during the LIA, but may also reflect changes in elevation and decreasing insolation . The cooling trend is followed by an increase in temperature in recent decades (Steig et al., 2009;Orsi et al., 2012) consistent with an increase in marine air mass intrusions . We note that 579 CE marks the onset of a decline in WDC isotope and borehole temperatures, in line with the observed anti-phase relationship of WDC with RICE and Siple Dome. No notable 15 change is observed in WDC water stable-isotope temperature data in the late 15 th Century. In contrast, WDC borehole temperature suggests the onset of a warming trend within the last 100 years, marking the modern divergence between WDC isotope and borehole records. The TALDICE water stable-isotope temperatures does not exhibit a long term trend over the past 2.7 ka. Yet colder water stable-isotope temperature anomalies have been associated with the LIA period (Stenni et al., 2002), which coincide approximately with the intensified warming at RICE and cooler conditions at WDC. 20 The similarity between the RICE and Siple Dome records suggests that the eastern Ross Sea was dominated by regionallycoherent climate drivers over the past 2.7 ka, perhaps receiving precipitation via similar air-mass trajectories. Overall this comparison shows that temperature trends in the eastern Ross Sea (warming at RICE and Siple Dome) and West Antarctica (WDC cooling) were anti-phased for over 2 ka (660 BCE to ~1500 CE), while the western Ross Sea (TALDICE) remained stable. From the 17 th Century onwards, while WDC water stable isotope temperatures continue to cool, the WDC borehole 25 temperature record a warming, in phase with RICE and Siple Dome and concomitant with warmer temperatures at TALDICE.

Snow accumulation variability
Investigating long-term trends in snow accumulation records (Figure 6b), the decadally-smoothed RICE (Winstrup et al., in preparation) show a discernible positive trend from about 600 CE. TALDICE data  show a long term 30 increase in snow accumulation rates for the eastern and western Ross Sea, respectively, while WDC  displays a decreasing trend for central West Antarctica. The RICE snow accumulation data reach a maximum in the 13 th Century, with a trend towards lower values from 1686 CE onward. Based on the strong, negative correlation between RICE snow accumulation and SIE in recent decades, we interpret the long-term increase in snow accumulation to represent a long term reduction in SIE the Ross Sea, consistent with a long term increase in RICE isotope temperature. The modern decadal  (Thompson and Wallace, 2000;Thompson and Solomon, 2002a) was developed during the late 20 th 10 Century, at a time when SAM was characterised by a strong positive trend. The SAMA reconstruction showed that the modern SAM is now at its most positive state of the past millennium (Abram et al., 2014). As a consequence, the reconstructed SAMA index is mainly negative. To investigate the influence of positive and negative anomalies of the SAMA relative to its average state over the past 1 ka, we plot the SAMA reconstruction with an above (light purple) / below (purple) value of its long term average of -1.3 (instead of '0'). The traditional positive SAMA values (above 0, dark purple) are also shown for reference. 15 Assessing the relationship between RICE, Siple Dome, WDC and TALDICE, we identify three major time periods of change.

Long-term baseline 660 BCE to 1367 CE
We find that for over 2 ka -from 660 ± 44 years BCE to 1367 ± 12 CE years, the eastern Ross Sea (RICE and Siple Dome) shows an enduring antiphase relationship with West Antarctica (WDC), while coastal East Antarctica in the western Ross Sea 20 (TALDICE) remains neutral ( Figure 6). Moreover, with some minor exceptions, isotope and snow accumulation records at RICE, WDC, and TALDICE, respectively, are positively correlated.

Negative SAMA -1368 CE to 1683 CE
The SAMA reconstruction suggests, that over the past millennium, the SAM was at its most negative (Abram et al., 2014) from 1368 ± 12 years CE to 1683 ± 8 CE years. As noted by Abram et al. (2014), the SAMA and Niño 3.4 reconstructions (Emile-25 Geay et al., 2013) are anti-phased on multi-decadal to centennial time scales with the Niño 3.4 index recording some of the warmest SSTs over the past 850 years during this period of negative SAMA.
During the negative SAMA, RICE shows a distinct and sudden increase in isotope temperature, while TALDICE records its coldest conditions over the past 2.7ka, consistent with a negative SAM-forced dipole change in meridional heat flux and surface wind anomalies in the Ross Sea (Marshall and Thompson, 2016). Previously published shorter records from the western 30 Ross Sea from Victoria Lower Glacier in the McMurdo Dry Valleys (Bertler et al., 2011) and Mt Erebus Saddle (Rhodes et al., 2012) also suggest colder conditions in the western Ross Sea during this period, with more extensive sea ice and stronger katabatic flow. We observe that WDC and TALDICE show below average snow accumulation values, while RICE snow accumulation changes from a long-term positive trend to neutral. Such trends are consistent with the reported increased SIE in the western Ross Sea (colder SAT, lower snow accumulation, at TALDICE; and cooler conditions with more extensive sea ice 35 and stronger katabatic winds at Victoria Lower Glacier and Mt Erebus) and Bellingshausen Sea (less snow accumulation, colder SAT at WDC). In contrast, RICE records warmer temperatures along with less and more variable snow accumulation, displaying a distinct Ross Sea Dipole. We suggest that the distinct antiphase relationship is caused by the SAM forcing of equatorward (poleward) heat flux anomalies in the western (eastern) Ross Sea. Coinciding with the sudden increase in RICE 11 Clim. Past Discuss., https://doi.org/10.5194/cp-2017-95 Manuscript under review for journal Clim. Past Discussion started: 1 August 2017 c Author(s) 2017. CC BY 4.0 License. δD in 1492 CE is the decoupling of local temperature from snow accumulation, evident from the diversion of the RICE snow accumulation and δD trends. The reduction in snow accumulation might be linked to a negative SAM-induced weakening of the ASL, leading to the development of fewer blocking events. Alternatively, the abrupt change to warmer temperatures at RICE might also point towards the development of the Roosevelt Island polynya. In recent decades, a Roosevelt Island polynya is observed and merges at times with the much larger Ross Sea polynya (Morales Maqueda et al., 2004). In contrast to the 5 Ross Sea Polynya (Sinclair et al., 2010), a local polynya could provide a potent source for isotopically enriched vapour to precipitation at RICE, perhaps exaggerating the actual warming of the area as interpreted from water stable isotope data. We expect the influence of a Roosevelt Island polynya to have a reduced effect on the more distant Siple Dome, which is consistent with our observations. 10

Onset of the positive SAM -1684 CE to 2012 CE
At 1684 CE ± 7 years, the SAMA increases and remains above its long-term average until modern times while the Niño 3.4 index suggests a change to the prevalence of strong La Niña-like conditions. RICE δD suggest the continuation of warm conditions, while snow accumulation drops below the long-term average, with the RICE snow accumulation trend now inphase with WDC ( Figure 6). In contrast, TALDICE records above average snow accumulation rates, which are out of phase 15 with RICE and WDC. Such an alignment is consistent with a positive SAM-forced dipole in meridional heat flux and surface winds (Marshall and Thompson, 2016) between the western Ross Sea (TALDICE) contrasting the eastern Ross Sea (RICE, Siple Dome) and western West Antarctica (WDC). The change to above-average SAMA (or even positive SAM) values does not coincide with notable changes in any of the isotope records of RICE, Siple Dome or WDC, but concurs with a modest warming of the TALDICE isotope record. However, it marks the onset of the diversion of WDC water isotope and borehole 20 temperature reconstructions. Coincident positive SAM (purple SAMA values in Figure 6c) and La Niña events have been linked in recent decades to increases in SIE in the western Ross Sea and decreases in the Bellingshausen Sea (Stammerjohn et al., 2008). This is consistent with the notable reduction in snow accumulation at RICE and the trend towards warmer conditions and increased marine air mass intrusions at WDC but is inconsistent with the reduction in snow precipitation at WDC and the trend to warmer conditions at RICE. We interpret the continuation of warm RICE isotope temperatures to reflect the persistence 25 of the Roosevelt Island polynya.

Dipole pattern on decadal to centennial time scales
To investigate the drivers of decadal to centennial variability, we compare the linearly detrended time series of RICE water stable isotope records with those from (i) Siple Dome and WDC (West Antarctica, Figure 8a) and (ii) TALDICE (East Antarctica, western Ross Sea, Figure 8b). The analysis suggests that until ~200 CE, RICE and Siple Dome (eastern Ross Sea) 30 are in phase with TALDICE variability and thus the East Antarctic climate of the western Ross Sea. From about 400 -1900 CE, RICE and Siple Dome variability are in-phase with West Antarctic climate variability (WDC). During the negative phase of the SAMA (Figure 8), the anti-correlation between RICE and TALDICE is particularly strong. Comparable periods of a strong anti-phase relationship occur during 800 -1200 CE and 300 -600 CE (grey boxes), perhaps indicative of earlier periods of a strong SAM forcing, albeit a strongly positive SAM with warmer conditions at TALDICE and cooler conditions at RICE. 35 The most recent decades experienced a strong positive SAM influence on the region, which also shows a distinct anticorrelation between RICE and TALDICE and is consistent with this hypothesis.
To assess the correlation of cyclicities apparent in the RICE, Siple Dome, WDC and TALDICE isotope records, wavelet coherence spectrum analyses were conducted (Figure 9)  70 year periodicity from 0-1000 CE. The high coherency suggests that RICE and Siple Dome respond to similar forcings. The coherence analysis between RICE and WDC shows an enduring in-phase correlations from ~1000 CE to today for the bandwidth of 200 -700 years. An anti-phase coherence is found from 0 -500 CE. The coherence analysis between RICE and TALDICE identifies strong relationships predominantly for the early part of the records, from 660 BCE to ~500 CE and a weak coherence from about 1100-1400 CE, when RICE leads by ~75-100 years. The analysis suggests that for the past 2.7 ka 5 the eastern Ross Sea (RICE, Siple Dome) and western West Antarctica (WDC) are climatologically closely linked in their response to forcings on decadal to centennial time scales. The relationship between the western (TALDICE) and eastern (RICE, Siple Dome) Ross Sea experienced a marked change, with a strongly coupled relationship until ~500 CE, which also coincides with opposing climate variability that we interpret to represent a strengthened response to positive and negative SAM forcing of the region. 10 7

Concluding Remarks
The recent change to a strongly negative SAM  in November 2016 coincided with a significant reduction of Antarctic SIE, including the Ross Sea, during the 2016/17 summer (Turner et al., 2017). Longer observations are necessary to assess whether this recent trend continues and indeed forces the reduced SIE, but it fuels questions on the potential 15 acceleration of future environmental change in the Antarctic / Southern Ocean region. To improve projections for the coming decades, an improved understanding of the interplay of teleconnections and local feedbacks is needed.
The Ross Sea region is a climatologically sensitive region that is exposed to tropical and mid-latitude climate drivers. In recent decades, SAM and PSA2 teleconnections lead to opposing effects in the eastern and western Ross Sea region with respect to meridional heat flux, surface wind fields (Marshall and Thompson, 2016), and sea ice extent (Stammerjohn et al., 2008) 20 exhibiting a Ross Sea Dipole. The ASL deepens during combined positive SAM and La Niña events, and weakens during negative SAM and El Niño events. Such interactions have far reaching implications on the regional atmospheric and ocean circulations and sea ice (Turner et al., 2015, Raphael et al., 2016. Additionally, a negative (positive) IPO leads to cooler (warmer) SSTs in the Ross, Amundsen and Bellingshausen seas and has the potential to strengthen (in phase) or weaken (out of phase) the ENSO teleconnection (Henley et al., 2015). Furthermore, the phasing and strength of ENSO events and SAM 25 have been shown to be dependent (Fogt and Bromwich, 2006).
Our data suggest that these dynamically linked climate patterns led to significant and abrupt changes in the past with implications for regional interpretations of trends, including temperature, mass balance and SIE. For over 2 ka, from 660 BCE to the late 14 th Century, climate trends in the eastern Ross Sea (RICE and Siple Dome, trend to warmer temperatures and higher precipitation) are anti-correlated with conditions in the western West Antarctica (WDC, cooling with reduced precipitation), 30 while coastal East Antarctica in the western Ross Sea appeared decoupled (TALDICE, no trend in temperature or precipitation). This regional pattern re-organised during a period with strong negative SAM conditions (SAMA) accompanied by weak La Niña-like conditions (Niño 3.4 index) when western West Antarctica (WDC borehole and isotope temperature) and western Ross Sea (TALDICE) show cold temperatures during the Little Ice Age, while the eastern Ross Sea (RICE, Siple Dome) show warmer or stable temperatures, respectively. In the late 17 th Century, the SAMA changes to above average values, 35 concurrent with a change to strong La Niña-like conditions. These changes establish a strong Ross Sea Dipole that emerges from long-term trend, with western Ross Sea (TALDICE) experiencing warmer temperatures and increased precipitation, while the eastern Ross Sea (RICE, Siple Dome) exhibit reduced precipitation and warmer temperatures. We interpret this pattern to reflect an increase in SIE in the western Ross Sea with perhaps the establishment of the modern Roosevelt Island polynya as a local moisture source for RICE. At the same time, western West Antarctica is showing a warming (WDC borehole 40 temperature) along with an increase in marine air mass intrusions (WDC isotopes) and a reduction in snow accumulation 13 Clim. Past Discuss., https://doi.org/10.5194/cp-2017-95 Manuscript under review for journal Clim. Past Discussion started: 1 August 2017 c Author(s) 2017. CC BY 4.0 License.
(WDC). However, when longer-term trends are removed from the correlations, we find earlier periods of a strong Ross Sea dipole which we interpret to reflect previous time periods dominated by strongly positive SAM conditions from the 3 rd -6 th Century and 9-12 th Century. Our observations are consistent with the reconstruction of the strongly negative SAM from the 14-18 th Century and the positive SAM from the 19 th Century to today. The continued improvements of array reconstructions Thomas et al., 2017) are an exciting development to further our knowledge of the drivers and effects of 5 past change and their implications for future projections.

Acknowledgments
Funding for this project was provided by the New Zealand Ministry of Business, Innovation, and Employment Grants through