Little Ice Age climate and oceanic conditions of the Ross Sea, Antarctica from a coastal ice core record

Increasing paleoclimatic evidence suggests that the Little Ice Age (LIA) was a global climate change event. Understanding the forcings and associated climate system feedbacks of the LIA is made difficult by the scarcity of Southern Hemisphere paleoclimate records. We use a new glaciochemical record of a coastal ice core from Mt. Erebus Saddle, Antarctica, to reconstruct atmospheric and oceanic conditions in the Ross Sea sector of Antarctica over the past five centuries. The LIA is identified in stable isotope ( D) and lithophile element records, which respectively demon- strate that the region experienced 1.6 ± 1.4 C cooler aver- age temperatures prior to 1850 AD than during the last 150 yr and strong (> 57 m s 1 ) prevailing katabatic winds between 1500 and 1800 AD. Al and Ti concentration increases of an order of magnitude (> 120 ppb Al) are linked to enhanced aeolian transport of complex silicate minerals and represent the strongest katabatic wind events of the LIA. These events are associated with three 12-30 yr intervals of cooler tem- peratures at ca. 1690 AD, 1770 AD and 1840 AD. Further- more, ice core concentrations of the biogenic sulphur species MS suggest that biological productivity in the Ross Sea polynya was 80 % higher prior to 1875 AD than at any sub- sequent time. We propose that cooler Antarctic temperatures promoted stronger katabatic winds across the Ross Ice Shelf, resulting in an enlarged Ross Sea polynya during the LIA.


Introduction
The Little Ice Age (LIA) is typically considered to be a Northern Hemisphere climate phenomenon characterised by alpine glacial advances and relatively cool temperatures observed between 15th to mid-19th centuries (Grove, 1988). However, the timing, magnitude and character of the LIA climate response all exhibit strong regional variations (Jones and Mann, 2004). The cause of the LIA has been considered widely, with explanations ranging from increased volcanism, reduced solar irradiance and ocean circulation changes (e.g. Ammann et al., 2007;Broecker, 2000;Crowley, 2000;Miller et al., 2012). Furthermore, paleoclimate records suggest that the LIA is the most recent cooling event of a series that punctuated the Holocene (Domack and Mayewski, 1999;Masson et al., 2000). This apparent oscillatory behaviour in Holocene climate has led to speculation about what role the thermohaline circulation of the world's oceans may have played in instigating or amplifying these climate changes (Broecker, 2000;Denton and Broecker, 2008). Currently, no consensus has been reached on this issue and a key concern is the spatial bias toward Northern Hemisphere proxy and meteorological records (Jones et al., 2009). An important tool to understand the LIA must therefore be the acquisition and interpretation of Southern Hemisphere paleoclimate records, but few are currently available (Jones and Mann, 2004). Antarctic ice cores are valuable archives of past climate, and several ice cores have recently been obtained with sufficient temporal resolution to record the LIA. Whilst significant spatial and temporal variability exists (Masson et al., 2000;Morgan and van Ommen, 1997;Mosley-Thompson and Thompson, 1990;Stenni et al., 2002), records suggest that Antarctica experienced cooling synchronous with the Northern Hemisphere during the LIA  an observation that is not consistent with a scenario involving oceanic thermohaline circulation reorganisation as the driver of climatic change (cf. Broecker, 2000;Denton and Broecker, 2008).
In this study, we examine the climate of the LIA in a key region of Antarctica using sub-annual resolution glaciochemical records from a coastal ice core from Mt. Erebus Saddle (MES), close to the edge of the Ross Ice Shelf in the southwestern Ross Sea sector (Fig. 1). The Ross Sea polynya, which is located along the edge of the Ross Ice Shelf, is a major site of Southern Ocean sea ice generation (Tamura et al., 2008), which drives bottom water formation (Jacobs and Giulivi, 1998). Additionally, the region is the most biologically productive area of the Southern Ocean (Arrigo and van Dijken, 2003). Consequently, climatic changes occurring here have the potential to produce feedbacks that influence both regional conditions and the global climate system (Mayewski et al., 2009). The MES ice core site is at a boundary between the influence of two contrasting meteorological regimes. Air masses originate from both the Southern Ocean, in the form of cyclones (Sinclair et al., 2010), and the interior of Antarctica, in the form of katabatic flow (Sinclair et al., 2010;Bromwich et al., 1992). The two dominant air masses have contrasting geochemical characteristics that allow them to be identified (Bertler et al., 2004b). Thus, the location of the MES ice core site allows the meteorological regimes of the interior of Antarctica and the Ross Sea region to be traced back through time in the same ice core record.
A continuous record of ice core lithophile element concentrations from this site spanning five centuries provides a proxy for katabatic wind speeds, which were stronger prior to 1850 AD than at any ensuing time. Furthermore, by linking the katabatic wind proxy with methylsulphonate (MS − ), a tracer of marine primary productivity, we demonstrate that stronger katabatic winds caused a larger or more persistent Ross Sea polynya between 1600 and 1875 AD.

Study site
MES is a local snow accumulation zone situated between Mt. Erebus and Mt. Terra Nova on Ross Island at an altitude of 1600 m ( Fig. 1) (77 • 30.90 S, 167 • 40.59 E). Ice flows from the saddle towards NNW and SSE directions. Ground penetrating radar reconnaissance indicated a depth to bedrock of 220 m and imaged parallel isochrones, suggesting favourable glaciological conditions for the preservation of a continuous, annually resolved ice core record. A 168m-deep ice core was drilled in the central zone of minimum ice flow in December 2006. Annual temperature at the site, ascertained from firn temperature logging, is −25.6 • C. The depth-density profile generated from density measurements conducted in the field is well explained by the densification model of Herron and Langway (1980) using a critical density of 550 kg m −3 and an initial density of 370 kg m −3 . The accumulation rate, calculated from the annual layer count, is 0.23 m water equivalent (w.e.) yr −1 .

Core processing
Here we present data from the uppermost 120 m of the MES ice core. The core was processed at the New Zealand Ice Core Laboratory, GNS Science, using a continuous melter system which sampled discrete aliquots from the pristine interior of the core for chemical analysis (Osterberg et al., 2006). Stable isotope ratios (δ 18 O and δD) were determined on samples from corresponding depths obtained from the outer section of core. The mean sampling resolution was 43 samples m −1 between 0 and 66 m depth and 61 samples m −1 between 66 m and 120 m depth. Inductively coupled plasma mass spectrometry (ICP-MS) sample resolution was reduced to 30.5 samples m −1 between 77 m and 120 m to improve the efficiency of processing.
All vials used to sample material intended for ion chromatography (IC) or ICP-MS analysis were rigorously cleaned to prevent chemical contamination. The polypropylene vials used for IC samples were triple-rinsed with ultrapure water (> 18.2 M ), soaked for three days and triplerinsed again. The polypropylene vials used for ICP-MS samples were first triple-rinsed with ultra-pure water then soaked in 5 wt. % HNO 3 for a minimum of 10 days, triple-rinsed, soaked in ultra-pure water for a minimum of 3 days and triple-rinsed with ultra-pure water.
Procedural blanks were used to monitor the potential contribution of the processing and sampling procedures to the measured concentrations of chemical species measured. The chemical concentrations recorded in the procedural blanks (Table S1 in the Supplement) are negligible compared to the variability in concentration exhibited in the ice core record for all chemical species except Zr and Ca (Supplement).

Stable isotope ratios
The stable isotope ratios δ 18 O and δD (where were measured at the Stable Isotope Laboratory, GNS Science. Samples from the upper 91 m of the ice core were analysed with a GVI Isoprime mass spectrometer by methods reported previously . The analytical precision on this instrument was ± 0.1 ‰ for δ 18 O and ± 1.0 ‰ for δD, resulting in an uncertainty of ± 1.3 ‰ on the second order parameter deuterium excess (d-excess = δD − 8 × δ 18 O), Dansgaard, 1964).
Samples from 92 m to 120 m depth were analysed using a Los Gatos Research Liquid Water Isotope Analyser, which simultaneously measures δ 18 O and δD by laser absorption spectroscopy. δ 18 O and δD were calibrated using internal standards: MM1 with reported values of −29.4 ‰ for δ 18 O and −231.51 ‰ for δD, and INS11 with reported values of −0.4 ‰ for δ 18 O and −4.6 ‰ for δD, and reported with respect to V-SMOW. Isotopic standard INS9 (−17.4 ‰ for δ 18 O and, −131.0 ‰ for δD) was analysed to check the accuracy of the calibration. Each sample is analysed six times in succession, enabling memory effects from previous samples to be identified and removed. The analytical precision on this instrument was ± 0.2 ‰ for δ 18 O and ± 0.6 ‰ for δD, resulting in an uncertainty of ± 1.7 ‰ on d-excess.
Comparison of stable isotope ratios of a batch of samples measured by mass spectrometry and by laser spectroscopy identified a systematic offset between the values determined by the two techniques. The stable isotope ratios of MES ice core samples measured by laser spectroscopy were therefore corrected for this offset. Details of this correction are provided in the Supplement. A revised analytical uncertainty for d-excess, which takes into account the errors on both instruments, is calculated as ± 2.1 ‰.

Ion chromatographic analysis of major ions
Ion chromatographic analysis of major ions (Na + , Mg 2+ , Ca 2+ , MS − , Cl − , NO − 3 , SO 2− 4 ) was conducted at the Climate Change Institute, University of Maine. Anions were measured using a Dionex AS-18 column, 400 µl sample loop with a KOH eluent concentration of 35 mM. Cations were measured using a Dionex CS-12A column and a 500 µl loop with 25 mM methanesulphonic acid eluent. Calibration using five standards bracketing the anticipated concentration range routinely achieved correlation coefficients of R 2 > 0.99 for all analytes. An external reference material ION-92 (Environment Canada) was analysed at dilutions of 100 and 1000 times to verify the calibration. Na + , Mg 2+ , Ca 2+ , MS − , Cl − , and SO 2− 4 measurements are accurate to within 5 %. The reproducibilities (2 × relative standard deviation, RSD) of Na + , Mg 2+ , Ca 2+ and MS − measurements are < 11 % and < 4 % for Cl − , and SO 2− 4 . NO − 3 , is not detectable in ION-92.

Inductively coupled plasma mass spectrometric analysis of trace elements
Ice core samples were analysed for concentrations of major, minor and trace elements (Na, Mg, Al, Ca, Ti, V, Mn, As, Rb, Sr, Y, Zr, Cs, Ba, La, Ce, Pr, Tl, Pb, Bi, Th, U) by ICP-MS (Agilent 7500cs Series) in the Geochemistry Laboratory of the School of Geography, Environment and Earth Sciences, Victoria University of Wellington. The operating conditions adopted for ICP-MS analysis are provided in Table S2 in the Supplement. Calibration standards were gravimetrically diluted daily from stock mixed element standards made up from mono-elemental 1000 ppm standards. Element concentrations in these standards varied according to the relative concentrations expected in the ice core samples. Additional tuning was required to calibrate measurements conducted by the two different detector types (pulse and analog) on elements which recorded > 1 000 000 counts s −1 (Na, Mg and Al). Samples were stored frozen prior to analysis, then thawed, and acidified to 1 wt. % HNO 3 with 68 wt. % HNO 3 (Seastar) at least 12 h prior to analysis. Counts recorded on 1 wt. % HNO 3 blanks were subtracted from those recorded on samples and standards to correct for background contributions. Detection limits are calculated as 3σ on the blank measurements (Table S1 in the Supplement). The calibration standards were run as bracketing standards every 12 samples to correct for instrumental drift. An external reference material, Canada), was run at 10 times dilution during each analytical session to test the accuracy and reproducibility of the measurements (Table 1).

Tritium analysis
Six ice core samples from a depth range of 11 m to 21.3 m were measured directly without prior tritium enrichment by ultra-low-level liquid scintillation spectrometry (Quantulus TM ) in a monitored tritium-free environment. The water samples were purified in a vacuum distillation system to exclude tritium contamination. Details of the tritium analysis procedure are described in Morgenstern and Taylor (2009).

Annual layer counting
A snow pit sampled at MES two years prior to drilling of the ice core was dated by identification of annual peaks in δ 18 O for the time period 2000 to 2005 AD (Rhodes et al., 2009). The equivalent section of the δ 18 O record was located in the ice core to facilitate dating of the uppermost section. Maximum values in the annual signals of δ 18 O and δD in the top 25 m of the ice core were identified as mid-summer (1 January) (Fig. 4). The annual cycles in δ 18 O and δD often appear asymmetric in form (i.e. not perfectly sinusoidal) as a result of a seasonal bias in accumulation events (Cuffey and  and/or winter snow ablation at the MES site. Harsh weather conditions at this site prevented deployment of a weather station to monitor precipitation and ablation events. NO − 3 displayed a well-defined annual signal with maximum values in winter-spring, possibly as the result of production in stratospheric clouds (Michalski et al., 2005). Molecular diffusion in the firn pore space resulted in significant smoothing of the annual δ 18 O and δD cycles deeper down the core (Johnsen, 1977;. The δ 18 O and δD time series were therefore deconvolved/back-diffused to obtain the original high frequency signal (Johnsen, 1977;Johnsen et al., 2000) (Fig. S4 in the Supplement). The back-diffused time series were used to count annual layers in the ice core to 61.1 m depth (1816 AD).

Age tie points
In order to verify the interpretation of δD peaks as annual layers in the MES core, six samples were analysed for tritium ( 3 H) content. Elevated tritium contents related to thermonuclear bomb testing were determined in the MES ice core between 1964 and 1972 AD, in line with tritium contents measured in snow pits on the Ross Ice Shelf and at Halley Bay, Antarctic Peninsula, and in New Zealand precipitation ( Fig. 2). MES tritium contents are lower than those of the other Antarctic locations plotted but are similar to those measured on Law Dome material (unpublished data, not shown). This may be the result of the strong marine cyclonic influence at both MES (Sinclair et al., 2010) and Law Dome (Mc-Morrow et al., 2002), which dilutes the stratospheric tritium signal. These tritium measurements confirmed the validity of our approach to annual layer counting in the absence of accumulation data. The depth-age model was not adjusted to fit to this age tie point.
Positive identification of volcanic nss (non-sea salt) SO 2− 4 deposition in the MES ice core for dating purposes (e.g. Traufetter et al., 2004) is challenging, because the nssSO 2− 4 signal is dominated by biogenic nssSO 2− 4 from the seasonally open ocean. As an alternative, we examined the deposition history of heavy, volatile trace elements, which are emitted by volcanoes (Hinkley et al., 1994) and may be deposited on ice sheets (Kaspari et al., 2007;Kellerhals et al., 2010). In the MES ice core, Pb, Bi and Tl are all strongly enriched relative to terrestrial element concentrations in at least 8 consecutive samples between 60.5 m and 61.1 m depth (Fig. 3a). The initial annual layer count dated this signal to the year 1824 AD, within 8 yr of the Mt. Tambora (Indonesia) eruption signal across Antarctica (Traufetter et al., 2004). However, Vallelonga et al. (2003) attributed elevated Pb and Bi concentrations in a Law Dome ice core (Fig. 1a) at this time to emissions from Mt. Erebus, on the basis of Pb isotopic signatures. As MES is located adjacent to Mt. Erebus, it is plausible that the trace element record of the MES ice core may record the eruption history of Mt. Erebus rather than that of global-scale stratospheric eruptions. In order to test this hypothesis, trace element ratios measured across the Pb, Tl and Bi peaks were compared to trace element ratios measured in the plume of Mt. Erebus (Fig. 3b). None of the trace element ratios exhibit any marked deviation from the median value of the ice core record (green line) during the proposed 41 Zealand, decay-corrected to December 2006. Error bars are 1σ. The detection limit for 3 analysis without prior enrichment is 2.5 TU. One tritium unit (TU) is defined as one atom of 4 tritium per 10 18 atoms of hydrogen, equivalent to 0.118 becquerels (Bq) per litre of water. Mt. Tambora eruption. This lends support to attribution of the Bi, Pb and Tl peaks at 61.1 m to the Mt. Tambora eruption. Moreover, the median values (Fig. 3b, green line) of the trace element ratios across the MES ice core record differ from those of the Mt. Erebus plume (Fig. 3b, purple line), providing evidence that the ice core trace element budget is predominantly influenced by input from other sources.

Ice flow model and dating uncertainty
The 1-D flow model of Dansgaard and Johnsen (1969), tuned to the Mt. Tambora age tie point (1816 AD, 61.1 m), was used to predict the depth-age relationship below this depth ( Fig. S5 in the Supplement) and gives an age of 1473 AD at 120 m depth. The outcome of this model is sensitive to the prescribed accumulation rate. We employ an accumulation rate of 0.23 m w.e. yr −1 , which is the mean 1950 to 2006 AD value calculated from the annual layer count. This value is therefore subject to a potential error in the annual layer count. For this reason, the ice flow model was also run with accumulation rates of 0.21 and 0.25 m w.e. yr −1 , which are equivalent to ± 5 yr errors in the annual layer count back to 1950 AD. These runs produced age differences at 120 m of −84 and +178 yr respectively, as compared to the age predicted using a 0.23 m w.e. yr −1 accumulation rate (Fig. S5 in the Supplement). However, we have confidence in the mean accumulation rate of 0.23 m w.e. yr −1 and its stability over time for two reasons: (1) when the model is tuned to the Tambora age tie, the location of the bomb tritium peak causes the bomb tritium peak location to be off by > 2 yr; (2) if the model is run with the alternative accumulation rates (0.21 and 0.25 m w.e. yr −1 ) but same flow-parameter tuning as used to fit the Tambora eruption with the 0.23 m w.e. yr −1 accumulation rate, the age offset for the Tambora eruption depth-age marker is > 17 yr (Fig. S5 in the Supplement), which is significantly larger than the estimated uncertainty on our annual layer count.
As the initial layer count did not need to be adjusted to match the tritium marker age tie point, dating is accurate to within ± 1 yr between 2006 and 1964 AD (19 m depth). The age offset between the the initial annual layer count and the date of deposition from Mt. Tambora was 8 yr and provides an age uncertainty estimate at 61.1 m depth. For comparison, annual layer counting was continued until 90 m depth, after which it became challenging to identify annual layers. The age difference between the annual layer count and the ice flow model (0.23 m w.e. yr −1 ) at 90 m was 9 yr. Although these dating methods are not entirely independent, this difference provides an estimate of the cumulative dating error at 90 m depth. However, the outcome of the ice flow model is heavily dependent on both the prescribed accumulation rate and the assumption that it is constant in time. Therefore, according to the tests described above, the absolute age uncertainty could be as high as ± 50 yr at 100 m depth and ± 130 yr at 120 m depth.

Stable isotopes
δD and δ 18 O records of the MES ice core display welldeveloped seasonal cycles (Fig. 5, δD shown). δD varies between a mean summer (1 January, fixed by dating) maximum of −198 ‰ and a mean winter (1 July) minimum of −224 ‰ in the 1970-2006 AD time period (Fig. 5). The δD and d-excess time series show different interannual variability (Fig. 4), and regression of their annual mean or maximum values produces no significant correlation.  duration occur prior to 1850 AD (Fig. 6). The most notable of these low δD periods are identified on Fig. 6 by the letters A to E and dated as 1492-1518 AD, 1613-1637 AD, 1672-1700 AD, 1757-1778 AD and 1831-1843 AD, but we note that these absolute ages are subject to error (Section 3.3). The mean δD of the pre-1850 AD period is 6.2 ‰ lower than that of the post-1850 AD record ( Table 2). The pre-and post-1850 AD periods also exhibit contrasting trends in dexcess. The 5-yr running mean of the d-excess record oscillates about the 1950-2006 AD mean value from 1875 to 2006 AD, whereas prior to 1875 AD the 5-yr mean remains almost consistently above the 1950-2006 AD mean, the only exception occurring at ca. 1795 AD. This relatively elevated state of d-excess is maintained back through the record until at least 1658 AD. The d-excess mean is 1.5 ‰ higher between 1658 to 1875 AD compared to the following time period, although this value is close to the analytical uncertainty for this section of the record (Sect. 2.3.1). Unfortunately, at 92 m depth in the ice core (1658 AD), the technique used for isotopic analysis was changed from mass spectrometry to laser spectroscopy and inter-calibration differences between the two instruments make comparison of absolute dexcess values difficult. The d-excess values measured by the LGR laser instrument have been adjusted for offsets between the instruments (Supplement) before plotting in Fig. 6. The LGR data suggest that the relatively elevated d-excess values, which characterised the 1658-1875 AD time period, do not persist further back in the record. However, the apparent switch in d-excess behaviour at 1658 AD is coincident with the change in measurement technique and, therefore, may be be an analytical artefact.

Major ion and trace element chemistry
The ICP-MS dataset and its component variables (elements) have negatively skewed distributions with small numbers of outlying points of high concentration (Fig. 7). These points are outliers in a statistical sense, but they show no evidence of contamination or analytical artefacts. Here we use the minimum covariance determinant (MCD) estimator to produce robust correlation matrices not influenced by outliers (Fig. 8a). Both the pre-1850 and post-1850 AD sections of the ice core record show clear delineation of marine and terrestrial (hereafter lithophile) elements. Na, Mg and Sr are strongly correlated (R 2 > 0.95), and concentration ratios of these elements correspond to the seawater concentration ratios, indicating a marine source. Concentrations of lithophile elements Al, Ti, Mn, Rb, Ba, Y, La and Ce all inter-correlate at R 2 > 0.6, whilst V exhibits some weak correlation with both marine and lithophile elements. As, Cs, Pr, Tl, Bi, Th and U are not considered because levels were frequently below the ICP-MS detection limit. Marine and lithophile elements show different temporal trends in concentration variability between 1970 and 2006 AD (Fig. 4). The mean annual cycles of element concentration suggest that the timings of marine and lithophile element deposition are anti-phased. Annual maximum Na concentrations (92 ppb mean) generally occur in summer, coincident with the annual peaks in δD (Fig. 5). This is the opposite phasing of annual Na peaks to that seen in ice cores from the Antarctic interior (e.g. Legrand and Mayewski, 1997), which record Na maxima in winter months. In contrast, the lithophile elements exhibit concentration maxima in the autumn and winter months (Fig. 5, Ce, Y, Al, Mn, Ti shown). The Al mean seasonal cycle of 1970-2006 AD shows a 45 % higher concentration in winter relative to summer. Despite the existence of significant interannual variability in the amplitude of the seasonal cycle (Fig. 4), the phasing is consistent across the 1970-2006 AD period and is evident for all lithophile elements. LGR laser measurements of δD and d-excess have been adjusted (Supplement). The d-excess analytical error applicable to the entire dataset is indicated. Al plot is clipped at 80 ppb and Mn plot is clipped at 600 ppt. Timing of a diatom bloom reported by Leventer and Dunbar (1988) in four sediment core records from the north-west of Ross Island in McMurdo Sound is indicated. The gap in ICP-MS data is due to the removal of anomalously high chemistry concentrations relating to a tephra layer in the ice core.
Marine and lithophile element concentration records show markedly different variability over the last five centuries in the MES record. Over the entire record, marine element concentrations remained relatively constant with no prominent amplitude changes (Fig. 6, Na shown). In contrast, lithophile element records clearly exhibit higher magnitude concentrations prior to 1850 AD compared to post 1850 AD ( Fig. 6 and Table 2, Al, Ce and Mn > 60 % higher prior to 1850 AD). To test whether or not this increase in lithophile element concentrations is in fact reflected in the entire dataset, not just in several outlying data points, we plot notched boxplots of successive 50-yr intervals of the MES record (Fig. 7). This method allows identification of significantly different median values, which are not themselves influenced by outliers. The notched boxplots confirm our observations of the Na record -no 50-yr interval has a significantly different median from any other in the record. However, the notched boxplots show that median concentrations of Al, Mn and Ce are significantly (p < 0.05) higher for every 50-yr interval between 1500 and 1800 AD compared to 1800 to 2000 AD. The 1800-1850 AD interval does not have significantly different median values for Al, Mn and Ce compared to 1850-2000 AD, despite some spikes in Al and Ti concentration reaching values an order of magnitude above the baseline (Fig. 6, Al shown). This highlights that, whilst the anomalously high lithophile concentration data points are outliers statistically, they are valid geochemical analyses that can provide useful information about paleo-environmental conditions. Substantial variability between the behaviour of different lithophile elements, not observed in the record of the last 35 yr (Fig. 4), can be discerned in the long-term record if the entire dataset is considered (Fig. 6). For example, the mean Ce concentrations decreased by 74 % after 1850 AD, whilst mean Mn concentration decreased by only 38 % (Table 2). A correlation matrix of the raw trace element data for the pre-1850 AD period displays three groups of lithophile elements: R 2 values of > 0.8 are produced between Ti and Al; R 2 > 0.7 between Mn and Rb; and R 2 > 0.9 between Y, La and Ce (p < 0.01) (Fig. 8b). Figures 4, 6 and 7 display data for a representative element from each of these groups. In contrast, the correlation matrix for the post-1850 AD period shows inter-correlation of all the lithophile elements, except Al and Ba, at R 2 > 0.5 (p < 0.01) (Fig. 8b). , 8, 1223-1238 Methylsulphonate concentrations in the MES ice core are available for the section of record dating from 1795 to 2006 AD and vary between 1 and 187 ppb about a mean of 54 ppb (Fig. 9). A period of significantly higher concentrations is identifiable between 1825 and 1875 AD. The mean concentration of this period is 81 % higher than that recorded between 1876 and 2006 AD.

Lithophile elements and katabatic wind strength
Lithophile trace element concentrations in the MES ice core are significantly higher than those measured on snow or ice samples from inland Antarctica. For example, Hur et al. (2007)  removes the effect of statistical outliers (Rousseeuw, 1984). B) Matrices on the right panel 5 display correlations performed on raw data of non-Gaussian distribution.  (Rousseeuw, 1984). (B) Matrices on the right panel display correlations performed on raw data of non-Gaussian distribution.
period are 1456 ppt, 4.0 ppt, 72.5 ppt and 33.1 ppt -higher by more than an order of magnitude. The outcome is the same if MES V, Mn, Ba and Pb concentrations are compared to those reported for samples dating back to the early 19th century from Coats Land, East Antarctica (Planchon et al., 2002). Lithophile element concentrations reflect the level of mineral dust present in ice core samples (Petit et al., 1981;Rhodes et al., 2011). Unlike most other Antarctic ice core locations, MES has several areas of outcropping rock located within 160 km, in the McMurdo Dry Valleys and along the Transantarctic Mountains (Fig. 1). We therefore surmise that the atypically high lithophile element concentrations recorded at MES are the result of mineral dust input from local (100 s of km distance away), rather than global (1000 s of km distance away), dust sources. Several studies of mineral dust deposition within the southwestern Ross Sea region support this assumption (Ayling and McGowan, 2006;Dunbar et al., 2009;Bull, 2009). We do not attempt to relate the lithophile element signatures recorded in the ice core to the various dust lithologies of the region as previous work has shown this does not provide a reliable indicator of provenance (Rhodes et al., 2011).
Increased lithophile element concentrations during the autumn and winter months (Fig. 5) indicate that strong (> 35 m s −1 at Scott Base; data available at www.cliflo.co. nz) katabatic winds originating from the continental interior Table 2. Mean (italicised), median and range (in parentheses) of selected stable isotopic and chemistry parameters within specified time intervals of the MES ice core record. δD and d-excess data measured by LGR laser instrument have been adjusted (Supplement). (> DL to 39.7) (0.21 to 401) 1 Figure 9: MSdata for the MES ice core plotted with the timing of a diatom bloom reported 2 by Leventer and Dunbar (1988) in two sediment core records from McMurdo Sound, to the 3 west of Ross Island. Representative analytical error bars (± 2σ) are displayed on selected 4 points. A gap in the time series exists because some samples were not analysed for MS -5 concentration. 6 7 Fig. 9. MS − data for the MES ice core plotted with the timing of a diatom bloom reported by Leventer and Dunbar (1988) in two sediment core records from McMurdo Sound, to the west of Ross Island. Representative analytical error bars (± 2σ ) are displayed on selected points. A gap in the time series exists because some samples were not analysed for MS − concentration.
are required to transport additional local mineral dust material to MES. To further investigate this relationship, we correlate the ice core data with NCEP/NCAR reanalysis meridional wind strength (Kalnay et al., 1996). In this region of Antarctica, the meridional (south to north) component of the wind field is an approximation of katabatic wind flow (Fig. 10a). No significant correlation between wind speed and lithophile element concentration could be achieved for the entire time period for which reliable Antarctic NCEP/NCAR data are available (1979. However, when individual decades are studied, a significant correlation with the MES Al concentration is observed for the 1990-2000 AD time period (Fig. 10b). In contrast to the 1980s, the 1990s were characterised by a relatively strong teleconnection between the El Niño Southern Oscillation and Antarctica (Fogt and Bromwich, 2006). This led to enhanced katabatic flow speeds in the Ross Sea region (Bertler et al., 2004a), and a stronger signal in the MES katabatic wind proxy. The area of significant correlation includes the Transantarctic Mountains to the south of MES, a source region for katabatic winds to the site, which further confirms katabatic flow as the mechanism for mineral dust transport to MES.
As lithophile element concentrations in the MES ice core record are proxies for katabatic wind strength, we conclude that the elevated lithophile concentrations prior to 1853 AD (± 8 yr) indicate greater katabatic wind activity. Based on this transition and a cooling trend prior to 1850 AD suggested Clim. Past, 8, 1223-1238  by δD (discussed below), we equate the pre-1850 AD portion of the MES record to the LIA, in line with several other Antarctic ice core records Masson et al., 2000;Kreutz et al., 1997;Mosley-Thompson, 1995).
Formulating an approximation of how much stronger katabatic winds were is challenging, principally because the relationship between dust entrainment and wind speed is not linear (Bagnold, 1941;Pye, 1987;Schulz et al., 1998). However, as the elevated lithophile element concentrations of the LIA are not repeated in the post-1850 AD ice core record, we can assume that wind speeds did not reach such strength again. At Scott Base the maximum daily gust speed recorded since 1972 was 57.7 m s −1 from the south, so we conclude that LIA maximum wind speeds must have significantly exceeded this value. Evidence for stronger katabatic winds across East Antarctica during this interval of the LIA is also found in an ice core microparticle record from the South Pole (Mosley-Thompson, 1995).
By examining the behaviour of the three groups of lithophile elements identified in Sect. 4.2 and their links to mineralogy and wind speed, considerable climate variability within the LIA is revealed. Al (Figs. 6 and 7) and Ti (Table 2) show increased median concentrations coupled with discrete periods of dramatic concentration increases of several orders of magnitude during the LIA. These elements are predominantly sourced from complex silicate minerals, such as feldspars, which have relatively high densities and grain sizes, and therefore require higher wind velocities to initiate transport in suspension (Bagnold, 1941). The discrete periods of increased Al and Ti concentrations therefore represent the most extreme wind speeds of the LIA period. In contrast, the lithophile elements of the second group, Mn and Rb (Table 2), which show statistically higher median concentration increases (Fig. 7) but relatively muted concentration spikes during the LIA, are common in Fe-Mn oxyhydroxides and sheet silicate minerals such as micas and clays. The relatively low densities and tendency of these phases to rapidly weather to sheet-like particles of a low, uniform grain size result in a lower wind velocity threshold to initiate transport of these minerals in suspension (Bagnold, 1941). It can therefore be assumed that Mn and Rb concentrations in the MES ice core do not reflect maximum wind strength in the LIA. The third group of lithophile elements, consisting of La, Ce and Y, shows characteristics of both the other two groups. During discrete time periods of the LIA, concentrations increase by an order of magnitude but the timings of these extreme events do not always match that of the greatest increases in Al (Fig. 6) and Ti. Similarly to Mn, La, Ce and Y are abundant in ferromanganese oxyhydroxide minerals but they are also common impurities in apatite. Given this information, the lithophile element records suggest that between 1800 and 1853 AD the prevalent or mean katabatic wind speed decreased to a level similar to that of the present day but that there were still some anomalously strong wind events capable of transporting Al-and Ti-rich silicate minerals.

Stable isotopes and temperature
The MES δD record suggests that colder temperatures, as well as stronger winds, than those experienced in the past 50 yr prevailed during the LIA. Decreases in δD indicate cooler air temperatures (Craig, 1961;Dansgaard, 1964), and all the periods of significant deviations below the 1950 to 2006 AD mean identified on Fig. 6 (letters A-E) date from the LIA. Additionally, sustained cold temperatures appear to be linked to the strongest katabatic flow events as the three most recent cold periods identified in the MES ice core (Fig. 6, letters C-E) occur synchronously with the strongest wind events of the LIA, as recorded by the concentrations of Al (Fig. 6) and Ti.
Direct calibration of the stable isotopic paleothermometer for the MES ice core record is challenging, because the weather conditions at MES are significantly different to those of nearby weather stations and the extreme topography of Ross Island is not well represented in gridded reanalysis datasets. As an alternative, we employ the Taylor Dome δ-borehole temperature calibration of 4 ± 1.5 ‰ δD per • C  to estimate the degree of cooling experienced at MES during the LIA. Taylor Dome is also located in the Ross Sea region (Fig. 1a) and is the most proximal ice core for which a local δ-temperature slope has been established. The difference between the mean δD of the LIA and the post-LIA period is 6.2 ‰ (Table 2), which equates to a mean cooling of 1.6 ± 0.60 • C during the LIA. The extensive review of Antarctic stable isotopic signatures by Masson-Delmotte et al. (2008) found that δ-temperature slopes can vary by 20 % between localities, which increases the uncertainty on our cooling estimate to ± 1.41 • C. Our estimate agrees with the ∼ 2 • C cooling reported for this time period at Victoria Lower Glacier in the northern McMurdo Dry Valleys (Fig. 1c)  . Furthermore, a temperature record obtained from the West Antarctic Ice Sheet (WAIS) Divide ice core borehole also found evidence for 0.52 ± 0.28 • C cooling between 1400 and 1800 AD relative to the last 100 yr (Orsi et al., 2012).

Marine aerosol input
The Na + /SO 2− 4 values of MES ice core samples are all lower than that of seawater, with only one exception. The mean Na + /SO 2− 4 is 0.56, compared to Na + /SO 2− 4 seawater of 3.97. This is consistent with marine aerosol sourced from the open ocean via the bubble bursting mechanism, rather than from frost flowers entrained from the sea ice surface (Rankin et al., 2004). The timing of marine aerosol deposition during the summer months (Fig. 5) of minimum sea ice extent is also suggestive of an open water source. We attribute the deposition of marine aerosol at MES to synoptic-or mesoscale cyclonic systems (Sinclair et al., 2010). The opposite seasonal cycle of lithophile and marine element concentrations at MES (Fig. 5) supports the assignment of two different meteorological systems, katabatic winds and marine cyclones respectively, as transport mechanisms for these chemical species.
We observe that in the 1970-2006 AD record d-excess peaks greater than the long-term mean have corresponding peaks in Na concentration in 90 % of cases (Fig. 4, grey shaded bars). Furthermore, the maximum summer values of Na and d-excess between 1970 and 2006 AD are significantly correlated (R 2 = 0.58, p < 0.01) suggesting a common source. Deuterium excess is thought to be controlled by conditions at the moisture source, which are principally relative humidity, sea surface temperature (SST) and wind speed (Merlivat and Jouzel, 1979;Petit et al., 1991), but is also controlled by the degree of super-saturation during snow flake formation and is therefore related to cloud temperature (Jouzel and Merlivat, 1984). As a result, detailed delineation of potential oceanic moisture source regions, coupled with isotopic modelling, is required to understand the origin of the MES d-excess signal. However, we have established that more positive d-excess values are associated with inputs of marine aerosol sourced from open water. Therefore, we tentatively speculate that more positive d-excess values are related to increased cyclonic activity in the Ross Sea region, which promotes marine aerosol generation by bubble bursting (Kreutz et al., 2000). A detailed study of back trajectories conducted for a site close to the Byrd Glacier (Fig. 1b) identified a statistical link between ice core d-excess and the frequency of oceanic cyclone systems, albeit with several complicating factors (Markle et al., 2012). We speculate that the elevated d-excess values observed prior to 1875 AD, until at least 1658 AD (Fig. 6), could be the result of more frequent intrusion of marine cyclones, but more evidence is required to evaluate this hypothesis. One argument against a stronger cyclonic intrusion is the lack of any accompanying increase in Na concentration (Fig. 8) prior to 1875 AD, which we would expect given the observed relationship between d-excess and Na over the past 35 yr.

Ross Sea polynya
The MS − record of the MES ice core provides a proxy record of Ross Sea polynya area (Rhodes et al., 2009). Although it has limited temporal coverage, the higher concentrations recorded between 1825 and 1875 AD indicate that the polynya area was greater during this time (Fig. 9). As variability in katabatic wind strength has been directly linked to polynya size (Morales Maqueda et al., 2004), we conclude that the stronger katabatic winds of the LIA were responsible for the increase in polynya size. Increased polynya area exerts a positive feedback on marine primary productivity (Arrigo and van Dijken, 2004). By extrapolating the linear relationship between MS − concentration (x) and primary production rate (y) established by Rhodes et al. (2009) (y = 1.35 x − 3.41), we calculate that between 1825 and 1875 AD primary productivity rates were approximately 80 % higher than the post-1875 AD level.
Diatom abundances from two sediment cores retrieved from the west of Ross Island reveal a period of increased primary productivity between 1825 and 1880 AD (Leventer and Dunbar, 1988). This timing closely corresponds to the interval of higher MS − concentrations observable in the MES record ( Fig. 9) suggesting that both records respond to the same signal. These sediment cores were dated by the extrapolation of accumulation rates for the first 100 yr of the record, calculated from Pb-210 measurements. All time ranges are referred to as "approximate" by Leventer and Dunbar (1988), and dating errors are likely to be greater than those of the MES core.
A second, more prolonged, time period (1600-1875 AD) of increased open water diatom populations was identified in four sediment cores retrieved from the north and northwest of Ross Island (Leventer and Dunbar, 1988). These records indicate that the Ross Sea polynya was greater in area, or temporally more persistent, through much of the time covered by the MES ice core and consequently much of the LIA. The 1600-1875 AD interval includes all of the low temperature, high wind speed events recorded in the core. We propose that colder temperatures in the Ross Sea region led to stronger katabatic flows that maintained a spatially larger and/or more temporally persistent polynya between 1600 and 1850 AD. We also have evidence for a greater polynya area between 1850 and 1875 AD relative to the ensuing time, but δD and lithophile element concentrations provide no indication of colder temperatures or windier conditions beyond 1853 AD. Here the d-excess record becomes intriguing because the time interval of relatively high values persists until 1875 AD, suggesting that whatever process is controlling dexcess, be it cyclonic activity or something else, is linked to polynya area. Further investigation into the controls on dexcess at MES is required to decipher this signal.

Conclusions
Stable isotope, major ion and trace element chemistry records from the MES ice core document the last five centuries of climate in the southwestern Ross Sea region of Antarctica at sub-annual resolution. Dating of the record is accomplished by a combination of annual layer counting and ice flow modeling. Absolute dating uncertainty could exceed ± 50 yr for the pre-1816 AD record as no age tie points are distinguishable and dating relies on a 1-D flow model. Examination of the monthly resolution record of the last 35 yr links lithophile elements concentrations to mineral dust transport by katabatic winds, whilst d-excess and marine element concentrations are tentatively related to incursion of marine cyclones.
The MES stable isotope record suggests that the Ross Sea region experienced 1.6 ± 1.4 • C cooler average temperatures during the LIA in comparison to the last 150 yr. For every 50-yr interval between 1500 and 1800 AD, we observed statistically higher median values of lithophile element concentrations than during the past 150 yr. This indicates that strong (> 57 m s −1 ) katabatic winds were prevalent during this time. By considering the variable temporal trends exhibited by three groups of lithophile elements, several time periods of especially strong katabatic winds are identifiable by the occurrence of Al and Ti concentration increases of over an order of magnitude. These strong wind events occur coincidently with cooler temperatures, as indicated by δD values, during time intervals centered at ca. 1690 AD, 1770 AD and 1840 AD.
The ice core MS − record and diatom blooms recorded in sediment cores (Leventer and Dunbar, 1988) indicate that marine primary productivity in the Ross Sea polynya was increased between 1600 and 1875 AD. We conclude that during the LIA colder temperatures promoted stronger katabatic winds across the Ross Ice Shelf, causing sea ice divergence and creating a greater polynya area in the Ross Sea.