2.1 Fieldwork, sample processing and analysis

Two field campaigns were conducted in the UG region in austral summers 2014 and 2015. UG is one of the major outlet glaciers within the Ellsworth Mountains and flows into the Ronne–Filchner Ice Shelf in the Weddell Sea sector of Antarctica (Fig. 1a and b). It is composed of several glacier tributaries covering a total area of 2561 km². UG has a total length of 86 km, a maximum ice thickness of 1540 m and a maximum depth of the snow–ice boundary layer of 120 m. The subglacial topography of the glacier valley is smooth with U-shaped flanks. The bedrock is located below sea level (−858 m; Rivera et al., 2014).

We examine six firn cores (GUPA-1, DOTT-1, SCH-1, SCH-2, BAL-1 and PASO-1) retrieved at different locations (Fig. 1b), ranging between 760 m a.s.l. (GUPA-1 and DOTT-1) and 1900 m a.s.l. (PASO-1) in altitude, using a portable solar-powered and electrically operated ice-core drill (Backpack Drill; icedrill.ch AG). GUPA-1 was drilled near the ice-landing strip on UG. DOTT-1 originates from an ice rise towards the Ronne–Filchner Ice Shelf. SCH-1 was retrieved from the ice divide between the glaciers Schneider and Schanz. SCH-2 and BAL-1 were both taken in U-shaped glacial valleys (Schneider Glacier and Balish Glacier). PASO-1 originates from a plateau west of the Gifford Peaks. Details on the drill locations and basic core characteristics are given in Table 1.

Table 1Details on drill locations and basic statistics of the stable water isotope composition and accumulation rates of the six firn cores retrieved from UG. In addition, minimum, mean and maximum values of stable oxygen isotope annual means and accumulation rates are given for the period covered by all cores (1999–2013). Mean values of stable water isotopes and accumulation rates are highlighted in bold.

Download Print Version | Download XLSX

For firn cores BAL-1 and PASO-1, high-resolution (< 1 mm) density profiles were obtained using X-ray microfocus computer tomography (ICE-CT; Freitag et al., 2013) at the ice-core-processing facilities of the Alfred Wegener Institute (AWI), Helmholtz Centre for Polar and Marine Research, in Bremerhaven, Germany. The cores were sampled at 0.025 m resolution and analysed for stable water isotopes using a cavity ring-down spectrometer (L2130-i; Picarro Inc.) coupled to an autosampler (PAL HTC-xt; CTC Analytics AG) at the Stable Isotope Laboratory of the AWI in Potsdam, Germany. Stable water isotope raw data were corrected for linear drift and memory effects following the procedures suggested by van Geldern and Barth (2012) using six repeated injections per sample, from which the first three were discarded. The drift- and memory-corrected stable water isotope compositions were calibrated with a linear regression analysis using four different in-house standards that have been calibrated to the international VSMOW2 (Vienna Standard Mean Ocean Water) –SLAP2 (Standard Light Antarctic Precipitation) scale. Stable water isotope ratios are reported in per mil (‰) versus VSMOW2. The precision of the measurements is ±0.08 ‰ for δ¹⁸O and ±0.5 ‰ for δD.

For firn cores GUPA-1, DOTT-1, SCH-1 and SCH-2, density profiles were constructed by section-wise determining the core volume and weight. Accordingly, the average resolution of density profiles is 0.25 m for GUPA-1, 0.4 m for DOTT-1, 0.27 m for SCH-1 and 0.78 m for SCH-2. Cores GUPA-1, DOTT-1 and SCH-1 were sampled at 0.05 m resolution for stable water isotope analysis carried out at the Stable Isotope Laboratory of the Universidad Nacional Andrés Bello (UNAB) in Viña del Mar, Chile. For the measurements an off-axis integrated cavity output spectrometer (TLWIA 45EP; Los Gatos Research) was used with a precision of 0.1 ‰ for δ¹⁸O and 0.8 ‰ for δD (Fernandoy et al., 2018). Each sample was measured twice in different days using 10 repeated injections, from which the first four were discarded on each measurement. Stable water isotope raw data were corrected for linear drift and memory effects and then normalized to the VSMOW2–SLAP2 scale using the software LIMS (Laboratory Information Management System; Coplen and Wassenaar, 2015). For data normalization, three different in-house standards and one USGS standard (USGS49) calibrated to the VSMOW2–SLAP2 scale were used.

Core SCH-2 was analysed at the British Antarctic Survey (BAS) in Cambridge, UK. Stable water isotopes were determined at 0.05 m resolution using a Picarro L2130-i analyser with measurement precision of 0.08 ‰ for δ¹⁸O and 0.5 ‰ for δD. Major ions (Cl⁻, ${\mathrm{NO}}_{\mathrm{3}}^{-}$, ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$, MSA⁻, Na⁺, K⁺, Mg²⁺, Ca²⁺) were measured at 0.05 m resolution using a Dionex reagent-free ion chromatography system (ICS-2000). Furthermore, longitudinal subsections of the core (32 mm×32 mm in size) were melted continuously on a chemically inert, ultra-clean melt head to measure electrical conductivity, dust and H₂O₂ at high resolution (∼1 mm; continuous flow analysis, CFA; McConnell et al., 2002). In addition, core PASO-1 was analysed for liquid conductivity, H₂O₂, ${\mathrm{NO}}_{\mathrm{3}}^{-}$, ${\mathrm{NH}}_{\mathrm{4}}^{+}$, black carbon and various chemical elements at the Trace Chemistry Laboratory of the Desert Research Institute (DRI) in Reno, Nevada, USA, according to methods described by Röthlisberger et al. (2000) and McConnell et al. (2002, 2007). Chemical elements were measured using two Thermo Finnigan Element2 inductively coupled plasma mass spectrometers (ICP-MS). Subsequently, values of ssNa (sea-salt Na) and nssS (non-sea-salt S), which are used for the core chronology of PASO-1, were calculated from Na and S concentrations according to Röthlisberger et al. (2002) and Sigl et al. (2013), respectively, applying relative abundances from Bowen (1979; Eqs. S1a–S1d in the Supplement).

The corrected and calibrated δ¹⁸O and δD values were used to calculate the d excess (d excess = δD$-\mathrm{8}\cdot {\mathit{\delta }}^{\mathrm{18}}\mathrm{O}$; Dansgaard, 1964) and to derive a co-isotopic relationship ($\mathit{\delta }\mathrm{D}=m\cdot {\mathit{\delta }}^{\mathrm{18}}\mathrm{O}+n$) for each core. Since there are no stable water isotope data of recent precipitation available for the UG region, a Local Meteoric Water Line (LMWL) was inferred from the co-isotopic relationship of all cores. Based on findings by Fernandoy et al. (2010) we believe that this composite co-isotopic relationship – at a first approximation – can be referred to as LMWL at UG.

2.2 Dating of firn cores and time series construction

Recent studies have used the seasonality of stable water isotopes and chemical parameters to reliably date firn cores from Antarctica (e.g. Vega et al., 2016; Caiazzo et al., 2016). Stable water isotope profiles (δ¹⁸O and δD) of the six firn cores are displayed with respect to depth in Fig. S1 in the Supplement. Data gaps in GUPA-1, DOTT-1 and SCH-1 are due to leaking sample bags (GUPA-1: 2 samples; DOTT-1: 3 samples; SCH-1: 2 samples). Firn cores for which glacio-chemical data are available, i.e. SCH-2 and PASO-1, were dated using annual layer counting (ALC) of stable water isotopes and chemical parameters focusing on H₂O₂ for SCH-2 (Fig. 2) and nssS for PASO-1 (Fig. 3). H₂O₂ and nssS exhibit both clear seasonal alternations between the highest and lowest values, since the former parameter is directly linked to the annual insolation cycle (Lee et al., 2000; Stewart, 2004) and the latter to marine biogenic activity (phytoplankton; Kaufmann et al., 2010). In addition, for the age model construction of SCH-2 methanesulfonic acid (MSA), ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ and the Cl∕Na ratio were used for corroborating the years identified by ALC of H₂O₂ and stable water isotopes (Vega et al., 2018). For the dating of PASO-1 the signal of the Mt. Pinatubo eruption (1991) could be found via the coincidence of an above-average peak in the nssS/ssNa record with a lacking winter minimum in the nssS record at 9.5–9.7 m depth (Fig. 3). Similar changes in wintertime nssS were observed in the well-dated ice cores PIG2010, DIV2010 and THW2010 from West Antarctica (Pasteris et al., 2014). Hence, the signal attributed to the Mt. Pinatubo eruption was used as an additional tie point for the age model construction of PASO-1. For cores GUPA-1, DOTT-1, SCH-1 and BAL-1 we applied ALC of stable water isotopes and then matched the preliminary age–depth relationships to the SCH-2 age scale.

Figure 2Age scale for firn core SCH-2 constructed by counting and inter-matching of maximum peaks (dashed lines) in CFA-derived profiles of stable water isotope composition (δD) and different chemical parameters (H₂O₂, Cl∕Na, MSA and ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$). Smoothed 2P is a two-point running average. Note that the maxima of the chemical parameters do not necessarily coincide with the respective maxima in δD due to the different seasonality of the proxies.

Download

Snow accumulation rates at the firn core sites were determined and converted to metres of water equivalent per year (m w.e. a⁻¹) based on measured densities (Fig. S2).

Composite stable water isotope and accumulation records were constructed for the entire UG region by combining time series of annually averaged stable water isotopes and accumulation rates of the individual firn cores. In order to assign each firn core the same weight and to not over-represent a certain firn core drill site, annually averaged data were standardized before stacking (Stenni et al., 2017; Eq. S2). Linear trends were calculated and tested for their statistical significance using the non-parametric Mann–Kendall tau and Sen slope (s) estimator trend test (Mann, 1945; Sen, 1968; Kendall, 1975) with correction for autocorrelation according to Yue and Wang (2004). Signal-to-noise ratios for stable water isotope and accumulation rate time series of the UG firn cores were estimated according to Fisher et al. (1985; Eq. S3).

2.3 Meteorological database and backward trajectory analysis

Meteorological data from an automatic weather station (AWS) located at the UG ice runway (station: Wx7; 79^∘46^′ S, 83^∘16^′ W; 705 m a.s.l.; Fig. 1b) cover the 4-year period from 1 February 2010 to 8 February 2014. The AWS records near-surface air temperature (2 m), wind speed, wind direction, relative humidity and air pressure every 10 min. Additionally, hourly resolved data of the same meteorological parameters are available from a second AWS operated on UG (station: Arigony; 79^∘46^′ S, 82^∘54^′ W; 693 m a.s.l.; Fig. 1b) covering the period from 14 December 2013 to 29 March 2018. Since differences between the Wx7 and the Arigony records are small throughout the overlapping period (14 December 2013 to 8 February 2014; e.g. the mean differences in air temperature and air pressure are 0.74 ^∘C and 5.76 hPa, respectively), the two datasets were combined in order to expand the meteorological record.

The meteorological data from the two AWSs (air temperature and air pressure) along with measured surface densities (Fig. S2) and derived accumulation rates (Fig. 6 and Table 1) were used to model depth-dependent diffusion lengths for each firn core across the entire length following the approach described by Münch and Laepple (2018) and Laepple et al. (2018), respectively. Local near-surface air temperatures at the firn core drill sites were estimated from the AWS measurements by rescaling using the altitude of the respective site and a lapse rate of 1 ^∘C per 100 m. Local surface air pressures were estimated from the barometric height formula.

Furthermore, we use fields of near-surface air temperature (2 m), precipitation–evaporation and geopotential heights (850 mbar) from the European Centre for Medium-Range Weather Forecasts (ECMWF) Interim Reanalysis (ERA-Interim; 1979–2019, spatial resolution: 79 km, temporal resolution: 6 h; Dee et al., 2011; available at: https://www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era-interim, last access: 1 October 2019) for comparison with UG composite records of stable water isotopes and accumulation rates. ERA-Interim reanalysis data were aggregated to daily, monthly or annual values.

Annually averaged stable water isotope compositions and accumulation rates were also related to time series of climate modes such as SAM and ENSO as well as to SIE and SIC in order to identify potential dominant drivers of climate variability in the UG region.

For comparison with the UG stable water isotope and accumulation records, mean monthly SIE for different Antarctic sectors (Weddell Sea: 60^∘ W–20^∘ E; Indian Ocean: 20–90^∘ E; Western Pacific: 90–160^∘ E; Ross Sea: 160^∘ E–130^∘ W; Bellingshausen–Amundsen Sea: 130–60^∘ W) was obtained from the National Aeronautics and Space Administration (NASA; Cavalieri et al., 1999, 2012; available at: https://neptune.gsfc.nasa.gov/csb/index.php, last access: 5 September 2019). Note that these data are only available for the period 1979–2012. Satellite-derived SIC data were acquired from the National Snow and Ice Data Center (NSIDC). The dataset NSIDC-0079 – Bootstrap Sea Ice Concentrations from Nimbus-7 SMMR and DMSP SSM/I-SSMIS, Version 3 – with a spatial resolution of 25 km×25 km was used (Comiso, 2017; available at: https://nsidc.org/data/nsidc-0079, last access; 1 October 2019). Furthermore, we use the Marshall SAM index as an indicator for the prevailing SAM phase (Marshall, 2003; available at: https://legacy.bas.ac.uk/met/gjma/sam.html, last access: 5 September 2019). The multivariate ENSO index (MEI) serves as an indicator for the occurrence and strength of El Niño and La Niña events (Wolter and Timlin, 1993, 1998; available at: https://www.esrl.noaa.gov/psd/enso/mei/table.html, last access: 5 September 2019).

Spatial and cross-correlation analyses between UG composite time series of stable water isotopes and accumulation rates and time series of ERA-Interim reanalysis data, SIE, SIC, the Marshall SAM index and the MEI, respectively, were carried out calculating Pearson correlation coefficients (r) and p values (p) at the 95 % confidence level (i.e. α=0.05). For calculating spatial correlations all time series were detrended.

We performed backward trajectory analysis using the Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT) model (Draxler and Hess, 1998; Stein et al., 2015; available at: https://ready.arl.noaa.gov/index.php, last access: 24 September 2019) in order to determine potential moisture source regions and transport pathways of precipitating air masses for the UG region. The drill site of firn core SCH-1 was taken as initial point for the calculation of 5 d backward trajectories. This location was selected as it is less influenced by post-depositional processes as well as wind-induced redistribution and removal of snow. Thus, it is among all available firn core drill sites the most representative for snowfall events in the UG region. ERA-Interim time series of daily precipitation extracted from the nearest grid point at 1061 m a.s.l. (Table 1) for the period 2010–2015 (resolution: $\mathrm{0.75}{}^{\circ }\times \mathrm{0.75}{}^{\circ }$; available at: http://apps.ecmwf.int/datasets/data/interim-full-daily/levtype=sfc/, last access: 24 September 2019) served as input to identify precipitation events on a daily scale. We used a minimum threshold corresponding to 1 % of the mean annual snow accumulation at UG (see below), a value that in terms of percentage is equivalent to the one applied by Thomas and Bracegirdle (2009). Based on input data from the Global Data Assimilation System (GDAS-1; available at: https://www.ready.noaa.gov/archives.php, last access: 24 September 2019) 121 backward trajectories were calculated corresponding to daily precipitation events occurring in the period 2010–2015. They were then subdivided into their respective seasons (DJF, MAM, JJA and SON). Individual backward trajectories were combined to four different clusters that group precipitation events with statistically similar transport pathways according to their spatial variance (Stein et al., 2015).

3.1 Meteorological data

The mean daily air temperature for the composite record of near-surface air temperature (2010–2018; Fig. 4) is $-\mathrm{21.3}\pm \mathrm{8.5}$ ^∘C with an absolute minimum of −46.8 ^∘C recorded on 17 July 2017 and an absolute maximum of +3.3 ^∘C measured on 21 February 2013. Daily air temperatures are the highest during December and January, with mean values of $-\mathrm{9.2}\pm \mathrm{2.9}$ ^∘C and $-\mathrm{9.1}\pm \mathrm{3.0}$ ^∘C, respectively, and lowest from April to September, with mean values below −25 ^∘C. The coldest month of the whole composite record period is July, with a mean air temperature of $-\mathrm{28.8}\pm \mathrm{5.6}$ ^∘C. Daily wind speeds have a mean value of $\mathrm{6.9}{\mathrm{ms}}^{-\mathrm{1}}\pm \mathrm{5.1}{\mathrm{ms}}^{-\mathrm{1}}$ with a predominant direction from SW (220^∘). The maximum wind speed recorded is 29.7 ms⁻¹. Wind speeds are higher than 1.2 ms⁻¹ in >75 % of all data. The observations are in line with those made by Rivera et al. (2014) for the period 2008–2012 and imply the possibility of substantial redistribution of snow due to wind drift.

Figure 3Age scale for firn core PASO-1 constructed by counting and inter-matching of maximum peaks (dashed lines) in profiles of stable water isotope composition (δD) as obtained from discrete sample measurements and in CFA-derived profiles of nssS and nssS/ssNa, respectively. The year of the eruption of Mt. Pinatubo (1991) that is used as tie point for annual layer counting is highlighted. Note that the maxima of nssS and nssS/ssNa do not necessarily coincide with the respective maxima in δD due to the different seasonality of the proxies.

Download

Figure 4Composite record of mean daily and mean monthly air temperatures recorded at two nearby UG AWS sites (stations: Wx7, Arigony) from February 2010 to March 2018. The mean daily air temperature for the entire composite record period (−21.3 ^∘C; dashed black line) and the period overlapping with UG firn core records (February 2010 to November 2015; grey bar) are also indicated.

Download

3.2 Firn core age model

The results of firn core dating are given in Table 1. The estimated dating uncertainty is ±1 year for cores dated with stable water isotopes and glacio-chemistry (SCH-2 and PASO-1) and ±2 years for cores dated with stable water isotopes only and subsequent matching to the SCH-2 age scale (GUPA-1, DOTT-1, SCH-1 and BAL-1). Core DOTT-1 (16 years, 1999–2014) exhibits the shortest and PASO-1 (43 years, 1973–2015) the longest record. Note that for the age model construction of GUPA-1 and BAL-1, 2 and 3 years, respectively (1990 and 2001 for GUPA-1; 1981, 1983 and 1994 for BAL-1), were identified from linear interpolation only and are not clearly visible in the stable water isotope records due to smoothing in the respective core sections. Furthermore, for SCH-1 the first year (1986) was identified by linear extrapolation at the lowest end of the core using the depth–age relationship between the previous two clearly identifiable peaks (years 1987 and 1988). For all cores the last year (either 2014 or 2015) was excluded from further analyses as it is incomplete.

3.3 Firn core stable water isotopes and accumulation rates

The mean stable oxygen isotope composition of the six firn cores ranges from −36.6 ‰ (PASO-1) to −29.9 ‰ (DOTT-1) with standard deviations varying between 2.3 ‰ (PASO-1) and 3.9 ‰ (DOTT-1; Table 1). Absolute minimum δ¹⁸O (δD) values are found in GUPA-1 and absolute maximum δ¹⁸O (δD) values in DOTT-1. Note that the results for GUPA-1 have to be handled with caution as this core was drilled next to the UG field camp and ice-landing strip. Therefore, snow relocation effects due to wind drift and/or human activities (e.g. runway maintenance) might have biased its stable water isotope composition. This is also indicated by less pronounced seasonal alternations in the GUPA-1 stable water isotope records (Fig. S1). Despite different drill locations and altitudes, the range in mean d-excess values of the six firn cores is small (from 4.9 ‰ for SCH-2 to 7.0 ‰ for PASO-1). The slope of the co-isotopic relationship (Table 1 and Fig. S3a–f) is close to that of the Global Meteoric Water Line (GMWL; Craig, 1961) for all cores ranging between 7.94 (GUPA-1) and 8.24 (PASO-1). Hence, the original (oceanic) stable water isotope signal is preserved during moisture transport and snow deposition at the study site (Clark and Fritz, 1997). For all cores δ¹⁸O and δD values are highly correlated (R²≥0.98), with the largest variation in d excess observed for SCH-2 (values range from −5.6 ‰ to 17.2 ‰). The LMWL of the UG region was determined as $\mathit{\delta }\mathrm{D}=\mathrm{8.02}\cdot {\mathit{\delta }}^{\mathrm{18}}\mathrm{O}$ + 6.57 (R²=0.99, p=0; Fig. 5).

Figure 5Composite co-isotopic relationship (δ¹⁸O vs. δD) based on all individual samples (n=2348; white dots) of the six firn cores from UG with the equation, the coefficient of determination (R²) and the p value (p) of the linear regression (red dashed line). The GMWL is indicated in blue. The composite co-isotopic relationship is referred to as the LMWL of the UG region.

Download

Time-series analysis of δ¹⁸O annual means (Fig. S4) reveals statistically significant trends only for cores SCH-1 ($s=+\mathrm{0.039}$ ‰ a⁻¹, p<0.05) and BAL-1 ($s=+\mathrm{0.054}$ ‰ a⁻¹, p<0.0001). Statistically significant positive trends of d-excess annual means (Fig. S5) have been found for GUPA-1 ($s=+\mathrm{0.085}$ ‰ a⁻¹, p<0.01), SCH-2 ($s=+\mathrm{0.085}$ ‰ a⁻¹, p<0.0001) and PASO-1 ($s=+\mathrm{0.016}$ ‰ a⁻¹, p<0.01), whereas for DOTT-1 ($s=-\mathrm{0.110}$ ‰ a⁻¹, p<0.05) and SCH-1 ($s=-\mathrm{0.052}$ ‰ a⁻¹, p<0.0001) the d-excess trend is negative. The BAL-1 d-excess record exhibits no trend.

Mean annual accumulation rates vary between ∼0.18 m w.e. a⁻¹ (GUPA-1 and PASO-1) and ∼0.29 m w.e. a⁻¹ (SCH-2), with the lowest standard deviations found at the DOTT-1 and PASO-1 sites (∼0.04–0.05 m w.e. a⁻¹) and the highest ones exhibited by cores SCH-2 and BAL-1 (∼0.08 m w.e. a⁻¹; Table 1). Annual minimum accumulation ranges between 0.1 and 0.2 m w.e. a⁻¹. Annual maximum accumulation reaches – except at the PASO-1 site – values of higher than 0.3 m w.e. a⁻¹, with the absolute maximum found at the SCH-2 site in 1985 (0.47 m w.e. a⁻¹). In the same year, snow accumulation reaches a local minimum of 0.098 m w.e. a⁻¹ at the PASO-1 site (Fig. 6). At the GUPA-1 and BAL-1 sites snow accumulation decreased at the same rate of −0.003 m w.e. a⁻¹ (p<0.01). At the SCH-1 and SCH-2 sites the decrease in snow accumulation amounts to −0.002 m w.e. a⁻¹ (p<0.05) and −0.004 m w.e. a⁻¹ (p<0.0001), respectively. In contrast, snow accumulation exhibits a slight albeit statistically significant increase at the PASO-1 site ($s=+\mathrm{0.001}$ m w.e. a⁻¹, p<0.0001).

4.1 Potential noises influencing UG firn core records

In order to properly assess the environmental signals in stable water isotope and accumulation data of the UG firn cores, several aspects that could potentially induce noise need to be taken into account: the intermittency of precipitation, the redistribution of snow by wind drift and the diffusion in firn.

The analysis of diffusion lengths for the maximum depth of the UG firn cores revealed values between 0.07 m (PASO-1) and 0.11 m (GUPA-1), which are much lower than the mean annual layer thicknesses ranging between 0.35 m (PASO-1) and 0.60 m (DOTT-1; Table S1 in the Supplement). Therefore, we assume diffusion to be of minor importance for inducing noise to the stable water isotope records of the UG firn cores.

In contrast, wind drift and redistribution of snow certainly play an important role in the UG region, which is supported by data on wind speed (on average 7 ms⁻¹) and wind direction (predominantly from SW) from the two AWSs. The location of the individual firn cores at different altitudes as well as in different topographic positions (near the ice-landing strip, valley vs. plateau) causes site-specific differences in the susceptibility towards wind drift and thus to the reallocation of snow. However, as there are AWS data from only one location, i.e. the GUPA-1 drill site, for the period February 2010–March 2018 available, we have no reliable meteorological information for the other firn core drill sites in order to assess the local effects of wind drift in detail. Rivera et al. (2014) report that in the UG region katabatic winds can cause strong snow drift due to acceleration by the slope of the glacial valleys feeding UG. Studies of drifting snow in Antarctica using satellite remote sensing (Palm et al., 2011) and regional climate modelling (Lenaerts et al., 2012; Lenaerts and van den Broeke, 2012) provide evidence for a drifting snow frequency (defined as the fraction of days with drifting snow) of about 15 %–30 % in the UG region. Also, studies from close-by areas such as Patriot Hills (Casassa et al., 1998) and the Ronne–Filchner Ice Shelf (Graf et al., 1988) as well as from DML (Schlosser and Oerter, 2002; Kaczmarska et al., 2004) revealed that wind drift and random sastrugi formation are the main reasons for large spatial and temporal variations in accumulation rates. However, linkages between the different firn core drill sites at UG are speculative, e.g. the potential uptake of snow at the PASO-1 drill site (local minimum accumulation) and its redistribution towards the lower-altitudinal core sites (SCH-2; absolute maximum accumulation) by the predominantly south-westerly winds in 1985. The calculation of signal-to-noise ratios for stable water isotopes and accumulation rates can help to further assess the extent to which the UG firn cores are influenced by noise-inducing processes such as wind drift and diffusion. In the following, firn core GUPA-1 is excluded from statistical evaluation due to the likely biasing and smoothing of its stable water isotope and accumulation records as a consequence of its position near the UG ice-landing strip. Based on the two-by-two signal-to-noise ratios between the individual cores (Tables S2 and S3), we obtained for the UG firn cores a δ¹⁸O signal-to noise ratio of 0.60 for the entire record period (1973–2014) and of 0.78 when referring to the overlapping period (1999–2013). The signal-to-noise ratios are similar when considering only the three neighbouring cores BAL-1, SCH-1 and SCH-2 situated within 10 km distance in similar valley positions. They yield 0.60 for 1973–2014 and 0.86 for 1999–2013. The signal-to-noise ratios of δ¹⁸O in the UG region of between 0.6 and 0.86 are quite high – i.e. they are similar or slightly higher than signal-to-noise ratios of δ¹⁸O on the WAIS for interannual timescales (∼0.5–0.7) – and much higher than those in DML (<0.2) for multiannual to decadal timescales (Münch and Laepple, 2018).

Figure 6Annual accumulation rates at the drill sites of the six firn cores from UG for the period covered by the respective firn core. Sen slopes (s in m w.e. a⁻¹) and p values (p) are given for all firn cores, indicating that snow accumulation has decreased at most sites since the beginning of the respective record period.

Download

Calculation of the signal-to-noise ratio for accumulation rate time series of the UG firn cores revealed a value of 0.27 for the entire record period and 0.33 for the overlapping period, respectively. Compared to the signal-to-noise ratios of δ¹⁸O, these values are much lower, probably reflecting the strong influence of the site-specific characteristics on accumulation rates (e.g. the different exposure to wind drift and the subsequent relocation of snow). The signal-to-noise ratio increases significantly when referring to the close-by cores BAL-1, SCH-1 and SCH-2 or to SCH-1 and SCH-2 only (Tables S2 and S3). It yields 0.79 and 1.67, respectively, when referring to the entire record period. This might be due to the proximity of the drill sites (within 10 km distance) and the similarity of the site-specific characteristics of the three cores, i.e. the location in northwest–southeast-oriented, U-shaped glacial valleys, leading to similar accumulation patterns. When referring to the overlapping period only, the signal-to-noise ratio stays at a low value of 0.24 for the three cores but remains high for SCH-1 and SCH-2 (1.2; Tables S2 and S3). In summary, the signal-to-noise ratios of accumulation rates at UG are consistent with those determined by Schlosser et al. (2014) and Altnau et al. (2015) for firn cores from Fimbul Ice Shelf, DML, and much higher than those found in low-accumulation areas of the Amundsenisen mountain range, DML (Graf et al., 2002; Altnau et al., 2015). Generally, the higher the accumulation rates at a specific site are, the higher the signal-to-noise ratios for stable water isotopes and accumulation rates are (Hoshina et al., 2014; Münch et al., 2016). Hence, the UG region with higher accumulation rates as compared to low-accumulation sites on the EAIS (e.g. Oerter et al., 2000) exhibits relatively high signal-to-noise ratios for both stable water isotopes and accumulation rates, despite the likely influence by post-depositional processes, in particular the redistribution of snow due to wind drift. Hence, we conclude that the UG firn core records allow one to deduce temporal changes in stable water isotopes and accumulation rates on a regional scale.

Figure 7Profiles of the stable water isotope composition (δ¹⁸O) of the six firn cores from UG with respect to time. Years with prominent maxima and minima in δ¹⁸O time series – although not visible in all cores – are highlighted by red and blue shading, respectively. The peak in summer 2002 is the only maximum found in all cores.

Download

Figure 8Composite records of mean annual δ¹⁸O, d excess and snow accumulation in the UG region for the period 1980–2014 derived from standardized data (black; excluding firn core GUPA-1). Sen slopes (s) and p values (p) are given for each record. For δ¹⁸O, the composite record derived from non-standardized annually averaged data (grey) is also displayed as it exhibits a statistically significant positive trend that is worth mentioning (s in ‰ a⁻¹).

Download

4.2 Spatial and temporal variability of firn core stable water isotope composition and relation to near-surface air temperatures, sea ice and climate modes

4.2.1 Spatial and temporal variability of stable water isotopes

From the inter-comparison of minimum, mean and maximum values of δ¹⁸O annual means for the overlapping period (1999–2013, excluding GUPA-1; Table 1 and Fig. S4), generally a depletion of the stable water isotope composition with increasing height (“altitudinal effect”) and distance from the sea (“continentality effect”) has been detected as expected for a Rayleigh distillation process (Clark and Fritz, 1997). Core PASO-1 situated at the highest altitude and greatest distance from the sea shows the lowest mean δ¹⁸O values (−36.3 ‰). In contrast, the low-altitude core DOTT-1, which is situated closest to the sea, displays a distinctly higher mean stable water isotope composition (−29.7 ‰). However, core SCH-2 exhibits less depleted mean δ¹⁸O values (−34.1 ‰) than SCH-1 (−34.7 ‰), although located at about 250 m higher altitude. This inverted pattern, which is also visible in the minimum, mean and maximum δ¹⁸O (δD) values derived from all samples of the respective core (Table 1), might originate from snow reallocation processes within the glacial valley due to katabatic winds. Nonetheless, the existence of an altitudinal effect is confirmed by the δ¹⁸O–altitude relationship calculated from δ¹⁸O annual means for the overlapping period (excluding GUPA-1) yielding altitude = $-\mathrm{152.7}\cdot {\mathit{\delta }}^{\mathrm{18}}\mathrm{O}$–3803.4 with R²=0.85 and p<0.05 (Fig. S6).

Figure 9Spatial correlations (r) of annually averaged ERA-Interim (a) near-surface air temperatures (2 m), (b) geopotential heights (850 mbar) and (c) precipitation–evaporation with standardized mean annual (a) δ¹⁸O, (b) d excess and (c) snow accumulation in the UG region (red star) for the period 1980–2014 (excluding firn core GUPA-1). All time series were detrended before calculating spatial correlations. Only statistically significant correlations (p<0.05) are displayed. For (a) and (c) December–November annual averages (calendar year) were used, whereas for (b) August–July annual averages (winter–winter) were considered as spatial correlations appear more significant than for December–November annual averages.

The individual firn cores are generally well correlated with each other for δ¹⁸O and δD, for both the maximum overlapping period between two individual cores and the overlapping period (1999–2013; Tables S2 and S3). All cores display a common δ¹⁸O maximum in summer 2002 (Fig. 7). Correlation coefficients are highest for nearby cores such as SCH-1 and SCH-2, reaching up to r=0.66 (1999–2013: r=0.58) for δ¹⁸O (p<0.05) and r=0.71 (1999–2013: r=0.66) for δD (p<0.01). Only core BAL-1 shows relatively low correlations with all other cores except for SCH-1 (r=0.44 for δ¹⁸O and r=0.43 for δD, p<0.05; 1999–2013: r=0.49 for δ¹⁸O and r=0.50 for δD, p<0.1; Tables S2 and S3). The results of the cross-correlation analysis, the negligible influence of diffusion and the relatively high signal-to-noise ratios allowed us to construct a composite δ¹⁸O record (UG δ¹⁸O stack) based on standardized annually averaged data spanning the period that comprises at least three core records per year (1980–2014, excluding GUPA-1; Fig. 8). From the UG δ¹⁸O stack a more regional picture of the isotopic characteristics of precipitation at UG can be drawn. As a result we found only a negligible positive trend in δ¹⁸O, which is statistically not significant. Thus, we infer that at UG regional changes in δ¹⁸O values must have been negligible during the last 35 years. This is in line with findings from ice cores retrieved from the nearby Ronne–Filchner Ice Shelf and the Weddell Sea sector, respectively, that show no statistically significant trends in their stable water isotope time series (e.g. Foundation Ice Stream, Graf et al., 1999; Berkner Island, Mulvaney et al., 2002, and Stenni et al., 2017). However, it is worth mentioning that stacking firn core δ¹⁸O time series without previous standardization (excluding GUPA-1) yields a statistically significant positive trend of +0.054 ‰ a⁻¹ (p<0.0001; Fig. 8). This trend most likely results from the dominance of SCH-1 and BAL-1 as these are the only cores showing statistically significant (positive) trends in their δ¹⁸O records (Fig. S4).

Table 2Results of cross-correlation analysis for stable water isotope and accumulation composite records from UG and time series of the SAM index, multivariate ENSO index (MEI) and sea ice extent (SIE) in five different Antarctic sectors (annual means). Cross-correlations were calculated considering the record period that is covered by a minimum of three firn cores (1980–2014; excluding firn core GUPA-1). Prominent correlations with a low p value (p<0.1) are marked in bold and if statistically significant (p<0.01, α=0.05) in bold and italic.

Download Print Version | Download XLSX

In order to detect possible changes in the origin of precipitating air masses reaching the UG region, d-excess time series of individual firn cores have been inter-compared (Fig. S5). In line with δ¹⁸O, the d excess does not show clear trends or similar contemporaneous changes among the firn cores, suggesting little change in the main moisture sources and the origin of air masses precipitating over the UG region at least during the last four decades. The composite d-excess record (UG d-excess stack; excluding GUPA-1), which has been calculated despite the individual firn cores showing no cross-correlations in the d excess (Tables S2 and S3), corroborates this finding. It shows no statistically significant trend for the period 1980–2014 (Fig. 8). Dissimilarities between d-excess trends of individual firn cores might be partly explained by the presence of orographic barriers that separate the different drill sites from each other. This might lead to dissimilarities in moisture sources, even though the drill sites are located within an only 50 km horizontal distance.

4.2.3 Relation of stable water isotopes to large-scale climate modes and sea ice variability

Mean annual d-excess values (UG d-excess stack) exhibit a weak positive correlation with the SAM index (r=0.38, p<0.05; Table 2). Stronger contraction of the polar vortex during positive SAM phases facilitates the advection of warm and moist air from mid- and lower latitudes, i.e. from regions with higher SST and lower relative humidity towards Antarctica (Thompson and Wallace, 2000; Thompson and Solomon, 2002; Gillett et al., 2006). Hence, atmospheric water vapour with higher d-excess values is expected to reach Antarctica (Uemura et al., 2008; Stenni et al., 2010). This relationship is confirmed by spatial correlations of mean annual d-excess values with ERA-Interim geopotential heights (850 mbar) calculated for the period 1980–2014 (up to r>0.4, p<0.05; Fig. 9b). Increased geopotential heights above the mid-latitudes, as occurring during positive SAM phases (Thompson and Wallace, 2000; Thompson and Solomon, 2002), imply increased d-excess values of precipitation in the UG region. However, mean annual δ¹⁸O (δD) values (UG δ¹⁸O [δD] stack) show no correlation with the SAM index (Table 2). Furthermore, none of the isotopic values (i.e. δ¹⁸O, δD, d excess) exhibits a significant correlation with the MEI (Table 2). Hence, oceanic circulation changes associated with the alternation between El Niño and La Niña events seem to have no detectable influence on the stable water isotope composition of precipitation in the UG region.

Kohyama and Hartmann (2016) showed that the SAM index is statistically significantly positively correlated with SIE in the Indian Ocean sector of Antarctica. When comparing SIE in the different Antarctic sectors (Weddell Sea, Bellingshausen–Amundsen Sea, Ross Sea, Western Pacific, Indian Ocean) with the UG δ¹⁸O and d-excess stacks, the only (weak) positive correlation is found between UG d excess and SIE in the Indian Ocean sector (r=0.32, p<0.1; Table 2). However, this correlation does not necessarily indicate predominant moisture transport from the Indian Ocean sector towards UG. Cluster analysis and seasonal frequency distribution of backward trajectories performed with the HYSPLIT model (Figs. 10 and 11) suggest that the Weddell Sea sector, including the Ronne–Filchner Ice Shelf, is the dominant source region for precipitating air masses reaching the UG site (46 %) throughout the year, closely followed by Coats Land and DML (35 %). The latter finding is consistent with the observation of a positive correlation between the UG δ¹⁸O stack and ERA-Interim near-surface air temperatures in these regions (Fig. 9a; see above). Nevertheless, during winter (JJA) and spring (SON) a small percentage of precipitating air masses reaching UG might also originate from the Indian Ocean sector (Fig. 11), supporting the findings from cross-correlation analysis (Table 2).

Figure 10Cluster analysis of 5 d backward trajectories transporting air masses towards the UG region (red star) as calculated for all days with precipitation events (in total 121, represented by thin grey lines) for the period 2010–2015 using the model HYSPLIT. Bold coloured lines represent the main transport pathways calculated as a percentage (%) of all trajectories associated with the respective cluster.

Figure 11Seasonal frequency distribution of single 5 d backward trajectories during precipitation events extracted from ERA-Interim time series of daily precipitation for the UG region (red star) for the period 2010–2015. In total 121 precipitation events with ≥1 % of the mean annual snow accumulation (∼2.5 mm d⁻¹) were considered for this analysis performed with the HYSPLIT model.

Spatial correlations with SIC yield a more specific regional picture of the interplay between sea ice distribution and UG moisture sources. We found that only SIC in the northern Weddell Sea exhibits a strong negative correlation with mean annual δ¹⁸O and d-excess values in the UG region (up to $r<-\mathrm{0.6}$, p<0.05; 1980–2014; Fig. 12a and b), corroborating the results of the backward trajectory analysis. Thus, higher or lower δ¹⁸O and d-excess values at the UG site correspond to a reduction or increase in SIC in the northern Weddell Sea, respectively. Consequently, reduced SIC in the northern Weddell Sea implies enhanced availability of proximal surface level moisture (Tsukernik and Lynch, 2013), which would support higher δ¹⁸O (δD) values in the UG region during low-SIC phases. For a detailed interpretation of the negative correlation between SIC and UG d excess, further data on the moisture sources' relative humidity are needed, which are not yet available.

Figure 12Spatial correlations (r) of mean annual sea ice concentrations in the Weddell and Bellingshausen Sea sectors with standardized mean annual (a) δ¹⁸O, (b) d excess and (c) snow accumulation in the UG region (red star) for the period 1980–2014 (excluding firn core GUPA-1). For the calculations all time series were detrended and December–November annual averages (calendar year) were considered. Only statistically significant correlations (p<0.05) are shown.

4.3 Spatial and temporal variability of accumulation rates and relation to sea ice and climate modes