1. Introduction
Polar ice cores can provide annually resolved proxy records of past atmospheric temperature, composition and circulation changes for thousands of years when instrumental records are not available. Ice core studies have enabled us to better understand human-induced warming on a global scale and the non-linear behaviour of the climate system at multiple timescales. However, the reliability of any palaeoclimate data is most dependent on the accuracy of its chronology and constructing a robust age model is essential for interpreting past climatic records from polar ice cores (Legrand and Mayewski, Reference Legrand and Mayewski1997; Sinclair and others, Reference Sinclair, Bertler and Trompetter2010). Considering the large spatial variability in Antarctic snowfall, the local estimation of snow accumulation rate is vital for estimates of the coastal Antarctic mass balance (Krinner and others, Reference Krinner, Magand, Simmonds, Genthon and Dufresne2007), assessments of regional climate models, validations of satellite measurements (Stenni and others, Reference Stenni2000; Goursaud and others, Reference Goursaud2017) and eventually deciphering interactions between the Antarctic ice sheet, atmosphere and ocean (Stenni and others, Reference Stenni2000; Sinclair and others, Reference Sinclair, Bertler and Trompetter2010).
Multiple and complementary methods are often used to establish more reliable ice core chronologies. When chronology for the past few centuries is established in coastal Antarctica, these methods are mainly a combination of annual layer counting, use of time markers and correlation with other dated time series, occasionally complemented with glaciological modelling (Stenni and others, Reference Stenni1999; Zhang and others, Reference Zhang2006; Thamban and others, Reference Thamban, Naik, Laluraj, Chaturvedi, Ravindra, Sinha and Ravindra2013). High snow accumulation sites in the coastal region offer excellent opportunities to reconstruct sub-annual to annual climatic records by counting annual layers of stable water isotopes (Naik and others, Reference Naik, Thamban, Laluraj, Redkar and Chaturvedi2010; Thamban and others, Reference Thamban, Naik, Laluraj, Chaturvedi, Ravindra, Sinha and Ravindra2013; Philippe and others, Reference Philippe2016). However, there are many complicating factors like the damping effect due to diffusion processes during firnification, especially where the snow accumulation rate is lower than 0.2 m w.e. (Johnsen, Reference Johnsen1977). More importantly, summer surface melting in coastal Antarctica can induce post-depositional effects in the stable isotope records. Therefore, a multiproxy approach is preferred for dating an ice core, generally using a combination of stable isotopes and chemistry records. Such an approach has limitations for coastal sites, which are influenced by storm activities and summertime melting, making the annual layer counting using stable isotope and chemistry records very complicated.
Visual stratigraphy (VS) obtained by line-scan imaging is commonly acquired information in many ice core studies. Light scattering is caused by micro-inclusions, such as solid particulate matter and firn structure (or air bubbles enclosed in the ice matrix at greater depths), and its characteristics depend on the size and concentration of these inclusions, which may change with the seasons, resulting in dark and bright bands. The VS can be acquired using many techniques ranging from simple ocular inspection to digital scanning at high resolution, using a variety of different light sources (Kameda and others, Reference Kameda1995; Alley and others, Reference Alley1997; Pohjola and others, Reference Pohjola2002; Takata and others, Reference Takata2004; Svensson and others, Reference Svensson2005; McGwire and others, Reference McGwire2008; Abram and others, Reference Abram2013; Morcillo and others, Reference Morcillo, Faria and Kipfstuhl2020). The high-resolution images obtained using digital scanning can differentiate the annual layers even at depths below the bubble hydrate transition zone (Svensson and others, Reference Svensson2005; Svensson and others, Reference Svensson2008; Winstrup and others, Reference Winstrup2012). Although VS has been extensively studied in ice cores from the Arctic and a few inland sites of Antarctica, studies from coastal Antarctica are limited. No studies have utilised VS as a proxy for annual layer counting of the firn section of ice cores from coastal Antarctica, where the climatic variability is higher due to the maritime influence and frequent episodes of extreme precipitation events, which in turn results in noise in the proxy records. These sites are also characterised by high-accumulation rates and occasional to frequent melt features, which add complexities to the VS record, making it harder to obtain a clear annual signal. While micro-inclusions are the significant source of variability within VS data in deeper core sections, gradual densification processes and summer melting also contribute to VS data changes in the upper firn sections (Sjögren and others, Reference Sjogren2007; Kinnard and others, Reference Kinnard2008). Summer melting is significant in the coastal regions of Antarctica (Das and Alley, Reference Das and Alley2005; van den Broeke, Reference van den Broeke2005; Kaczmarska and others, Reference Kaczmarska2006; Tedesco and Monaghan, Reference Tedesco and Monaghan2009; Lenaerts and others, Reference Lenaerts2017). Although there are several uncertainties related to the transport and refreezing of meltwater within the snowpack, the refrozen ice can be easily differentiated from the surrounding firn and ice in the VS profiles as they appear as dark and transparent layers with few or no bubbles. While surface melting is affected by numerous factors like surface albedo and cloud type, summer temperature is the primary factor responsible for melting the glacial surface (Winski and others, Reference Winski2018). Melt layers in ice cores provide a valuable record of the summertime warming trend over recent times (Kaczmarska and others, Reference Kaczmarska2006; Kelsey and others, Reference Kelsey, Wake, Kreutz and Osterberg2010). Thus, a record of melt history from a coastal site can be a proxy for summer temperature trends in the past and complement surface air temperature reconstructions using water-stable isotopes. In this study, we examined VS profiles of a high-accumulation firn core from coastal Dronning Maud Land (DML) to assess the suitability of VS method for annual layer counting in the snow and firn sections of coastal Antarctic ice cores.
2. Materials and methods
2.1 Ice core
A new 122 m long firn core IND-36/B9 (hereafter IND36/9) was retrieved during the 2016–17 austral summer from the summit of the Djupranen Ice Rise (70.18° S, 9.18° E; elevation 321 m a.s.l.), at the western margin of the Nivlisen Ice Shelf, the central DML coast in East Antarctica (Fig. 1). The DML coast is characterised by distinct topographic features like ice rises having their specific local flow, climate regime and surface mass-balance variability (Lenaerts and others, Reference Lenaerts2014; Drews and others, Reference Drews2015; Matsuoka and others, Reference Matsuoka2015; Goel and others, Reference Goel, Brown and Matsuoka2017; Rignot and others, Reference Rignot2019; Pratap and others, Reference Pratap2021). The coastal DML also offers opportunities for reconstructing annually resolved decadal to multi-decadal variability in temperature, westerly winds, sea ice and their interactions with various regional climatic modes (Naik and others, Reference Naik, Thamban, Laluraj, Redkar and Chaturvedi2010; Philippe and others, Reference Philippe2016; Goel and others, Reference Goel2020; Laluraj and others, Reference Laluraj, Rahaman, Thamban and Srivastava2020; Thamban and others, Reference Thamban, Rahaman and Laluraj2020; Ejaz and others, Reference Ejaz, Rahaman, Laluraj, Mahalinganathan and Thamban2021; Pratap and others, Reference Pratap2021).
Fig. 1. Regional settings of the IND36/9 core site (green circle) at the summit of the Djupranen Ice Rise in coastal DML. The background image is a hillshade extracted from the Reference Elevation Model of Antarctica (Howat and others, Reference Howat, Porter, Smith, Noh and Morin2019), with grounding line (Mouginot and others, Reference Mouginot, Scheuchl and Rignot2017) and ice shelf fronts digitised from Radarsat-2 imagery taken between 2012 and 2014 (Goel and others, Reference Goel2020). This map was made using Quantarctica (Matsuoka and others, Reference Matsuoka2021).
2.2 Ice core processing and line scanning
For this study, we used the top 50 m of the firn section of the IND36/9 ice core. The ice core was cut into 3.5 cm thick, 10 cm wide slabs for line scanning. Both sides of these slabs were polished using a pre-cleaned handheld microtome blade to remove surface undulations resulting from cutting the ice slab using a bandsaw. The VS measurements were done using an intermediate layer core scanner (ILCS), which was designed to perform a transect of a planar ice core at a predetermined depth, providing an intermediate layer scan of the ice core (Krischke and others, Reference Krischke, Oechsner and Kipfstuhl2015). The ILCS works on the principle of dark field microscopy with two LED line illumination sources of wavelength 640 ± 20 nm. Dark field is a method of illuminating a sample with an oblique light source, so the camera only records the light scattered through the sample. The direct light from the light source falls outside the camera's field of view and therefore does not appear in the image, thus resulting in a dark, almost black background with bright objects. The two illumination sources are placed below the polished ice core slab at an incident angle of 45°, and the camera is placed above the sample to measure obliquely forward scattering (Fig. 2). Areas of the ice core with a high concentration of micro-inclusions will scatter more light and be recorded as bright bands, whereas areas of clear ice like melt layers appear dark. While the core sample was fixed, the 2048-pixel monochrome 8-bit line-scan camera and light sources were moved synchronously along the ice core using a computer-controlled motorised unit. The maximum scan length is 1200 mm with a single scan and 1700 mm with two overlapping scans, having a scan width of 105 mm and an effective resolution of 51 μm pixel−1. The length of the overlapping scans is restricted to the length of the ice core tray (1750 mm). The image data are transferred via a Gigabit Ethernet interface to SKan-G-ILCS software used for image acquisition and inspection (Krischke and others, Reference Krischke, Oechsner and Kipfstuhl2015). The top 1.67 m of the core is composed of low-density firn and was too fragile to fit into the core tray and was therefore not scanned. After line scanning, the ice slabs were cut at 5 cm resolution for stable isotope and chemical analysis. The samples for chemical analysis were cut into cuboids; the three dimensions of the samples were measured using a calliper and weighed using a weighing balance. The density for the samples was calculated as mass divided by the volume of each sample. The error in measuring the sample dimensions was ±0.5 mm, while the weighing balance's uncertainty was ±0.1 g for measurements up to 100 g. As a result, the density measurements have a propagated uncertainty of 5%.
Fig. 2. (a) ILCS line-scanning set up at the Ice Core Lab of the National Centre for Polar and Ocean Research. The bright object in front of the person is the ice core being scanned. (b) Schematic cross-sectional representation of ILCS imaging technique. The line-scan camera above the ice slab moves synchronously (perpendicular to the page) with the two line-LED illumination sources mounted below the slab at an incident angle of 45°.
2.3 Image processing
The monochrome line-scan camera records 8-bit greyscale images of the ice core section (Fig. 3a). The intensity of the scattered light, recorded as pixel intensity values with the 8-bit resolution, ranges from 0 (black) to 255 (white, saturated). Theoretically, the pixel intensity value of a section of clear ice should be zero, which is rarely observed in our study. All image processing was done using MATLAB (version 2018a, The MathWorks, Inc., Natick, Massachusetts, United States), as summarised below. To remove the image section outside the core sample, we first assign a value of 255 to pixels with a value greater than half of the mean value of all pixels in the image (Fig. 3b). A 7 × 7 structural element was then used to erode and then subsequently dilate the greyscale image morphologically. This greyscale image was then converted to a binary image, and the holes were flood-filled. A 250-pixel disc structuring element was then used to erode morphologically and then dilate the binary image to remove any objects smaller than 250 pixels in width, resulting in a mask of the ice core section (Fig. 3c). Finally, the central 7 cm wide section of the sample object was extracted (Fig. 3d). The image contrast was enhanced using contrast-limited adaptive histogram equalisation (CLAHE) with exponential distribution (Fig. 3e). The CLAHE is a variant of adaptive histogram equalisation in which contrast amplification is limited to reduce problems of noise amplification (Magudeeswaran and Singh, Reference Magudeeswaran and Singh2017). Breaks in the ice core sometimes appear as saturated pixels, and these regions were excluded from further analysis. A total of 1.5% of the total ice core length was thus excluded. We used the arithmetic mean of grey values along the width of the core sections as a parameter for further analysis.
Fig. 3. Line-scan image processing steps from the raw image (a) to the final image (e) are presented for a core sample from 48.1 to 48.97 m depth. The contrast of the final image is exaggerated to make the annual layer visible, and the approximate seasonal positions are labelled. The core section is 87 cm long and 10 cm wide. The identical greyscale is used for all panels. See text for image processing procedures corresponding to individual panels. Two minor melt layers are identified in (e) and indicated with red arrows.
A continuous VS profile was merged for the 50 m length of the ice core. Considering the sub-millimetre-scale (51 μm pixel−1) scanning of ILCS, the VS profiles have the potential to provide an extremely detailed record of snow accumulation and sea-salt inclusions. Due to the inherent noise involved in such ultra-high-resolution data of line scanning and to make it comparable with the δ18O records, the VS profiles were downsampled at 5 cm resolution by calculating the mean of greyscale values for every 5 cm intervals at which discrete isotopic and chemical analysis were performed. This enabled us to enhance the signal-to-noise ratio and improve confidence in annual layer counting.
3. Results and discussion
3.1 Preliminary chronology and age constraints
A preliminary chronology for the IND36/9 ice core was developed using annual layer counting of the δ18O record, using winter troughs in the stable isotope record to identify the potential annual layers (Fig. 4a). Next, this preliminary chronology was compared with the non-sea-salt sulphate ${\rm ( nssSO}_4^{2-} = {\rm SO}_4^{2-} -0.25{\rm N}{\rm a}^ + )$
records of the core to identify the well-known volcanic eruption events. The δ18O measurements were made using a Triple Isotope Water Analyzer (TIWA-45EP, Los Gatos Research), with an external precision of ±0.1‰ (1σ), while ionic concentrations were measured using an ion-exchange chromatograph (Dionex ICS-5000) according to the method detailed by Thamban and others (Reference Thamban2010).
Fig. 4. Establishment of chronology for IND36/9 core using multiple proxies. (a) A preliminary age model using δ18O records. Annual layers based on the winter minima of the δ18O record are marked with red dots and red dotted lines. (b) Measured ${\rm nssSO}_4^{2-}$
records (black) exceeding the detection threshold (red curve) are considered potential volcanic events. Identified volcanic peaks (yellow bands and ages) and the tritium anomaly attributed to the atomic bomb testing of 1961 (blue band) were used as age tie points to constrain the chronology. The final chronology (c) is obtained by annual layer counting, taking into account the seasonal variability in δ18O, ssNa+, ${\rm nssSO}_4^{2-}$
, ${\rm NO}_3^-$
and ${\rm NH}_4^ +$
. Grey bars represent the annual layers from StratiCounter. Note that the y-axes of the chemistry records have been cropped to show maximum variability, and the cropped peaks are marked with breaks to differentiate them from missing data points. (d) Seasonality of chemical proxies and δ18O for the whole studied section. The black dot shows the median, and the black triangles show the 95% confidence interval of the median. Vertical bars show the interquartile range.
Identification of volcanic eruptions with absolute certainty in ice cores from coastal regions is challenging due to the highly variable ${\rm nssSO}_4^{2-}$
background signals attributed to the presence of a natural background caused by marine biogenic sulphur deposition (Philippe and others, Reference Philippe2016). To separate this natural background, we used the detection threshold method of Sigl and others (Reference Sigl2013) using a 100 sample running median (RM) and median absolute deviation (MAD) of the whole record (Fig. 4b). RM and MAD are considered a robust measure of natural background as they are not strongly affected by the presence of outliers, as compared to mean and standard deviation. The detection threshold is thus defined as the sum of RM and three times MAD, and any ${\rm nssSO}_4^{2-}$
value exceeding this threshold is considered a potential volcanic eruption peak. While multiple values exceeding this threshold were observed in the core, we used only four volcanic marker events that are extensively used in Antarctic ice cores. Accordingly, we identified the ${\rm nssSO}_4^{2-}$
anomaly peaks related to Pinatubo/Cerro Hudson (1991), El Chichón (1982), Agung (1963) and Cerro Azul (1932) volcanic eruptions in the top 50 m of the ice core (Fig. 4b). Previous studies have shown that the ${\rm nssSO}_4^{2-} \;$
records in ice cores can have a delay of 1–2 years from the time of the volcanic eruption, but this delay may be different for different volcanic events and core locations (Li and others, Reference Li, Xiao, Sneed and Yan2012; Sigl and others, Reference Sigl2013; Emanuelsson and others, Reference Emanuelsson, Thomas, Tetzner, Humby and Vladimirova2022). To simplify, we assume a delay of 1 year for all identified volcanic events. To further constrain with an absolute age marker, we measured the tritium concentrations in a selected 12 m section (inset of Fig. 4b). The tritium profile shows an anomalous peak at 27.25 m depth, coinciding with the well-known tritium anomaly attributed to the atomic bomb testing of Tsar Bomba in 1961 (Cauquoin and others, Reference Cauquoin, Jean-Baptiste, Risi, Fourré and Landais2016).
3.2 Multiproxy approach and chronology validation
We used three additional ionic records (ssNa+, ${\rm NO}_3^-$
and ${\rm NH}_4^ +$
) together with ${\rm nssSO}_4^{2-}$
and δ18O records to refine the chronology between the volcanic and tritium tie points (Fig. 4c). Supplementary Figure S1b gives an expanded version of the top 20 m section of the core. A multiproxy approach is needed because the climatic conditions in the coastal Antarctic sites are highly variable, with the snowfall influenced by extreme precipitation events (Turner and others, Reference Turner2019) and repeated occurrence of summertime melting (Lenaerts and others, Reference Lenaerts2017), which can disturb the annual signals in δ18O records. Moreover, firn diffusion can also lead to smoothening of weak annual signals in stable isotope records. We, therefore, used multiple chemical proxy records for annual layer counting to account for the high degree of variability. Winters can be identified with more negative stable isotope values and low concentration of ${\rm nssSO}_4^{2-}$
, ssNa+, ${\rm NO}_3^-$
and ${\rm NH}_4^ +$
and annual layers were marked as concurrent winter troughs.
To ascertain the robustness of the chronology, we used the StratiCounter program (Winstrup and others, Reference Winstrup2012) using multiple proxies for the IND36/9 ice core. StratiCounter works on the statistical framework of hidden Markov models and the expectation–maximisation algorithm to automatically recognise annual layers in palaeoclimate archives. It requires training data, and user-defined annual layer counts. We constrained the StratiCounter chronology using volcanic events and the tritium anomaly identified in Figure 4b. Details of the StratiCounter settings are given in Supplementary section S1, and the initial set of manual layer counts used as an input for StratiCounter are shown in Supplementary Figure S1. These manual counts are used to provide a generalised framework. StratiCounter uses this generalised template for annual layers and applies an expectation–maximisation algorithm to continuously update and refine the statistical description of an annual layer, which in turn allows for changes in layer characteristics with depth (Winstrup and others, Reference Winstrup2012). To further decrease the dependence of StratiCounter on the initial manual counts, we re-run the program using an improved layer template derived from the algorithm output. Matching the volcanic events and tritium anomaly with the final chronology gives an estimated age error of ±2 years for the past century (Supplementary Table ST1).
We then use this chronology to examine the seasonality of the proxies (Fig. 4d). We calculate anomalies of each month's value from the annual mean value. This is repeated for all the proxies for the period 1919–2016 (Fig. 4d). Since the sampling resolution in the time domain is highly variable and identifying monthly separation in proxy data is debatable, we do not claim monthly resolution, but the median of all monthly values should represent the seasonality of the proxies. We find that the troughs in the δ18O record typically correspond to the troughs in most chemical species studied here. Although a high degree of variability is observed in the chemical species due to the close proximity of the core site to the open ocean, they are not biannual in nature.
While the accumulation history of the core site is outside the scope of this manuscript, we do observe several anomalously thick and sometimes thin annual layers. This is possibly due to the highly variable precipitation regime of coastal DML, which has been observed to result in anomalous snowfall events, like in 2009, which was recorded using both satellite records (Boening and others, Reference Boening, Lebsock, Landerer and Stephens2012) as well as station-based observations (Gorodetskaya and others, Reference Gorodetskaya2013). Regional climate models like RACMO2 also captured this event in their simulations (Lenaerts and others, Reference Lenaerts2013). A similar degree of variability in snow accumulation record from an ice core has also been observed at a nearby site in the Fimbul Ice Shelf (Kaczmarska and others, Reference Kaczmarska2004), with accumulation rates varying between 0.08 and 0.58 m w.e. during 1737–2000.
3.3 Melt layer quantification and masking
Previous studies have used the melt layer's maximum thickness to calculate the total melt proportion in ice cores (Kaczmarska and others, Reference Kaczmarska2006; Kinnard and others, Reference Kinnard2008; Winski and others, Reference Winski2018). While this method may work well in understanding the temporal variability of annual melt, it is prone to overestimation of the melt proportion. In this study, we used a different approach by manually identifying and masking each melt layer in line scans using a simple algorithm for detecting connected pixels using a prescribed threshold. This image recognition algorithm comprises two primary parts, threshold assignment and pixel connectivity procedure (Fig. 5). First, an adaptive threshold is initially applied to the image by manually assigning a reference point within a melt layer (Fig. 5b). The algorithm then separates the melt layers from the background by considering the empirical distribution of the dark (melt layers) and bright (background) pixels. This results in a binary image (Fig. 5c) of potential melt layers within the core section. Second, since the threshold-based segmentation is not definitive, an eight-neighbourhood pixel connectivity operation is carried out on the binary image. A pixel Pi +1,j+1 is said to be an eight neighbour of a pixel Pi ,j if the pixels share either an edge or a vertex, while a region is said to be eight-connected if every pixel in the region can be reached by a combination of moves in the two vertical, two horizontal and four diagonal directions. A mask is created that includes all pixel values that are eight-connected to the reference point and have a grey value within a tolerance range, where tolerance (scalar value) defines the range of pixel values to be included in the mask (grey value of reference point ± tolerance). The tolerance is by default set to 32, but this is adjusted according to each image section, depending on how well the grey values of melt layers are resolved in them. This helps in the identification of any stray pixels segmented during the thresholding process (Fig. 5d). Finally, the holes in the binary mask are flood-filled (Fig. 5e), and the mask is morphologically eroded and then dilated to produce a smooth mask (Fig. 5f). Because the reference point and tolerance were chosen manually, the mask results may vary from iteration to iteration. To ensure accurate results, the melt masks were manually checked against raw images to ensure appropriate masking, and the procedure was repeated (Figs 5b–f) if any differences were identified. The threshold-based connected pixel recognition algorithm ensures that the morphology of the melt layer is well-preserved in the mask and since the VS record has a resolution of 51 μm pixel−1, even melt layers of sub-millimetre-scale would be morphologically well-resolved.
Fig. 5. Melt layer masking with a threshold-based connected pixel algorithm. An ice core section with two visible melts is shown in panel (a), with the red square showing the section zoomed in for detailed reference in the following panels. Panels (b–f) show individual processing steps described in the text. The green dot in panel (b) represents the user input reference point, while the red polygon represents the final mask outline. For this section, a tolerance of 40 is utilised. Inset in panel (d) shows a schematic depiction of an eight-neighbourhood (dark blue) of a pixel (light blue).
These melt layers appear as clear, dark bands of ice with or without bubbles in VS record. This causes the core section's grey values (pixel intensity) to decrease abruptly, unlike the systematic and smooth changes shown in annual accumulation layers. Our study employed a melt layer mask to exclude the melt section from ice core images while preparing the VS record. This ensures that no bias is introduced in the VS records due to a higher melt in an ice core section. The melt layer mask created for the ice core was used to prepare and quantify the mass proportion of the melted and non-melted segments (Fig. 6a). A melt layer was considered for calculating melt proportion only if it covers at least 50% width of the ice core (Abram and others, Reference Abram2013). The melt proportion is calculated as the percentage of the area of the core section comprised of melt layers by mass (assuming that the melt layer is consistent on the z-axis) to each annual layer identified using multi-proxy-based chronology (Fig. 4). The annual proportion of melt in the studied ice core varies between 0 and 4.4%, with a median melt proportion of 0.25% (Fig. 6). We define the melt proportion by the area of melt polygons, while other studies use the representative thickness of individual melt layers. This methodological difference may affect the quantification of annual melt proportion. To examine this issue, we also calculated the melt proportion using thickness and compared it with the melt proportion calculated with polygon area (Figs 6, 7). We find that while both methods provide similar temporal change patterns, the melt proportion using layer thickness is overestimated by approximately two times compared to the estimates using melt layer polygons. The higher degree of overestimation generally occurs when the melt layers are not perfectly rectangular in shape or are not consistently well-formed through the width of the core.
Fig. 6. Melt layer distribution in the ice core was calculated using melt layer polygons (a) and melt layer thickness (b). Estimated annual melt proportion (blue curve) plotted against the age. The blue dashed line shows the mean melt proportion (0.5 and 1.3% by using melt layer polygon and melt layer thickness, respectively) for the time period from 1919 to 2011.
Fig. 7. Scatter plot between annual melt proportion obtained using melt layer polygon and melt layer thickness. The dashed black line is the 1:1 correspondence line. All scatter points are restricted to the left of the black line, showing that melt proportion estimation using layer thickness is overestimated compared to the estimation from melt layer polygons.
Melt layers on polar ice sheets indicate warmer summer air temperatures and are a useful climatic proxy (Kelsey and others, Reference Kelsey, Wake, Kreutz and Osterberg2010; Winski and others, Reference Winski2018). Melt layers appear as clear dark patches and are visually closer to ice than firn. These melt layers are primarily perpendicular to the length of the core, and their thickness can vary from less than a millimetre to several centimetres along the width of the core (Fig. 8). Since most melt layers are sub-millimetre to less than a centimetre in width, they do not affect the VS record significantly. However, melt layers with thickness greater than a centimetre or the presence of multiple small melt layers at close intervals can significantly affect the VS record and create the illusion of an annual layer. We thus use the melt mask to remove all melt layer sections from the VS record. Melt frequency and melt percentage are two of the most common metrics for quantifying melt layers, as they are not affected by annual layer thinning (Winski and others, Reference Winski2018). We used melt percentage to quantify the melt layer record in this core. We suggest that using a melt layer polygon to calculate annual melt proportion instead of melt layer thickness would provide a better estimate of melt history from an ice core.
Fig. 8. Line-scan image profile of a core section of 0.65 m long and 7 cm wide. The greyscale (same as Fig. 3) highlights melt layers (marked with blue arrows) of sizes ranging from a few centimetres (first from left) to less than a millimetre (first from right). The mean pixel value along the width of the core is shown (solid red curve). A sharp fall in mean pixel values is observed when melt layers are present, even if less than a millimetre wide.
The occurrence of summertime melting is known to affect the isotopic composition of the near-surface snow. When snow particles come in contact with meltwater, isotopic fractionation occurs at the snow–melt interface. Melting in the early summer affects the highly depleted winter/spring snow, which is present at the surface. The meltwater may also percolate down and refreeze. Although a detailed analysis of the effect of surface melt on the δ18O record is beyond the scope of the manuscript, the melt features observed in this study are on average 0.5% of the annual layer, with the highest melt proportion being 4.4%. Since the melt proportion at the present core site is low and no percolation of meltwater is observed in the studied section, we conclude that the melt features do not significantly affect the use of stable isotope and chemical profiles from this ice core as a proxy of seasonal cycles. It is also safe to assume that the effect of summertime melting on the snowpack, if any, is restricted to the annual layer.
3.4 Effect of firn density on transmitted light intensity
The intensity of scattered light is maximum near the top, and it gradually decreases to less than half at 50 m depth (Fig. 9a). Firn density gradually increases over this depth range from 420 to ~840 kg m−3 (Fig. 9b). The firn density controls this systematic depth trend of the scattered light intensity, as has been previously observed by Sjogren and others (Reference Sjogren2007) and Kinnard and others (Reference Kinnard2008). This density effect is removed using circulant single-spectrum analysis (Bógalo and others, Reference Bógalo, Poncela and Senra2021). This method works by extracting the underlying signals in a time series by identifying their frequency of oscillation in an automated way by simply introducing the data and a suitable window length. We used the VS record for this purpose and a window length of ten, giving six reconstructed components (RCs). We found that increasing the window length above ten did not significantly affect the outcome and, therefore, proceeded with the same. The first reconstructed component, RC1, replicated this general decreasing trend of VS with depth (Fig. 9a) and has a clear dependence on density (Fig. 9c). We further derived a second-order polynomial transfer function between density and RC1 (R 2 = 0.92) (Fig. 9c). The highest order component, RC6, shows largely random noise. Therefore, we consider the sum of the intermediate four components, RC2–RC5, as the representative component of VS variability attributed to the seasonal variability (Fig. 9d).
Fig. 9. Depth profiles of the VS record and density. (a) Depth profiles of VS (blue curve) and the RC1 (red curve). (b) Depth profiles of the measured, 5 cm resolution density data (blue). (c) Density dependence of RC1 scattering intensity (blue dots), which is fitted with a second-order polynomial transfer function (solid red curve). (d) The reconstructed seasonal component of VS as a cumulative sum of RC2–RC5.
Variations in several parameters, such as density and micro-inclusions, influence the intensity of light scattered through the ice core section. While density in the firn section has a long-term secular trend (Sjögren and others, Reference Sjogren2007), micro-inclusions such as dust and sea-salt inclusions exhibit seasonal variability (Svensson and others, Reference Svensson2005; Winstrup and others, Reference Winstrup2012), which may be used to count annual layers. The quantity of air and air–ice contacts in the core substantially impacts the amount of light dispersed by the core. The density of the firn is likewise controlled by its air content; hence, there is a substantial link between density and the intensity of light transmitted through it. A non-linear relationship between the mean pixel intensity and density over the length of the ice cores has also been previously observed by Sjogren and others (Reference Sjogren2007) and Kinnard and others (Reference Kinnard2008). We also observe a similar non-linear relationship and find that once the trend due to the gradual increase in density is separated from the VS record, the resultant component would possibly represent the seasonal variability and can be used for annual layer counting.
3.5 VS from line-scan images as a tool for firn core chronology
To examine the potential of high-resolution VS records as a reliable tool for annual layer counting in coastal Antarctica, we first construct an age–depth model based on layer counting of the VS record and then compared the age model against the ${\rm nssSO}_4^{2-}$
profile that has volcanic events and tritium marker horizons demarcated (Fig. 10). Winters can be identified as peaks in the VS record. The processed VS record has relatively less noise, but ambiguous peaks and shoulders are still present. If all such ambiguous peaks and shoulders are identified as annual layers, it can result in significant overcounting. To ascertain the robustness of the VS age–depth model, we used StratiCounter for annual layer identification in the VS record. An initial annual layer template is provided for the algorithm, with certain and uncertain layers marked and constrained using the time markers identified in Section 3.1. The input layer template and StratiCounter annual layer outputs are shown in Supplementary Figures S2 and S3. We find that the age–depth model from VS is highly robust, with the error in the chronology being <±2 years for the past century (Supplementary Table ST1).
Fig. 10. Comparison of age models using VS record and volcanic events. (a) Depth profile of VS (sum of the RC2–RC5). Annual layers (winter peaks) using the StratiCounter program are marked with grey bars, and every fifth year is labelled. (b) Same as Figure 4b. (c) Seasonality plot of VS record plotted similar to Figure 4d. Note that the VS record uses arbitrary units (a.u.).
The seasonality of the VS record is calculated as explained in Section 3.2 and shown in Figure 10c. The reported age model error for the majority of high-accumulation ice core records from coastal Antarctica for the past century is ±2 years or greater (Isaksson and others, Reference Isaksson1999; Thamban and others, Reference Thamban2006; Naik and others, Reference Naik, Thamban, Laluraj, Redkar and Chaturvedi2010; Thamban and others, Reference Thamban, Naik, Laluraj, Chaturvedi, Ravindra, Sinha and Ravindra2013; Laluraj and others, Reference Laluraj, Thamban and Satheesan2014; Philippe and others, Reference Philippe2016). Therefore, the high-resolution VS profile studied here provides a reliable means to constrain the ice core chronology with similar accuracy to previous studies. Our study shows that VS profiles have high potential as a preliminary and supplemental age tool due to their clear seasonality (Fig. 10c), especially considering that analysis of isotopic and chemical proxies requires more time and resources. Considering the extensive summertime melting in low-elevation sites of coastal Antarctica, where annual layer counting and chronological constraints using stable isotope and chemical records are susceptible to larger uncertainties, the VS method would be advantageous when used in tandem with other proxies.
The ultra-high-resolution VS profile of the ice core also provides a detailed picture of the physical processes and environmental variability during the period. Throughout the length of the core, clear ice core layering is visible. The dark and bright horizons observed in the VS profiles are formed due to variation in micro-inclusions in the ice cores, and these micro-inclusions are known to have seasonal variations (Winstrup and others, Reference Winstrup2012). However, in the top snow/firn section of the ice core, where annual layer thickness is large, the high resolution of VS data makes it challenging to differentiate annual signals from the noise. However, with proper filtering and downsampling, it is possible to extract the annual signals from the VS profile. VS in ice cores can be challenging due to non-uniform firnification or depositional events and becomes less clear with increasing age and depth, owing to densification and other physical changes. Thus, identification of annual layers solely based on VS can be tricky due to the presence of multiple peaks within an annual layer; however, when several of the above methods are combined, such discrepancies can be identified and removed. Hence, VS provides valuable support in identifying seasonal cycles with higher confidence, resulting in a more robust chronology.
4. Conclusions
Our study demonstrated that VS records reconstructed using line-scan images from the firn section of an ice core can be successfully used to establish an independent high-resolution age–depth model. Traditional methods of annual layer counting using stable isotope and major ion chemistry records are laborious, potentially susceptible to artefacts related to summertime melting and harder to interpret for annual layer counting in the coastal region due to the close proximity to the oceanic processes and the influence of storm activities. VS profiles are proposed as an important supplemental dating tool in coastal regions of Antarctica. Also, VS profiles can be used to develop a preliminary chronology in a short analysis time, while the other chemical and isotope methods require more time and resources to acquire. The VS profile of the firn section of the ice core is dominantly affected by the temporal changes in density, but automated signal extraction procedures can separate it to distil the seasonal signals attributed to the dust and sea-salt inclusions. The line-scan images can also be used to identify and quantify the melt layers routinely noticed in ice cores from low-elevation sites of Antarctica. Such melt layers can be appropriately masked during the pre-processing of the line-scan images, enabling us to reconstruct the chronology of ice cores from coastal Antarctica. We demonstrate that quantifying annual melt proportion using only the thickness of the melt layer can lead to overestimations and propose using the melt layer polygon area for calculating annual melt proportions. The reconstructed VS profile of the top 50 m section of an ice core from coastal DML revealed an excellent match with the chronology of the ice core based on a multi-proxy approach, supported with independent age markers of volcanic events and tritium anomaly, with the highest error being ±2 years for the past century. We conclude that ultra-high-resolution VS records from coastal Antarctica could be valuable in reconstructing the chronology of ice cores.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/jog.2022.59
Data
The IND36/9 ice core records of δ18O, ${\rm nssSO}_4^{2-}$
, ssNa+, ${\rm NO}_3^-$
, ${\rm NH}_4^ +$
and VS are available at National Polar Data Centre (http://ramadda.npdc.ncpor.res.in:8080/repository/entry/show?entryid=ea609882–3911).
Acknowledgements
Ice core drilling and field expedition have been made successful with the support of the NCPOR logistics leaders, Maitri logistics team and NPI logistic personnel. Ashish Paiginkar and Tariq Ejaz are acknowledged for their contributions to ice core processing and chemical analysis. RD acknowledges University Grant Commission for a research fellowship grant and Prof. Kotha Mahender (Goa University) for valuable guidance. We thank the scientific editor Mathieu Casado and two anonymous reviewers for constructive comments and helpful suggestions, which have improved the manuscript. This is NCPOR contribution number J-17/2022-23.
Author contributions
MT and RD defined the study objectives. MT, CML, KM and BLR collected the ice core. RD led the ice core processing, line scanning and analysis with support from BLR, KM and CML. RD analysed and interpreted the data with contributions from MT and KeM. All authors contributed to the manuscript. MT and KeM were the project leaders.
Financial support
This research has been supported by the Ministry of Earth Sciences, India (grant no. MoES/Indo-Nor/PS-3/2015) and the Research Council of Norway (grant no. 248780) for the joint India–Norway project, ‘Mass balance, dynamics, and climate of the central Dronning Maud Land coast, East Antarctica’ (MADICE).
Code availability
The MATLAB codes, used for (i) pre-processing of line-scan images and (ii) quantification of melt proportion, are available on the GitHub page of Rahul Dey (www.github.com/dey-rahul).
Open access
