EGRIP is a deep ice core drilling project located on the Northeast Greenland Ice Stream (NEGIS), the largest ice stream in Greenland . The drill site is located at 75∘37.820∘ N and 35∘59.556∘ W, 2704 m a.s.l., 440 km to the southeast of the NEEM site. The ice flow velocity at the drill site is ∼55 m yr-1 . To date, 2418 m of ice has been drilled and partly processed, with ∼250 m remaining to bedrock.

Ice grain size was measured at EGRIP on discrete samples every 5–15 m of depth. The 55 cm samples were cut into six samples (parallel to the core axis) with dimensions of ∼90×70×0.3 mm. Samples were polished and measured with an automated G50 fabric analyser (Russell Head type ). Details about the procedure and data processing are found in .

Visual stratigraphy measurements, i.e. line scans, were conducted in the EGRIP trench shortly after the core retrieval to avoid signal alteration , with some buffer time in between to account for differences in atmospheric pressure. Brittle zone ice was measured a year after retrieval to lower the risk of ice breaking during processing.

Processed line scan data are available from 13.75 to 2120 m depth . The instrument used was a Schäfter+Kirchhoff GmbH line scanner, developed in cooperation with the Alfred Wegener Institute Helmholtz Centre for Polar and Marine Research (AWI) and the University of Copenhagen (details in ). Slabs of ice were polished from both sides and illuminated from below (“dark field” imaging). The light is scattered by solid impurities, fractures, and bubbles and directed back into the camera making them visible. In glacials, the main scattering objects are cloudy bands and fractures.

The remaining pieces of the fabric analyser samples were cut into cubes of ca. 2×2 cm (Table ) and polished from two sides with a microtome to enable successful microstructure mapping and Raman spectroscopy analyses .

Raman spectroscopy measurements were conducted in a cold lab (-17 ∘C) at AWI in Bremerhaven, Germany. The spectrometer, excitation laser, and control unit are located at room temperature close to the cold lab, which contains the microscope unit. A 100 µm fibre was used for good signal intensity and confocality. We used a WITec alpha300 M+ combined with a Nd:YAG laser (λ=532 nm) and a ultra-high throughput spectrometer (UHTS) 300 spectrometer with a 600 grooves per millimetre grating. The pixel resolution is <3 cm-1 and the spectral range is >3700 cm-1. A Hg/Ar spectral calibration lamp was used to calibrate the system. Spectra were background corrected and identified using the RRUFF database and reference spectra e.g..

Following Raman spectroscopy analysis at AWI, the samples were transported via a commercial freezer transport to Ca'Foscari University of Venice. Micrometre-resolution LA-ICP-MS 2D imaging was performed with a set-up comprised of an Analyte Excite ArF excimer 193 nm laser (Teledyne CETAC Photon Machines) with a HelEx II two-volume ablation chamber. The ablated material is transported to an iCAP-RQ quadrupole ICP-MS (Thermo Scientific) via a rapid aerosol transfer line. Samples surfaces are cleaned and polished with ceramic ZrO2 blades (American Cutting Edge, USA) before placing them on a cryogenic sample holder. Before each measurement, 1–2 pre-ablation runs are conducted with an 80×80 µm spot to clean the surface of interest before the analysis. Before and after each measurement one scan line on a standard (NIST glass SRM 612) is acquired to monitor potential instrumental drifts. We used laser spot sizes of 35 and 20 µm on the following samples analysed earlier with Raman spectroscopy: S2, S4, S7, S8, S10, S11, and S12 (Table ). We followed a newly refined multi-element imaging method including dust-related elements as described in . In order to consider species with mostly soluble and mostly insoluble behaviour, we measured the following analytes: 23Na, 24Mg, 27Al, 48Ti, 56Fe and 88Sr.

Mean grain size decreases from 3.6 mm2 at 1340 m depth to 1–2 mm2 by 1800 m (except 2–2.7 mm2 at 1600 m depth) (Fig. ). The mean grain size is at its minimum between 1500 and 1700 m, correlating with the Last Glacial Maximum (LGM). Below this point, values increase and spread with depth and are between 0.8 and 5.2 mm2. Analysed thin sections contain between 471 and 4550 grains.

The greyscale analysis (Fig. ), i.e. the brightness variations of line scan images, is a proxy for the solid particle concentration and shows fluctuations following the glacial stadials and interstadials. The data enable classifying the brightness of cloudy bands: <100 is dark, 100 to 150 is medium dark, around 150 is medium, 150 to 200 is medium bright, and 200 to 255 is bright.

We classified 1267 cloudy bands into seven types (Fig. ). The type single is the most abundant one making up 29 % of all cases (Table ). The types up and bottom, i.e. a bright layer at the top or bottom of a homogeneous layer, add up to 15 % and 17 %, respectively. The types confined (9 %), heterogeneous (10 %), and centre (8 %) make up another third of all cloudy band types. Rarest is the thick homogeneous (3 %) type. Almost 10 % could not be distinguished clearly (unknown).

Throughout the analysed depth, the relative distribution of cloudy bands varies per 55 cm section, especially for single, homogeneous and heterogeneous types (Fig. a). Single cloudy bands occur more often with depth and range from 5 % to above 50 % of identified types per 55 cm ice core. The types up, centre, and bottom are fairly constant. The type unknown occurs more frequently in deeper ice (Fig. a) as layers thin and categorisation becomes more difficult. Furthermore, the darkness of the image plays a role. For consistency, throughout the glacial the brightness is kept constant for all images, thus making our analysis more favourable for stadials, i.e. cold periods with a higher dust concentration, within the Last Glacial Period.

We identified 989 and 278 cloudy bands in Greenland stadials and interstadials, respectively. Single dominates both period types (if identifiable) with a similar relative abundance (Fig. b). Especially bottom, heterogeneous, and confined cloudy bands are more common in stadial ice. However, 27.3 % of cloudy bands in interstadials could not be identified clearly (type unknown) compared to 4.6 % in stadials.

We observed strong differences in the microstructural localisation of the analysed elements (Figs. , ). Soluble Na is usually located at grain boundaries. However, in samples containing strong cloudy bands, Na can also be located in the grain interior on some occasions. Mg is predominantly located in the grain boundaries but also occurs infrequently in the grain interior. Al is the prime indicator of insoluble particles and is found commonly in the grain interior where it often accumulates in particle clusters, i.e. containing several pixels, with additional presence of Ti and Fe. The latter two also show a comparatively weaker presence at grain boundaries, potentially indicating a soluble component. It is important to note that mass 48 (48Ti) can also contain a small contribution of Ca, which would also have a substantial soluble component.

Constituting the most important finding, cloudy bands can be clearly distinguished in the chemical images through the presence of insoluble particles mostly indicated by Al, Ti, and Fe at locations that are consistent with the findings from cryo-Raman analysis (S2 and S10 in Figs. , ). Samples with distinct cloudy bands, i.e. S8 and S10, also show a higher amount of impurities in the grain interiors than shallower samples with less distinct cloudy bands, i.e. S2 and S4. The images generally show a high degree of heterogeneity in elemental ratios, indicating a non-homogeneous composition of particle clusters, analogous to the findings by .

Mean grain sizes are relatively constant with depth until 1700 m except for the larger grains at 1600 m (Fig. ). This evolution may be due to sub-grain formation resulting in smaller grain sizes than in shallower ice e.g.. The steady increase in grain size and variability between 1700 and 2121 m is probably due to grain growth and dynamic recrystallisation. Climatic features are visible in the grain size in more detail than in previous studies e.g.. Grain size changes strongly in Termination 1 (∼15–11 ka b2k) (Fig. ) and is minimal in the coldest phase of the Last Glacial, i.e. the LGM (∼31–16 ka b2k), likely due to hampered recrystallisation by ice lattice defects caused by the high amount of dust particles and Zener pinning of grain boundaries by particles e.g.. Thus, mean grain area and greyscale brightness are often anti-correlated (Fig. ), displaying the potential impact of high insoluble content on grain size .

The grain size evolution at EGRIP is similar to the NEEM ice core (Fig. ) but shifted slightly upwards. This could be related to the different boundary conditions (temperature, elevation) at the drill sites, the impact of extensional deformation and high strain inside NEGIS and the hard shearing in flow direction , and the strong dynamic recrystallisation observed at EGRIP . However, the grain size evolution within both cores is still comparable, and the fast-flowing ice within NEGIS does not (yet) intensely affect the grain size in the upper 2121 m (∼80% of ice sheet thickness).

Our approach combines data from three methods for the first time. Hence, data integration is far from straightforward. We applied the following structure: (1) a visual characterisation of cloudy bands via visual stratigraphy, (2) an analysis of the localisation and mineralogy of insoluble particles in these defined bands with Raman spectroscopy, and (3) a detailed characterisation of the total impurity content of specific areas with LA-ICP-MS 2D imaging. Integrating chemical data yields results with higher concentrations of impurities (especially insoluble ones) in cloudy bands concurrent with visual data. We further show that insoluble particles are particularly abundant in cloudy bands, strengthening and extending the first steps in combining Raman spectroscopy and LA-ICP-MS 2D imaging . Overlapping LA-ICP-MS with microstructure mapping and Raman spectroscopy data displays that element intensities are higher in cloudy bands than around them (Figs. , ). Additionally, particle clusters with high Fe and Ti intensities are often located where Fe- (magnetite, hematite) and Ti-bearing minerals (rutile, anatase, and titanite) were observed (Figs. , ), supporting the identification of relatively inconspicuous spectra interfered with by the overlying ice spectrum (e.g. magnetite) via Raman spectroscopy. Al, Ti, and Fe vary in concentration throughout samples and are located inhomogeneously (Figs. , ), often as clustered, intra-grain inclusions. Soluble, mobile impurities (Na, Mg, Sr) are at grain boundaries. Some triple junctions display an accumulation of elements while the surroundings are comparably pure. Compared to initial LA-ICP-MS investigations based on spot sizes between 280 and 128 µm , these detailed insights can only be obtained from high-resolution imaging combined with Raman spectroscopy, expanding previous research e.g. and marking substantial progress towards a holistic characterisation of microstructural impurity localisation.

Our visual and chemical investigations show that cloudy bands are more complex and diverse than previously understood. We here discuss processes involved in their formation. Cloudy bands are commonly interpreted as storm events in spring or summer, transporting large amounts of dust from Asian deserts across the Greenland ice sheet . If cloudy bands were solely from dust deposition events in spring and summer, dark layers would dominate the visual stratigraphy. However, the images are dominated by bright layers of high insoluble particle content, as displayed by our Raman spectroscopy and LA-ICP-MS data. The greater thickness of cloudy bands compared to neighbouring dark layers could be due to the high accumulation in spring and summer in Greenland, but this does not explain the observed variation in cloudy band thickness. The abundant unknown cloudy bands at 35, 45, and 48 ka b2k (Fig. ) all originate from interstadials with low dust concentrations hampering their characterisation.

To understand the anatomy and origin of cloudy bands, the main types of dust deposition are important: (1) dry deposition, i.e. fallout of solid impurities over the ice sheet by atmospheric transportation, and (2) wet deposition, i.e. snowfall and accumulation, including washout of atmospheric dust particles.

We interpret thin and bright layers (single) as long-lasting dry precipitation events either formed by precipitation or by wind redistribution of surface snow, similar to . further suggest that very thin and bright cloudy bands are associated with enhanced scavenging early in snowfall events or with dry deposition. A single thin and bright layer indicates dry deposition (Fig. a), as the layer thickness does not increase by accumulation. Still, dust particles are deposited on the same surface. The greater the number of deposited particles in a single layer, the brighter it appears. Some thin bands show a distinct concentrated layering of insoluble elements and specific minerals, such as hematite and carbonaceous particles, and insoluble elements (Al, Fe, Ti) (S10 in Fig. ). At low wind speeds, i.e. no redistribution of snow, the thin, dry deposition layer is covered by wet precipitation adding new snow (Fig. b). This leads to an increasing layer thickness, resulting in a thick cloudy band (Fig. c) with roughly evenly distributed impurities (e.g. S1, S5, S11, S12 in Figs. , ). Snowfall events last a few days, and their sequence thus creates alternating patterns of bright and dark layers (Fig. g).

Post-depositional surface processes, such as wind-driven redistribution of surface snow, can lead to well-mixed surface snow layers containing a mix of previously deposited impurities (e.g. S5, S7, and S9 in Fig. ). Thus, bright and dark layers are mixed into homogeneous and moderately bright layers (Fig. d and e). Snow layers impacted by wind redistribution (Fig. e) could lead to a homogeneous layer (up, centre, bottom, and confined). Further, wind dunes creating high-density snow layers could serve as lids isolating high-impurity layers and creating a distinct type of cloudy band (Fig. f), but information on these layers in Greenland is rare. The redistribution and mixing of surface snow by drifting (<2 m above ground level) and blowing (>2 m above ground level) seem to be the main parameters in creating the seven distinguished cloudy band types.

Redistribution of surface snow does not impact variations between seasonal signals in ice cores, as only the seasonal snow is mixed. Mixing snow between different years seems unfeasible, as accumulation in Greenland is high enough for layers to become covered and preserved quickly. Yet this mixing affects the detailed analysis of small-scale distributions of impurities. A single thin cloudy band contains the insoluble particles from dry deposition, maybe including some precipitation, but not significant redistribution by surface processes such as wind. Thick homogeneous cloudy bands are usually dimmer than single thin cloudy bands because their impurity concentration is inhomogeneous due to redistributed surface snow. Dark layers do not contain many visible impurities and thus represent autumn or winter precipitation unmixed with spring or summer layers.

Single bands mostly appear together with dimmer cloudy bands (up, bottom, centre, confined, or heterogeneous). Assuming that a stack of bright layers between two dark layers, i.e. our definition of a cloudy band, represents 1 year, then the most common case is one bright layer per year (single, up, bottom, and centre). In some cases, we find multiple thin and bright cloudy bands within the stratigraphy of 1 year (confined and heterogeneous). The thin cloudy band is missing in other cases (homogeneous). Assuming a constant influx of dust particles over the ice sheet on an annual scale with seasonal variations, the absence of a single thin cloudy band could be a proxy for years with strong surface snow redistribution processes. Cloudy band intensity thus indicates the chronology of events on the ice sheet.

We show the value and future potential of a multi-method and multi-scale approach to better understand the development of cloudy bands. To our knowledge, continuous visual stratigraphy data are rare and only available for the EDML ice core , potentially enabling a similar study of Antarctic ice. Comparing our results with cloudy bands in NEEM, NGRIP, and RECAP ice is promising. However, the impact of inclusion size, shape, and chemistry on the brightness in the visual stratigraphy needs more research to clarify if deriving more information on impurities from visual stratigraphy is possible. Future LA-ICP-MS applications could be analysing (1) larger areas (several centimetres) covering entire cloudy bands and their surroundings and (2) at higher resolution (1–5 µm) to distinguish analytes in the observed impurity clusters more clearly. Finally, correlating cloudy band types, chemistry, millimetre-scale grain size, borehole deformation, and c-axis orientation data would enlighten us about internal deformation within the EGRIP ice core.