Microstructure, Micro-inclusions and Mineralogy along the EGRIP ice core-Part 2: Implications for paleo-mineralogy

Impurities in polar ice do not only allow the reconstruction of past atmospheric aerosol concentration, but also influence the physical properties of the ice. However, the mineralogy and location of impurities in ice and the involved processes are poorly understood. We use Continuous Flow Analysis to derive the dust particle concentration and optical microscopy and Cryo-Raman spectroscopy to systematically locate and analyse the mineralogy of micro-inclusions in situ inside eleven solid ice samples from the upper 1340 m of the East Greenland Ice Core Project ice core. Micro-inclusions are more variable in min5 eralogy than previously observed and are mainly composed of mineral dust (quartz, mica and feldspar) and sulphates (mainly gypsum). Inclusions of the same composition tend to cluster, but clustering frequency and mineralogy changes considerably with depth. A variety of sulphates dominate the upper 900 m while gypsum is the only sulphate in deeper samples, which however contain more mineral dust, nitrates and dolomite. The analysed part of the core can thus be divided into two depth regimes of different mineralogy, and to a lesser degree of spatial distribution, which could originate from different chemical 10 reactions in the ice or large-scale changes of ice cover in NE-Greenland during the Mid-Holocene. The complexity of impurity mineralogy on the metreand centimetre-scale in polar ice is still underestimated and new methodological approaches are necessary to establish a comprehensive understanding of the role of impurities.

by e.g., Kipfstuhl et al. (2006); Faria et al. (2010) focused mainly on the location of insoluble particles Scanning Electron Microscope (SEM) coupled with energy dispersive X-ray spectroscopy (EDS) (e.g., Wolff et al., 1988;Barnes et al., 2002;Baker et al., 2003;Barnes et al., 2003a) and LA-ICP-MS (e.g., Reinhardt et al., 2001;Della Lunga et al., 2014;Bohleber et al., 2020) studies also allowed the identification of impurities. They mainly found chlorine, sulphur and sodium at several 60 locations, ranging from grain boundaries to filaments and the grain interior. However, these methods are limited to the investigation of elemental concentrations. Identifying the mineralogy of inclusions is however possible with Raman spectroscopy (e.g., Fukazawa et al., 1997;Ohno et al., 2005;Sakurai et al., 2011;Ohno et al., 2014;Eichler et al., 2019) while preserving the microstructural location , a crucial aspect regarding their impact on the physical properties of the host material. Identified inclusions in Antarctic and Greenlandic ice mainly consist of terrestrial dust (e.g., quartz, feldspar, mica, hematite) and sulphates of different chemical composition (Ca, Na, Mg, K, Al) (Ohno et al., 2005;Sakurai et al., 2009Sakurai et al., , 2011Ohno et al., 2014;Eichler et al., 2019). Calcium carbonate (CaCO 3 ) was identified by Sakurai et al. (2009) mainly in the last glacial period from the Greenland Ice Core Project (GRIP) ice core. All these studies are demanding in time and costly and thus often rely on a small number of samples taken at arbitrary depths. This results in a poor understanding of the underlying processes and in a lack of identified species involved hampering the development of realistic generalisations of impurities for polar ice (Stoll 70 et al., 2021).
In summary, commonly used methods, developed to derive (paleo-) climate information on impurities such as Continuous Flow Analysis (CFA) (e.g., Kaufmann et al., 2008), apply their analysis on melted sample water for the sake of decontamination and time constraints. Doing so, information on the phase and localisation of impurities are not considered. On the other hand, dedicated studies on localisation are spatially very limited and do not catch the overall chemistry of a certain sample. However, 75 in order to develop a profound understanding on the mineralogy and localisation of impurities in polar ice, a combination of both approaches is needed.
Our approach is a combination of methods from microstructure and impurity research to perform a systematic high-resolution analysis along one deep ice core. The East Greenland Ice Core Project (EGRIP) ice core, the first ice core drilled through the Northeast Greenland Ice Stream (NEGIS), enables such an interdisciplinary undertaking. A companion study investigates the 80 location of micro-inclusions and their role regarding deformation and microstructure. In this study we investigate the mineralogy of these visible micro-inclusions. We apply bulk chemistry from CFA and optical microscopy and Raman spectroscopy on eleven samples along the EGRIP ice core. With this combined data set covering the Holocene and Late Glacial, i.e. the upper 1340 m, we aim to give a systematic overview of the evolution of the mineralogy of the EGRIP ice core by identifying a relevant number of micro-inclusions at equal depth intervals. We discuss the interpretation of the mineralogy of micro-inclusions 85 in the EGRIP ice core with respect to environmental boundary conditions and survey implications for future research.

The East Greenland Ice Core Project
The EGRIP ice core drilling project is located at 75°38' N and 35°58' W, 2704 m a.s.l. (2015) on NEGIS, the largest ice stream in Greenland (Joughin et al., 2010;. At the drill site the ice flows with a velocity of 55 m a −1 90 (Hvidberg et al., 2020) offering a unique possibility to study ice rheology and physical parameters contributing to deformation 4 https://doi. org/10.5194/tc-2021-190 Preprint. Discussion started: 6 July 2021 c Author(s) 2021. CC BY 4.0 License. such as crystal preferred orientation and impurity content. While first results regarding the physical properties of EGRIP ice were recently published (Westhoff et al., 2020) we investigate microstructure and micro-inclusions in detail in a companion paper.
The present day accumulation in water equivalent at the EGRIP drill site is between 13.8 and 14.9 cm/yr (Nakazawa et al.,95 2021) and the averaged annual layer thicknesses for the period 1607-2011 is 0.11 m ice equivalent . In the last 8 kyr before 2000 CE annual layer thicknesses were almost constant, probably due to a combination of flow-induced thinning and increased upstream accumulation (Mojtabavi et al., 2020). Traced radar layers from the North Greenland Ice Core Project (NGRIP) drill site indicate an undisturbed climatic record of at least 51 kyr .

Continuous Flow Analysis
CFA data were obtained during three measurement campaigns (2018-2020) at the University of Bern using the CFA setup 105 described in detail in Kaufmann et al. (2008) with a few minor improvements. For the analysis a longitudinal 35 x 35 mm section of the ice core was cut and melted down-core and analysed using continuous measurement methods to produce mmcm resolution records of soluble and insoluble impurities in the ice. Micro-particle concentrations where determined using an Abakus (Fa Klotz) Laser Particle Sizer (e.g., Ruth et al., 2003) operating in the size range between 1-15 µm, which covers the size range of optical microscopy. The depth resolution of the data is limited by signal dispersion in the analytical system to 110 approximately 0.5-1 cm. Depth co-registration to the thin sections is limited by the accuracy of the depth assignment in the field and is typically on the order of a few mm. We calculated the arithmetic mean dust particle concentration of each bag for an overview.

Sample preparation
Aiming to derive a systematic overview over the EGRIP core we selected samples roughly every 100 m of depth. Samples 115 were chosen based on a combination of depth-representative and prominent features in the CFA (e.g., high insoluble particle concentration), grain size (e.g., fine) and crystal preferred orientation data. The latter two are presented in a companion paper, we here focus on the chemistry. Exact regions of interests in these samples for microstructural and Raman analysis were defined using high-resolution CFA data (Fig. 1). The chosen areas included areas without prominent properties and specific areas of interest, such as high dust content. We analysed eleven samples between depths of 138.92 and 1339.75 m, taken every~110 m 120 ( Fig. 1 and Table 1). The nine shallower samples are from the Holocene and the two deepest from the last glacial termination (Mojtabavi et al., 2020).
As described in a companion paper we used the remaining ice of the physical properties samples analysed at the EGRIP camp and created thick sections following a standard procedure (Kipfstuhl et al., 2006). No silicon oil was used to avoid ar- tificial Raman spectra masking the micro-inclusion spectra. Most samples were approximately 10 x 10 mm, but dimensions 125 varied depending on the CFA data (Table 1). A Leica microtome was used to polish the top and bottom surface of the samples, which was followed by 1.5-2 hours of sublimation under controlled humidity and temperature conditions. Thus, a good sample surface quality was obtained and small-scale disturbances (e.g., microtome scratches) were erased while grain-boundary grooves became more distinct. Polished surfaces allow the localisation of micro-inclusions 500 µm below the sample surface and successful Raman spectroscopy measurements with strong signals.

Microstructure mapping and impurity maps
Microstructure mapping was performed following Kipfstuhl et al. (2006) and Eichler et al. (2017) and the detailed investigation of the microstructural location of micro-inclusions is explained in detail in a companion paper. Samples were placed under an optical Leica DMLM microscope with an attached CCD camera (Hamamatsu C5405) and several hundred individual photomicrographs were created grid-wise with a scanning resolution of 3 µmpix -1 . The creation of high-resolution maps to detect 135 micro-inclusions provided the basis for a structured Raman analysis. These inclusions are usually 1-2 µm in diameter ( Fig.   3A) and consist of droplets, dust particles or salts. We focused inside the ice using transmission light mode and different focus depths to locate micro-inclusions below the surface (Fig. 2B). Microstructure maps with indicated micro-inclusions are from now on called "impurity maps". We mapped grain boundaries on the sample surface and translated them into the impurity map ( Fig. 2) by applying a grain boundary width of 300 µm. This is an upper-limit assumption and compensates for vertically tilted 140 grain boundaries and internal light diffraction with depth (Eichler et al., 2017). Micro-inclusions inside these grain boundaries were classified as "at the grain boundary" (Fig. 2)C.

Cryo-Raman spectroscopy
The Raman effect is the inelastic scattering of light caused by the excitation of vibrational modes of crystals or molecules.
This results in the Raman shift -a loss of scattered light energy, which is specific for each vibrational mode. The Raman 145 shift is used to identify chemical impurities in samples in a non-destructive way -Raman spectroscopy. It is well suited for light-transparent materials, such as ice. Using confocal optics, it is possible to focus the excitation laser on a small volume segment (few µm 3 ) inside the sample (Fig. 3A). The same confocal optics collect the backscattered light, which is conducted into a spectrometer to resolve the spectral distribution. Due to the low probability of a photon being Raman-scattered long integration-times, high incident light intensities and sufficient particle concentration is needed. To avoid melting, laser power 150 is limited and thus, well-prepared sample surfaces and precise focusing of the laser is essential to obtain identifiable spectra  When using Raman spectroscopy on micro-inclusions inside ice, the ice spectrum is always present (Fig. 3B-D). Other factors hampering clear Raman signals are: 1) small sizes of micro-inclusions, 2) laser beam path length, 3) quality of sample surface, and 4) acquisition time. Normally, one micro-inclusion was measured ten times with an acquisition time of 0.5-2 s per 155 spectrum, resulting in a total acquisition time of at least 5 s. Weak signals were remeasured with up to 3 s per spectrum and thus, up to 30 s in total. The potential cumulative heating of the micro-inclusion limits the total acquisition time, because it could result in melting of the sample or the destruction of the inclusion.
We measured in a maximal focus depth of 500 µm below the surface to obtain high-quality signals (Fig. 2B). Surface pollution and possible reactions or rearrangements of elements on the surface were avoided by focusing into the sample, thus only analysing inclusions in the ice. Nevertheless, this leads to a permanent superimposition of the ice spectrum. Impurity peaks were well distinguishable in most cases, even though some signals with strong ice-vibrational bands were hidden by the Raman-active ice spectra. This is often the case for O-H signals of hydrates, which are luckily shifted to higher frequencies in minerals and normally very strong -distinguishing them was usually possible.
Spectroscopy analysis was performed at the Alfred Wegener Institute Helmholtz Centre for Polar-and Marine Research, 165 Bremerhaven with a WITec alpha 300 M+ combined with a NdYAG laser (λ = 532 nm) and a UHts 300 spectrometer with a 600 grooves mm -1 grating. The used Raman system is further described by Weikusat et al. (2015). The microscope unit is located in a small cold lab inside an insulated cabin with a temperature of -15°C. The control unit, spectrometer and excitation laser are located close to the cell at room temperature and connected with the microscope. We used a 100 µm fibre to obtain a good compromise between signal intensity and confocality. If necessary, samples were re-polished in the same cabin, allowing fast 170 and controlled sample sublimation and a low risk of contamination. The small cabin volume resulted in increasing temperatures over time; measurements were stopped at the very latest at -8°C. To avoid extensive sublimation measurements were continued on the next day. Sample analysis was carried out until a maximum of~500 µm from the sample surface sublimated. However, we were able to analyse high numbers of micro-inclusions, often exceeding 100 micro-inclusion measurements per sample.

Evolution of insoluble particle number with depth
To put the samples analysed for micro-inclusions into a broader context, Fig. 1A shows a profile of 55 cm average microparticle concentrations over the entire depth range of the core. Generally, during the Holocene, particle concentrations are 180 relatively low, compared to the Preboral and Younger Dryas periods in good agreement with previous studies of micro-particle concentrations ins Greenland ice (Steffensen, 1997;Ruth et al., 2003). However, a detailed climatological interpretation of the data, including e.g., possible upstream and effects, is outside the scope of this study and will follow at a later time.
Mean particle numbers are generally consistent and vary most at a depth of 250 m. The general increase in insoluble particle numbers with depth also occurs in our chosen samples and is up to two orders of magnitude from the shallowest to the deepest 185 samples ( Fig. 1B-L).
The four shallowest samples show insoluble particle peaks between 1400 and 8000 ml −1 (Fig. 1B-E), while the highest values are observed in the three deepest samples (up to >250000 ml −1 ) ( Fig. 1J-L). There is a high variability of the insoluble particle concentration on small spatial scales and significant changes occur within centimetres due to the pronounced seasonality in the dust deposition in Greenland. Fig. 1K) and 1L) display insoluble particle numbers skyrocketing and dropping over 190 a few centimetres.

Raman spectra and derived mineralogy
We conducted Raman spectroscopy on several of the previously located micro-inclusions from all eleven samples. Analysed inclusions were chosen randomly, but with the intention to represent the entire area of the sample. The use of confocal mode and a 20x and 50x lens allowed us to find more than 97 % of the micro-inclusions indicated in our impurity maps. The number 195 of analysed micro-inclusions per sample ranged from 47 to 134 and a total of 791 spectra were identified. Luminescence was present in 131 measurements, which were excluded. Micro-inclusions with no obtained Raman signal could be Raman-inactive or the Raman signal might be superimposed by the ice signal, i.e. too low in intensity to be identified with confidence. Some analysed micro-inclusions showed Raman spectra of more than one mineral. We classified these complex micro-inclusions We identified 26 different Raman signals, which can be mainly differentiated into mineral dust and sulphates. An overview is presented in Fig. 4 and the most abundant minerals are displayed in Table 2. Mineral dust minerals are primarily silicates such as quartz and members of the feldspar or mica group. Less abundant dust minerals are hematite, rutile and anatase (both T iO 2 , rutile is the high-temperature form), titanite (CaT iSiO 5 ), jacobsite (M nF e 2 O 4 ). Sulphates can be difficult to 205 distinguish due to tiny differences in their Raman spectra, but most sulphates were identified (Fig. 4B). Most abundant are gypsum, Mg-sulphate, and bloedite. Less abundant are krohnkite, Na-, and K-sulphates. Spectra with a strong peak at 1050 cm −1 probably indicate K-nitrates while a strong peak around 1070 cm −1 might indicate Na-nitrates (Ohno et al., 2005).

Mineral diversity with depth
Between eight and 17 different spectra were identified per sample, the median value is 10 different spectra per sample. To compare our samples despite the varying amount of total identified Raman spectra per sample we calculated a diversity index 215 I var with the total amount of identified micro-inclusions per sample (n i ) and the amount of different minerals per sample (n m ): Mineralogy diversity increases slightly with depth (Fig. 5). Deeper samples are slightly more diverse in mineralogy even though the large diversity of sulphates is only found in the upper 900 m (Fig. 4B). As expected, samples with larger numbers 220 of identified Raman spectra tend to be more diverse in mineralogy supporting the use of I var . The majority of sulphates was found in the seven shallowest samples below a depth of 900 m (Fig. 4B). Over the core sulphate spectra represent between 9.4 % (1256.98 m) and 87.7 % (899.94 m) of all identified micro-inclusions at one depth.
As displayed in Fig. 4B

Small-scale: Clusters
Dense clusters of micro-inclusions (minimum of three proximal inclusions) of the same mineralogy were observed throughout the core, but especially above 900 m (e.g., at 138. 92, 514.44, 757.21, and 899.98 m). Specifically sulphates at these depths 240 were often found in such clusters ( Fig. 6 and 7A, B). The abundance of such clusters of micro-inclusions with similar chemistry declines with depth and is rarely observed below 900 m. Certain minerals, such as mica, only occur in a limited area of the sample (e.g., at 1339.75 m) without creating distinct layers or clusters. Gypsum and nitrates were found throughout the entire sample area while no specific localisation was identified for minerals such as quartz, hematite or nitrates (Table 2).

Small-scale: Horizontal layers 245
Often micro-inclusions of the same mineralogy are found in distinct layers horizontal to the core axis (distinct vertical layers were not observed) or broader bands. Fig. 7A shows that at a depth of 757.2 m mineralogy changes on the centimetre-scale. The shallowest part of the sample is characterised by a mixture of sulphates (mainly gypsum) and terrestrial dust (quartz, feldspar, mica). Some centimetres deeper a small layer of gypsum is followed by a layer of mainly Mg-sulphates. At the bottom of the sample Mg-sulphates dominate, accompanied by gypsum and mica.

250
A broad layer is observed at 899.98 m (Fig. 7B). The upper three quarters of the sample are dominated by different sulphates, especially the middle section shows a clear layer. Sulphates in this layer are often found in clusters. The deepest part of the sample inhibits a broader variety of minerals, ranging from different sulphates to mica and quartz (Fig. 7B). Below 900 m feldspar, mica and quartz were more common at grain boundaries than sulphates, i.e. gypsum. However, at 1339.75 m 37.9 % of all identified micro-inclusions at grain boundaries were gypsum.

Insoluble particles at EGRIP
We here present the 55 cm bag mean CFA results of the insoluble particle number (Fig. 1), representing the dust concentration in the ice. Though, detailed studies will follow in the future, we can already see, that the insoluble micro-particle concentration, interpreted as the mineral dust aerosol particle record, seems to vary through the Holocene. The concentration is comparable 275 to e.g., the North Greenland Ice Core Project ice core (Steffensen et al., 2008) and is low in overall concentration compared to Glacial ice (e.g., Steffensen, 1997;Ruth et al., 2003).

High concentrations of insoluble particles measured with CFA correlate with areas of numerous micro-inclusions observed
with optical microscopy and Raman spectroscopy. The samples with the highest insoluble particle counts, i.e. at depths of 1256.96 and 1339.75 m, are from the Bølling Allerød. This explains the much higher insoluble particle concentration, compared 280 to our Holocene samples (Fig. 1), which is closer to resemble concentrations found in Glacial ice (Ruth et al., 2003). The highest amount of insoluble particles measured with CFA is at 1256.96 m of depth (Fig. 1K), inside a cloudy band. Cloudy bands originate from seasonal events, such as atmospheric storms, resulting in increased transport and deposition of impurities on the ice sheet (Svensson et al., 2005). This explains the high number of mineral dust minerals, the low number of sulphates measured with Raman spectroscopy (Fig. 4A) and the resulting lower I var compared to the deepest sample (1339.75 m). It 285 further supports the hypothesis that insoluble dust contributes significantly to cloudy bands (Svensson et al., 2005). The unique location of the EGRIP drill site on an ice stream and the assumed high impact of high-impurity layers on the deformation of the ice (e.g., Dahl-Jensen and Gundestrup, 1987;Paterson, 1991;Miyamoto et al., 1999) highlight the need for an in-depth study of cloudy bands at EGRIP.
We adjusted the approach of using CFA data for efficient Raman analysis (Eichler et al., 2019) and conclude that analysing 290 insoluble particle data first on the large-scale, i.e. along one ice core (Fig. 1A), followed by small-scale analysis, i.e. over

295
In general, our results show a large abundance of minerals across all samples while depth-related differences occur in total, and relative, mineral abundance. A detailed discussion of the observed shift in mineralogy over the EGRIP ice core is presented in Sect. 4.2.1.
The most abundant minerals are sulphates and terrestrial dust minerals, which agrees with observations by e.g., Ohno et al. spectra). We identified most sulphates as gypsum or Na-sulphates, which agrees with e.g., Ohno et al. (2005); Eichler et al.
We observed a total of 170 gypsum micro-inclusions making it the most abundant mineral and the only sulphate which occurs in every sample supporting e.g., Legrand and Mayewski (1997) and Sakurai et al. (2009). Legrand and Mayewski (1997) show that CaSO 4 and M gSO4 can be produced if terrestrial compounds, such as M gCO 3 and CaCO 3 , are neutralised with H 2 SO 4 . They also found CaSO 4 in Holocene ice from Greenland despite a low Ca 2+ concentration.
Mineral dust, such as feldspar, mica and quartz, were abundant at all depths, but generally in lower abundances than gypsum and IC (Eichler et al., 2019) analyses, but there is a lack of understanding in which form they are present in ice. For example, sulfuric acid competes with other acids to react with the relative rare cations, replaces other acids in their salts and thus forms a variety of sulphate salts (Iizuka et al., 2008). The relative abundance of nitrates at a certain depth (Fig. 4A) indicates that similar processes occur with nitrates. However, nitrate ions seem to be more likely to exist in dissolved forms than in particle forms as suggested by Eichler et al. (2019).

Shift in mineralogy
We have observed a pronounced shift in mineralogy from a sulphate and terrestrial dust-dominated regime in the upper 900 m to a terrestrial dust-dominated regime with partially high amounts of gypsum. Similar findings were reported by earlier studies (e.g., Ohno et al., 2005;Sakurai et al., 2009;Eichler et al., 2019) who found sulphates and mineral dust in varying numbers, but our systematic approach exposes the mineral diversity already present in Holocene ice in more detail.

325
Sulphates and terrestrial dust minerals are found throughout the entire ice core, but in varying abundance ( Fig. 4 and simplified in Fig. 8). Rare minerals, such as hematite, anatase and titanite were found at all depths. Other minerals e.g., rutile and epidote, were only found in shallow samples while e.g., dolomite and nitrates were only found in deep samples. The high abundance of sulphates in many samples can be explained by the formation of sulphates after initial deposition (Ohno et al., 2005). N a + and M g 2+ , the main cations combining with SO 2− 4 , could originate from blown-in sea salt (N aCl and M gCl 2 ), 330 which was deposited on the ice sheet and, during transport, has partially chemically reacted with acidic compounds as proposed by Artaxo et al. (1992); Kerminen et al. (2000) for Antarctic sea salt. However, deeper samples show smaller amounts, and varieties, of sulphates, which almost completely disappear below 900 m. In the four deepest samples gypsum is the only sulphate, except for one Krohnkite micro-inclusion at 1141.17 m. Data is scarce, but e.g., Sakurai et al. (2009, 2011) and Eichler et al. (2019 show that sulphates are found in deeper parts of ice sheets, also consisting almost entirely of gypsum supporting 335 our results.
Even though we found a shift in mineralogy, this shift does not correlate with a change of the climate period, i.e. from Glacial to Holocene ice. Eichler et al. (2019) found a strong difference in mineralogy between Interglacial and Glacial ice in the EDML ice core. However, main minerals in Interglacial ice are sulphates (especially Na-sulphate and gypsum) while mineral dust and gypsum dominate in Glacial ice (Eichler et al., 2019). Ohno et al. (2005) show that salt inclusions consist

Detailed comparison with mineralogy in the GRIP ice core
A limited number of polar ice cores have been drilled over the last decades and mineralogy data is only available for a few of those. The relatively close to EGRIP located drill site of the GRIP ice core was analysed with Raman spectroscopy and EDS by Sakurai et al. (2009). We thus compare both cores here in detail.

350
We found gypsum at all depths at EGRIP resembling the results by Sakurai et al. (2009). Similar to our findings, Na-and The higher amount of terrestrial minerals in our samples is difficult to explain conclusively, because the GRIP and EGRIP sites are comparably close to each other and in similar distances to the coast. At certain depths, e.g., at 613.3, 1062.65, and 360 1256.96 m, the majority of micro-inclusions consists of mineral dust, which might indicate strong dust storms or deposition events as proposed for cloudy bands by e.g., Svensson et al. (2005). However, we show that the diversity in mineralogy can be high within centimetres (e.g., Fig. 7). The observed differences are thus probably caused by natural variability amplified by methodological differences, such as different sample sizes.

Methodological drawbacks 365
We can not determine if rarely found minerals only occur at certain depths, or if this is related to statistics drawbacks. Minerals, such as hematite and anastase, were only found in a few samples while rutile, titanite, epidote, and jacobsite have only been found once. Unfortunately, the problem of statistics reoccurs in microstructural impurity research, which can only be partly solved by analysing larger samples, and with a higher spatial resolution.
pounds, the superimposition of the ice spectra, and the focus on visible, undissolved impurities. Especially the main component of sea salt, NaCl, is inaccessible with Raman spectroscopy and thus not considered in our analysis.
Some Raman spectra from a depth of 415.3 m have been difficult to identify. We thus had to exclude several spectra, further lowering the already small number of observations. However, analysing inclusions inside the ice prevents contamination and preserves their in situ locations and is thus a well justified approach. 375 We could not identify all measured Raman spectra similar to other studies using this method. As an example we measured the same unknown spectra as presented in Fig. 2G by Sakurai et al. (2011). Obtaining good Raman spectra and identifying them with enough confidence will always be a challenge. These technical limitations hamper a fully holistic analysis and are difficult to overcome. A partial solution would be an analysis of the same samples with a subsidiary method, such as SEM coupled with EDS or LA-ICP-MS. Such a combined analysis would enable a deeper insight into the EGRIP ice core and into 380 impurities in polar ice in general.

Possible reasons for the diversity of minerals in EGRIP ice
We observed a prominent diversity in mineralogy of micro-inclusions in EGRIP ice across all scales. Mineral abundance changes throughout the core, i.e. the observed shift in sulphate diversity (Sect. 4.2.1), and across samples, i.e. in distinct layers and clusters (Sect. 3.4). This can partly be explained methodologically, e.g., by the number of analysed micro-inclusions per 385 sample, by different deposition conditions, or by chemical reactions taking place in the ice.
A difference in the chemistry at the ice sheet surface at the time of deposition and different atmospheric circulation patterns, and thus varying aerosol input, could be indicated by our CFA data (Fig. 1). Iizuka et al. (2012) concluded that aerosol sulfatisation in Antarctica is more likely to occur in the atmosphere or during fallout than after deposition. For Greenland, the major dust source areas for the Last Glacial and the Younger Dryas were the East Asian deserts (Svensson et al., 2000;Vallelonga 390 and  with minor contributions from the Sahara (Han et al., 2018). Aerosols deposited in the Holocene are likely from the Takla Makan and Gobi desert (Bory et al., 2003;. These differences in dust input between the Holocene and the Last Glacial are represented in our results, e.g., the highest I var value was measured in the deepest sample originating from the Bølling Allerød. The strong layering and clustering of sulphates could originate from dry deposition events, which form deposition crusts mainly containing sulphates. Another possible aspect is the unique impact of 395 NEGIS. Older samples originate from further upstream than younger samples, which could be reflected in the micro-inclusion mineralogy. However, we do not expect any major upstream deposition effects during the Holocene, but this should be verified in future studies. A systematic follow-up study on EGRIP Glacial ice is needed to investigate if the observed trends, e.g., of mineral diversity and clustering, continue with depth. Another possibility explaining the diversity in chemical compounds are chemical reactions occurring inside the ice as sum-400 marised for snow by Bartels-Rausch et al. (2014), and discussed for ice by Steffensen (1997) de Angelis et al. (2013) suggest movement of chloride and sulphate ions after deposition while Barnes and Wolff (2004) report of possible reactions between dust particles and soluble impurities. Baccolo et al. (2018) suggest that, for iron, oxidation 405 and dissolution are the main processes taking place in deep samples from the Talos Dome, Antarctica ice core. The deep ice sheet environment is not anoxic and dissolved oxygen and liquid water veins might support the oxidation and dissolution of specific mineral phases (Baccolo et al., 2018). Fe-minerals occur relatively often in polar ice and were identified in this study and by e.g., Baccolo et al. (2018);Eichler et al. (2019); Fe was found by e.g., Obbard and Baker (2007); Della Lunga et al.

Holocene climate as derived from ice cores and its possible imprint in mineralogy
Previous studies have suggested two possible scenarios for the Holocene climate evolution. It was either a a stable period of climate with the Holocene Thermal Maximum in the Mid-Holocene (9-5 ka BP), or a period with a rather long-term cooling trend and an earlier, stronger Holocene Thermal Maximum (Axford et al., 2021). The prominent change of abundant minerals 420 at a depth of 900 m could be explained by these two scenarios leaving different imprints on the mineralogy.
The extension of the Greenland Ice Sheet during the Holocene cannot be constrained exactly, and it probably differed significantly between regions (Lecavalier et al., 2014;Young and Briner, 2015). The rapid retreat of the ice sheet in NE-Greenland 12 ka ago lead to ice-free coastal areas in NE-Greenland after~9 ka BP (Lecavalier et al., 2014). Air temperatures over Greenland increased in the Holocene Thermal Maximum and peaked~7.8 ka BP (Lecavalier et al., 2014;Axford et al., 425 2021) exposing bedrock and thus enabling local dust sources. Syring et al. (2020) show that the sea ice cover in NE-Greenland transformed from a reduced state towards a marginal, almost extended, state between Early (~11.7-9 ka BP) and Mid Holocene (~9-5 ka BP) possibly impacting the input of sea salt, a major source of sulphates. Even though the main dust input to Greenland was from East Asia (Svensson et al., 2000;Bory et al., 2003), regional changes (e.g., exposed bedrock) might have had an impact on the availability of chemical compounds in the air, and thus during deposition. However, this is unlikely due to the 430 high elevation of EGRIP (2700 m) and the upstream effect caused by NEGIS, but detailed isotopic and mineralogical analyses are needed to test this.
The occurrence of other sulphates than gypsum in 9.8 ka old GRIP ice (Sakurai et al., 2009) contradict a general atmospheric signal for Greenland and support a site specific difference at EGRIP. A major change in aerosol chemistry and ion composition deposited at EGRIP is however unlikely, but upcoming studies are needed to discuss the chemical evolution at the site. The 435 average EGRIP annual layer thickness is surprisingly constant during the Holocene and a significant change occurs around 7-8 ka b2k (Mojtabavi et al., 2020), but the involved upstream effect of NEGIS and the general history of the ice stream are not well understood, thus hampering the further investigation of these, spatial and temporal, large-scale processes. Visual Stratigraphy and the core break record show that the brittle zone, the transition from air bubbles to clathrates hyrates (Ohno et al., 2004;Neff, 2014), at EGRIP is roughly at 550-1000 m of depth. Defects in the ice matrix, grain boundaries and micro-inclusions 440 can act as nucleation sites for hydrate nucleations . We found most clathrates below 900 m evoking the possibility that this transition zone also impacts sulphate chemistry, but details are unknown.
This brief evaluation of possibilities shows that chemical reactions and differences in the salt-formation conditions in the ice are the most likely explanation, as e.g., suggested for differences in micro-inclusion chemistry between Dome Fuji Holocene and Interstadial ice (Ohno et al., 2005). It is thus necessary to investigate the redistribution of impurities, and their chemical 445 reactions (including their rates), in ice sheets. Samples with a high number of identified Raman spectra mostly showed a higher diversity in mineralogy, which leads to the conclusion that more extensive impurity studies are needed to get comprehensive results. At this point, we can only suggest explanations for the change in diversity of sulphates, but upcoming studies on e.g., the bulk chemistry of EGRIP Holocene ice and flow effects of NEGIS might deliver crucial information to fully explain it.

450
The data presented in this study and its companion paper show the need to aim for a more holistic and in-depth understanding of the, macro-and micro-scale, processes in polar ice. Progressing towards an improved synoptic view is only possible by broader collaborations between the involved diverse research fields (microstructure, paleo-processes, impurities) applying different methods. Therefore, it is needed to clearly define method-related limitations, to develop concepts to tackle these limitations while systematically combining their strengths. Especially the broad field of cryo-related impurity research has a variety of 455 methods to measure dissolved and undissolved impurities, but they are mainly applied solitarily and results are often only partly transferable. To illustrate this point we briefly summarise the (simplified) capabilities of established methods regarding their ability to measure dissolved or undissolved impurities. IC measures dissolved (and a fraction of undissolved) impurities while Raman spectroscopy measures a limited number of undissolved particles, such as the micro-inclusions analysed in this study. CFA melts the ice, where an unknown fraction of insoluble particles in the ice go into solution. The remaining insoluble 460 particles are measured by the particle sensor of the CFA, giving a relative number of particle concentration of the original ice.
Finally, ICP-MS and LA-ICP-MS measure all, dissolved and undissolved, particles but information on the fraction of both impurity states are lost. Recent work by Ng (2021) demonstrates the challenge up ahead of better understanding climate signal records in deep polar ice and how these signals could be affected by processes involving grain boundaries, microstructure, veins, and impurities.

465
Our presented data and drawn conclusions demonstrate that our systematic approach is a good way towards a more detailed understanding of impurities, i.e. their mineralogy and location, in polar ice. It emphasises the importance of systematically analysing different spatial scales with a variety of methods to achieve holistic results. Despite the comparably high absolute number of identified micro-inclusions statistics are still improvable, especially for minerals which occur rarely such as hematite or anatase. Further in-depth analysis of EGRIP Glacial ice could help to improve statistical certainty. A comprehensive un-derstanding of the mineralogy and location of impurities in our samples could be reached by analysing our samples with a complimentary method, such as LA-ICP-MS or SEM and EDS.

Conclusions
We here derive the first systematic analysis of the mineralogy of micro-inclusions in Holocene and Late Glacial ice from the EGRIP ice core, the first deep ice core from a fast flowing ice stream. More specific, we derived dust particle concentration 475 along the upper 1340 m via CFA and analysed micro-inclusions in detail with a combination of optical microscopy and Cryo-Raman spectroscopy at eleven different depths.
By identifying almost 800 Raman spectra we obtained new qualitative, and quantitative, insights into the mineralogy of micro-inclusions in polar ice. 26 different spectra were identified, which indicates a more diverse mineralogy in ice from the last 14 ka than previously observed. In general most inclusions are sulphates, especially gypsum, and mineral dust, such as Our study emphasises the need to overcome technical limitations, which could be achieved by an inter-method comparison of e.g., Raman spectroscopy, LA-ICP-MS and SEM. However, our systematic overview of the mineralogy of micro-inclusions throughout a large part of one deep ice core helps to develop a better understanding of the role of impurities in polar ice.
Data availability. High-resolution microstructure and impurity maps were submitted to PANGAEA and will be available eventually. Raman data are available on request. CFA data will be published separately with an in-depth CFA study.