Microstructure, micro-inclusions, and mineralogy along the EGRIP ice core – Part 1: Localisation of inclusions and deformation patterns

Impurities deposited in polar ice enable the reconstruction of the atmospheric aerosol concentration of the past. At the same time they impact the physical properties of the ice itself such as its deformation behaviour. Impurities are thought to enhance ice deformation, but observations are ambiguous due to a shortage of comprehensive microstructural analyses. For the first time, we systematically analyse micro-inclusions in polar fast flowing ice, i.e. from the East Greenland Ice Core Project ice core drilled through the Northeast Greenland Ice Stream. In direct relation to the inclusions we derive the crystal preferred orientation, fabric, grain size, and microstructural features at 10 depths, covering the Holocene and Late Glacial. We use optical microscopy to create microstructure maps to analyse the in situ locations of inclusions in the polycrystalline, solid ice samples. Microinclusions are more variable in spatial distribution than previously observed and show various distributional patterns ranging from centimetre-thick layers to clusters and solitary particles, independent of depth. In half of all samples, micro-inclusions are more often located at or close to the grain boundaries by a slight margin (in the areas occupied by grain boundaries). Throughout all samples we find strong indications of dynamic recrystallisation, such as grain islands, bulging grains, and different types of sub-grain boundaries. We discuss the spatial variability in micro-inclusions, the link between spatial variability and mineralogy, and possible effects on the microstructure and deformation behaviour of the ice. Our results emphasise the need for holistic approaches in future studies, combining microstructure and impurity analysis.

or volume of micro-inclusions. Locations of micro-inclusions were mapped manually following Eichler et al. (2017) which provides an "impurity map". Impurity maps enable a structured and fast localisation of micro-inclusions with a confocal Raman microscope, which otherwise would be tedious in impurity-poor Holocene ice. They further allow the identification of micro-inclusions in the microstructure and thus preserve important spatial information. Grain boundaries were mapped on the sample surface and translated to the impurity map (Fig. 2). We applied a grain boundary width of 300 µm to compensate for light diffraction with depth and vertically tilted grain boundaries as done by Eichler et al. (2017). We created microstructure maps of all samples and located micro-inclusions in all of them. Micro-inclusions located in the 300 µm large grain boundary area are classified as "at the grain boundary" and serve as upper-limit assumptions.
Small grains might enhance the probability of micro-inclusions being located close to grain boundaries. To compare samples with different grain sizes we measured the total area occupied by grain boundaries per sample for ten samples resulting in the 145 ratio of grain boundary area to micro-inclusions at grain boundaries (R GB ): I GB is the percentage of micro-inclusions at grain boundaries, and A GB is the accumulated area occupied by grain boundaries per sample in percent using the upper limit assumption. A ratio of 1 implies a coherent relation of micro-inclusions at grain boundaries, while R<1 implies less micro-inclusions at grain boundaries than assumed from the grain boundary area of 150 the sample, R>1 implies the opposite, i.e. more micro-inclusions at grain boundaries than implied by the grain boundary area.
Furthermore, we performed a two-sided binomial test to derive the statistical significance of micro-inclusions being located at grain boundaries. The amount of micro-inclusions at grain boundaries and the area occupied by grain boundaries were used to calculate the respective p-value.

Evolution of grain size, CPO and microstructure with depth
We derive a profile of the grain size with depth, displaying the grain size evolution of the upper 1340 m of the EGRIP ice core and microstructural data from the depth regimes analysed with optical microscopy (Fig. 1).
Grain size: The mean grain sizes derived from the EGRIP core per 55 cm bag vary between 2.21 and 10.8 mm 2 (Fig. 1A).
Starting from 3.9 mm 2 at the depth of 111 m, it steadily increases in the shallowest part and peaks around 500 m. Grain size 160 decreases uniformly until 900 m, where it remains at~4.75 mm 2 until 1100 m of depth. The following 260 m are characterised by a steady decrease towards the minimum grain size of 2.21 mm 2 .
We find the grain size to be highly variable on the centimetre-scale, mean values from neighbouring bags can vary up to 5 mm 2 (e.g., at 240 and 540 m). This occurs primarily in the shallower part of the Holocene ice, mean grain size values are more similar towards the Last Glacial (Fig. 1A). Fabric images ( Fig. 1B-

Localisation of micro-inclusions
In total, 5728 micro-inclusions (small yellow circles in Fig. 4) were located within ten analysed samples. In total, the spatial distribution of micro-inclusions in EGRIP ice is highly variable and micro-inclusions are heterogeneously distributed (Fig. 5).
They have been found, solitary or in clusters, at grain boundaries, triple junctions and in the grain interior (Fig. 4). Distinct spatial distributional patterns have been observed throughout the core in varying strength and number, but no clear trend

Micro-inclusions at grain boundaries
1909 micro-inclusions were found in the proximity of grain boundaries, i.e. in the previously defined 300 µm large area 200 occupied by grain boundaries (Fig. 6). We further measured the grain boundary area for all samples. Our results are upper-limit assumption since real grain boundaries are basically interfaces, with an "amorphous zone" of only a few nm. nmi=amount of localised micro-inclusions, n GB =micro-inclusions at grain boundaries, I GB =percentage of micro-inclusions at grain boundaries, A GB =accumulated area occupied by grain boundaries, R GB =ratio of grain boundary area to percentage of micro-inclusions at grain boundaries, b2k=before 2000 (Mojtabavi et al., 2020) at depths of 415.3, 757.21 and 1062.65 m, respectively. The deepest sample from a depth of 1339.75 m showed the highest amount of micro-inclusions at grain boundaries (42.4%) and the largest area occupied by grain boundaries (43.2%) as here we have the smallest grain size of all inspected samples (Fig. 1A). In general, ratios do not vary much throughout our samples and are close to 1 indicating that the relative amount of micro-inclusions at grain boundaries is comparable at all depths. The exceptionally high ratio of 1.99 at 415.3 m is most likely caused by the low total amount of micro-inclusions (51) at this depth.

215
Excluding this sample lowers the average R GB to 1.2.
A GB and I GB were used to perform null hypothesis significance testing for ten samples with an alpha of 0.05 as the cutoff for significance. If the p-value is less than 0.05, we reject the null hypothesis that micro-inclusions are preferably located at grain boundaries and that a significant difference exists. If the p-value is larger than 0.05 no significant difference exists.
Calculated p-values vary from sample to sample and range from 0.962 to 3.8 · 10 −9 (  Fig. 6). Thus, in 50% of our samples there are significantly more micro-inclusions at grain boundaries than expected, the majority is however located in the grain interior. A small p-value often correlates with a high R GB while a high p-value usually correlates with a low R GB .

Identified micro-inclusions at grain boundaries
A companion paper presents the results of the analysis of the mineralogy of almost 800 micro-inclusions with Raman spec-225 troscopy. Here we only present the mineralogy of micro-inclusions located at grain boundaries at all depths. In total, 181 of all identified micro-inclusions were located at grain boundaries (i.e. 22.9%). 92 sulphate particles were found at grain boundaries, followed by gypsum (47), quartz (28), mica (22), feldspar (21), nitrates (9), hematite (8), and titanite and anatase (both 1).
31.3% of all feldspar was found at grain boundaries, followed by 27.6% of all gypsum , 27.2% of all mica ( We briefly discuss the evolution of the mean grain size per 55 cm bag with depth, which will be analysed in more detail in a future study. The observed grain size evolution with depth is similar to findings from the NEEM ice core (Eichler, 2013;Montagnat et al., 2014), but grain size can vary highly on the centimetre-scale. The broad single maximum CPO indicates Microstructural features are discussed in detail in Sec. 4.4. In-depth studies on the physical properties of the EGRIP ice core 250 will follow and are a chance to enhance our understanding of ice stream dynamics.

General findings
We localised more than 5700 micro-inclusions within ten samples from the upper 1340 m of the EGRIP ice core. Combining optical microscopy and Raman spectroscopy confirmed that mapped micro-inclusions are indeed visible impurities below the 255 sample surface supporting previous studies by Ohno et al. (2005Ohno et al. ( , 2006; Eichler et al. (2017Eichler et al. ( , 2019. The CFA data presented in a companion paper supports our micro-inclusion counts since CFA dust particle peaks correlate with areas of highly abundant micro-inclusions.
4.2.2 Localisation as found with microstructure mapping and Raman spectroscopy Stoll et al. (2021) found that the observed locations of impurities in polar ice are highly diverse and results seem to be strongly 260 influenced by the applied method. Similar to other studies applying Raman spectroscopy (e.g., Ohno et al., 2005;Sakurai et al., 2009;Eichler et al., 2017Eichler et al., , 2019 micro-inclusions are heterogeneously distributed throughout our samples and the amount of micro-inclusions close to grain boundaries varies from sample to sample. However, only a small amount of micro-inclusions is located directly on grain boundaries, the majority is located close to them. To quantitatively compare our results we calculated the ratio of micro-inclusions at grain boundaries to the area of grain boundaries per sample, which reduces the impact of grain-265 size. Our estimates are likely too large due to the exaggerated thickness of grain boundaries, which are much smaller than 300 µm. The ratio oscillates around 1, which describes a coherent distribution and tends slightly towards larger values indicating that grain boundaries may be often preferred locations of micro-inclusions. However, the derived p-values (Table 1, p-value) support these findings and emphasise the high variability between samples. The localisation of solid micro-inclusions seems to be much weaker than recently observed for dissolved impurities (Bohleber et al., 2020).

Localisation of micro-inclusions at different depths
We rarely observed distinct horizontal bands of micro-inclusions except at a depth of 757 m (Fig. 5). 7-11% of all microinclusion in EDML ice analysed by Eichler et al. (2017) were at grain boundaries, these samples are from the early Eeminan (MIS 5.5), i.e. "warm period ice", and thus comparable to our Holocene samples. Their NEEM samples showed a more spread-300 out distribution across the samples, and between 18 and 24% of micro-inclusions were at grain boundaries. This supports our findings (22-41% at grain boundaries) in EGRIP Holocene ice even though our values are generally higher. Eichler et al. (2017) analysed NEEM samples from a depth of 739.9 and 740.2 m, which is close to our samples from 613.3 and 757.21 m of depth with values of 30.3 and 31.5 %, respectively. Our higher value can be explained by sample selection from depths with high dust content as shown in a companion paper. Our results indicate that the spatial variability of micro-inclusions within Holocene ice 305 is larger than previously thought emphasising the difficulty in generalising spatial patterns as suggested by Stoll et al. (2021).
Summarising, our observations regarding the location of micro-inclusions show that the spatial variability is very high and changes on the millimetre-to centimetre-scale. Thus, generalisations about the location of micro-inclusions should be made with care and have to be specified for the depth intervals considered. However, the majority of micro-inclusions is found in the grain interior even though grain boundaries at some depths locate comparably high numbers of micro-inclusions. Interestingly, 310 we were not able to link physical properties of the ice, such as grain size and CPO, to distinct distributional patterns of microinclusions. Investigating these relationships remains challenging due to e.g., 1) the different sizes of ice grains and samples and 2) the heterogeneity in grain size and micro-inclusion distribution on the mm to cm-scale. We thus suggest in-depth investigations of cloudy bands for future research.

Impact of micro-inclusions on ice properties
315 Impurities affect the physical and mechanical properties of ice, especially the deformation and the flow of ice (Paterson, 1991;Cullen and Baker, 2001), and most studies show that impurity-rich ice is easier to deform than impurity-poor ice (e.g., Fisher and Koerner, 1986;Paterson, 1991;Cuffey et al., 2000). The potential influence of impurities on other, partly related, ice properties, such as CPO and grain size, are complex and manifold (Stoll et al., 2021). We discuss potential processes involving micro-inclusions and observed microstructural features and their implications for deformation via dominant mechanisms, and 320 dynamic recrystallization.
Zener pinning and drag of grain boundaries by impurities are suggested to be major mechanisms impacting grain size (e.g., Smith, 1948;Alley et al., 1986a;Fisher and Koerner, 1986;Alley and Woods, 1996;Paterson, 1991;Weiss et al., 2002;Durand et al., 2006). Even though roughly one third of all mapped micro-inclusions were located in proximity (300 µm upper limit assumption) to grain boundaries (1), we did not observe any direct indications of micro-inclusions involved in Zener pinning 325 or drag. Direct indications would be e.g., a sharp edge at a second phase particle in an otherwise curved grain boundary segment or a bulging grain boundary restricted by a second phase particle (e.g., Fig. 2 in Stoll et al. (2021)) (Passchier and Trouw, 2005). This agrees with other studies (Ohno et al., 2005;Faria et al., 2010;Eichler et al., 2017Eichler et al., , 2019 and indicates that the grain evolution is not affected strongly by the strict Zener pining process of micro-inclusions as proposed by Alley et al. (1986b), but if effected via reduced grain growth at all, it may rather be effected by a reduced grain boundary mobility.
Fisher and Koerner (1986); Alley et al. (1986b); Li et al. (1998); Iliescu and Baker (2008) suggest that impurity concentrations must be above a threshold to result in counteracted grain boundary mobility and restricted grain growth by Zener pinning.
However, quantitative thresholds are vaguely defined and partly ambiguous and thus difficult to discuss. It is possible that EGRIP Holocene dust concentrations are below this vague threshold to impact grain size development significantly via pining, but unlikely because we also analysed samples comparable to Glacial ice, i.e. with high dust concentrations (discussed in a 335 companion paper) and small grains (e.g., Fig. 1J and K).
High impurity layers showed no microstructural evidence for grain boundary sliding, such as linked-up grain boundaries or rectangular and lath-shaped grains (Fig. 6 in Goldsby and Kohlstedt (1997), Fig. 2B in Kuiper et al. (2020b)). Our results ( Fig. 3) indicate that solid micro-inclusions are not a main driver of e.g., grain size change via localised deformation along grain boundaries. However, it is possible that further methodological progress is needed to directly identify the impact of solid 340 inclusions on grain size and potential localised deformation via other deformation mechanisms e.g., Frank-Read sources and the multiplication and entanglement of dislocations, heterogeneous strain distribution within grains or dislocation pile-up on inclusions representing glide obstacles (Frank and Read, 1950;Ahmad et al., 1986;Weertman and Weertman, 1992).
The localisation of micro-inclusions at grain boundaries in our samples seems to be somewhat related to their mineralogy (Table 2, Fig. 8 in companion paper). In the case of solid micro-inclusions the interface properties between the ice and the 345 particle play a crucial role and are thus probably influenced by the mineralogy and thus the surface properties of the inclusion at a grain boundary. We know from anti-freeze proteins (e.g., Bayer-Giraldi et al., 2018) that "ice-binding" properties result from similar surface structures (on the crystal lattice scale) of the second phase particle and the ice crystal. Less than one third of all feldspar, gypsum, and mica inclusions were located at grain boundaries, which is in good agreement with our average derived percentage of micro-inclusions at grain boundaries. These minerals have partly comparable crystallographic lattice 350 parameters to ice Ih (Table 2). Especially the properties of feldspar are similar, or multiples, to the properties of ice while the crystallographic properties of gypsum and mica are less similar. Fenter et al. (2000) show that, after removing the outermost K ions, remaining Si atoms of K-rich feldspars become attached to OH or O resulting in a surface prone to interact with water molecules via hydrogen bonding. In addition, electrostatic interactions between the feldspar surface and the dipole moment of water might be enabled by the charged crystal lattice of feldspars (Yakobi-Hancock et al., 2013). Investigating the impact of 355 these crystallographic properties in depth goes beyond the scope of this study, but might be of interest for future studies.
While Bohleber et al. (2020) found element-specific localisation trends for dissolved impurities we show that it is more complex for solid micro-inclusions. While the composition of the micro-inclusions may play a certain role, as explained above, it seems to become clear that its state, i.e. solid, is more important especially when compared to studies on dissolved impurities (e.g., Bohleber et al., 2020). Bohleber et al. (2020) show that N a + intensity peaks at grain boundaries, as proposed 360 by e.g., Barnes and Wolff (2004), which supports the probable difference between dissolved and undissolved impurities (Stoll et al., 2021). A comparison of different methods by analysing the same samples is thus of great interest to clarify the role of the 1) state of impurities and 2) of the applied method. With respect to the interface processes mentioned above, another aspect is the poorly understood structure of grain boundaries in ice. The density anomaly of water, in contrast to metals or minerals, leads to molecules in grain boundaries being packed more closely than in the lattice and amorphous water veins (liquid-like (e.g., Mader, 1992)  almost constant temperature regime, do not tend to affix to grain boundaries, is indicated by our finding that the relative amount of mineral particles found at grain boundaries does not change significantly with depth (Table 1).

375
Deforming ice might enable palaeorecord alteration and enhance impurities to change locations with time and thus, mix and react in ice as discussed in detail in a companion paper and suggested by e.g., large clusters related to grain boundary migration during extreme grain growth conditions. Our samples are from a different temperature regime and not from bottom ice and processes, such as abnormal grain growth, thus do not take place, but the 385 overall mechanisms could be similar and should be investigated further. de Angelis et al. (2013) suggest that in EPICA Dome C bottom ice acid-salt interactions, ion relocation and salt formation occur in situ in relation with ice recrystallization, but secondary salt formation is limited to acids relocated at grain boundaries at the surface of primary salts in inclusions. A large inclusion (~800 µm) analysed with high resolution synchrotron X-Ray micro-fluorescence was a complex structure containing a variety of minerals and tens of particle aggregates formed by gathering and mixing of several liquid phases and solid particles 390 (de Angelis et al., 2013). Oversaturation in residual pockets containing carbonate and calcium ions resulted in the precipitation of calcium carbonate. The observation that micro-inclusions in our samples, especially sulphates, form clusters indicates that these processes have to be considered in EGRIP ice as well. Such detailed investigations are very valuable, but highly complex and costly and thus not suitable for extensive studies on micro-inclusions at several depths of one ice core as aimed for in this study.