Temperature and strain controls on ice deformation mechanisms: insights from the microstructures of samples deformed to progressively higher strains at -10, -20 and -30 °C

In order to better understand ice deformation mechanisms, we document the microstructural evolution of ice with increasing strain. We include data from experiments at relatively low temperatures (-20 and -30 °C), where the microstructural evolution with axial strain has never before been documented. Polycrystalline pure water ice was deformed under a constant 15 displacement rate (strain rate ~1.0 × 10 s) to progressively higher strains (~3, 5, 8, 12 and 20%) at temperatures of -10, 20 and -30 °C. Microstructural data are generated from cryogenic electron backscattered diffraction (cryo-EBSD) analyses. All deformed samples contain sub-grain (low-angle misorientations) structures with misorientation axes that lie dominantly in the basal plane suggesting the activity of dislocation creep (glide primarily on the basal plane), recovery and subgrain rotation. Grain boundaries are lobate in all experiments suggesting the operation of strain induced grain boundary migration (GBM). 20 Deformed ice samples are characterised by interlocking big and small grains and are, on average, finer grained than undeformed samples. Misorientation analyses between nearby grains in the 2-D EBSD maps are consistent with some 2-D grains being different limbs of the same irregular grain in the 3-D volume. The proportion of repeated (i.e. interconnected) grains is greater in the higher-temperature experiments suggesting that grains have more irregular shapes, probably because GBM is more effective at higher temperatures. The number of grains per unit area (accounting for multiple occurrences of the 25 same 3-D grain) are higher in deformed samples than undeformed samples, and this increases with strain, suggesting that nucleation is involved in recrystallisation. “Core-and-mantle” structures (rings of small grains surrounding big grains) occur in -20 and -30 °C experiments, suggesting that subgrain rotation recrystallization is active. At temperatures warmer than -20 °C, c-axes develop a crystallographic preferred orientation (CPO) characterized by a cone (i.e., small circle) around the compression axis. We suggest the c-axis cone forms via the selective growth of grains in easy slip orientations (i.e., ~45° to 30 shortening direction) by GBM. The opening-angle of the c-axis cone decreases with strain, suggesting strain-induced GBM is balanced by grain rotation. Furthermore, the opening-angle of the c-axis cone decreases with temperature. At -30 °C, the c-


Introduction
Glaciers and ice sheets play key roles in shaping planetary surfaces, and form important feedbacks with climate, both on Earth (Hudleston, 2015) and elsewhere in the solar system (Hartmann, 1980;Whalley and Azizi, 2003). Understanding the controls on the flow rate of terrestrial glaciers and ice sheets is crucial, as this will be a major control on future sea level change (Bindschadler et al., 2013;Dutton et al., 2015;Bamber et al., 2019). Terrestrial glacial ice flow is driven by gravity. Flow is 10 facilitated both by sliding at the base of the ice and by the internal (creep) deformation of ice masses. The contribution of creep deformation to the total flow rate is controlled primarily by differential stress and temperature within the ice body (Rignot et al., 2011;Hudleston, 2015). Creep experiments show a change in the mechanical behaviour as initially isotropic polycrystalline ice is deformed (Faria et al., 2014;Hudleston, 2015). Mechanical weakening occurs during the transition from secondary creep (minimum strain rate) to tertiary creep (quasi-constant strain rate) in constant load experiments (e.g., Weertman, 1983; Budd 15 & Jacka, 1989;Montagnat et al., 2015) and from peak stress to steady-state stress in constant displacement experiments (e.g., Durham et al., 1983Durham et al., , 2010Vaughan et al., 2017;Qi et al., 2017). This mechanical weakening is often referred to as strain rate "enhancement" in the glaciological and ice sheet literature (Alley, 1992;Placidi et al., 2010;Treverrow et al 2012).
Enhancement correlates with the development of a crystallographic preferred orientation (CPO) (Jacka and Maccagnan, 1984;Vaughan et al., 2017) and also with other microstructural changes, in particular grain size reduction (Craw et al., 2018;Qi et 20 al., 2019). Understanding the deformation and recrystallization mechanisms responsible for ice microstructure and CPO development is therefore essential for quantifying how different mechanisms contribute to ice creep enhancement in nature.
Moreover, the relative roles of intracrystalline plasticity, recrystallization and grain size sensitive mechanisms, especially at low temperatures, are not well known.
In this contribution we present microstructural analyses of samples deformed to successively higher strains through the 25 transition from peak stress (secondary creep) to steady-state stress (tertiary creep) at -10, -20 and -30 °C. These conditions were chosen so that the experiments included evolution of CPO towards a cone (high temperature) and towards a cluster (low temperature). For the first time, we present ice microstructural data from samples deformed to progressively higher strains at -20 and -30 °C. Our objectives were to study the influences of temperature and strain on ice mechanical behaviour, microstructure and CPO development.

Mechanical data processing
During each experimental run, time, displacement and load were recorded at a frequency of 0.14-0.2 Hz. The axial stress was calculated from the load divided by cross-sectional area of the ice sample. The stress has been corrected for the change of sample cross-sectional area, assuming constant sample volume during the deformation. For each ice sample, we define the axial stretch l (Eq. (1)) as the ratio of the sample length ( ( )) at time t and the initial sample length ( 1 ). The sample length 5 ( ) at time t is calculated from the displacement and the initial sample length ( 1 ). The true axial strain ( ) is defined in Eq.

Cryo-EBSD data 10
The recent development of cryo-EBSD technique provides an access to full crystallographic orientations and significant microstructural detail in analysing ice samples with large sizes (up to 70 mm by 30 mm) (Prior et al., 2015). We prepared the ice samples and acquired the cryo-EBSD data following the procedures described by Prior and others (2015). The samples were cut in half along the cylindrical long-axis by a band saw in a -20 °C cold room and a 5 mm slice was extracted from half of the sample. One side of the slice, at a temperature of ~-30 to -50 °C, was placed against a copper ingot (70 mm by 35 mm) 15 at ~5 °C. The samples were placed in a polystyrene sample transfer box (~-100 °C) as soon as a bond formed between the ice sample and the ingot (Prior et al., 2015). We acquired a polished sample surface for cryo-EBSD by hand lapping on grit paper.
The samples were polished at ~-40 °C to 3.0-6.0 mm thick using grit sizes of 80, 240, 600, 1200 and 2400. Soon after, the sample-ingot assemblies were transferred to the polystyrene sample transfer box and cooled to close to liquid nitrogen temperature, before they were transferred to the SEM for the collection of cryo-EBSD data. 20 EBSD data were acquired using a Zeiss Sigma VP FEGSEM combined with a NordlysF EBSD camera from Oxford Instruments. The sample-ingot assembly was transferred to a cold stage (maintained at ~-100 °C) fixed in the SEM chamber.
We used pressure cycling in the SEM chamber to activate surface sublimation to remove frost and create a damage-free sample surface (Prior et al., 2015). EBSD data were acquired at a stage temperature of ~-95 °C, with 5-7 Pa nitrogen gas pressure, 30kV accelerating voltage and ~60 nA beam current. For each ice sample, we collected a reconnaissance map with a step size 25 of 30 µm for the whole section. For detailed microanalysis, we also collected a map with the step size of 5 µm from a selected sub-area. We acquired and montaged the raw EBSD data by using the Aztec software. Details of the raw EBSD data have been summarized in Table 2. https://doi.org/10.5194/tc-2020-2 Preprint. Discussion started: 22 January 2020 c Author(s) 2020. CC BY 4.0 License.

Processing of the cryo-EBSD data
Ice microstructural parameters including grain size and grain boundary geometry are important indicators for inferring deformation processes. We quantified the microstructural parameters of ice grains from raw EBSD data using the MTEX toolbox (Bachmann et al., 2011;Mainprice et al., 2015) in MATLAB. The ice grains were reconstructed from the raw EBSD pixel map with 5 µm step size. We defined the ice grain boundaries where the misorientations between the neighbouring pixels 5 were larger than 10°. We removed noise from the EBSD data by removing the grains containing fewer than 4 indexed pixels.
No pixel interpolation was applied to the EBSD pixel map, preserving any non-indexed space. Grain size was determined as the equivalent diameter of a circle with the area equal to the measured area of each grain.
We quantified the geometrically necessary dislocations (GNDs) (Ashby, 1970) using the Weighted Burgers Vector (WBV) method (Wheeler et al., 2009). GNDs are the dislocations required to generate an observed lattice distortion (for example, a 10 sub-grain boundary). The WBV method gives a weighted measure of the Burgers vectors of GNDs by defining the WBV as the sum, over all types of dislocations, of [(density of intersections of dislocation lines with a map) × (Burgers vector)]. Note that the 3D density for each type of the dislocation is weighted by a parameter l, which depends on the angle between the dislocation line and the EBSD map plane. The magnitude of l ranges from 1 (when the dislocation line is perpendicular to the EBSD map plane) to 0 (when the dislocation lies within the EBSD map plane). 15 We applied the point-by-point WBV calculations (Wheeler et al., 2009) on the EBSD maps with 5 µm step size. The WBV magnitude of each pixel (||WBV||), which provides a minimum estimation for the magnitude of dislocation density tensor, was calculated from the Euclidean norm of the WBV. The sub-grain boundaries were defined at where the ||WBV|| is higher than 0.0026 µm -1 . This threshold for defining subgrain boundaries is equivalent to a misorientation of 0.75°, which is higher than the limit of angular error of EBSD data. We calculated the c-component WBV (WBVc) of the pixels on the sub-grain 20 boundaries (Chauve et al., 2017). The proportion of c-component WBV ( 9:;< ) was quantified from the ratio between |WBVc| and ||WBV||. The value of 9:;< ranges from 0 (when the dislocation is dominated by a-component Burgers vectors) to 1 (when the dislocation is dominated by c-component Burgers vectors).
The EBSD maps with 5 µm and 30 µm step size have been used to generate the crystallographic preferred orientation (CPO) data with one point per pixel. The CPO data were contoured with a half-width of 7.5° based on the maximum of multiples of 25 a uniform distribution (MUD) of the points, for a better recognition of the CPO pattern. The MUD value is higher at the orientation that contains a higher density of points. The CPO intensity was quantified by the M-index (Skemer et al., 2005).
The M-indices and eigenvalues are consistent between CPOs generated from the EBSD maps with 30 μm and 5 μm step sizes.
The CPOs of ice deformed under uniaxial compression at high temperatures are often characterised by c-axes aligning in an open cone (i.e., small circle) ( Fig. 1(a)), around the compression axis (Kamb, 1972;Jacka and Maccagnan, 1984;Wilson et 30 al., 2014;Jacka and Li, 2000;Qi et al., 2017). In order to quantify cone opening-angles, we counted the number of c-axes that lie at a given angle ("co-latitude") from the compression axis (method adapted from Maccagnan, 1984, andPiazolo et al 2013). In practice we counted the c-axes between two co-latitudes separated by a 4° interval (selected by trial, see section https://doi.org/10.5194/tc-2020-2 Preprint. Discussion started: 22 January 2020 c Author(s) 2020. CC BY 4.0 License. S1 of the supplementary material for the sensitivity test of co-latitude range) and calculated the MUD for this co-latitude range to plot on a graph of MUD as a function of co-latitude (Fig. 1).

Mechanical data
The stress-strain curves for all the deformed ice samples are plotted in Fig. 2. Imposed initial strain rate and temperature are 5 shown in Table 1 together with peak and final stresses and corresponding strain rates. The strain rate increases slightly with strain (Table 1) as is required for a shortening sample at constant displacement rate. For all the deformation runs, stress initially increases as a function of strain, before reaching the "peak stress" at axial strains of 0.01 ≤ ≤ 0.04. Beyond the peak stress, stress deceases with increasing strain, with the rate of stress drop decreasing with increasing strain. At strains larger than ~0.1, stresses reduce only a modest amount, with steady-state reached at a strain of ~0.2. Peak and final stresses are larger at colder 10 temperatures and the peak stresses are better defined at -30 °C than at the warmer temperatures.

Microstructure
EBSD data are used to generate the illustrative grain orientation maps, grain sub-structure maps, as highlighted by WBV analysis, grain size distributions and subgrain size distributions shown in Fig. 3 angle, ≥ 2°. Note that the quantitative microstructural analyses and CPO data are based on larger areas than those presented in the EBSD maps (Table 2).

Sub-structure
Distinct intragranular distortions and sub-grain boundaries can be observed in all the samples (Fig. 3 (c), 4 (c) and 5 (c)).
Kinking, which breaks large grains into irregular shapes and sizes, can be observed in samples deformed at -30 °C to strains 20 higher than ~3%. At -20 and -30 °C, samples with strains higher than ~12% show a "core-and-mantle" structure (Gifkins, 1976;White, 1976), which is characterised by a "net" structure formed by finer grains encircling larger grains with slightly curved boundaries. CPO development, indicated by grain colours in microstructure maps, occurs with increasing strain. At -10 °C, grains with near-pink-and-orange colours dominate the IPF maps ( Fig. 3(a-b)) at strains higher than ~8%. At -20 and -30 °C, grains with 25 red, pink and orange colours predominant in the IPF maps ( Fig. 5(a-b) and 6 (a-b)) at ~20% strain.

Grain size
The undeformed ice samples all exhibit < 1% porosity and a homogeneous foam-like structure, with a mean grain size of ~230 µm (Qi et al., 2017). For all the samples deformed to ~3% strain, the grain size distributions are strongly skewed or possibly bimodal, with the main peak at finer grain sizes and a tail of coarser sizes with secondary peaks including grain sizes corresponding to the mean grain size of starting material ( Fig. 3(d), 4(d) and 5(d)). As the strain increases, the grain size 5 generally gets smaller. The range of the grain size distribution generally narrows, with the proportion of finer grains increasing and the coarse grain tail shortening ( Fig. 3(d), 4(d) and 5(d)).
For each sample, we calculated the mean grain diameter ( B ), the peak grain diameter ( CDEF ) and square mean root diameter ( GHI = (√ KKKK ) L ), presented in Table 3. While the mean diameter, B , is commonly used for grain size analyses in the literature (e.g., Jacka, 1994;Piazolo et al., 2013;Vaughan et al., 2017;Qi et al., 2017;Qi et al., 2019), Lopez-Sanchez and Llana-Fúnez 10 (2015) showed that the frequency peak ( CDEF ) of a grain size distribution provides the most robust measure of the recrystallized grain size. However, the population of grains smaller than CDEF is too small to provide representative CPO information for most of the data sets in this study. Meanwhile, GHI minimizes the bias from very large grains in the calculation of an average and is therefore better suited than B and CDEF for separating small and large grain populations.
B , GHI and CDEF have the relation of B > GHI > CDEF , and converge as the strain increases, with B and GHI becoming 15 stable after ~12% strain ( Fig. 3 (d), 4 (d), 5(d) and Table 3). For each temperature series, we defined a threshold grain size, equals to the GHI of the sample deformed to ~12% strain. The grains with the grain sizes greater than the threshold were classified as "big grains". Grains smaller than or equal to the threshold were classified as "small grains". The "small grains" are likely include all the recrystallized grains (Lopez-Sanchez and Llana-Fúnez, 2015) and some of the remnant grains. We calculated the mean diameters of "big grains" ( MNO KKKKKK ) and "small grains" ( PQERR KKKKKKKK ) for all deformed samples. The evolutions of 20 B , MNO KKKKKK and PQERR KKKKKKKK as a function of strain at different temperatures are illustrated in Fig. 12(a). B generally decreases with strain, and it is lower at colder temperatures relative to higher temperatures for samples deformed to strains higher than ~8%.
PQERR KKKKKKKK for each temperature series barely changes with strain, and it is higher at -10 °C than at -20 and -30 °C. At strains lower than ~12%, MNO KKKKKK generally decreases with strain, and it is larger at warmer temperatures.
As the boundary misorientation decreases, the mean subgrain size ( ̅ ) drops. At each , ̅ generally gets smaller as the strain 30 increases but the trend is not simple. For each sample, the distribution of subgrain size at boundary misorientation angle ≥ https://doi.org/10.5194/tc-2020-2 Preprint. Discussion started: 22 January 2020 c Author(s) 2020. CC BY 4.0 License. 2° ( Fig. 3(e), 4(e) and 5(e)) is similar to the grain size distribution (Fig. 3(d), 4(d) and 5(d)), but the main peak in the subgrain size distribution is at a finer size than the main peak of the grain size distribution (Table 3). At lower temperature, the boundary hierarchy distribution has the same shape but at smaller subgrain or grain sizes (Fig. 6). It is worth noticing that the PQERR KKKKKKKK is closer to ̅ at ≥ 2° at ~20% strain as the temperature decreases (Fig. 6). The ratio between mean recrystallized grain size and mean subgrain size has been considered as a useful parameter to quantify the recrystallization mechanism (Halfpenny et 5 al., 2012): because subgrain rotation recrystallization should produce grains that have similar sizes with subgrains, while bulging nucleation should produce grains that have smaller sizes than subgrains (Halfpenny et al., 2012). The "small grains" in this study probably contain all the recrystallized grains and some of the remnant grains. Therefore, PQERR KKKKKKKK can be considered as a maximum estimate of the mean recrystallized grain size. We calculated the ratio VWXX (= PQERR KKKKKKKK / ̅ ) for all deformed samples (Table 3). VWXX values are between 0.4 and 0.6 at strains up to 8% at -10 and -20 °C. Low strain VWXX values are 10 slightly higher at -30 °C. At -10 °C, the largest VWXX of 0.6 is at high strains (~12% and ~20%). VWXX values are higher (0.8-1.2) at high strains (12% and 20%) at lower temperatures (-20 and -30 °C).

Grain boundary geometry
The microstructures of deformed ice samples are characterised by larger grains interlocking with finer grains. At -10 °C, the boundaries of larger grains are generally more irregular (lobate) ( Fig. 3(a) and 3(b)) than the boundaries of larger grains at - 30 15 °C (Fig. 5(a) and 5(b)). Quantification of grain boundary lobateness has been considered as a useful microstructural parameter (Krul and Nega, 1996;Takahashi et al., 1998) to infer processes and to correlate with deformation conditions. In this study, we quantified the boundary shape of each ice grain by introducing a sphericity parameter Ψ, which is calculated from grain area A, grain boundary perimeter P and grain radius R. Ψ is defined as: The sphericity parameter Ψ can be considered as a useful indicator for grain boundary lobateness because Ψ measures how closely the ice grain boundary shape resembles that of a perfect circle. Ψ decreases from 0.5, where the grain has a perfect round shape, to 0, where the grain is infinitely serrated. Examples of grains and specific geometric shapes that present different Ψ values are illustrated in Fig. 7. This method is similar to that applied by Takahashi and others (1998) to quantify grain boundary shapes. We plotted the Ψ values as a function of grain sizes for all deformed samples grouped by deformation 25 temperature and strain ( Fig. 8). At any given temperature and strain, there is an inverse relationship between Ψ and grain size, indicating that "big grains" are generally more lobate than "small grains". As strain increases, the lobateness of "small grains" remains more-or-less constant, while "big grains" become more lobate and shift to progressively lower values of Ψ. Thus, the grain boundary lobateness of "small" and "big" grains become more distinct with increasing strain. As temperature increases, Ψ decreases more rapidly with increasing grain size, and "big grains" evolve to smaller values of Ψ, reflecting an overall

Crystallographic preferred orientations
The contoured c-, a-and m-axes pole figures are illustrated in Fig. 9-11. The contoured pole figures are presented with (1) the compression axis vertical and (2) with the compression axis perpendicular to the page. These two reference frames, which are commonly used by different communities, enable different elements of symmetry to be illustrated. We calculated the M-indices for the three grain size categories, including "all grains", "big grains" and "small grains" from the EBSD maps with 5 µm step 5 size ( Fig. 12(b)). The M-indices for all grains are consistent between CPOs generated from the EBSD pixel maps with 30 μm and 5 μm step sizes (Table 2): small grains cannot be separated robustly in the 30 μm data sets. The undeformed ice samples exhibit a random CPO with the M-index of 0.0026 (Qi et al., 2017). For all the deformed samples, the M-indices of "big grains" present a similar evolution pattern to "all grains" as the strain increases. The "small grains" generally exhibit lower M-indices at strains of ≥ ~5%. To illustrate the CPO differences between "big grains" and "small grains", we present the contoured c-10 axes CPOs for both big and small grains for the samples deformed to ~12% strain (Fig. 13). The "big grains" and "small grains" have similar patterns of c-axes CPOs. At -10 °C, the CPO intensity of "small grains" is lower than "big grains", and this contrast becomes strengthened as the temperature decreases.

-10 °C series
The CPO of the sample (PIL176) deformed to ~3% strain at -10 °C is characterized by several weak maxima of c-axes with 15 similar angles relative to the compression direction, and random distributions of a-and m-axes. As the strain increases from ~5%, the CPO becomes clearer, with c-axes aligned in a cone (small circle), and the a-and m-axes align in a broad swath along the plane perpendicular to the compression axis and bound by the c-axis cone. The opening-angle of the c-axis cone decreases with strain, from ~40° at ~5% strain to ~34 ° at ~20% strain. CPO intensity rises with strain in all grain size categories. At ~12% strain, the c-axes of "big grains" align in interconnected elongated maxima that form a cone. The "small grains" exhibit 20 a slightly weaker c-axis CPO, which is characterised by an overall cone of overlapping distinct maxima.

-20 °C series
The CPO of the sample (PIL183) deformed to ~3% strain at -20 °C is approximately random. At ~5% strain (PIL182), the CPO is characterized by a blurred cone formed by several weak maxima of c-axes, and randomly distributed a-and m-axes.
As the strain increases from ~8% to ~12%, the CPO becomes clearer, with c-axes aligned in distinct clusters superposed on a 25 blurred broad cone. The a-and m-axes align in weak a broad swath along the plane perpendicular to the compression axis and bound by the c-axis small cone. At ~20% strain, the c-axes align in a cone (small circle), and the a-and m-axes align in broad swath along the plane perpendicular to the compression axis and bound by the c-axis small cone. The opening-angle of the c-axes cone is ~29±3° and showing no clear pattern with strain. CPO intensity increases with strain for all grain size categories.
At ~12% strain, the c-axes of "big grains" align in distinct maxima superposed on a blurred cone. The "small grains" exhibit

-30 °C series
The CPOs of the samples (PIL165, PIL162) deformed to ~3% and ~5% strain at -30 °C are approximately random. As the strain increases from ~8% to ~12%, the c-axis CPO exhibits a pattern of a distinct narrow cone superposed on an overall broad cluster, the a-and m-axes align in a broad swath along the plane perpendicular to the compression axis. At ~20% strain, the caxes align in distinct clusters superposed on an overall broad cluster, and the a-and m-axes align in a broad swath along the 5 plane perpendicular to the compression axis and bound by the c-axis narrow cone. The opening-angle of the c-axes cone is ~15±1° at ~8% and ~12% strain, and it decreases to ~8° at ~20% strain. CPO intensity increases mildly with strain for "all grains" and "big grains". For "small grains" the CPO intensity remains at similar values to the starting material before ~12% strain, and it increases slightly at ~20% strain. At ~12% strain, the c-axes of "big grains" align in distinct maxima superposed on a blurred broad cluster. The "small grains" exhibit a significantly weaker c-axis CPO, which is characterised by a blurred 10 broad cone with overlapping weak and blurred maxima.

The opening-angle of the c-axis cone
We compare the evolution of the opening-angle, , of the cone-shaped c-axis CPO with the results from previous studies (Table 4 and Fig. 14). For data from the literature, we digitised the c-axis orientations from published stereonets (Jacka and Maccagnan, 1984;Jacka and Li, 2000) and calculated using the same method described in Section two (Method).  (2013) were conducted on D2O ice at -7 °C, which is a direct analogue for deforming H2O ice at −10 °C (Wilson et al., 2019). These angles were analysed using methods similar to ours. In order to make a direct comparison 20 with the data reported from this study and Qi and others (2017), we converted the reported axial engineering strain ( ) and strain rate () (Piazolo et al., 2013;Montagnat et al., 2015;Vaughan et al., 2017) to true axial strain ( ) and strain rate () using the equations: Equation (6) and (7) were used to forward model axial engineering strain ( ) and strain rate () from octahedral shear strain ( ) and strain rate () (Jacka and Maccagnan, 1984;Jacka and Li, 2000). (7) https://doi.org/10.5194/tc-2020-2 Preprint. Discussion started: 22 January 2020 c Author(s) 2020. CC BY 4.0 License.
After that, the axial engineering strain ( ) and strain rate () were converted to true axial strain ( ) and strain rate () using Eq.
To our knowledge, Fig. 14 and Russell-Head, 1982;Gao and Jacka, 1987;Treverrow et al., 2012;Wilson et al., 2019). These data are consistent with the pattern shown in Fig. 14  The stress-strain curves (Fig. 1) at all temperatures first rise to the peak stresses and then relax to approach near-constant stresses with strains. This pattern matches published constant-displacement-rate experiments (Mellor and Cole, 1982;Durham et al., 1983;Durham et al., 1992;Piazolo et al., 2013;Vaughan et al., 2017;Qi et al., 2017;Craw et al., 2018;Qi et al., 2019), and is comparable to the constant-load experiments (Budd and Jacka, 1989;Jacka and Li, 2000;Treverrow et al, 2012;Wilson and Peternell, 2012) where strain rate first decreases to a minimum and then increases to approach a near-constant strain rate. 25 Much of the stress increase prior to peak stress relates to elastic strain. However, linear portions have slopes of ~1GPa and this is below the published value of Young's modulus (~9GPa: Gammon et al, 1983). This and the curvature of the stress strain line at the start of each experiment suggests that there is also some dissipative deformation here. This can include porosity loss (Vaughan et al., 2017) and the intergranular stress redistribution used to explain primary creep in constant load experiments (Duval et al, 1983). The drop of stress after peak correlates with dynamic recrystallization driven grain size reduction and CPO 30 development (Jacka and Maccagnan, 1984;Vaughan et al., 2017;Qi et al., 2019). Experiments with initial grain size as a https://doi.org/10.5194/tc-2020-2 Preprint. Discussion started: 22 January 2020 c Author(s) 2020. CC BY 4.0 License.
variable, under comparable conditions to our experiments, suggest that grain size sensitive mechanisms are important (Qi et al., 2017). Grain boundary sliding (GBS) is kinematically required for all grain size sensitive mechanisms (Stevens,1971;Gates and Stevens, 1974), including diffusion creep (Boullier and Gueguen, 1975;Behrmann and Mainprice, 1987) and dislocation slide accompanied by GBS (disGBS) (Warren and Hirth, 2006). Goldsby and Kohlstedt (1997, 2001, 2002 suggest a general importance of GBS on the basis of the constitutive law parameters required to fit the mechanical data from 5 experimentally deformed fine-grained ice. Recent studies suggest GBS in fine-grained ice layers has a key role in controlling the Greenland ice flow (Kuiper et al., 2019a(Kuiper et al., , 2019b by applying the Goldsby-Kohlstedt flow law Kohlstedt, 1997, 2001) to modelling the deformation in the NEEM (North Greenland Eemian Ice Drilling) deep ice core. The grain size reduction resulting from dynamic recrystallization is thought to cause mechanical weakening by increasing the strain rate contribution of grain size sensitive deformation mechanisms (De Bresser et al., 2001). A development of strong CPO can also 10 lead to mechanical weakening in viscously anisotropic materials (Durham and Goetze, 1977;Hansen et al., 2012) such as ice.
Our experiments show higher peak and steady-state stress values at colder temperatures than at warmer temperatures. This phenomenon is well known, and the temperature dependence of creep rate is commonly parameterised using an Arrhenius relationship with an activation enthalpy (sometimes called entropy) (Homer and Glen, 1978;Durham et al., 1983Durham et al., , 2010Cuffey & Paterson, 2010;Scapozza & Bartlett, 2003). 15

Inferences from grain size distribution and microstructure
For all the deformed samples, the grain size distributions are characterised by peaks at finer grain sizes, and a smaller mean grain size compared with the undeformed sample. These observations suggest that nucleation, which generates grains with smaller sizes, is involved in the dynamic recrystallization processes. Microstructure maps and WBV analyses show subgrain boundaries in all deformed samples ( Fig. 3(a-c), 4(a-c) and 5(a-c)). Moreover, the plots of mean subgrain size ( ̅ ) as a function 20 of boundary misorientation angle ( ) for all samples (Fig. 6) show ̅ drops with decreasing , indicating a development of continuous boundary hierarchy (Trimby et al., 1998). Study in deformed quartz mylonites infer that a continuous boundary hierarchy development with can be correlated with a simultaneous operation of recovery and subgrain rotation (Trimby et al., 1998). Therefore, we infer the recovery and subgrain rotation were active in all deformed samples.
At -10 °C, the distribution of grain sphericity parameter Ψ for "big grains" reduces with increasing strain (Fig. 8(a)), and the 25 IQRs of Ψ values for "big grains" show lower median point values at larger grain sizes at strains higher than ~5% (Fig. 8(b)).
These phenomena suggest the larger grains become more lobate with strain, indicating a continuous operation of strain-induced grain boundary migration (GBM) (Urai et al., 1986) with increasing strain. At -20 and -30 °C, the IQRs of Ψ values for "big grains" show a stabilisation of median points at ~0.35 at strains lower than ~12% (Fig. 8 (d) and (f)), suggesting many of the "big grains" are of low lobateness. This makes sense as GBM is thermally activated (Merkle et al., 2004), and it should be less 30 active when temperature drops (Urai et al., 1986;Piazolo, et al., 2006). At lower temperature, the boundary hierarchy distribution has the same shape but at smaller subgrain or grain sizes (Fig. 6). This is likely a consequence of the higher stresses https://doi.org/10.5194/tc-2020-2 Preprint. Discussion started: 22 January 2020 c Author(s) 2020. CC BY 4.0 License. of the lower temperature experiments resulting in smaller subgrain and recrystallised grain sizes through a piezometer or similar relationship.
At -10 °C, the ratios ( VWXX ) between the mean grain sizes of "small grains" ( PQERR KKKKKKKK ) and mean subgrain sizes ( ̅ ) at boundary misorientation angle ( ) higher than 2° are lower than or equal to 0.6 (Table 3). This phenomenon suggests the "small grains" are smaller than subgrains. However, at lower temperatures, the VWXX values are higher, indicating that the grain sizes of "small 5 grains" and subgrains become similar at lower temperatures. Comparing recrystallized and subgrain size has been used to discriminate nucleation mechanisms, including subgrain rotation recrystallization and bulging (Halfpenny et al., 2006;Halfpenny et al., 2012;Platt and De Bresser, 2017). The VWXX provides a maximum estimation of the ratio between recrystallized grains and subgrains because the "small grains" used for the calculation probably contain all of the recrystallized grains and some of the remnant grains. Subgrain rotation is a process that involves an increase in the misorientation across the 10 subgrain boundary resulting from continuous addition of dislocations (Lallemant, 1985;Placidi et al., 2004). New grains will form as the misorientation across the subgrain boundary becomes large enough, with the subgrain boundary eventually dividing its parent grain (Poirier and Nicolas, 1975;Guillope and Poirier, 1979;Urai et al., 1986;Gomez-Rivas et al., 2017). This process is known as the subgrain rotation recrystallization (Hirth and Tullis, 1992;Stipp et al., 2002;Passchier and Trouw, 2005). When subgrain rotation recrystallization is responsible for the nucleation, the recrystallized daughter grains should be 15 initially of a similar size to the internal subgrain size of the parent grain (Urai et al., 1986). Our data show the "small grains" are smaller than subgrains for all samples deformed at -10 °C and samples deformed to lower strains at -20 and -30 °C. This observation suggests subgrain rotation recrystallization alone is unlikely to be the nucleation mechanism. others (2006, 2012) suggest the bulge nucleation is important in facilitating recrystallization during strain-induced GBM, based on the observation of smaller bulges compared with subgrains in deformed quartz mylonite. The size of the bulge is not controlled 20 by the subgrain size because the bulge can be achieved by the development of a bridging subgrain boundary across the neck of the bulge and its conversion by progressive misorientation into a high-angle grain boundary (Urai et al., 1986). Therefore, we suggest a larger subgrain size compared with the grain size of "small grains" should result from the activation of bulge nucleation in samples deformed at -10 °C. The bulging in this case is a consequence of strain-induced GBM that is favored by high boundary mobility at higher temperatures. The ratios ( VWXX ) between the mean grain sizes of "small grains" and mean 25 subgrain sizes suggest that bulging becomes less important at -20 and -30 °C (Table 3). At -20 and -30 °C, network of smaller grains encircles bigger grains at strains higher than 12%. Previous studies on deformed metals and quartzites describe the structure of smaller grains encircling larger grains as "core-and-mantle" structure (Gifkins, 1976;White, 1976). The production of smaller grains that form the "mantle" region was considered as a result of continual rotation of subgrains to develop small strain free grains (White, 1976;Jacka and Li, 2000). Therefore, we suggest the subgrain rotation recrystallization alone is the

Inferences from the CPO development
The CPO intensity and opening-angle of the c-axis CPO decrease as the temperature drops. Previous studies suggest the CPO development is mainly controlled by the deformation and recrystallization mechanisms (Alley, 1992;Qi et al., 2017). Fig. 15 explains how key processes (Fig. 15(b)) involved in the deformation and recrystallization mechanisms (Fig. 15(a)) may affect the CPO development as a function of strain and temperature (Fig. 15(c)). 5 At -10 °C, the prevalence of larger grains with lobate grain boundaries and the increase in lobateness with grain size and with strain (as indicated by parameter Ψ), suggest that strain-induced GBM is an important process. Our -10 °C series CPO data show a continuous enhancement of CPO intensity as indicated by M-index, and a clearer cone-shaped pattern of the c-axes with increasing strain. Similar phenomena have been observed in previous ice deformation experiments, and they were used to infer that the strain-induced GBM favourites the growth of grains with easy slip orientations (Vaughan et al., 2017;Qi et 10 al., 2017). Because grains with hard slip orientations should have greater internal distortions (Bestmann and Prior 2003), and therefore store higher internal strain energy, they are likely to be consumed by grains with easy slip orientations through GBM (Piazolo et al., 2006;Killian et al., 2011;Qi et al., 2017;Xia et al., 2018). The cone opening-angles for samples deformed at -10 °C decrease from 40° at ~5% strain to 34° at ~20% strain, rather than stabilise at the easiest slip orientation of 45°. This observation suggests GBM alone cannot be the mechanism that controls the CPO development. The narrowing of cone-shaped 15 c-axis CPO has been explained by an activation of grain rotation (Vaughan et al., 2017;Qi et al., 2017). Similarly, we infer the decreasing of opening-angle as a function of strain is likely to result from an acceleration in the rate of grain rotation driven by intracrystalline glide on the basal plane. The observation of subgrains and the continuous boundary hierarchies suggest that recovery and subgrain rotation operated in parallel with strain-induced GBM. The dislocation activity required to generate subgrain structures and to provide the strain energy driving force for strain-induced GBM is likely the primary control 20 on grain rotation (Duval and Castelnau, 1995;Llorens et al., 2016). The CPO intensity (as indicated by M-index) of "small grains" is generally lower than "larger grains" (Fig. 12 (b)). At ~12% strain, both "big grains" and "small grains" develop cone-shaped c-axis CPO, but the c-axes cone is slightly clearer for "big grains". These observations suggest a mechanism that weakens the CPO development may be involved in the deformation of "small grains". Ice deformation experiments applied under comparable conditions suggest that grain size sensitive mechanisms are important in samples deformed at -10 °C (Qi et 25 al., 2017). Previous rock and ice deformation studies reported small recrystallized grains have CPOs that are randomly dispersed equivalents of the stronger host-grain CPOs (Jiang et al., 2000;Bestmann and Prior, 2003;Storey and Prior, 2005;Warren and Hirth, 2006;Craw et al., 2018). These observations are interpreted as the result of an increase in the contribution of GBS in fine grains. Therefore, we infer GBS, a grain size sensitive mechanism, may be more active in "smaller grains". "Big grains" are less lobate at lower temperatures, indicating a less effective strain-induced grain boundary migration (GBM). 30 The CPO data suggest the opening-angle of the c-axes cone as well as the CPO intensity decrease with decreasing temperature. We infer the selective growth of the grains oriented for easy slip orientations becomes less active due to the reduction of GBM activity. Consequently, grain rotation, driven by intracrystalline glide on the basal plane, becomes more https://doi.org/10.5194/tc-2020-2 Preprint. Discussion started: 22 January 2020 c Author(s) 2020. CC BY 4.0 License. prominent. The less effective GBM together with the more active grain rotation can lead to a closure of c-axis cone at lower temperatures. Microstructural data show bands formed by finer grains that encircle larger grains are better developed at colder temperatures. "Small grains" that contain all recrystallized grains and part of the remnant grains have a lower CPO intensity than "big grains", and this contrast strengthens from -20 to -30°C (Fig. 13). These phenomena suggest that the overall CPO weakening at lower temperatures corresponds to CPO weakening in "small grains". Craw and others (2018) reported similar 5 observations in uniaxially deformed Antarctic ice, and the reduction of CPO intensity in grains with finer sizes was attributed to GBS. Similarly, we suggest the grain size sensitivity of GBS (Goldsby and Kohlstedt, 1997;Goldsby and Kohlstedt, 2001) favours a faster strain rate in "small grains" relative to large grains.

Some thoughts on weakening/enhancement mechanisms
All experiments show weakening from peak stress to steady-state stress that correlates with CPO and microstructural 10 development. At -10 °C the CPO developed includes many grains with basal plane orientations that would facilitate further axial shortening and it is intuitive that the CPO development could provide a cause for the weakening. However, at -30 °C the CPO developed at steady-state stress is a cluster with many basal planes sub-perpendicular to compression. In this case the CPO would hinder further axial shortening and it is intuitive that the CPO should cause strengthening. Nevertheless, weakening occurs at -30 °C. Development of CPO cannot provide a uniform explanation for weakening across the range of laboratory 15 experiments presented here. In all cases grain sizes are reduced during deformation. The weakening associated with the increased strain rate contribution of grain size sensitive mechanisms as grain sizes reduce could provide a uniform explanation for weakening across the range of laboratory experiments presented here. This is a topic requiring significant further analysis.
2. In all samples stress rises to a peak stress at ~ 1 to 3% strain and then drops to a lower steady-state stress. Samples deformed at colder temperatures show higher peak and steady-state stresses, as expected for the temperature dependency of creep. 4. All deformed samples have skewed grain size distributions with a strong peak at small (<100 µm) sizes and a tail to larger sizes. The peak and mean grain sizes are small compared to undeformed samples (230 µm), suggesting nucleation is involved in dynamic recrystallization processes.
5. For each temperature series, we separated populations of "small grains" and "big grains". The "small grains" are smaller than subgrains for all samples deformed at -10 °C. "Small grains" and subgrains are similar in size at -20 and -30 °C. 5 These observations suggest bulge nucleation facilitates the recrystallization process at warmer temperature, and it becomes less important at colder temperatures. "Core-and-mantle" structure are clearer at -20 and -30 °C, suggesting greater activity of subgrain rotation recrystallization.
6. For all deformed samples, "big grains" are more lobate than "small grains". "Big grains" become more lobate with increasing grain size and with strain at -10 °C. The lobateness of "big grains" is close to constant across the grain size 10 range at low strains at -20°C and at all strains at -30 °C. These observations suggest grain boundary migration (GBM) is more prominent at warmer temperatures.

Many of the deformed samples have CPOs defined by open cones (small circles) of c-axes. At -10 °C, CPO intensity and
the definition of the c-axis cone increase as the strain increases, suggesting strain-induced GBM favours the growth of grains at easy slip orientations. The opening-angle of the cone decreases with strain, suggesting strain-induced GBM is 15 balanced by grain rotation facilitated by basal slip as the strain increases. The CPO intensity and opening-angle of the caxis CPO decrease as the temperature drops from -10 °C to -30 °C. The closure of c-axes cone at -30 °C is interpreted as the result of a more active grain rotation together with a less effective GBM. The overall CPO weakening with decreasing temperature correlates with CPO weakening in "small grains". We suggest grain boundary sliding (GBS), a grain size sensitive mechanism, becomes more important at colder temperatures. 20 8. Weakening (enhancement) cannot be uniformly caused by CPO development. It is likely that grain size reduction plays a significant role in weakening.
Author contributions. DJP, DLG and SF designed the research. DJP, SF, TH, AJC and CQ performed experiments. SF, DJP and MN collected the cryo-EBSD data. SF, DJP and JW analysed data. SF and DJP wrote the draft. All authors edited the paper. 25 Acknowledgements. We are thankful to Pat Langhorne for providing the cold room facility at University of Otago. This work was supported by a NASA fund (NNX15AM69G) and two Marsden Funds of the Royal Society of New Zealand (UOO1116 and UOO052). SF was supported by the University of Otago doctoral scholarship, the Antarctica New Zealand doctoral scholarship and the University of Otago PERT (Polar Environment Research Theme) seed funding.
Competing interests. The authors declare that they have no conflict of interest.    Qi and others (2017). 2 True axial strain. 3 Mean subgrain size. 4 Boundary misorientation angle. 5 Mean grain size. 6 Square mean root grain size. 7 Mean grain size of "big grains". 8 Mean grain size of "small grains". 9 Peak grain size in grain size distribution. 10 Peak grain size in subgrain size (with ≥ 2°) distribution. 11 Ratio between PQERR KKKKKKKK and 5 ̅ at ≥ 2°. https://doi.org/10.5194/tc-2020-2 Preprint. Discussion started: 22 January 2020 c Author(s) 2020. CC BY 4.0 License. * ̇ is the true axial strain rate, is the true axial strain, ̇ is the octahedral shear strain rate, is the octahedral shear strain, ̇ is the engineering axial strain rate, is the engineering axial strain, is the initial stress. latitudes separated by 4 degrees' interval were drawn. The points lying between the given co-latitudes (red area) are counted. The frequency density of the points is calculated from the normalised counts divided by the normalised area between the given co-latitudes. (c) The distribution of c-axes frequency density as a function of angle to the compression axis. The angle corresponds to the peak in the distribution is taken as the opening half-angle for the cone (small circle) shaped c-axes distribution. Throughout the text this is referred to as the opening-angle. 10 Figure 2. The stress-strain curves for all the deformed ice samples. The x-axis is the true axial strain (Eq. (2)). The y-axis is the uniaxial stress. The stress has been corrected for the change of sample cross-sectional area, assuming constant sample volume during the deformation. green line marks the threshold grain size between "big grains" and "small grains" (see text). (e) Distribution of subgrain size presented in 4 µm bins. The subgrain size is calculated by applying the boundary misorientation angle of ≥ 2°.       Figure 13. Comparison of contoured [0001] (c-axis) CPOs of (a) "big grains" and (b) "small grains", for the samples deformed 5 to ~12% strain at different temperatures. The numbers of "big grains" and "small grains" are marked on the bottom left of pole figures. The c-axis CPOs are calculated based on all pixels taken from the EBSD data with 5 µm step size. Compression axis is in the centre of the stereonets. https://doi.org/10.5194/tc-2020-2 Preprint. Discussion started: 22 January 2020 c Author(s) 2020. CC BY 4.0 License. Figure 14. Plot of the relationship between the opening-angle, , of the cone-shaped c-axis CPO and the true strain. The data 5 come from this study and the literature ( Table 4). The data from constant displacement rate experiments on D2O ice (Piazolo et al, 2013) are illustrated by hollow circles. The deformation of D2O ice at -7 °C is a direct analogue for deforming H2O ice at −10 °C (Wilson et al., 2019). The data from constant displacement rate experiments on H2O ice (this study, Vaughan et al., 2017, Qi et al., 2017, Craw et al., 2018 are illustrated by filled circles. Data from this study are highlighted by orange-black edges. The data from constant load experiments (Jacka and Maccagnan, 1984;Jacka and Li, 2000;Montagnat et al., 2015) are 10 illustrated by solid squares. Each marker is sized and coloured by the corresponding true strain rate and temperature, respectively. For all experiments the strain rate shown is the strain rate at the end of the experiment.