Physical and optical characteristics of heavily melted “rotten” Arctic sea ice

Field investigations of the properties of heavily melted “rotten” Arctic sea ice were carried out on shorefast and 1 drifting ice off the coast of Utqiaġvik (formerly Barrow), Alaska during the melt season. While no formal criteria exist to 2 qualify when ice becomes “rotten”, the objective of this study was to sample melting ice at the point where its structural and 3 optical properties are sufficiently advanced beyond the peak of the summer season. Baseline data on the physical 4 (temperature, salinity, density, microstructure) and optical (light scattering) properties of shorefast ice were recorded in May 5 and June 2015. In July of both 2015 and 2017, small boats were used to access drifting “rotten” ice within ~32 km of 6 Utqiaġvik. Measurements showed that pore space increased as ice temperature increased (-8 °C to 0 °C), ice salinity 7 decreased (10 ppt to 0 ppt), and bulk density decreased (0.9 g cm-3 to 0.6 g cm-3). Changes in pore space were characterized 8 with thin-section microphotography and X-ray micro-computed tomography in the laboratory. These analyses yielded 9 changes in average brine inclusion number density (which decreased from 32 mm-3 to 0.01 mm-3), mean pore size (which 10 increased from 80 m to 3 mm) as well as total porosity (increased from 0% to > 45%) and structural anisotropy (variable, 11 with values generally less than 0.7). Additionally, light scattering coefficients of the ice increased from approximately 0.06 12 cm-1 to > 0.35 cm-1 as the ice melt progressed. Together, these findings indicate that the properties of Arctic sea ice at the 13 end of melt season are significantly distinct from those of often-studied summertime ice. If such rotten ice were to become 14 more prevalent in a warmer Arctic with longer melt seasons, this could have implications for the exchange of fluid and heat 15 at the ocean surface. 16


Introduction
The seasonal evolution of Arctic sea ice follows a fairly predictable annual pattern: winter, snow melt, pond formation, pond 17 drainage, rotten ice [DeAbreu et al., 2001]. Considerable attention has been given to characterization of these various states 18 and their transitions. In situ observations during the summer melt season are typically straightforward through the pond 19 drainage stage, but, as ice conditions deteriorate, it becomes increasingly difficult to work on or around the most fragile 20 state, rotten ice. During the summer of 1894, Nansen, in his seminal work Farthest North (1897, p. 433) described it well, 21 "Everything is in a state of disintegration, and one's foothold gives way at every step." Extensive areas of rotten ice in the 22 Beaufort Sea pack were encountered in September 2009 [Barber et al. 2009], where the ice cover was found to be composed 23 of small remnants of decayed and drained ice floes interspersed with new ice. The remotely sensed radiometric 24 characteristics of this ice cover appeared indistinguishable from old, thick multiyear ice. Such characterization is largely 25 indicative of the physical properties of the ice on meter to decameter scales, but the microstructural properties of melting sea 26 ice at the very end of its summer melt remain largely undocumented. 27 The relatively high temperatures and abundant sunlight of summer cause sea ice to "rot". While the microstructure of winter 28 ice is characterized by small, isolated brine inclusions, with brine convection restricted to the lower reaches of the ice, and 29 spring ice is characterized by increased permeability and brine convection through the full depth of the ice cover [Jardon et 30 al., 2013;Zhou et al., 2013], the defining characteristics of rotten ice may be its high porosity and enhanced permeability. 31 Warming causes changes in the ice structure including enlarged and merged brine and gas inclusions (see, e.g., Weeks and 32 Ackley, 1986;. Columnar ice permeability increases drastically for fluid transport when the brine volume 33 fraction exceeds approximately 5% [Golden et al., 2007;Pringle et al., 2009]. In a previous study on shorefast ice, brine 34 volume fractions were found to exceed this 5% threshold for permeability through the entire depth of the ice from early May 35 onwards [Zhou et al., 2013]. While the term "rotten ice" is used in this manuscript to refer to heavily melted summer ice that 36 has diminished structural integrity, relatively large voids, and is highly permeable, it is also noted that this work is intended 37 to provide a more refined and quantitative definition of this ice type. 38 Connectivity of the pore space in sea ice is known to contribute to ocean-atmosphere heat transfer [Weeks and Ackley, 1986;39 Hudier et al., 1995;Lytle and Ackley, 1996;Weeks, 1998;Eicken et al., 2002], exchange of dissolved and particulate matter 40 [Freitag, 1999;Krembs et al., 2000] including nutrients [Fritsen et al., 1994], salinity evolution of the ice cover 41 [Untersteiner, 1968;Wettlaufer et al., 2000;Vancoppenolle et al., 2007], and surface melt pond distribution [Eicken et al., 42 2002]. As a result of this notable connectivity, rotten ice also has reduced structural integrity, which can have implications 43 for ice dynamics. Though it is known to have diminished tensile and flexural strength [Richter-Menge and Jones, 1993;44 Timco and O'Brien, 1994;Timco and Johnston, 2002], such details have not been well-characterized. Measurements by 45 Timco and Johnston [2002] demonstrated that in mid-May, the ice had about 70% of its mid-winter strength. By early June, 46 about 50% and by the end of June, 15%-20% of its mid-winter strength. The ice strength during July was only about 10% of 47 average in situ core temperatures (-5 °C in May, -2 °C in June, -1 °C in July), referred to subsequently in this text as 75 "working" temperatures. 76  Table A1). Flat, snow-covered ice 102 with no noticeable ridging was visible for many kilometers in all directions (Fig. 2a). Once cleared of snow (depth of 14−18 103 cm, -7 °C at 9 cm below the surface), the ice appeared flat and uniform. Snow was cleared prior to the collection of samples. 104 The measured ice thickness ranged from 141-150 cm at the sampling site, which is substantially thicker than the 105 cm 105 thickness reported at the nearby MBS. The uppermost ~10 cm (7 %) of the ice was above freeboard. At the time of our 106 sampling, the altimeter at the MBS had failed, so ice thickness was estimated from the thermistor string and was considered 107 to have large uncertainty. Ambient air temperature on the date of sampling was -9 °C at 11:00 AM. Samples collected for 108 analysis were subsectioned in the field at depths of 0-20 cm (top horizon), 32-52 cm (middle horizon), and the bottom 20 cm 109 of each core (bottom horizon). 110

June 2015
The second set of samples was collected on 3 June 2015 from within 30 m of the site sampled in May (Fig. 1b). The ice had 111 begun to form melt ponds (Fig. 2b), which we avoided during sampling. The June ice had thickness ranging from 149-159 112 cm, with ~21 cm above freeboard (14 %). It is likely that some of the increased ice thickness observed, compared to what 113 was measured in May, was due to the addition of retextured snow at the surface following a significant rain event during the 114 last week of May (SIZONet, 2017, observations for Utqiaġvik by Billy Adams, https://eloka-arctic.org/sizonet/), which 115 manifested as a layer of granular ice. Ambient air temperature on the date of sampling was -1.6 °C at 12:00 PM. Samples 116 collected for analysis were subsectioned in the field at depths of 0−20 cm (top horizon), 45-65 cm (middle horizon), and the 117 bottom 20 cm of each core (bottom horizon). 118

July 2015
In 2015, the landfast ice broke away from the local coastline during the third week of June (Fig. 1f). July samples were 119 drilled from isolated floes accessed by small boats within a radius of 32 km from Point Barrow (Fig. 1c). Floes in July varied 120 greatly in size, thickness, and character. 121 On 3 July 2015, the sea was ice-free within an ~8 km radius of Pt. Barrow; beyond this were regions of mixed ice, with both 122 sediment-rich and sediment-poor floes. Ice encountered near the barrier islands bounding Elson Lagoon included many 123 apparently grounded floes as well as some small (~7 m above freeboard), blue icebergs. Wildlife was abundant in the region, 124 with king and common eider, walrus, bearded seals, a grey whale, and a large pod of ~100 beluga whales observed. Cores 125 were drilled in three different floes: a thick (170 cm) "clean" floe (JY3-L1; with naming convention month (M, JN, JY), day 126 of month -location number), a small (~2 m 2 , 86-150 cm thick in the center) sediment-rich floe (JY3-L2), and a large, 127 heavily-ponded floe (JY3-L3; the single core collected from this floe measured 145 cm long). At all sites, freeboard depth 128 was difficult to determine due to the high variability in the underside of the ice, however, roughly 10-12% of ice cored was 129 above freeboard. Other than the variable thickness of the floes and high sediment content in some floes, the character of the 130 ice was similar to the ice observed in June. 131 Cores collected on 10 July 2015 (JY10) came from a sediment-rich, heavily-ponded floe with an ice thickness measured in a 132 non-ponded part of the floe of 190 cm. Ice in non-ponded areas was solid and saline, similar to what was observed in June 133 and on 3 July. Cores from ponded areas of the floe (collected from ponds ~18 cm deep) were visibly highly porous (rotten) 134 and ranged from 58-90 cm in length. During the course of two hours of sample collection, the floe began to break up under 135 light wave action (winds in the region increased from ~10 to 15 knots during the course of sample collection), forcing our 136 team to retreat to the boat. In one case, a crack developed that connected core holes drilled across a ponded area; in another, 137 a large crack developed across the width of the floe. 138 On 11 July 2015, additional cores (64-90 cm length) were sampled from a ponded area of a clean (sediment-poor) floe of 139 rotten ice (JY11). As with the 10 July floe, ice in non-ponded areas was solid and saline, partially drained but not heavily 140 rotted. The upper portion of the ice was pitted and drained. Ice beneath melt ponds (cores collected were submerged under 141 8-15 cm water) was heavily rotted and drained rapidly when cored. Ambient air temperature during sampling was -1. One core from each sampling site was used to measure vertical profiles of temperature, density, salinity, and pH. Ice 153 temperature was measured in the field immediately following core removal. The core was placed on a PVC cradle, and 154 temperature was measured using a field temperature probe (Traceable™ Total-Range Thermometer, Fisher Scientific; 155 accuracy ±1 ℃, resolution 0.1 ℃) inserted promptly into 3 mm diameter holes drilled into the center of the core at 5 cm 156 intervals. Horizontal pucks of the ice were then sawed at the 5 cm marks, and caliper measurements (± 0.01 cm) were taken 157 of two thicknesses and two diameters across the puck to estimate puck volume. Volume error values were calculated by 158 propagating relative variability in the thickness and diameter measurements. Pucks were then sealed in Whirlpak bags and 159 returned to the lab, where mass and salinity measurements were taken of melted pucks using a digital scale and conductivity 160 meter (YSI Model 30, accuracy ± 2 %, resolution 0.1 ppt). Bulk density was computed from the measured mass (accuracy ± 161 0.1 g) and estimated puck volumes. 162

Thin section microphotography
Representative horizontal and vertical sections were prepared from each horizon of ice for each of the three time points 163 sampled in 2015. Thin sections (~ 2 mm thickness) were prepared using a microtome (Leica), with the exception of some 164 July samples that were too fragile for microtome cutting. These fragile sections were cut as thin as possible on a chop saw (~ 165 1 cm thickness). All cut sections were then photographed on a light table at working temperatures for each month as well as 166 at -15 °C. An LED epifluorescence microscope (AxioScope.A1 LED, Carl Zeiss, with EC Plan-Neofluar phase contrast 167 objectives) specially adapted for cold room work was used to image the thin section samples. Transmitted light 168 photomicrograph mosaic images were constructed from 50x magnification snapshots taken at the working temperatures for 169 each time point. Image software (ImageJ, Adobe Photoshop CC) and manual image analysis were used to highlight pore 170 spaces for pore size analysis. 171

X-ray micro-computed tomography
To prepare samples for x-ray micro-computed tomography ("micro-CT") imaging, 10-cm subsections of ice cores returned 172 from the field were stored overnight in insulated coolers in a walk-in freezer set to working temperatures. Subsections were 173 then placed upright in Teflon centrifuge cups (500 mL bottles with tops cut off) and spun out at -5 °C for 5 minutes at 1500 174 rpm to remove brine using a Thermoscientific S40R centrifuge. Samples were kept at working temperature right up to the 175 time they were centrifuged. The 5 minutes in the centrifuge at -5 C was assumed to be brief enough that sample 176 temperatures, and thus brine volume, were not significantly altered. 177

178
The masses of brine and spun-out ice were determined, and brines saved for later biological and chemical analysis. Spun-out 179 ice horizons were returned to the working-temperature walk-in freezer, where they were then placed upright on top of 180 corrugated cardboard circles placed inside the Teflon centrifuge cups. Samples were casted with dimethyl phthalate (DMP) 181 in an attempt to minimize structural changes during transport, storage, and processing and in order to use methods consistent 182 with prior micro-CT work on snow. Working temperature dimethyl phthalate (DMP) was then carefully poured down the 183 sides of the container in order to flood the ice samples and form casts of the brine networks in contact with the borders of the 184 ice core as described by Heggli et al., [2009] for casting snow. The DMP was left to penetrate brine networks and slowly 185 freeze at the working temperature for at least 12 hours before freezing fully at -20 °C. Casted cores were then removed from 186 the Teflon cups, sealed in Whirlpak bags, and stored at -20 °C for later micro-CT imaging. In addition, several archived 187 cores from July 2015 that had been stored at -20 °C were scanned without casting to assess the effect of DMP casting on 188 tomography measurements. 189 Prepared samples were imaged at the U.S. Army Cold Regions Research Laboratory using a micro-computed tomography 190 high-energy x-ray scanner (SkyScan 1173, Bruker) housed in a -10°C walk-in freezer. Scans were run at 60 kV, 123 µA, 191 with a 200 ms exposure time and 0.6 ° rotation step. The nominal resolution was set to 142 µm pixel -1 in a 560 x 560 pixel 192 field of view, which permitted fast scan times (18 minutes), resulting in low exposure of samples to excess radiation and 193 egregious warming (scanner chamber temperatures were recorded as ~2 °C during runs). 194 Shadow images generated by micro-CT were reconstructed into 2D horizontal slices using the software NRecon (Bruker). 195 Thermal abnormalities were corrected by performing x/y alignment with a reference scan. Samples with x/y shifts greater 196 than |Δx| = 5 were re-scanned. Following x/y alignment, reconstructed image histograms of linear attenuation coefficients 197 were clipped to 0.000 -0.005 and the following correction factors were applied: 50 % defect pixel masking, 20 % beam-198 hardening correction, smoothing level 2 using Gaussian kernel. Post-alignment shifts were determined manually and were 199 between -2 and 2. The ring-artifact reduction parameter was also chosen manually to minimize artifacts and was between 2 200 and 10 for all processed samples. 201 Reconstructed 8-bit 2D images were analyzed using the software CTAn (Bruker). Cylindrical subvolumes (height = 4.0 cm, 202 diameter = 4.97 cm) centered on the scanned sample's z-axis were selected from the original scanned samples and positioned 203 to capture a representative segment of the sample, avoiding sample edges. Reconstructed images were parsed into four 204 phases using brightness thresholding determined manually at well-defined phase local minima for each scan: air (black), ice 205 (dark grey), DMP (bright grey), and brine (bright). Phases were manually parsed using cutoffs based on greyscale intensity 206 histograms picked by a single analyst. A preliminary sensitivity analysis indicated that manual thresholding by a single 207 analyst was found to give more reliable results than automated thresholding methods due to relatively large variability in 208 brightness and contrast in reconstructed images as well as poor brightness separation between the ice and DMP phases. 209 Noise reduction was then applied using a despeckle of 8 voxels for ice, brine, and air, and 10 6 voxels for DMP (a high 210 despeckle value ensured that only DMP-thresholded regions that connected to subvolume boundaries were included, as any 211 DMP "islands" are, by definition, artifacts). During the casting using DMP, air bubbles were trapped inside the solidifying 212 DMP. Due to the brightness order of phases (air > ice > DMP > brine) the gradient between air and DMP is incorrectly 213 identified as ice creating thin ice "rings" inside DMP regions of the 2D slices. This problem was resolved by using a 214 morphological operation called "closing", where thin (1-2 pixel) threads of ice were dilated then eroded, thus removing the 215 features [Soille, 2003]. Ultimately, DMP casting introduced artefacts in the analyzed samples, so the analyses presented in 216 the results of this manuscript focus only on the ice phase and the combined air + brine + DMP ("not-ice") phases. 217 CTAn was then used to calculate properties of the parsed phases, including 3D volume, number of 3D objects, closed and 218 open porosity, and anisotropy. A description of the mathematical basis for these parameters as well as detailed best-practice 219 methods for micro-CT imaging of sea ice can be found in Lieb-Lappen et al. [2017]. 220 Further, 3D prints of the reconstructed ice-only phase were made from the micro-CT reconstructions using polylactic acid 221 fused deposition modeling (Flashforge Creator Pro, FDM print with Makerbot print program and layer height 0.1 mm). 222

Optical properties
Field measurements of optical properties are generally limited to estimation of apparent optical properties (AOPs), e.g., 223 albedo, transmittance, and extinction. Due to the tenuous working conditions on rotten sea ice floes and instrument reliability 224 problems, we were not successful at obtaining estimates of in situ AOPs of rotten ice. Instead, we focused on assessing the 225 optical properties of extracted ice samples in the laboratory. Inherent optical properties (IOPs), such as scattering and 226 absorption coefficients and scattering phase functions, are intrinsically difficult to measure in multiple-scattering media, but 227 estimates from laboratory measurements can build a picture of the evolution of sea ice optical properties. In fact, estimates of 228 IOPs are particularly useful since they are independent of boundary conditions (e.g., ice thickness and floe size) and the 229 magnitude, directionality, and spectral character of the incident light field (see e.g., Katlein et al., 2014;Light et al., 2015). 230 The evolution of light scattering coefficients for sea ice as it melts determines the partitioning of solar radiation in the ice- To track the evolution of how the ice in this study partitioned sunlight, a laboratory optics study was carried out. Cores for 236 optical property assessment were sampled alongside cores for other characterizations, returned to the lab, and stored intact at 237 -20 °C. The May and June cores were stored for 2-3 days prior to running the optics experiments. The July cores were 238 shipped back to the freezer laboratory at the University of Washington and stored for 16 months prior to optical 239 measurement. 240 To carry out optical property assessment, each core was cut into 10 cm long sections. Each section was placed in a chamber 241 for the measurement of light transmittance using a technique developed to infer inherent scattering properties of a sea ice 242 sample from a simple measurement and a corresponding model calculation (see Light et al., 2015). Figure 4 shows a 243 schematic of this laboratory measurement, where ice samples are placed in a dark housing and illuminated from above. 244 Spectral light transmittance between 400 and 1000 nm wavelength of each subsample was recorded relative to the 245 transmittance through pure liquid water. The relative transmittance was then compared with results from numerical radiative 246 transport simulations using the model described by  for a wide range of scattering coefficients. The 247 scattering coefficient producing relative transmittance (at 550 nm) closest to the observed relative transmittance was then 248 chosen. When subsamples from a full length of ice core are measured, this technique estimates the vertical profile of the light 249 scattering coefficient through the depth of the ice. By directly assessing scattering coefficient, an IOP, we avoid 250 complications introduced by the interpretation of AOPs (e.g., albedo, total transmittance measured in situ), notably 251 differences in total ice thickness and incident solar radiation conditions (e.g., diffuse/direct), as well as other physical 252 boundary conditions. In each case, samples taken from ice sitting below freeboard were placed into the sample chamber and 253 then gently flooded with a sodium chloride and water mixture in freezing equilibrium (temperature and salinity) with the 254 sample. Light transmission was measured while the sample was flooded. Sample measurement was fast, with each sample in 255 the chamber for less than one minute. It is probable that the liquid did not completely fill the pore structure of the ice 256 samples, however, the visible appearance of the samples indicated a dramatic reduction in backscatter during the flooding 257 process, suggesting that flooding was effective. 258 Samples were run in two modes. In the first mode, samples were analyzed promptly after removal from the ice. These 259 samples represent snapshots of the rotting process as it occurs naturally. The second mode was run in attempt to use light 260 scattering measurements to inform our understanding of ice rotting processes. To do this, an archived May core was cut into 261 10 cm thick sections and placed in an insulated box in the freezer laboratory. The sections were stored standing upright and 262 were placed on a wire rack such that the melt water drained away from the remaining sample material. Initially, the freezer 263 temperature was set to -8 °C, but once the experiment commenced, the temperature was increased gradually every 24 hours. 264 The sample density and vertical scattering profile were measured at each temperature step (-6, -5, -4, -3, -2 °C over a one 265 week period.) This attempt to artificially rot the ice was documented using the optical measurements with the hope that such 266 a measurement would inform our efforts to simulate rotten ice in the laboratory. Uniformly thin floes were rotten throughout in both ponded and non-ponded regions. Visually, rotten ice was devoid of the 293 microstructural inclusions that characterized the May and June ice interior, instead appearing to have large, isolated pores, 294 and a more chaotic structure. When cored, rotten ice crumbled or broke at many points along the length of cores, rendering it 295 difficult to handle. Rotten ice drained copiously when cores were removed from drill holes, and the bottom portion of rotten 296 cores consisted of optically clear, fresh ice drained of brine and characterized by large (cm-scale) voids. Figure 6 shows 297 photomosaics of cores sampled on 14 July 2015 at Location 3. Images show variations in ice texture depending on whether 298 the ice was ponded or non-ponded, although both types do appear to have at least some scattering layer with bright white 299 appearance. Many cores had holes exiting the bottom of the ice that were large enough to stick a finger into, although we 300 did not have a means to quantitatively assess how vertically extensive these drainage tubes were. 301 302 Figure 5. Photomosaics of representative cores collected and analyzed in this study showing the sequence of rot. Core names correspond to samples discussed elsewhere in this paper and are coded by sample site (as shown in Figure 1). The measured ice thickness at each core hole is indicated. For the JY14 samples, measured core length is indicated instead of ice thickness. Due to variability in the ice bottom, spreading or compression of weak layers, and artefacts of image stitching, core images, which are shown to scale, may not match the measured ice thickness. Asterisks (*) indicate cores collected from submerged ice. Note the brown algal bloom layer visible in the bottom of the May core and faintly visible at the bottom of the June core, and the bright layer of retextured snow at the top of the June cores.

Temperature, salinity, and density profiles
The May temperature profile had values as low as -8 °C at the snow-ice interface; below that, temperatures increased with 312 depth (Fig. 7). By June, the entire depth of the ice had warmed above -1 °C, with the lowest temperatures measured in the 313 middle sections of cores. By July, the ice was approximately isothermal, with temperature 0 °C. These profiles generally 314 agree with observations at the MBS and are typical of other investigations in the area (e.g., Zhou et al., 2013). 315 Bulk salinity profiles (Fig. 7b) were also consistent with prior published observations. May ice showed the classical C-

Thin section microphotography
Thin sections show the evolution of the ice structure as it warmed (Fig. 8). Each of the microphotographs in Fig. 8

X-ray micro-computed tomography
Calculations done on 3D reconstructions generated from micro-CT show a significant evolution in the internal structure of 356 ice during the course of melt and help define "rotten" ice. Figure 9 shows reconstructions of the ice-only phase (top row), 357 reconstruction of the not-ice phase (air + brine + DMP) with objects of different sizes color coded as blue (<0.11 cm 3 ), green 358 (0.11-1.15 cm 3 ), and red (>1.15 cm 3 ) showing the evolution toward larger pores and channels in rotting ice (middle row). 359 Note that micro-CT analyses only resolve structures with a short dimension > 284 μm, (derived from the 8 voxel despeckle 360 that was applied) which is significantly larger than the average inclusion size observed in the microscope imagery for both 361 May and June. The bottom row shows monochrome photographs of 3D prints made from the four reconstructions. 362 Porosity is defined as the percentage of total volume occupied by pores, as measured from the ice-only phase perspective 363 such that the porous space is derived from air, brine, and DMP. Porosity in DMP-casted May and June horizons (excluding 364 June top horizons determined to represent a retextured snow layer) ranged from 0.5-7.5 % by volume (Fig. 10a). In contrast, 365 the DMP-casted rotten core (JY11-06) had a range in porosity of 37.5-47.9 %. For non-casted rotten cores measured, the 366 porosity ranged from 7.6-23.1% (mean = 15.5 %) in a sample collected from below a melt pond (JY11-19), and from 5. In addition to becoming generally more porous, the nature of pores in the ice changed as melt progressed (Fig. 10b). Open 371 pores were those pores connected to the exterior surface of the volume analyzed, while closed pores were those fully interior 372 within the 77.6 cm 3 volumes analyzed. In May, closed pores comprised 26-72 % of the total pore volume (mean = 51%). In 373 June, the percent by volume of closed pores was similar (mean = 42 %) except for the uppermost retextured snow layer (0-3 374 % closed pores by volume). In July, this was markedly changed: >74 % of pore volume in all samples (casted and uncasted) 375 addition, the number of closed pores in the normalized unit volume decreases from May and June to July (Fig. 10c). The 378 May cores and June middle horizons have the highest closed pore densities (31-94 cm -3 with mean = 61 cm -3 , and 2-63 cm -3 379 with mean = 26 cm -3 in May cores and 44-63 cm -3 measured in June middle horizons). In June, the density of closed pores in 380 the top and bottom (8-11 cm -3 , and 2-10 cm -3 , respectively) decrease, creating a reverse C-shaped profile. In July, the 381 density of closed pores is uniformly low throughout the cores measured (16-36 cm -3 with mean = 24 cm -3 , and 1-16 cm -3 382 with mean = 6 cm -3 in casted and non-casted July cores, respectively). Both metrics indicate that connected (open) pores 383 dominate in July. This follows from larger pore sizes, as quantified by the 2D structure thickness metric, which measures the 384 mean maximum diameter of 3D objects. In May and June, pores averaged <5 mm along their longest axis (1.7-3.3 mm, 385 mean = 2.4 mm, again with the exception of the June retextured surface snow and a 32 mm outlier value in one June middle 386 horizon). In rotten July cores, pores enlarge substantially (4.2-17.0 mm, mean = 8.1 mm). The trend toward more connected 387 pores is most pronounced in the upper-and lowermost layers of the core. Pasta shells (rounded) would be a good way to visualize an isotropic assembly of pores. Spaghetti (pre-cooked) is clearly 417 anisotropic. However, the strongest anisotropy could be represented by pre-cooked spaghetti still in the box. If the uncooked 418 spaghetti were spilled on the floor, it would become more isotropic, even though each individual piece is anisotropic. 419 Spaghetti in the box is a good analogy for the pore spaces in the mid-horizon. Horizontal connectivity in the bottom horizon 420 makes that pore space less anisotropic. 421 In the rotten July cores, the C-shaped profile disappeared entirely. In the JY11 sample analyzed (from ponded ice), the 422 middle portion of the core became more isotropic (0.38-0.57 in the DMP-casted sample, 0.24-34 in the uncasted sample), 423 indicating a rounding of the core center brine channels. This trend was not apparent in the JY14 (thinner rotten floe of non-424 ponded ice) sample, however, in all July cores analyzed, the upper layer had a generally greater anisotropy value than core 425 middle values, perhaps indicative of vertical channel formation in the upper portion of the ice due to melt and draining from 426 the upper portion of the ice. 427

Optical properties
As the sea ice cover progresses through the onset and duration of melt season, its optical properties respond to increased 428 temperature and the absorption of increasing amounts of solar radiation. Typically, the albedo of the ice cover decreases (less 429 light backscattered to the atmosphere) and its transmittance increases (more light propagating into the ocean). The bulk of 430 this effect, however, is due to the loss of accumulated snow and the widespread formation of melt puddles on the ice surface 431 . While this net effect dominates the surface radiation balance, it overlooks effects due to changes in 432 the properties of the ice itself. As the ice warms and becomes porous, permeable, and rotten, increases in void space increase 433 the total amount of internal ice / liquid and ice / air boundary, and would thus be expected to increase total scattering. 434 Increases in ice scattering should promote higher albedo and lower transmittance-exactly opposite the behavior of the 435 aggregate ice cover. 436 The results of the laboratory optical measurements are shown in Fig. 11. Vertically resolved profiles of scattering coefficient 437 are shown for ice obtained in April, May, June, and July. The April ice was extracted in the same vicinity as the May and 438 June samples during an unrelated field campaign in 2012. In addition to the temporal trend of sampled ice, optical property 439 assessment was also carried out for a May sample subjected to controlled melt in the laboratory (open circles). Scattering 440 coefficients generally increased with time and individual profiles were typified by the characteristic c-shape (higher 441 scattering at top and bottom of the column, lower scattering in the middle) also seen in typical salinity profiles. 442 443 Figure 11. Vertically resolved scattering coefficients of sea ice measured during each phase of the field campaign. Coefficients are inferred from laboratory optical transmittance measurements (after Light et al., 2015) and interpretation of a radiative transport model in cylindrical domain . April profile included to show spring ice was measured on ice sampled in 2012 at a comparable geographic location. The May lab rot profile is for ice extracted in May during field campaign, and then warmed in the lab prior to sub-sample preparation. Shaded area shows the range of measurements on melting multiyear ice (Light et al., 2008) and melting first-year ice (Light et al., 2015).

Physical characteristics of rotten ice
As sea ice warms, its microstructure changes as inclusions of brine and gas enlarge as required to maintain freezing 445 equilibrium. This has been well established theoretically [Cox and Weeks, 1983], as well as in laboratory experiments 446 [Perovich and Gow, 1996; for ice with isolated inclusions of brine and gas. This study addresses the 447 limits of sea ice microstructure when natural ice is in advanced stages of melt, where these inclusions are typically no longer 448 isolated, but rather are in connection with the ocean and/or the atmosphere. 449 The equations of Cox and Weeks [1983] describe the phase relations of sea ice for temperatures less than or equal to -2 ℃ 450 and where the bulk ice density describes a volume containing liquid brine and gas−both in equilibrium (freezing equilibrium 451 with the ice in the case of brine and phase equilibrium with the brine in the case of gas). Lepparanta and Manninen [1988] 452 expanded this treatment to include temperatures above -2 ℃. In the case of ice in advanced melt, the ice temperature would 453 be expected to be always close to 0 ℃. Furthermore, most sample volumes will typically include void spaces that are in 454 connection with the atmosphere or ocean and hence may not conform to the requirements of freezing or phase equilibrium 455 (e.g., brine inclusion size will not necessarily shrink if the temperature decreases). As a result, expected changes in the 456 microstructure−and ultimately, the mechanical behavior−of sea ice at most times of the year should not be expected to 457 pertain to changes experienced during late summer. 458 Photos of ice core samples shown in Fig. 5 illustrate the evolution of the ice structure. Early in the season, the majority of the 459 interior ice (areas away from the top and bottom) appears mostly translucent and often milky with the exception of isolated 460 bright, bubble-rich weak layers. As the season progresses, more of the ice appears opaque, losing its transparency (Fig. 5). 461 This highly scattering ice results from merging, connecting, and draining inclusions. This effect is clearly seen in the cores 462 that were submerged when extracted (e.g., the cores indicated with * in Fig. 5, and cores shown in Fig. 6), but can also be 463 seen in the JY13-L1 and JY13-L2 cores, which were not submerged when sampled. 464 Submerged cores appear to have more porous ice structure. We hypothesize this is due to additional heating of submerged 465 ice. This heating may come as a result of increased absorption of radiation as swamped or ponded ice will not maintain a 466 substantial surface scattering layer, and as a result, its albedo is typically lower [Light et al., 2015], and more sunlight is 467 absorbed within its interior. Or it may result simply from the contact between this ice and sunlight-warmed water. It is also 468 possible this additional melting serves to enhance the connectivity of this ice to the ocean, promoting the invasion of 469 seawater−and any associated heat−from beneath. 470

Temperature, salinity, and density profiles
Rotten ice is isothermal, having warmed to approximately the freezing temperature (0 ℃) of fresh water. Correspondingly, 471 core samples of rotten ice extracted from the ocean typically drain any associated liquid rapidly. Accordingly, this ice is 472 much fresher than earlier-season ice, with salinity values < 3 ppt through most of the core, indicative of a loss of much of the 473 brine that characterizes earlier-season ice (see Fig. 7b). The May salinity measurements show the classic 'c-shaped' salinity 474 profile indicative of first-year ice yet to experience summer melt. By June, the salinity profile shows freshening at the ice 475 bottom, likely associated with the onset of bottom ablation. It is also possible that this freshening resulted from increased 476 brine drainage during core sampling of ice with enlarged pore space. However, the optical transparency of the bottom 477 portion of the ice when sampled as well as the micro-CT data imply that little closed porosity remains in rotten ice-the ice 478 is snaked through with large drainage tubes. Additionally, the top of the June ice shows significant freshening. In this 479 particular year at this location, this change is likely related to the presence of retextured snow at the ice surface, which would 480 be expected to be very fresh. It may also result, in part, from the onset of surface ablation and the ensuing fresh water 481 flushing that would be expected this time of year. The July ice was almost completely devoid of salt. This is expected, due to 482 the prevalence of a connected pore structure and the significant flushing and drainage of virtually all salt in the ice. 483 Density profiles (Fig. 7c) reflect changes in temperature, bulk salinity, and structure. We observed a marked decrease in 484 density corresponding to summer melt, a result of the dramatic increase in porosity that defines rotten ice. May and June 485 profiles had density measurements centered around 0.9 g cm -3 and showed little variability except for reduced density in the 486 upper portions of the June core, likely resulting from the prevalence of the observed retextured snow. July profiles had even 487 further reduced density, with values reaching as low as 0.6 g cm -3 , reflecting void spaces in the ice following the rapid 488 draining of seawater from the ice, and was much more variable. For comparison, the density of core horizons (measured 489 using the same technique) taken in melting Arctic pack ice in July 2011 had similar values between 0.625-0.909 g cm -3 490 [Light et al., 2015]. Normally, sea ice with significantly smaller bulk density would be expected to float higher in the water 491 and thus have larger freeboard. But the density reductions that occur during advanced melt result from large void spaces 492 within the ice that are typically in connection with the ocean. As a result, such ice can have small freeboard, even if total ice 493 thickness is still relatively large. 494 It is worth noting that sediment loading did not appear to influence the density and structure of rotten ice. Rotten cores 495 collected on 10 July 2015 came from a floe with a visibly high sediment load, while rotten cores collected on 11 July 2015 496 and in July 2017 had much less sediment (Fig. 5). For all July cores, measured density values were similar within the large 497 range of measurement error. Salinity in the core collected from a sediment-rich floe was, however, somewhat higher than the 498 cores collected from "clean" floes. 499

Internal structure: porosity, connectivity and implications of rot
The number and size of brine inclusions identified in this study through the microscope imagery is commensurate with the 500 number and size of inclusions documented by . That study reported a brine inclusion number density range 501 of 24 mm -3 to 50 mm -3 from ice sampled in May, offshore from Utqiaġvik in a similar vicinity as the present study. The 502 number densities observed in May ice in the present study were 32 mm -3 in May, well within the range identified by the 503 earlier study. The earlier study showed brine inclusion number densities to decrease with increasing temperature, up to a 504 point, but did not follow the ice into advanced melt. The present study documents decreases in inclusion number density 505 from 32 mm -3 in May to 19 mm -3 in June to 0.01 mm -3 in July. While these values are consistent with the earlier findings, 506 they also extend the results much further into melt than has been previously attempted. In particular, the micro-CT work is 507 useful for sampling much larger sample volumes, and thus central for estimating size and number distributions for the July 508 ice. 509 Porosity (Fig. 10a) is low in May, with values less than 10 %, and increases as the ice warms and melts. By July, the micro-510 CT-determined porosity approached 50 %, commensurate with densities measured as low as 0.6 g cm -3 and our general 511 observations that this ice was highly porous, containing obvious channel structures with that were clearly connected. There 512 were differences in the handling of cores used for direct density measurement and cores used for micro-CT imaging. In 513 particular, cores used for density measurement were extracted from the ice immediately prior to measuring their dimensions. 514 In contrast, samples taken for micro-CT imaging spent several hours transiting to the laboratory, which may have enhanced 515 brine loss and structural change. In addition, samples casted for micro-CT imaging were centrifuged prior to casting. It 516 would thus be expected that the micro-CT-derived porosity measurements could yield estimates with less included fluid than 517 the density measurements made closer to in situ conditions. Similarly executed micro-CT measurements have quantified 518 included air volumes in growing winter sea ice [Crabeck et al., 2016], where the gas phase was clearly distinguished from 519 the brine phase, but the total pore space did not increase above 11 %, which is far smaller than the ultimate pore space 520 observed in this study. 521 The permeability, and hence pore structure, is central to the hydrological evolution of summer sea ice [Eicken et al., 2002]. 522 This suggests that the documentation of highly permeable ice with large porosity may be central to understanding the mass 523 balance of modern ice covers late in the summer melt season. In particular, Eicken et al. [2002] outlined a mechanism for 524 significant ice melt whereby warmed surface waters run off the ice and accumulate beneath areas with shallow draft late in 525 summer, and this pool of warmed fresh water experiences convective overturn and is entrained within the open structure of 526 melting ice. It is expected that further melting from this additional heat could exacerbate the decay and structural frailty of 527 the melting ice, literally melting it from the inside out. 528 The pore anisotropy results shown in Fig. 10d reinforce the overall trend that as the season progresses, the ice structure 529 homogenizes, losing its characteristic c-shape. Where strong vertical gradients in anisotropy existed in May and June, the 530 July ice is more uniform. Our findings are consistent with those of Jones et al. [2012], which used cross-borehole DC 531 resistivity tomography to observe increasing anisotropy of brine structure as early spring (April) ice transitioned to early 532 summer (June) ice. In that work, the brine phase was found to be connected both vertically and horizontally and the 533 dimensions of vertically oriented brine channels gradually increased as the ice warmed. 534 There remain notable limitations associated with the characterization of sea ice using micro-CT techniques. Many small 535 brine inclusions were not counted owing to the limited spatial resolution of the technique. Furthermore, the casting technique 536 that was employed appears to have introduced artifacts, especially in connectivity. From all the derived properties (porosity, 537 connectivity, and anisotropy), it appears that the introduction of the casting media may have forced channel connections 538 where perhaps they did not exist naturally. However, the trend in casted samples and the values measured for uncasted 539 samples reflect the substantial changes in ice character that are apparent in the field. 540

Optical evolution of rotting ice
Increases in effective light scattering coefficient over the course of seasonal warming are shown to be approximately 5-fold 541 for the interior ice studied here (Fig. 11). The overall trend of increasing scattering with time as the melt progresses is a 542 result of the connecting and draining microstructure, as assessed in the microstructure and tomography analyses. Relative 543 increases in the scattering would be expected to scale by the inclusion number density multiplied by the square of the 544 effective inclusion radius (see . Observed mean inclusion sizes increased from average May size of 80 m 545 to average June size of 221 m to average July size of 3 mm. Observed number densities decreased from 32 mm -3 (May) to 546 19 mm -3 (June) to 0.01 mm -3 (July). These changes correspond to relative scattering coefficient magnitude changes of 1: 4.5: 547 0.4, which would predict a scattering coefficient increase from May to June by a factor of 4.5, and a decrease in July by 548 more than half. The increased scattering shown in Fig. 11 from May to June is consistent with this observed average size 549 increase, but there is no decrease seen in July scattering. The large variability in both size and number for July makes 550 prediction of observed scattering increases very problematic. This suggests that when the ice is truly rotten and porous, and 551 the pores are very large, as was observed in July, that light scattering cannot be well represented by a simple evaluation of 552 average pore size and number density. 553 Early in the season, the larger scattering near the ice bottom likely reflects the higher brine content (larger and/or more 554 numerous brine inclusions) near the growth interface. The larger scattering near the top ice surface likely results from the 555 less organized ice structure that forms prior to the onset of congelation growth during initial ice formation. As the melt 556 season progresses, this uppermost portion of the ice has additional enhanced scattering due to the drainage of above-557 freeboard ice and the eventual development of a surface scattering layer. The enhanced scattering at the top and bottom of 558 the ice results in a C-shaped profile, consistent with observed salinity profiles. This C-shape appears to dominate the profiles 559 for April, May and June, but the July sample appears to have no memory of the characteristic C-shape found earlier in the 560 season. Given the significant structural retexturing that occurred by July, this should not be surprising. 561 Laboratory optical measurements made analogously to the ones in this study were carried out for melting first-year sea ice in 562 the open pack (see Light et al., 2015). That data set included little information about the temporal progression of the ice, as 563 no one location was sampled more than once. However, interior ice scattering coefficients between 0.1−0.3 cm -1 were found 564 for that ice in June and July, and these values are comparable to what was found in this study. 565 In an effort to use light scattering measurements to inform our understanding of ice rotting processes, we monitored the 566 optical properties of natural ice samples as they melted. Since most of the May core had in situ temperature > -5 ℃, only 567 small changes in sample density and light scattering properties were observed until the ice warmed to -2 ℃ (Fig. 11, dashed  568 curve). The lab-rot core shows significantly enhanced scattering, although not as large as the naturally rotted ice. This was 569 viewed as a preliminary attempt to create rotten ice in the laboratory. Differences between ice rotted in air and floating in the 570 ocean would likely be the rate of rot, and the relative abundance of gas-filled pore space relative to liquid pore space. 571 Refractive index contrasts mean that gas pores scatter more effectively than brine filled pores; thus, lab-rotted samples were 572 flooded in order to best mimic in situ rotted ice. 573

Conclusions
As Arctic sea ice melts during the summer season, its microstructure, porosity, bulk density, salinity, and permeability 574 undergo significant evolution. In situ measurements of sea ice documented off the northern coast of Alaska in May, June, 575 and July, indicate that sea ice transitioned from having 4−10 ppt salinity in May to near zero salt content in July. The ice 576 became extremely porous, with porosity values exceeding 10 % through most of the depth of the ice compared to <10 % for 577 ice collected in May and June. Some July porosity values approached 50 % at places in the ice interior. Brine pockets in 578 rotten ice are few; the ice is essentially fresh in composition and characterized by large, visible voids and channels on the 579 order of several millimeters in diameter. These changes result from increased air temperature, ocean heat, and prolonged 580 exposure to sunlight and leave the ice with dramatically increased porosity, pore space with increased connectivity, and 581 increased capacity to backscatter light. These changes have potential implications for the structural integrity, permeability to 582 surface melt water as well as ocean water, light partitioning, habitability, and melting behavior of late summer ice. 583 Specifically, increased connectivity with the ocean may affect how material (e.g., dissolved and particulate material, 584 including biological organisms and their byproducts) is exchanged at the ice/ocean boundary. Subsequent surface meltwater 585 flushing may in turn effectively rinse these constituents from the ice, making this enhanced connectivity central to the 586 control of ice-associated constituents well into the summer season. Rotten ice is a very different physical and chemical 587 habitat for microbial communities than earlier-season ice. 588 Reductions in bulk density were observed to occur from values approximately 0.90−0.94 g cm -3 to values as low as 0.6 g cm -589 3 . Pore spaces within this low density ice, however, were typically well connected to the ocean. This left the low-density 590 summer ice to generally have very small freeboard and frequently be flooded by ambient seawater. Finally, and significantly, 591 field observations stress the lack of structural integrity of this porous, fragile ice, indicating that thickness-based models of 592 ice behavior may not accurately predict the behavior of late-season sea ice. 593 In addition to sampling naturally rotted sea ice, we have also attempted to simulate the rotting process in the laboratory. Our 594 laboratory optics measurements suggest that natural samples extracted early in the season can be at least partially rotted in 595 the laboratory. To achieve ice that is as rotted and structurally compromised as was observed to occur in nature, the 596 absorption of solar radiation may be a necessary parameter. Sunlight is key to the formation of surface scattering layers at 597 the air−ice interface. In the lab, ice was permitted to rot in air, so any melt that was produced would quickly drain away. In 598 nature, the ice necessarily floats in its own melt, and this may be a critical difference in the way that heat is delivered to the 599 ice. Increases in melt season length may bring increased occurrence of rotten ice, and the timing and character of the 600 seasonal demise of sea ice may be related to the evolution of the ice microstructure. and SF analyzed the microstructure images. CF performed micro-CT measurements, and the micro-CT analyses were 609 designed and conducted by CF, SF, RL, and ZC. CF and BL prepared the manuscript with contributions from all co-authors.