Morphological and Spectral Characterization of Lunar Regolith Breakdown due to Water Ice

Remote sensing observations of the Moon suggest that the lunar polar regolith environment is affected by several natural processes that may cause the regolith in these regions to become more porous and fine particulate. One of these processes may be the mechanical breakdown of regolith particles through the interaction of water ice and regolith by frost wedging. We present morphological and spectral analyses of high-fidelity lunar regolith simulants LHS-1 (lunar highlands simulant-1) and LMS-1 (lunar mare simulant-1) that have been exposed to varying concentrations of water ice (1, 10, and 30 wt%) over extended periods of time (1, 3, and 6 months) to evaluate the extent at which lunar regolith may be weathered by ice-regolith interactions in the Moon’s polar regions. To characterize changes in regolith particle morphology, we explored grain size and shape parameters with the CILAS ExpertShape suite and characterized the abundance and evolution of clinging fines with scanning electron microscopy and energy dispersive X-ray spectroscopy. Reflectance spectra were taken from 1.0–22.5 μm (444.4–10,000 cm−1) to characterize any differences in spectral features that may occur as a result of regolith breakdown. Both the morphological and spectral investigations display trends that show simulant particle degradation as a function of composition, increasing water concentration, and freezing time. Our study demonstrates that the lunar regolith is susceptible to mechanical breakdown in the presence of water ice and that water ice is likely a contributor to the weathering environment within permanently shadowed regions on the lunar surface.


Introduction
Our current understanding of the presence of water ice at the lunar poles and its relationship with the lunar regolith is wholly dependent on remote sensing observations from instruments like the Moon Mineralogy Mapper (M3), the Diviner Lunar Radiometer Experiment (Diviner), the Lyman Alpha Mapping Project (LAMP), the Lunar Orbiter Laser Altimeter (LOLA), and the Lunar Prospector Neutron Spectrometer (LP-NS; e.g., Hubbard 1998;Pieters et al. 2009;Gladstone et al. 2010;Paige et al. 2010;Smith et al. 2010).The first search for water ice near the lunar poles was conducted with the Clementine bistatic radar experiment.While the magnitude and polarization of the reflected radar signals were suggestive of the presence of volatile ices, the results from this study were deemed inconclusive (e.g., Nozette et al. 1996).The search continued with LP-NS thermal and epithermal neutron data, which showed enhanced hydrogen in permanently shadowed craters at the south pole (e.g., Feldman et al. 2000Feldman et al. , 2001)).Years later, M3 on Chandrayaan-1 made the first tentative detection of water on the lunar surface through absorption features that were attributed to the presence of silicate materials that bear hydroxyl (OH) and/or molecular water (H 2 O; e.g., Clark 2009;Pieters et al. 2009;Sunshine et al. 2009;Li & Milliken 2017).Following this discovery, complimentary polarimetric radar studies from the Lunar Reconnaissance Orbiter (LRO) and Chandrayaan-1 found anomalous craters at the lunar poles with interior surfaces that possessed high circular polarization ratio (CPR) values (CPR > 1) and perimeters that possessed low CPR values (e.g., Spudis et al. 2010;Neish et al. 2011).This discrepancy in CPR values was presumed to be due to the presence of water ice, as the radar scattering properties of weakly absorbing media like water ice lead to a coherent backscatter effect that can generate CPR > 1 (e.g., Peters 1992;Hagfors et al. 1997).Analyses from the Lunar Crater Observation and Sensing Satellite impact at Cabeus crater were coupled with these radar studies to show that water ice may be present in the form of either discrete ice grains mixed into the lunar regolith or as thin, icy coatings on regolith grains (Colaprete et al. 2010;Neish et al. 2011).It should be noted that more recent studies claim that high CPR values could be produced from subsurface roughness (e.g., Fa & Cai 2013;Fa & Eke 2018).Conversely, ice could also exist in smooth areas that display low CPR values (Jozwiak et al. 2022).
As LRO continued its survey of the lunar surface, more promising evidence for water ice arose, especially in permanently shadowed regions (PSRs) at the lunar poles.Temperature and ultraviolet (UV) albedo measurements from Diviner and LAMP, respectively, were combined to study regions where water ice would be thermodynamically stable below ∼110 K (Hayne et al. 2015).LAMP UV spectra of these regions are consistent with the presence of surficial water ice, and constraints were placed on abundance; if the ice is intimately mixed with the lunar soil, cold traps may contain ∼0.1-2.0 wt% of water ice by mass, but pure ice exposures may yield concentrations up to 10% of the surface area of LAMP's 250 m scale measurements (Hayne et al. 2015).Additionally, Diviner temperature and LOLA reflectance measurements at 1064 nm show a trend of increasing reflectance with decreasing temperature at the lunar south pole, again suggesting that the regolith at polar latitudes is different, namely from the presence of OH, H2O, and other volatile species (e.g., Zuber et al. 2012;Fisher et al. 2017).
The aforementioned studies were all indicative of the presence of water ice, but no direct, unique detections of water ice were made until M 3 visible-to-near-infrared reflectance data were used to identify water ice based on diagnostic spectral features near 1.3, 1.5, and 2 μm (Li et al. 2018).Ice signatures were seen in ∼3.5% of the lunar polar cold traps using this method, and spectral modeling results from this study indicated that intimately mixed water ice could be present in the regolith at ∼5 wt% to ∼30 wt% or higher.These water abundances, when compared to the much higher concentrations modeled for cold traps identified on other airless bodies like Mercury and Ceres (e.g., Platz et al. 2016;Deutsch et al. 2017), suggest that the Moon has a unique volatile environment, perhaps due to comparatively lower rates of water supply and more intense regolith gardening processes (Li et al. 2018).
Volatile delivery, transport, and retention mechanisms are not the only processes that regularly act at the lunar polar regolith environment; vaporization and gardening from solar wind and micrometeorite impacts commonly alters the lunar regolith environment.The likelihood of increased regolith porosity within PSRs (e.g., Schultz et al. 2010;Hayne et al. 2017) suggests that other methods of regolith breakdown may be occurring.Recent work has suggested that pulverization and fragmentation of lunar dust via dielectric breakdown may be acting at the lunar poles (e.g., Jordan et al. 2015), and could explain an increase in both the percentage of fine particulate materials and the generation of "fairy-castle" regolith structures (Hapke & van Hoen 1963) within PSRs.While this form of breakdown occurs along particle boundaries, other forms of mechanical breakdown can occur within particles, particularly along microscopic fractures and other points of weakness (e.g., McGreevy 1981).
Along a particle's weak points, external forces can cause stress corrosion of the particle (e.g., Tharp 1987).One such process that can cause stress corrosion is frost wedging, or the mechanical disintegration of a material by the pressure of water freezing within cracks, pores, etc. in a material (e.g., Earle 2019).Over repeated exposures, the particle's material strength is further diminished, causing the particle to break apart and produce an increasing amount of fine particulate material.The initial freezing rate and temperature also have an impact on the extent of this breakdown.Higher freezing temperatures and longer freezing times generally result in more extensive damage to a material (e.g., Tharp 1987).Additionally, soils that include fine particulate material (<20 μm in diameter), as opposed to coarse soils that are uniform in particle size, have been shown to be more susceptible to the effects of weathering via frost (e.g., Everett 1961).With this in mind, we propose that another regolith breakdown process may be occurring in PSRs: mechanical breakdown of the particulate lunar regolith and by water ice.
As preparations continue for humans to return to the Moon through Artemis, all processes acting to alter the polar regolith, in particular breaking it down into smaller particle sizes, should be investigated to better prepare for conducting successful, long-term lunar surface science.Apollo astronauts faced many challenges while working with and in the lunar regolith environment.In such a low-gravity environment, the dusty regolith becomes lofted quickly and can easily cover any systems utilized during an extra-vehicular activity, including spacesuits, tools, scientific packages/experiments, and transportation.While dust-mitigation controls were provided to the astronauts, these tools were incapable of removing the finest fraction of lunar dust from the many surfaces it clung to (e.g., Jacobs et al. 1971;Gaier & Jaworske 2007;Gaier 2020).Not only does the finest fraction of lunar dust significantly impact efficiency and reliability of lunar surface science payloads and transportation (McKay 1971(McKay , 1972(McKay , 1973;;Gaier 2020), but it also directly impacts the health and safety of our astronauts (e.g., Gaier 2005).Multiple lunar regolith simulants have been tested against mammalian cells to understand how the lunar soil may affect human health in the long term and were found to cause cytotoxic and genotoxic effects on both cultured neuronal and lung-derived cells (Caston et al. 2018).Therefore, dustmitigation technologies must be considered and implemented when designing new tools, rovers, suits, and other necessary payloads, and that consideration begins with understanding the nature of dust and processes that form it.
Here we investigate how water ice alters the lunar polar regolith by characterizing the morphologic and spectral changes of two high-fidelity lunar regolith simulants after exposing them to freezing temperatures and three different concentrations of water ice (1, 10, and 30 wt%) for three separate lengths of time (1, 3, and 6 months).While many experiments have attempted to replicate the regolith found in PSRs (e.g., Cooper et al. 2011;Mantovani et al. 2015;Slumba et al. 2022), this work's focus on mechanical breakdown via water ice has yet to be explored.We note that ice-particle interactions would vary in their breakdown efficiency depending on the conditions under which they occur (phase of water ice, rate of freezing, temperature, etc.).Here, we explore a scenario wherein liquid water infiltrates natural weaknesses in a particle, expands upon freezing, and subsequently causes those weaknesses to propagate over time through stress corrosion.With this process, we observed the generation of fine particulate material, or "fines" (particles <2 μm in diameter) that began to cling to one another or larger particles to form "clumps" with increasing water concentration and freezing time.Size and shape analyses for changes in individual particle morphology were conducted using the CILAS ExpertShape suite, fine particulate behavior, generation, and abundance were characterized using scanning electron microscopy (SEM) and energy dispersive X-ray spectroscopy (EDS), and bulk simulant spectral features were studied in reflectance space across the 1-22.5 μm (10,000-444.4cm −1 ) spectral region.

Sample Creation, Storage, and Preparation for Analyses
Two lunar regolith simulants were used for this work: LMS-1, a high-fidelity, mineral-based mare simulant, and LHS-1, a high-fidelity, mineral-based highlands simulant produced by Exolith Lab at the University of Central Florida (Long-Fox et al. 2023).The term "high-fidelity" is used here to emphasize that these simulants are robust lunar soil analogs made by combining mineral and rock fragments to accurately simulate the texture, particle size, and composition of Apollo soils (Exolith Labs 2021).While this study primarily aims to provide context for future missions to the lunar south pole where highlands regolith dominates, volatile species have been proposed to exist throughout the lunar regolith environment (e.g., Pieters et al. 2009;Hayne et al. 2021), so we chose to study both lunar highlands and lunar mare simulants.
Both simulants were weighed out into air-tight containers in 2 g aliquots and were placed into a freezer for 24 hr to ensure that they did not immediately turn into mud upon the addition of water.In accordance with the spectral modeling estimates from Li et al. (2018), each aliquot of simulant was mixed with 1 wt%, 10 wt%, or 30 wt% of deionized water from a vial suspended in an ice bath.The samples were then placed in a Midea freezer on the lowest temperature setting and left undisturbed for 1, 3, or 6 months.In addition, we analyzed two references: frozen simulant with no water added (referred to as "dry" samples) and the base, unaltered simulant (referred to as "unaltered" samples, kept at atmosphere in standard lab conditions) for a total of 26 individual aliquots.
The temperature and humidity of the sample storage freezer was measured both by a commercial temperature sensor and an Arduino fit with a DHT-22 sensor with the ability to stream temperature and humidity data to a server.The DHS-22 output was monitored on two separate days.On the first day, the sensor was tested for 28 minutes, long enough to notice that there was some level of temperature cycling.The next day, the output was measured once every 2 minutes for 6.75 hr.The lab freezer had a temperature of −22 °C according to the commercial sensor placed among the sample containers in the center of the freezer, and this value appeared to be consistent upon visual checks that occurred during sample removal.This value seemingly only varied when the freezer door was opened for a prolonged period and was never seen above −12 °C.The DHT-22 sensor, placed just inside the freezer door, read a minimum temperature of −17 °C with minimum increases up to −14 °C and maximum increases up to −11 °C.The minimum freezer humidity was measured to be 34.5% with minimum increases up to 38.9% and maximum increases up to 39.2%.The observed increases in temperature and humidity occurred simultaneously and cyclically every 24 minutes.This is expected due to the natural cycling of the freezer, and these fluctuations are not expected to affect the phase of the ice (e.g., Hansen 2021).The samples were not disturbed in any way until they were removed from the freezer to limit any freeze-thaw cycling that could significantly weather the simulants.
Once the allotted time for freezing was up for each sample, the sample container lid was removed, and both the container containing the sample and the lid were placed into a gravity convection oven set to 50 °C (122°F) to slowly bake off the water without shocking the sample with heat.For the dry frozen and 1 wt% samples, this typically took an hour, while the 10 and 30 wt% samples took roughly 2 hr to fully dry.After the water evaporated, the containers were removed, cooled at room temperature until they were no longer warm to the touch, and their lids were replaced.They were stored under ambient conditions until they were ready for analysis, the methods for which are outlined in the order in which they were conducted below.

Sample Analysis
To preserve sample integrity and mitigate the loss of fines, scanning electron microscopy and reflectance spectroscopy were performed before any sieving occurred for other size and shape analyses.

Clinging Fines and Clumping: Scanning Electron Microscopy Analysis
We used scanning electron microscopy images to quantify the number and behavior of fines present in the samples.A Zeiss ULTRA-55 Field Emission Gun Scanning Electron Microscope (FEG SEM) with a Schottky field emission source and resolution of 1 nm at 15 kV or 1.7 nm at 1 kV was used alongside its Noran System 7 Energy Dispersive X-ray Spectrometer.
Samples were prepared by pressing a small puck covered in carbon tape into the simulants and tapping away the excess.The pucks were then placed into a Quorum Technologies EMS 150T ES Sputter Coater, which was set to deposit ∼10 nm (or a 45 s sputter) of gold onto the surface of the samples, which is sufficient for SEM analyses of powdered samples.Once the SEM was focused on the sample puck at 40× magnification to align with the spatial resolution of the EDS, the EDS elemental mapping feature was used to identify concentrations of calcium, aluminum, and magnesium.These elemental concentrations were then overlaid on an SEM image as an RGB color map, as shown in Figure 1.When calcium (blue) and aluminum (red) are present in high concentrations, their overlap points to the existence of plagioclase, whereas the presence of magnesium (green) is indicative of pyroxene or olivine.These mineral phases were chosen for analysis because plagioclase and pyroxene/olivine are the main constituents of LHS-1 and LMS-1, respectively, and should weather at differing rates (e.g., Goldich 1938;Birkeland 1999).Once these maps were collected, a "pink" particle (calcium + aluminum) and a "green" (magnesium) particle with relatively flat surfaces were chosen to image.For each of the selected particles, two image classes have been defined: "Clumps" images of the particle surface were made at 2000× magnification and "fines" images were made at 5000× magnification.Creating the two classes of fine particulate SEM images for individual particles of differing mineral phases allowed us to determine the abundances of clinging fines and the extent to which these fines aggregate or clump together for each sample.
When comparing images of particles from unaltered and water ice-mixed samples, we noticed that the surfaces of the water ice-mixed particles were covered by small units of clinging fine particulate material that had clumped together.Carrier (2005) stated that lunar regolith particles are so irregular in shape that they mechanically interlock with one another, giving lunar regolith a "cohesive" behavior.This concept prompted us to explore fine particulate "clumping" as an indicator of weathering; Perhaps if more fine particulate material was being generated through mechanical breakdown, this material was more susceptible to cohesion.We organized "clumps" in their respective images into two subgroups: "primary clumps" and "secondary clumps" (Figure 2).Primary clumps are a conglomerate of particles greater than 2 μm in diameter that can exist in one of two forms: they may either be composed of a unit of more than one particle that features 10 or more submicron clinging fines or a unit of at least four clinging particles.Secondary clumps are < 2 μm in diameter and are made up of least four clinging particles.Primary and secondary clump identifications and counts are conducted for each image using Photoshop's counting tool.To further characterize the primary clumping unit behavior, we measured their maximum Feret diameter, or largest end-to-end diameter (e.g., Merkus 2009), as well as an estimate for the area they encompass, which is done by loosely drawing rectangles or right-triangles around the clumps and calculating the area of the assigned polygon(s).It should be emphasized that the area and maximum Feret diameter are estimates.The length of a line measured with the ruler tool in pixels was converted to length  in microns using the scale bar present on each SEM image.An example of each of these measurements and identifications for both clump classes are shown in Figure 2.
We defined "fines" in their respective images as an individual particle less than 2 μm in diameter.For each fines image (e.g., Figure 3), all fine particulates across the image area were counted by eye using the counting tool in Photoshop.Shadowing, charging, and general particle topography in SEM images may obscure or hide submicron fines from view resulting in counts that are likely underestimated.Additionally, fines may be overestimated or underestimated on particles with highly textured surfaces, where fine particulate material becomes difficult to differentiate from surface texture.
To better understand how human interpretation of what a fine is affects the resulting counting statistics, three individuals (two of which were internal to the project and one of which was external) counted fines over the same 382 ×382 pixel area in one image (LHS Ca/Al, frozen for 3 months with 10 wt% of water).Within this area, individuals #1, #2, and #3 counted 168, 152, and 140 fines, respectively.The count from individual #1, who imaged the samples and provided counts for each image, was taken as the "true" number of counts.This count is believed to be higher than the others because of the individual's familiarity with the imaging process, which bettered their ability to distinguish fines from surface texture.The counts from individuals #2 and #3 were averaged (for a total of 146 fines) and compared against the "true" count by individual #1 (168 fines) to give a percent error on the fines counts of 13.1%.This percent error was taken as an average percent error for all fines images and is shown as error bars on the counts presented in Figure 11.
Finally, on both clumps and fines images, there may be regions we denote as "uncountable areas" (outlined in purple, as shown in Figure 2) that are affected by charging and/or particle topography.Any fines or clumps in these "uncountable areas" are not used in our analyses.

Reflectance Spectroscopy: Experimental Setup
Near infrared (NIR), short-wave infrared (SWIR), and midinfrared (MIR) diffuse reflectance spectra were obtained for unsieved samples with a Nicolet iS50 Fourier Transform Infrared Spectrometer (FTIR) and a Pike EasiDiff diffuse reflectance accessory set inside the main sample compartment.For maximum reflectance in the infrared, the optics within the reflectance accessory consist of uncoated aluminum on a glass substrate.The spectrometer source beam enters the EasiDiff accessory and is initially reflected off a 45°mirror to encounter an ellipsoidal reflector, which subsequently focuses the beam onto the sample or reflectance standard.The diffusely reflected energy is collected by the ellipsoidal reflector and passes to a mirror at the bottom of the assembly before encountering a final 45°mirror that directs the light back into the spectrometer and to the chosen detector.
Three detector, beamsplitter, source, and diffuse standard combinations were used to obtain our spectral measurements.To measure diffuse reflectance from 1-2.5 μm, we used an InGaAs detector and quartz beamsplitter alongside a Tungsten-Halogen NIR/Vis (white light) source and Spectralon diffuse standard.Spectralon was chosen because it displays a nearly flat reflectance spectrum with values near 1 for NIR wavelengths (e.g., Georgiev & Butler 2007).For the 2.5-22.5 μm range, measurements were made with a DLaTGS-KBr detector, Polaris Long-lifetime mid-IR source, and a diffuse gold standard.Gold was used across this range for its nearly flat, high-reflectance spectrum (e.g., Hanssen & Kaplan 1998).For the 2.5-5 μm spectral range we utilized a calcium fluoride (CaF2) beamsplitter for its low dispersion properties in this region, while spectra from 5-22.5 μm were collected with a potassium bromide (KBr) beamsplitter for transparency in the MIR.Spectral experiment setup file specifications are shown in Table 1.
Samples were prepared by pouring the simulant into the 0.18 cm 3 capacity Pike EasiDiff sample cup and using a straight edge to level the sample surface.The sample cup was then gently tapped against the preparation table to aid in the removal of any surface texture.Each sample was measured three times under ambient conditions for each of the three experimental setups described above (Table 1).The sample cup was rotated between measurements to create an average spectral measurement, and a background spectrum was made for each sample spectral collection.All spectra are made with the Thermo Scientific OMNIC software suite, which allows for direct interfacing with the FTIR and its components.This software automatically subtracts background spectra from the corresponding sample spectra.Each spectral measurement included a total of 128 interferograms or scans at a spectral resolution of 2 cm −1 .The optical velocity, or speed of the moving mirror within the FTIR, for each experimental setup defaults to the OMNIC recommended setting for the corresponding spectral range to obtain a signal-to-noise ratio (S/N) capable of detecting and separating 6% features from the noise across the spectral range.Due to the relatively low albedo of LMS-1, a detector gain of 8.0 was used to obtain spectra from 2.5-22.5 μm and a gain of 4.0 was used from 1-2.5 μm.

Reflectance Spectroscopy: Normalization
To identify changes in spectral shape or shifts of spectral features, we normalized our spectra so that these changes can be directly compared to one another.NIR (1-2.5 μm) measurements were normalized in reflectance space at 2.4 μm and SWIR measurements (2.5-5 μm) were normalized in reflectance space at 2.55 μm.These normalization wavelengths were chosen to avoid normalizing over any expected spectral features or normalizing too close to the InGaAs and DTGS detector limits (Equation ( 1)).The MIR (5-22.5 μm) spectra were normalized in 1-reflectance space at their respective Christiansen feature (CF) positions (i.e., the maximum near 8 μm in 1-reflectance space; see Section 4.3 for further detail) so that they appear as emissivity spectra for ease of comparison with existing emissivity spectra of constituent minerals across the MIR (Equation ( 2)).Spectra were normalized according to the normalization Equations (1) and ( 2 The AccuScope, a tabletop microscope fit with a camera, was used in conjunction with the CILAS Laser Particle Size Analyzer's ExpertShape software to characterize changes in particle size and shape.Each sample was sieved into four particle size fractions: >250 μm, 125-250 μm, 63-125 μm, and <63 μm.Then, particles were placed onto a microscope slide to be imaged at 4× magnification one particle size fraction at a time using an IDS uEye camera (Model UI149xLE-C) and the uEye Cockpit, which provides access to all camera settings for imaging.Several particles were imaged at once and images were analyzed via the CILAS ExpertShape software.Within the software, the thresholding feature was used to find the perimeters of each particle and distinguish them from the image background.If any particles were touching, a built-in eraser tool was used to create a line between them that was one pixel in width, effectively separating them for analysis.Similarly, if particle boundaries were not accurately recognized by the simulant software or if particles were touching the image boundaries, they were removed from the analysis.Once the particle boundaries were correctly defined, a batch treatment was conducted wherein over 90 size and shape parameters were measured for each particle.From these parameters, we selected three that were most suitable for determining the extent of particle weathering.The first, equivalent circular diameter (ECD), provides insight into changes in particle size.We also chose to investigate  sphericity and elongation, which indicate changes in particle shape.These parameters are described in Figure 4 and Table 2.
The number of particles imaged and analyzed by Expert-Shape varied per particle size fraction and sample, for example, the 63-125 μm particle size fraction of unaltered LMS-1 has 301 particles that make up the median size and shape values we present, whereas the same particle size fraction of 1 wt% LMS-1 frozen for 1 month only contains a sample of 135 particles.This is due to the natural, random selection of particles that make it onto the microscope slide to be imaged.For each sample, every particle imaged and analyzed by ExpertShape has a unique value for sphericity, elongation, and ECD.To explore how these parameters changed with water concentration and time, we present the median values for these parameters for each sample.Error bars are given as the standard deviation on these parameters for each sample.
Finally, it should be noted that imaging particles with the AccuScope only provides accurate particle size results for particles >63 μm in diameter.The presence of clinging fines in the smallest particle size fraction makes imaging in 2D-space more difficult, as the particles become nearly impossible to differentiate by eye and by the ExpertShape software, which can only detect particles >2.5 μm in diameter.To better understand the nature of the fines in our samples, we decided to use SEM (described in Section 2.2.1) to characterize the particle sizes of the <63 μm particle size fraction of our samples.

Results
To provide context for our spectral results, we first present the data from our ExpertShape and SEM/EDS analyses.

Size and Shape Results
Across the three largest particle size fractions (>250 μm, 125-250 μm, and 63-125 μm), no consistent trends in apparent particle degradation were observed as a function of time or water content.Therefore, we only discuss the results for the <63 μm particle size fraction (Figures 5 and 6).Results for the remaining particle size fractions are included in the supplementary materials in Figures A1-A6.All comparisons in the text between size and shape parameters for unaltered and weathered simulant were made using the percent difference formula (Equation (3)):

LMS-1
For LMS-1, we observed an overall trend of increasing particle degradation with increasing added water content and time.Median particle sphericity values in the <63 particle size fraction of LMS-1 mixed with 30 wt% of water increased by ∼25% compared to unaltered simulant over the course of 6 months of freezing.For the same sample, elongation decreased by ∼24%, and ECD decreased by ∼39%. Figure 5 shows median particle sphericity, elongation, and ECD values for the <63 μm particle size fraction of each aliquot of LMS-1.

LHS-1
For the <63 μm particle size fraction of LHS-1, we see the same general trend of increasing degradation with increasing water ice content and freezing time.Median particle ECD in the <63 particle size fraction of LHS-1 mixed with 30 wt% of water ice decreased by ∼57% compared to unaltered simulant over the course of 6 months of freezing.For the same sample, sphericity increased by ∼23% and elongation decreased by ∼22%. Figure 6 shows median particle sphericity, elongation, and ECD values for the <63 μm particle size fraction of each aliquot of LHS-1.

Clinging Fines and Clumping Results
Primary and secondary clump counts will be presented first, followed by counts for clinging fines, results for primary clump maximum Feret diameter, and finally, primary clump-covering area percentages.For analysis, we split the clumps and fines into one final identification subset based off the type of particle imaged: magnesium (Mg, or pyroxene/olivine) and calcium/ aluminum (Ca/Al, or plagioclase).As previously mentioned, this distinction allows us to quantify differences in weathering rate that may depend on mineral phase.

Primary Clumps
Between Mg and Ca/Al particle types, more primary clumps (maximum Feret diameter >2 μm) are accumulated on Mg particles (maximum of 64) compared to Ca/Al particles (maximum of 55).Between simulants, LMS-1 accumulates more primary clumps on both particle types as a function of time and increasing water content, as shown in Figure 7.In the case of the Mg primary clump count, LMS-1 developed up to ∼75% more primary clumps than LHS-1 (64 clumps versus 29 clumps, respectively, under the same sample conditions of 6 months of freezing time with 10 wt% of added water).For the Ca/Al count, LMS-1 developed up to ∼56% more primary clumps than LHS-1 (55 clumps versus 31 clumps, respectively, under the same sample conditions of 6 months of freezing time with 10 wt% of added water).Both Mg and Ca/Al primary clump counts show general upward trends in clump number with increasing amounts of time and added water content, so the duration of freezing and amount of water present in the sample has a significant impact on the number of these clumping fines units that we identify on particle surfaces.Figure 7 shows the primary clump counts for Mg and Ca/Al particles.

Secondary Clumps
Like the primary clumps, more secondary clumps (maximum Feret diameter <2 μm) are accumulated on Mg particles than Ca/Al particles, with a maximum total of 62 versus 37 secondary clumps, respectively.LMS-1 developed more secondary clumps on Mg particles in every sample case, and

Primary Clump Maximum Feret Diameter
To determine how clump size changes with increasing time and water concentration, we measured the maximum Feret diameter of each primary clump then took the median of these values for each particle imaged.We note a downward trend in the median maximum Feret diameter of the primary clumps, as shown in Figure 9. Thus, as the abundance of primary clumps increases for a particle, the average clump maximum Feret diameter decreases.However, this trend does not hold for every case.From the Mg images, we see that many of the LHS samples are outliers, particularly the 1 month dry, frozen sample.Additionally, Mg images typically display smaller clumps than their Ca/Al counterparts.The Ca/Al images only display a clear trend in the 6 month sample set, suggesting (alongside Figure 8) that particles of this composition may weather more unevenly or less readily than Mg particles.

Primary Clump-covering Area
The decrease in primary clump mean maximum Feret diameter with increasing water concentration and time led us  to investigate how much area these clumps inhabit on a particle's surface.To calculate the percent area that the primary clumps occupy across the countable area of the clump, we used a simple proportion (Equation ( 4)):

=
x Sum of primary clump areas in pixels Full countable area on image in pixels 100 4 where the full countable area in pixels was found by subtracting the combined area of the uncountable regions from the total number of pixels in each image.Figure 10 shows the total clumpcovering area of both particle types (Mg and Ca/Al) covered by the respective primary clumps.The relationship between increasing clump number with increasing water concentration and time becomes less clear here.We might expect that the more weathered particles with more clumps would have more of their area dominated by these clumps, but some samples with a higher number of clumps have less of their measurable area dominated by them.For example, LMS-1 3 month 30 wt% has ∼21% more    primary clumps than the same sample of LHS-1, but LHS-1ʼs primary clumps cover a ∼47% larger area.We note that several potential clumps that overlap with image edges go uncaptured, as their full maximum Feret diameters and areas cannot be measured.Some images contain a chosen particle that fully fill the image boundaries (i.e., all pixels can be counted over), while other images may contain a chosen particle along with some background (i.e., only the main particle pixels are counted over) as described in Section 2.2.1.

Clinging Fines
For fines images, Mg particles have a greater abundance of singular fines (<2 μm in diameter) in almost every case, aligning with the finding that a greater number of clumps also occurs on the Mg particles.There is also a general upward trend in the number of fines with freezing time and water concentration.The trend of increasing number of fines with freezing time and water concentration only holds for Ca/Al particles for the 6 month sample of LHS.Otherwise, Ca/Al particles do not seem to weather according to a trend.For example, all Ca/Al LMS images save the 6 month, 30 wt% have fewer fines than the unaltered simulant image.Additionally, the 1 month 30 wt% of LMS-1 has fewer fines than any of the other LMS-1 1 month samples, while the LMS-1 6 month 30 wt% sample has nearly twice that of any other Ca/Al particle imaged.Overall, LMS-1 produces a higher number of fines than LHS-1 across both Mg and Ca/Al particles with no exceptions, whether a trend with time and water content is observed.Figure 11 shows the counts for Mg and Ca/Al fines for both simulants.Table 3 provides the number of fines per image.

Spectral Results
We present reflectance spectra for samples of LHS-1 and LMS-1 of all water concentrations that were frozen for 1, 3, and 6 months.The MIR spectra are plotted in 1-reflectance space from ∼7-22.5 μm to emphasize any changes in band depth.To characterize changes in band depth for both simulants, we chose to analyze the first reststrahlen band (RB) that occurs just after the Christiansen feature (CF; see Table 4), as well as the transparency feature (TF) that occurs longward of the CF between 10 and 12 μm.Positions of key features for both simulants are calculated and shown in Table 4. Band depths are calculated according to the equation below (Equation ( 5)), where R min is the minimum reflectance value at the wavelength at which the bands occur.
The calculated depths are presented for both simulants in Tables 5 and 6.For LHS-1, when comparing our spectra to the "Dry" (unaltered) simulant spectra in the MIR, we see a trend  of increasing TF and RB depth with increasing water ice concentration and time, as shown in Figure 12.No trends are observed in the LMS-1 MIR spectra.
The NIR and SWIR spectra, shown in Figures 13 and 14, show no differences in absorption band strengths or spectral slope with increasing breakdown of the sample.We note that these spectra have several non-real features from external factors, designated by the absorption bands present at 3, 3.5, 4, and 4.3 μm.The deep 3 μm band is from adsorbed water within the samples, as these measurements were taken under ambient conditions and were not heated to remove any accumulated adsorbed water before data acquisition.The doublet absorption bands centered around 3.5 μm, as well as the features near 4 μm, are likely due to some organic material that may have been introduced during simulant production.Finally, the 4.3 μm feature is attributed to CO2 in the atmosphere, as we noticed a decrease in the depth of these features or total removal of these features the longer we kept a sample in the spectrometer's measurement chamber before taking a sample measurement.

Simulant Mineral Endmembers
Here we discuss the simulant mineral endmembers and how the choice of these mineral endmembers may have impacted our results.Our batch of LMS-1 is reported to have a mean particle size of 50 μm, a median particle size of 45 μm, and a particle size range of <0.04-300 μm, while the batch of LHS-1 should have a mean particle size of 60 μm, median particle size of 50 μm, and a particle size range of <0.04-400 μm (Exolith Labs 2021).We note that we measured simulant particles with diameters of up to 1 mm, and personal communications with Exolith revealed that the maximum sieve size used for the basaltic materials during this time was 1 mm.We do not expect this discrepancy in range of particle sizes to impact our analyses, as the finest particle size fraction of the materials was shown to weather the most drastically.
We show that the magnesium component mineralogies, olivine and pyroxene, tended to weather faster than others within the simulants.We also note that the plagioclase seems to weather more unevenly with water ice content and time when compared to the pyroxene/olivine.While it is generally expected that pyroxene and olivine would break down more slowly than plagioclase (e.g., Goldich 1938;Birkeland 1999), there are several reasons why this may not be the case in the simulants, which we discuss further in Section 4.3.

Creation of Icy Samples
Currently, there is no standard for creating icy simulant, and many methods have been utilized (e.g., Cooper et al. 2011;Mantovani et al. 2015;Slumba et al. 2022).In many of the studies exploring in situ resource utilization capabilities and geotechnical properties of icy simulants, the techniques used focus on the generation and study of different forms of ices, e.g., shaved ice particles mixed with simulant, ice created by vapor deposition onto simulant under vacuum, etc.Our samples were created with a method similar to the "mud pie" (e.g., Sargeant et al. 2022) method of generating icy simulant in that they were not made under vacuum conditions.However, our simulant and water ice mixtures were not made under entirely ambient temperature conditions either (e.g., Pitcher et al. 2016).As described in Section 2.1, the simulants were frozen dry for 24 hr before being quickly mixed with cold water and returned to the freezer.We chose to create our samples this way to mitigate against the creation of "mud pies" and to reduce the amount of abrupt temperature shock that the simulants received.
We developed our method to ensure that the water ice would be in direct contact with the simulant, thus facilitating the breakdown process.This method, though tailored to our analyses, is not perfect.Some of our samples could have experienced uneven water mixing due to their method of creation, even if we attempted to ensure that even mixing would occur.While our samples were stored in air-tight containers during freezing, we recognize the possibility that the humidity conditions within the freezer could allow for collection of some amount of adsorbed water.However, measurements across our analytical methods show that the spectral and clumping behaviors of the "unaltered" and "dry, frozen" simulants do not change drastically when compared (e.g., Figures 7 and 12), implying that any adsorbed water does not have a drastic effect on mechanical degradation of our simulants.

Sample Weathering: Extent and Differences
As mentioned in Section 2.2.4,we had difficulties with the AccuScope/ExpertShape method of size and shape analyses for the <63 μm particle size fraction of our samples.While we were able to obtain sphericity and elongation measurements for the particles we imaged, we knew we were not accurately representing the submicron clinging fines in our sample, as we were unable to separate them from their host particles for analysis.This skewed our results in favor of the larger, more easily imaged particles.Still, the shape parameters we were able to investigate yielded some unexpected results.LHS-1, the simulant with more plagioclase, should weather faster simply due to its composition (e.g., Goldich 1938;Birkeland 1999), yet LMS-1 seemed to undergo more degradation across all samples even after only 1 month of freezing.
The SEM/EDS analyses told the same story, albeit with more detail.LMS-1 weathered more than LHS-1 in nearly every instance, and the Mg, or pyroxene/olivine, component images showed a higher number of both primary and secondary clumps and clinging fines than the Ca/Al, or plagioclase, images.This result, however unexpected, is likely for a number of reasons: (1) the pyroxene and olivine inside the included basalts is generally of a smaller particle size than the larger   Ice-particle interactions on the lunar surface are expected to function differently than the ice-particle interactions that occurred in our lab freezer.The speed of the freezing process and any temperature fluctuations that occur as a result of the freezing process impact how ice crystals form.Rate of ice crystal formation and crystal size are both impacted by the freezing process, and both can alter the extent of mechanical damage a material may undergo (e.g., Tan et al. 2021).The water ice that weathered our samples is expected to be larger in size due to its much slower freezing rate, likely causing more extensive damage to our samples.In contrast, water on the lunar surface is expected to rapidly sublimate and condense through a variety of processes (Schörghofer et al. 2021), possibly creating smaller crystals or forming as amorphous ice (e.g., Seki & Hasegawa 1983).Therefore, the extent that water ice alters the lunar regolith could be less severe than the extent that it altered our analogs.The differences in breakdown behaviors between our simulants and actual lunar soils, as discussed in the previous Section, further complicate our ability to draw conclusions on the extent that lunar soils within the lunar polar regions will be impacted by water ice.Still, our work suggests that these ice-particle interactions may be a mechanical process breaking down the lunar regolith creating a more heavily weathered regolith environment at the poles.

Implications for Future Lunar Surface Science
As the Artemis missions begin in earnest, the obvious detriment that lunar soils pose to mission success and safety must be recognized.Artemis will send the first human missions to the lunar south pole, and if water-ice interactions do indeed generate a higher concentration of fine particulate material on the lunar surface as they did for our samples, astronauts may face greater difficulties when working in the lunar regolith environment than they did during the Apollo era.Much research is currently being conducted regarding in situ resource utilization (e.g., Purrington et al. 2022;Olthoff et al. 2023;Slumba 2023), dust mitigation and interactions (e.g., Cao et al. 2023;Hirabayashi et al. 2023;Wells et al. 2023), rover and tool development (e.g., Budzyn et al. 2023;Long-Fox et al. 2023), and astronaut health and safety (e.g., Pohlen et al. 2022).However, our understanding of the particle size distribution of the lunar regolith and the processes acting to mechanically break down the regolith in the lunar polar regions and the PSRs, which greatly affects all aforementioned areas of research, is limited.Therefore, for the future of lunar exploration, the complications and hazards that lunar dust creates must be well-characterized through understanding the nature of the regolith itself.
We now propose avenues of future study to close the gaps in our understanding of the particle size distribution of the lunar regolith and mechanical breakdown of the regolith.Based on the analyses in this work, lunar fine particulate material is more easily detected in MIR spectra than NIR or SWIR spectra.Therefore, hyperspectral MIR observations of the lunar surface should be prioritized in future in situ and remote sensing studies for characterizing the presence of submicron material.Additionally, we emphasize that the generation of fine particulate material is not limited to the single mechanical breakdown method explored in this study.The space weathering environment at the lunar surface complicates in situ regolith breakdown by introducing multiple processes that may mechanically alter the lunar regolith further, such as comminution via impacts and structural fatigue via diurnal thermal cycling (e.g., Pieters & Noble 2016).Laboratory experiments focused on how these alternative breakdown phenomena compare to mechanical breakdown via water ice or how these types of breakdown compound on one another could better elucidate the full extent of mechanical weathering that the lunar regolith may be experiencing.

Conclusions
With increasing lengths of freezing time and increasing water concentration, we saw general trends that point to the generation of increasing amounts of submicron fine particulate material, as well as changes in particle shape.The amount of fine particulate material generated with increasing water content and time is shown to be dependent on component mineral phases of the simulants.Simulant particles mapped to have more Mg-rich composition showed more weathering, i.e., the generation of more clumps and fines, than particles with predominantly Ca/Al-rich composition.For the more Mg-rich simulant, LMS-1, we saw increasing particle sphericity and decreasing particle elongation with increasing water content and time for particles <63 μm in diameter.While the shapes of LHS-1 particles changed, they did not weather according to any trend.Spectral evidence of weathering of LHS-1 was seen in the MIR via increases in band depth with increasing water concentration and time, though no such relationship was observed for LMS-1.No evidence of weathering was seen for either simulant in NIR or SWIR spectra.Although these weathering relationships are complex and vary across our methods, this study indicates that the presence of water ice at the lunar poles may cause the lunar regolith to mechanically weather in an observable way.Therefore, we must consider how higher concentrations of fine particulates and changes in particle shape at the poles may affect the spectral, thermal, and geotechnical properties of polar regolith for the success of future lunar surface operations and science.

Figure 1 .
Figure1.Energy Dispersive X-ray Spectroscopy (EDS) elemental map (at left) with magnesium (green), calcium (blue), and aluminum (red) counts overlaid onto an SEM image (at right) of LMS-1, frozen for 1 month with 1 wt% of water taken at 40× magnification.Green regions are indicative of pyroxene/olivine content, while pink regions are indicative of plagioclase.The Ca/Al or plagioclase particle we imaged to obtain our "clumps" and "fines" images for this sample is circled in red.The scale bar in the image at the right applies to both images.

Figure 2 .
Figure2.Clump analysis on a Ca/Al image of LHS-1 that was frozen for 6 months with 30 wt% of water ice.Purple boxes: uncountable areas.Green lines: maximum Feret diameters.Orange boxes: primary clump area-defining polygons.Pink markers: primary clump counts.Yellow markers: secondary clump counts.The inset illustrates the difference between primary clumps (conglomerates >2 μm in diameter) and secondary clumps (conglomerates <2 μm in diameter).A "fine," or single particle <2 μm in diameter, is shown circled in red.

Figure 3 .
Figure3.Magnesium fines image of LMS-1 that was frozen for 1 month with 1 wt% of water ice.Counts are shown here within an inset, as displaying all counts renders the image unreadable due to the number of markers.For a resolution reference, particle #42 in the inset is only 0.13 μm in diameter.

Figure 4 .
Figure 4. ECD, sphericity, and elongation, respectively, as defined by the ExpertShape software, as well as multiple examples of "thresholded" particles.
large across the 6 month sample suite, wherein the 6 month, 10 wt% sample of LMS-1 developed ∼3.4 times the number of secondary clumps than LHS-1 did (58 versus 17, respectively).The Mg secondary clumps accumulate with the same steady trend seen in the Mg and Ca/Al primary clumps: increasing time and water concentration aligns itself with an increase in Mg secondary clumps.The secondary clump count for the Ca/Al particles, however, does not seem to follow a clear trend, though LMS-1 still develops more secondary clumps than LHS-1 in all but one sample case (1 month, 30 wt%).Figures 8 shows the secondary clump counts for Mg and Ca/Al particles.

Figure 5 .
Figure 5. Median values for particle sphericity (top), elongation (middle), and ECD (bottom) for the <63 μm particle size fraction of LMS-1.Water ice content increases from unaltered to 30 wt% from left to right.The different symbols show the amount of time that the sample was frozen.

Figure 6 .
Figure 6.Median values for particle sphericity (top), elongation (middle), and ECD (bottom) of the <63 μm particle size fraction of LHS-1.Water ice content increases from unaltered to 30 wt% from left to right.The different symbols show the amount of time that the sample was frozen.

Figure 7 .
Figure 7.Primary clump count for Mg clump (top) and Ca/Al clump (bottom) images of all samples.

Figure 8 .
Figure 8. Secondary clump count for Mg clump (top) and Ca/Al clump (bottom) images of all samples.

Figure 10 .
Figure 10.Primary clump-covering area given in percent (%).Markers with a purple outline indicate that the imaged particle did not fully fill the image boundaries.

Figure 11 .
Figure 11.Number of <2 μm fines per particle imaged.The error bars on the counts are given according to the methodology outlined in Section 2.2.1.

Figure 12 .
Figure 12.LHS-1 (left) and LMS-1 (right) samples after 1, 3, and 6 months of freezing compared to unaltered (dry) simulant.MIR spectra are normalized to their respective Christiansen feature (CF), designated by a black, dotted line, and are offset on the y-axis for clarity.The transparency features (longward of the CF) are indicated by a black, dashed line.The first reststrahlen band for each simulant is indicated by a solid blue line.

Figure 13 .
Figure 13.LHS-1 (left) and LMS-1 (right) samples after 1, 3, and 6 months of freezing compared to unaltered (dry) simulant.SWIR spectra are normalized to 2.4 μm and are offset on the y-axis for clarity.

Figure 14 .
Figure 14.LHS-1 (left) and LMS-1 (right) samples after 1, 3, and 6 months of freezing compared to unaltered (dry) simulant.Intermediate-infrared spectra are normalized to 2.55 μm and are offset on the y-axis for clarity.

Figure A1 .
Figure A1.Median values for particle sphericity (top), elongation (middle), and ECD (bottom) for the 63-125 μm particle size fraction of LHS-1.Water ice content increases from unaltered to 30 wt% from left to right.The different symbols show the amount of time that the sample was frozen.

Figure A2 .
Figure A2.Median values for particle sphericity (top), elongation (middle), and ECD (bottom) for the 125-250 μm particle size fraction of LHS-1.Water ice content increases from unaltered to 30 wt% from left to right.The different symbols show the amount of time that the sample was frozen.

Figure A3 .
Figure A3.Median values for particle sphericity (top), elongation (middle), and ECD (bottom) for the >250 μm particle size fraction of LHS-1.Water ice content increases from unaltered to 30 wt% from left to right.The different symbols show the amount of time that the sample was frozen.

Figure A4 .
Figure A4.Median values for particle sphericity (top), elongation (middle), and ECD (bottom) for the 63-125 μm particle size fraction of LMS-1.Water ice content increases from unaltered to 30 wt% from left to right.The different symbols show the amount of time that the sample was frozen.

Figure A5 .
Figure A5.Median values for particle sphericity (top), elongation (middle), and ECD (bottom) for the 125-250 μm particle size fraction of LMS-1.Water ice content increases from unaltered to 30 wt% from left to right.The different symbols show the amount of time that the sample was frozen.

Figure A6 .
Figure A6.Median values for particle sphericity (top), elongation (middle), and ECD (bottom) for the >250 μm particle size fraction of LMS-1.Water ice content increases from unaltered to 30 wt% from left to right.The different symbols show the amount of time that the sample was frozen.
(Long 2014;he spectra and remove noise, we utilized AstroPy's convolution functionality(Long 2014; The Astropy Collaboration et al. 2022)to perform "boxcar" smoothing of our spectra.Boxcar smoothing takes a signal s[t] and creates a new signal s'[t] over an average of w adjacent signal elements.We selected a boxcar kernel width of w = 11 to smooth our data, which we found sufficient to remove noise near the detector limits without erasing the smallest distinguishable features.2.2.4.Size and Shape: ExpertShape Analysis

Table 1
FTIR Experimental Setup Specifications

Table 2
ExpertShape Equations for Size and Shape Analysis ECD (in μm)

Table 3
Number of Fines per Particle for Ca/Al and Mg Particles

Table 4
Features (in cm −1 ) Identified in Simulant Spectra

Table 5
Feature Depths Calculated for RB and TF of LHS-1 MIR Spectra