Cloud Properties and Boundary Layer Stability Above Southern Ocean Sea Ice and Coastal Antarctica

Significant variability in climate predictions originates from the simulated cloud cover over the Southern Ocean. Historically, Southern Ocean cloud and aerosol properties have been less studied than their northern hemisphere counterparts, and cloud‐sea‐ice interactions over the Southern Ocean also remain largely unexamined. We used data from combined radar, lidar, radiometer, radiosonde, and ERA5 reanalysis profiles to investigate cloud property relationships to cloud temperature, sea‐ice concentration, and boundary layer stability. Our findings show correlations between both cloud macrophysical properties and radiative effects and sea‐ice concentration, and that the marine atmospheric boundary layer is more stable over higher sea‐ice concentrations. Mixed‐phase cloud frequency of occurrence was highest over the sea‐ice zone at 15%, three times higher than over cold water south of the Antarctic Polar Front. For temperatures greater than −15°C, low‐level, single‐layer clouds were more likely to precipitate ice if they were coupled to cold‐water or sea‐ice surfaces than if they were decoupled from these surfaces, with the highest percentage of clouds precipitating ice observed over sea ice. These findings suggest a surface source of ice‐nucleating particles at high southern latitudes that increases cloud glaciation probability. We discuss the implications of our results for future studies into the relationship between cloud properties, aerosols, sea ice, and boundary layer stability at high latitudes over the Southern Ocean.


Introduction
The Southern Ocean (SO) influences atmospheric and oceanic conditions throughout the southern hemisphere.Strong, circumpolar westerlies interact with the sea surface while katabatic winds flowing off Antarctica drive the expansion of sea ice outward from the continent (Hall & Visbeck, 2002;Thompson et al., 2020).The isolation of the SO and Antarctica from most anthropogenic aerosol effects makes this region a unique natural laboratory for the study of atmosphere-ocean interactions (Hamilton et al., 2014).Cloud-aerosol, -radiation, and -ice interactions such as ice nucleation and the resultant precipitation are of particular interest in this region (McFarquhar et al., 2021;Uetake et al., 2020).The primary drivers of weather and climate are solar heating and atmospheric circulation along the storm track, though ocean, ice shelf, and continental (orographic) influences are also present (Mayewski et al., 2009).The influence of maritime and continental airmasses are readily apparent in the observational records made at coastal Antarctic sites.
The SO region is an active area of research in climate simulations, as significant variability in climate predictions originates in the simulated cloud cover and albedo for this region (Zelinka et al., 2020).Cloud phase, fraction, and precipitation have been identified as poorly represented in coupled global climate models (Trenberth & Fasullo, 2010), numerical weather prediction output (Bromwich et al., 2013;Protat et al., 2017), and atmospheric reanalysis products (Naud et al., 2014).Specifically, many state-of-the-art models consistently underestimate cloud cover in the lower troposphere (Bodas-Salcedo et al., 2014).
The surface of the southern hemisphere is largely oceanic, so clouds have a marked impact on the energy balance as their high albedo contrasts with the low surface albedo of water.In particular, with their smaller average droplet size, the shortwave albedo of liquid water clouds tends to be higher than that of ice clouds (Hartmann, 2016).Much of the model cloud cover bias is associated with an under-representation of low-level clouds comprised predominantly of supercooled liquid water (SLW), which are more prevalent in the cold sectors of the cyclones than elsewhere over the SO (Hyder et al., 2018;Naud et al., 2014).The resultant biases in sea surface temperature (SST), up to 3 K too warm in the subtropics (Zhang et al., 2023), lead directly to incorrect simulations of sea-ice extent, the position of the tropopause jet, and storm track locations (Ceppi et al., 2012(Ceppi et al., , 2014)), as well as generating substantial bias in the top-of-atmosphere reflected shortwave radiation (Bodas-Salcedo et al., 2014).
Despite their central importance to radiative balance, knowledge of cloud thermodynamic phase transitions, including the conditions required to form, sustain, and precipitate SLW clouds, remains incomplete over the SO region.Studies of the region as a whole often rely on satellite data.For example, using combined radar-lidar satellite products, Listowski et al. (2019) found that sea-ice presence over the high-latitude SO is associated with decreases of approximately 20% in the fraction of low-level, SLW-containing clouds in autumn and winter, and that this decrease is mostly observed in mixed-phase clouds (MPCs; clouds containing both liquid and ice particles) rather than pure (unglaciated) SLW clouds.This contrasts with a smaller association (approximately 10%) in spring and negligible association in summer (Listowski et al., 2019).However, Listowski et al. (2019) also found that the seasonal prevalence of MPCs versus unglaciated SLW clouds does not vary over the Antarctic continent itself.This suggests that seasonal changes in marine influence, along with sea-ice fraction and temperature, are partly responsible for phase partitioning in low-level clouds originating over the SO.
Cloud phase transitions are fixed at very cold temperatures, as ice crystals form spontaneously through homogeneous ice nucleation at temperatures of 40°C and below.However, in the absence of ice particle formation caused by precipitation from above, for the heterogeneous freezing of liquid droplets to occur between 40°and 0°C, ice nucleating particles (INPs) are required to act as nuclei for crystal formation.This process is known as primary ice production (PIP).These particles include aerosols such as terrestrial dust, crystalline salts from sea spray, and airborne biological material known as bioaerosols (Kanji et al., 2017).Using a global climate model, DeMott et al. (2010) found that global net cloud radiative forcing increased by approximately 1 Wm 2 for every order of magnitude increase in INP concentrations, a finding supported by Vergara-Temprado et al. (2018).Bigg (1973) determined from data gathered by ship-borne instruments between 1969 and 1972 that INPs over the SO must be at least partially oceanic in origin, with concentrations ranging from 3 to 250 m 3 at temperatures of 15°C.A more recent study found that INP concentrations at similar temperatures over the SO were approximately two orders of magnitude lower than those reported by Bigg (1973), at 0.38-4.6m 3 (McCluskey et al., 2018).Uetake et al. (2020) found that, while concentrations of SO INPs are low relative to those over the northern hemisphere and global subtropical regions, the ocean produces most INPs found in the region, and bioaerosols in the form of marine bacteria are the most common of these.Such INP are ejected into the atmosphere when bubbles burst at the ocean surface (McCluskey et al., 2018).In their study of satellite products, Listowski et al. (2019) observed a relation between MPC fraction and monthly changes in aerosol concentrations around the Antarctic coast, noting that the latter varies with the activity of marine biology.
Cloud phase is also controlled through mechanisms unrelated to local INP concentrations, such as the Wegener-Bergeron-Findeisen (WBF) process in MPCs (Korolev, 2007) and seeding from higher-altitude clouds (Bergeron, 1950), both of which lead to depletion of SLW clouds through ice growth.MPCs are also inherently unstable due to the potential for runaway riming processes to convert liquid drops into ice particles, which then collide with other liquid drops to trigger precipitation inside the cloud.This is known as secondary ice production (SIP), an example of which is the rapid dispersion of ice particles through a SLW layer between temperatures of 3°to 8°C in a process known as Hallett-Mossop (HM) rime splintering (Field et al., 2017).Atlas et al. (2022) found that the inclusion of the HM process in global cloud-resolving models lead to a decrease in cloud radiative effect over the SO, more closely matching satellite observations.Hu, Geerts, et al. (2023) examined the thermodynamic phase partitioning of shallow convective clouds over the SO, including the effects of INP concentration on precipitation rates.They found that ice presence is poorly predicted by turbulent and convective processes, that cold-cloud processes such as WBF and riming are likely responsible for much of the observed precipitation, and that INP concentration is also among the best indicators of ice growth.In subsequent modeling work examining a specific case study, Hu, Lebo, et al. (2023) found evidence of seeder-feeder mechanisms operating within low-level clouds themselves and confirmed the importance of rime splintering to produce ice within these clouds.The cases examined by Hu, Geerts, et al. (2023) and Hu, Lebo, et al. (2023) are limited to late summer/early autumn, when the sea ice has largely retreated from even the highlatitude SO.Furthermore, Hu, Geerts, et al. (2023) did not directly examine the role of boundary layer coupling in affecting cloud phase changes and precipitation rates.Given the focus on low-level clouds in the pursuit of reducing shortwave radiation bias in models and improving climate projections in response to global warming, it is important to consider the role of atmospheric structure and dynamics close to the surface.
The marine atmospheric boundary layer (MABL) is the lowest layer of the troposphere, in direct contact with the surface, and is also the layer that contains most SO clouds (Huang, Siems, Manton, Hande, & Haynes, 2012;Huang, Siems, Manton, Protat, & Delanoë, 2012).The state of the MABL is typically defined by whether or not the cloud layer capping the MABL is convectively coupled with the subcloud layer beneath it (Stull, 1988).When the prevailing state is one of stability, stratification and decoupling of the cloud and subcloud layers occurs, limiting vertical mixing within the MABL.Causes of decoupling include weakened radiative cooling at the cloud layer capping the boundary layer, stabilization as a direct result of precipitation, and the ventilation of warm air from the free troposphere into the MABL, all of which reduce the available turbulent kinetic energy (TKE) below the threshold required to maintain the overturning circulation that defines a well-mixed, coupled MABL (Zheng & Li, 2019).
In the northern hemisphere, Griesche et al. (2021) investigated the hypothesis that ice cloud formation over the Arctic is linked with coupling of MABL clouds to the ocean surface.They found that surface-coupled clouds containing ice were 2-3 times more likely to occur than decoupled clouds containing ice at temperatures above 10°C.However, they also found evidence that the reservoir of INPs in the free Arctic troposphere is drawn from terrestrial sources, in contrast with the oceanic reservoirs over the SO (Uetake et al., 2020).Given the potential for marine sources of INPs to influence clouds phase in low-level SO clouds, analysis of MABL properties driven by observational data provides a practical basis from which to study SO cloud processes, especially those relating to MABL coupling status (Zheng & Li, 2019).Truong et al. (2020) used radio-and drop-sonde data from four observational campaigns that took place in the period 2016-2018 to develop a climatology of the SO MABL, using k-means clustering to group data profiles from individual soundings.They identified seven clusters: one from data exclusively north of the ocean polar front; one for high-pressure systems; three for the primary sectors of SO midlatitude cyclones; one for the highlatitude SO; and one in close proximity to the Antarctic coast.These clusters vary in their average temperatures, pressures, wind velocities, and humidity levels, as well as their MABL properties and temperature inversion strengths.Truong et al. (2022) then used this framework to quantify biases in ERA5 reanalysis data, specifically in the structure of the SO MABL.They found that ERA5 biases were highest over the high-latitude SO, where biases in specific humidity result in underestimation of cloud layer number, layer thickness, and top height.This demonstrates the utility of observational data of the SO MABL in validating model output.
Over the SO, cloud properties and regimes are observed remotely by satellites with broad coverage but low spatial resolution (Listowski et al., 2019).High temporal resolution measurements of low-level clouds are made by instruments on ships, aircraft, or the ground (Alexander & Protat, 2018;McFarquhar et al., 2021) and by high vertical resolution station-and ship-launched radiosondes (Sato et al., 2018(Sato et al., , 2020)).Ground-based and in-situ measurements are therefore required to produce data sets to evaluate and improve cloud parameterization schemes such that models may more closely match observations using realistic physical mechanisms.Satellite observations of the SO also show association between sea-ice fraction and the prevalence of SLW clouds (Listowski et al., 2019), yet the role of sea ice in limiting the emission of marine aerosols into the MABL over the SO or affecting the structure of the MABL itself has yet to be closely examined, especially using data from shipbased campaigns.
Addressing the aforementioned issues in climate models requires observational data made within and beneath low-level cloud layers over the SO (Morrison et al., 2020).The Measurements of Aerosols, Radiation, and Clouds over the SO (MARCUS) campaign gathered atmospheric data while the Australian icebreaking resupply vessel (RSV) Aurora Australis transited the SO from Hobart, Tasmania to Casey, Mawson, and Davis Antarctic stations between October 2017and March 2018(McFarquhar et al., 2021).Data from these voyages was used by Hu, Geerts, et al. (2023) and Truong et al. (2022) to examine SO cloud and MABL processes, though neither included the possible effects of Antarctic sea ice at high latitudes.Given the importance of clouds to climate change model variability and the evolving link between climate change and Antarctic sea-ice extent (Hobbs et al., 2016), understanding cloud-sea-ice interactions will be increasingly important for ocean and atmosphere simulations in both coupled and uncoupled climate and weather models.Campaigns such as MARCUS provide vital data in this endeavor.
Our objective is to use data collected during the MARCUS campaign to improve understanding of cloud processes and interactions over an extended summer season in the high-latitude SO.We will also incorporate reanalysis data and satellite retrievals of sea-ice concentration in analyzing the structure of the MABL to see whether it varies with surface type.We have provided an overview of this study's background and goals in Section 1.In Section 2, we outline the methods used, including instrumentation and data sources, pre-processing, filtering, and analysis of the data.Following this, we describe and discuss the results of our analysis in Section 3. In Section 4, we conclude by summarizing the key results from this paper, as well as recommendations for further research.

Overview of the MARCUS Campaign
During the austral spring and summer months between 29 October 2017 and 23 March 2018, the RSV Aurora Australis spent a combined total of 11 weeks and 2 days in the sea-ice zone collecting atmospheric data for the MARCUS campaign (McFarquhar et al., 2021).Instruments on board the ship measured and recorded precipitation, cloud and aerosol properties, and radiative fluxes (Alexander et al., 2021;Humphries et al., 2021;Mace, Protat, Humphries, et al., 2021).These data provide a unique opportunity to study the atmosphere-sea-ice interactions in the region.Figure 1 shows the tracks of the first three voyages made by the ship during the MARCUS campaign.

Instruments and Data
Table 1 lists the subset of Atmospheric Radiation Measurement (ARM) mobile facility instruments installed on board the RSV Aurora Australis that are used in this study.Combined cloud radar and lidar data provided cloud pixel data, including phase and type, as a function of time and altitude.Hereafter referred to as the lidar-radar product, these data were used to analyze cloud phase and generate cloud base height (CBH) and cloud top height (CTH) timeseries.A balloon-borne radiosonde was deployed every 6 hr on average to produce highresolution vertical profiles from in-situ observations of atmospheric thermodynamic quantities above the ship.Ship location data were used in conjunction with satellite retrievals of sea-ice data to determine the mean sea-ice concentration (SIC) around the ship.Downwelling shortwave (SW) and longwave (LW) radiation data used for cloud radiative effect (CRE) calculations (explained in Section 2.3) were provided by radiometers.A total sky imager provided cloud fraction data for all clouds (the imager does not discriminate between clouds within and above the MABL), and surface meteorological measurements were also recorded (McFarquhar et al., 2021).Three ancillary data sets were used in addition to the MARCUS data sets, and are also included in Table 1.Daily updated SIC at 6.25 km grid resolution was provided by the University of Bremen's Advanced Microwave Scanning Radiometer 2 (AMSR2) satellite sensors (Melsheimer & Spreen, 2019).Reanalysis temperature, pressure, radiation, geopotential, heat flux, and MABL height timeseries from the European Centre for Medium-Range Weather Forecasts' (ECMWF) ERA5 data set were also employed in the analysis (Hersbach et al., 2020).Finally, Global Data Assimilation System (GDAS1) meteorological data were used to create airmass backtrajectories (Stein et al., 2015).

Data Processing and Analysis
AMSR2 provides SIC data at daily resolution.For the purposes of this study, the data were interpolated to hourly resolution by assuming that SIC does not change significantly within a 24-hr time period with respect to the atmospheric variables considered here.Mean SIC was then calculated within a 10 km × 10 km box centred on the ship.Each hour of the campaign was assigned a SIC group based on the following criteria: 0%-15% ∼open water; 15%-80% ∼open ice; and 80%-100% ∼close ice, following Armstrong (1972).A further distinction was made using the mean climatological position of the Antarctic Polar Front (APF) to describe warm and cold water north and south of the APF respectively.Orsi et al. (1995) describe the historical, multi-year, hydrographic data used to determine the position of the APF, which was updated by Park et al. (2019) using more recent altimetry data.While there is MARCUS campaign data as far north as 42°S, all observations of sea ice occurred between 62 and 69°S, south of the APF.As such, open ocean south of the APF was contrasted with the SIC groups when considering atmospheric properties above the ship.Therefore, the surface types selected for analysis were warm water, which comprises all observations made north of the APF, and cold water, open ice, and close ice, which comprise all observations made south of the APF.
In order to analyze potential sea-ice-atmosphere relations, determination of cloud phase and vertical extent from the lidar and radar data was first required.Raw backscatter data from the micropulse lidar were processed according to the algorithms described in Guyot et al. (2022), including background noise removal and beam overlap correction.The lidar-radar cloud phase retrieval approach was initially developed to process CloudSat-CALIPSO data (Delanoë & Hogan, 2010).This ground-based retrieval has also been used to evaluate MPC detection from  geostationary satellites (Noh et al., 2019).Calibration of the lidar uses periods of optically thick stratocumulus, in which the lidar ratio is a known constant (O'Connor et al., 2004).Calibration of the W-band radar was performed using a 4.5 dB offset, as the ARM millimeter radar deployed in MARCUS had not been calibrated before departure (Mace, Protat, Humphries, et al., 2021).The accuracy of this calibration estimate is not expected to affect the results of this study as cloud phase determination depends upon detection versus non-detection of clouds by the cloud radar in the presence of SLW detected by the lidar, whatever the calibration accuracy.An issue exists with the lidar depolarization ratio for the first two MARCUS voyages, in that the MPL instrument window artificially changed the polarization of the laser light (Mace, Protat, Humphries, et al., 2021).As such, the depolarization ratio could not be used for cloud phase determination as it was for previous campaigns (e.g., Alexander & Protat, 2018).Instead, we apply a depolarization ratio cutoff of 0.3 to separate ice from liquid during these voyages, as depolarization ratios were consistently greater than 0.3 for cloud bases known to contain ice as recommended in Mace, Protat, Humphries, et al. (2021).For voyage 3, ice presence was determined as a function of layer-integrated attenuated backscatter versus lidar depolarization ratio, as described in Hu et al. (2009), their Figure 4 (left panel).
First, cloud boundaries were determined following the detection algorithm described in Alexander et al. (2021) and first developed by Wang and Sassen (2001).A cloud base detection requires that the backscatter signal increase continuously across three altitude bins in addition to meeting signal-to-noise ratio conditions, while a cloud top detection requires that the backscatter gradient fall once more below the clear-sky gradient defined by the three altitude bins beneath the cloud base.A true cloud top cannot always be defined by lidar alone due to strong attenuation within optically thick clouds, reinforcing the importance of combining lidar and radar data sets to compensate for instrument limitations.The specific cloud phase was then calculated using the relationship between the layer-integrated lidar backscatter and depolarization ratio, the theoretical basis for which is full described by Hu et al. (2009).Vertical temperature profiles drawn from ERA5 were used at this stage to distinguish between SLW and warm liquid.Reanalysis data were used because the rapidly changing nature of low-level clouds would be difficult to accurately capture with the relatively coarse resolution of 6-hourly radiosonde profiles.Furthermore, assessment of ERA5 temperature data against radiosonde observations at the same times and altitudes showed a mean bias error of less than 0.5°C for altitudes less than 15 km, demonstrating that ERA5 data are suitable for distinguishing between warm liquid and SLW.The final cloud phase is then fully described by the merged lidar-radar product.When the lidar detects liquid particles at sub-freezing temperatures, SLW clouds are classified as distinct from MPCs by the presence of a lidar signal and the absence of a reflected radar signal.This is because the millimeter radar is insufficiently sensitive to detect the very small SLW droplets.
As the radar detects only ice, a reflected radar signal coincident with lidar backscatter indicative of SLW results in a mixed-phase classification.For altitudes bins where the lidar signal is fully attenuated and where the radar detects ice, the "Ice or mixed phase" category is assigned, which assumes the possibility of undetected liquid water.The merging of the lidar-radar data is described in detail in Noh et al. (2019), the flow chart for cloud phase determination is shown in Figure 2, and two case-studies (Figures S1 and S2 in Supporting Information S1) provide visual examples of raw lidar and radar data alongside the resultant cloud phase for each pixel.The final lidar-radar product uses 15 m vertical and 1 min temporal resolution.
Regular profiles of atmospheric quantities derived from radiosonde data were generated by interpolating irregular observations onto 5 m intervals between launch and the maximum altitude recorded for each radiosonde.These profiles were then used to generate three measurements, following Naud et al. (2020): lower tropospheric stability (LTS), a property of the MABL; estimated inversion strength (EIS); and the marine cold air outbreak M parameter (M), an alternative inversion strength.The formulae for calculating the LTS are described by Bolton (1980) while Naud et al. (2020) and Wood and Bretherton (2006) provide the formulae for the EIS and M calculations.
These measurements were calculated for conditions where the CTH was less than the current ERA5-defined MABL height plus one standard deviation of the MABL height for the current surface type (hereafter referred to as the "MABL limit"; see Table 2, column 4).Guo et al. (2021) report that, out of four major reanalysis data sets, the ERA5-defined MABL height is the most reliable, though it does consistently underestimate the MABL height by approximately 130 m when compared with data from radiosondes.Furthermore, visual examination of ERA5 MABL height overlayed on plots of cloud phase showed that low-level clouds capping the MABL sometimes occurred within several hundred metres above the ERA5 MABL height.Such clouds may exist within a "buffer zone" where MABL airmasses mix with those from the free troposphere Russell et al. (1998), and observational data from Macquarie Island suggests that SO clouds are commonly observed within this zone (Hande et al., 2012).Therefore, to capture clouds capping the MABL (including those within the buffer layer) while excluding those within and above the free troposphere, clouds above the MABL limit are excluded in all analyses involving inversion strength and surface coupling.
Radiosonde data were further used to determine the coupling status of the MABL.Coupling status was calculated by comparing the potential temperature at the altitude of the lowest CBH and the cumulative mean potential temperature from the surface to this point.If the difference between these two values exceeded 0.5 K, then the cloud was determined to be decoupled from the surface and capping a stable, stratified MABL (Griesche et al., 2021;Sotiropoulou et al., 2014).In addition to using radiosonde data, MABL coupling status, the LTS, EIS, and M parameter were also calculated using ERA5 data to increase the amount of data available for analysis.A comparison of 174 hr of hourly radiosonde-and ERA5-derived coupling status showed a 79% agreement between the pair with a standard deviation of 21%.
Net short-and long-wave cloud radiative effect (SW CRE and LW CRE respectively) were calculated as a function of SIC.Downwelling radiation was provided by two identical instruments on the port and starboard sides of the ship, and the radiation value used at each timestamp was the greater of both instruments to reduce shadowing from the ship's superstructure.For both SW and LW data, the modelled clear-sky downwelling radiation (provided by ERA5) was subtracted from the observed downwelling radiation measured by the radiometers.For the SW radiation, the resultant CRE was then multiplied by one minus the local albedo, assumed to interpolate linearly between 0.07 for open water and 0.8 for 100% SIC (Brandt et al., 2005).For LW radiation, the resultant CRE was multiplied by the local emissivity, assumed to interpolate linearly between 0.98 for open water and 0.95 for 100% SIC (Hori et al., 2006).Cloud base and top heights and cloud minimum and top temperatures were extracted from the merged lidar-radar product as follows.First, lidar-radar cloud data were subsetted to include only low, SLW-containing clouds in order to isolate clouds of interest, with the lowermost cloud pixel in each 1-min timestamp designated the CBH.
For cases where the CBH was used in subsequent analysis such as MABL property, inversion strength, and cloudpercentage-ice calculations, only cloud columns with a CTH lower than MABL limit were selected in order to further restrict analysis to clouds within the MABL.The CTH was found by locating the highest non-precipitating cloud pixel in a continuous cloud column above the CBH, irrespective of cloud phase, and the CTT was assigned using the ERA5 temperature of this pixel.In order to match the methodology used by Naud et al. (2020), cloud top temperature was used in Section 3.5 instead of cloud minimum temperature, while cloud minimum temperature, the coldest temperature in the column of pixels between the CBH and CTH, was employed in Section 3.8 in order to remain consistent with Griesche et al. (2021).Griesche et al. (2021) note that using the temperature of the topmost cloud pixel (i.e., CTT) in directly comparing cloud phase and temperature may introduce positive temperature biases due to the increased probability of drawing temperatures from above the inversion capping the MABL.Regardless, both cloud temperature measures were calculated from the combined lidar-radar product and ERA5 data used in this study, and the difference between the two was found to be small, with cloud top temperatures 0.07 K greater on average than the cloud minimum temperatures, with a standard deviation of 0.25 K.
The first analysis of cloud phase using the lidar-radar product was performed by determining the cloud frequency of occurrence (CFO) as a function of altitude.For the CFO calculations, the cloud phase data were grouped into surface types and by voyage prior to determining the frequency of occurrence in order to compare the relative CFO of specific cloud phases across different surface types and to account for seasonal changes.In calculating CFO, ice virga were considered "Ice" and subsumed into that category.
It is also essential to account for cloud influences beyond the local meteorology, as airmasses traveling from further north over the SO, where complex frontal systems are more frequently encountered (Berry et al., 2011), will likely have different properties to those originating over the Antarctic coast or continent.This problem was addressed by analyzing airmass HYSPLIT back-trajectories in GDAS1 meteorological data (Stein et al., 2015) using the splitr package in R (Iannone, 2021).The back-trajectories were initialized hourly at the median coordinates of the ship using the median CBH determined by the lidar-radar product for that hour, or else the ceilometer if the former did not return a valid CBH.72-hr airmass back trajectory results were used to visualize the distribution of recent airmass locations across the SO and Antarctic continent, and separation of these results by surface type provides a qualitative means by which to account for recent differences in the location and properties of airmasses above different surface types.A minor technical issue with the HYSPLIT code meant that back-trajectories for the last 2 hr of each week of each month could not be determined.However, as this issue only affected 36 out of the 2,740 hr of data available from voyages 1-3 (approximately 1.3%), it was deemed insignificant and consequently ignored.
The second analysis of cloud phase was performed by calculating the percentage of liquid-containing clouds precipitating ice (hereafter cloud-percentage-ice, CPI) as a function of cloud minimum temperature, relative to all clouds with an observable liquid-containing layer, whether "SLW" or "Mixed phase."Following Mace, Protat, and Benson (2021), we scanned for the presence of ice (including the categories "Ice virga," "Ice," and "Ice or mixed phase") pixels 100 m below a liquid layer for clouds with tops less than the MABL limit.A cloud column might contain ice above the lowermost liquid layer, but only contributed to the CPI count if it was actively precipitating ice.In this way, a single cloud column was counted as either 1 in the numerator if ice was found or 0 if it was not.We also excluded multilayer cloud regimes, including clouds with top heights above the MABL limit and clouds receiving precipitation from aloft in seeder-feeder fashion, in order to restrict the analysis to simple, low-level, single-layer, SLW-containing clouds.The cloud minimum temperature for each column was rounded to the nearest 5°C for temperatures below 10°C or to the nearest 2.5°C for temperatures above 10°C.
Error bars were calculated as follows.A single cloud may include any number of 1-min observations, and observations made of the same cloud are not independent and potentially autocorrelated.For a timeseries containing n observations that is autocorrelated, Mitchell et al. (1966) provides a method for calculating the effective observation count, n eff , which is n eff = n × (1 ρ 1 )/(1 + ρ 1 ).ρ 1 is calculated using a lag of one time step in the autocorrelation function.The standard error bars were calculated using n eff determined separately for each cloud minimum temperature bin.

Sea-Ice Concentration and Surface Type
We established four surface type categories to divide the MARCUS campaign data based on SIC and position relative to the APF.As each voyage took place further into the austral summer, it was also important to determine both how the surface types were distributed across the voyages and the latitudes covered by each voyage.For example, close ice was observed during voyages 1 and 3 but not during voyage 2, though voyage 2 did not travel farther south than 66°S.Latitudes 64°C to 66°S are dominated by observations from voyage 2, and therefore by data from early mid summer.Figure 3 presents a full breakdown of how the data varied by sea-ice concentration, surface type, latitude, and voyage.

Radiosonde Profiles
Radiosonde observations made during the MARCUS campaign show differences in meteorological conditions in the MABL across surface types (Figure 4).Lower-atmosphere air temperature decreases poleward, and dew point temperature follows the same trend.The salient feature is a sharp decrease in the dew point between 800 and 600 hPa over close ice corresponding to a drop in relative humidity directly measured by the radiosondes (not shown).Westerly winds prevail at all altitudes over warm and cold waters and above 500 hPa over open and close ice.Easterly winds prevail in the lower troposphere over open and close ice, and are potentially associated with dry, katabatic outflow from the Antarctic continent (Grazioli et al., 2017).Such winds also increase SIC along the Antarctic coast (Thompson et al., 2020), limiting surface sensible heat flux and evaporation (Figure S8 in Supporting Information S1) and further contributing to the decrease in relative humidity at high latitudes.These profiles show that airmasses over higher SIC surface types are colder and drier than those over lower SIC and warmer waters, especially in the MABL.

Inversion Strength Distributions
Radiosonde observations also reveal that the mean state of the SO MABL is different across surface types.We analyzed measures of lower-troposphere inversion strength and MABL stability by surface type using statistics derived from both radiosonde profiles and ERA5 data (Figure 5).More positive LTS (Figure 5a) and EIS (Figure 5b) values indicate increased MABL stability and a stronger temperature inversion respectively.A more negative M parameter (Figure 5c) also corresponds to a stronger inversion.While there is some indication of a stronger inversion over cold water than warm water, most prominently in the M parameter results, these surface types are broadly similar.Much larger differences exist moving from cold water to open ice, with stronger inversions above both higher-SIC surface types shown in the LTS and EIS results.This suggests increased stability and less mixing of the MABL over sea ice than over open ocean.A discrepancy exists in the M parameter results between the radiosonde and ERA5 data, with ERA5 reporting MABL stability similar to warm and cold waters.
The skin temperature values used to calculate the M parameter were sourced from ERA5 data for both the radiosonde and ERA5 calculations, which means that disparities between results for different data sources are solely the result of differences between radiosonde and ERA5 potential temperature values at 800 hPa.However, after removing clouds above the MABL limit, only 8 data points remain for ERA5 and only 1 for radiosondes over close ice, and this small sample size means that further data are required to be confident in the close ice results.
For open and close ice, the interaction between the cold, dry, surface winds blowing from the east (Figures 4c and  4d), typically associated with strong katabatic outflow from Antarctica, and MABL stability is complex.These winds descend and pour outwards from the continent, stripping heat from the ocean surface and causing sea-ice growth and equatorward expansion (Thompson et al., 2020).If the MABL is initially in an unstable state, such winds can enhance turbulence and vertical mixing, maintaining the convective overturning.However, under stable conditions, strong, cold winds can cause the MABL to become more stratified by decreasing the amount of TKE available and required to maintain convective overturning, especially when the underlying water is warmer than the air itself (Zheng & Li, 2019).The fact that the MABL is consistently less stable over open ice than close ice suggests that the sea ice itself may play a role in increasing MABL stability.This introduces other factors such as surface wind drag and heat flux that also affect MABL stability, and we now consider possible limitations to this analysis of MABL stability due to these factors.

Surface-Boundary Layer Physical Connections
When investigating possible links between surface conditions and the microphysical properties of clouds capping the MABL, it is important to understand the underlying physical processes connecting surface properties such as SIC with the atmospheric layers above.Surface heat fluxes contribute to the thermodynamic conditions that determine the strength of convective mixing in the lower atmosphere and will therefore influence cloud conditions atop the MABL.We examined ERA5 surface sensible and latent heat fluxes around the ship as a function of both latitude (Figure S7 in Supporting Information S1) and surface type (Figure S8 in Supporting Information S1).ERA5 yields heat flux results in line with those produced by the foremost ocean reanalysis for the SO and uses a resolution of 0.25°× 0.25°, which is high relative to the global reanalysis standard (Josey et al., 2023).Average surface heat fluxes during the MARCUS campaign were generally negative, denoting a consistent loss of heat from the ocean to the atmosphere through both direct heat exchange (sensible) and phase changes such as condensation and evaporation (latent).Heat fluxes were also larger for warm waters further north, falling to approximately 20 Wm 2 , which persists across most of the sea-ice zone.Median surface sensible and latent heat fluxes were similar across the three SIC surface types ( 20 Wm 2 and 5 Wm 2 respectively).Median sensible heat flux diminished slightly from cold water to close ice, as did its variability, though the changes were small.Latent heat fluxes were rarely positive over open and close ice, a realistic result given the inhibiting effect of sea ice on ocean surface evaporation.This is significant because surface latent heat fluxes are a primary source of TKE (Zheng & Li, 2019), and the reduction of this surface buoyancy likely results in the increased MABL stability over higher-SIC surface types (Figure 5).However, there was no consistent trend in the heat flux changes between the SIC surface types, whereas heat fluxes for warm waters were distinct in both their magnitude and variability.Therefore, as our investigation focuses on differences in cloud properties between the SIC surface types, we will proceed with examining such correlations without further consideration of surface heat fluxes.
Another factor to consider is surface wind drag over sea ice.Overland (1985) used aircraft measurements to estimate air-sea drag over multiple sea ice types in the Arctic and found that drag coefficients maximize in conditions of strong (>5 ms 1 ) off-ice winds over the marginal ice zone (MIZ) and especially broken ice.More recently, aircraft observations over the Arctic MIZ have also shown that surface roughness (and therefore wind drag) peak for SIC between 0.6 and 0.8 (Elvidge et al., 2016).Consideration of wind drag is important because the buoyancy provided the strong, cold winds typical of katabatic outflow tends to increase MABL coupling, and the surface roughness of sea ice further enhances this effect under such windy conditions.However, there are currently no data on Antarctic sea-ice roughness available for determining atmospheric drag around the ship during the MARCUS campaign.Furthermore, Renfrew et al. (2019) note that the effect of sea ice on surface drag is indirect, as ice morphology is the salient variable rather than SIC itself.The lack of available data places a full consideration of wind drag outside the scope of this paper, and we echo Johnson et al. (2022) in calling for an extension of existing and proven mapping techniques using satellite observations to retrieve sea-ice roughness over the Antarctic ice sheets.We acknowledge the limitations that this places on our work tying SIC to cloud macrophysical properties, but contend that meaningful correlations may still be observed and reported.
Certain cloud properties also provide insight into the state of the MABL, so we now examine how cloud layer altitude, temperature, and total cloud fraction vary with MABL stability.

Cloud and Radiation Properties
In Figure 6, we show the distributions of SW and LW cloud radiative effect, total cloud fraction, lowermost cloud layer boundaries, and cloud top temperatures of all clouds observed during the MARCUS campaign for each surface type.
The close ice results are the most distinct across the cloud properties examined.Clouds are least prevalent over close ice, with a median cloud fraction of approximately 70%, compared with almost 100% for all other surface types, shown in Figure 6c.Less cloud cover results in smaller SW and LW CREs, as can be seen in Figures 6a and  6b.The close ice group also sees higher cloud bases and top heights (Figure 6d) and colder top temperatures (Figure 6e) on average than clouds over other surface types.From this, we can infer that clouds over close ice are higher, colder, and less prevalent than clouds over other surface types.This appears at first to contradict the earlier finding that the MABL is most stable over close ice (Figure 5).However, MABL stability was only calculated cloud tops less than the MABL limit in order to prevent high-altitude clouds entirely isolated from the MABL (e.g., icy cirrus) from impacting the results.Therefore, while low-level clouds over close ice are less prevalent than over other surface types, the MABL beneath such clouds tends to be more stable.
Regarding the other surface types, median CBH and CTH both decrease from cold water to open ice, shown in Figure 6d.In Figure 6e, CTTs are similar for cold water and open ice and larger over warm waters.In the SW data in Figure 6a, the median CRE over open ice (∼ 60 Wm 2 ) is smaller than over cold water (∼ 100 Wm 2 ), and the interquartile range in measurements decreases for the same comparison.In the LW data in Figure 6b, the median CRE over open ice (∼75 Wm 2 ) is slightly larger than over cold water (∼77 Wm 2 ) but the range of measurements covers lower values than for the same comparison.
Factors affecting CRE include cloud fraction, cloud top height, cloud layer count, and SIC.Southern Ocean cloud systems are often complex, with a high frequency of multi-layer cloud occurrence involving both low-(cloud top <3 km) and mid-(cloud top from 3 to 6 km) level clouds (Mace et al., 2009).Such clouds likely contribute to the large CRE variability, especially over warm and cold waters, which motivates their exclusion from certain analyses.The relative cloud fraction of cold water and open ice explains some of the SW CRE difference between these two groups, with less cloud cover over higher-SIC surface types logically resulting in a smaller SW CRE.Low clouds also increase surface LW radiative forcing more than high clouds (Harrop & Hartmann, 2016), which may contribute to the higher median LW CRE for lower CBH and CTH over open ice.The exact role of Antarctic sea ice in affecting radiative transfer has yet to be investigated in-depth.Using data from Hori et al. (2006), we determined that 100% SIC absorbs only 3% more LW radiation than open water.The direct effect of sea ice on surface LW radiative forcing is therefore marginal.However, with up to 80% of incident SW radiation reflected off sea ice, it is likely that surface type affects high-latitude SW CRE over the SO, factoring into both the open and close ice SW CRE values.

Cloud Phase Frequency of Occurrence
At this point, we have not yet examined cloud phase as a function of surface type.As discussed, the liquid versus ice content of clouds affects not only their radiative effects but also their longevity.Cloud temperature is an important factor here, as the number concentration of INPs, an important factor in determining the efficiency of primary ice production (PIP) within liquid clouds, increases approximately tenfold for each 5 K decrease in temperature (DeMott et al., 2015).Given that liquid clouds reflect more SW radiation back to space than icecontaining clouds (Hartmann, 2016), and that sea ice also increases surface SW reflectivity, we would expect interplay between downwelling SW radiation, cloud phase, and sea ice.The smaller radiative effects of clouds above open and close ice may yet be fully explained by the factors discussed above.However, we will now examine the cloud phase data from the MARCUS campaign that are available for analysis.
The CFO for different cloud phase categories by surface type is shown in Figure 7. Below 1 km, "All clouds" were observed at a similar maximum frequency (∼55%) for each surface type except for close ice (∼30%), and generally decreased in frequency of observation above 1 km (Figure 7a).The cold water and open ice surface types have similar profiles for "All clouds," with slightly less cloud occurrence over open ice at most altitudes (in agreement with Figure 6c).Close to the surface, "Ice" clouds are more common over cold water than over open ice surfaces by approximately 10% (Figure 7b), while the reverse is true for "Ice or mixed phase" clouds (Figure 7c).The salient observation from this plot is that "Mixed phase" clouds are approximately three times more prevalent over open ice (∼15%) than over cold water (∼5%) below 1 km, and are scarce over both warm water and, notably, close ice (Figure 7e).The "SLW" CFO is generally less than 2% regardless of surface type (Figure 7d).However, this category also shows similarly shaped curves to the "Mixed phase" (which does contain supercooled liquid water) CFO across all four surface types.
The "Ice or mixed phase" category contains all cloud phase data where the radar detected ice but where no lidar signal was detected, meaning the presence of liquid water could not be confirmed.This is demonstrated by the fact that, below 3 km, the trend for the "Ice" cloud category is one of decreasing CFO with altitude, while the reverse is generally true for "Ice or mixed phase."The "Mixed phase" CFO is under 5%, excepting the 15% peak for clouds at less than 1 km altitude over open ice.This means that the ratio of clouds confirmed to contain both SLW and ice to clouds where SLW presence cannot be confirmed is typically no greater than 1:6, with some variation by surface type.Notably, when the open ice curve for "Ice or mixed phase" dips at approximately 500 m and again around 1.5 km, corresponding increases are visible in both the "Mixed phase" and "SLW" CFO curves.Additionally, while the peak in the "Mixed phase" CFO centred around 500 m over open ice is also visible in the "SLW" CFO, it is one-tenth as as large in the latter.Altogether, this means that the likelihood of there being significant quantities of undetected, pure SLW in the "Ice or mixed phase" category is low, and that much of the data in this category is simply ice, where either the lidar backscatter was extinguished before reaching the layer or where sparse ice particles were invisible to the lidar (as evidenced in Figures S1 and S2 in Supporting Information S1).
The CFO observations in Figure 7 suggest a higher prevalence of low-level clouds containing SLW, especially MPCs, over open ice than over other high-latitude surface types.This result may be compared with the findings of Listowski et al. (2019), who used satellite lidar-radar data to investigate seasonal variation in low-level (under 3 km altitude) cloud phase over the high-latitude SO and Antarctica.Listowski et al. (2019) found that the prevalence of SLW-containing clouds was lower over sea ice than open ocean, in apparent contrast with our result; however, they also noted that this decrease was small in spring (∼10%), negligible in summer, and more apparent in MPCs than unglaciated SLW clouds.Overall, the number of low-level, unglaciated SLW clouds observed in Figure 7 is small relative to mixed-phase clouds, irrespective of surface type.The higher mixedphased CFO over open ice than cold water persists from late spring through early/mid-summer (voyages 1 and 2, Figures S3 and S4 in Supporting Information S1 respectively).In late summer (voyage 3, Figure S5 in Supporting Information S1), the mixed-phase CFO decreases, with loose convergence observed between the cold water and open ice curves, while an increased MPC prevalence is observed over close ice between 1 and 3 km.confirm with the voyage 2 data (Figure S4 in Supporting Information S1).We also observe that the prevalence of unglaciated SLW clouds is similar for cold water and open ice from late spring through early/mid-summer, though with greater representation at higher altitudes over cold water than close ice (Figure S4, panel d in Supporting Information S1).In late summer (Figure S5, panel d in Supporting Information S1), low-level SLW clouds are more common over cold water than open ice.Altogether, this shows that some of the discrepancy between our results and the cloud phase observations made by Listowski et al. ( 2019) disappears when seasonality is accounted for, while remaining differences may yet be explained by unavoidable differences between groundbased and satellite observational techniques.Surface instruments are advantaged in observing low-level clouds due solely to their proximity and a lack of intervening clouds, but are spatially limited, while satellites can aggregate data across a much larger spatial domain but with observation potentially obscured by mid-and highlevel cloud cover.This reinforces the fact that, when working together, top-down and bottom-up cloud observations produce the best combined picture of cloud phase.Mace, Protat, and Benson (2021) investigated the occurrence of MPCs over the SO using both satellite and surface-based data, focusing their analysis of surface observations on clouds with CBHs below 2 km and with sub-zero temperatures.Incorporating data from the CAPRICORN I, and CAPRICORN II campaigns, they found that the frequency with which liquid water was observed in these clouds was greater for clouds with top heights below 1 km than for clouds extending to altitudes as high as 5 km.While this result agrees with our finding of more prevalent MPCs below 1 km (specifically over open ice), Mace, Protat, and Benson (2021) found that this result was not as strong in the satellite data examined, once more underlining the essential role of groundbased observations in gathering data on low-level clouds in the mid-and high-latitudes.
Because SLW presence is associated with higher cloud shortwave reflectivity, we might expect to see differences in the SW CRE for different SLW and MPC occurrence frequencies.For example, the increased presence of low- level MPCs over open ice should correspond to a greater SW CRE. Figure 6a appears to show the opposite, with SW CRE decreasing from cold water to open ice.Furthermore, while sea ice enhances surface albedo, previous studies have demonstrated that multiple scattering between clouds and reflective surfaces can actually increase SW CRE (Fitzpatrick et al., 2004), a finding that also contrasts with the results of Figure 6a.The patchier cloud cover over open ice (Figure 6c) may minimize or even reverse this effect of multiple scattering by allowing reflected radiation to escape upwards, decreasing SW CRE.However, as previously mentioned, the high prevalence of multi-layer clouds over the SO also complicates radiative transfer to the point where their frequency correlates with biases in model surface radiative flux (Trenberth & Fasullo, 2010).Addressing the problem of SO multi-layer clouds is beyond the scope of this paper, so we return our focus now to possible connections with sea ice.
With close ice observed at more southerly latitudes (Figure 3d) and fewer low-level clouds observed over close ice generally (Figures 6c and 7a), the dearth of SLW-containing clouds observed over this surface type may be due to the recent meteorological history of the airmasses contributing to these datapoints.As observed in Figure 4, the close ice data are more associated with the dry, cold easterlies originating over the Antarctic continent than with coastal and open-ocean airmasses.Once again, the close ice category is therefore an outlier.The lower fraction of clouds in the lower troposphere over close ice confirms the total cloud fraction results in Figure 6c.Furthermore, clouds over close ice exist largely in the "Ice" and "Ice or mixed phase" categories (Figures 7b and  7c).The differential persistence of sea ice at different longitudes may provide an explanation for the differences in the "Ice or mixed phase" data between close ice and the other southerly surface types.There is very little SLW over close ice (Figures 7d and 7e), and this reduces the possibility of multiple scattering.Close ice also reflects the most SW radiation of the four surface types.Combined with the overall decrease in cloud fraction and more southerly latitude, these factors account for the very small SW CRE and relatively small LW CRE over close ice relative to other surface types.We now turn to airmass origin as another factor influencing observed cloud properties.

Airmass Back-Trajectories
In order to account for the possibility of airmasses with disparate origins and compositions preventing robust comparisons between cloud phase and surface type, the density of 72-hr HYSPLIT airmass back-trajectories over the SO and Antarctic continent was assessed by surface type, shown in Figure 8. Airmasses over warm water originate largely over water north of the APF.Airmasses over cold water have spent significant time over both open ocean south of the APF and coastal waters around Antarctica, with a small contribution from over the continent itself.This contrasts somewhat with both the open and close ice results, which tend to show airmass histories more closely confined to the Antarctic coast, though cold water was still observed for extended periods in close proximity to Antarctica, largely during voyage 3 in mid-to late-summer.Overall, airmasses over open and close ice have similar origins, with a higher density of back-trajectories around the coastline.This means that disparate results for these two surface types may be associated with the sea-ice concentration itself, rather than confounded by recent meteorological history.
A notable observation in the CFO results (Figure 7) is the increased low-level MPC prevalence above open ice versus the other surface types, especially close ice, persisting through late spring and early/mid-summer (Figures S3 and S4 in Supporting Information S1 respectively).Further clear disparities exist between the open and close ice results in the radiosonde profiles (Figure 4), MABL stability measurements (Figure 5), and cloud macrophysical properties (Figure 6).Furthermore, Listowski et al. (2019)'s satellite data study found evidence that seaice fraction, marine influences, and temperature (which vary seasonally) all affect the thermodynamic phase of low-level clouds over the SO.Our results support these findings, with warmer, moister air and more SLWcontaining clouds observed over open ice compared with a more stable boundary layer and fewer clouds overall over close ice.The marine influence, including evaporation and the emission of aerosols from the ocean into the atmosphere, may be directly controlled by sea-ice concentration, where the almost-total coverage of close ice blocks this influence in a way that the incomplete coverage of open ice does not.However, these results do inform of possible relationships between cloud ice formation rates (especially SIP), surface type, and MABL coupling status.

Percentage of Clouds Precipitating Ice
To investigate whether differences in SIP between MABL-coupled and -decoupled clouds exist, we examined the percentage of clouds precipitating ice (CPI) as a function of cloud minimum temperature, shown in Figure 9.The observation count refers to the number of 1-min lidar-radar observations.All cloud pixels counted as either "Ice" or "Ice or mixed phase" contribute to the CPI in Figure 9.However, as clouds in Figure 9 are derived from data columns containing multiple pixels of potentially different phase, we cannot expect CPI values to reflect the CFO values shown in Figure 7.
Between 10 and 5°C, the CPI is highest above open ice (50%-95%), followed by warm water (35%-75%) then cold water (25%-50%).The available close ice data suggests a CPI ranging from 100% to 30% between 20°and 5°C, but this surface type is excluded from subsequent discussion due to the low quantity of and lack of temperatures covered covered by the data (Figure S6 in Supporting Information S1).The available data for the other surface types indicates an association between surface type and CPI for temperatures greater than 20°C.If convective processes beneath and within MABL-coupled clouds are distributing surface-sourced INPs into the free troposphere, then this may enhance PIP in these clouds, thereby increasing the CPI of MABL-coupled clouds  as cold water, coupled cloud ice precipitation is consistently above 85%, even at temperatures as high as 5°C.One possible explanation for this is that, while SIP drives ice precipitation in both coupled and decoupled clouds, PIP is less active in decoupled clouds, resulting in higher ice precipitation rates for coupled clouds and a SIP associated curve visible in the decoupled data.
Causes of glaciation in SLW-containing clouds include PIP, controlled by INP availability, and SIP, controlled by splintering of existing ice crystals within the cloud and incident precipitation from higher-altitude clouds.However, we removed cases of higher-level cloud seeding from our data set prior to analysis.Our results suggest the possible existence of INPs at high latitudes over the SO with the strongest effects observed over the sea-ice zone.Furthermore, our results agree with those of Griesche et al. (2021), who found that Arctic clouds coupled to the surface were more likely than decoupled clouds to precipitate ice for temperatures between 10°and 0°C.PIP in cold clouds is generally dominated by biogenic aerosols, though bacteria are only active as INP at temperatures above 10°C (Kanji et al., 2017).Griesche et al. (2021) compared their Artic CPI results with observations made over Leipzig, Germany, and found evidence that Arctic INP concentrations are strongly influenced by continental aerosols.Griesche et al. (2021) reinforced this point using in-situ INP measurements made in the Arctic MABL, and found that higher INP concentrations existed in coupled airmasses than decoupled.
Despite the fact that Griesche et al. (2021) performed their study in the Arctic and over a more homogeneous domain, our results show the same pattern of relative CPI in coupled versus decoupled clouds in the high-latitude SO.Mace, Protat, Humphries, et al. (2021) examined the microphysical properties of non-precipitating clouds over the Southern Ocean using MARCUS and CAPRICORN I and II data.They found a distinct bimodality in the properties of clouds along the East Antarctic coastline, where SIC is generally higher.They identified cases of sudden changes in cloud microphysical properties between modes consistent with more northerly MABL clouds and modes with markedly increased CCN.The latter mode maintains higher liquid water paths, resulting in higher SW CREs from these clouds as well as greater cloud top reflectivity.They also found no evidence that these airmasses were affected by continental aerosol emission, which suggests that this high-latitude bimodality is not the product of landmasses contributing continental aerosols but a consequence of oceanic emission of biogenic aerosols into the MABL.
In a recent study of SIP over the mid-latitude SO, Atlas et al. (2022) investigated the effects of including HM rime splintering in storm-resolving simulations.They found that rime splintering reduced cumulus MPC occurrence, leading to a correspondingly weaker SW CRE by increasing ice crystal concentrations in MABL clouds.In observing and modelling ice particle formation in cumulus clouds over the United Kingdom, Crawford et al. (2012) found that INP concentrations of at least 10 5 cm 3 are required for HM rime splintering to commence, a finding reinforced by Huang et al. (2017).INP concentrations exceeding this threshold were also reported during the MARCUS campaign when the ship was located close to Mawson Station (Vignon et al., 2021).Our results suggest that, at higher latitudes and especially over SO sea ice, SIP occurs in both MABL-coupled and -decoupled clouds.If SIP is to some extent PIP-dependent, then coupled clouds may experience higher levels of SIP due to the enhanced presence of primary ice particles and/or INPs mixed from below, which could be raised by the generating cells within clouds present at these latitudes (Alexander et al., 2021).Regardless, as observed by Atlas et al. (2022), the relationship between rates of PIP and SIP is poorly constrained and should remain a focal point for future investigation into SIP over the SO.
We have found possible evidence of SIP in SO clouds at high latitudes over the sea-ice zone, especially where clouds are decoupled from the MABL.SIP may also enhance the percentage of clouds precipitating ice at temperatures between 10°and 0°C, though the results shown in Figure 9 do not provide direct evidence of this, showing only that the CPI of coupled clouds is greater than that of decoupled clouds.The difference in CPI over cold waters and over open ice suggests that high-latitude INP concentrations vary with surface type and are possibly linked to the sea ice itself.While these data are distinct from results north of the APF, where the effect of MABL coupling status on CPI is less clear, INP transport may play a confounding role.The bimodality in highlatitude SO clouds discovered by Mace, Protat, Humphries, et al. (2021) is an important link in this story, as the enhanced CCN levels and resultant likelihood of SLW presence in the mode characterized by elevated aerosol presence may provide the right kind of environment for elevated PIP levels to lead to enhanced SIP and an overall larger CPI of the kind we observed over open ice.

Conclusions and Future Work
We have presented an analysis of cloud properties and boundary layer strength over Southern Ocean waters and sea ice using data from the ship-based MARCUS field campaign that took place from November 2017 to March 2018.We sought to examine how changing surface and atmospheric conditions could influence cloud phase and frequency of occurrence.We found that the lower-troposphere is generally colder and drier over higher sea-ice concentrations than over open waters.The stability of the marine atmospheric boundary layer is also associated with sea-ice presence, as the strength of the inversion between the boundary layer and the free troposphere was positively correlated with higher SIC surface types.
In general, higher-altitude, colder clouds are more strongly associated with stronger inversions and a stratified troposphere, especially over high-concentration sea ice.This is important for cloud phase and radiative effect, as the circulation of ice-nucleating particles through clouds to initiate primary ice production and alter cloud phase may depend on surface sources, which are more likely to affect cloud phase in a coupled than a decoupled boundary layer.We found that the frequency of occurrence of MPCs is three times higher over open ice than other surface types and that the percentage of SLW-containing clouds precipitating ice is also highest over open ice.Boundary layer coupling also plays an important role, as the percentage of ice-precipitating south of the APF is higher coupled than for decoupled clouds, especially over sea ice.There are indications of secondary ice production, specifically the HM process, occurring in these clouds, evidenced by an uptick in the percentage of clouds precipitating ice between approximately 5°and 10°C, in agreement with the HM activity window of 3°to 8°C.
Based on these results, we believe there are differences between the characteristics of high-latitude coupled and decoupled clouds over the SO in terms of their thermodynamic phase.South of the APF, the boundary layer tends to be more stable for higher SICs, suggesting less mixing of the lower troposphere over these surface types, though we do not draw any causal conclusions.Truong et al. (2020) produced seven synoptic clusters to describe the varying conditions of the SO MABL, rather than the simple "coupled-decoupled" differentiation used here.With sufficient data to populate all MABL clusters and surface types, their framework could be used to more thoroughly analyze cloud properties as a function of both sea-ice concentration and detailed MABL state.
Evidence of association between sea-ice concentration and cloud phase implies that sea ice may influence surface radiative forcing through more than just reflection of shortwave radiation.We recommend that this study be succeeded by ship-based campaigns to gather year-round atmospheric data over Antarctic sea ice to further investigate its involvement in cloud phase changes.As observed at the beginning of this paper, the Southern Ocean's isolation has meant that there has historically been less observational data available to researchers intending to improve simulations of weather and climate in the region.Fortunately, we expect to see a marked increased in data availability in the coming years, with inter-seasonal, mobile campaigns and multi-year studies planned up to 2027, listed in more detail in Mallet et al. (2023).We hope that this data will be utilized in climate model sensitivity testing to further investigate the link between clouds, sea-ice, aerosols, and the properties of the planetary boundary layer over the Southern Ocean.

Figure 1 .
Figure 1.Tracks for the RSV Aurora Australis' first three voyages in 2017/18 as part of the MARCUS campaign.The dates for these three voyages were: voyage 1, 29/10/2017 to 03/12/2017 (late spring); voyage 2, 13/12/2017 to 11/01/2018 (early/ mid-summer); and voyage 3, 16/01/2018 to 05/03/2018 (mid-/late summer).The fourth voyage, to Macquarie Island, was not included as it did not transit the sea-ice zone.The dashed black line shows the climatological-mean latitude of the Antarctic Polar Front, while the orange and red dotted lines indicate the AMSR2 median sea-ice extent for November 2017 and February 2018 respectively.Landmasses are shaded gray and SO bathymetry is shown in blue.

Figure 2 .
Figure 2. Flow chart showing the determination of cloud pixel phase from lidar and radar observations and ERA5 data.The return signal gradient algorithm used to determine the initial cloud boundary is described byWang and Sassen (2001).The radar minimum detectable signal (MDS) is calculated as a function of range: MDS(range) = MDS 1km + 20× log 10 (range), with MDS in dBZ and range in kilometers.*Due to the depolarization issue affecting the first two voyages, a cutoff of 0.3 was used for these data.However, during the third voyage, ice presence in the depolarization data was determined following the theoretical relationship shown.This is fully described inHu et al. (2009); Figure4(left panel), who used CALIPSO/ CALIOP data pointing 0.3°off nadir, which is most similar to the MARCUS instrumentation.

Figure 3 .
Figure 3. (a) The number of hours for each voyage spent in each 10% SIC bracket on a logarithmic scale, (b) the number of hours of data available for the four surface type categories defined for analysis, total and by voyage, (c) the relative percentage of time spent in each 1°latitude band by voyage, and (d) the relative percentage of time each surface type was observed in each 1°latitude band.Because sea ice is the focus of this study, and as no warm water was observed south of 56°S, latitudes north of this are omitted for visual clarity in panels c and d.

Figure 4 .
Figure 4. Skew-T logP mean profiles of radiosonde-derived temperature (red) and dew point temperature (blue) from 407 soundings by surface type: (a) warm water, (b) cold water, (c) open ice, and (d) close ice.Wind barbs at pressure levels are included on the right of each panel.

Figure 5 .
Figure 5. Marine atmospheric boundary layer measurements derived from radiosonde profiles and ERA5 data, divided by surface type.Shown here are (a) lower tropospheric stability, (b) estimated inversion strength, and (c) the marine cold air outbreak M parameter.

Figure 6 .
Figure 6.Cloud and radiation properties derived from the broadband short wave (SW) and long wave (LW) radiometers, total sky imager, lidar-radar, and ERA5 temperature data.(a) SW cloud radiative effect, (b) LW cloud radiative effect, (c) total cloud fraction, (d) cloud base height and cloud top height, and (e) cloud top temperature.
Listowski et al. (2019) report a maximum in the unglaciated SLW fraction at the beginning of summer, which we

Figure 7 .
Figure 7. Cloud frequency of occurrence vs. altitude derived from the lidar-radar product, grouped by surface type.(a) "All clouds," (b) "Ice," (c) "Ice or mixed phase."(d) "SLW," (e) "Mixed phase," and (f) "Warm liquid" are shown as a percentage of time spent in each surface type.Note that some cloud phase categories were subsumed into others where appropriate: for example, ice virga observations were counted under "Ice."Note also the different x-axis range for panel d, intended to make visible the vertical distribution of the relatively small "SLW" percentages.

Figure 8 .
Figure 8.The latitude-longitude density of 72-hr airmass back-trajectories for each of the four surface types.The climatological-mean latitude of the Antarctic Polar Front is shown by the black dashed line, while the orange and red dotted lines indicate the AMSR2 median sea-ice extent for November 2017 and February 2018 respectively.Landmasses, predominantly Antarctica, are also shown.

Figure 9 .
Figure 9.The percentage of clouds precipitating ice as a function of cloud minimum temperature.Standard error of the mean bars are shown, and were calculated using the effective observation count.Shown here are (a) data by surface type, (b) warm water by coupling status, (c) cold water by coupling status, (d) open ice by coupling status, and (e) close ice by coupling status.Cloud minimum temperatures were rounded to the nearest 5°C below 10°C, and to the nearest 2.5°C above 10°C.

Table 1
Data Sources Used in This Study, Including ARM-MARCUS Instruments Installed on Board the RSV Aurora Australis, the Satellite-Borne Advanced Microwave Scanning Radiometer 2, ERA5 Reanalysis Data, and GDAS1 Reanalysis Data

Table 2
Computed Statistics for the ERA5-Defined MABL Height for Each Surface Type