The AMF2 instruments, measurements, and their corresponding uncertainties and references are listed in Table 1. Because AMF2 was designed to support shipboard deployments, the baseline suite of instruments are marine-focused, including the 95 GHz W-band Atmospheric Radiation Measurement (ARM) user-facility Cloud Radar (WACR), ceilometer, micropulse lidar (MPL), microwave radiometer (MWR), aerosol observation system (AOS), meteorological measurements (MET, which includes the following data: temperature, pressure, specific humidity, and wind direction and speed) on the ship, rain gauge, and the radiosonde soundings. The combined cloud radar and ceilometer measurements can provide the cloud boundaries as long as there are no optically thin clouds and the cloud base heights (Hbase) are not greater than the upper limit (7.7 km) of the ceilometer. The micropulse lidar will be used to identify optically thin clouds and the clouds with Hbase>7.7 km. A previous study has shown that these additional clouds detected by the micropulse lidar can be a non-negligible supplement to the total cloud fraction (Mace et al., 2021). A detailed description of the instruments and the cloud parameters during MARCUS can be found in Mace et al. (2021) and McFarquhar et al. (2016, 2021).

The cloud occurrence frequency can be determined through the following two steps: the column cloud fraction is simply the ratio of cloudy samples to the total observations every 5 min, and the occurrence frequency for each type of cloud during the entire time period equals the ratio of the number where the column cloud fraction is greater than zero for the total 5 min samples. In order to accurately estimate the cloud temperatures, we adopted a linear interpolation method based on the daily balloon soundings (4 to 5 times per day) to achieve a better temporal resolution of the vertical profiles of temperature, pressure, and specific humidity. The method considers MET measurements to ensure vertical continuity and adjacent soundings for temporal continuity. Using these interpolated atmospheric profiles, cloud temperatures can be obtained at a 5 min temporal resolution.

The cloud liquid water path (LWP) and atmospheric precipitable water vapor (PWV) are retrieved based on a physical-iterative algorithm using observations of the microwave radiometer brightness temperatures at 23.8 and 31.4 GHz, with uncertainties ranging from 15 to 30 g m-2 (Marchand et al., 2003). It is important to note that the brightness temperature biases switch signs among different climatological regions because a threshold of 5 ∘C in the cloud base temperature was used in their physical retrievals. Since the retrieved LWP and PWV are based on the MWR measured brightness temperatures at two frequencies, any biases in the brightness temperatures will affect these retrievals. Therefore, we propose an extra step to determine the uncertainties during MARCUS. Based on the temperature profiles, we can identify clouds that are not likely to contain liquid (e.g., pure ice cloud), and then we can estimate the LWP uncertainty based on their corresponding retrieved LWP values. From the probability density function (PDF_analysis), the LWP uncertainty is estimated as 10 g m-2 for MARCUS.

To determine the precipitation status, the AOS and rain gauge measurements were used to determine whether rain is reaching the surface qualitatively, but not quantitatively, in this study. All the measurements were averaged over 5 min, except the radar reflectivity, Doppler velocity, and spectrum width used in Sect. 4.3.

A classification method developed in Xi et al. (2010) was used to categorize different types of clouds using ARM radar–lidar-estimated cloud base (Hbase) and top (Htop) heights and cloud thickness (ΔH). A brief description of the classification of cloud types is as follows (Table 1 and Fig. 6 in Xi et al., 2010). The single-layered, low-level clouds (LOW) are the fraction of time when low clouds with Htop≤3 km occur without clouds above them. Middle clouds (MID) range from 3 to 6 km, without any clouds below and above, while high clouds (HGH) have Hbase>6 km, with no cloud underneath. Other types of clouds are defined by different combinations of the above three types, including middle over low (MOL), high over low (HOL), and high over middle (HOM), and the cloud column through the entire troposphere is defined as HML. The three types, MOL, HOM, and HML, include both contiguous and non-contiguous cloud layers, and their thicknesses may be overestimated when clear layer(s) are present between any two cloud layers.

Furthermore, we used the measurements of interpolated sounding, microwave- radiometer-retrieved LWP, radar reflectivity, Doppler velocity, and spectrum width to classify the cloud phase in each radar range volume of low-level clouds during MARCUS. The detailed classification method will be introduced in Sect. 4.1. We also used ERA-Interim reanalysis data to study the environmental conditions during MARCUS. The lower tropospheric stability (LTS) is calculated from the potential temperature difference between the surface and 700 hPa to assess the boundary layer stabilities when the low-level clouds appeared along the ship tracks. The relative contributions of mixed-phase, liquid, and ice clouds to the single-layered, low-level clouds, as well as their drizzling status, are also analyzed in this study. The latitudinal and longitudinal variations in the single-layered, low-level clouds and their vertical distributions are further explored in this study.

In this study, cloud boundaries are determined by combining cloud radar, ceilometer, and micropulse lidar measurements at a temporal resolution of 5 min. The cloud phase, liquid water droplets, or ice particles are determined in each radar range volume. A flow chart for classifying the phases of single-layered, low-level clouds is shown in Fig. 5. The determination of warm liquid clouds is straightforward using both cloud base (Tbase) and cloud-top (Ttop) temperatures greater than 0 ∘C, and cloud LWPs greater than the threshold (10 g m-2). The determination of supercooled liquid clouds is slightly complicated. When either Tbase or Ttop is below 0∘ C, and cloud LWPs are greater than the threshold, then the radar Doppler spectrum width (WID) and velocity (Vd) are used for the determination of supercooled liquid water clouds. If the majority (10 s of original radar measurements) of WID within a 5 min period are less than 0.4 m s-1 and Vd are equal to or less than 0.0 m s-1 (updrafts) in the volume, then this range volume is defined as supercooled liquid clouds.

Mixed-phase clouds are determined when the median (calculated from 10 s of original radar measurements) of WID is greater than 0.4 m s-1 or Vd is greater than 0.0 m s-1 (downdrafts) due to the existence of large ice particles in the clouds. If cloud LWP is below the threshold, then it is defined as an ice cloud; otherwise, it is defined as a mixed-phase cloud. It is worth mentioning that large ice particles, which grow through vapor deposition or rime processes, dominate the radar reflectivity and are heavier than cloud droplets. Therefore, these large ice particles not only broaden the spectrum width but also have relatively large fall speeds.

To further evaluate our classification method, we compared the classified mixed-phase and ice clouds with the micropulse lidar linear depolarization ratios (LDRs) as an extra measure. The LDR ranges follow the method in Shupe et al. (2005), which are 0.11<LDR<0.15 for mixed-phase clouds and LDR >0.15 for ice clouds, as listed in Table 1. Table 3a shows the quantitative comparison of the cloud-phase identifications between these two classification methods. The numbers represent the counts of each matched 5 min sample, where the diagonal numbers indicate that both methods are identifying the same type of cloud phase. In general, the two methods have 89 % agreement on the phase identification. Second, we performed the phase classification directly from microphysical probes on board the G-1 aircraft during SOCRATES and treated them as ground truth (Mohrmann et al., 2021). Since the in situ cloud microphysical measurements can tell us the phase of the cloud, the measurements allows us to see the percentage variations in cloud phase, by changing integration time of in situ sampling to mimic what the radar may observe the cloud for each range volume. Table 3b shows possible cloud phase partitionings that may be detected by cloud radar. As sampling time increases from 1 to 30 s, more mixed-phase clouds and fewer single-phase clouds can be observed.

Figure 6 shows the determination of mixed-phase and ice clouds through combined measurements of radar reflectivity and spectrum width, lidar LDR and backscatter, and cloud LWP. For the classified ice clouds, cloud LWPs are lower than 10 g m-2 (Fig. 6f), most of the Doppler spectrum widths range from 0.08 to 0.16 m s-1 (Fig. 6b), and the LDR ratios (Fig. 6d) can be greater than ∼0.15, representing a narrow range of ice particle size distribution with higher LDR ratios. For the classified mixed-phase clouds, cloud LWPs are greater than 10 g m-2, and most of the Doppler spectrum widths range from 0.15 to 0.5 m s-1, representing a broad particle size distribution resulting from the mixture of liquid droplets and ice particles. An interesting result occurs where both LDR signals (>0.2) and LWPs are much higher during the drizzling periods (Fig. 6a), indicating a mixed-phase cloud with cloud droplets within the cloud layer and large liquid drizzle drops and ice crystals below cloud base.

Based on the Doppler velocity, the mode values for both mixed-phase and ice clouds occur at ∼0.5 m s-1, where the ice particles are dominant in both types of clouds. The broader particle size distribution with lower LDR ratios for mixed-phase clouds and narrower particle size distribution with higher LDR ratios for ice clouds further corroborate that the classified results from this study are consistent with the traditional micropulse lidar LDR method.

It is important to note that the micropulse lidar signals are usually attenuated and cannot provide a meaningful signal when the liquid cloud layer is thicker than a couple of hundred meters (Sassen, 1991). Arctic mixed-phase clouds are typical with the liquid-dominant layer on the top of the mixed-phase clouds and the ice-dominant layer underneath. The ceilometer-derived cloud base height represents the base of the liquid-dominant layer near the cloud top, while the MPL-derived cloud base height represents the base of the lower ice-dominant layer (Qiu et al., 2015; Shupe, 2007; Shupe et al., 2005). Over the Arctic, the micropulse lidar signals can penetrate through the ice-dominant layer to the liquid-dominant layer. However, the mixed-phase clouds over the Southern Ocean are totally different from those over the Arctic region because they are well mixed (liquid droplets and ice particles) from cloud base to cloud top, which is found in this study. Thus, the micropulse lidar signals can be attenuated in the mixed-phase clouds over the Southern Ocean. Statistical results show that 43 % of micropulse lidar signals were attenuated during MARCUS compared to our classified results.

This classification method is further supported by the on-board cloud radar measurements during the Southern Ocean Clouds, Radiation, Aerosol Transport Experimental Study (SOCRATES; not shown). In that campaign, the reflectivity measurements were usually greater, and the spectrum widths were much wider when the aircraft observed large ice particles compared to the time periods when liquid cloud droplets were observed. Although the wider spectrum widths might be caused by Doppler broadening of the moving aircraft, further analysis shows that the on-board radar sends the signals (assuming the time of transmitted and received signals is short enough comparing to aircraft speed) in the perpendicular to the movement of the aircraft; that is, there is no relative movement between radar signals and clouds. Thus, the on-board radar spectrum width measurements should be not significantly impacted by Doppler broadening (relative movement in the same direction).

In this study, a total of 6934 5 min single-layered, low-level cloud samples were determined using our classification method, including 697 liquid cloud samples, 3777 mixed, 1205 ice, and 1255 other clouds. The category of other clouds represents more than one phase in each column. Note that although other also refers to a mixed-phase cloud, it has different vertical distribution of liquid compared to the mixed cloud. It is also worth mentioning that about 5.5 % of single-layered, low-level cloud phases could be determined when the radar measurements were not available during MARCUS, and those were not accounted to the category of other clouds.

Figure 7a shows the drizzling status for each categorized cloud type, i.e., no rain (green), virga (brown), and rain (navy blue). The definition of drizzling status follows the method in Wu et al. (2015, 2017), where there are radar reflectivity measurements below the ceilometer-/lidar-determined cloud base. The major difference for drizzle in the studies of Wu et al. (2015, 2017) and this study is that drizzle is in the liquid phase at the ARM ENA site but could be both liquid and ice phases in this study.

The percentages shown below the x axis represent the portion of drizzling status in each type of cloud, such as liquid, mixed-phase, ice, and other clouds. Figure 7b also shows the percentages and vertical distributions of classified liquid, mixed-phase, ice, and other clouds for each column in the single-layered, low-level clouds, represented by different colors. After classification, the samples in each category are sorted by their Htop. In detail, Fig. 7 demonstrates that the mixed-phase clouds dominate the single-layered, low-level cloud category with an occurrence frequency of 54.5 %. The other and ice clouds have similar occurrence frequencies of 18.1 % and 17.4 %, respectively, while the liquid clouds have the lowest occurrence frequency of 10.1 %. The liquid-topped mixed-phase clouds (included in other), which frequently occur in the Arctic region (Qiu et al., 2015), are rarely found over the SO. The existence of ice particles in mixed-phase clouds should strongly depend on the distribution of ice nuclei (IN), whereas spatially unevenly distributed IN may result in the other type of cloud.

Based on the results in Fig. 7, we draw the following conclusions. Most of the ice clouds are without icy precipitation, and the percentages with virga and precipitation below the cloud base are 12 % and 15 %, respectively. The percentages of non-drizzling, virga, and drizzling mixed-phase clouds are 50 %, 21 %, and 29 %. The liquid and other clouds have similar percentages; they are 36 %, 25 %, and 39 % for liquid clouds and 35 %, 22 %, and 44 % for other clouds. For liquid and other clouds, the drizzling frequencies are independent of Htop. In contrast, for mixed-phase and ice clouds, the drizzling frequencies strongly depend on Htop, i.e., higher drizzling frequencies occur mostly at higher Htop.

The properties of single-layered, low-level clouds are summarized in Table 4. The liquid clouds have the lowest Hbase and Htop but more available water vapor than other types of clouds. Since the other clouds are a transitional stage among mixed-phase, liquid, and ice clouds, they have the highest Htop, deepest cloud layer, and largest LWP. The ice clouds occur in relatively dry environments and have the highest Hbase at 1.218 km and thinnest cloud layer. The cloud variables for mixed-phase clouds fall between liquid and other. Since LWPs in mixed-phase clouds have a larger standard deviation, which implies that SLW is more common at higher LWPs and ice is more common at lower LWPs.

Figure 8 shows the meridional variation in single-layered, low-level cloud properties during MARCUS. As illustrated in Fig. 8a, the meridional distributions of Hbase, Htop, and ΔH are nearly independent of latitude; however, their corresponding temperatures (Tbase and Ttop) increased by about 8 K from 69 to 43∘ S, although there were slight fluctuations. These results suggest that the cloud and sea surface temperatures have minimal impact on the cloud boundaries over the SO, which is consistent with the findings in McFarquhar et al. (2016). The meridional variation of LWPs follows those of Tbase and Ttop, with an increasing trend from south to north. It is important to point out that a big drop in LWP at ∼50∘ S results from fewer occurrences of low-level clouds there, indicating that the cloud samples at some latitudes are not statistically significant. The atmospheric PWV increased dramatically from ∼5 mm at 69∘ S to ∼18 mm at 43∘ S, presumably due to increased sea surface and atmospheric temperatures.

Figure 9 shows the latitudinal and meridional distributions of categorized liquid, mixed-phase, ice, and other in single-layered, low-level clouds over the SO during MARCUS. Each circle represents the exact location and time along the ship track. Mixed-phase clouds occurred everywhere over the SO during the MARCUS field campaign and became dominant in November, December, and February. Liquid clouds dominated in March, while ice clouds dominated in January. The other clouds are a kind of transitional phase falling in between the mixed-phase and ice/liquid clouds because there are no standalone occurrences in any month during MARCUS.

The vertical distributions of classified liquid, mixed-phase, and ice clouds in LOW category are presented in Figs. 10–12. The focus of this section will be comparisons of cloud macrophysical properties between the north (NSO) and south (SSO) regions of the domain. Figure 10a shows the vertical distributions of liquid clouds, which were capped at ∼1.6 km, mostly in the marine boundary layer. The vertical occurrence frequencies are up to 27 % over NSO, while they were less than 4 % over SSO, i.e., liquid clouds occurred fairly often over the midlatitude region, but very few occurred over the polar region. On the contrary, the occurrence frequencies of mixed-phase clouds between NSO and SSO are opposite to liquid clouds, as illustrated in Fig. 10b, although the differences are not so obvious. Mixed-phase clouds increased with altitude until ∼1.6 km and then decreased monotonically towards 3 km. The highest frequencies were ∼37 % at 0.6 km over SSO and ∼27 % at 1.5 km over NSO. The vertical distributions of ice clouds are similar to those of mixed-phase clouds (Fig. 10c); however, there were no significant differences between NSO and SSO. It is worth mentioning that the vertical distributions of mixed-phased clouds over SO are quite different to those from the DOE ARM North Slope of Alaska (NSA) site, where the low-level mixed-phase clouds are commonly featured with a liquid-topped layer (e.g., Qiu et al., 2015).

To further investigate the vertical distributions of classified liquid, mixed-phase, and ice clouds over NSO and SSO, we plot the normalized vertical distributions (cloud base as 0; cloud top as 1) of radar reflectivity, Doppler velocity, and spectrum width in Figs. 11 and 12, respectively. In this study, the threshold of -50 dBZ was used to determine the cloud boundary over the SO instead of the threshold of -40 dBZ radar reflectivity used at the ARM ENA site (Dong et al., 2014). If we used the threshold of -40 dBZ over the SO, then there would be only 73 % of cloud samples available. If we used the threshold of -50 dBZ, then we would have 90.4 % of cloud samples, which gained an additional 17.4 % on top of the -40 dBZ threshold. About 9.6 % of the radar reflectivities during MARCUS are lower than -50 dBZ for all LOW cloud samples but without ceilometer and MPL lidar signals. Thus, these 9.6 % cloud samples were eliminated in Figs. 11–12.

Figures 11a–c represent the normalized vertical distributions of the radar reflectivity, Doppler velocity, and spectrum width of liquid clouds. Liquid clouds had the lowest reflectivity near the cloud top because of cloud-top entrainment. The reflectivity had a nearly constant median value of ∼-22 dBZ from cloud top height (∼0.8 for normalized height) of the cloud layer to the cloud base. Most of the reflectivities were less than -15 dBZ, which is a threshold to distinguish cloud droplets and drizzle drops in each radar range volume (Wu et al., 2020). Most of the Doppler velocities were greater than 0.0 m s-1, indicating that downwelling motion is dominant in liquid clouds. The profiles of Doppler velocity and spectrum width increased smoothly from the cloud top to base, suggesting that larger cloud droplets and broader size distributions exist near the cloud base, which is attributable to more drizzle drops near the cloud base, as illustrated in Fig. 7.

The vertical distributions of mixed-phase clouds in Fig. 11d–f are similar to those of liquid clouds. The more occurrences of larger reflectivity measurements and larger median values of spectrum width near the cloud base are most likely due to the presence of moderate ice particles and/or drizzle drops. The nearly same median values of reflectivity, Doppler velocity, and spectrum width (but slightly larger standard deviations in each level in mixed-phase clouds) in both liquid and mixed-phase clouds suggest that the ice particle sizes in mixed-phase clouds are comparable to cloud droplets and drizzle drops. The nearly uniform vertical distributions of Doppler velocity and spectrum width indicate well-mixed liquid cloud droplets and ice particles throughout the cloud layer in the mixed-phase clouds over NSO.

Compared to liquid and mixed-phase clouds, ice clouds had much lower reflectivities and a narrower spectrum width, as shown in Fig. 11g–i. Almost all reflectivity measurements were less than -25 dBZ, with a median value of -35 dBZ at the cloud base, resulting from small or moderate ice particles but much lower concentration. A nearly constant Doppler velocity within the cloud layer further supports the discussion of mixed-phase clouds above, i.e., that the ice particle sizes are independent of cloud height and comparable to liquid cloud droplets in the low-level clouds over the SO. Because there are no mechanisms for growing large ice particles in such shallow ice clouds, the accretion process cannot take place. From the statistical results in Fig. 7, these ice particles have relatively little chance to become virga or raindrops and usually dissipate or transition to other types of clouds.

Since there are not enough liquid cloud samples over the polar region, only the mixed-phase and ice clouds results are shown in Fig. 12. Compared to the vertical distributions of ice clouds over NSO, the median values of reflectivity and Doppler spectrum width over SSO were lower and narrower, indicating a lack of large ice particles in the polar region. The small ice particles in the polar region were also reflected in their mixed-phase clouds. Compared to the vertical distributions of the mixed-phased clouds over NSO, the median values of reflectivity and Doppler spectrum width over SSO were dramatically lower (-35 dBZ at SSO vs. -22 dBZ at NSO; 0.25 m s-1 at SSO vs. 0.32 m s-1 at NSO). Figure 12 illustrates that the ice particle sizes over SSO are smaller and their size distributions are narrower than those over NSO, indicative of lack of large ice particles over SSO.