The properties of clouds derived using a suite of remote sensors on board the Australian research vessel (R/V) Investigator during the 5-week Clouds, Aerosols, Precipitation, Radiation, and Atmospheric Composition over the Southern Ocean (CAPRICORN) voyage south of Australia during March and April 2016 are examined and compared to similar measurements collected by CloudSat and CALIPSO (CC) and from data collected at Graciosa Island, Azores (GRW). In addition, we use depolarization lidar data to examine the thermodynamic phase partitioning as a function of temperature and compare those statistics to similar information reported from the CALIPSO lidar in low-Earth orbit. We find that cloud cover during CAPRICORN was 76%, dominated by clouds based in the marine boundary layer. This was lower than comparable measurements collected by CC during these months, although the CC dataset observed significantly more high clouds. In the surface-based data, approximately 2/3 (1/2) of all low-level layers observed had a reflectivity below −20 dBZ in the CAPRICORN data (GRW) with 30% (20%) of the layers observed only by the lidar. The phase partitioning in layers based in the lower 4 km of the atmosphere was similar in the two surface-based datasets, indicating a greater occurrence of the ice phase in subfreezing low clouds than what is reported from analysis of CALIPSO data.
A persistent and long-standing high bias in absorbed solar radiation at the surface of the Southern Ocean (SO) due to low-biased cloud cover is at the root of many challenges in simulating the response of the southern mid- and high latitudes to climate change (Trenberth and Fasullo 2010; Hwang and Frierson 2013; Armour et al. 2016). While the SO region is known for frequent and deep midlatitude cyclones that drive persistently strong surface winds and heavy seas, it is the widespread convective and spatially extended stratiform cloudiness in the marine boundary layer (MBL) that give this region one of the highest cloud-cover fractions of any comparable region on Earth (Mace et al. 2009; Mace and Zhang 2014). It is the extended shallow MBL cloud systems that drive the albedo of the maritime storm tracks in general (Norris and Iacobellis 2005) and the Southern Hemisphere in particular (e.g., Haynes et al. 2011; Bodas-Salcedo et al. 2012), and it is the MBL cloud systems, especially in the cold sectors of cyclones, that are most poorly represented in global models (Bodas-Salcedo et al. 2014, 2016; Franklin et al. 2013), regional forecast models (Protat et al. 2017), and even modern reanalyses (Naud et al. 2014).
Cloud systems that occur in the cold air portion of midlatitude cyclones and that appear from measurements to be primarily composed of supercooled liquid water near their tops (Huang et al. 2015; Huang et al. 2016; Hu et al. 2010) are central to our understanding of the SO climate system (Bodas-Salcedo et al. 2012, 2014, 2016) and the solar radiation biases they cause (Kay et al. 2016; Frey and Kay 2018; Tan et al. 2016). Mace (2010) and Bodas-Salcedo et al. (2016), using A-Train data (Stephens et al. 2008), find that shallow clouds based in the marine boundary layer are a significant if not the dominating influence on the albedo of this region as well as in northern maritime storm tracks. While this cloud genre is predominant over the midlatitude oceans in both hemispheres, the radiation bias in models seems to be focused on situations when these clouds exist in thermodynamic conditions that cause their tops to be supercooled and the associated supercooled tops appear to be far more prevalent over the SO than the Northern Hemisphere (Hu et al. 2010).
Phase partitioning between ice and liquid in the SO low- and midlevel clouds is a characteristic that has emerged as crucial to our ability to accurately simulate the energy budget of this region (Vergara-Temprado et al. 2018). CALIPSO lidar observations of cloud-top phase (Hu et al. 2010) suggest that the tops of these cloud systems contain much more supercooled liquid than what had been parameterized in models and more than what is observed over the Northern Hemisphere high latitudes. Kay et al. (2016) show that the radiation bias can be reduced by forcing models to have the phase partitioning observed by CALIPSO. Tan et al. (2016) find that the climate sensitivity of Earth is coupled to this phase partitioning where models that are forced to have a phase partitioning similar to CALIPSO are ~30% more sensitive (i.e., have 30% more warming for doubled CO2) than models that allow more rapid glaciation of these clouds. This sensitivity is due to a cloud-phase feedback that tends to be more negative in models that allow more rapid glaciation than implied by measurements (McCoy et al. 2014). More recently, however, Kay et al. (2016) have shown that this sensitivity may be overstated when allowing the atmosphere to be coupled to a dynamic ocean.
Ultimately, the feedbacks and their impact on climate sensitivity telescope down the scale continuum to cloud systems and cloud elements that go through their life cycles interacting with the local aerosol, thermodynamic, and turbulent environments. We contend that developing a requisite understanding of the processes that govern the climate sensitivity has, at its foundation, an understanding of the processes that occur on the cloud scale where models must be capable of simulating what is observed in nature. In this and a companion paper (Mace and Protat 2018, hereafter Part II) we describe initial results from a unique dataset collected by ship-based remote sensors that sailed aboard the Australian research vessel (R/V) Investigator during a voyage into the SO during March and April 2016.
Our objectives in this study are twofold. We first seek to describe the cloud occurrence statistics observed from remote sensors aboard Investigator and compare them with other similar datasets. Second, we examine the phase partitioning of ubiquitous lower-tropospheric clouds. In Part II we consider the properties of a specific class of clouds that may be particularly important in explaining the absorbed solar radiation bias in the higher-latitude SO. To accomplish these objectives, we compare the SO measurements to data collected from a remote maritime site in the Northern Hemisphere midlatitudes and we also compare with cloud occurrence data collected by the active remote sensing satellites CloudSat and CALIPSO.
2. The 2016 CAPRICORN voyage
The second voyage of the Clouds, Aerosols, Precipitation, Radiation, and Atmospheric Composition over the Southern Ocean (CAPRICORN) experiment was held using R/V Investigator from 13 March 2016 to 15 April 2016 south of Tasmania [Fig. 1; see Protat et al. (2017) for a description of the first CAPRICORN voyage]. The objectives of CAPRICORN were to (i) characterize the cloud, aerosol, and precipitation properties, boundary layer structure, biological production and cycling of dimethyl sulfide (DMS) in the upper ocean, atmospheric composition, and surface energy budget, as well as their latitudinal variability; (ii) evaluate and improve satellite estimations of these properties; and (iii) evaluate and improve the representation of these properties in the regional and global versions of the Australian Community Climate and Earth-System Simulator (ACCESS) model (Puri et al. 2013; Franklin et al. 2013).
In this and Part II, we seek to address a narrower set of questions associated with cloud occurrence in the SO study region with a particular emphasis on boundary layer clouds. Therefore, we analyze only a few of the data streams collected during CAPRICORN that address our specific objectives. In particular, zenith-pointing cloud measurements were collected on R/V Investigator with the Bistatic Radar System for Atmospheric Studies (BASTA) 95-GHz cloud radar (Delanoë et al. 2016) that has a reported sensitivity of −36 dBZ at 4 km, a vertical resolution of 25 m, and a temporal resolution of 12 s; a cloud-aerosol Leosphere RMAN-511 mini-Raman lidar operating at 355 nm and using a vertical resolution of 15 m and temporal resolution of 35 s (Royer et al. 2014; see Table B1 for lidar specifications); and a Radiometrics 2-channel microwave radiometer (MWR; Liljegren 1994; Liljegren et al. 2001) that collected 31- and 23-GHz microwave brightness temperatures at a temporal resolution of 2 s. These measurements allow for a comprehensive documentation of the macrophysical and microphysical properties of clouds that are relevant to the climate model biases in the region. As we show below and in Part II, supercooled liquid water clouds, which are potentially one of the main causes of climate model shortwave radiation biases in the Southern Ocean, were frequently observed during the voyage using the cloud radar–lidar combination.
Beyond the data streams that we consider here and in Part II, a large suite of additional atmospheric and surface flux measurements were collected during this CAPRICORN voyage. In lieu of a more comprehensive publication summarizing the larger objectives of CAPRICORN, the additional data streams not directly used in this study are listed briefly in appendix A.
3. Data and methods
a. Additional data
To complement the CAPRICORN data, we compare with a surface-based dataset collected on the island of Graciosa (GRW) in the Azores by the Atmospheric Radiation Measurement (ARM) Program known as the Clouds, Aerosols, and Precipitation in the Marine Boundary Layer Experiment (Wood et al. 2015).
Located at 39°N and 28°W, the Azores region is known primarily for being dominated by warm marine stratocumulus clouds associated with the Atlantic subtropical high pressure zone. However, the midlatitude storm track dips equatorward into the Azores region from autumn through spring (Rémillard and Tselioudis 2015) and produces a range of boundary layer and frontal clouds. We use the W-band cloud radar (WACR), the Vaisala ceilometer, the 2-channel microwave radiometer, and the micropulse lidar (MPL) datasets as described in Wood et al. (2015, their Table 2) to compare with similar measurements collected during CAPRICORN. While the GRW data extend over 19 months, we limit our focus to data collected during the cold-season months of December, January, and February. Our motivation for comparing CAPRICORN with GRW is to examine, when present, the phase partitioning and microphysical properties of supercooled liquid- and mixed-phase clouds from different maritime regimes derived from independent sets of ground-based remote sensors. As we discuss below, the GRW data had the most supercooled liquid- and mixed-phase layers during the winter months of December, January, and February. Therefore, we use the December, January, February period from GRW for comparison. While we comment on the broader cloud occurrence statistics observed in these regions, our main motivation is on the supercooled liquid- and mixed-phase layers.
b. Cloud occurrence statistics and observing system considerations
In addition to the surface-based measurements, we analyze data collected with CloudSat and CALIPSO (CC)—the active remote sensors in the A-Train. Specifically, we use the radar–lidar geometrical profile data described in Mace and Zhang (2014). Our primary science objective is to examine the occurrence statistics and phase partitioning of stratiform and shallow convective clouds based in the marine boundary layer. Deeper convective clouds, frontal systems, and cirrus, while important components of the SO cloud climatology, are not specifically examined here. Therefore, we set 4 km as an upper boundary for observed cloud-layer bases in our analysis unless otherwise noted. A base height of 4 km encompasses the boundary layer and allows us to extend our analysis to supercooled liquid layers observed by the remote sensors, although boundary layer tops were always less than 2 km, as determined from radiosondes. A hydrometeor layer is defined to be any consecutive set of active remote sensor range bins in a vertical profile that has measurable (i.e., nonnoise) backscatter detection. For a set of vertically contiguous range bins to be considered a layer, gaps (i.e., hydrometeor-free range bins) must be less than 100 m (240 m for A-Train).
It is worth noting that the observing systems we consider (zenith- and nadir-pointing W-band radar and elastic lidar at the surface and in low-Earth orbit) have significant limitations and differences that influence how we interpret the results. W-band radar, while very sensitive relative to typical weather radars, is not sufficiently sensitive to observe all nonprecipitating cloudy layers, as we show later. Furthermore, the radars used all have different sensitivity thresholds. Lidar, on the other hand, is extraordinarily sensitive to nonprecipitating cloud volumes but attenuates fully after about 3 visible optical depths (note that attenuation of the radars is not an issue for the clouds we consider). This means that from the surface, the lidar is unable to penetrate more than a few tens of meters into a typical cumulus cloud, while from space the lidar is unable to identify cloud base and often attenuates very near cloud top. W-band radars are also very poor at identifying cloud base either from space or the surface because of an inability to distinguish cloud from ubiquitous light precipitation. The upshot of this discussion is that from the surface, we have a very good idea where the lowest cloud base is and whether there is precipitation falling from it. However, from the surface we are unable to confidently identify cloud top. Conversely, from space we are able to identify the highest cloud top and, in many situations, provide some measure of the hydrometeor-layer depth when larger droplets or ice crystals are present. However, never are we confidently able to identify cloud base from space. See Protat et al. (2006, 2014) for additional discussion.
There are significant differences also in the spatial and temporal resolution of the measurements and how the instruments sample the atmosphere. The CC data are reported along the subsatellite track in 240-m vertical range bins that nominally correspond to footprints that are 2 km along track and 1 km across track. The surface-based instruments provide data in ~30-m range bins every few tens of seconds within fields of view that are nominally a few tenths of a degree. To make things even more interesting, the W-band radar on CloudSat is unable to sense confidently below 1 km above the surface because of the transmit pulse interacting with the very reflective sea surface (Marchand et al. 2008). When the CALIPSO lidar beam can penetrate to this blind zone, we are able to observe from space layer occurrences below 1 km. None of the active instruments we consider scan beyond the vertical (with the possible obvious exception that a ship pitches and rolls up to ~10°—typically much less) meaning that the surface-based instruments provide a curtain of measurements in the vertical and temporal direction while CloudSat and CALIPSO provide a spatial curtain along the subsatellite tracks. To make sense of the statistics, therefore, we establish analysis domains about the regions of interest and then collect the CloudSat and CALIPSO data within these domains during the months of interest for years that include 2007–11 during the period when CloudSat operated both during night and day (following 2011, CloudSat operated only in daylight). All of these issues need to be kept in mind when interpreting the results shown below.
The thermodynamic structure of the boundary layer is a critical aspect of this study and Part II. While soundings were launched from Investigator, their frequency was irregular and typically limited to 1 or 2 per day depending on meteorology and other local weather constraints. Since the temporal resolution of the soundings was not sufficient to characterize the thermodynamic properties of the observed cloud layers, we extract thermodynamic information from a combination of the shipborne surface meteorological data and the Modern-Era Retrospective Analysis for Research and Applications (MERRA-2; Molod et al. 2015) extracted along the track of the ship. We compared the MERRA temperature and humidity to the soundings collected during CAPRICORN. We found that MERRA was warm biased compared to the soundings at all levels by ~1.2 K with a standard deviation of 0.7 K, while the humidity was slightly dry biased (~8%) with a standard deviation of 15%. Observed soundings (4 day−1) are used for our analysis of the GRW dataset.
d. Cloud-phase determination
Our understanding of mixed-phase clouds is primarily derived from analyses of higher-latitude ground-based remote sensors where typically ice is present in the mixed-phase region and consists of precipitation-sized ice crystals that are sedimenting from the bases of liquid water clouds (Shupe et al. 2008; De Boer et al. 2009; Morrison et al. 2012). The presence or absence of ice precipitation is a fundamental property of these clouds that is diagnostic of the processes modulating their life cycles. It is the cloud life cycle as represented ultimately through the budget of total water that impacts the radiative balance—a factor directly associated with various climate feedback mechanisms (Kay et al. 2016; Tan et al. 2016). Therefore, we seek to determine the thermodynamic phase of condensed water at and near the bases of clouds that are colder than freezing at their observed layer tops using lidar depolarization ratio (δL). We identify cloud base from the lidar data as a step increase in the attenuated backscatter using the methodology described by Wang and Sassen (2001). The approach we take to calibrating the lidar data from CAPRICORN and GRW and the methodology for establishing the δL thresholds is discussed in appendix B. The critical detail is that we identify a liquid-dominant layer as having δL ≤ 0.02 and an ice-dominant layer as having δL ≥ 0.03. Volumes that have δL between 0.02 and 0.03 are deemed to be ambiguous and are not assigned a phase determination. In all cases with tops below 4 km, ice dominated layers were found to have liquid water in the layer as determined by measureable increases in the downwelling microwave brightness temperatures.
We are careful to consider more than just δL at cloud base in establishing the likely phase properties of supercooled layers. Because the measured δL is a backscatter-weighted integral property of the hydrometeors in a resolution volume, if the scattering at cloud base is dominated by the typically much higher total cross-sectional area of the water droplet population, then the presence of ice could easily be missed. Therefore, if δL at cloud base is less than 0.02, we also look below cloud base in the range bins immediately below the identified base (up to 200 m) and identify the maximum in δL that is associated with a radar reflectivity in excess of −20 dBZe. If the maximum in δL in that subcloud layer exceeds 0.03 (is less than or equal to 0.02) and there is a measureable radar reflectivity, then we deem that supercooled layer to be producing ice-phase (liquid phase) precipitation.
a. Cloud occurrence and case studies
The 2016 voyage can be divided into two periods ( appendix A) with the first occurring in the first two weeks from 15 to 27 March when the ship serviced oceanic buoys near 47°S and 142°E and the second after 28 March when the ship spent time sampling oceanic eddies along the Sub-Antarctic Front south of 50°S and broadly in the 148°E region. The cloud-cover frequencies along with other daily averaged meteorological quantities are shown in Fig. 2, while Fig. 3 shows a time series of radar reflectivity observed by the W-band radar on Investigator with lidar cloud base overlaid. During most of the voyage, the air temperature was colder than the SST except for the period when Investigator sampled a cold mesoscale marine eddy during the first few days of April depicted by the star pattern in Fig. 1 near 50°S. Cloud cover varied with meteorological conditions as increasingly deeper and more intense midlatitude cyclones passed over the ship as Investigator worked southward and deeper into the seasonally intensifying westerlies. The qualitative change in the weather was noticeable and is evident in Figs. 2 and 3 as storms became deeper, stronger, and more frequent with time. Figure 2 shows a decrease in surface pressure, temperature, SST, and an increase in wind speed and an increase in frequency of precipitation. Daily averaged surface wind speeds in excess of 15 m s−1 (~30 kt) during days 100–103 were notable.
The first period when the ship was in the vicinity of 47°S was characterized by frontal passages on 17, 22, and 24 March. The sensible weather was largely quiescent otherwise with mostly broken clouds in the MBL that were predominantly open-cell stratocumulus with shallow scattered cumulus in regions of cold advection. As depicted in Fig. 2, the several-day period following the frontal passage on 24 March produced a several day period of cold air advection and open-cellular cumulus persisted. In Fig. 4, we present a 6-h interval on 26 March that is typical. Note that we discuss this case in more detail in Part II. The period was marked by short intervals when the ship passed under shallow convective showers that were observed by the cloud radar. The Micro Rain Radar (MRR), with much less sensitivity than the cloud radar (~+5 dB), observed several of these showers during this period. Peak rain rates measured by the disdrometer aboard Investigator were near 1 mm h−1 for the showers at 0145, 0350, and 0445 UTC and 1.5 mm h−1 for the shower at 0045 UTC. The lidar illustrates much of what the radars missed. In particular, a layer of stratocumulus near the tops of the convection was persistent during most of this 6-h period with only occasional breaks when the lidar penetrated beyond the MBL top near 1.8 km.
An interesting aspect of this case was that the phase of the precipitation in the shallow convective showers is reasonably characterized by the radars and lidar. With the freezing level near 1 km as observed by radiosonde (Part II), we find evidence for a melting layer near that height depicted by a jump in dBZe in the W-band and in the MRR reflectivity factors due to the increase in refractive index as the phase of the scatterers changed from ice to liquid. This is evident in the showers at 0250 and 0430 UTC. The Doppler velocity (not shown) depicts a step increase at the melting level to values typical of precipitation-sized liquid hydrometeors from values typical of ice-phase hydrometeors. The lidar depolarization ratios near cloud base of the precipitation events in Fig. 4 are in excess of 0.1 in each of these showers indicating that ice is present. This was a typical characteristic of the shallow convection throughout the voyage with the convective showers showing signs of glaciation when the tops extended above the freezing level similar to the findings of Huang et al. (2016) who report on recent aircraft in situ measurements collected southwest of Tasmania. Lidar depolarization, however, shows that the supercooled stratocumulus near the top of the MBL remained mostly liquid, although the lidar did not reliably penetrate the heavier liquid showers to cloud base.
The second period of the deployment took place mostly south of 50°S (Fig. 1). Relative to the earlier period, the meteorology in this region was more active in part because of the advancing season, the higher latitude, a different large-scale pattern that featured a more migratory and low-amplitude pattern, a colder sea surface, etc. The cloud field in the cold sectors of cyclones tended to evolve rapidly from shallow convection to extended areas of stratocumulus that existed near the top of the MBL. Overall, the shallow convection was noticeably deeper immediately behind cold fronts (5, 7, 9, and 10 April) with several episodes of graupel showers at the surface in the coldest air masses on 6 April at 2100 and 2130 UTC and from 9 to 10 April intermittently from 2100 to 0300 UTC.
We present a typical case from data collected on 6 April in Fig. 5 (see also Part II). The days around 6 April were characterized by a rapid succession of fronts and troughs that brought quite variable conditions. The frontal passages were marked by deeper-layer clouds and rain followed by open-cell cumulus in the cold air advection that often transitioned rapidly to stratocumulus that tended to be found near the tops of deep boundary layers. As in the previous example, the cumulus often demonstrated evidence for ice-phase precipitation with obvious melting layers and occasional graupel at the surface. The stratocumulus, while found at subfreezing temperatures, were primarily liquid phase with some drizzle and pockets of ice-phase precipitation as can be seen in Fig. 5 early on 6 April. We see from lidar depolarization that there were brief periods of ice-phase precipitation present interspersed in what appears to be liquid-phase drizzle. In particular, the higher reflectivity period early in the case presented more extensive periods of ice-phase precipitation.
b. Comparison to CloudSat–CALIPSO and GRW
Table 1 lists cloud occurrence statistics for the CAPRICORN dataset, GRW, and the CC data. Overall, the data collected along the CAPRICORN track by the CAPRICORN remote sensors show that 76% of all profiles observed by the radar and lidar on the ship contained measureable hydrometeors with a similar occurrence frequency at GRW. The major difference evident from these statistics is that the GRW data have a higher fraction because of layers with tops above 4 km. By differencing the first two columns, layer tops in excess of 4 km are found 23% of the time at GRW while higher layers are found only 11% of the time in the CAPRICORN data. By comparison, the CC data aggregated over the 2007–11 period show a higher coverage fraction in the CAPRICORN and GRW regions by 11% and 5%, respectively. Of the layers observed, the GRW region had more layers with tops in excess of 4 km (16%) than found in the CAPRICORN domain (4%). The fraction of columns that have hydrometeors observed only by the lidars are higher in the CAPRICORN region compared to GRW although the ship-based lidar has a significantly higher fractional occurrence of lidar only layers than CC in the SO region.
It is significant that between 20% and 30% of all layers in these regions are observed only by the lidar and are below the sensitivity or reach of the cloud radars. In Fig. 6 we show the cumulative frequency distribution of layer-averaged dBZ for the two surface-based datasets. Figure 6 allows us to easily determine what fraction of the layers is below an arbitrary reflectivity threshold. For instance, the radar on CloudSat had a reflectivity threshold near −30 dB suggesting that it would miss approximately 30% of layers in both analysis regions. If we assume that −20 dBZe indicates the presence of light precipitation, then 60% (53%) of the low-based layers in the CAPRICORN (GRW) data are nonprecipitating. These results highlight the utility of having both millimeter radar and lidar in a collocated observing system in order to fully characterize the cloud climatology of a region (Protat et al. 2006; Berry and Mace 2014; Mace and Berry 2017).
We gain a more complete understanding of the broad statistics in Table 1 by examining a set of conditional probability plots that show how the radar reflectivity is distributed vertically between the locations and among the datasets—commonly known as contour frequency by altitude diagrams (CFADs; Fig. 7 and a cumulative frequency version of a CFAD in Fig. 8). Of relevance to the present study are the similarities and differences in the MBL cloud occurrence statistics. Considering first the CC data, the statistics from both regions show maxima in what would be considered nonprecipitating clouds with dBZ less than −20, although this peak is stronger in the CAPRICORN data. The MBL clouds in the CC CFAD from the GRW region demonstrate a more bimodal distribution with a stronger minimum in the reflectivity between −10 and 0 dBZ than what is shown in the CC CFAD from the CAPRICORN regions. This difference suggests that the MBL clouds in the Azores region are either nonprecipitating or drizzling rather strongly. We speculate that this difference may imply different rates of precipitation process production with the MBL clouds in the Azores region creating precipitation rapidly once the process starts, whereas clouds in the CAPRICORN region create precipitation more gradually. While this idea requires additional study, it is an intriguing possibility that may explain much about the differences in occurrence frequency of MBL clouds noted in Table 1.
The CAPRICORN surface data are missing a large fraction of the high clouds and this can be explained by the sensitivity of the radar that is evident in the CFAD plot. The lack of sensitivity of the BASTA radar was due to the radome that was required to protect the radar from sea spray and precipitation inducing at least a 2 dB loss when dry (and more when wet, which was often). The GRW dataset, while also demonstrating reduced sensitivity in the upper troposphere, does better at capturing the high-level clouds observed by CC. Comparing the BASTA CFAD to the CC CFAD we find that the statistics of the low-level clouds are similar. Both datasets show the broad distribution in dBZe between the lowest ranges sampled and 2 km, and both datasets show a minimum in occurrence between −5 and +5 dBZe. The greater sensitivity and resolution of the BASTA radar allows us to assess the degree to which CloudSat accurately characterizes the occurrence of nonprecipitating boundary layer clouds in the analysis domain. The BASTA radar data show that the clouds are distributed in range between 750 m to 2 km and that the dBZe values tend to be in excess of −30. While CloudSat clearly misses some fraction of the nonprecipitating clouds in this region, CloudSat tends to characterize their occurrence reasonably well.
However, the distribution of low-level clouds in the GRW surface-based data is very different from the CC CFAD for the Azores region. The GRW surface-based data seem to be largely missing the nonprecipitating boundary layer clouds that are predominant in CC CFAD. The W-band radar was clearly sensitive enough to capture those clouds had they been present over the radar. We suspect that the cloud occurrence may have been modified form what occurs over the open water by the effect of the local topography.
CFADs are useful but limited in that they do not provide information regarding hydrometeor layers in the atmosphere. Figure 9 provides a conditional frequency diagram in terms of layer base height and layer physical thickness as measured by the combined radar and lidar remote sensors. Comparing first the CC layer frequency diagrams, we find slight but interesting differences in the two regions that are consistent with the differences noted in the CFADS (Figs. 7 and 8). Layers based above 6 km are important components of the cloud climatologies. Cirrus are more prevalent in the GRW data (Fig. 8) and the high-based layers tend to be geometrically thicker at GRW. Otherwise, the marginal distributions of high-based layers are similar. Low-level clouds in the CC data also show subtle differences between the regions with an overall higher relative frequency of low-based clouds in the CAPRICORN region. The CAPRICORN region CC data also show a higher frequency of geometrically thinner and lower-based layers than in the Azores region. The distribution of low-based deep layers in the CC data that mark frontal systems seems to be similar in the two regions.
Comparing the layer occurrence statistics of the CC and BASTA data from the CAPRICORN region, a major difference that is relevant to our present focus on low-level clouds is the higher frequency of occurrence of geometrically thin layers based between 1 and 2 km in the BASTA data. This peak is missing in the CC CAPRICORN region data, although it is qualitatively consistent with the ubiquitous presence of geometrically thin layers near the top of the MBL noted in our case study discussion and in Part II. The absence of this peak in the CC data is consistent with the findings of Protat et al. (2014) and is due to the vertical resolution of the CC data and the geometrically thin layers that make up much of the MBL cloud cover. We also note that multilayer stratiform layers were rare in the surface-based CAPRICORN data. There was a 3-h period on 16 March when a true multilayer stratiform layer was observed below 4 km and both layers were below the sensitivity of the radar. Cumulus under stratocumulus occurred more often, but generally the cumulus or the stratocumulus or both were below the sensitivity of the radar.
c. Phase partitioning
We refer to supercooled liquid clouds as those clouds with a base below freezing and with no evidence of ice either at layer base or in any subcloud precipitation (i.e., liquid–liquid in Table 1). Mixed phase is defined as those layers that are supercooled and liquid at their base but show evidence for ice-phase precipitation below their base. Ice layers are those that appear to be ice both at base and in any subcloud precipitation. Note, however, that all layers with bases below 4 km that seem to be ice dominant in the lidar depolarization ratio at their bases also show evidence of liquid water being present in the layer based on analysis of the MWR brightness temperatures. We also limit our attention to layers with bases below 4 km above the surface so as to exclude layers where homogeneous freezing processes might be occurring. While supercooled liquid clouds occur less often in an absolute sense at GRW than at CAPRICORN, when present the fraction that is supercooled liquid seems to have a similar frequency of occurrence of just more than about one-half of the time. Both datasets also have unambiguous ice phase (the sum of mixed phase and ice dominant) present about a third of the time with the remainder of cases being in the ambiguous depolarization range between ice and liquid. While we hesitate to draw broad conclusions from the similarities in the phase partitioning at cloud base, it is an intriguing result since the Southern Ocean generally is known to have a significantly higher occurrence of supercooled liquid as we have discussed. It would seem, at least as implied by these datasets, that the difference is not in the phase partitioning at cloud base when clouds are present but in the occurrence frequencies of the clouds themselves. The difference that we do find is that at GRW, it is much more likely to find ice precipitation below what appears from lidar depolarization statistics to be liquid clouds. This is rare in the CAPRICORN dataset. When ice is observed in the precipitation below CAPRICORN cold clouds, ice is also diagnosed at cloud base.
The phase partitioning findings derived from CALIPSO data (Hu et al. 2010) have been influential in motivating the global modeling community to examine assumptions regarding the freezing process of cold clouds that are warmer than the homogenous freezing limit. Of course, knowledge of phase at cloud top does not preclude the possibility that ice-phase precipitation processes are occurring just a few optical depths below cloud top beyond the reach of a lidar viewing from above. The statistics in Table 1 suggest that ice-phase processes are indeed occurring within these supercooled layers a third of the time.
We address this issue in more detail by expanding the cloud-base phase statistics in Table 1 as a function of cloud-top temperature in Fig. 10. Cloud top is determined from either the cloud radar, when the layer is observed by the radar, or the top of the lidar cloud layer (even if attenuated) if the radar does not observe the layer. This is clearly an imperfect method but it is somewhat more comparable to the CALIPSO results than just using the cloud-base temperature. Note that determining the phase deeper within the layer from surface-based lidar is not possible because of multiple scattering and attenuation. Not reaching the actual tops of layers explains why the GRW and CAPRICORN curves in Fig. 10 do not approach 1 at the freezing level. Also plotted in Fig. 10 are the results of Hu et al. (2010) taken from their Fig. 7 at 45°S (their results at 45°N are very similar). As might be inferred from Table 1, the CAPRICORN and GRW datasets show strong similarities with each other, although the GRW clouds seem to have slightly more tendency for supercooled liquid. Overall, we find an increasing likelihood of liquid phase with increasing cloud-top temperature. The variable tendencies at the coldest temperatures are due to small sample sizes. Comparing the cloud-top phase results from CALIPSO with the cloud-base phase results that we derive from below, however, suggest that ice-phase processes are not entirely absent in these supercooled clouds as might be inferred from CALIPSO cloud-top results. Assuming that it is reasonable to compare the surface-based data to the zonal averages from CALIPSO, we can state that, on average, between 20% and 40% of the cold layers that seem to be liquid at their tops do have ice-phase processes occurring within them.
To further expand on the phase partitioning, we examine how δL is distributed as a function of cloud-base temperature and W-band reflectivity in Fig. 11 where only results from CAPRICORN are shown because the GRW results are not qualitatively different. Recall that the thresholds we established earlier were δL ≤ 0.02 and δL ≥ 0.03 for resolution volumes to be diagnosed as unambiguous liquid and ice phase, respectively (Fig. B2). Note that the frequency distributions in Fig. 11a are normalized across each row so that it can be clearly seen how dBZ and δL are distributed as functions δL and temperature, respectively. In terms of cloud-base temperature, we find a tendency for an increasing frequency of ice-phase volumes with decreasing cloud-base temperature. Note that because we are examining cloud-base temperature in Fig. 11b, ice precipitation is observed to temperatures warmer than freezing although this is rare. Figure 11a documents an intriguing relationship between δL and cold-layer dBZ. While there is a degree of ambiguity for lower values of dBZ, we find that a W-band dBZe in excess of −10 almost guarantees that the volume will have a value of δL indicative of ice-phase hydrometeors. Similar relationships between δL, dBZ, and temperature were recently shown in Zhang et al. (2017) using data collected at the Barrow, Alaska, ARM site. Such a relationship between δL and dBZe would allow for dBZe to be used a proxy for the occurrence of the ice phase in optically thick cold MBL layers.
5. Summary and conclusions
Observations of cloud and aerosol properties with ship-based suites of sophisticated remote and in situ sensors over the Southern Ocean have been limited until recently to measurements collected by satellites. While the space-based datasets have proven immensely valuable, they are inherently limited in numerous ways, leaving many unanswerable questions that can only be addressed by collecting measurements on location and looking at clouds from below. The incidence of surface-based and in situ sensing over the remote oceans has been increasing. The Southern Ocean is beginning to see more attention from airborne, island-based, and ship-based measurement campaigns. In particular, the U.S. National Science Foundation in coordination with the Australian Marine National Facility and Bureau of Meteorology conducted the Southern Ocean Clouds, Radiation and Aerosol Transport Experimental Study (SOCRATES) in January–February 2018 that featured flights by the NCAR Gulfstream V (GV) combined with an Antarctic voyage by the R/V Investigator. Additionally, the ARM Program, the Australian Antarctic Division (AAD) and the Bureau of Meteorology are currently collecting measurements on Macquarie Island [Macquarie Island Clouds and Radiation Experiment (MICRE)] and are collecting ship-based data during the Antarctic resupply voyages of the R/V Aurora Australis starting in late 2017.
In this and a companion paper (Part II), we present an initial analysis of the first of these comprehensive datasets that was collected by a suite of modern remote and in situ meteorological, aerosol, and cloud and precipitation microphysical instrumentation that deployed for 5 weeks on the Australian R/V Investigator during the Austral autumn of 2016 (Fig. 1, appendix A). For comparison, we also analyze winter data from a deployment by the ARM mobile facility to Graciosa Island, the Azores (GRW; Wood et al. 2015).
The frequency of occurrence of vertical columns containing hydrometeors over the SO during CAPRICORN was 76%. This occurrence frequency is lower than what the CloudSat and CALIPSO (CC) combined datasets record for the March–April period during 2007–11 seasons where an occurrence of 87% (Table 1) is observed. The frequency of occurrence for the surface-based data at GRW was nearly identical to that found from the ship-based data over the SO (76%), although a greater fraction of high-based (above 4 km) layers were observed at GRW for a lower relative occurrence of boundary layer clouds.
The cloud occurrence frequencies of both regions are dominated by clouds in the lower 2 km of the atmosphere (Figs. 3, 7–9) although the CC data demonstrate that high-based layers are significant components of the cloud climatologies of these regions—more so for the Azores. A clear difference between the CAPRICORN and Azores CC datasets was the distribution of radar reflectivity in the lowest 2 km. The SO data (Figs. 7a,c) suggest that the radar reflectivity is much more uniformly distributed in dBZ space than is found in the CC data over the Azores region. In Fig. 7b we find a distinct bimodal frequency distribution of low-level dBZ indicating a population of nonprecipitating clouds and clouds that are precipitating. This result is intriguing in that it implies that low-level clouds that occur over the Azores region transition rapidly from nonprecipitating to precipitating, while over the SO the transition seems to be more gradual as suggested by the more uniform distribution of dBZe. Because it would impact cloud lifetime, such a precipitation process rate difference would contribute to the higher occurrence frequencies of low clouds over the SO. Clearly this hypothesis requires more detailed examination but the differences in the CFADS are unambiguous.
A large fraction of the cloud layers (31% in the SO and 20% at GRW) were observed only by the lidars because the hydrometeor layers were below the sensitivity of the ground-based radar systems, which were on the order of −35 dBZ in the lowest few km of the atmosphere. Setting −20 dBZ as a conservative layer-mean threshold for the occurrence of drizzle, the data suggest that 60% (53%) of the layers in the CAPRICORN (GRW) datasets are composed on nonprecipitating clouds (Fig. 6). This is a significant finding since it suggests that the cloud cover of these regions is dominated by layers that are quite tenuous. The CAPRICORN dataset, especially, shows that geometrically thin layers between 1 and 2 km are a major component of the low-level cloud population. The extent to which these geometrically thin layers influence the radiation budget of these regions and the degree to which their presence or absence in simulations influences the well-documented shortwave radiation bias is an interesting question that requires further research with more comprehensive datasets and model simulations.
The partitioning between ice and liquid in subfreezing clouds that are warmer than the homogeneous freezing point has emerged as an important issue because the occurrence of supercooled liquid in climate model simulations are significantly different from what is observed by CALIPSO (Bodas-Salcedo et al. 2012, 2014, 2016; Kay et al. 2016; Frey and Kay 2018; Tan et al. 2016). When present, the partitioning between supercooled liquid and ice is similar between CAPRICORN and GRW and both sites display a trend with temperature that is similar to CALIPSO, but the ground-based data show that ice-phase processes occur 20%–40% more often than implied by the CALIPSO data. While we do not question the validity of the CALIPSO results, we note that airborne and spaceborne lidars are unable to probe more than a few optical depths into the tops of the supercooled clouds, and they likely miss the production of ice in some fraction of vertical columns.
Because lidar is limited in diagnosing processes in optically thick clouds, finding other proxies for thermodynamic phase is needed. The relationship between δL and dBZ shown in Fig. 11a and in Zhang et al. (2017) may point a way forward in this regard. For layers colder than freezing, the occurrence of radar reflectivity greater than −10 dBZ seems to occur mostly in volumes where δL indicates an unambiguous presence of the ice phase. Therefore, it may be possible to use dBZe in subfreezing volumes beyond the reach of CALIPSO as a proxy for the presence of the ice phase. Layers that then show unambiguous δL indicating liquid at cloud top can be diagnosed with reasonable certainty as being mixed phase if the dBZ at levels beyond the first few optical depths and in subfreezing volumes exceeds −10. Of course such an indirect proxy would be subject to uncertainty, require more extensive in situ validation, and represent a lower limit for the occurrence of the ice phase, but the approach could provide information on likely phase where there is presently none.
We acknowledge the able assistance, hard work, dedication, and generous spirits of our colleagues aboard the R/V Investigator during an arduous five weeks in the Southern Ocean. The authors thank the CSIRO Marine National Facility (MNF) for its grant of sea time on Investigator and associated personnel, scientific equipment, and data management. All cloud data are currently available upon request to Alain Protat (firstname.lastname@example.org) and will be publicly available by June 2018 in accordance with MNF policy from the CSIRO Data Access Portal (https://data.csiro.au/dap/). Sally Benson assisted in generating many of the graphics used in this paper. Support for this work (GM) was provided by NASA Grant NNX13AQ34G. Data were used from the Atmospheric Radiation Measurement (ARM) Program provided by the ARM archive.
Additional CAPRICORN Measurements
Figure A1 provides additional information on the measurements recorded during CAPRICORN during the March and April 2016 voyage. In addition, bulk and turbulent fluxes of momentum, sensible and latent heat, and shortwave and longwave downwelling radiation were measured using the U.S. National Oceanic and Atmospheric Administration (NOAA) surface energy flux package (e.g., Fairall et al. 2008). Downwelling radiation was also measured with two additional radiometers on the port and starboard sides. Local rainfall measurements were collected using a Metek 24-GHz MRR (Klugmann et al. 1996) that employed 50-m range bins from the surface to 1.5 km and 30-s averaging with a minimum detectable signal of +5 dBZe. In addition, drop-size distribution measurements were collected on the main front mast using an Ocean Rain and Ice-Phase Precipitation Measurement Network (OceanRAIN) disdrometer (Klepp 2015) specifically developed for ship deployments.
A suite of instruments were deployed to measure in situ aerosol properties and atmospheric composition of the boundary layer: aerosol microphysical properties including total particle number and aerosol size distribution; aerosol chemical composition, including real-time organic and inorganic species; aerosol optical properties including absorption, scattering, and aerosol optical depth; cloud condensation nuclei (CCN); ice nucleating particle (INP) concentrations; reactive gases including volatile organic compound (VOC) concentrations; real-time INP concentration measurements via continuous flow diffusion chamber (CFDC); filter collections for offline INP concentration measurements via ice spectrometry, with biological INP composition assessment; and bioaerosol measurements [wideband integrated bioaerosol sensor (WIBS) instrument]. Radiosondes were also launched approximately once a day to measure the thermodynamic state of the atmosphere.
Lidar Data Processing
The primary difference between the GRW and CAPRICORN lidar datasets is the different wavelengths of the instruments, with the GRW MPL system operating at 532 nm and the CAPRICORN R-MAN 510 lidar operating at 355 nm (Royer et al. 2014; see Table B1 for technical specifications). The lidar depolarization ratio (δL) is a fundamental measurement that allows us to estimate the phase of hydrometeors at and near cloud base. For δL the total transmitted power is not a concern because of normalization; however, accurate δL requires that both the copolar and cross-polar receiver channels are calibrated. Furthermore, because we are using the attenuated backscatter (βL) in a retrieval algorithm of thin liquid-phase clouds Part II, the absolute calibration of the copolar power return is important. In general, we follow the approach described by Reagan et al. (2002) and compare the uncalibrated power returns against theoretical molecular attenuated backscatter using the algorithms described by Bucholtz (1995), Frohlich and Shaw (1980), and Sneep and Ubachs (2005). While it is most desirable to evaluate a lidar against molecular returns in the lower and middle stratosphere where nonmolecular scattering can be assumed negligible, neither lidar used in this study is sufficiently sensitive to observe stratospheric molecular signals reliably. Therefore, we take advantage of the fact that the marine free troposphere over the remote oceans during the periods that we consider tends to be largely devoid of aerosol signals, and we examine the returns from the 3–6-km vertical interval. To ensure that our assumption is valid, we evaluated the time series of lidar returns and excluded periods when evident structure in the attenuated backscatter was observed. Furthermore, the MPL, at 532 nm, suffers from significant noise from sunlight. Therefore, only the nighttime periods were used. The Raman lidar from CAPRICORN, working at a wavelength of 355 nm, was not significantly contaminated by ambient sunlight.
In practice, we adjusted both cross- and copolar channels of both lidars to agree with Rayleigh theory accounting for the difference in wavelength during periods when no hydrometeors were present below 6-km altitude using the cross-polarization factors d as appropriate for clear air as reported by Bucholtz (1995) and Bates (1984). The procedure we adopt is as follows. Beginning at 3 km to ensure that we are outside the marine boundary layer, we step up the profile to 6 km examining the correction to Rayleigh βL in groups of twenty 15-m range bins choosing the group of 20 range bins that imply a correction with the least scaled variance in the correction factor for that profile. We neglect the profile if the minimum-scaled correction variance is greater than 10%. An example of a representative lidar profile from CAPRICORN is shown in Fig. B1.
We find that a single set of calibrations C (10.2 ± 3.2 and 3.55 ± 2.5 for co- and cross-polar channels, respectively) are valid for the GRW MPL data with the backscatter coefficient as
where, as reported in the grwmplpolM1.b1 data files, S and B are the signal return and background signal in counts per microsecond, Pw is the pulse width in hertz, R is the range in kilometers, E is the monitored energy output in counts per microsecond, h is the Planck constant and η = c/λ or the frequency (and c is the speed of light and λ the wavelength or 532 nm) and 1012 is a conversion factor so that βG is given in m−1, and A is the receiver or telescope area assumed to be 10 cm.
For the CAPRICORN lidar data, a similar procedure is followed with the obvious difference in d. However, for the CAPRICORN data we find that C increased monotonically during the course of the Voyage. For instance, copolar C increased from 15 to 30 to 72 and cross-polar C increased from 1.25 to 3.25 to 14 from 14 to 30 March to 10 April 2016. The reason for this temporal evolution in calibration constant is not known.
In developing δL thresholds for estimating thermodynamic phases near cloud base, we assume that multiple scattering can be neglected in precipitation and in the first range bin above cloud base. We then examine the distributions of δL for layers that are certainly liquid and layers that are certainly ice phase. For liquid-certain layers we require base temperature > 275 K and radar dBZ near base less than −20 ensuring that we only consider nonprecipitating warm layers. Predominant ice layers are potentially more ambiguous since data with bases colder than the homogeneous freezing limit are rare in both datasets and subject to greater uncertainty because of the limited sensitivity of the lidars. Therefore, we examine layers that are colder than 253 K (also quite rare) but we also require the radar dBZ near cloud base to be in excess of −10 dBZe. As shown by Zhang et al. (2017), the precipitation from such layers are highly likely to be ice phase. Figure B2 shows the frequency distributions of liquid-certain and ice-certain layers for the two datasets. We find that no ice layers are found to have δL less than 0.03 in either dataset. While a longer tail to δL in excess of 0.03 (likely due to multiple scattering at cloud base) is found for the liquid layers, only a small fraction (<5%) of the liquid layers are found to have δL > 0.03.
This article has a companion article which can be found at http://journals.ametsoc.org/doi/abs/10.1175/JAMC-D-17-0195.1