Aircraft observations of ﬁve pockets of open cells sampled during VOCALS REx

Five pockets of open cells (POCs) are studied using aircraft ﬂights from the VOCALS Regional Experiment, conducted in October and November 2008 over the southeast Paciﬁc Ocean. Satellite imagery from the geostationary satellite GOES-10 is used to distinguish POC areas and measurements from the aircraft ﬂights are used to com- 5 pare cloud, aerosol, and boundary layer conditions inside and outside of POCs and conditions found across individual POC cases. POCs are observed in boundary layers with a wide range of inversion heights (1250 to 1600 m) and surface wind speeds (5 to 11 ms − 1 ) and show no remarkable di ﬀ erence from the observed surface and free tropospheric conditions during the two months of 10 the ﬁeld campaign. In all cases, compared to the surrounding overcast region the POC boundary layer is more decoupled, supporting both thin stratiform and deeper cumulus clouds. Although cloud-base precipitation rates are higher in the POC than the overcast region in each case, a threshold precipitation rate that di ﬀ erentiates POC precipitation from that in overcast precipitation does not exist. Mean cloud-base precipitation rates 15 in POCs can range from 1.7 to 5.8 mmd − 1 across di ﬀ erent POC cases. The occurrence of heavy drizzle ( > 0 dBZ) lower in the boundary layer better di ﬀ erentiates POC precipitation from precipitation in the surrounding overcast regions, likely leading to the more active cold pool formation in POCs. Cloud droplet number concentration is at least a factor of eight smaller in the POC clouds,


Introduction
During most of the year, a large stratocumulus cloud deck covers the southeast Pacific Ocean (SEP), extending westward from the South American continent. Due to a strong 10 albedo and weak greenhouse effect, the stratocumulus exerts a strong cooling effect on the ocean and the atmospheric boundary layer (de Szoeke et al., 2012). The extensive stratocumulus cloud deck is occasionally interrupted by cloud breaks in the form of pockets of open cells, where the lower cloud cover exposes the darker sea surface, effectively lowering the albedo of the area where these features form.
Pockets of open cells (POCs) have previously been defined as regions of opencell convection completely or largely embedded within a larger region of closed-cell convection (Stevens et al., 2005;Wood et al., 2008). In regions of open-cell convection, patches of descending clear air are ringed by ascending cloudy air, giving it a honeycomb-like pattern when viewed from above. On the other hand, closed-cell con- 20 vection are characterized by patches of cloudy ascending air that are ringed by clearer descending air, leading to higher albedos (Atkinson and Zhang, 1996;Wood and Hartmann, 2006). Subsequent studies have found, however, that POCs also support thin and extensive stratiform clouds in addition to the honeycomb-like patterns or isolated clusters of deeper cumulus clouds (Wood et al., 2011a) and that horizontal gradients in cloud microphysical properties exist at the transition from closed to open cellular convection, with often dramatic decreases in cloud droplet number concentrations in-8289 Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | number concentration to initiate POC formation (e.g., Wang and Feingold, 2009a, b;Berner et al., 2011). The simulations point to the importance of cold pools in transforming closed-cell circulation to open-cell circulation (Wang et al., 2010), and indicate horizontal coupling of inversion heights that allows for the coexistence of closed-cell and open-cell clouds within a region of hundreds of kilometers (Berner et al., 2011). Kazil 5 et al. (2011) implemented a more sophisticated model of the sulfur cycle and aerosol processing within an LES and was able to reproduce the observed structure of the aerosol-depleted "ultraclean layer" (as detailed in observations by Wood et al., 2011a) and its potential for supporting spontaneous nucleation of ultrafine aerosol. More recently, Berner et al. (2013) showed that a large-domain cloud-resolving simulation with 10 interactive aerosols can develop long-lived, mutually-supporting regions of open and closed cells existing side by side, suggesting that under some large-scale meteorological and aerosol conditions, the POC-overcast system is a self-maintaining form of mesoscale cloud organization. POC modeling studies have been initialized based on one DYCOMS-II RF02 case (Wang and Feingold, 2009a, b) and one VOCALS case,

15
RF06 (Wang et al., 2010;Berner et al., 2011;Kazil et al., 2011). The boundary layer heights of the two cases range from 800 m (Van Zanten and Stevens, 2005) to 1600 m (Wood et al., 2011a), suggesting diverse environments in which POCs exist. More observational analyses of POCs are therefore necessary to identify the boundary layer properties that are common across different POC cases and those properties that are 20 more variable. Early observations of POCs were fortuitous because their formation cannot be easily forecast and because they form over remote oceans. The VAMOS Ocean Cloud Atmosphere Land Study Regional Experiment (VOCALS REx), conducted over the southeast Pacific Ocean during October and November of 2008, was the first concerted field 25 campaign for which POCs developing in a larger region were systematically targeted for observational sampling (Wood et al., 2011b;Mechoso et al., 2014). Wood et al. (2011a) provides a detailed analysis of one VOCALS case (RF06) using data from two aircraft. The current study investigates the commonalities and variations across five dif-8291

Data
During VOCALS REx, six research flights flown by the NSF/NCAR C-130 sampled the atmospheric conditions across transitions from closed-to-open cellular clouds accompanied by clear corresponding changes in cloud microphysics (Wood et al., 2011b). Five of those six research flights are used in this analysis, because the diurnal break-15 up of the closed-cell stratocumulus made it difficult to distinguish the closed-cell from the open-cell regions during RF14 (15 November). The five remaining research flights examined here are RF06 (28 October), RF07 (31 October), RF08 (2 November), RF09 (4 November), and RF13 (13 November). During VOCALS-REx, if a suspected POC was present within aircraft range when flight plans were made for a given day, the C-Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | by most flights, except for RF06 and RF13, in which an additional subcloud and cloud level flight leg was flown.
For all flight legs, a suite of instruments measured thermodynamic quantities (temperature, water mixing ratio, pressure, etc., described in Wood et al., 2011b) to help characterize the structure of the boundary layer and lower free troposphere. Be-5 cause the humidity measurements were found to be biased high in previous analyses (Bretherton et al., 2010), the humidity measurements from the Lyman-alpha hygrometer are corrected according to Bretherton et al. (2010). The Wyoming Cloud Radar (WCR), also aboard the C-130, provided radar reflectivity from cloud and precipitation hydrometeors above and below the aircraft. The column maximum reflectivity from the 10 WCR is used to estimate precipitation rates at cloud base inside and outside the POCs using the Z-R relationship from Comstock et al. (2004). From the subcloud flight legs, the cloud fraction, cloud base height, and cloud liquid water path are estimated using the upward pointing Wyoming Cloud Lidar (WCL) and G-band Vapor Radiometer (GVR) (Zuidema et al., 2012), respectively. Sea surface temperatures (SSTs) are re-15 motely sensed by downward viewing Heimann radiometric sensors and corrected by 1 K, to correct a warm bias found by Bretherton et al. (2010).
Concentrations of aerosols of size > 0.1 µm in clear air are obtained from the Passive Cavity Aerosol Spectrometer Probe (PCASP) that measures the size distribution of aerosols with diameter between 0.1 and 3 µm. This size range encompasses most of 20 the accumulation-mode aerosols and is representative of concentrations of cloud condensation nuclei (CCN) at supersaturations typical of those in marine stratocumulus (Martin et al., 1994;Terai et al., 2012). Total aerosol concentrations (CN) for particle diameters > 10 nm were obtained by the 3010 CN Counter (Clarke et al., 2007). Aerosol data are all filtered for possible splashing and shattering events from drizzle 25 drops by removing aerosol data if any of the following conditions are met: liquid water content > 0.04 g m −3 (as measured by the Gerber PV-100 Probe); drizzle drop concentration > 1 L −1 (as measured by the 2DC Probe); a 10 s-forward-lagged, 11 s-movingwindow-mean drizzle water content > 10 −4 g m −3 (as measured by the 2DC Probe).
8293 at 1 Hz time resolution, corresponding to a spatial resolution of ∼ 100 m. Cloudy air is distinguished from clear air in the cloud-level flight legs when the liquid water mixing ratio from either the Gerber PV-100 Probe or the CDP+2DC exceeded 0.03 g kg −1 .
In order to locate the POC boundaries regardless of the time of day, infrared brightness temperature imagery from Channel 4 (centered on 11 µm) of the Geostationary Operational Environmental Satellite imager (GOES-10), obtained roughly every fifteen minutes, is used to locate the flight legs in relation to the POC boundary.

Differentiating between POC and overcast regions
We divide each relevant flight leg into POC, transition, and overcast (OVC) subsections by visual inspection of the GOES-10 infrared (IR) imagery. This allows us to compare 15 the various boundary layer properties in the POC with those in the overcast region. This approach is similar, but not identical to Wood et al. (2011a) and Berner et al. (2011), who instead used radar reflectivity to distinguish the regions. Figure 2 shows subcloud flight legs from each of the five POC cases overlaid on GOES-10 IR images taken closest to the leg time. The transition region is identified to ensure that POC and overcast 20 sections of the flight leg solely sampled the corresponding cloud types, with its width determined from the sharpness in the transition from overcast to broken clouds along the leg. In the following analysis, we compare the thermodynamic, macrophysical, and microphysical characteristics in the POC region and the overcast region.

Geographical and diurnal context
Four of the POC flights in this study were located at 80 • W and just to the north and south of 20 • S. The RF13 POC was located slightly to the east at 78 • W (see Table 1 and Fig. 2). The region around 20 • S and 80 • W has been characterized by a combina-5 tion of aircraft and shipboard measurements made during VOCALS REx (Bretherton et al., 2010;Allen et al., 2011) Leon et al., 2008;de Szoeke et al., 2012), and because a goal of VOCALS-REx was to observe the precipitation characteristics in POCs, four of the five POC-sampling research flights were conducted in the early morning hours (Table 1). Large differences in cloud properties between the one POC 20 sampled during mid-day (RF13) and the four POCs sampled during early morning are apparent in the precipitation rate and in-cloud turbulence (Table 3 and Sect. 5.3).

Synoptic setting
The C-130 VOCALS sampling period of 15 October to 15 November 2008 was characterized as a quiescent period, in which little synoptic activity propagated from the 25 mid-latitudes into the VOCALS region (Toniazzo et al., 2011). During this quiescent pe- riod, however, POCs contributed substantially to cloud cover variability (Toniazzo et al., 2011). Although no evidence has been found to show that the synoptic waves directly cause the POC features to form or that they help maintain POC features, because midlatitude storms have the potential of affecting air masses advecting into VOCALS region (George and Wood, 2010), we cannot discount a possible influence from mid-latitude 5 synoptic activity on the formation of POCs.

Free troposphere
Lower tropospheric stability (LTS) over the subtropical stratocumulus regions is positively correlated with low cloud cover (Klein and Hartmann, 1993;Leon et al., 2008).
In Table 2, we compare the LTS, calculated as the potential temperature difference be-10 tween 700 hPa level and the surface, over the POCs and surrounding overcast region. We find no indication that the VOCALS POCs formed under remarkably high or low LTS conditions. Because the aircraft did not fly many profiles up to 700 hPa (∼ 3100 m), the same 700 hPa potential temperatures are used to calculate LTS in POCs and overcast regions in Table 2. The surface potential temperatures are the mean potential temper-15 ature from the subcloud legs in each region. Because surface potential temperatures did not vary by much between separate POC cases (287-289 K), variations in LTS between cases are largely due to differences in the 700 hPa temperatures. Stevens et al. (2005) observed a moister free troposphere (FT) above a northeast Pacific POC, which led to the hypothesis that POCs formed under regions with moister 20 FT. Wood et al. (2008), however, found no evidence in the southeast Pacific that POCs formed preferentially in regions with a moister FT. Likewise, if we compare the water vapor mixing ratio (q v ) from the above-cloud flight legs flown over the POC and overcast region, then we find that q v is not consistently higher above the VOCALS POCs ( Air-sea temperature differences and wind speeds determine the temperature flux at the surface. Except for RF13, the SSTs are slightly lower in the POC than in the overcast region (Table 2). However, the apparent SST decrease in the POC might be a sampling 5 artifact, because the intervening water vapor between the aircraft and sea surface can affect radiometric SST estimates from the pyrometer (Stevens et al., 2005). Given that the subcloud temperature in the POC is approximately 1 K lower and the water vapor mixing ratio is 1 g kg −1 higher, we would expect SSTs retrieved in the POC region to be biased lower than SSTs retrieved in the overcast region. Despite this potential low 10 bias of the POC SSTs, the estimated air-sea temperature difference inside the POC is larger than in the overcast region ( Table 2). The cooler surface air temperatures in the POC, likely a result of cold pool formation, increase the air-sea temperature difference in the POC. Relative changes in wind speed across the POC-overcast transition are small ( Ta-15 ble 2). Therefore, any systematic changes in the surface sensible and latent heat fluxes are likely due to air-sea temperature and moisture differences in the POC. Across different POCs, a large range in surface-level horizontal wind speeds was measured (Table 2), which indicates that the maintenance of POCs does not require stronger or weaker winds.  (Table 2), which is on the higher end of the range of z i, base observed at 80 • W during all the VOCALS-REx flights (Bretherton et al., 2010), but there is no indication that POCs only form in deeper boundary layers. If we compare the POC with the surrounding overcast region, the inversion base height is 25 to 100 m lower in the POC than in the overcast region in RF06, RF07, RF08, and RF09, 5 but 50 m higher for RF13, the one POC sampled during the daytime. Inadequate sampling precludes attributing this difference to the diurnal cycle. The z i, base observed in the VOCALS POCs are substantially deeper than the 700-800 m inversion base heights of POCs observed previously over the northeast Pacific (see Table 2; Van Zanten and Stevens, 2005;Sharon et al., 2006). We conclude that POCs can be produced 10 in boundary layers with a range of inversion heights. The strength of the inversion determines how easily free tropospheric air can be entrained into the MBL. We compare the inversion structure over the POC and overcast region in Fig. 3. A mean profile for the overcast and POC region is constructed for each case from averaging the potential temperature from the profiles in 25 m height bins. To 15 compare results from different POC cases, the mean inversion height (z i ) from each POC and overcast case found in Table 2 is subtracted from each profile. Rather than using z i, base for this purpose (which would align the bottom of the gradient regions in the profiles), we define the z i in each flight profile to correspond to the maximum of the 5 s-running-mean of dθ/dz and subtract the mean of those heights from each 20 averaged profile. Table 2 gives this mean z i for the POC and overcast (OVC) regions of each flight; for the four nocturnal flights, the mean z i is quite similar in the POC and OVC regions, even though z i, base is lower in the POC. Figure 3 shows that the inversion is typically sharper in the OVC region than in the POC. This is expected given how radiative cooling from thick overcast clouds helps to sharpen the inversion. However, 25 the inversion structure also differs from case to case. For example, the inversion in the overcast profile from RF09 is especially hard to identify. This is partly a result of averaging six different overcast profiles into one, but visual inspection shows that the inversions in individual profiles from the overcast region of RF09 are also less sharp than in other cases.

Decoupling
The degree to which the boundary layer is decoupled can be quantified by taking the difference (∆z) between the cloud base height and the lifting condensation level (LCL) 5 from the subcloud flight legs (Jones et al., 2011). Defined this way, the decoupling gives an indication of the relative timescales between boundary layer mixing and processes that act to stabilize the boundary layer. From Table 3, we see that for all cases, the POC has a significantly (150-350 m) lower LCL. In the nocturnal cases, the ∆z is larger (indicating stronger decoupling) in the POC than the surrounding OVC region. In the one daytime flight (RF13), the OVC is more decoupled, consistent with observational and large eddy simulation studies that find that solar insolation strongly decouples the daytime overcast boundary layer (Turton and Nicholls, 1987;Bretherton et al., 2004;Caldwell and Bretherton, 2009 is not shared across the other POC cases. The stronger decoupling in POCs is likely a manifestation of strong drizzle evaporation in the subcloud layer that leads to the stabilization and decoupling of the lower boundary layer. This traps surface fluxes in the lower levels of the boundary layer and leads to lower LCL. LES studies have shown that the increased stability from subcloud 20 drizzle evaporation enhances the transition from closed to open cellular convection (Savic-Jovcic and Stevens, 2008) and that cold pools help to organize cumulus cloud convection (Wang et al., 2010;Berner et al., 2011). It should however be noted that cold pools are not exclusive to POCs, but are also observed under non-POC forming overcast stratocumulus, and hence do not serve as a sufficient condition for POC 25 formation (Terai and Wood, 2013

Vertical wind variance (w 2 )
Counteracting the stabilization of the boundary layer, the radiative cooling at cloud top drives turbulence and enhances the mixing between the cloud layer and underlying surface mixed layer. The strength of the turbulence can be quantified by the vertical velocity variance (w 2 ) measured within cloud-layer legs. Corroborating Wood et al.

5
(2011a), we find that w 2 is lower in the POC across all cases (Table 3). The very low w 2 values in both POC and OVC regions in the RF13 case sampled during mid-day are probably due to absorption of solar radiation, which reduces the net cloud-layer radiative cooling that helps drive turbulence. It is illuminating to see local variations in w 2 over the course of a flight leg, which we calculate based on a 21 s (∼ 2 km) moving window, which acts as a crude high-pass filter to isolate turbulent motions from mesoscale fluctuations. In Fig. 4, measurements from the cloud-level flight leg of the RF08 POC are shown to illustrate the observed changes in cloud, drizzle, and vertical wind speed across the POC-to-OVC transition. Figure 4b shows that w 2 is intermittent in the POC. Long stretches of very low w 2 are interrupted by short bursts of elevated 15 w 2 , which are associated with cumulus clouds. Although there are fluctuations in w 2 in the overcast region, the values are generally high. Cloud level w 2 also gives an indication of how much turbulence is available to entrain air from the FT. The much lower values of cloud-layer w 2 within POC regions compared to surrounding overcast regions with similar inversion temperature jumps suggest that the entrainment rates Introduction

Cloud macrophysics
A defining difference between the POC and overcast regions in a satellite image is the difference in cloud fraction. For cloud fraction estimates we use the upward-pointing Wyoming Cloud Lidar to detect overlying cloud during the C-130 subcloud flight legs.

5
Cloud fraction in overcast regions are nearly 100 %, while on average, the POC regions have more variable cloud fraction between 56 % and 83 % ( Table 3). The all-sky mean LWP can be higher or lower in the POC than in the OVC region, but the relative standard deviation of LWP (the ratio of the standard deviation of LWP to its mean, both given in Table 3), tends to be larger in the POC. This, together with enhanced decoupling, is an 10 indicator of more cumulus-like convection within the POC (Jones et al., 2011). These differences in cloud structures also manifest themselves in the precipitation rates. The mean precipitation rates are higher in the POCs than in the surrounding OVC regions (Table 3). POC radar reflectivities are also more broadly distributed. The column maximum radar reflectivity (Z max ) is plotted against the fraction of columns with 15 reflectivity greater than that value in Fig. 5a. The broader distribution in the POC is evident from the shallower slope of the POC cumulative distribution, compared to the OVC. In particular, there is a much larger fractional coverage of strong drizzle > 10 dBZ for all POC cases. This contrast is consistent with statistics of the distribution of reflectivities within open and closed cell stratocumulus observed previously over the SEP 20 with shipboard cloud radar (Comstock et al., 2007) and over the global oceans with CloudSat (Muhlbauer et al., 2014). Terai and Wood (2013) found that precipitation at lower elevations is a better indicator of cold pool formation than the occurrence of precipitation at cloud base (where the column maximum reflectivity is observed). Similarly, by examining the reflectivity at 25 250 m (Z 250 ) in Fig. 5b, instead of the column maximum (Z max ), we find that reflectivities greater than 0 dBZ occur on average 10 % of the time in POCs compared to only ∼ 1.5 % of the time in the OVC region. Among the different precipitation characteristics, 8301 Introduction this best differentiates the POC precipitation from the OVC precipitation. The reflectivities at 250 m are likely influenced by the amount of subcloud evaporation between the cloud base and 250 m. Therefore, the higher occurrence of > 0 dBZ precipitation at 250 m in the POC may reflect lower cloud base heights in the POCs rather than a difference in the drizzle drop size distributions.

5
The macrophysical signatures of the cloud and precipitation properties in POCs all point to more cumuliform convection and correspondingly patchier and heavier precipitation within the POC than the surrounding overcast region.

Cloud microphysics
We now consider the cloud microphysical properties in the POCs. Consistent with pre-10 vious POC observations (Petters et al., 2006;Sharon et al., 2006;Van Zanten and Stevens, 2005;Wood et al., 2011a), all five of the POC cases examined here show cloud droplet number concentrations (N d ) decrease by more than a factor of eight from the OVC region (Table 4). Whereas the mean cloud-layer N d inside the POCs has a range between 4 and 13 cm −3 across cases, the N d in the surrounding OVC region 15 range from 52 to 207 cm −3 , further showing that POCs can persist when surrounded by a wide range of microphysical environments. In the surrounding OVC, the case-tocase variations in N d are highly correlated (r = 0.98) with the variations in subcloud PCASP aerosol concentrations (N a ). In the POCs, the relative changes in N d is much more spatially variable than in OVC clouds, and N d is at least a factor of three lower 20 than the subcloud N a . As we discuss below (Sect. 5.3), this is due to dramatically lower N d in "quiescent" stratocumulus clouds than in the "active" cumulus clouds that bring up the subcloud-layer aerosol. Despite this, the mean subcloud N a is correlated to N d in the POCs (r = 0.76). The horizontal-mean cloud water mixing ratio (q c ) drops drastically inside the POCs 25 (Table 4), due in part to a lower cloud fraction. In contrast, the horizontal-mean drizzle water mixing ratio (q D ) at least doubles inside the POCs, such that in all five POC cases 8302 the mass of water in the drizzle mode is larger than that in the cloud mode; remarkably, the q D /q c ratio is at least twenty-fold larger in the five POCs than in the OVC regions. The increase in the estimated precipitation rate from the surrounding OVC region to the POC is not as large as that in q D . This apparent discrepancy is explained by the distinct size distribution of cloud and 5 drizzle drops in the POC (Fig. 6). If we examine the cloud and drizzle drop size distributions measured by the CDP and 2DC in Fig. 6, we note a large increase in the drizzle drop concentrations in the POCs. However, since these are the smallest of drizzle drops, while they may contribute to a substantial proportion of the drizzle mass, they do not contribute strongly to the precipitation rate. The increase in the number of 10 large cloud and small drizzle drops is curious and may be related to the very low N d and different cloud types in POCs.

Active vs. quiescent clouds
A closer examination of the clouds found in POCs reveals two general types of clouds: thinner, stratiform, quiescent clouds and cumulus-like active clouds (Wood et al., 15 2011a). We use a w 2 threshold of 0.1 m 2 s −2 as a means of distinguishing between the quiescent and active clouds inside the POC, except for the daytime case RF13, where the overall cloud-level vertical wind variance is low and hence where a threshold of 0.03 m 2 s −2 is used. In the RF08 example (Fig. 4), one can see lower N d , lower vertical velocities, and consistently elevated drizzle drop number concentrations (N D ) in the 20 quiescent clouds than in the active clouds. in the active clouds, the latter having mean N d on the order of 15 cm −3 . RF07 is an apparent exception, but active clouds were rarely sampled on this flight, with only 2 km of the flight leg sampling active clouds. The ratio between q c and q D differ between the two regions, much like the change in ratio between overcast and POC clouds; in active clouds, q c and q D are comparable, whereas in quiescent clouds, q D is far greater.

5
Nonetheless, it is important to note that the q D in both active and quiescent clouds of the POC is far greater than in the overcast clouds. The high q D is likely to increase the accretion and self-collection rates in POCs. Table 5 shows, however, that in each POC case higher q D in quiescent clouds does not always translate to higher precipitation rates, which are estimated from the in situ 2DC-measured drizzle size distribution 10 averaged over the appropriate cloud type. As in the POC-overcast comparison, this discrepancy exists because much of the drizzle mass difference resides at the smaller sizes, which have a smaller influence on precipitation rate.  (Table 4). The cases (e.g., RF08) with higher N a in the overcast region are related to hook features associated with local-25 ized offshore aerosol transport in the free-troposphere and boundary layer, described 8304 by George et al. (2013). The subcloud N a in the POC is uncorrelated with N a in the surrounding overcast and shows remarkable small variation (24-40 cm −3 ) across the five POC cases. If we extend the comparison to other previously observed POC cases (Sharon et al., 2006;Petters et al., 2006;Wood et al., 2008), the similarity in subcloud N a still holds (Table 4). Spatially, across the overcast-to-POC transition, subcloud N a 5 gradually decreases over tens of kilometers across the transition (Fig. 2), whose gradients across the transition vary from case to case. Even within the POC, Fig. 2 and Table 4 show that subcloud N a can spatially vary by more than 25 %. Whereas we have mostly examined data collected during the level flight legs to examine horizontal gradients in N a , we use both the profiles and level flight legs to ex-10 amine the vertical distribution of N a (Fig. 7). In the RF06 case, Wood et al. (2011a) noted a layer of ultraclean air in the POC region where mean PCASP aerosol concentrations are considerably lower than 10 cm −3 . We find that an ultraclean layer exists in all of the other POC cases examined during VOCALS. The ultraclean layers typically extend from roughly the stratocumulus cloud base height to the inversion base, with a 15 broad range of thicknesses (from 200 m for RF06 to 700 m for RF07). The concentrations in the ultraclean layer also vary from case to case, ranging from the very clean RF09 case, where a 400 m layer with concentrations < 1 cm −3 was observed, to the more polluted RF08 case, where concentrations mostly hovered around 5 cm −3 and those concentrations < 1 cm −3 were rarely observed. Concerning the structure of the 20 ultraclean layer, with the exception of RF08, the vertical gradient in aerosol concentration is sharper at the inversion compared to the more gradual change from relatively well-mixed subcloud layer to the ultraclean layer.

Aitken-mode aerosol
Whereas N a is much lower in the POC at all heights, the same is not true for concentra- that the low aerosol concentrations in the POC allow for the nucleation of Aitken-mode aerosols. Their growth to accumulation-mode sizes may potentially provide a source of CCN in the POC (Kazil et al., 2011). The Aitken-mode aerosol concentration is estimated from our C-130 observations as the difference between total aerosol concentrations for particles with diameter > 10 nm (CN) and those from the PCASP (> 100 nm).
From our observations of the five POC cases, no clear picture of enhanced or depleted Aitken-mode aerosols emerges. Subcloud Aitken-mode aerosol concentrations in the POC only increase for three of the five cases (RF06, RF08, and RF09; Table 4), and their concentrations do not correlate with subcloud N a . We also find that except for the daytime RF13 flight, the concentration of Aitken-mode aerosols decreases in the ultraclean layer, where we expect the highest nucleation rates due to low aerosol surface concentrations. It must be noted that four of the POCs were sampled during the night, when photochemical production of nucleation mode precursor gases is absent.
What we can say is that unlike the accumulation-mode aerosols, which have a large sink in incloud coalescence scavenging across all POCs, which defines their vertical 15 horizontal structure, the Aitken-mode aerosols appear to be modulated by a variety of processes that are not as well constrained across different POC cases.

Accumulation-mode aerosol budget in POCs
Although we should entertain the possibility that the relatively small variation between cases in POC subcloud N a is coincidental, it certainly warrants attention. The similarity 20 across different POCs indicates that the source and sink balance of accumulationmode aerosols somehow gives rise to relatively constant concentrations. The small spread between cases is surprising given that the observations are strongly suggestive of a surface source of aerosols and that mean surface wind speeds vary between 5 and 11 m s −1 between cases. Surface aerosol fluxes from sea-salt flux parameterizations where N + is the CCN concentration just above the subcloud layer, F is the surface source of CCN-sized aerosols, and M Cu is the entrained mass flux into the subcloud 5 layer from above, that keeps the surface layer partially coupled to the cloud layer in the POC. This must equal the cumulus mass flux out of the subcloud layer. In other words, the surface flux of CCN at the surface is balanced by the mixing of CCN-depleted air into the top of the surface layer.
We may obtain estimates of N + , F , and M Cu to verify whether this balance equa-10 tion applies to the observed POC cases. N + is estimated as the mean PCASP aerosol concentration from the below cloud flight legs. F is estimated using the sea salt flux parameterization from Clarke et al. (2006), integrated over the size distribution of PCASPsize aerosols. The 10 m mean wind speed (u 10 ) that is needed for the calculation is estimated from the mean wind speed measured in the subcloud legs and assuming a 15 log-wind profile with a surface roughness of 1.86 × 10 −4 m (Wood et al., 2011a). Because the parameterized aerosol flux has a dependence of u 3.41 10 , an order of magnitude difference in fluxes exists between the low and high wind conditions (see Table 6). The cumulus mass flux (M Cu ) can be estimated from examining the water balance at the top of the subcloud layer. If we assume that the upward moisture flux in the cumulus 20 clouds is balanced by precipitation rates, then M Cu can be estimated as where R 500 is the precipitation rate at 500 m, ∆q is the moisture difference between cloudy and clear parcels at the top of the subcloud layer, and ρ is the air density, which we assume to be 1 kg m −3 . The precipitation rate at 500 m is chosen, because this is the approximate height of the LCL ( height of the top of the subcloud layer. The estimated cumulus mass fluxes, reported in Table 6, range between 1.0 and 2.8 cm s −1 . Except in RF09, the estimated F/M Cu is much smaller than the observed N s − N + , inconsistent with the balance (Eq. 1). There are several possible reasons for this discrepancy. First, determining an appropriate N + is made difficult by the vertical gradient in N a at the top of the subcloud layer, as 5 evident in Fig. 7. Rather than decreasing abruptly at the top of the subcloud layer, N a gradually decreases upward into the cloud layer. Second, we have neglected any other sources of CCN-sized aerosols, such as the growth of Aitken-mode aerosols to accumulation-mode sized particles. Third, other budget terms such as losses from precipitation scavenging of aerosols in the subcloud layer could be significant. What 10 we can say from the values in Table 6 is that F /M Cu can easily vary by more than a factor of four, and their variations do not correlate with N s or N s − N + . We are therefore left without an explanation to why the subcloud N a in POC varies so weakly across different POC cases.

15
Two cases of cloud breaks which may appear as contenders for POCs, but which are excluded from our definition of POCs, are examined here. The first case of cloud breakup sampled during RF02 (18 October) and shown in Fig. 8a was sampled to the south of a hook-like cloud feature of enhanced N d from the transport and entrainment from the free troposphere of anthropogenic aerosols (George et al., 2013). Despite the break 20 in the cloud to the south of the hook feature, we do not identify this feature as a POC, because it lacks the microphysical change (a substantial increase in effective radius at its edges) that we associate with POCs. In situ observations show that instead of a decrease in accumulation-mode aerosols, there is a subsequent increase in subcloud N a from ∼ 200 to ∼ 300 cm −3 to the south. Likewise, the N d increases from ∼ 100 to Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | at approximately 1000 m around 83 • W is anomalously low, and the decrease in cloud cell size to the south in the satellite visible imagery hints at a lowering of the inversion height to the south that might mean a lowering of cloud top heights to the south. The second case (Fig. 8b) corresponds to western edge of the stratocumulus cloud deck observed during RF04 (23 October), where large drizzle rates were observed.

5
High radar reflectivities (> 10 dBZ), representing heavy drizzle, cold pool formation, and an increase in cloud effective radius towards the edge of the cloud deck were observed in daytime satellite retrievals of this case, but no open cellular convection was observed beyond the edge of the overcast clouds, excluding this case from our definition of POCs. We also do not observe a decrease in N a to the west of the break up, 10 although a decrease of aerosols in the clear region is not part of the definition of POCs in this study. This suggests that higher precipitation rates at the edges of overcast clouds do not always indicate that the strong aerosol-cloud-precipitation interactions that we observe in POCs are the cause for the break-up in the clouds.

15
Five pockets of open cells (POCs) sampled during VOCALS-REx by the NSF/NCAR C-130 are comprehensively compared with their surrounding overcast regions. Free tropospheric water vapor and temperature, sea surface temperatures, inversion height, and near-surface winds were not found to differ appreciably in the POCs compared with the surrounding overcast stratocumulus areas, and are not very different from 20 the REx-mean conditions in the sampled regions. This latter fact indicates that POCs can be maintained under typical large-scale meteorological conditions found over the Southeast Pacific.
A consistent feature of POCs is the presence of heavy drizzle rates (> 10 dBZ) and associated cold pools, which are largely absent in the surrounding overcast regions.

25
All five of the observed POCs also contain both active, cumuliform and quiescent, stratiform clouds. Active clouds associated with cumulus-like convection, make up a 8309 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | smaller proportion of the cloud cover (CF < 25 %) in the POC and have larger cloud droplet number concentrations (N d ) and larger cloud water mixing ratios (q c ). Although collision-coalescence rates are expected to be high in the active clouds, their comparatively high N d (20 cm −3 ) suggests that active clouds loft relatively aerosol-rich subcloud-layer air to the cloud layer and detrain some of this aerosol into surround-5 ing quiescent clouds, which cover more of the POCs and are characterized by very low N d (< 10 cm −3 ) and high drizzle water mixing ratio (q D ). We cannot determine from these observations whether quiescent clouds are spatial extensions of active cells, as in the trailing stratiform region of larger mesoscale convective systems, are remnants of decaying active clouds, or are formed in situ. Snapshots of POC cloud fields from LES 10 studies of Berner et al. (2011) andKazil et al. (2011) suggest that quiescent clouds are spatial extensions of active cumulus clouds, but a closer examination of the time evolution of quiescent clouds will be necessary to resolve this issue.
The efficiency of the in-cloud coalescence scavenging in cleaning out accumulationmode aerosols in POCs is most evident in the apparently omnipresent ultraclean layer 15 observed across all five POCs. The presence of the ultraclean layer, combined with the low values of vertical velocity variance in the cloud layer within the POC and increased aerosol concentrations at lower altitudes, also strongly suggests that the surface source of accumulation-mode aerosols, and hence of cloud condensation nuclei, is more important than a free tropospheric source. In addition, we find that the spread of mean 20 subcloud-layer accumulation-mode aerosol concentration (N a ) in the POCs is remarkably narrow, with a mean value (∼ 30 cm −3 ) that is close to values found in previously studied cases from other locations and times (Sharon et al., 2006;Petters et al., 2006;Wood et al., 2008). This mean value appears to be insensitive to N a in the surrounding overcast region, suggesting that the factors controlling the aerosol budget in the POC 25 are independent of those in the surrounding overcast. The factors that determine the N a budget in the POC subcloud layer in the simple aerosol budget of Eq. (1) are quantified in Table 6. We find that our simple budget calculation does not shed light on explaining the narrow range of observed subcloud 8310 N a in POCs. Instead it leaves us with a puzzle as to how the narrow range of N a exists when the surface source of aerosols can differ by more than an order of magnitude. Because high resolution cloud resolving model studies with interactive aerosols also show similar accumulation-mode aerosol concentrations in the subcloud layer (Kazil et al., 2011;Berner et al., 2013), further modeling studies with variable surface source 5 functions may help shed light on whether there is indeed a physical explanation for the similarity.