An aircraft case study of the spatial transition from closed to open mesoscale cellular convection over the Southeast Pacific

An aircraft case study of the spatial transition from closed to open mesoscale cellular convection over the Southeast Pacific R. Wood, C. S. Bretherton, D. Leon, A. D. Clarke, P. Zuidema, G. Allen, and H. Coe Atmospheric Sciences, University of Washington, Seattle, USA Atmospheric Science, University of Wyoming, Laramie, USA Department of Oceanography, University Hawai’i, Honolulu, USA Rosenstiel School of Marine and Atmospheric Science, University Miami, Miami, USA School of Earth, Atmospheric and Environmental Sciences, University Manchester, Manchester, UK

sharp spatial transition in marine boundary layer (MBL), cloud, and aerosol structure across the boundary between a well-mixed MBL containing overcast closed mesoscale cellular stratocumulus, and a pocket of open cells (POC) with significantly lower cloud cover. Long (∼190-250 km) straight and level flight legs at three levels in the marine boundary layer and one level in the lower free troposphere permit sampling of 10 the closed cells, the POC, and a 20-30 km wide transition zone with distinctly different structure from the two airmasses on either side. The POC region consists of intermittent active and strongly precipitating cumulus clouds rising and detraining into patches of drizzling but quiescent stratiform cloud which is optically thin especially toward its edges. 15 Mean cloud-base precipitation rates inside the POC are several mm d −1 , but rates in the closed cell region are not greatly lower than this, which suggests that precipitation is not a sufficient condition for POC formation from overcast stratocumulus. Despite similar cloud-base precipitation rates in the POC and overcast region, much of the precipitation (>90%) evaporates below cloud in the overcast region, while there is sig-20 nificant surface precipitation inside the POC. In the POC and transition region, although the majority of the condensate is in the form of drizzle, the integrated liquid water path is remarkably close to that expected for a moist adiabatic parcel rising from cloud base to top.
The transition zone between the POC and the closed cells often consists of thick

Introduction
The influence of the structure, dynamics, and microphysics of marine stratocumulus clouds on the nature and quantity of the precipitation that they produce has been a fo-25 cus of research stretching back over seventy years to pioneering studies by Walter Findeisen in the late 1930s which showed that even relatively thin, low liquid water 17913 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | content clouds could produce drizzle-sized drops through collision-coalescence (see Mason, 1957). Simpson (1941) and Squires (1952) discuss observer reports of precipitation falling from warm clouds in the subtropics and tropics. Mason (1952) showed theoretically that clouds thinner than 1 km can produce drizzle that can reach the surface especially if the cloud is turbulent. The aircraft observations of Mason and 5 Howorth (1952) and Singleton (1960) demonstrated unequivocally that warm stratocumulus clouds as thin as 300-600 m can produce surface drizzle. Squires (1958a,b) opened up the study of factors controlling warm rain formation including the importance of increased cloud droplet concentration, and hence aerosol concentration, for reducing the propensity for precipitation formation in low cloud. Very 10 little work then followed until interest in stratocumulus clouds grew and a significant number of aircraft case studies of marine stratocumulus showed drizzle to be a common feature (Brost et al., 1982;Nicholls, 1984;Nicholls and Leighton, 1986;Austin et al., 1995;Bretherton et al., 1995) and, importantly, that precipitation rates were frequently sufficient enough to play an important role in the cloud moisture budget. Sen- 15 sitive millimeter radar studies also began to shed light on the structure of precipitation in marine low cloud (Frisch et al., 1995;Miller and Albrecht, 1995;Vali et al., 1998) and were central in establishing that drizzle is intermittent and can be locally strong, particularly where cumulus clouds are growing and detraining into a marine stratocumulus layer above. 20 Together, these aircraft and radar studies engendered the idea that stratocumulus precipitation might exert an influence on the stability of the boundary layer, and thus cloud dynamics and structure. This notion fuelled various modeling studies of varying complexity (e.g., Liou and Ou, 1989;Albrecht, 1989;Wang et al., 1993;Feingold et al., 1997;Stevens et al., 1998;Savic-Jovcic and Stevens, 2008) which all showed sensitivity of cloud cover, thickness, and MBL structure to drizzle. Many of these modeling studies suggest that drizzle, particularly if it is strong (cloud base precipitation rates on the order of 1 mm d −1 ), has a tendency to decrease cloud cover because drizzle acts to stabilize the MBL by suppressing moisture transport into the cloud and pro-moting decoupling. Recent radar and satellite observations have revealed a striking connection between the mesoscale morphology of marine stratocumulus and the degree to which they precipitate, with spatial transitions between regions of closed and open mesoscale cellular convection frequently associated with an increase in precipitation strength Comstock et al., 2005, 5 2007). It is thus reasonable to suppose that drizzle plays a significant role in setting the climatological cloud cover. However, since there have been only very limited aircraft missions contrasting precipitation and cloud structure transitions between open and closed cells Sharon et al., 2006), there are many remaining questions regarding the way in which precipitation impacts MBL cloud cov-10 erage and thickness.
Cloud modeling studies have advanced the assertion (Squires, 1958a,b) that increased cloud droplet concentration could limit precipitation and have opened a pathway through which increasing atmospheric aerosols might alter cloud cover regionally (Liou and Ou, 1989;Albrecht, 1989). Studies collating many aircraft and re- 15 mote sensing measurements have confirmed the hypothesis that stratocumulus drizzle rates are sensitive to cloud droplet concentration (Pawlowska and Brenguier, 2003;Comstock et al., 2004;Wood, 2005;Geoffroy et al., 2008). Without adequate controls, it has proven difficult to establish observationally whether microphysically-suppressed precipitation is able to increase stratocumulus cloud cover 20 or thickness. Cloud resolving model studies suggest that increased cloud condensation nucleus (CCN) concentrations can induce non-monotonic responses in cloud cover (Xue et al., 2008) and cloud thickness (Ackerman et al., 2004). These responses are the result of competing effects with increased CCN suppressing precipitation formation but, as a consequence, increasing the lateral and cloud-top entrainment that tends to 25 reduce cloud volume (Wood, 2007;Xue et al., 2008;Stevens and Feingold, 2009).
Ship and aircraft observations show large reductions in CCN concentration, and consequently lower cloud droplet concentration, within pockets of open cells (POCs) embedded in overcast marine stratocumulus (Sharon et al., 2006; Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Wood et al., 2008). Spaceborne remote sensing measurements of cloud droplet effective radius are consistent with such microphysical contrasts Rosenfeld et al., 2006;Wood et al., 2008), and there is a tendency for regions of open cells to be embedded within overcast cloud regions with low cloud droplet concentrations (Wood et al., 2008). This, together with the finding that open cells form- 5 ing within marine subtropical stratocumulus are rarely found close to the coasts (Wood and Hartmann, 2006), may tempt one to attribute the formation of POCs to low CCN concentrations since low CCN would drive stronger precipitation all else being equal. Cloud resolving model simulations do indeed show that spatial CCN gradients can induce precipitation gradients that lead to closed/open cell cloud morphology gradients 10 (Wang and Feingold, 2009). The conclusion that CCN differentials are the primary trigger of POCs in nature, however, may be difficult to establish since we also know from simple calculations that mean precipitation rates within POC regions (∼1 mm d −1 ) are sufficient to drive significant rates of depletion of CCN through coalescence scavenging (Wood, 2006). For a 1 km deep boundary layer the timescale for CCN re- 15 moval through coalescence scavenging is only ∼1.5 d for a cloud base precipitation rate of 1 mm d −1 (Wood, 2006). Therefore, we would expect precipitation differentials, which might be driven by dynamics, to drive CCN differentials. The tight coupling between microphysics and dynamics within POCs, and perhaps within drizzling stratocumulus boundary layers in general is therefore an intriguing area where improved 20 measurements may shed important light on POC formation mechanisms.
Here, we present aircraft measurements from two aircraft flights sampling the same broad region of transition between overcast stratocumulus and a pocket of open cells, taken over an 18 h period (27/28 October 2008) during the VOCALS Regional Experiment (REx) 1 . A major focus of VOCALS-REx is an improved understanding aerosol- The VOCALS (Variability of the American Monsoon Systems Ocean-Cloud-Atmosphere-Land Study) Regional Experiment (VOCALS-REx) is an international field program designed to make observations of poorly understood but critical components of the coupled climate system of the southeast subtropical Pacific, a region dominated by strong coastal upwelling, extensive cloud-precipitation interactions within drizzling stratocumulus clouds and the role that these interactions play in driving mesoscale cloud variability. As such, the aircraft missions documented here are the first dedicated specifically to documenting the transitions between an overcast region of closed mesoscale cells and a pocket of open cells embedded within it. 5 This paper describes observations from an aircraft case study contrasting the structure and dynamics of the spatial transition from overcast marine stratocumulus to a pocket of open cells. The organization of the paper is as follows. Section 2 describes the sampling, instrumentation, and data. Section 3 describes the mesoscale structure and the rationale for breakdown into different regions, while Sect. 4 describes the large 10 scale context. The mean structure and the cloud/aerosol microphysics are presented in Sects. 5 and 6, respectively. Section 7 is a discussion and introduces a conceptual model of the transition from overcast to open cells. Conclusions are drawn in Sect. 8. 15 Data from two research flights are used in this study.  Figure 1 shows visible satellite imagery of the sampled POC and environs approximately 3 h after the end of the C-130 measurements. As seen in the GOES whole-disk 20 image ( Fig. 1 Trajectories estimated using NCEP GFS analysis and short-range forecast fields were used to position the C-130 in the same airmass as that sampled by the BAe-146 during B409 roughly 12 h earlier. Flight RF06 was conducted during the early morning hours of 28 October with the science sampling taking place from 08:00 to 13:30 UTC 10 (03:00-08:30 LT). The flight consisted of a series of six straight and level legs 190-250 km in length which were designed to sample the transition between a POC and the surrounding overcast closed cellular stratocumulus. In addition, a sawtooth run with the aircraft climbing and descending from 100 m above the inversion to approximately 100 m below stratocumulus cloud base was carried out, and profiles were taken well 15 into the POC and overcast regions. All runs and profiles are oriented NNE-SSW with headings of 30/210 • , which resulted in approximately Lagrangian airmass sampling over 5.5 h. At the end of the legs, 180 • turns of opposite handedness (left then right then left, etc.) were used, which led to a slight upstream lateral drift of the legs with respect to the mean flow (∼60 km over the 5.5 h). This ensured that the sampling of the 20 boundary, which as Fig. 1 shows is quite heterogeneous on the <100 km scale, was not biased by repeatedly sampling the same set of cells throughout the flight. Figures 3 and 4 show sections of the BAe-146 and C-130 flight tracks overlaid on near-coincident GOES IR and (when sufficient sunlight permits) visible imagery, during missions B409 and RF06, respectively. Patches of thin high cloud are evident on both flights, particularly in the IR, and preclude good spaceborne visualization of the boundary especially in the hours before sunrise (10:42 UTC) on RF06. Both flights sample the boundary between a pocket of open cells to the SW and overcast stratocumulus to the NE, although the cloud layer straight and level run in B409 appears to skim the eastward edge of the POC feature. On both flights, the boundary between the open and closed cells is rather uneven and consists of cellular cloud features with horizontal scales of 20-40 km.

Instrumentation
Full details of the instrumentation flown on the C-130 and BAe-146 during  REx are given in Wood and coauthors (2010).
We present data from numerous instruments including atmospheric state variables and winds (standard C-130 or BAe-146 instruments), cloud liquid water content (Particle Volume Monitor, PVM on the C-130 and Johnson-Williams/Nevzorov on the BAe-146), drizzle liquid water content and drop size distribution (r>30 µm, 2-D-C optical 10 array probe), and cloud droplet concentration N d and size distribution (1<r<23.5 µm, Droplet Measurement Technologies Cloud Droplet Probe, CDP). All measurements were analyzed at 1 Hz time resolution unless otherwise stated. The 2-D-C probe flown on the C-130 in VOCALS was modified by doubling the number of diodes from 32 on the conventional probe to 64. The electronics were updated which reduces the number 15 of particles lost due to slow signal ramping. Counts from the first two size bins were excluded since the sample volume for these particles is difficult to determine. The 2-D-C thus sampled drops with radii larger than approximately 30 µm.
Size distributions of dry particles measured on the C-130 are reported at 60 s resolution by merging cabin-sampled size distributions from the following instruments: 20 a custom build radial differential mobility analyzer (RDMA, dry diameters 10-150 nm), a "long" DMA (LDMA, diameters 10-500 nm), a custom modified PMS LAS-X optical particle counter (OPC, diameters 150-1000 nm) and a TSI model 3321 aerodynamic particle sizer (APS, diameters 0.78-10 µm). Details are given in Clarke et al. (2007). In addition, an externally-mounted PCASP gives nominally dry size distributions for di-25 ameters 120-3000 nm. Total aerosol concentrations N CN are measured with TSI 3025 and 3010 CN counters (>3 nm and >10 nm, respectively). Aerosol size distributions on the BAe-146 are measured using an SMPS system with total concentrations from 17919 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | a TSI CN counter. On the C-130 one of the inlets to the 3010 CN counters is heated to ∼350 • C which is sufficient to volatilize sulfuric acid and ammonium/sulfate salts and provide the concentration N CN,hot of refractory CN (these measurements are described in Clarke et al., 2007). From this we estimate the non-refractory (volitilizable) fraction f non-ref of total particles larger than 10 nm.

5
The C-130 flew a substantial remote sensing suite during VOCALS. Measurements used in this study are upwelling and downwelling broadband fluxes (from solar and IR pyranometers), profiles of radar reflectivity Z from the zenith and nadir-viewing 95 GHz University of Wyoming Cloud Radar (WCR), lidar backscatter profiles from the zenithviewing Wyoming Cloud Lidar (WCL), and microwave radiances from a wing-mounted 10 zenith-viewing 183 GHz G-band microwave radiometer (GVR).

Derived data products
Several derived data products are used in this study. Data at 1 Hz time resolution (100 m horizontal resolution) from the WCR on the C-130 are used to derive (a) cloud top height z top for those clouds with significant radar echoes, using a threshold of 15 −35 dBZ; (b) column maximum radar reflectivity Z max ; (c) near-surface reflectivity Z sfc . Precipitation rates are estimated from Z max and Z sfc using Z-R relationships appropriate for drizzling stratocumulus . From the maximum reflectivity we derive the column maximum precipitation rate R max =(Z max /25) 0.77 and from the near-surface (250 m) reflectivity Z 250 we derive a 250 m precipitation rate 20 R 250 =(Z 250 /57) 0.91 . In addition, the in-situ drop size distribution measurements from the 2-D-C probe are used to estimate the precipitation rate assuming terminal velocities from Pruppacher and Klett (1997).
Cloud top height taken to be the highest gate where the WCR return is determined to reflect scattering from hydrometeors rather than receiver noise or other artifacts.

25
To avoid spurious cloud top height determinations caused by random variations and non-ideal behavior of the receiver noise, a combination of thresholds is used for each range gate: over half of the samples within the 1 s average must exceed the standard deviation of the noise and the 1 s average must exceed 1.5 times the standard deviation of the noise; the variance of the Doppler velocities within the second must be less than 0.4 time the variance associated with noise; finally, cloud top height must be within ±100 m of the median cloud top height for the leg.
The C-130 WCL lidar backscatter profile is used to determine the cloud base altitude 5 using a maximum gradient method. Cloud cover is also determined from subcloud legs using the WCL by identifying cloud features above the aircraft. Liquid water path (LWP) is derived from the GVR on the C-130 using the 183.31±14 GHz channel as described in (Zuidema and coauthors, 2010). Such estimates are only available for the two subcloud legs (Table 1). 10

Distinguishing cloudy, clear, and drizzling samples
Each 1 Hz data sample is classified as being either cloudy, clear, or containing drizzle (but not cloudy) using a combination of sensors. To be classified as cloudy, either the PVM or CDP liquid water mixing ratio has to exceed 0.03 g kg −1 . To be classified as containing drizzle, the sample must not be cloudy as defined above, but the 2-D- 15 C probe must indicate a drizzle drop concentration of at least 1 L −1 . All samples not classified as cloudy or drizzle-containing are deemed to be clear.
In this paper, we refer to the PVM-measured liquid water as the cloud liquid water mixing-ratio q L since the PVM instrument is relatively insensitive to droplets with radii greater than approximately 25 µm (Wendisch et al., 2002). Since the 2-D-C counts 20 drops with radii larger than 30 µm, there is little overlap between the PVM and the 2-D-C, and so we refer to the 2-D-C measured liquid water as the drizzle liquid water mixing ratio q D in this study.

17921
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 3 Mesoscale structure Figure 5 shows radar reflectivity from the straight and level and sawtooth legs in RF06 as a function of distance along the run. A zone of strong echoes was frequently present in the transition zone between the overcast stratocumulus and the more broken cloud 5 associated with the POC. This region of strong precipitation represents the transition zone between overcast and POC, and we term the mesoscale cloud structure associated with this the boundary cell. The POC-ward extent of the boundary cell we term the leading edge and the location where the overcast stratocumulus clouds begin we refer to as the back edge. The leading edge was diagnosed subjectively based upon 10 the WCR imagery. The images in Fig. 5 are aligned so that the leading edge is defined as being zero distance. The location of a back edge is harder to define since for four of the runs the boundary cell is not all that clearly separated from the echoes associated with the overcast stratocumulus (e.g. runs SC1, CB, SC2, C2). For other legs (AC, C1 and S), there is a clear gap between the strongly precipitating echo and the overcast 15 stratocumulus. Figure 5 denotes the location of the back edge by a vertical dashed line. We refer to the region between the leading and back edge as the transition region separating the POC from the overcast stratocumulus. The width of the transition region ranges from 16 to 43 km (see also Table 1). The location of the transition region moved roughly consistently with the mean wind in the MBL on RF06, and the edge observed 20 with the satellite although a conclusive assessment of this is not possible with the flight strategy used.

Delineation of POC, overcast and transition regions
The key features contrasting similarities and differences in the MBL, cloud, precipitation and aerosol structure between the POC and overcast regions are summarized in Table 2. The following sections of the paper constitute a presentation of the obser-Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

Overcast region
Since the column maximum reflectivity Z max is typically found near the cloud base , Z max is a good indicator of the precipitation rate near cloud base R CB . Almost all of the overcast region can be characterized by columns with maximum reflectivity Z max greater than −20 dBZ (Fig. 5). Remarkably, 92% of columns 5 in the overcast region sampled by the WCR have Z max >−15 dBZ (Fig. 6), which is the threshold commonly assumed for the presence of significant drizzle (−15 dBZ corresponds to roughly 0.1 mm d −1 precipitation rate according to Comstock et al., 2004).
For roughly a quarter of all columns in the overcast region Z max >0 dBZ, corresponding to cloud base precipitation rates R CB of ∼2 mm d −1 (see upper axis in Fig. 6), and 2% 10 of the columns have Z max as high as 10 dBZ. It is clear, therefore, that the overcast stratocumulus clouds surrounding the POC are producing substantial drizzle. Figure 5 also shows that the precipitation within the overcast region is not uniform but is organized into cellular structures containing core regions with high Z max surrounded by regions with much lower Z max . This cellularity in marine stratocumulus is consis-15 tent with the closed mesoscale cells seen in the satellite imagery ( Fig. 1), which have a dominant horizontal scale of some 30-40 km, a typical value for stratocumulus over the Southeast Pacific (Wood and Hartmann, 2006).

POC region
The POC region is characterized by lower cloud cover and the POC clouds are orga-20 nized into open cells (e.g., Fig. 1). Only 55-60% of the POC columns sampled contain reflectivities exceeding −30 dBZ (Fig. 6), while the more sensitive WCL detected clouds 60% of the time from subcloud runs in the POC. As in the overcast region, roughly a quarter of the POC has Z max >0 dBZ (R CB > 2 mm d −1 ) but the occurrence of heavy drizzle (Z max >10 dBZ) is three times more likely inside the POC than in the surround-25 ing overcast regions. There are also relatively fewer incidences of weak precipitation (−20<Z max <−10 dBZ) inside the POC. Thus, there is a fundamentally different nature 17923 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | to the precipitation inside the POC compared with the surrounding overcast clouds, with a broader distribution of Z max and locally stronger precipitation.

Transition region
The distribution of Z max within the transition region is shifted to higher Z max values than in the POC and overcast regions (Fig. 6). Around 65% of the transition re-5 gion has Z max >0 dBZ, 30% has Z max >10 dBZ (R CB >10 mm d −1 ), and roughly 5% has Z max >20 dBZ (R CB >70 mm d −1 ), which emphasizes that the drizzle in the transition region can be locally very heavy indeed. The distribution of Z max is not as broad as in the POC region, being closer to the distribution in the overcast region except shifted to higher reflectivities. Inspection of Fig. 5 confirms that the strongest precipitation in 10 most flight legs is associated with boundary cells that are frequently present in the transition region.

Large scale context
Tables 3 for flight B409 and 4 for flight RF06 show a number of key large scale variables, some of which are estimated separately for the POC and overcast regions, which con- between the flights. The sea-surface temperature (SST) is slightly warmer than the air temperature. The near-surface air temperature inside the POC is cooler than in the overcast region 20 (by 0.3 K on B409 and 0.9 K on RF06), and some of this difference can be explained by cooler SSTs in the POC (see Table 4). Surface estimated sensible heat fluxes (SHF) are very small (<15 W m

25
The primary driver of turbulence in the MBL is the longwave radiative flux divergence across the MBL, which averages 91 W m −2 in the overcast stratocumulus in RF06 (Table 4), a particularly high value due to low values of downwelling flux (F ↓) above the MBL caused by the very dry free-troposphere (q<1 g kg −1 on both B409 and RF06).
Despite the visual impression of a lack of cloud in the POC region (e.g., Fig. 1), the 5 actual cloud cover is ∼60% (Sect. 3.3 above) on RF06 leading to 72 W m −2 of radiative flux divergence across the MBL in the POC region. However, much of the extensive cloud in the POC consists significantly lower liquid water path (see Sect. 5.3 below), and contains very large droplets. Therefore the optical thickness and emissivity of the extensive POC clouds will be lower than that in the overcast region, so that longwave 10 radiative flux divergence will be distributed over a thicker layer and may therefore be less effective at driving buoyant circulations in the POC. This behavior has been found in large-eddy simulation studies for clouds with similar characteristics to those in the POC studied here (Ackerman et al., 1993). The most marked difference between the two flights is the MBL depth which in- 15 creases from 1280 m on B409 to 1375 m on RF06. The MBL is capped with a very strong inversion throughout. The difference in potential temperature between 200 m above the inversion and the inversion base is ∼12-13 K in the POC and ∼15 K in the overcast regions (Tables 4 and 3). Visually, the cloud top in the overcast region was quite flat. From the various inversion-crossings, no significant difference in MBL depth 20 was detected between the overcast and POC regions on either flight. In both regions the mean cloud top height is located very close to the inversion base. The MBL is under the influence of large scale divergence, at least in the daily mean (Table 4). It is interesting that this is the case despite negative d v/d y (the strongly positive d u/dx is double the magnitude of the negative d v/d y). Since the POC boundary

Conserved variables
Profiles of liquid potential temperature (θ l ) and total water content (q t , the sum of the vapor (dewpoint hygrometer), cloud liquid (PVM/CDP), and drizzle liquid (2-DC/2-DS) water mixing ratios) from both leg-mean data and representative individual profiles are 5 shown in Fig. 7. On both flights the overcast region MBL is remarkably well-mixed given the significant drizzle that the clouds are producing. The remarkably strong inversion (∼15 K inversion jump in θ l ) is clearly evident. The inversion jump in the POC region is closer to ∼12 K, in the mean, although the RF06 POC profile happens to be through a cumulus cell and so is warmer than the mean conditions. 10 In the mean, on RF06 the POC region is more strongly stratified than the overcast region ( Fig. 7), both in θ l and in q t , and the transition region even more so, with strong cold/moist pooling of air near the surface. Liquid potential temperature in the upper MBL in the POC exceeds that in the overcast region by 1.5 K, but is colder by roughly the same amount near the surface. The mean total water content in the upper MBL is 15 greater in the overcast region by some 0.3-0.6 g kg −1 , but near the surface, the POC is moister than the overcast region by close to 1 g kg −1 . The POC region sampled on B409 is actually slightly drier than the overcast region ( Fig. 7) Approximately two thirds of the way up the MBL θ l and q t are roughly the same in the overcast and POC. It is interesting that the MBL mean value of θ l is slightly lower and q t slightly higher inside 20 the POC than outside. This would be consistent with weaker entrainment inside the POC than outside, although differences in radiative heating or surface fluxes over the airmass history preclude definitive attribution. Data taken on flight B409 12 h prior to the RF06 measurements show a very similar structure with a well-mixed boundary layer and very similar mean values of θ l and q t (Table 3 and Fig. 7), but with an inversion base that is approximately 100 m lower than that observed during RF06. The latter finding is consistent with entrainment deepening of the MBL over the period between the two flights (see Sect. 7 below).
Tracers with higher concentrations in the free-troposphere than in the MBL such as carbon monoxide (CO) and ozone are less abundant in the POC MBL than in the overcast MBL. For example, on RF06 the CO concentration is 0.5-2 ppb lower inside the POC, and the ozone concentration 2-4 ppb lower. Similar overcast/POC differences in CO (1.8 ppb) and ozone (1.6 ppb) are found for the subcloud layer leg on flight B409, 5 implying longevity in the horizontal gradients and structure across the POC-overcast boundary. Since the dry deposition rate of CO is small, and the free-tropospheric values measured in RF06 were approximately equal above the POC and the overcast regions, the observed MBL differences could represent integrated differences in entrainment rate. However, since we do not know either the history of the air entrained 10 into the MBL prior to RF06, or the concentrations in the MBL before the POC formed, this cannot be concluded with certainty. The same is the case for ozone, which has the added complication of significant dry deposition and short timescale photochemical sources/sinks. 15 Profiles of cloud (q L ) and drizzle (q D ) liquid water mixing ratio from the straight and level runs on RF06 are shown in Fig. 8, and profiles of precipitation rate (2-D-C plus radar-estimated), relative humidity, drizzle drop size and number are shown in Fig. 9. In the cloud layer of the overcast region, q L significantly exceeds q D , q L increases upward, and q D downwards in the cloud. This is typical behavior for overcast drizzling 20 stratocumulus (Wood, 2005). By 300-400 m below cloud base, little drizzle water remains in the overcast region. Since the cloudbase precipitation rates (Fig. 9a) in the overcast region are substantial at 2-4 mm d −1 , this indicates strong evaporation in the layer immediately below cloud base.

Condensate and drizzle on RF06
In the POC and transition regions, as inferred from cloud-level flight legs, q D >q L 25 ( Fig. 8) indicating that the majority of the condensate is present as precipitation rather than cloud water. Averaged over the two cloud layer legs, the ratio q D /q L is approximately 3 in the POC region and 4 in the transition region (in contrast with the overcast 17927 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | region where it is only 0.25). Therefore, a marked transition from cloud to drizzle condensate appears to be a signature of the transition from closed to open cellular convection. The mean total condensate mixing ratio in the cloud layer of the POC region is significantly lower than that in the overcast region when weighted by the fraction of samples that are cloudy (35% of POC samples from the two cloud legs are cloudy, in 5 contrast to 92% of those in the overcast region), but the in-cloud means are approximately equal. The transition region observed on RF06 contains the heaviest precipitation ( Fig. 9a), where mean values are 10-20 mm d −1 at all levels. In this region there is also a substantial quantity of drizzle liquid water present close to the surface (Fig. 8b), with some-10 what less in the POC and almost none in the the overcast region. This largely reflects the lower cloud base and high relative humidity (Fig. 9b) both of which hinder drizzle evaporation. Interestingly, the drizzle drop volume radius r v,D is largely independent of region but is a strong function of height ( Fig. 9c), so that the mean size of the drizzle drops leaving cloud base increases from the overcast (∼80 µm) to POC (∼100 µm) to transition (∼140 µm) region. The strong height scaling of r v,D independent of region suggests that the mean size of drizzle drops is largely a function of the depth of cloud that produces them rather than other factors. One consequence of this is that since the depth of the layer that a drizzle drop can fall before evaporating scales strongly with r v,D , it is not surprising that the fraction of the cloudbase precipi-20 tation rate reaching the surface increases strongly from the overcast (<10% reaching surface) to the POC (∼25% reaching surface), and is higher still in the transition region where most of the drizzle reaches the surface (Fig. 9a).
A striking feature, observed with both in-situ and radar data, is that the mean precipitation rate at the cloud level in the overcast region is significant (3-4 mm d −1 ) and 25 about three quarters of that in the POC (4-5 mm d −1 ). Thus, the thinking that POCs delineate regions in marine stratocumulus with locally enhanced precipitation ) may need to be revised. Examination of MODIS-retrieved liquid water path ( Fig. 10) from 4 h after RF06 shows that the POC feature sampled is itself embedded in a broader swath of thick overcast stratocumulus with peak values of liquid water path at the center of the mesoscale cells exceeding 400 g m −2 and a mean LWP of 150-200 g m −2 . Our observations from RF06 are indicating that this surrounding cloud contains significant cloud base precipitation. Since elevated LWP is known to be associated with enhanced precipitation in marine stratocumulus 5 Wood, 2005;Kubar et al., 2009, and see also Sect. 5.4 below) it seems reasonable to conclude that the high LWP swath, rather than simply the POC, may delineate regions of locally enhanced precipitation and that this broad region of precipitation supports the formation of the POC. This is consistent with existing satellite analysis over the southeast Pacific (Wood et al., 2008) showing that open cellular convection is positively correlated with elevated LWP in overcast clouds surrounding the open cells.
Since r v,D is similar in all regions at a given height, variations in precipitation rate at a given height level are more strongly reflective of variations in the drizzle drop concentration N D (Fig. 9d). This is consistent with previous aircraft observations (e.g.,  and with observed exponents relatively close to unity in the 15 relationship between reflectivity and rain rate . and only a few instances where LWP exceeds 500 g m −2 . The POC LWP distribution has a mode at close to zero, a mean of 210 g m −2 , and is broader and more positively skewed, with some 15% of POC clouds with LWP exceeding 500 g m −2 . The transition region has the largest mean LWP (510 g m −2 ) with a broad distribution and little skewness. with the largest values of the column maximum reflectivity Z max , and thus are the most strongly precipitating cells.

Cloud vertical structure on RF06
Cloud base height z cb distributions (Fig. 11c) are much broader than those in cloud top in all three regions. This result is in accordance with previous studies of mesoscale variations of overcast stratocumulus (Wood and Taylor, 2001). The narrowest distribu-10 tion is in the overcast region. In the POC region there is a bimodal distribution of z cb , with the lowest bases corresponding to cumuli which grow from the top of the surface mixed layer, and the highest bases consistent with stratocumulus and thin stratus aloft. The cloud base height in the transition region is unimodal and somewhat higher than the cumulus bases in the POC. 15 Further insight into the vertical structure of the clouds in the MBL across the POCovercast boundary is provided in Fig. 12 which shows zenith-viewing radar and lidar backscatter for the entirety of the subcloud run SC2, together with liquid water path observations from the microwave radiometer. Radar-derived cloud top height and maximum radar reflectivity are overlaid on the lidar backscatter which indicates a remark-20 able degree of agreement between the height of the cloud base z b (delineated by strong lidar backscatter in regions with moderate or no drizzle) and the height of the maximum radar reflectivity z Zmax . The coincidence of the z b and z Zmax is to be expected because drizzle drops grow by collection monotonically downward in the cloud but begin to evaporate as soon as they fall into the subsaturated region below cloud 25 base Wood, 2005), but the agreement is striking nonetheless. Further, this agreement suggests that z Zmax may be used instead of a lidar-derived z b in cases where the drizzle is sufficiently heavy to cause substantial lidar backscatter thus making it difficult to objectively determine a cloud base (for example, throughout much of the transition region in Fig. 12). Figure 12 further exemplifies the narrow distribution of LWP in the overcast region, the high values in the transition region, and the intermittent distribution in the POC. What is particularly interesting is that in all regions and for all observed values of LWP, the clouds appear to be well-modeled with an adiabatic assumption. This is 5 true whether the model uses the lidar-derived or, in strong drizzle, the maximum radar reflectivity-derived cloud base height. That even very strongly drizzling clouds, in which much of the condensate is in the form of drizzle (Fig. 8), retain an adiabatic liquid water path presumably indicates that the replenishment time for liquid water path through cloud updrafts is sufficiently strong to overcome the loss of liquid water to precipitation 10 (see e.g., the simple adiabaticity model in Wood, 2005). In addition, the vertical motions may also prevent loss to drizzle by keeping drizzle drops lofted, thus hindering fallout. However, since there are substantial lateral mesoscale motions, regions of high liquid water content could be advected into regions with elevated cloud base, leading to superadiabatic liquid water contents as deduced from a one-dimensional perspective. 15

Factors controlling precipitation
The clouds with the strongest radar echoes correspond to those with the greatest liquid water path (Fig. 12), emphasizing the key role that the liquid water path plays in driving precipitation formation in these clouds. Figure 13 builds on this by showing the relationship between cloud liquid water path 20 (LWP) and the median column-maximum radar reflectivity Z max for clouds in the POC, overcast, and transition regions. In general, the strongest precipitation is associated with the clouds with the highest LWP, which the analysis of the previous section indicates are the thickest clouds. This result is not particularly surprising in the light of previous measurements indicating that LWP/cloud thickness is a strong driver of pre-25 cipitation in marine stratocumulus (see review by Brenguier and Wood, 2009). What is interesting is that in the POC and transition regions the relationship between LWP and Z max diverges from that in the overcast, but that the divergence only occurs for 17931 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | LWP values lower than ∼400 g m −2 . This indicates that for Z max greater than 5 dBZ (equivalent to a cloud base precipitation rate of a few mm d −1 ), the precipitation rate appears to be explained primarily by LWP. There is little evidence that the significant differences in N d (see Sect. 6) between POC and overcast regions have any bearing on the ability of the thicker clouds (400 g m −2 adiabatic LWP is equivalent to a 600 m 5 thick cloud) to precipitate. On the other hand, lighter precipitation rates typical of thinner clouds cannot be explained by variability in LWP alone, which leaves open a role for the lower N d values in the POC to play a role in enhancing the precipitation efficiency compared with the overcast clouds. These results are broadly consistent with satellite observations, large eddy, and heuristic drizzle models (Feingold and Siebert, 10 2009;Kubar et al., 2009;Wood et al., 2009b;Sorooshian et al., 2009).

Cloud microphysics and aerosols
There are striking cloud microphysical and aerosol differences between the POC and the overcast regions that are as dramatic as those in the cloud and precipitation macrostructure. Table 5 provides cloud droplet concentrations, cloud droplet effective 15 radii, and mean clear-air aerosol for selected levels in the POC, the transition region, and the overcast region on flight RF06.

Cloud droplet concentration and effective radius
Mean cloud droplet concentration N d for clouds in the cloud layer of the overcast region is approximately 70 cm −3 , while in the POC it is an order of magnitude lower. Similar 20 gradients were observed on the B409 flight 12 h earlier, implying that the microphysical gradients can be maintained over significant periods of time. This is a remarkable contrast, and results in strong differences in the cloud effective radii r e . If drizzle drops are included in the r e estimates ( Table 5) the differences are even more striking. While drizzle drops can contribute significantly to r e (Wood, 2000), the magnitude of the con-tribution in the POC and the transition regions is a doubling of r e (Table 5). Figure 14 shows the MODIS-derived r e near cloud top, showing values of 12-25 µm in the overcast region to the north of the POC where the aircraft sampled overcast clouds roughly 4 h earlier. These values are somewhat larger than those measured on the cloud legs in the overcast region, but it must be borne in mind that the cloud legs were quite close 5 to the base of the cloud and that in well-mixed overcast stratocumulus clouds, the effective radius increases with height (e.g., Martin et al., 1994). Given the cloud top and base heights in the overcast region, and since the impact of drizzle near cloud top is relatively small (Wood, 2005), we estimate that the cloud top effective radius should be roughly 30-40% greater than that for the cloud droplets at the height of the cloud legs, 10 i.e. 13-14.5 µm or so. This is somewhat lower than the mean MODIS-derived value in the overcast region (17 µm), consistent with previous comparisons (e.g., Nakajima et al., 1991). In the POC region MODIS retrievals are not performed in most cases and so comparison with in-situ data is not meaningful. 15 Aerosol size distributions for different heights in the POC and overcast regions from RF06 using the in-cabin University of Hawaii system are shown in Fig. 15. Both individual and mean spectra are shown to give a sense of the variability of the distribution. Aerosol distributions from B409 and RF06 in the subcloud region of the POC and overcast regions are shown in Fig. 16 and indicate dramatic differences in the aerosol 20 properties between the two regions. We describe particular characteristics of the two primary submicron modes in the following subsections.

Accumulation mode particles
In the subcloud layer, the concentration N a of particles with diameters 0.12-3.12 µm (measured with the PCASP during RF06) is three times higher in the overcast re-  Table 5). The size distributions confirm this strong reduction in accumulation mode aerosol in the subcloud layer of the POC (Fig. 15d,e). A remarkably similar contrast between the accumulation mode of the aerosol distribution in the POC and the overcast regions is also observed on flight B409 approximately 12 h earlier (Fig. 16). There is good evidence It is interesting that although there are considerably lower accumulation mode aerosol concentrations in the POC subcloud layer, the lidar returns from these aerosols are significantly stronger (by a factor of 2 or 3) than those in the overcast region (see 20 Fig. 20a). This may reflect the higher relative humidity there (Figs. 9b and 20c), but may also reflect changes in the concentration of supermicron sized particles. Given that near-surface wind speed inside and outside the POC are almost identical (Table 4), the surface source would be expected to be quite similar. However, the decoupled nature of the POC region may mean that the surface-produced (mainly supermicron) particles 25 are mixed over a shallower depth than in the overcast, more well-mixed region. This issue warrants further exploration in future studies.
It is interesting that despite considerable coalescence scavenging that would be expected from the drizzle rates observed in the overcast region, the concentrations of aerosols and the size distribution measured in the overcast region subcloud layer on B409 and RF06 are very similar, indicating little time change in CCN between flights. Based on the analysis of Wood (2006), and drizzle rates observed on the flights, we would expect coalescence scavenging to remove ∼50 cm −3 CCN over a 12 h period. That we do not see decreases of this magnitude indicates that CCN must be being re-5 plenished, either through a surface source or by entrainment from the free-troposphere. Our estimates of entrainment rate (see Sect. 7 below) together with the concentrations of potential CCN in the FT (which are substantially lower on B409 than on RF06, not shown), indicate that entrainment is probably provides an insufficient supply, and that a significant surface CCN source may be necessary. 10 Above the subcloud layer, while N a stays roughly constant in the overcast region (i.e. determined on the CB leg below cloud) consistent with the well-mixed nature of the MBL there, N a drops dramatically with height in the POC, reaching minimum values roughly 125-275 m below the inversion base (roughly 1100-1250 m altitude) of less than 1 cm −3 (Table 5 and Fig. 15c). Data from seven separate profiles within the POC  (Table 5). Figure 17 shows aerosol concentrations for different size classes from a representative profile in the POC region, which highlights the ultraclean layer that exists within and near the top of the cumulus-coupled layer. Within the 20 ultra-clean layer there has been a near-complete removal of the accumulation mode.
Since concentrations in the lower free-troposphere are substantially higher than those anywhere in the POC MBL, entrainment cannot account for the low N a in the ultraclean layer, and so it is reasonable to conclude that the aerosols that would have been transported from the subcloud layer and detrained into this layer by cumulus have been 25 almost completely scavenged and removed by precipitation. Indeed, the low mean N a in the ultra-clean layer implies that entrainment is a particularly inefficient process in the POC for buffering MBL aerosols with FT aerosols. In the overcast region, however, the MBL is more well-mixed, with N a in the FT lower than that in the MBL. However,

17935
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | the total aerosol concentrations in the free-troposphere exceed those in the MBL, and the size distributions (Fig. 15b,e) are consistent with a free-tropospheric source of new particles that then grow by aqueous phase processing.

Small aerosol particles
Total aerosol concentrations N CN (Table 5) in the subcloud layer of the POC in 5 RF06 are marginally greater (N CN =150 cm −3 ) than those in the overcast region (N CN =140 cm −3 ), which implies that the concentration of Aitken mode particles inside the POC (∼120 cm −3 ) is almost three times the concentration of such particles outside the POC. This is clearly shown as a mode at ∼20 nm diameter in the size distributions (Fig. 15d,e). This suggests either a source of new particles inside the POC that is not 10 present in the overcast, or a sink of small particles in the overcast region that is not occurring in the POC. That there is a distinct mode and not a monotonically increasing concentration with decreasing size indicates that any new particle formation is not taking place at the time of either the B409 or RF06 measurements, and may have occurred some time earlier. Concentrations in the 20 nm mode were significantly higher in the POC 12 h earlier (compare B409 and RF06 spectra in Fig. 16), but the mode shapes are very similar. This is consistent with the removal of the 20 nm mode over time by Brownian coagulation on cloud droplets, perhaps following a daytime nucleation event prior to B409. Figure 15a,b shows that it is unlikely, although not impossible, that the free-troposphere could be a significant source of the small particles, since the concen-20 trations there are typically smaller than those in the subcloud layer in the POC. Data from the B409 flight, where concentrations of aerosols in the FT were particularly low (not shown), also suggest that the FT is an unlikely source for the 20 nm mode particles in the POC.
The dominant sinks of small particles are Brownian diffusion onto existing particles 25 or coalescence scavenging (Wood, 2006). Because there is a good relationship between the droplet concentration in updrafts and the subcloud PCASP concentration (not shown) it appears that the population of particles smaller than the PCASP range largely remains unactivated. It is thus likely that coalescence scavenging is a relatively unimportant sink of sub-PCASP sized particles in either region. Brownian diffusion losses on the larger overall surface area in the overcast region depends extremely strongly upon the size of the small particles. Aerosols with radii of 20 nm and greater 5 have Brownian scavenging efficiencies that are much lower than those of freshly nucleated aerosol (Seinfeld and Pandis, 1996), but it is not possible to entirely rule out Brownian scavenging by cloud droplets as a cause for removal of the small particles. Brownian scavenging would certainly be a stronger sink in the overcast region where the total available cloud droplet surface area is significantly greater than that in the 10

POC.
Additional evidence regarding the reasons for the larger CN concentration inside the POC can be found in the non-refractory (volatilizable) fraction of total CN, estimated as f non-ref =(N CN −N CN,hot )/N CN ( Table 5). Values of f non-ref in the POC are 0.5-0.7, much greater than the 0.25-0.3 found in the overcast. Since the CN concentrations in 15 the layers below cloud in the POC and overcast are similar, this means that there is a substantially greater number of non-refractory particles in the POC than outside the POC. It seems to be difficult to account for these differences as a result of either particle losses or entrainment, and so we hypothesize that the enhanced CN in the POC are a result of new particle formation in the POC. Since the ultrafine (3-10 nm) concen-20 trations in the POC were not strongly elevated, it is suggested that the particles formed during the previous day, with precursor build up perhaps requiring photochemicallyproduced oxidants which may have been subsequently depleted by deposition onto the new particles. Elevated CN concentrations were found during a POC passage measured during a ship cruise to the same region (Wood et al., 2008), suggesting that 25 this case is not unique, and evidence of new particle formation and size growth from the same passage was also presented in Tomlinson et al. (2007).
It is puzzling why the mode at 20 nm diameter is more dominant in the subcloud run than in the cloud layer (Fig. 15), since one would expect stronger nucleation to 17937 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | have occurred in the ultraclean layer where aerosol surface area is very low. However, if these particles were nucleated hours earlier, a stronger sink from coagulation onto cloud droplets may have depleted concentrations there. There is also considerable variability in the concentrations around 20 nm in the subcloud layer (Fig. 15d), so uncertainties associated with sampling a highly heterogeneous field must be borne in 5 mind. We note that locally high CN concentrations (∼300 cm −3 ) are actually observed in the POC on the C1 leg even though the mean value in this layer (Table 5) is considerably lower. Since precipitation events are highly intermittent, the scavenging would also be expected to be patchy. to approximately 50 cm −3 , with the highest values in active cumulus updrafts, and (although to a lesser extent) in the compensating downdrafts of the cumuli (Fig. 19), and the lowest values in air with near-zero vertical motion. The cumuli are often quite broad (few km) in horizontal extent and essentially constitute the bright-  Table 5). This result confirms that the subcloud layer represents the source of the air in the cumuli 25 in the POC. The subcloud layer PCASP concentrations are much more homogeneous in space than those in the upper two-thirds of the MBL (>500 m), which indicates that the upper MBL air that is strongly depleted of aerosols mixes down to the subcloud layer very slowly. This seems consistent with recent large eddy simulations (Jonker et al., 2008) which show that most of the compensating downward motion in cumulus fields is not in the form of slow subsidence in the clear regions between clouds, but is confined primarily to relatively strong but localized downdrafts close to the cumuli. Since we observe relatively high N d values in the strong compensating cumulus down-5 drafts (Fig. 19), this suggests that the extremely low mean values of N a (and N d ) in the ultra-clean layer may be less closely linked to the active cumuli than one might suppose given the strong association of much of the POC precipitation with the active cumuli (Fig. 19b). Perhaps there is another loss mechanism involving the widespread and quiescent clouds, including the thin stratus, that does not depend upon coales-10 cence scavenging in the strongly-precipitating cumuli. This might be a slow and steady but widespread sedimentation of large cloud and small drizzle drops in the quiescent clouds that results in the formation of the ultra-clean layer near the top of the MBL in the POC.

Variability of cloud and aerosol microphysics
Another interesting feature seen in Fig. 18 is the dramatic decline in N d which begins 15 in earnest approximately 20-30 km from the transition region. At the back edge of the transition region N d is only a few drops cm −3 even though the subcloud layer N a in this region does not fall below about 30 cm −3 (although N a does show a decrease towards the transition region). The low N d may indicate signficant coalescence scavenging by drizzle in this region. This region of decreasing N d is also one in which CO concentra-20 tions drop from values typical of the overcast region to values typical of the POC (not shown). This suggests that, in the overcast region, as the transition is approached, more and more air from the POC is present at the cloud level. One hypothesis is that this region comprises air detrained from the boundary cell, which is itself a mixture of subcloud air from the POC and the overcast regions. Model simulations should shed 25 considerably more light on the mesoscale dynamics, but these are beyond the scope of this study. Figure 20a shows remarkable variability in the lidar backscatter due to aerosols in the POC, with very little scattering above the subcloud layer, and considerable variability 17939 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | within it, with curiously strong returns in the POC region immediately adjacent (e.g. −20 to 0 km from the leading edge) to the boundary cell, and very weak returns on the overcast side (e.g. 25-35 km from the leading edge). On SC1 at least, N a is actually higher here than on overcast side of the boundary cell (Fig. 18), consistent with the aerosol scattering differences being driven not only by relative humidity differences but 5 by significant changes in aerosol concentrations.

Turbulence and entrainment rate estimates combining the two flights
Using the 12 h time difference between B409 and the middle of the sampling on RF06, we estimate the mean entrainment rate using three different budgets: (a) mass; (b) energy; (c) moisture. The basic approach for this is outlined in Caldwell et al. (2005), 10 but here we apply it to the budgets between the two flights. Since a Lagrangian sampling strategy was used, we assume that the advective terms are zero. Large-eddy simulation results (Berner and Bretherton, 2010) show that there is considerable lateral exchange between the overcast and POC regions, and so we choose to estimate a single entrainment rate for the POC and overcast regions together. The mass budget 15 uses the observed increase in MBL depth between B409 and RF06, together with the subsidence rate (Table 4), to estimate the rate at which the MBL must have entrained over the 12 h period between the two flights. The MBL depth was found to increase from 1285 m to 1375 m over this period, and the subsidence rate over this period is estimated to be 0.1-0.2 cm s −1 (note that the value given in Table 4 is a diurnal mean), 20 resulting in an entrainment rate of w e =0.3-0.5 cm s −1 . The chief source of error is in the subsidence rate which has a significant diurnal cycle in the SEP region that is still not fully understood (Wood et al., 2009a). The energy budget requires assessment of the infrared radiative flux divergence over the MBL, the surface precipitation rate, and the surface sensible heat flux, for which we 25 use the measurements in RF06 since there was more extensive sampling. From Table 4 the LW flux divergence is −78 W m −2 and the sensible heat flux is 12 W m −2 .
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | We estimate that the mean surface precipitation rate is approximately 1 mm d −1 , or 28 W m −2 in energy terms. The storage term is estimated to be slightly negative (−5 W m −2 ). The jump in liquid potential temperature across the inversion is approximately 12 K. The largest source of error is likely to be the surface precipitation rate estimate, and the storage term, which combined may result in an uncertainty in the 5 entrainment rate of a factor of two, i.e. w e =0.2-0.4 cm s −1 . The water budget requires estimates of the surface precipitation rate, the latent heat flux at the surface. As with the energy budget the moisture storage term was found to be close to zero. From Table 4 the latent heat flux is some 140 W m −2 , and the jump in total water across the inversion approximately 7.5 g kg −1 , from which we esti-10 mate w e =0.45-0.7 cm s −1 , with the largest source of uncertainty being the storage term (0.5 g kg −1 uncertainty in the MBL mean temperature change between flights leads to a 40% change in the derived w e . Given the three estimates and their uncertainties, our best estimate of the mean entrainment rate for the POC, overcast and transition regions between the two flights 15 is 0.45±0.1 cm s −1 . This is consistent with mean nighttime entrainment rates of 0.4-0.6 cm s −1 estimated from the EPIC Stratocumulus Cruise in 2001 (Caldwell et al., 2005), and suggests that the POC-overcast boundary region is actively entraining despite the strong cloudbase drizzle rates. These estimates should prove useful for testing model simulations of the POC/overcast boundary. In general, modeling studies give 20 conflicting information on the degree to which strong drizzle suppresses entrainment in stratocumulus, with Stevens et al. (1998) suggesting a marked impact, consistent with theoretical ideas for mixed layers (Wood, 2007), while the more recent intercomparison of large eddy models by Ackerman et al. (2009) finds that drizzle does not have a marked effect on entrainment.

25
Since there are no significant differences in the MBL depth between the POC and overcast regions on either of the flights, this implies either that the entrainment in the POC is the same as that in the overcast region, or that there is a different subsidence rate in the POC and the overcast regions. The vertical velocity variance measured in 17941 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | the overcast cloud layer is almost double that in the POC (0.60 m 2 s −2 in the overcast vs. 0.32 m 2 s −2 in the POC on RF06), while the mean inversion strength is only 10-20% greater in the overcast region. Since cloud top entrainment rates in stratocumulus scale with the ratio of the two, we would expect that the stronger turbulence in the overcast region would lead to substantially stronger entrainment there. That this does not result 5 in a more rapidly growing MBL in the overcast region suggests that the subsidence rate is weaker in the POC region than in the overcast region. This is unlikely to be caused by the large scale divergence which would be expected to vary only weakly in space, and so it suggests that there may be a compensating circulation between the overcast and POC regions which effectively "holds-up" the MBL in the POC despite 10 weaker entrainment. Model simulations of this case (Berner and Bretherton, 2010) do indeed show differences in entrainment rate across the boundary and yet an inversion that rises in concert.

Discussion and conceptual model
A conceptual diagram of the POC-transition-overcast boundary layer is shown in 15 Fig. 21. In-situ and radar doppler wind data (Fig. 23) indicate significant mid-level inflow (±2 m s −1 ) into the boundary cells in the transition region, consistent with the typical behavior observed in strongly precipitating stratocumulus cells in the southeast Pacific (Comstock et al., 2007). Outflow is observed in radar winds at the cloud level which then detrains air into the surrounding POC and overcast regions. Evaporation 20 of strong precipitation with rates of 10-20 mm d −1 below cloudbase appears to lead to the strong cold/moist pool below the boundary cell (see Fig. 20c and is also evident in the doppler winds in Fig. 23), which drives outflow from the transition region at low levels (see Fig. 20e), although we note that significant variability was found from leg SC1 to SC2. The cold pool in SC1 is 20 km in extent with >1 K temperature depression, a strongly-precipitating boundary cell is not present near to the stratocumulus edge. When the cold/moist pool is evident, the air in it has elevated equivalent potential temperature θ e (Fig. 20d) since the humidity increase dominates over the suppression in temperature. Similarly elevated θ e was found under strongly drizzling open cells in the northeastern Pacific , and a similar inferences can be made 5 from drizzling data from the southeastern Pacific (Comstock et al., 2005), although Paluch and Lenschow (1991) show a case of drizzling stratocumulus in which fluctuations associated with evaporating precipitation were moist and cool by an amount that almost canceled in θ e . Smaller and weaker cold pools are also evident in the POC region and the overcast region in this study (Fig. 20d), and these too have elevated θ e . 10 Thick clouds in the POC consist of both broad active (turbulent) cumulus clouds containing one or more updraft cores, together with more quiescent cells that may be the decaying remnants of earlier active cells. Figure 22 shows examples of these two types of cells, both of which contain significant drizzle. The strongest echoes appear to be found in the more active cells, although quantifying this is difficult given the limited 15 sample size (visual inspection of Fig. 5 demonstrates how few of the cells are sampled in total). Despite the stronger echoes in the active cells, the precipitation rates appear to be equally high in the quiescent cells, reflecting higher drizzle water masses in these cells. Cloud water content and droplet concentrations are very low in the quiescent cells, so this may represent a more mature stage of the conversion of cloud to drizzle 20 via coalescence, and may also be responsible for much of the accumulation mode aerosol depletion in the ultra-clean layer. However, it should be noted that the largest drizzle drops (volume radii in excess of 100 µm) of all in the cells shown in Figure 22 are actually found in the updraft core of the active cell. Here, drizzle drops may be growing particularly large by being suspended in the upward moving cloud water and 25 growing rapidly via coalescence. Such drops would be expected to fall quite rapidly once the updraft ceases, whereas the drizzle drops in the quiescent cell are smaller and would take longer to reach the ground.
An earlier study of stratocumulus drizzle cells in the same region (Comstock et al., 17943 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 2005) found that typical lifetimes of large precipitating cells are around two hours. What seems remarkable, if the quiescent cells are indeed decaying cells, is that they can still contain large amounts of condensate. The total condensate in the quiescent cell in Figure 22 is approximately the same as that in the active cell, although much of it is in the form of drizzle drops. The liquid water path is estimated from adiabatic considerations 5 (Sect. 5.3) to be ∼500 g m −2 for this cell, while the precipitation rate measured from leg C1 (Fig. 22d) is approximately 30 mm d −1 , which would lead to a rainout timescale of 20-30 min, consistent with estimates from Comstock et al. (2005). Thus, without significant updraft activity to replenish the cell, it seems reasonable to interpret the quiescent cell as a decaying remnant of an earlier active cell. It is possible that the ultra-clean 10 layer is partly the result of hydrometeors in the decaying cells slowly sedimenting in quiescent air. The optically-thin stratus in the POC may be remnants of cells that take some time to be completely removed by sedimentation. Another possibility for the quiescent cells is that they represent a marine boundary layer cloud analogue to the trailing stratiform regions found in deep mesoscale con- 15 vective systems (MCSs). Since the stratiform areas in MCSs form partly as a result of weakening and merging active convective cells (Houze Jr., 2004), this interpretation may not be at odds with the decaying remnant idea above. However, stratiform regions in MCSs are also supported by mesoscale ascent, and there is evidence for that in the quiescent cell in the low frequency (smoothed) vertical wind in Fig. 22f, which shows 20 a 3-4 km wide region where the mean vertical wind is 10-30 cm s −1 , which is sufficient to loft 50 µm drizzle particles, similar to the mean volume radius of such drops in the drizzle cell (Fig. 22c). This supports an earlier finding from (Van Zanten and ) that the drizzling regions in a POC are associated with mean vertical ascent. These mesoscale updrafts may thus prolong the timescale over which drizzle can re-25 main in cloud, and potentially the cloud lifetime itself. Cloud resolving modeling will be useful in determining the importance of broad scale ascent in this context. To what extent are the precipitating cells inside the POC different from the boundary cells in the transition zone? Figure 5 demonstrates that peak radar reflectivities in the POC cells can be as high as those in the boundary cells. The primary difference, it appears, is the horizontal extent of the precipitation features in the boundary cells (typically 10-20 km wide) in contrast to those in the POC which have a typical scale of ∼5 km. That said, the boundary cells in some cases appear to be two distinct cells that are sufficiently close that their precipitating cores overlap (e.g. CB run in Fig. 5). As 5 Figure 8 shows, there do not appear to be marked differences in either cloud or drizzle condensate amounts in the transition cells and the POC clouds, and drizzle drop sizes and concentrations (Fig. 9c,d) are similar. So it appears that precipitating cells in the POC and the boundary cell differ largely in their horizontal extent and are otherwise quite similar in nature.

Conclusions
Aircraft flight legs ∼200 km long, from two flights approximately 12 h apart, crossing a spatial transition between a pocket of open cells and the surrounding closed cell stratocumulus, are used in conjunction with satellite data to document aspects of the structure of the cloud, precipitation and aerosol macrostructure and microphysics in 15 the marine boundary layer over the southeast Pacific approximately 1000 km from the Chilean/Peruvian coast. The long flight legs allowed good sampling of both the open and closed cell region, and of the transition between them. This case study was one of several POC studies with a similar sampling strategy by aircraft during VOCALS-REx, and analysis of these cases will provide insight into whether the features observed in 20 this case are ubiquitous POC features or peculiar to this specific case.
Remarkable contrasts in both the macrostructure and the microphysical structure are observed between the open and closed cell regions on both flights. Table 2 serves to describe the key conclusions regarding similarities and differences in the MBL, cloud, precipitation, and aerosol structure between the two regions. In general, these findings 25 echo those of the few existing studies that have sampled POC-overcast transitions Sharon et al., 2006;Comstock et al., 2007), but the 17945 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | comprehensive instrument suite used here is able to quantify these differences with more certainty than before. Because a cloud base precipitation rate of 1 mm d −1 is equivalent to ∼30 W m −2 of cloud layer warming, a particularly pertinent finding here is that persistent drizzle with a magnitude (∼3 mm d −1 ) comparable to the longwave driving (70-90 W m −2 of flux divergence) is present in the overcast region surrounding 5 the POC. Since satellite imagery several hours after the flight shows that the surrounding clouds remained overcast, this suggests that drizzle is not a sufficient condition for POC formation. These results make determining the critical factors responsible for POC formation all the more challenging, but this case is ripe for testing cloud resolving models and using them to explore controls on POC formation. 10 There is significant evidence that coalescence scavenging by precipitation formation in the POC is responsible for the extremely low concentrations of cloud droplets and aerosols found in the upper MBL there. Given that mean precipitation rates do not differ strongly between the overcast and POC regions and that coalescence scavening is essentially controlled by the mean rate (Wood, 2006), it seems quite surprising 15 that commensurate depletion is not observed in the overcast region. Cloud resolving models with interactive aerosols will help to establish the reasons for this behavior.
Acknowledgements. The authors are extremely grateful to the staff, aircrew, and groundcrew of the Research Aviation Facility at the National Center for Atmospheric Research and those of the FAAM, Direct Flight, and Avalon, who operated the two aircraft used in this study. We 20 would like to thank the instrument scientists and operators from the two aircraft who worked tirelessly to provide high-quality datasets used here, and the staff of the University of Wyoming King Air National Facility for supporting the WCR instrument. In addition, we are indebted to all the scientists involved in the VOCALS Program for their guidance, discussion, and insights that have helped improve this manuscript. Graham Feingold,Hailong Wang,Jan Kazil,Chris Terai,25 and Andy Berner are thanked for constructive comments on the results and manuscript. US financial support for the analysis presented here is from the National Science Foundation (Grant 0745702). UK financial support was provided by the National Environment Research Council Grant number NE/F019874/1. path and droplet concentration at the scale of a stratocumulus cloud system, Atmos. Chem. Phys., 8, 4641-4654, doi:10.5194/acp-8-4641-2008Phys., 8, 4641-4654, doi:10.5194/acp-8-4641- , 2008 Houze Jr., R. A.: Mesoscale convective systems, Rev. Geophys., 42, RG4003, doi: 10.102910. /2004RG000150, 2004 Jonker, H. J. J., Heus, T., and Sullivan, P. P.: A refined view of vertical mass transport by  Table 1. Overview of C-130 flight legs, and the earlier BAe-146 flight legs. SC: subcloud leg; CB: leg around cumulus cloud base level; C: leg near the center of the stratocumulus cloud layer; AC: above cloud leg in the free-troposphere above the marine inversion; S: sawtooth run from above inversion to base of cloud. The mean run altitude z is given. Directions are given as from POC to overcast (P→O) or vice-versa. Times are in UTC (LT+3 h). Run lengths L in the three different regions are given in kilometers.  Vertically uniform, values fairly typical for clean marine conditions. Low concentration of Aitken mode particles.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |      . Symbolia are the same as for Fig. 7 Fig. 11. Probability distribution functions from RF06 for (a) liquid water path; (b) cloud top height; (c) cloud base height, for overcast, POC, and transition regions. Cloud base height data are from the WCL. LWP data are from the microwave radiometer data on the two subcloud legs. Cloud top height data are from the WCR on the subcloud, above-cloud, and cloud-base legs. Cloud-level legs were not used to prevent the chance of the cloud top occurring in the radar dead-zone. Cloud base data are from the two subcloud legs. In heavy drizzle the lidar algorithm is unable to detect a clear cloud base, and in these cases we use the height of the maximum radar reflectivity as a proxy for the cloud base. For the POC and transition regions this substitution accounts for roughly 50% of all data points, while for the overcast radar data are used only 5% of the time.    Fig. 15. Aerosol size distributions from RF06 for clear-air samples measured with the University of Hawaii in-cabin sampling system, at different levels in the POC region (a,c,d for the freetroposphere, cloud level, and subcloud layer, respectively) and for the overcast region (b and e for the free-troposphere and subcloud layers, respectively). We show both individual spectra (typically taken over 10 s samples) and the mean spectrum for each level/region, to emphasize variability. No data are available from the cloud level in the overcast region since almost the entire leg was cloudy.