A Lagrangian Analysis of Pockets of Open Cells over the Southeast Pacific

. Pockets of open cells (POCs) have been shown to develop within closed-cell stratocumulus (StCu) and a large body of evidence suggests that the development of POCs result from changes in small-scale processes internal to the boundary layer rather than large-scale forcings. Precipitation is widely viewed as a key process important to POC development and maintenance. In this study, GOES-16 satellite observations are used in conjunction with MERRA-2 winds to track and compare the microphysical and environmental evolution of two populations of closed-cell StCu selected by visual inspection over the 5 southeast Pacific Ocean: one group that transitions to POCs and another control group that does not. The high spatio-temporal resolution of the new GOES-16 data allows for a detailed examination of the temporal evolution of POCs in this region. We find that POCs tend to develop near the coast, last tens of hours, are larger than 104 km2, and often (88% of cases) do not re-close before they exit the StCu deck. Most POCs are observed to form at night and tend to exit the StCu during the day when the StCu is contracting in area. Relative to the control trajectories, POCs have systematically larger effective radii, lower 10 cloud drop number concentrations, comparable conditional in-cloud liquid water path, and a higher frequency of more intense rainfall. Meanwhile, no systematic environmental differences other than boundary-layer height are observed between POC and control trajectories. These results support the consensus view regarding the importance of precipitation on the formation and maintenance of POCs and demonstrate the utility of modern geostationary remote sensing data in evaluating POC lifecycle.

observations of POCs between 2005 and 2018 to analyze POC characteristics. They found POCs have a larger effective radius, lower cloud optical depth, and smaller cloud water path than the surrounding cloud.
A foundational aspect of Wood et al. (2008) was their use of GOES-8 to investigate POC development. By using geosyn-60 chronous observations, they were not limited to instantaneous snapshots of POCs from instruments such as MODIS (e.g. Eastman and Wood, 2016;Watson-Parris et al., 2021). Wood et al. (2008) found that two-thirds of the POC cases identified between September and October 2001 formed in the early morning hours when cloud drop and aerosol concentrations were lowest. However, the cloud microphysical characteristics were not derived from GOES-8; instead, Wood et al. (2008) qualitatively compared the GOES-8 visible, near-infrared, and infrared observations to MODIS-derived retrievals and aircraft 65 observations. Abel et al. (2020) made quantitative use of the Spinning Enhanced Visible and Infrared Imager (Aminou, 2002) onboard the Meteosat Second Generation geosynchronous satellites and MODIS to investigate the influence of biomass burning on POCs. They found that the boundary layer within POCs is ultra-clean even in columns containing aerosols emitted from biomass burning, suggesting that open-cellular convection does not efficiently entrain free-tropospheric aerosols from immediately above the inversion into the boundary layer. 70 In this study, we add a Lagrangian perspective of the full POC lifecycle from satellites by using GOES-16 passive measurements of StCu in the southeast Pacific (SEPAC). GOES-16 makes full-disk observations at 10-minute time intervals with a horizontal resolution between 0.5-2 km. The continuous observations afforded by a geostationary orbit allow for a characterization of POC cloud properties throughout the POC lifetime. Furthermore, the data allow for a comparison of the cloud properties along Lagrangian trajectories that develop into POCs with similar trajectories that remain closed-cell. We use these 75 observations to demonstrate that POCs and closed-cell StCu experience indistinguishable large-scale forcing yet markedly different cloud microphysical properties, thereby supporting the consensus view regarding the role of precipitation in POC formation and maintenance.

80
We identify POCs visually by creating true-color RGBs using 0.47-µm, 0.64-µm, and 0.86-µm reflectance during the day and 10.3-µm -3.9-µm brightness temperature difference (TB 10.3µm˘3.9µm ) images at night using Satpy (Raspaud et al., 2018) from 10-minute observations of GOES-16 ABI top-of-atmosphere solar reflectance and infrared brightness temperatures (IR; Schmit et al., 2017). The 0.64-µm reflectance is sampled at 0.5 km, while 0.47-µm and 0.86-µm reflectances are sampled at 1 km so we resample the 0.64µm reflectance to 1 km resolution before creating the true-color RGBs. The 3.9-µm and 10. 3-µm 85 TBs are sampled at 2 km. We focus on the southeastern Pacific Ocean (SEPAC) defined as the region spanning 45 • S -5 • N and 70 • W -120 • W during September -November of 2019, and we create weekly animations of the true-color RGBs and TB 10.3µm˘3.9µm images to visually identify POCs. For consistency, we classify any region of clearing completely enclosed within the StCu deck as a POC with the following conditions: 1) regions of clearing at the StCu edge that become completely enclosed within the StCu deck are not classified as POCs, and 2) any potential POCs that appear to develop in response to 90 gravity waves (Allen et al., 2013) are not included. This visual identification method is admittedly subjective, and the overall classification of POCs under this framework may slightly differ by person. However, our overall results, as discussed later, are consistent with prior POC studies and a subjective approach is common in this literature (e.g. Wood et al., 2008Wood et al., , 2011Terai et al., 2014;Watson-Parris et al., 2021).  Once we visually identify POCs, we develop an overcast-Sc mask that is used to track the evolution of each POC within 95 the larger cloud field. As shown in Figure 1, the initial overcast-Sc mask is defined using the clear-sky mask level-2 product from GOES-16 by filtering out clear pixels (Heidinger and Straka, 2012). The next step is different depending on if there is daylight. During the day, 1.6-µm reflectance (R 1.6 ) (resampled from 1 km to 2 km) is used to filter out ice-phase clouds and many shallow cumulus clouds from the overcast-Sc mask. R 1.6 has two useful tendencies in this regard. First, because water droplets are more reflective at 1.6-µm (Miller et al., 2014), R 1.6 tends to be larger for water droplets than ice crystals. Second, 100 R 1.6 tends to be brighter in StCu than cumulus pixels (e.g. Zinner and Mayer, 2006;Wolters et al., 2010). To exploit these tendencies, we find the median R 1.6 of all cloudy pixels within the SEPAC region and exclude the lowest 50% of remaining cloud pixels that tend to be associated with ice-phase and liquid phase cumulus clouds (Figure 2c-d). At night, the initial filter uses TB 10.3µm˘3.9µm as an initial overcast-Sc mask filter. Considering that warm cloud emissivity is smaller at 3.9-µm than ice cloud emissivity but similar at 10.3-µm (Hunt, 1973), TB 10.3µm˘3.9µm has been used to separate both cloud types (e.g.

POC
105 Jedlovec et al., 2008). Therefore, any pixels with TB 10.3µm˘3.9µm < 0.3 K are excluded (Figures 3c-d). Once the overcast-Sc mask is conditioned using either R 1.6 or TB 10.3µm˘3.9µm , we remove any remaining cold clouds using a threshold 10.3 µm TB of 273 K and clouds over land using the Global 1-km Base Elevation dataset (Hastings and Dunbar, 1999).
The filters above effectively precondition the overcast-Sc mask but it still likely includes the brightest shallow Cu and it can be noisy near the edge of the StCu deck. To smooth the overcast-Sc mask, we calculate the mean binary mask value within a 110 25x25 pixel window centered on each pixel. We filter out any overcast-Sc mask pixels with a window-mean mask value < 0.5  (Figures 2k and 3k). We then map the potential POCs to the observed cloud field to visually confirm if a potential POC is meeting our subjective POC criteria defined above. Comparing Figure 2a to 2l and Figure 3a to 3l, we see that our algorithm can effectively identify both StCu decks and the POCs they enclose. We identify a total of 258 POCs over the study period.

Defining Lagrangian Trajectories
We use trajectory analysis to track the evolution of each POC. We define the initial POC time by identifying a time when 120 each POC is larger than 3136 km 2 (approximately one 0.5 • x0.5 • gridbox) and visually not too irregularly shaped such that the POC centroid is contained within the POC region and not in the surrounding StCu. From this time and location, we run forward and backward trajectories using the nearest 3-hourly 925-mb horizontal winds from Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2; Gelaro et al., 2017), with a time step of 10 minutes to follow the Lagrangian evolution of POCs. The trajectories are run forward and backward from the initial time up to 6 hours before 125 POC development or after POC dissipation. If any trajectory intersects any other POC in the 6-hour timeframe before POC development or after POC dissipation, it is terminated prematurely.
We classify the POC start time as the time along each trajectory when CF begins to decrease. POC end time is determined using two separate criteria: 1) CF increases back to 100% and does not change for more than an hour (hereby known as the POCs that re-close) or 2) at least one POC edge reaches the edge of the StCu deck (hereby known as the POCs that never 130 re-close). The first criterion is based on the calculated changes in CF along each trajectory, while the second criterion must be satisfied visually. Throughout the rest of the paper, these trajectories will be labeled as POC cases/trajectories. Level-2 Binary Cloud Mask   Level-2 Binary Cloud Mask  the POC forms, and we define the "after" time as the time after the CONTROL trajectory reaches the location where the POC dissipates.

Cloud Properties
We compare the following cloud properties along the POC and CONTROL trajectories: CF, cloud optical depth (COD), cloud top effective radius (r e ), liquid water path (LWP), and cloud drop number concentration (N). COD, r e , LWP, and N are composited from cloudy pixels within a 0.5 • x 0.5 • window surrounding each trajectory point because this window size is close to the same size as a MERRA-2 gridbox. Finally, CF is defined as the number of cloudy pixels divided by the total number of pixels 145 within each window. As mentioned prior, different channels and algorithms are used to retrieve COD and r e during the day and night. During the day, a combination of 0.64-µm and 2.25-µm reflectance are used (Walther et al., 2013), while 3.9-µm, 11.2-µm, and 12.3-µm brightness temperatures are used at night (Minnis and Heck, 2012). The day/night retrieval algorithms are fundamentally different, therefore the daytime and nighttime cloud properties are separated in the remainder of the paper.
At night, COD is limited from 0 to 16 and r e is limited to 2 µm -78 µm, whereas, during the day, COD can be retrieved from 150 0.25 to 158 and r e can be retrieved from 2 µm to 100 µm. The dynamic range is smaller at night because the emissivity of larger particles is similar at 11.2-µm and 12.3-µm, resulting in a smaller range of COD and r e values that can be discerned (Lin and Coakley, 1993). An effect of the limited range of nighttime optical depth retrievals is that a noticeable fraction of nighttime CODs are exactly 16. Finally, any values retrieved during twilight (solar zenith angles from 65 • -90 • ) are of degraded quality (Minnis and Heck, 2012;Walther et al., 2013) and are therefore removed resulting in a short diurnal sampling gap in the cloud 155 property data. LWP is derived using equation 8, and N is derived using equation 9 from from (Wood, 2006).
One potential limitation of the data is cirrus contamination. To account for this, we use different sets of tests, that are not used in the POC identification algorithm, during the day and night to remove cirrus. During the day, cirrus removal is based on the GEOS 1.37-µm channel (Schmit et al., 2018), and all cloud-property values within any 0.5 circ x0.5 circ window containing any 1.37-µm reflectance values < 5 are removed. We subjectively chose this value, because we visually found a 1.37-µm 160 reflectance threshold of 5 results in the lowest amount of non-cirrus cloud removal, while removing the most cirrus. We also use 8.4-µm -10.3-µm brightness temperature difference (TB 8.4µm˘10.3µm ) to remove cirrus because TB 8.4µm˘10.3µm tends to have positive values for ice clouds and small negative values for low water clouds (Baum et al., 2000;Giannakos and Feidas, 2013); previous studies (Krebs et al., 2007;Strandgren et al., 2017) have removed retrievals with TB 8.4µm˘10.3µm > -5 K. At night, we use a stricter TB 10.3µm˘3.9µm threshold to remove cirrus based on the assumption that any negative TB 10.3µm˘3.9µm 165 are likely representative of cold clouds (Jedlovec et al., 2008), such that any cloud retrievals with TB 10.3µm˘3.9µm < 0 K are removed. This algorithm effectively removes thick cirrus; however, it struggles to remove thin cirrus. Despite this, we find any potential influence of thin cirrus does not affect the overall statistics discussed in our results. Note, these tests are not used to flag cirrus clouds moving over the StCu deck that our POC identification algorithm may identify as a potential POC, because we can visually distinguish cirrus from POCs in the weekly animations.

Precipitation
We compare precipitation intensity along both the POC and CONTROL trajectories using the Advanced Microwave Scanning Radiometer 2 (AMSR-2) precipitation product from (Eastman et al., 2019). This dataset is based on statistical relationships between 4 x 6 km 2 AMSR-E 89 GHz microwave brightness temperatures and collocated CloudSat rain rates, applied to 3 x 5 km 2 AMSR-2 observations. These statistics are used to derive CloudSat-like precipitation across the microwave radiometer swath, 175 which allows for significantly more potential overlap with GOES-16 identified POC cases than does CloudSat. We co-locate AMSR-2 with GOES-16 by identifying any time within 20 minutes of a GOES-16 timestamp where AMSR-2 precipitation is observed within any 0.5 • x 0.5 • POC box. Figure ??a visually demonstrates this, showing variations in matched precipitation within an example POC. However, due to the statistical nature of the AMSR-2 product, we find that the data identify 98% of all POC and CONTROL AMSR-2 pixels as possibly raining over SEPAC, which is much higher than the typical rain fractions 180 we found of 5% using the rain certain classification from CloudSat. To correct this, we use a precipitation threshold of 0.1 mm day-1, which corresponds to rain probabilities in the AMSR-2 product typically below 5% (Figure ??b) and is consistent with the minimum observable value of surface rain (Comstock et al., 2004;Rapp et al., 2013  where the solid black line represents the median probability at a given rain rate, grey fill represents the 10 th -90 th percentile spread at a given rain rate, and the dashed red line represents the rain rate threshold of 0.1 mm day −1 .

Environmental Conditions
We classify the large-scale environment at each point along the POC and CONTROL trajectories using sea-level pressure 185 (SLP), estimated-inversion strength (EIS), 700-mb water vapor mixing ratio (q v ), 700-mb omega (ω), planetary-boundary layer (PBL) height, PBL mean q v , lifted condensation level (LCL) height, aerosol-optical depth (AOD), 50-m winds, and 925-mb wind speed/direction, all of which are derived from MERRA-2. EIS is calculated using equation 4 from (Wood and Bretherton, 2006), where lower-tropospheric stability (LTS; Slingo, 1987;Klein and Hartmann, 1993) represents the difference between potential temperature at 700 mb and sea-surface temperature is taken from MERRA-2. The 850-mb moist adiabatic lapse rate (γ 850 m ) is derived using Metpy (May and Bruick, 2019). The LCL height is derived using MERRA-2 surface temperature and the formulation of Romps (2017). MERRA-2 outputs two PBL Height variables, one (PBLH) based on the total-eddy diffusion coefficient of heat (PBLH), and another based on the bulk Richardson number (TCZPBL; McGrath-Spangler and Molod, 2014). Ding et al. (2021) found both generate PBL depths shallower than those derived directly from satellites, but PBLH is closer. Therefore, we use PBLH as our proxy for PBL height.

General POC Characteristics
Of the 147 POCs that have valid CONTROLs, Figure 5a shows that most POCs traverse between 5 • S-25 • S, and 80 • W-100 • W, similar to prior satellite-based studies (Wood et al., 2008;Watson-Parris et al., 2021). In comparison, Figure 5b shows the CONTROL trajectories take similar paths to the POC trajectories, increasing our confidence that both the POC and CONTROL Interestingly, 88% of all POCs never re-close, meaning that they remain open-celled until they exit typically north/northwest of the StCu deck. Further breaking this down, 129 POCs never re-close, 12 POCs re-close, and 6 of the calculated trajectories leave their associated POC area prematurely. Note that the 6 POC trajectories that prematurely exit POCs are not included in the remainder of this analysis. Here we note that we could use the POC centroids themselves to define the trajectory to salvage 205 these discarded trajectories. However, that method would only work during the POC lifetime whereas the use of reanalysis trajectories allows us to extend the trajectories both before and after the POC lifetime. Using a cloud-resolving model, Feingold et al. (2015) found that the recovery from open-to closed-cell StCu is much slower than the transition from closed-to open-cell Sc, and it depends on the replenishment of aerosols resulting in cloud water increases exceeding precipitation loss. Our results suggest this does not frequently happen in the SEPAC, and as a result, POCs that never re-close and those that do are grouped 210 together throughout the remainder of this paper. Figure 5c shows the location where POCs typically begin and end, showing that POCs tend to form between 10 • S-20 • S, and 80 • W-90 • W and that POC starting locations tend to cluster more than POC ending location. This implies that there is relatively high variability in POC duration. To quantify this, Figure 6 shows a histogram of POC duration. POCs are generally long-lived and last an average of 20 hours, similar to previous observational (Stevens et al., 2005)   To determine the diurnal cycle in start and end times of POCs, Figure 7 shows the relative frequency of POC start and end times, with POCs typically forming at night and ending during the day. Night time formation is consistent with prior literature that found POC formation is most likely at night when precipitation is most intense and StCu are thickest (Wood et al., 2008; previous results. Why might POCs preferentially end (exit the Sc) during the day? We find that StCu area reaches a minimum around 12 local time (Figure s-1). As a result, we hypothesize that this may simply be the result of a general reduction in StCu extent during the daylit hours (Burleyson and Yuter, 2015), so that the StCu edge effectively moves towards the POC during sunlight.
(2021). The differences may result from different algorithm sensitivity or regional sampling differences between the SEPAC and their global statistics. Figure 8b shows changes in POC maximum area as a function of POC duration. POCs tend to grow larger the longer they last, with POCs lasting > 20 hours having a maximum area approximately 10 times larger (3.01x10 5 km 2 ) than those lasting less than 10 hours (2.69x10 4 km 2 ).

POC Occurrence and Associated Environment
In this section the large-scale environmental conditions are contrasted between periods when POCs occur frequently and when they do not occur. For this analysis we include all 258 identified POCs. Figure 11 shows that there are usually between 0 and 7 POCs on any given day. However, this activity is not random, with extended periods of frequent POC occurence, followed by sustained periods with few POCs. We find the interquartile range of the daily POC count to be 0 and 7. Therefore, we 235 grouped days when no POCs developed and days when > 7 POCs developed together separately, and composited the average environments for both groups as shown in Figure 10.
Setting the stage synoptically, we find that a surface high pressure system (Figures 10a and 110b) Figure 10x shows AOD is generally lower, except very close to the continent, on days with the most POCs. This is consistent with prior studies that found POC air tends to be cleaner than non-POC air (e.g. Wood et al., 2011;Terai et al., 2014). However, the pattern of the composite difference is still complicated and we must exercise some caution in the interpretation of an aerosol field derived from a reanalysis.
On days with the most POCs, a shallower and moister boundary layer, lower LCLs, large-scale subsidence, and a strong there are several days with no POCs and several days with many may be that environmental conditions more favorable for StCu development result in a higher likelihood of POC development.

Cloud-Field Characteristics
We now turn to a comparison of CF and cloud microphysical properties for the POC and CONTROL cases. Figure 12 shows the 255 evolution of example POC and CONTROL trajectories in terms of CF, average COD, average in-cloud LWP, average r e , and average N. The largest changes along the POC trajectory are in CF, COD, r e , and N relative to the CONTROL trajectory, while LWP remains relatively constant. Specifically, Figure 12a shows that CF decreases as the POC develops. This decrease in CF is accompanied by decreasing COD (Figure 12b), increasing effective radius (Figure 12d), and decreasing N (Figure 12e). These variables then remain relatively constant during the POC's lifetime before approaching pre-POC values after POC dissipation.

260
Larger r e and lower N inside of this non-re-closing POC is consistent with the development of precipitation, resulting in the closed-to open-cell transition (Comstock et al., 2007;Savic-Jovcic and Stevens, 2008;Wang et al., 2010b;Glassmeier and Feingold, 2017).    One case gives us a glimpse at how cloud field properties vary throughout a POC's lifetime, and how they compare to a CONTROL case. However, more cases are needed to make more definitive conclusions. Therefore, we composite all the cloud 265 field properties for all POC cases and their corresponding CONTROL case, and we then compare the distribution of each variable before POC formation, during POC lifetime, and after POC dissipation. We then separate daytime and nighttime data because the retrieval algorithms are subject to very different sensitivity and uncertainty limitations. Figure 13 shows the 10 th -90 th percentile spread (see Figure S2 for the full daytime distributions) in the cloud field and precipitation characteristics of all POCs and their associated CONTROL trajectory. For all POC cases during the day, CF and 270 COD are similar to the CONTROL cases before POC formation. There are indications of elevated LWP, r e and N relative to the CONTROL trajectory before POC formation suggesting drizzle initiation before a visible cloud regime transition. Once the POCs form, CF decreases, COD decreases, LWP decreases, r e increases, and N decreases for all POC cases relative to the CONTROL cases. However, LWP remains similar to the CONTROL cases for the POC cases during POC lifetime. Switching to the nighttime data, Figure 13 (see Figure S3 for the full nighttime distributions) shows that the overall patterns 275 in CF, COD, r e , and N and how they compare to the CONTROL cases are similar to the daytime characteristics of all POCs, with one main difference. The COD and derived LWP are substantially smaller than in the daytime data, and the CONTROL and POC LWP are essentially indistinguishable at night. This is likely an artifact of the limited dynamic range of the GOES-16 nighttime algorithms. Nevertheless, given the substantial algorithmic differences in retrieved COD and r e at night versus during the day, it is encouraging that the microphysical patterns (i.e. r e and N) are similar. In particular, one might be concerned that 280 the daytime differences between the POC and CONTROL trajectories are merely an artifact of 3D radiative transfer artifacts differentially affecting the two regimes (Zhang et al., 2012;Liang et al., 2015). However, the fact that these signals are also observed in the emission-based nighttime data lends credence to their sign, while uncertainty in their magnitude remains.

Precipitation Characteristics
Figures 13p-r show the precipitation statistics in a manner similar to the cloud statistics. To further illustrate the differences 285 between POC and CONTROL precipitation, Figure 14 shows the distribution of rate-weighted rain rate. Specifically, each bin is multiplied by bin-center rain rate, which places a higher weight on bins with more intense rainfall such that the area under the histogram is equal to the accumulated rainfall. After Figure 14. Histograms of rain rate along the POC (red) and CONTROL (blue) trajectories before (A), during (b), and after (c) POC lifetime are shown, where each bin is multiplied by bin-center rain rate (e.g. accumulation weighted) and the area under each curve is normalized to mean rain rate for each distribution Precipitation rates are typically low with median values between 0.14 and 0.44 mm day −1 and they remain relatively constant from before POC development until after POC dissipation. Precipitation is generally more intense at night than during the day 290 (Figure 13p-r) which is consistent with prior works (Wood et al., 2008;Burleyson et al., 2013). Notably, there is a larger spread in rain rate, with more frequent intense precipitation for the POC cases compared to the CONTROL cases. Although evident during both day and night, the differences in the occurrence of the most intense precipitation between the POC and CONTROL cases are most distinct before POC development at night. Furthermore, Figure 14 clearly highlights that there is substantially more precipitation accumulation for POC then control trajectories during the POC lifetime. Together these 295 finding are consistent with larger re (Figures 13j-l) and lower N (Figures 13m-o) retrieved from GOES-16 for POCs than controls, which suggest that intensity and quantity of precipitation is key to the formation and maintenance of POCs.

Environmental Characteristics
In section 3.2, we evaluated the synoptic environment during periods of frequent POC occurrence and rare occurrence in an attempt to identify conditions conducive to POC formation. Most notably we found that the largest numbers of POCs tended to 300 form when AOD is low and the StCu area is high. Here we will instead evaluate the differences in the environmental conditions along the POC trajectories with the CONTROLs to identify whether there are systematic differences between the large-scale forcing experienced by the two trajectories. We focus primarily on PBL characteristics because of their potential importance of rain to POC development. We find that PBL height (Figures 15a-c around 800-1400 m, moisture around 7-11 g kg −1 , AOD around 0.05-0.15, LCL height around 400-600 m, and winds from the southeast at 20 knots. Additionally, we find there are no significant differences between the POC and CONTROL cases in EIS, 700-mb q v , and 700-mb ω (Figure s-4). Holistically, these results are consistent with the expectation that few differences exist between large-scale forcing where POCs form and those where they do not. This implies that the process relevant to the formation and maintenance of POCs are small-scale processes within the PBL.

4 Summary and Discussion
This study develops a novel methodology to identify POCs and then uses a Lagrangian analysis to track their evolution and how that compares to CONTROL trajectories of closed-cell StCu that never transition. Along the trajectories, we analyze the cloud field characteristics, and environmental characteristics of both cases. We find that POCs tend to last on average 20 hours, have a maximum area larger than 104 km 2 , and exit the StCu deck without re-closing 88% of the time. 315 We find that POC development and maintenance are most highly correlated with processes that influence cloud microphysical state such as precipitation Burleyson et al., 2013;Burleyson and Yuter, 2015;Eastman et al., 2021) as opposed to large-scale forcing, which does not appear to have an impact on POC formation (Sharon et al., 2006;Bretherton et al., 2010;Berner et al., 2011Berner et al., , 2013. Modeling studies (Feingold et al., 2010; have inferred that POC formation may be related to the clustering of raining closed-cell StCu that drives the development of interacting cold  Figure 16. Wind rose plots are used to show the most common 925-mb wind speed and direction before a POC (panels A and D), during a POC (panels B and E), and after a POC ends (panels C and F). The wind rose plots in the top row represent the trajectories that intersect a POC, while those in the bottom row represent the trajectories that do not intersect a POC. pools, which subsequently causes more intense precipitation and initiates the transition to open cells. Even though we cannot observe changes in precipitation organization alone from GOES-16, our results show that re is typically higher, N is lower, more precipitation falls and is more intense for POCs than the CONTROL population before POC formation. These findings are consistent with precipitation occurrence and intensity driving likely cold-pool development and subsequent reductions in CF.  used a cloud-resolving model to show that the distance between precipitating StCu cells 325 is important to open-cell development. After POC formation our results show that more frequent rain persists, while effective radius becomes even larger and N decreases further than before POC formation. It is, however, also possible that at least some of the observed increases in effective radius during the POC period could be related to retrieval artifacts associated with increasing sub-pixel heterogeneity (Zhang et al., 2012;Liang et al., 2015).
Overall, these results appear robust and suggest precipitation as a key driver of POC development. However, we found 330 evidence, similar to Allen et al. (2013), suggesting gravity waves may also influence POC development. We found this happened infrequently over SEPAC during September-November, 2019. Therefore, more observations over a longer timeframe are needed to assess their influence.
The general understanding of broader StCu to cumulus transitions is that they occur when StCu drift over warmer sea surfaces, leading to deeper and decoupled boundary layers (Albrecht et al., 1995;Wyant et al., 1997;Bretherton and Wyant, 1997;335 Stevens, 2000;Wood and Bretherton, 2004), in which cumulus clouds begin to develop that penetrate the overlying StCu layer and mix drier free-tropospheric air into the cloud layer (e.g. Wyant et al., 1997). These studies have argued that precipitation is not necessary for these transitions to occur, however there has been some debate over the importance of precipitation in the timing of the transition (e.g. Paluch and Lenschow, 1991;Eastman and Wood, 2016;de Roode et al., 2016;Yamaguchi et al., 2017). Considering most POCs identified in this study never re-close, our results suggest that 340 the development of POCs driven by organizing precipitation can mediate the timing of the stratocumulus to cumulus transition. Yamaguchi et al. (2017) used LES to investigate the influence of precipitation on StCu transitions, finding that, when aerosol/drop number concentrations are low due to precipitation, the StCu transition can be rapidly accelerated. Additionally, as the open-cells within POCs continue to organize and precipitation intensity increases, the StCu deck transitions to a precipitating shallow cumulus field (Yamaguchi et al., 2017). These results may explain why most POC cases we observe never 345 re-close, but more observational studies tracking POCs over a longer timeframe and across more regions are needed to draw more general conclusions.
The results herein have important implications for aerosol-cloud-interactions and climate change mitigation through marine cloud brightening. Importantly, current models do a poor job of representing the warm rain process (e.g. Sun et al., 2006;Kharin et al., 2007;Wehner et al., 2014;Christopoulos and Schneider, 2021;Witte et al., 2021) and the scavenging effect of 350 precipitation on aerosol (e.g. Tost et al., 2010;Grandey et al., 2014;Gettelman et al., 2015;Michibata et al., 2019;Jing et al., 2019). Our finding that POCs tend to accelerate the StCu -cumulus transition and not to reclose suggests that the ability of aerosol to enhance cloud albedo is highly dependent on the current state of the cloud field. Specifically, the efficiency of cloud condensation nuclei at influencing cloud albedo over the cloud lifetime would be maximized prior to the nascent formation of drizzle that precedes POC development.

355
Even though our analysis is limited to a 3-month period over the SEPAC and may not generalize to other marine stratocumulusdominated regions of the globe (i.e., northeast Pacific and southeast Atlantic basins). We believe, most importantly, that this study demonstrates that the improved spatio-temporal resolution of the current generation of geostationary sensors and the associated data product suite provides an important tool in evaluating the temporal dimension of POCs for future studies.