Aircraft observations of cold pools under marine stratocumulus

in the C-130 measurements by identifying step-like decreases in the potential temperature. Examination of their mesoscale environment shows that the observed cold pools tend to form under heavier precipitation, thicker clouds, and in cleaner environments. Cold pools are also found to form under clouds with high LWP values over the night of or before sampling. When they form, cold pools often form in clusters or on top of 10


Introduction
The formation of pools of cold, relatively high-density air at the surface, or cold pools, represents one key way in which precipitation from convective clouds organizes cloud fields. Because evaporating and melting precipitation is the primary driver of cold 25 pools (Srivastava, 1987), the strongest cold pools are typically found under heavily precipitating cumulonimbus clouds, the characteristics of which have been studied extensively (e.g., Charba, 1974;Goff, 1976;Wakimoto, 1982;Engerer et al., 2008). Cold pools are found close to the surface and spread outward from the source of cold air (Droegemeier and Wilhelmson, 1987). They play a role in the organization of precipitation cells by triggering new convection at their edges via mechanical lifting at the edges 5 (Rotunno et al., 1988) or by priming surface air with high concentrations of water vapor (Tompkins, 2001).
Cold pool formation is not limited to heavily precipitating deep cumulonimbus but is also observed under shallow clouds with cloud tops below 2 km (Jensen et al., 2000;Van Zanten and Stevens, 2005;Comstock et al., 2005Comstock et al., , 2007Wood et al., 2011a;10 Zuidema et al., 2012a). Modeling studies have highlighted the importance of cold pools in determining the mesoscale organization of cloud fields across a range of cloud types, scales and environments (e.g. stratocumulus in Savic-Jovcic and Stevens, 2008; deep maritime convection in Khairoutdinov et al., 2011;trade cumulus in Seifert and Heus, 2013). Under marine stratocumulus in particular, cold pools are characterized by re- 15 duced temperatures, increased water vapor mixing ratios, and increased equivalent potential temperatures (θ e ), and have a characteristic depth of a few hundred meters (Jensen et al., 2000;Comstock et al., 2007;Feingold et al., 2010). The elevated θ e inside cold pools was initially surprising (Van Zanten and Stevens, 2005) given that precipitation evaporation alone cannot increase θ e , because evaporation conserves 20 the moist static energy. In an earlier modeling study, Mechem and Kogan (2003) noted the possibility that stratification induced by the cold pool can trap surface fluxes near the surface and increase θ e , a hypothesis later confirmed by Savic-Jovcic and Stevens (2008). Elevated θ e is not always associated with cold pools forming under precipitating clouds. In their observational study of cold pools under precipitating trade cumuli, 25 Zuidema et al. (2012a) found reduced θ e in cold pools and showed how this is achieved by precipitation-driven downdrafts that pull down drier air from aloft, which is also found in cold pools under deep cumulonimbus (Engerer et al., 2008).  (Stevens et al., 2005;Comstock et al., 2007;Wood et al., 2011a). All POC observations have found evidence of cold pools (Van 5 Zanten and Stevens, 2005;Comstock et al., 2007;Wood et al., 2011a), and large eddy simulations of POCs have shown how cold pools concentrate the areas of ascent to drive the open cell convection in POCs (Wang and Feingold, 2009;Feingold et al., 2010;Wang et al., 2010;Berner et al., 2011). Whereas precipitation is a necessary ingredient for cold pool formation and appears necessary for POC formation, it remains 10 to be seen whether cold pool formation in marine stratocumulus is a sufficient condition for the formation of open cells. It is also unclear under what range of marine boundary layer (MBL) and cloud conditions cold pools can form.
The rate at which precipitation falls from stratocumulus clouds is largely controlled by the liquid water path (LWP), or alternatively cloud thickness, and the cloud droplet 15 number concentration (Geoffroy et al., 2008;Wood, 2012). The latter is closely related to the accumulation mode aerosol concentration in marine environments (Martin et al., 1994), and this is also the case over the southeastern Pacific (SEP, Terai et al., 2012). In this region, the cloud droplet number concentration decreases and the cloud thickness increases moving away from the South American coast (Bretherton et al., 2010). In 20 response, mean cloud base precipitation rates increase from < 0.2 mm d −1 close to the coast to values in excess of 1 mm d −1 further offshore around 85 • W (Bretherton et al., 2010;Wood, 2012). This would suggest that cold pools will be preferentially found further away from the coast.
In this study we incorporate measurements from thirteen flights of the NSF/NCAR C- 25 130 research aircraft to study cold pools that form under the clouds over the SEP. The flights took place as part of the VOCALS Regional Experiment (Wood et al., 2011b) and sampled a wide range of stratocumulus clouds including both thin, overcast stratocumulus close to the Chilean coast to thick and more strongly precipitating clouds within POCs. Section 2 introduces the dataset and the criteria by which we identify cold pools. Section 3 examines the mesoscale features that accompany cold pools compared with cases without cold pools. Section 4 then examines cold pool characteristics by comparing the air inside the cold pool to that of the environment, examining closely the kilometer scale changes at the cold pool edges. Section 5 provides further 5 discussion on the mean characteristics of the cold pools and what observations tell us about their role in affecting MBL structure.

Data
The VOCALS (VAMOS Ocean-Cloud-Atmosphere-Land Study) Regional Experiment 10 (REx) is an international field experiment conducted in October to November 2008 to better understand processes and interactions that maintain the climate system of the SEP (Wood et al., 2011b). The analysis in this study is based on aircraft observations from thirteen research flights (RF02-RF14) that the NSF/NCAR C-130 flew during REx. Because our focus is on cold pools that form in the lower MBL, data from the subcloud 15 flight legs, flown at approximately 150 m altitude are the primary source of data. Because previously observed depths of cold pools under stratocumuli range from 100 to 500 m (Jensen et al., 2000;Comstock et al., 2007;Feingold et al., 2010), we expect the subcloud flight legs to sample most cold pools present along flight legs. The in situ thermodynamic, chemical, and dynamic measurements, as well as radar and lidar es-20 timates of the overlying cloud and precipitation properties, are retrieved at 1 Hz, which approximates to a spatial resolution of 100 m. We use 87 subcloud flight legs, most of which are 60 km in length, except for legs flown in POCs, which were 150-250 km in length. Their locations span from 70 to 85 • W and from 17 to 28 • S (see Fig. 6). This dataset is appropriate for studying cold 25 pools in a large range of potential stratocumulus conditions. The instruments used 11027 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | to measure the thermodynamic and dynamic state variables (temperature, pressure, water vapor mixing ratio, and horizontal and vertical winds) are given in Wood et al. (2011b). The humidity measurements are taken from the fast response UV humidity sensor because of its fast response time that is necessary to capture any changes in humidity across a cold pool edge. Because the UV humidity sensor was inoperable 5 during RF06, humidity values from the reference thermoelectric dew point sensor are used for that flight.
Since previous modeling studies have suggested that cold pools trap surface heat fluxes, we also examine measurements of chemical and aerosol tracers with sources at the surface. To study the aerosol properties within the cold pools, we use measure-10 ments from the Passive Cavity Aerosol Spectrometer Probe (PCASP) and the Cloud Droplet Probe (CDP). The PCASP measures aerosols in the dry size (diameter) range of 0.1-3 µm, whereas the CDP, which is normally used to measure cloud droplets in the size range of 1-24 µm, can also measure large or hydrated aerosols in the same size range. Heaters in the PCASP are used to heat the airstream to below 30 % relative hu-15 midity (RH), whereas the CDP measures the aerosols at ambient RH (typically > 70 % for subcloud legs). To ensure that splashing/shattering drizzle drops do not affect the aerosol measurements, a drizzle filter is applied to remove aerosol data if any of the following conditions are met: liquid water content > 0.04 g m −3 (as measured by the Gerber PV-100 Probe); drizzle drop concentration > 1 L −1 (as measured by the 2DC 20 Probe); a 10 s-lagged, 11 s-mean drizzle water content > 10 −4 g m −3 (as measured by the 2DC Probe).
Properties of clouds above the cold pools are obtained using retrievals from the Wyoming Cloud Lidar (WCL), the Wyoming Cloud Radar (WCR), and the G-band Radiometer Probe (GVR). Cloud base height is estimated from the WCL by determining 25 the height of maximum gradient in lidar backscatter . When heavy drizzle rates preclude accurate cloud base height estimation from the WCL, the altitude of maximum reflectivity from the WCR is used as the cloud base height, because the two are found to track each other with a bias of 22 m and root-mean-square error of 280 m from the VOCALS data. Cloud top height is retrieved from the WCR by determining the maximum height where the power return and the variance of the Doppler velocity exceed threshold values. Note that cloud top heights were not retrievable for clouds with low radar reflectivity. Hence, only 80 % of cloud top heights were obtained for cloud columns in which the maximum column reflectivity exceeded −20 dBZ. The 5 liquid water path (LWP) of the cloud is retrieved from the GVR, following Zuidema et al. (2012b). Drizzle occurrence at the flight level is obtained from the 2DC probe. We use a drizzle water content threshold of 0.02 g kg −1 to indicate drizzle occurrence, which corresponds to a precipitation rate of roughly 1 mm d −1 .
Satellite data are used to obtain information about the mesoscale cloud fields in 10 which cold pools form. Visible radiance images and infrared brightness temperatures (centered on a wavelength of 11 µm) from the Geostationary Operational Environmental Satellite imager (GOES-10), obtained roughly every fifteen minutes, are used to study the cloud morphology around cold pools. A four-times daily dataset (∼ 6 a.m./p.m. and 2 a.m./p.m. local time) of LWP retrievals are obtained at a 0.25 • resolution from 15 a number of polar orbiting passive microwave satellites (Advanced Microwave Sounding Radiometer-EOS, AMSR-E; WindSat; Special Sensor Microwave/Imager on the Defense Meteorological Satellite Program platforms) that was constructed over the SEP using data from Remote Sensing Systems, Santa Rosa (http://www.ssmi.com) and the methods described by Wentz (1997), Wentz and Meissner (2000), and Hilburn 20 and Wentz (2008).
To track changes in mesoscale cloud morphology before and after the time of cold pool sampling, 2-D isobaric backward and forward trajectories at 950 hPa were calculated from European Centre for Medium Range Weather Forecasting (ECMWF) operational analyses as described by George et al. (2013). shows that in most cases where there is a drop in θ, there is a subsequent increase in water vapor mixing ratio (q v ). This is indicative of drizzle evaporation (Paluch and Lenschow, 1991;Jensen et al., 2000;Comstock et al., 2005). The drops in θ also coincide at times with thick clouds, which are more likely to produce heavy drizzle 5 (Geoffroy et al., 2008;Terai et al., 2012). In this study changes in θ alone are used to identify the edges of cold pools, which we henceforth refer to as cold pool edges.
To capture the step-like change in θ across a cold pool edge while excluding spurious turbulent spikes in θ, we experimented with different θ threshold criteria to define a cold pool edge. To help visualize how we identify the cold pool edges, an idealized θ time 10 series is shown in Fig. 2. We designate a particular data point at time t in the time series a cold pool edge if the following conditions are all satisfied where θ 1 = 0.06 K and θ 2 = 0.12 K. Whereas the previous criteria ensure that random temperature spikes in the data are not identified as cold pool edges, we allow for two sets of spatial intervals, where either t 1 = 2 s and t 2 = 4 s or t 1 = 3 s and t 2 = 6 s, to pre-20 vent the turbulent spikes from hindering the identification of potential cold pool edges. Roughly speaking, an edge is identified if four θ measurements over 200 m or 300 m intervals each decrease by more than 0.06 K. If consecutive data points in a time series satisfy the criteria, then the median among those data points is chosen as the cold pool edge. The same process is run in reverse to identify cold pool edges that involve θ in- 25 creasing in time. The values of θ 1 and θ 2 are chosen such that the step-wise changes identified with the criteria largely match those identified by eye. The number of cold pool edges and magnitude of the changes observed across edges is sensitive to the choice of θ 1 and θ 2 but the key conclusions of this study are unaffected by the choice. 11030 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Cold pool edges identified in this way are indicated by red triangles in the θ time series of Fig. 1c. Despite the subjective choice of thresholds, the criteria can be applied to the data from all 87 subcloud flight legs of the C-130, and the cold pool edges can then be used to composite the data. The main disadvantage to using the drop in θ is that it misses cold pools whose abrupt θ drops over the sub-kilometer scale have recovered 5 either due to mixing or surface fluxes. The method is also unable to identify small scale cold pools where the temperature jumps occur over spatial scales < 200 m. Despite these potential shortcomings, the θ criteria provide a basis to compare cold pools and their surroundings across 90 cold pool edges distributed over 22 of the 87 flight legs.
3 Atmospheric conditions when cold pools are observed 10

Size
The size L of cold pools on a given flight leg is determined by examining the two sets of cold pool edges that involve θ-increasing and θ-decreasing with time along the leg. For each θ-decreasing edge, we examine the θ time series at subsequent times up to the time (if one exists) when the temperature recovers to the value it had at the edge 15 itself. We denote this time the recovery time. If a second θ-increasing edge is found within 10 s (∼ 1 km) of the recovery time for the first edge, then we denote the pair as being two edges of the same cold pool. The time between these two cold pool edges is reported as its size. However, there are instances where a θ-increasing edge is not detected. For such cases, the cold pool size is reported as the time between the θ 20 decreasing edge and the recovery time for that edge. If θ does not recover before the end of a flight leg, then a cold pool size is not reported. Indeed, sizes for 15 of the total 80 cold pools identified were excluded for this reason. We expect therefore that our cold pool size distribution (Fig. 4) has a low bias for larger cold pools. Since the aircraft flew straight legs at a rate of ∼ 100 m s −1 , the time interval is then converted into a distance. 25 The size of the cold pools identified in this way for the flight leg in Fig. 1 is shown in 11031 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Fig. 3. Note that cold pools determined using our approach can be overlapping, such that smaller cold pools can be found embedded within larger cold pools. If we run our algorithm in the reverse direction, the differences are found to be negligible.
The cumulative size distribution f (L) of all the observed cold pools is shown in Fig. 4. We correct the measured distribution f m (L) on a leg-by-leg basis to account for the 5 undersampling of larger cold pools by limited leg length L leg using the method used in Wood and Field (2011): For cold pools smaller than 20 km (90 % of the observed cold pools), this weight is typically between 1.0 and 1.5. 10 The median size of the corrected size distribution is ∼ 6 km, and the middle 50th percentile of the distribution lies between 2 km and 16 km. The large cold pools (> 50 km) are mostly cold pools associated with POCs, whose contribution may appear anomalous. That said, Comstock et al. (2005) found similarly anti-correlated mesoscale variations on scales of 50-100 km in surface layer temperature and moisture over the SEP. 15 The size distribution suggests that for computer simulations to accurately represent the interaction of cold pools under marine stratocumulus, their domain size needs to be significantly greater than 10 km. Figure 4 also shows that f (L) largely follows a lognormal distribution. It should also be noted that since the aircraft did not necessarily intersect the centers of the cold pools, if we assume that the cold pools are circular 20 and that flight transects are uniformly distributed along the diameter of cold pools, we expect the lengths to be underestimated by approximately 25 % (a factor of 4/π).
We find a strong negative correlation between the cold pool size and the mean temperature difference between inside and outside the cold pool (Spearman rank correlation of −0.76, Fig. 5). This strong correlation suggests that larger cold pools require sustained cooling. When taken with the finding above ( Fig. 3) that large cold pools are associated with clusters of cold pools, we speculate that large cold pools are not formed from larger precipitation cells, but from the clustering of multiple cells. As a result, the larger cold pools can last beyond the lifetime (∼ 2.5 h, Comstock et al., 2005) of the individual precipitation cells that form the smaller cold pools, because the larger cold pools are made of multiple cold pools forming over preexisting cold pools.

5
Because mean precipitation rate increases westward from the South American coast (Leon et al., 2008;Bretherton et al., 2010;Wood et al., 2011b) and because strong mesoscale variability in temperature is associated with precipitation rates > 1 mm h −1 (Comstock et al., 2005), we expect cold pools to preferentially form further away from the coast. Our observations confirm this (Fig. 6), with no cold pools observed within 10 500 km of the Chilean coast (i.e. east of 75 • W). One caveat is that the coastal pollution flight legs that extended from 20 • S to 28 • S were conducted during the day when drizzle is less frequently observed (Leon et al., 2008) and less heavy (Burleyson et al., 2013). However, many of the legs along 20 • S east of 75 • W were at night when precipitation is favored. The numbers of cold pool edges in each leg ( Fig. 6) indicates that cold 15 pools are usually clustered together, such that if one cold pool is observed on a flight leg, there is high likelihood that another cold pool is observed on the same leg. Thus the example shown in Fig. 3 is actually quite typical; cold pools are not typically found by themselves, and interactions between cold pools must be studied when trying to understand their effect on the MBL.

Boundary layer characteristics
Cold pools form under precipitating clouds and so we expect cold pools to form preferentially in MBLs with strong precipitation. In Fig. 7, we have plotted the flight leg mean cloud base precipitation rate against the leg mean cloud thickness. As in other studies (Bretherton et al., 2010;Wood et al., 2011a;Terai et al., 2012), the mean precipita-25 tion rate is estimated from the cloud base WCR reflectivity using the Z-R relationship 11033 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | from Comstock et al. (2004). Open circles differentiate legs that traversed POCs from those that did not. Thicker clouds tend to produce higher precipitation rates and those legs with cold pools clearly are those with thicker clouds and heavier precipitation. That said, Fig. 7 also shows that cold pools were not observed on a number of flight legs where the mean precipitation rate exceeded 1 mm d −1 . The reader should be aware of 5 the limits to interpreting this figure. Precipitation rates and cloud thickness are spatially variable and cold pools can still be observed after the precipitation event that formed them has dissipated, so one should not expect the association between precipitation, cloud thickness, and cold pool occurrence to be a perfect one given the nature of the aircraft sampling. 10 We would also expect cold pools to form preferentially in deeper MBLs and when CCN concentrations are low, enhancing precipitation (Terai et al., 2012). This is especially pertinent to the currently unresolved issue about the conditions necessary and sufficient for POCs formation. Observational studies of POCs have all found cold pools in POCs (Van Zanten and Stevens, 2005;Comstock et al., 2007;Wood et al., 2011a) 15 and results from modeling studies point to the importance of cold pools in maintaining and forming POCs (Savic-Jovcic and Stevens, 2008;Wang and Feingold, 2009;Berner et al., 2011). Although we will not address whether cold pool formation eventually leads to POC formation in this study, we can address whether clean environments and deeper MBLs are associated with cold pool formation. Figure 8 shows that cold

Satellite derived cloud field
Although satellite images, in conjunction with the aircraft cold-pool observations, provide a way of showing case studies of where cold pools are observed in relation to cloud features (Fig. 3), they can also be used to examine cloud mesoscale morphology associated with cold pools. Previous studies have found that heavier drizzle is ob-5 served more frequently under open cellular regions (clouds broken on the mesoscale), compared to closed-cellular regions (associated with overcast, homogeneous clouds) (Stevens et al., 2005;Sharon et al., 2006;Comstock et al., 2007). Because surface precipitation is a good indicator of cold pools (Sect. 4.1; Comstock et al., 2005), we expect to preferentially find cold pools under more broken clouds. 10 We compare cloud heterogeneity around flight legs with and without cold pools. To quantify the degree of heterogeneity, we calculate the standard deviation of 4 km resolution GOES-10 IR Channel 4 (λ ∼ 11 µm) brightness temperature (T B ) over a 1 • ×1 • box centered on the flight leg (see Fig. 3a for an example). The standard deviation of T B of the box in Fig. 3a is 1.73 K. The T B of adjacent pixels in an overcast cloud deck does 15 not vary by much (< 3 K), but the variability is much higher for broken clouds (up to 8 K). The 8 K difference is close to the temperature difference between the cloud top and the sea surface below. Most of the T B variability comes from the degree of brokenness in the cloud field, rather than the variation in cloud top temperatures, since those with high variance in the T B exhibit a bimodal distribution in T B (see also Coakley 20 and Bretherton, 1982). A histogram of the standard deviation of T B for the two populations of flight legs (with cold pools and without cold pools) shows that the population of legs with cold pools has a markedly different T B distribution ( Fig. 9). Legs with cold pools are more preferentially located in areas with broken clouds. On the other hand, some cold pools are also found in regions with overcast cloud fields. We conclude that 25 while drizzle is a necessary condition for the formation of cold pools, a broken cloud field, usually associated with precipitating cumulus clouds, is not a necessary condition for the formation of cold pools. Furthermore, if we follow the temporal evolution of 11035 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | cloud field T B using forward trajectories calculated from ECMWF wind fields (George et al., 2013), we find no indication that cold pool formation preferentially leads to clouds breaking up (not shown), suggesting that cold pool formation does not necessarily lead to cloud breakup. We can pursue the relationship between drizzle and cold pools further by looking at 5 the satellite passive microwave dataset. Because drizzle rates increase with increasing liquid water path (LWP) under marine stratocumuli (e.g., Comstock et al., 2004), we compare the satellite derived LWP for legs with and without cold pools. Since continuous passive microwave retrievals cannot be obtained, as we can from the geostationary GOES-10, we determine the location at the time of the previous satellite overpass 10 of the air mass sampled by the flight leg using backward trajectories calculated from ECMWF. The backward trajectories also allow us to follow the 24 h evolution of LWP leading up to the sampling of the flight leg. Figure 10 plots satellite-derived LWP averaged over 2 • × 2 • boxes surrounding the tracked location of the leg air mass against the time in UTC. We plot LWP against UTC instead of against hours before/after the 15 aircraft leg time to avoid mixing diurnal variability with other variability. Cloud LWP over legs with cold pools is in most cases higher than the cloud LWP of clouds over legs without cold pools. The night-time maximum LWP (∼ 08:00 and ∼ 12:00 UTC retrievals) before aircraft sampling has some predictive power of cold pool formation, but the LWP of the previous afternoon (∼ 20:00 and ∼ 00:00 UTC retrievals) does not differentiate 20 legs with and without cold pools.

Thermodynamics
We now switch from looking at the large scale differences to small scale differences seen across cold pools. Looking at small scale changes across cold pool edges can 25 inform us how cold pools differ from the environment and how they affect boundary layer processes. Overlying cloud properties, as well as thermodynamic, dynamic, and chemical properties of the air mass are composited as a function of distance from the cold pool edge for the 90 identified edges to examine the characteristic differences between the cold pool and the ambient subcloud layer. Since we are primarily interested in how the air inside the cold pool differs from the air outside, rather than the mean 5 state, we composite the deviations of the state variables from the value at the edge, unless noted otherwise. For all composite plots we show the mean values and the 95 % confidence interval of the mean calculated using Student t-statistics, assuming that each cold pool edge is independent. The composites are oriented with the cold pool to the right. A mean drop of 0.4 K in potential temperature θ (Fig. 11a and Table 1) is observed across the cold pool edge, which is substantially larger than the minimum threshold of 0.24 K set by the θ criteria (Sect. 2.2). The mean drop is sensitive to the choice of threshold, but this does not change the general conclusions in this study. As θ drops, water vapor mixing ratio (q v ) increases inside the cold pool by approximately 0.5 g kg −1 (Fig. 11b and Table 1). An increase in q v inside the cold pool is consistent 15 with the findings of Jensen et al. (2000) and Comstock et al. (2007). The spatial scale of the sharp q v and θ change is similar, lending confidence that our approach is able to capture the edge between two distinct air masses.
If drizzle evaporation is solely responsible for the θ and q v jumps, then θ e would be conserved, such that c p θ = −Lq v , where c p is the specific heat of air, L is the latent 20 heat of vaporization of water, and θ indicates the difference between inside and outside of the cold pool. Figure 11b clearly shows that Lq v exceeds c p θ by roughly a factor of three, confirming that drizzle evaporation is not the only process that determines the temperature and water vapor inside the cold pool. Another way of stating this is that θ e increases inside cold pools ( Fig. 11c and Table 1). This is not a new finding as previ- 25 ous modeling and observational studies have reported increases of θ e inside marine stratocumulus cold pools (Jensen et al., 2000;Mechem and Kogan, 2003;Van Zanten and Stevens, 2005;Savic-Jovcic and Stevens, 2008). These studies have attributed the θ e increase to an increase in stability in the lower MBL. This shows that it is likely 11037 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | a universal feature of cold pools found under stratocumulus. Because air inside the cold pool is colder and hence negatively buoyant compared to the environmental air, air inside the cold pool is mixed less efficiently with the rest of the MBL, allowing the energy from the surface fluxes to be trapped within a shallow layer close to the surface.
We also estimate the mean depth of the cold pools using the density jump across 5 the cold pool boundary, the radar altimetry data, and the pressure difference inside and outside of the cold pool. At the same altitude inside and outside of the cold pool, the denser air inside the cold pool will manifest itself as a positive pressure perturbation. We determine pressure perturbations by removing those due to small changes in aircraft altitude z using the radar altimetry, and estimate the depth of the cold pool h as: where p is the in situ static pressure, g is the gravitational acceleration, and θ is the in situ density. Subscripts 1 and 2 denote air outside and inside the cold pool, respectively. Taking the mean values of the observations made between the cold pool edge and

Dynamics
The cold pool air is denser than the environmental air, having a virtual potential temperature approximately 0.3 K lower than the environment. The cold pool therefore spreads out as a density current along the surface. The result is convergence at the cold pool edge. To estimate this convergence, the along-track wind speed (wind speed 5 along the direction of the flight track) is composited across the edge and plotted in Fig. 13. The convergence at the cold pool edge is calculated by taking the along track wind speed difference from adjacent time steps. As was the case for cold pool size (Sect. 3.1), the along track wind speed is expected to underestimate the true cold pool outflow velocity by a factor of 4/π, because the flight track is not always aligned nor-10 mal to the edge of the cold pool. The estimated convergence with the correction is 1.2 × 10 −3 ± 1.0 × 10 −3 s −1 at the cold pool edge. Figure 13b shows that the maximum convergence occurs ∼ 0.2 km inside the cold pool, with a value of 1.9 × 10 −3 s −1 .
We can then compare the corrected convergence we observe with the theoretical outflow velocity V f at the edge of a density current: where k is the Froude number, h is the cold pool depth, ∆θ is the density difference inside and outside of the cold pool, and θ is the environmental density. If we use the maximum-corrected outflow difference between inside and outside of the cold pool (0.97 m s −1 ), we obtain a k-value of 0.53. This lies on the lower end of k values reported 20 in past studies of density currents acquired from lab experiments and observations of gust fronts and sea breezes (Table 8 of Wakimoto, 1982). Since k represents the ratio of the inertial force to the gravitational force, this suggests that for the observed cold pools, whose density differences with the environment are smaller than those summarized by Wakimoto (1982), a smaller fraction of the potential energy from the The updraft at the cold pool edge is consistent with both observational (Jensen et al., 2000) and modeling (Wang and Feingold, 2009) studies that show positive vertical wind anomalies at cold pool edges. This lifting from the horizontal convergence at the cold pool edge can potentially help lift warm moist air from the surface into the cloud layer to help maintain a steady source of moist air into the precipitating clouds.

Cloud and precipitation
In Sect. 3.3 we showed that at scales of 10-100 km, cold pools preferentially formed under heavier precipitation and thicker clouds. We now examine whether the same changes in cloud and precipitation properties also occur across the cold pool edges at the 1 km scale, using retrievals from the Wyoming Cloud Lidar (WCL), Wyoming 15 Cloud Radar (WCR), and the G-band Radiometer Probe (GVR). There is no discernible change in mean cloud top height across the cold pool edge (Fig. 14a). The mean cloud top height over cold pool legs is approximately 300 m deeper than the mean cloud top height derived from all the subcloud legs in REx (see Fig. 8), but this largely reflects the fact that air masses with cold pools are generally found in the deeper boundary layer in 20 the west of the domain. On the other hand, the cloud base is approximately 100 m lower over the cold pool than it is outside (Fig. 14a), which results in the LWP increasing from approximately 250 to 350 g m −2 inside the cold pool (Fig. 14b). From the composites alone we cannot conclude that the decrease in cloud base height over the cold pool is a sign of the dynamical impact that cold pools have on the clouds, since the lowered 25 cloud base could appear because cold pools form from heavy drizzle produced by thick clouds. Figure 14a also shows that the LCL dramatically decreases inside the cold pool, evidence of the cooler and moister subcloud layer in the cold pool. The mean LCL also lies more than 200 m below the mean cloud base height, even outside of the cold pool. Thus cold pools are generally found in decoupled MBLs and have the potential to lift the decoupled surface layer above its LCL to form cumulus clouds that can penetrate 5 into the overlying stratocumulus.
To examine precipitation characteristics in and below the clouds, we composite radar reflectivity from the Wyoming Cloud Radar (WCR). From Fig. 12a, which shows the Z max composite, we can see that a majority of the columns exceed the −15 dBZ threshold that we use to distinguish drizzling from non-drizzling clouds (Bretherton et al., 2010). Since the altitude of Z max largely corresponds to the cloud base height (Sect. 2.1; Wood et al., 2011a), Z max largely reflect the cloud base precipitation rate. Previous studies report using threshold values ranging between −20 and −10 dBZ . Regardless of the choice of threshold, we can see that a majority of the clouds across the cold pool edge are drizzling, and that there is a gradual increase 15 in precipitation over the cold pool. However, there is a more abrupt increase across the cold pool edge in radar reflectivity below the mean cloud base at 500 m altitude, showing that precipitation rates at 500 m are better indicators of cold pool formation (Fig. 12b). At the flight level (∼ 150 m), there is a three-fold increase in drizzle occurrence inside the cold pool (Fig. 12c). Drizzle occurrence at the flight level is defined 20 as a drizzle water content > 0.02 g m −3 , which corresponds to the 98th percentile of all drizzle water contents observed by the C-130 at 150 m. To compare the occurrence of similarly infrequent precipitation rates at the other two levels, the occurrence of 98th percentile precipitation rates at 500 m and column maximum are also plotted in Fig. 14c. The drizzle occurrence at 500 m also increases three-fold over the cold pool, 25 but the occurrence of Z max only increases by two-fold inside the cold pool, supporting our previous observation that precipitation occurrence lower in the MBL is a better indicator of cold pool formation.

11041
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

Aerosols and DMS
In Sect. 4.1 we discussed the increase in θ e inside the cold pools, which previous studies have attributed to the trapping of surface fluxes within the cold pool. Both coarse mode (sea-spray) aerosols and dimethyl sulfide (DMS) have their primary sources at the ocean surface (Yang et al. (2009)  tions. The composites of both PCASP-and CDP-measured aerosol concentrations increase inside the cold pool (Fig. 15a), but the relative magnitude of the increase is substantially greater for the coarse particles. The mean concentration 2.5 km inside the cold pool exceeds that of the mean concentration 2.5 km outside the cold pool, by only 7 % (98 to 106 cm −3 ) for the PCASP, but by 70 % (0.09 to 0.16 cm −3 ) for the CDP (see 20   Table 1). The increase in PCASP aerosol concentration is not significant at the 95 % level, whereas the increase in CDP aerosol concentration is. Along with the coarse mode aerosol concentration, the mean DMS concentration also increases across the cold pool edge (Fig. 15c and Table 1).
To test whether the capping of surface fluxes by the cold pool-induced stratification 25 can account for enhancements in coarse mode aerosol and DMS concentrations inside the cold pool, we employ a set of two simple box models and study the differences in coarse mode aerosol and DMS concentration in the two boxes. Previous studies have established that the majority of coarse mode aerosols in the MBL come from the sea salt flux at the ocean surface and that they are strongly dependent on wind speed (Gong, 2003;Clarke et al., 2006;O'Dowd and de Leeuw, 2007). One box represents a column of air outside of the cold pool, and the other, of air inside the cold pool. Two sea salt parameterizations from Gong (2003) and Clarke et al. (2006) are used 5 to explore sensitivity to different sea salt flux parameterizations. To drive the sea salt fluxes, we use 10 m wind speeds by taking the C-130 observed mean wind speeds and interpolating them down to the surface by assuming a log-wind profile with a surface roughness of 1.86 × 10 −4 m (Wood et al., 2011a). Although wind speeds increase slightly in the cold pool (8.35 to 8.45 m s −1 ), we find that this has a negligible effect on 10 the sea salt fluxes. Because the CDP measured aerosols in ambient relative humidity and because MBL aerosols in this region are mostly hygroscopic (Allen et al., 2011), we account for the hygroscopic growth of the measured aerosols by estimating the dry diameters of the measured aerosols using the hygroscopic growth factor of sodium chloride from Tang (1996). For the DMS flux, we use a previously observed mean flux 15 estimate from the VOCALS region (Yang et al., 2009). The aircraft measurements provide no direct knowledge of the temporal evolution of cold pools, and so we must make a number of assumptions in our model. First, we assume that the cold pool forms instantaneously at t = 0 and that there is no difference in the concentrations prior to cold pool formation. Second, we assume that the heights 20 of the two boxes are fixed in time. Third, we assume that the only exchange determining changes in aerosol and DMS concentrations in the boxes are due to surface fluxes. Therefore, the different heights of the two boxes drive any concentration differences and these differences grow linearly with time.
Using Eq. (7) for estimating the depth of the cold pool, we obtain a height of 226 m 25 for the box representing the cold pool. This depth is lower than the previous estimate of 335 m, because here we are estimating the mean cold pool depth only for those legs in which CDP measurements are available. For the second box, a mixing height of 1171 m is used, derived using the following basis. The mixing height of a well-mixed 11043 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | MBL corresponds to the cloud top height, whereas the mixing height of a decoupled MBL is generally below stratocumulus cloud base (Jones et al., 2011). In the latter, the lifting condensation level (LCL) does not correspond to the stratocumulus cloud base, because θ is higher and q v lower in cloud layer than the surface layer. We define a decoupled MBL in this analysis as a MBL where the difference between the cloud base 5 height and the LCL is greater than 300 m (Comstock et al., 2005). We take the LCL as the mixing height in decoupled cases and the cloud top height as the mixing height for the coupled MBLs. The mean of these weighted by the frequency of occurrence of coupled and decoupled MBLs is used as our mixing height for the second box.
We find that the observed concentration difference of aerosols in the CDP size range is reached in 17 min using the parameterization of Clarke et al. (2006) and 23 min using that of Gong (2003). Similarly the DMS enhancement inside the cold pool, using the mean DMS flux of 3.2 µmoles m −2 d −1 in the VOCALS region as reported by Yang et al. (2009) is reached in 29 min. We take the good agreement in the time scales for coarse aerosols and DMS as evidence that trapping may be the critical process governing 15 the build-up of scalar quantities in the stable air in cold pools. Jensen et al. (2000) using a mixed layer model estimate that it takes approximately 3.5 h for the cold pool to lose its thermodynamic identity due to surface fluxes and mixing. Therefore, both the enhancements in aerosol and DMS concentrations can be explained by the trapping of surface fluxes by the cold pool. The time scales also give us an estimate of the 20 "mean" age of the observed cold pools. Since we observe that the cold pools spread out at a mean velocity of ∼ 1 m s −1 (Fig. 13a), this indicates that the mean cold pool has likely spread out a mean distance of approximately 1-2 km on either side at the time of sampling.
Attempting to estimate the time scale necessary to explain the enhanced θ e in the 25 cold pool surface layer is complicated by the dependence of the surface fluxes on the temperature and humidity of the air inside the cold pool. The fluxes cannot be considered constant inside and outside the cold pool as we have assumed for the aerosol and DMS fluxes, where we found that the difference in the wind speed, the main driver of the two fluxes, to be negligible inside and outside the cold pool. We must make the additional assumption that the cold pool air is initially cooled and moistened at t = 0 by the spontaneous evaporation of drizzle water into the air. If we make these assumptions and use observed SST and wind speed to drive the two box model, we obtain a time scale of 1.5 h. This is at least a factor of 3 times longer than the time 5 scales estimated from the sea salt or DMS flux. Given the strong sensitivity of the surface fluxes to the surface air properties, we cannot make any definitive statements as to whether stratification alone explains the enhancement of θ e in the cold pool. What we can say is that the enhancement of θ e is likely largely dominated by latent heat fluxes.

Discussion and conclusions
Modeling and observational studies have suggested that cold pools initiate transitions from closed to open cellular stratocumulus clouds (Savic-Jovcic and Stevens, 2008;Xue et al., 2008;Wang and Feingold, 2009;Feingold et al., 2010;Berner et al., 2011). This study systematically examines cold pools that form in the southeast Pacific us- 15 ing aircraft data. We first examined the range of MBL and cloud conditions that accompany cold pools, finding that cold pools form preferentially under heavily drizzling clouds (> 1 mm d −1 ). This explains why cold pools are common further offshore where the MBL is deeper, clouds are thicker, and the aerosol concentrations are lower. We find that precipitation near the surface is a better indicator of cold pool formation than 20 is precipitation at cloud base. Satellite microwave data show high values of cloud LWP during the night prior to the observed cold pools, but do not show high values of LWP during the previous afternoon. This suggests that understanding the factors controlling how high values of LWP are produced overnight will be important for predicting heavy precipitation and cold pool formation. In terms of size, the observed cold pools 25 in this study have horizontal extents that are roughly lognormally distributed with 50 % of them between 2 and 16 km. Some cold pools are larger than 100 km and these tend 11045 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | to be associated with POCs. Instead of forming separately, cold pools tend to cluster together with new ones often forming on top of older ones, stressing the importance of understanding how cold pools interact with each other to affect boundary layer processes.
Composite cold pools show that numerous variables change between the cold pool air and its environment, as summarized in Table 1 and Fig. 16. Consistent with previous measurements of cold pools under marine stratocumulus, drops in temperature are accompanied by increased water vapor, convergence and associated uplift at the edges, and enhanced θ e values inside the cold pool. Additionally, we find an increase in DMS and coarse-mode aerosol concentrations, both of which provide important evidence 10 that cold pool-induced stratification concentrates surface fluxes in the cold air near the surface. From the observations alone, we cannot address whether the dynamic or thermodynamic effects of the cold pools are more important for changing MBL structure. We do find, however, that the significant drop in LCL inside the cold pool is largely due to the q v increases instead of the θ decreases (not shown). This means that if cold pools vertically and horizontally concentrate moisture (thermodynamic) and how that gets lifted into the cloud layer (dynamic). We also find that unlike cold pools that form under deeper cumulonimbus clouds (Goff, 1976), wind gusts associated with the cold pools under stratocumulus are almost always smaller than the mean wind speed in the SEP. This means that, by themselves, cold pools cannot drive upstream propagation of POC boundaries (Fig. 2 of Wood et al., 2011a). It also means that changes in surface fluxes inside the cold pool that are driven by changes in wind speed are small. There remain a number of questions that cannot be addressed from observations alone. We find that cold pools indeed exert dynamic and thermodynamic effects on the MBL. What then is their role in organizing and influencing further precipitation? Can the dynamic effect of cold pools alone increase precipitation in a cloud field? Modeling studies of Xue et al. (2008), Feingold et al. (2010), and Wang et al. (2010 have tackled this using a variety of experiments, but it remains to be seen whether a dynamic or thermodynamic feedback between precipitation and cold pools is necessary to transition tion. http://data.eol.ucar.edu/. AMSR-E data are produced by Remote Sensing Systems and sponsored by the NASA Earth Science MEaSUREs DISCOVER Project and the AMSR-E Science Team. WindSat data are produced by Remote Sensing Systems and sponsored by the NASA Earth Science MEaSUREs DISCOVER Project and the NASA Earth Science Physical Oceanography Program. SSM/I data are produced by Remote Sensing Systems and sponsored 15 by the NASA Earth Science MEaSUREs DISCOVER Project. AMSR-E, WindSat, and SSM/I data are available at http://www.remss.com. The authors thank Anthony Clarke, Romain Blot, Jorgen Jensen, Sandra Yuter, Robert Houze, and Christopher Bretherton for fruitful discussions on various aspects of the study. Rhea George and Gallia Painter are thanked for providing the code and help to run the forward and backward trajectories. 20 Stevens, B., Vali, G., Comstock, K., Wood, R., Van Zanten, M. C., Austin, P. H., Bretherton, C. S., and Lenschow, D. H.: Pockets of open cells and drizzle in marine stratocumulus, B. Am. Meteorol. Soc., 86, 51-57, doi:10.1175/BAMS-86-1-51, 2005 Chemical and size effects on hygroscopic aerosols on light scattering coefficients, J. Geophys. Res., 101, 19245-19250, 1996. 11043 Terai, C. R., Wood, R., Leon, D. C., and Zuidema, P.: Does precipitation susceptibility vary with increasing cloud thickness in marine stratocumulus?, Atmos. Chem. Phys., 12, 4567-4583, doi:10.5194/acp-12-4567-2012, 2012 (2003) and Clarke et al. (2006). c Same as a , but using observed mean DMS flux from Yang et al. (2009) (Bolton, 1980, dotted); (b) cloud base radar reflectivity, as measured by the WCR, and corresponding precipitation rate using a Z-R relationship from Comstock et al. (2004); (c) in situ potential temperature (θ), where the red triangles indicate the location of cold pool edges, as identified by the criteria described in the text; (d) in situ water vapor mixing ratio (q v ). The distance along the x-axis is approximated from the time series assuming the aircraft flies at a mean ground speed of 100 m s −1 .

T B standard deviation (K) Fraction of total
Legs without cold pools Legs with cold pools   Fig. 11. Composite thermodynamic measurements of (a) potential temperature (θ), (b) water vapor mixing ratio (q v ), and (c) equivalent potential temperature (θ e ) from 90 cold pool edges. Anomalies relative to the value at the cold pool edge are plotted. The air mass within the cold pool is found to the right of zero, while air mass outside of the cold pool air mass is found to the left of zero. Bold lines indicate mean values, while thinner lines indicate the 95 % confidence intervals of the mean, calculated using a standard Student t test. The gray line in (b) indicates the change in q v based on observed θ in the case where c p θ = −Lq v .