The Marine Stratus/Stratocumulus Experiment (MASE) field campaign was undertaken in July 2005 off the coast of Monterey, California to evaluate aerosol-cloud relationships in the climatically important regime of eastern Pacific marine stratocumulus. Aerosol and cloud properties were measured onboard the Center for Interdisciplinary Remotely-Piloted Aircraft Studies (CIRPAS) Twin Otter aircraft. One cloud that was clearly impacted by ship emissions as well as the ensemble of clouds observed over the entire mission are analyzed in detail. Results at both the individual and ensemble scales clearly confirm the Twomey effect (first indirect effect of aerosols) and demonstrate drizzle suppression at elevated aerosol number concentration. For the ship track impacted cloud, suppressed drizzle in the track led to a larger cloud liquid water path (LWP) at the same cloud thickness, in accord with the so-called second indirect effect. Ensemble averages over all clouds sampled over the entire 13-flight mission show the opposite effect of aerosol number concentration on LWP, presumably the result of other dynamic influences (e.g., updraft velocity and ambient sounding profile). Individual polluted clouds were found to exhibit a narrower cloud drop spectral width in accord with theoretical prediction (M.-L. Lu and J. H. Seinfeld, Effect of aerosol number concentration on cloud droplet dispersion: A large-eddy simulation study and implications for aerosol indirect forcing, Journal of Geophysical Research, 2006). This field experiment demonstrates both the indirect aerosol effect on ship track perturbed clouds, as well as the subtleties involved in extracting these effects over an ensemble of clouds sampled over a 1-month period.
 One of the largest uncertainties in quantitative estimates of radiative forcing of the climate system is the interaction between aerosols and clouds, the so-called aerosol indirect effect [Intergovernmental Panel on Climate Change (IPCC), 2001; Lohmann and Feichter, 2005]. Aerosol indirect forcing of liquid (warm) clouds has generally been divided into two types. The first type, also referred to as the Twomey effect [Twomey, 1977], is that in which an increase in the number of subcloud aerosols at constant liquid water content leads to more numerous, smaller cloud droplets and a more reflective cloud. This effect, referred to as the first indirect effect, has strong observational support from in situ measurements [Warner and Twomey, 1967; Brenguier et al., 2000; Durkee et al., 2000b], satellite remote sensing [Kaufman and Nakajima, 1993; Han et al., 1998; Nakajima et al., 2001; Breon et al., 2002; Schwartz et al., 2002; Schreier et al., 2006], surface-based remote sensing [Feingold et al., 2003; Kim et al., 2003], and all of the above [Feingold et al., 2006]. The second type of indirect effect, first described by Albrecht , addresses the dynamical response of the cloud resulting from an increase in aerosol concentration. In it, more numerous, smaller cloud droplets reduce droplet growth by collision and coalescence, and the consequent suppressed precipitation formation leads to enhanced cloud albedo. In this respect, in an early work, Warner  attributed reduction in rainfall as a result of smoke particles from sugarcane fires; this effect has been confirmed in other observations [Rosenfeld, 1999]. There is less clear observational evidence concerning the response of total liquid water path (LWP) owing to the difficulty in isolating the aerosol effect from those arising from concurrent variations in meteorology. On the whole, there is limited observational support for the second indirect effect [Rosenfeld, 2000; Han et al., 2002; Kaufman et al., 2005; Rosenfeld et al., 2006; Schreier et al., 2006]. With respect to marine stratocumulus, drizzle formation can exert a strong influence on cloud structure, thickness, and coverage [Albrecht, 1989; Bretherton et al., 2004]. Variability in cloud droplet number concentration driven by variability in the number concentration of subcloud aerosol can lead to substantial variability in in-cloud drizzle characteristics. Drizzle formation can generate feedback to cloud dynamics such as decreasing boundary layer turbulence intensity and decreasing cloud top entrainment [Stevens et al., 1998]. These drizzle-cloud dynamical interactions inevitably lead to changes in the cloud LWP, the sign and magnitude of which remain unclear.
 As noted above, strong observational support exists for the first indirect (Twomey) effect, although the quantitative relationship linking aerosol number concentration, Na, in the accumulation mode [that in which the predominant number of cloud condensation nuclei (CCN) reside] and cloud droplet number concentration (CDNC) depends on aerosol concentration, size distribution, and composition, as well as meteorological conditions, such as updraft velocity. Once a perturbation occurs as a result of an increase in subcloud aerosol concentration, a series of cloud responses is induced, only the first of which is an increase in cloud droplet number activated. Cloud thickness and liquid water path, cloud top entrainment, and drizzle production can all be affected as a result of the initial perturbation, acting in concert to amplify or suppresses the original response of a brighter cloud. The moisture budget in cloud is determined by the balance between cloud top entrainment drying, surface evaporation, and cloud base drizzle rate. For example, Stevens et al.  showed that with moist air above the boundary layer, drizzle formation actually leads to a higher LWP because the reduced cloud top entrainment drying exceeds the amount of direct cloud water loss by drizzle. On the basis of this result, in marine stratocumulus, an increase in subcloud aerosol leads to a decrease in LWP for a relatively dry free troposphere, whereas the opposite is predicted in a moist free troposphere [Ackerman et al., 2004; Lu and Seinfeld, 2005]. A similar diverse trend in aerosol-induced cloud depth change, but under different criteria, is noted by Wood (Cancellation of aerosol indirect effects in marine stratocumulus through cloud thinning, submitted manuscript, 2006), such that on the timescale within a day, the aerosol induces a cloud-thickening effect if cloud base is low (<400 m), and a cloud-thinning effect if cloud base is high.
 Changes in subcloud aerosol also affect the cloud droplet size spectrum (the so-called dispersion effect) and thereby also the cloud albedo. Liu and Daum  and Peng and Lohmann  used cloud data from field studies and showed that the relative dispersion (which is defined as the ratio of the standard deviation to the mean radius of the cloud drop size distribution) apparently increased as cloud drop number concentration increased. Liu and Daum argued that the more numerous smaller droplets formed in a polluted cloud compete for water vapor and broaden the droplet size distribution compared with the unperturbed cloud. Lu and Seinfeld  carried out detailed large-eddy simulations (LES) of marine stratocumulus with varying aerosol number concentrations under the same meteorological condition (sounding). On this basis, they found that cloud spectral relative dispersion actually decreases with increasing aerosol number concentration. This is explained by suppressed drizzle at increased aerosol loadings, which results in less spectral broadening by collision and coalescence processes, and more spectral narrowing by droplet condensational growth at the higher updraft velocities dynamically induced by suppressed drizzle. Thus, the actual direction of the spectral response resulting from an increase of aerosol depends on prevailing conditions.
 In this work, we report on a dedicated field experiment designed to probe aerosol-cloud relationships in clean and perturbed marine stratocumulus. The goal of the work is to provide a comprehensive set of data to evaluate aerosol indirect effects in marine stratocumulus. Section 2 describes the experiment. In section 3, we focus on a case study involving ship emissions. Section 4 is devoted to a statistical analysis of the ensemble of clouds sampled over the entire experiment.
2. Marine Stratus/Stratocumulus Experiment
 The Marine Stratus/Stratocumulus Experiment (MASE) field campaign was undertaken during July 2005 off the coast of Monterey, California. The goal of MASE was to make state-of-the-art measurements of aerosols and clouds in the climatically important regime of eastern Pacific marine stratocumulus clouds. The cold ocean surface, in combination with warm, dry air aloft, induces the formation of one of the world’s most persistent stratocumulus cloud decks. The region of the Pacific atmosphere adjacent to the coast of California experiences air masses of very clean, background air as well as ones having strong anthropogenic influences. In addition, the prevalence of shipping lanes in this region offers the opportunity to study directly the effect of significant localized aerosol perturbations from ship emissions on marine stratocumulus properties, so-called ship tracks [Coakley et al., 1987; Durkee et al., 2000b, 2000c; Schreier et al., 2006]. Indeed, ship emissions released below a stratocumulus deck provide a tailor-made realization of the indirect climatic effect of aerosols. The eastern Pacific region adjacent to the coast of California is, therefore, an ideal test bed for studying aerosol-cloud interactions. The July time frame was selected because it is the month during which the coverage of stratocumulus in this region is at its maximum. A total of 13 science flights were conducted between 2–17 July (Table 1). Measurements were carried out on board the Center for Interdisciplinary Remotely-Piloted Aircraft Studies (CIRPAS) Twin Otter aircraft during the mission. The instrumentation on board the Twin Otter is summarized in Table 2, and the data analysis process for the study is briefly described in the Appendix.
Table 1. Summary of Cloud Properties From Sampled Stratocumulus Cloudsa
Separation of Cloud Droplets From Interstitial Aerosol
Counterflow Virtual Impactor
Cloud Droplet Size Distribution
Phase Doppler Interferometer (PDI)
Cloud and Drizzle Drop Size Distribution
Cloud, Aerosol, and Precipitation Spectrometer (CAPS)b
0.4 μm–1.6 mm
Cloud Droplet Size distribution
Forward Scattering Spectrometer Probe (FSSP)
Cloud Droplet Liquid Water Content
Light diffraction (Gerber PVM-100 probe)
Aerosol Bulk Ionic Composition and Soluble Organic Composition
Particle-into-Liquid Sampler (PILS)
0.02–0.28 μg m−3 (depending on species)
Aerosol Bulk Composition (Nonrefractory Species)
Aerodyne Time-of-Flight Aerosol Mass Spectrometer (TOF-AMS)
1 s or 15 s
<0.25 μg m−3 (all species)
Dva ∼40 nm–1 μm
Photoacoustic Absorption Spectrometer (PAS)
10 nm–5 μm
Particle Soot Absorption Photometer (PSAP)
1 s or higher
 Of the 13 Twin Otter flights that were conducted during MASE, six encountered strong, localized perturbations in aerosol concentration, size, and composition consistent with ship emissions. The subcloud aerosol and its impact on the stratocumulus layer were analyzed using a detailed aircraft cloud profiling strategy (Figure 1). Clear effects of enhanced aerosol loading in these cases were observed in cloud droplet concentration and droplet size distributions. The flight on July 5 (see Table 1) is selected as a case study because of the particularly pronounced effect of ship emissions.
 The MASE experiment was undertaken to address following scientific questions:
 (1) Does cloud droplet number concentration increase in response to increasing subcloud aerosol amount? And, is cloud droplet size reduced?
 (2) Is the drizzle rate suppressed owing to more numerous cloud droplets? Furthermore, does the cloud liquid water path or averaged liquid water content increase because of drizzle inhibition?
 (3) Do we observe an aerosol-induced droplet spectral dispersion effect?
Table 3 lists key questions concerning aerosol indirect effects on marine stratocumulus that have been addressed previously by in situ measurements and LES simulations.
 The region observed on July 5 (denoted cloud 5A in Table 1), sampled over 17:30–19:00 UTC (local time = UTC − 7h), is selected for a detailed case study. Measurements of both aerosol number concentration and composition indicate this region was impacted by ship emissions. Aerosol chemistry measurements, comprising the Aerodyne Time-of-Flight Aerosol Mass Spectrometer (TOF-AMS) (flown for the first time during MASE), the Particle-into-Liquid Spectrometer (PILS), and the Photoacoustic Absorption Spectrometer (from which black carbon concentration is inferred) (Table 2), showed a number of significant features: (1) a fairly concentrated layer of organic carbon aerosol (∼3 μg m−3) overlaid the marine boundary layer (MBL); (2) within the MBL, organic aerosol loading was lower (∼0.5 μg m−3); however, sulfate aerosol loading was comparable to that overlying the MBL; (3) the ship tracks are associated with a strong enhancement in MBL sulfate aerosol (∼0.5 to 2 μg m−3) above the background but with virtually no observable enhancement in organic or black carbon aerosol.
 Extensive horizontal legs flown through the ambient cloud and ship tracks provide fine-scale (∼1 Hz or 50-m resolution) three-dimensional data on the cloud properties. Long flight legs of ∼30 km enabled sufficient averaging to reduce sampling uncertainty and provided insights into the spatial variability at each flight level. We study the impacts of aerosol on the cloud properties at the fine scale by comparing data in the ship track regions against those in the adjacent unperturbed clouds. The aircraft flew through an extensive region of stratocumulus cloud features with several cloud rolls. The flight path is shown against the satellite imagery (Figure 2) from the Geostationary Operational Environmental Satellite (GOES) visible channel (channel 1, 0.65 μm). The cloud rolls were aligned approximately parallel to the mean wind. The ship tracks are most apparent in the GOES near-infrared (IR) channel (channel 2, centered at 3.9 μm, Figure 3). Because liquid water at this wavelength is moderately absorbing, sunlight does not penetrate the cloud deeper than about 100 m, so that the cloud albedo depends mainly on the cloud droplet size in the upper region of the cloud. The near-IR imagery reveals two marked brighter (high reflectance) narrow, curvilinear cloud lines in the sampling period. Between and outside the two brighter ship track impacted clouds are darker regions that are still cloudy from the visible channel picture. The ship track on the west was sampled with nine horizontal traverses (Figure 1), seven through the cloud layers and two above and below clouds, respectively. The horizontal flight legs are numbered based on the time sequence.
3.2. SubCloud Aerosol Number Concentration
 The flight path delineated according to the magnitude of aerosol and cloud droplet number concentrations is presented in Figure 4. The horizontal distance (X) includes a shift by the mean horizontal wind so that the spatially coincident structures of aerosol and cloud drop number concentrations (i.e., near- or below-cloud base aerosol number concentration versus cloud droplet number concentration) are revealed. A noticeable peak (orange to red color) in aerosol number concentration located around X = 59 km is shown below and above cloud (Figure 4a), which is coherent with the apparent peak in cloud drop number concentration in the same location (Figure 4b). Figure 4c shows clearly that the horizontal distribution of cloud droplet number concentration varies with the horizontal distribution of below-cloud aerosol number concentration. Both Condensation Particle Counter (CPC) and Ultrafine CPC (UFCPC) data exhibit covariabilities with cloud droplet number concentration. The difference between data from the UFCPC and the CPC instruments represents the number concentration of ultrafine particles between 3 and 12 nm diameter; this indicates that freshly nucleated particles in the Aitken mode are significant in the major ship track region. Another smaller peak in CPC and UFCPC located around X = 48 − 54 km is also seen at a similar X location in the cloud drop number concentration data. The spatially coherent peaks in aerosol and cloud number concentrations and the narrow curvilinear cloud lines on the satellite imagery suggest that increases of cloud droplet number concentration are associated with and can be ascribed to increases in below-cloud CCN concentration by particles emitted or formed from ship effluent.
 The relationship between cloud droplet number concentration, CDNC, and subcloud aerosol number concentration, Na, is shown in Figure 5. Although the spatial distributions of UFCPC and CPC data measured below cloud visually match the patterns of cloud drop number concentration in the middle to upper cloud (Figure 4), some mismatches do exist. This is because the below-cloud and the middle-to-upper cloud measurements were not carried out simultaneously. The satellite imagery (Figure 3) shows a somewhat broadened line feature in the downstream direction (relative to the northwest mean wind). Meanwhile, the CDNC measurement carried out at the later mission time than the subcloud aerosol measurement (Figure 4) also shows a wider CDNC peak at about X = 48 − 56 km, as compared with the peaks in the CPC and UFCPC data at the similar horizontal location. The wider peak in CDNC is a result of the dispersion of the ship plume from its source. Therefore those CDNC data showing some widening and mismatch with upstream subcloud aerosol data are removed from the regression calculation (Figure 5). The regression results (Figure 5b) suggest that CDNC is highly positively correlated with Na over a wide range of Na, while the corresponding meteorological condition is about constant.
3.3. Cloud and Drizzle
Figure 6 shows the cloud droplet number concentration on each horizontal flight leg. The cloud droplet number distribution exhibits some wave-like structures with several apparent peaks through four horizontal traverses (legs 1, 5, 6, 7, and 8) and more random distribution at legs near cloud base (legs 3 and 4, not shown), probably because the airplane penetrated in and out of cloud base. The width of the major peak is about 5 km (X = 56 − 61 km); the total width of the ship track that includes several wave-like structures is about 21 km (X = 44 − 65 km). (From the Monterey Area Ship Track Experiment (MAST), averaged ship track widths were about 9 ± 5 km [Durkee et al., 2000a]). Outside this broad ship track region, lower cloud droplet number concentrations exist (denoted “Clean west” and “Clean east” in the figure), which are considered to be clean, undisturbed regions. These three cloud regions have about the same cloud depth, facilitating the study of the aerosol indirect effect at essentially constant cloud depth.
 Cloud properties along two horizontal leg flights in the upper regions of the cloud through the solid stratocumulus deck are now explored (legs 7 and 8 in Figures 7 and 8, respectively). Averaged cloud properties in two unperturbed regions and the ship track region shown in Figure 6 are also given in Table 4. Although the horizontal distribution of cloud liquid water content (LWC) seems in phase with that of CDNC in Figures 7 and 8, carefully averaging the cloud LWC over the ship track and clean regions, respectively, shows a higher cloud LWC in the ship track than the clean regions for both legs (Table 4). The averaged drizzle drop number concentration (Nd) is lower in the ship track than the clean regions for both legs. The drizzle LWC is higher in the ship track region for leg 7 but lower in the ship track region for leg 8. For both legs, the drizzle rate is higher in the ship track region.
Table 4. Summary of the Averaged Cloud Properties Sampled in (a) Leg 7 and (b) Leg 8 for the Ship Track Region and the Neighboring Relatively Cleaner Regionsa
cloud LWC, g m−3
drizzle LWC, g L−1
Fdrz, mm day−1
cloud rm, μm
drizzle rm, μm
cloud re, μm
cloud σ, μm
See Figure 6, source for definitions of the regions.
Value in the parenthesis is integrated over the combined cloud and drizzle spectrum.
 In sum, we see cloud LWC at the upper portion of the cloud increased in the ship track region because of drizzle suppression as reflected by drizzle drop number concentration. We also note that this selected cloud case (cloud 5A) is one of the most strongly drizzling clouds (cloud base drizzle rate ∼0.5 mm day−1) among all those sampled.
3.4. Cloud Drop Spectral Dispersion
 As noted in section 1, changes in subcloud aerosol number concentration can affect the shape of the cloud drop size distribution, the so-called dispersion effect, influencing cloud albedo. Cloud droplet effective radius (re) is a key parameter that links cloud microphysics with cloud optical properties in large-scale models [Slingo and Schrecker, 1982]. Parameterization of re is generally represented through the parameter k, defined by
where rv is the volume mean radius of the cloud droplet size distribution. By the definition that rv = (3LWC/4πρw CDNC)1/3, where ρw is the density of water,
The coefficient k can be expressed analytically as a function of droplet relative dispersion, d = σ/rm, as [Lu and Seinfeld, 2006]
where σ is the standard deviation (or cloud drop spectral width), rm is the mean radius of the droplet distribution, and s is the skewness of the drop spectrum, generally assumed to be zero.
 Through the relationship between k and d, cloud spectral shape affects cloud optical properties. Liu and Daum  suggested that an increase in aerosol number concentration would lead to an increase in relative dispersion and a cloud darkening. In their work, not all cloud cases show the suggested increasing trend of d as a result of pollution. The polluted and clean cloud cases were not necessarily subject to identical meteorological soundings and may have experienced various updraft velocities. By contrast, on the basis of LES simulations assuming a constant meteorological condition but varying only aerosol number concentration, Lu and Seinfeld  demonstrated that stratocumulus cloud would evidence the opposite effect, namely, a decrease in relative dispersion and a consequent cloud brightening.
 Patterns in Figures 7 and 8 reveal interesting spatial correlations of several variables with cloud droplet number concentration. For example, the horizontal distributions of d and σ are out of phase with that of CDNC, while the parameter k approximately varies directly with CDNC. The relationship between relative dispersion and CDNC displayed in Figure 9 for both cloud and cloud-and-drizzle (referred to as combined cloud and drizzle spectra, hereafter) is more pronounced for leg 7 than for leg 8 but is still discernable for leg 8; similar results hold for the relationship between k and CDNC. The value of k is about 10% smaller when drizzle drops are taken into account. The value of k is found to depend inversely on relative dispersion because spectral skewness (s) is generally less than unity as shown in Table 4. Averaged dispersion effects in ship track and clean regions show that the former features smaller d and σ and larger k than its cleaner counterpart for legs 7 and 8; this is also the case when the drizzle size distribution is taken into account in calculating d, σ, and k.
 The range of values of sampled rm, σ, and d in legs 7 and 8 (Figures 7, 8, and 9) are generally in agreement with those reported from several stratocumulus clouds in the Second Aerosol Characterization Experiment (ACE2) shown in Figure 1 (line) of Pawlowska et al.  compared at the same range of CDNC. Furthermore, the fine-scale inverse relationships rm-CDNC, σ-CDNC, and d-CDNC sampled in horizontal legs 7 and 8 (scale ∼50 m) are consistent with in-flight results (scale ∼10 m) from ACE2. Pawlowska et al. assumed no in-flight horizontal variability of aerosol and attributed the in-flight observations of the decrease of σ and d with CDNC to the variability of cloud base updraft velocity. However, because of negligible variations in the cloud base updraft between ship track and clean regions, the MASE in-flight relationships of rm-CDNC, σ-CDNC, and d-CDNC can be attributed to the subcloud aerosol number concentration. Therefore given the strong correlation of subcloud aerosol number concentration with CDNC (Figure 5), we illustrate the possible dependence of rm (inverse), σ (inverse), d (inverse), and k (direct) on subcloud aerosol number concentration. The relationships σ-Na, d-Na, k-Na, and k-d also agree with those derived from the LES studies by Lu and Seinfeld . We will return to the dependence of d on CDNC at the “ensemble cloud scale” from all clouds sampled in MASE in Section 4.
3.5. Case Study Summary
 In summary, the upper region of the ship track-influenced cloud is characterized by higher cloud droplet number concentration, smaller relative dispersion, larger k, fewer drizzle drops, smaller cloud mean radius, and narrower cloud drop spectral width. Drizzle suppression by the ship track effluent is identified, which shows a lower drizzle drop number concentration in the ship track region. This likely leads to a higher upper cloud LWC in the ship track region.
4. Ensemble-Averaged Clouds Sampled in MASE
 Ensemble scale (∼several tens of kilometers; i.e., by averaging data along horizontal legs) impacts of subcloud aerosol on cloud properties are now explored. Specifically, we address the relationships of cloud droplet number concentration, cloud effective radius, drizzle rate, cloud LWP, and relative dispersion on subcloud aerosol number concentration. We select 13 sampled clouds from the set of MASE flights for analysis. The clouds exhibit a wide range of subcloud aerosol number concentrations, from 70 to 1300 cm−3, where in some cases ship tracks were present. Cloud base and top also show some degree of variation among cases. This offers an opportunity to study aerosol impacts on cloud microphysics and dynamics under a broad range of conditions. Table 1 summarizes properties of the MASE clouds studied.
4.1. Ensemble-Averaged Cloud and Aerosol Number Concentration
 The number of cloud droplets formed is controlled principally by the number of CCN at cloud base and vertical velocity. We determine the subcloud aerosol number concentration by averaging the measured aerosol number concentration within about 100 m below cloud base. The vertical distribution of aerosol number concentration below cloud base is more or less uniform if the cloud is not decoupled from the surface or the aerosol is not influenced by ship emissions. In ship track regions, the subcloud aerosol consists of a mixture of ambient marine aerosol and ship emissions. Nevertheless, the chemical composition of the subcloud aerosol composition for all MASE clouds is basically dominated by ammonium bisulfate, (NH4)HSO4. The subcloud dry aerosol size distributions do not show significant variations in shape among all clouds, and they can be approximately characterized by two evident modes with mode radius around 25 nm and 100 nm. Measured vertical distributions of cloud droplet number concentrations tend to be roughly constant in the broad middle portion of the cloud with smaller values near cloud top and base owing to cloud-clear air interfacial mixing. Some cases, for example that on 5 July, which exhibited a constant CDNC in a narrower middle region of the cloud, also exhibited much smaller CDNC concentration in the lower portion of the cloud, which could be associated with scattered cumulus below the well-mixed stratocumulus. The ensemble average of aircraft sampled data for each case shows a direct dependence of CDNC on subcloud aerosol number concentration, as expected (Figure 10). This relationship can be approximately represented with the power law function shown in Figure 10. The correlation is improved when cloud base updraft velocity variability (using σw, see Table 1) is taken into account,
Results from observations of stratocumulus by Martin et al.  are also displayed in the figure, where the clouds were sampled over the eastern Pacific Ocean near California, South Atlantic, sea around the British Isles, and North Atlantic near Azores.
4.2. Ensemble-Averaged Cloud Drop Effective Radius
 Cloud radiative properties, e.g., cloud optical depth and albedo, are most influential near cloud top; therefore, re shown in Table 1 is calculated by averaging observations over the upper third of the cloud depth. The effective radius integrated over the cloud-and-drizzle spectrum is larger than that integrated over the cloud spectrum by about 0.1 to 0.9μm; the difference is approximately proportional to the drizzle drop number concentration. Figure 11 shows clearly the decreasing trend of cloud droplet effective radius with increasing cloud droplet number concentration and subcloud aerosol number concentration for all 13 clouds. These trends continue to hold when the combined cloud and drizzle effective radius is used (not shown). We further group the data based on the cloud LWP. Figure 11 shows that the expected inverse relationship between re and CDNC is still strong for each LWP category. Therefore the overall feature of the MASE clouds is a smaller effective radius in more polluted clouds at the same LWP, a clear demonstration of the first indirect effect.
 The measured vertical profile of drizzle number concentration increases roughly linearly with height in the cloud and reaches a maximum value near cloud top for 11 out of 13 clouds sampled (not shown). The vertical profile of drizzle drop mean radius increases asymptotically with decreasing height, and the largest drizzle drops occur near or below cloud base. The maximum value of Nd near cloud top, together with the smallest drizzle drop mean size, largest cloud droplet size, and spectral distribution, suggests that the drizzle drops near the cloud top are those freshly formed by cloud droplet coalescence and condensational growth. The few but large drizzle drops near the cloud base are a result of drizzle accretion of cloud droplets and drizzle self-collection. To examine the relationship between drizzle and aerosol number concentration, we separate the problem into two steps: first, we explore the dependence of embryonic drizzle drops on CDNC (given that we have shown a strong positive correlation between Na and CDNC) to see whether the sources of drizzle drops are affected by Na. Secondly, we explore the dependence of both cloud base drizzle rate and LWP on CDNC (section 4.4) to see whether drizzle depletes in-cloud water in responses to CDNC changes.
Figure 12a displays the observed relationship between drizzle drop number concentration (Nd) and CDNC. The drizzle drop number concentration is averaged over the upper third of the cloud layer to represent precipitation embryos. The data points in Figure 12a can be basically separated into two groups (by two circles): The cleaner cloud group (CDNC ∼70–200 cm−3 and Nd ∼0.2–0.8 cm−3) clearly is characterized by more drizzle drops than the polluted cloud group (CDNC ∼220–350 cm−3 and Nd < 0.2 cm−3). There is less distinguishable dependence of Nd on CDNC on Figure 12c. This is because the number of embryonic drizzle drops is influenced both by the CDNC and the LWC amounts. Therefore, multiple regression of these three variables in log-log space yields a beter correlation than that in Figure 12c:
This form of relationship is similar to the Kessler  type parameterization of the warm rain autoconversion rate, which is the rate of drizzle drop production employed in numerical models (e.g., Wood [2005b]). We note that cloud 10 is not consistent with the other clouds, as it shows particularly low Nd with lowest CDNC. This case also exhibits notably smaller drizzle drops (Figure 12b). It is possible that this case has evident decoupled (discontinuous) features in LWC and CDNC vertical profiles (not shown) and significantly less intensive subcloud turbulence (σw in Table 1). Figure 12b shows an anti-correlation between drizzle drop mean radius (rd) and CDNC with moderate correlation.
 As drizzle drops fall below cloud base, cloud LWP is reduced, in the absence of other dynamical changes. The cloud base drizzle rate (RCB) is a proxy for the extent to which drizzle depletes liquid water in cloud. As discussed above, cloud base drizzle drops are formed as a result of collision and coalescence of cloud top embryonic drizzle falling through the cloud layer. Hence RCB shows a strong positive dependence on cloud top Nd (Figure 12c). Together with the correlations for Nd-RCB and CDNC-Nd, it is not surprising that the observational relationship between RCB and CDNC in Figure 13 shows that cleaner clouds have a larger cloud base drizzle rate. In this respect, the current results are consistent with other in situ aircraft data of marine stratocumulus, e.g., the second Dynamics and Chemistry of Marine Stratocumulus (DYCOMS-II) [van Zanten et al., 2005], and the Atlantic Stratocumulus Transition Experiment (ASTEX) first Lagrangian experiment and the U.K. Met Office C-130 aircraft studies [Wood, 2005a]. We note that the maximum ensemble-averaged RCB listed in Table 1 is about 0.5 mm day−1, so MASE clouds are classified to exhibit light to moderate drizzle rate based on a 1 mm day−1 heavy drizzle threshold [e.g., Lu and Seinfeld, 2005 and van Zanten et al., 2005]. Summarizing Figures 12 and 13, the MASE data show that more polluted (thereby higher CDNC) clouds have fewer and smaller drizzle drops near cloud top and a smaller cloud base drizzle rate than their cleaner counterparts, which exemplifies drizzle suppression by pollution.
4.4. Ensemble-Averaged Liquid Water Path
 The ensemble-averaged LWP as a function of subcloud aerosol number concentration is displayed in Figure 14. In this figure, the cloud depth (H)-adjusted mean LWC, the ratio of LWP to H, is used as a surrogate for LWP, so that the dependence of LWP on H is eliminated. The amount of drizzle LWP is significantly less (the difference between the circles and triangles) for the two most polluted clouds. The entire data set appears to be separable into two groups (dotted line) at Na = 200 cm−3. For conditions with Na > 200 cm−3, there exists a clear negative correlation of either cloud or cloud-and-drizzle LWP with Na; that is, the more polluted cloud has a smaller cloud LWP. Even though evidence is found for drizzle suppression, the MASE clouds exhibit a weak to moderate drizzle rate so that drizzle does not appear to deplete enough in-cloud LWP to produce a positive LWP-versus-Na relationship. This indicates that when the cloud base drizzle is weak, drizzle has negligible impact on the MBL moisture budget, which is then more likely determined by the cloud top entrainment drying [Ackerman et al., 2004; Lu and Seinfeld, 2005]. There is, however, a hint of a positive LWP-versus-Na relationship by comparing the LWP of the cleanest cloud (cloud 10, Na < 200 cm−3) with those of Na = 200–850 cm−3. Cloud 10 occurred in a decoupled MBL with relatively weaker subcloud turbulence than other clouds (Table 1). On the basis of only one cloud, one cannot draw an overly general conclusion.
 Extracting any simple and direct mechanism of LWP reduction by drizzle is complicated by drizzle intensity, evaporation of drizzle below cloud base, and dynamical adjustment of cloud when drizzling, so the above conclusions should be viewed with caution.
Lu and Seinfeld  carried out LES simulations of marine stratocumulus and investigated the factors that control the cloud droplet spectral relative dispersion in response cloud doplet soctral relative dispersion to increasing aerosol number concentration. They found that cloud spectral relative dispersion decreases with increasing aerosol number concentration particularly at low Na (Na ≤ 1000 cm−3), in agreement with the observational data derived from the work of Miles et al.  of 17 separate studies. They showed that this trend of decreasing d with increasing Na is a result of more rapid decrease of σ than rm with Na. They also found that the coefficient k depends inversely on d. They attributed this result to suppressed drizzle at increased aerosol loading, which results in less collision and coalescence spectral broadening and more spectral narrowing by droplet condensational growth at higher updraft velocities dynamically induced by suppressed drizzle.
 Ensemble-averaged vertical profiles of cloud mean radius, cloud spectral width, and relative dispersion are shown in Figure 15 and compared against data from ACE2 [Pawlowska et al., 2006]. Both MASE (Figure 15a) and ACE2 (their Figures 2 and 3) show rm increases with altitude. Most clouds sampled in MASE start with rm of 2.5–3.5 μm at cloud base, and grow into a larger range of 4.5–10 μm in response to variations in Na. Vertical profiles of σ (Figure 15b) are generally nearly constant with height, with some cases showing a slight increasing trend with altitude near cloud top (clouds 5A, 5B, 8, 9, and 14). These features are consistent with the polluted clouds (both adiabatic and nonadiabatic regions) and clean clouds (in the nonadiabatic regions) of ACE2 [Pawlowska et al., 2006]. Unlike the nondrizzling stratocumulus described in Martin et al. , in which relative dispersion is constant with height, data from both MASE and ACE2 show vertical variations in d. The MASE vertical profiles of d are similar (Figure 15c, except cloud 10) to that of the ACE2 marine cloud (their Figure 2), in which dispersion decreases with increasing altitude. The large value of d near cloud base is a result of cloud base mixing and updraft velocity variance. We also note that MASE vertical profiles show d = 0.2–0.6 for 13 clouds, while the ACE2 marine cloud shows d = 0.1–0.2. (The smaller vertical variation in d for ACE2 might be due to the fact that there is only marine cloud considered).
 The relative dispersion calculated over the combined cloud and drizzle drop spectra is larger than that calculated over the cloud spectrum only (Table 1), and the difference is roughly proportional to the drizzle drop number concentration.
 The ensemble-averaged values of cloud and drizzle combined spectral width (cloud-and-drizzle σ) versus CDNC (Figure 16a) show that cleaner clouds (low CDNC) have broader spectral width. The mean radius (rm) is similar to re that decreases with CDNC or Na (not shown). However, the relationship between cloud-and-drizzle d with CDNC in Figure 16b is less obvious. We note that the data can be roughly separated into three groups: a particularly high value of d for the cleanest cloud (cloud 10, d = 0.41), a relatively small value of d for the two points at CDNC ≈ 280 cm−3 (cloud 15, d = 0.20; cloud 17, d = 0.26), and in between, a group of data points with medium value of d. The points in the middle group seem to show a positive correlation between d and CDNC. Although we see less robust patterns in the d-CDNC plot, Figure 17 shows that MASE d values are basically in agreement with observations of several stratiform clouds in the study of Miles et al.  and LES simulations of the First International Satellite Cloud Climatology Project Regional Experiment (FIRE) and ASTEX [Lu and Seinfeld, 2006]. This figure also shows that d is relatively larger for CDNC <100 cm−3 than d for CDNC >100 cm−3. The impact of aerosol number concentration on cloud relative dispersion can be best represented by the plot of d-Na because Na is the basic parameter for the level of pollution and CDNC is indirectly determined by Na and updraft velocity. The d-Na relation is presented in Figure 16c, which shows a positive correlation between d and Na. The ensemble-averaged results seem to suggest that more polluted clouds have a larger relative dispersion, in agreement with other flight-averaged data [Martin et al., 1994; Liu and Daum, 2002; Pawlowska et al., 2006]. However, it should be pointed out that studies by Liu and Daum  and Pawlowska et al.  use CDNC instead of Na as the basis for the dispersion correlation. On the basis of the good relationship of ensemble-averaged results of d-Na and the previous reasoning, we suggest that d-Na maybe a better surrogate for the dispersion effect. We also note that the “in-flight” relationship of d-CDNC for MASE and ACE2 [Pawlowska et al., 2006] as analyzed in section 3.4 shows the opposite trend from the “ensemble-averaged” relationship of d-Na shown here. The LES study of Lu and Seinfeld  under the same sounding profile also predicts the in-flight relationship between d and Na. These results suggest that the ensemble-averaged relationships between d and Na (or CDNC) are affected not only by Na but also by dynamical conditions such as cloud base updraft velocity and cloud top mixing, which vary between cloud cases because of different sampling locations (Table 1).
 The coefficient k is calculated over the upper third of the cloud depth following the same reasoning as for the effective radius. The flight-averaged cloud (cloud-and-drizzle) k is shown to correlate highly with cloud (cloud-and-drizzle) d (Figure 18a); which suggests that relative dispersion is a suitable proxy for parameterization of k. For this reason, the scatterplot of cloud-and-drizzle k versus CDNC (Figure 18b), is similar to that of d versus CDNC, which shows no simple dependence. It is shown that k is smaller for those cases with CDNC <250 cm−3 than k for CDNC >250 cm−3. The k-Na relationship is shown in Figure 18c; as expected from results of d, k shows a strong inverse dependence with Na. For cloud with Na about 1000 cm−3, k is about 0.65–0.7, and for Na = 300–400 cm−3, k is about 0.8. MASE flight-averaged results are close to the analysis of stratocumulus by Martin et al. , who showed cloud k = 0.67 for continental cloud (aerosol number concentration = 375–1500 cm−3) and k = 0.80 for maritime cloud (aerosol number concentration = 36–280 cm−3). However, again, the regression between k and Na has been performed in an Na region that is close to Martin’s continental cloud conditions.
5. Summary and Conclusion
 The MASE field campaign was undertaken in July 2005 to evaluate aerosol-cloud relationships in the climatically important regime of eastern Pacific marine stratocumulus. Thirteen clouds sampled in the region 123.5–121.5°W and 35.75–36.75°N were selected for detailed analysis; subcloud aerosol number concentrations varied from 70 to 1300 cm−3, with some cases exhibiting ship tracks. Among the clouds sampled, that observed on July 5 was clearly impacted by a ship track, as confirmed by in situ aircraft measurements and GOES near-IR satellite imagery. Multiple airborne horizontal traverses (∼30 km) through the ship track and unperturbed regions provide insights into variations of cloud properties on a 50-m (∼1 Hz) scale both horizontally and vertically. Comparison of ship track and clean regions in the upper portion of the cloud shows that the ship track region exhibited a smaller cloud drop effective radius, reduced drizzle drop number concentration, and larger cloud LWC than the adjacent clean regions. Therefore the data from this cloud provide evidence for the first indirect effect, as well as drizzle suppression by enhanced aerosol number concentration and corresponding larger cloud LWC as a consequence of reduced depletion of cloud water by drizzle. The ship track region also exhibits a smaller cloud drop spectral width and relative dispersion, in accord with LES predictions of Lu and Seinfeld  based on meteorological conditions in the ASTEX and FIRE experiments.
 The ensemble average of aerosol and cloud conditions over the 13 cloud regions sampled were also computed. For these, as subcloud aerosol number concentration increases, cloud drop number concentration increases. The magnitudes of aerosol and cloud drop number concentrations and their dependences lie roughly in the range as those reported by Martin et al.  for a variety of locations. Averaging over all 13 clouds, as Na (or CDNC) increases, the cloud drop effective radius decreases, a robust demonstration of the Twomey effect.
 Drizzle was prevalent in virtually all the clouds sampled. Drizzle drops were found to be most numerous and smallest near cloud top. The data suggest that more polluted clouds have fewer embryonic drizzle drops near cloud top, resulting in a smaller cloud base drizzle rate. The smaller cloud base drizzle rate, however, does not result in a larger LWP when compared with the clean clouds. It appears that direct conversion of cloud drops to drizzle is insufficient to explain the dependence of LWP on aerosol number concentration; drizzle intensity and the dynamic adjustment of the cloud in response to drizzle in-cloud latent heating, subcloud evaporative cooling, and cloud top entrainment would need to be taken into consideration [Ackerman et al., 2004; Lu and Seinfeld, 2005].
 Averaged results over all 13 clouds show that more polluted clouds have narrower spectral width. The MASE data exhibit a clear positive correlation between cloud drop spectral dispersion d as a function of Na.
 In general, on the ensemble cloud scale, increases in aerosol number concentration result in cloud LWP decreases and dispersion broadening acting together to diminish the Twomey effect; while on the scale of a single ship track-perturbed cloud, cloud LWP enhancement and dispersion narrowing act to increase the overall aerosol indirect effects. The contrast is due to the fact that, at the ensemble scale, clouds are influenced by different meteorological conditions that result in different cloud top entrainment, cloud base updraft variance, drizzle intensity, and surface evaporation; at the scale of a single cloud, the ship track and clean regions are embedded in the similar sounding profile, and the aerosol-induced changes are compared irrespective of the variations in LWP and dispersion due to ambient conditions.
Appendix A:: Data Analysis
 The instrumentation payload on the Twin Otter aircraft is listed in Table 2. The nominal airspeed of the Twin Otter is 50 m s−1. Figure 1 shows the flight path on 5 July 2005; this flight path was typical of those in the experiment. Below- and above-cloud aerosols were measured, both size and composition. Cloud droplet size distributions were measured by the forward scattering spectrometer probe (FSSP), and cloud droplets are defined as those having radius greater than 1.56 μm and smaller than drizzle drops. Drizzle drop size distribution was measured by the cloud-imaging probe (CIP), and drizzle drops are typically defined as those having radius greater than 20 μm. Cloud base and top are determined using the FSSP threshold of >5 cm−3, LWC >0.001 g cm−3, and RH must be greater than 100%. The aerosol number concentration, Na, was obtained by the CPC. Wind velocities were measured by the five-hole turbulence probe on the nose of the aircraft, C-MIGITS or TANS-Vector inertial navigation systems, and C-MIGITS or Nova Tel GPS following Lenschow . A Gaussian filter was applied to the vertical wind velocity data to remove any long-period (scale greater than 2.5 km) drift possibly associated with unmeasured changes in the aircraft state. The standard deviation of cloud base updraft velocity (σw) is calculated by the standard deviation of vertical velocity >0 along the flight leg near cloud base.
 For the ensemble-averaged cloud properties presented in section 4, we first calculate the mean value of cloud properties (e.g., CDNC, …etc.) in each altitude section of 30 m, and then perform the vertical average. Cloud droplet number concentrations are averaged over the cloud depth. Liquid water path is calculated by vertically integrating the averaged liquid water content in each altitude bin of 30 m. Drizzle drop number concentration, cloud LWC effective radius, relative dispersion, and k coefficient are averaged over the upper third of cloud depth. Drizzle rate is averaged over the lower third of cloud depth.
 This work was supported by the Office of Naval Research (N-00014-04-1-0018) and National Science Foundation grant ATM-0340832. PILS data were provided by Armin Sorooshian. We appreciate valuable comments from Patrick Chuang and Robert Wood.