• Ackerman, A. S., , O. B. Toon, , D. E. Stevens, , and J. A. Coakley Jr., 2003: Enhancement of cloud cover and suppression of nocturnal drizzle in stratocumulus polluted by haze. Geophys. Res. Lett., 30, 1381, doi:10.1029/2002GL016634.

    • Search Google Scholar
    • Export Citation
  • Ackerman, A. S., , M. P. Kirkpatrick, , D. E. Stevens, , and O. B. Toon, 2004: The impact of humidity above stratiform clouds on indirect aerosol climate forcing. Nature, 432, 10141017.

    • Search Google Scholar
    • Export Citation
  • Albrecht, B. A., 1989: Aerosols, cloud microphysics, and fractional cloudiness. Science, 245, 12271230.

  • Christensen, M. W., , and G. Stephens, 2011: Microphysical and macrophysical responses of marine stratocumulus polluted by underlying ships: Evidence of cloud deepening. J. Geophys. Res., 116, D03201, doi:10.1029/2010JD014638.

    • Search Google Scholar
    • Export Citation
  • Christensen, M. W., , J. A. Coakley Jr., , and W. R. Tahnk, 2009: Morning-to-afternoon evolution of marine stratus polluted by underlying ships: Implications for the relative lifetimes of polluted and unpolluted clouds. J. Atmos. Sci., 66, 20972106.

    • Search Google Scholar
    • Export Citation
  • Coakley, J. A., Jr., , and C. D. Walsh, 2002: Limits to the aerosol indirect radiative effect derived from observations of ship tracks. J. Atmos. Sci., 59, 668680.

    • Search Google Scholar
    • Export Citation
  • Coakley, J. A., Jr., , R. L. Bernstein, , and P. A. Durkee, 1987: Effect of ship-stack effluents on cloud reflectivity. Science, 237, 10201022.

    • Search Google Scholar
    • Export Citation
  • Coakley, J. A., Jr., , M. A. Friedman, , and W. R. Tahnk, 2005: Retrievals of cloud properties for partly cloudy imager pixels. J. Atmos. Oceanic Technol., 22, 317.

    • Search Google Scholar
    • Export Citation
  • Conant, W. C., and Coauthors, 2004: Aerosol–cloud drop concentration closure in warm cumulus. J. Geophys. Res., 109, D13204, doi:10.1029/2003JD004324.

    • Search Google Scholar
    • Export Citation
  • Ferek, R. J., and Coauthors, 2000: Drizzle suppression of ship tracks. J. Atmos. Sci., 57, 27072728.

  • Fountoukis, C., and Coauthors, 2007: Aerosol–cloud drop concentration closure for clouds sampled during the International Consortium for Atmospheric Research on Transport and Transformation 2004 campaign. J. Geophys. Res., 112, D10S30, doi:10.1029/2006JD007272.

    • Search Google Scholar
    • Export Citation
  • Hayes, C. R., , J. A. Coakley Jr., , and W. R. Tahnk, 2010: Relationships among properties of marine stratocumulus derived from collocated CALIPSO and MODIS observations. J. Geophys. Res., 115, D00H17, doi:10.1029/2009JD012046.

    • Search Google Scholar
    • Export Citation
  • Kaufman, Y., , I. Koren, , A. Remer, , D. Rosenfeld, , and Y. Rudich, 2005: The effect of smoke, dust, and pollution aerosol on shallow cloud development over the Atlantic Ocean. Proc. Natl. Acad. Sci. USA, 102, 11 20711 212.

    • Search Google Scholar
    • Export Citation
  • King, M. D., , L. F. Radke, , and P. V. Hobbs, 1993: Optical properties of marine stratocumulus clouds modified by ships. J. Geophys. Res., 98, 27292739.

    • Search Google Scholar
    • Export Citation
  • Loeb, N. G., , and G. L. Schuster, 2008: An observational study of the relationship between cloud, aerosol and meteorology in broken low-level cloud conditions. J. Geophys. Res., 113, D14214, doi:10.1029/2007JD009763.

    • Search Google Scholar
    • Export Citation
  • Martin, G. M., , D. W. Johnson, , and A. Spice, 1994: The measurement and parameterization of effective radius of droplets in warm stratocumulus clouds. J. Atmos. Sci., 51, 18231842.

    • Search Google Scholar
    • Export Citation
  • Matheson, M. A., , J. A. Coakley Jr., , and W. R. Tahnk, 2005: Aerosol and cloud property relationships for summertime stratiform clouds in the northeastern Atlantic from Advanced Very High Resolution Radiometer observations. J. Geophys. Res., 110, D24204, doi:10.1029/2005JD006165.

    • Search Google Scholar
    • Export Citation
  • Matheson, M. A., , J. A. Coakley Jr., , and W. R. Tahnk, 2006: Effects of threshold retrievals on estimates of the aerosol indirect radiative forcing. Geophys. Res. Lett., 33, L07705, doi:10.1029/2005GL025614.

    • Search Google Scholar
    • Export Citation
  • Meskhidze, N., , A. Nenes, , W. C. Conant, , and J. H. Seinfeld, 2005: Evaluation of a new cloud droplet activation parameterization with in situ data from CRYSTAL-FACE and CSTRIPE. J. Geophys. Res., 110, D16202, doi:10.1029/2004JD005703.

    • Search Google Scholar
    • Export Citation
  • Platnick, S., , and S. Twomey, 1994: Determining the susceptibility of cloud albedo to changes in droplet concentration with the Advanced Very High Resolution Radiometer. J. Appl. Meteor., 33, 334347.

    • Search Google Scholar
    • Export Citation
  • Platnick, S., and Coauthors, 2000: The role of background cloud microphysics in the radiative formation of ship tracks. J. Atmos. Sci., 57, 26072624.

    • Search Google Scholar
    • Export Citation
  • Quaas, J., , O. Boucher, , N. Bellouin, , and S. Kinne, 2008: Satellite-based estimate of the direct and indirect aerosol climate forcing. J. Geophys. Res., 113, D05204, doi:10.1029/2007JD008962.

    • Search Google Scholar
    • Export Citation
  • Radke, L. F., , J. A. Coakley Jr., , and M. D. King, 1989: Direct and remote sensing observations of the effects of ships on clouds. Science, 246, 11461149.

    • Search Google Scholar
    • Export Citation
  • Segrin, M. S., , J. A. Coakley Jr., , and W. R. Tahnk, 2007: MODIS observations of ship tracks in summertime stratus off the west coast of the United States. J. Atmos. Sci., 64, 43304345.

    • Search Google Scholar
    • Export Citation
  • Snider, J. R., , and J.-L. Brenguier, 2000: Cloud condensation nuclei and cloud droplet measurements during ACE-2. Tellus, 52B, 828842.

    • Search Google Scholar
    • Export Citation
  • Snider, J. R., , S. Guibert, , J.-L. Brenguier, , and J.-P. Putaud, 2003: Aerosol activation in marine stratocumulus clouds: 2. Köhler and parcel theory closure studies. J. Geophys. Res., 108, 8629, doi:10.1029/2002JD002692.

    • Search Google Scholar
    • Export Citation
  • Stevens, B., , and G. Feingold, 2009: Untangling aerosol effects on clouds and precipitation in a buffered system. Nature, 461, 607613.

    • Search Google Scholar
    • Export Citation
  • Twomey, S., 1959: The nuclei of natural cloud formation. Part II: The supersaturation in natural clouds and the variation of the cloud droplet concentration. Geofis. Pura Appl., 43, 243249.

    • Search Google Scholar
    • Export Citation
  • Twomey, S., 1974: Pollution and the planetary albedo. Atmos. Environ., 8, 12511256.

  • Zhang, Z., , and S. Platnick, 2011: An assessment of differences between cloud effective particle radius retrievals for marine water clouds from three MODIS spectral bands. J. Geophys. Res., 116, D20215, doi:10.1029/2011JD016216.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Image of ship track crossings constructed from 3.7-μm radiances and the automated identification of the ship track pixels (blue), the control pixels (red), and the crossing pixels (green) for tracks 10 and 11. The observations were for the MODIS Aqua at 2155 UTC 29 Jul 2003, ranging from 39° to 41°N and 127° to 132°W. Tracks 10 and 11 cross near the center of the image, while tracks 10 and 12 cross near the right edge.

  • View in gallery

    Schematic illustrating the procedures used to analyze the crossing of two ship tracks. The solid lines represent the boundaries of the ship tracks. The portions of the ship tracks analyzed in this study are rendered in light and dark blue. The ship tracks cross at the box near the center of the diagram rendered in green. The dashed lines indicate the extent of the pixels on both sides of the ship tracks that contain the unpolluted clouds that were used as controls. The control pixels analyzed in this study are those rendered in red. Each rectangle in the figure represents a five-pixel segment along the ship track. Properties for the clouds in the ship tracks and for the nearby unpolluted clouds were averaged separately within each five-pixel segment along the ship track. At the crossing, the properties of each ship track not affected by the second ship were estimated using linear least squares fits of the properties along the track for up to four of the five-pixel segments nearest the crossing on both sides. The estimated cloud properties were then compared with those of the pixels at the crossing to deduce the effects of the second ship. The completely colored five-pixel segments represent the 20-km segments used to infer the effects of the individual ships on the unpolluted clouds. These segments were at least 20 km from the pixels associated with the crossing.

  • View in gallery

    Average droplet effective radius for five-pixel along-track segments derived using 3.7-μm radiances, and distance from the center of the crossing for the ship (○), control 1 (+), control 2 (×), and crossing (•) pixels of tracks 10 and 11 shown in Fig. 1. Control 1 and control 2 refer to the control pixels on opposite sides of the ship tracks. All averages were for overcast pixels. Least squares trend lines are shown for the averages of the ship track pixels within the first four five-pixel along-track segments on each side nearest the crossing that satisfied the conditions required for inclusion in the analysis as described in the text. Negative distances from the crossing are associated with the leg of the track that contains the head, the pixels nearest the polluting ship.

  • View in gallery

    (a) Average 3.7-μm derived cloud droplet effective radius Re for the ship track pixels (solid line) and the control pixels (dashed line), (b) visible optical depth, (c) liquid water path, and (d) column droplet number concentration. The averages are for the overcast pixels in the first 20-km along-track segments that fell on either side of a ship track crossing and were more than 20 km from the crossing. Number of samples, means, and 95% confidence intervals are shown.

  • View in gallery

    As in Fig. 4, but for the average differences between the overcast ship track and control pixels (ship − control; solid line) and the overcast control pixels on either side of the ship track (control 1 − control 2; dashed line).

  • View in gallery

    Differences in average cloud droplet effective radius Re between the overcast pixels in the ship tracks and those in the nearby controls for 20-km ship track segments 20 km from the ship track crossing (solid line) and average differences for the overcast crossing pixels and the projected values at the crossing (dashed lines) for the (a) subordinate and (b) dominant ship track. The number of crossings, mean differences, and 95% confidence intervals for the differences are given.

  • View in gallery

    As in Fig. 6, but for visible optical depth.

  • View in gallery

    As in Fig. 6, but for liquid water path.

  • View in gallery

    As in Fig. 6, but for column droplet number concentration.

  • View in gallery

    Average changes in column droplet number concentration due to pollution from the ship and the existing droplet number concentration for the overcast control pixels in 20-km ship track segments 20 km from the crossing for the dominant and subordinate ship tracks and for the crossing pixels. At the crossing, the changes for the dominant ship track are from the values projected at the crossing for the subordinate ship and the changes for the subordinate ship track are from the values projected for the dominant ship. The line shows the result of the least squares fit given by (3).

  • View in gallery

    (a) Image of 2.1-μm reflectivity from the Aqua MODIS off the coast of Northern California at 2140 UTC 26 Jul 2004, also shown with (b) the hand-logged track positions overlaid, (c) pixels identified as polluted by the automated scheme shown in color indicating the 2.1-μm derived droplet effective radius, and (d) pixels identified as uncontaminated control pixels on both sides of the polluted pixels shown in color. The tracks identified as 9 and 11 crossed at 40.5°N, 134.4°W at the time of the overpass. The × symbols in (c) and (d) indicate the central locations associated with the polluted pixels of the five-pixel segments along the track used to accumulate information on both the polluted and uncontaminated control pixels. To reduce clutter only the centers of five-pixel segments separated by 20 pixels along the track are shown.

  • View in gallery

    The 2.1-μm reflectivity (%, small dots) of polluted pixels and along-track distances to the crossing for tracks 9 and 11 shown in Fig. A1. Negative distances are used for the leg of the track between the track head, the point nearest the ship, and the crossing. The smallest reflectivity for pixels identified as polluted are given by the means minus 1.5 times the standard deviations of the pixel-scale reflectivity accumulated in the five-pixel segments along the track and are indicated by large dots joined by lines. The dashed line is a least squares estimate of the smallest reflectivity associated with polluted pixels for each track in the vicinity of the crossing. It was used to predict the lower bound of the reflectivity for the polluted pixels at the crossing.

  • View in gallery

    As in Fig. A2, but for the cross-track widths of the domain containing pixels identified as polluted using the automated scheme. The dots give the means of the widths and the error bars give the standard deviations for five-pixel along-track segments. Pixels that had sufficiently large 2.1-μm reflectivities and fell within the predicted cross-track width were identified as polluted.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 23 23 2
PDF Downloads 7 7 0

Effects of Additional Particles on Already Polluted Marine Stratus

View More View Less
  • 1 College of Oceanic and Atmospheric Sciences, Oregon State University, Corvallis, Oregon
© Get Permissions
Full access

Abstract

The response of already polluted marine stratocumulus to additional particles was examined by studying the clouds where two ship tracks cross. Nearly 100 such crossings were collected and analyzed using Terra and Aqua Moderate Resolution Imaging Spectroradiometer (MODIS) multispectral imagery for the daytime passes off the western coast of the United States during the summer months of 4 years. To reduce biases in the retrieved cloud properties caused by the subpixel spatial structure of the clouds, results are presented only for ship tracks found in regions overcast by extensive layers of marine stratus. When two ship tracks cross, one of the tracks exhibits much larger changes in droplet radii when compared with the surrounding unpolluted clouds and is referred to as the dominant ship track. The clouds at the crossing typically exhibit properties that are closer to those of the dominant than to those of the subordinate ship track. To determine whether the additional particles at the crossing affect the dominant track, local gradients in the retrieved cloud properties near the crossing were determined for both ship tracks. Based on the gradients, the clouds at the junction were found to have significantly smaller droplet radii and significantly larger column droplet number concentrations than were predicted based on their values in both ship tracks on either side of the crossing. Comparing the effects of particle loading at the crossings and elsewhere along the ship tracks revealed that the effects decreased as the column droplet number concentration of the clouds being affected increased.

Corresponding author address: James A. Coakley Jr., College of Oceanic and Atmospheric Sciences, 104 COAS Admin. Bldg., Oregon State University, Corvallis, OR 97331-5503. E-mail: coakley@coas.oregonstate.edu

Abstract

The response of already polluted marine stratocumulus to additional particles was examined by studying the clouds where two ship tracks cross. Nearly 100 such crossings were collected and analyzed using Terra and Aqua Moderate Resolution Imaging Spectroradiometer (MODIS) multispectral imagery for the daytime passes off the western coast of the United States during the summer months of 4 years. To reduce biases in the retrieved cloud properties caused by the subpixel spatial structure of the clouds, results are presented only for ship tracks found in regions overcast by extensive layers of marine stratus. When two ship tracks cross, one of the tracks exhibits much larger changes in droplet radii when compared with the surrounding unpolluted clouds and is referred to as the dominant ship track. The clouds at the crossing typically exhibit properties that are closer to those of the dominant than to those of the subordinate ship track. To determine whether the additional particles at the crossing affect the dominant track, local gradients in the retrieved cloud properties near the crossing were determined for both ship tracks. Based on the gradients, the clouds at the junction were found to have significantly smaller droplet radii and significantly larger column droplet number concentrations than were predicted based on their values in both ship tracks on either side of the crossing. Comparing the effects of particle loading at the crossings and elsewhere along the ship tracks revealed that the effects decreased as the column droplet number concentration of the clouds being affected increased.

Corresponding author address: James A. Coakley Jr., College of Oceanic and Atmospheric Sciences, 104 COAS Admin. Bldg., Oregon State University, Corvallis, OR 97331-5503. E-mail: coakley@coas.oregonstate.edu

1. Introduction

Ship tracks have been used to determine the response of marine stratocumulus to increases in aerosol particle concentrations produced by underlying ships. They have revealed some of the complex interactions that occur as part of the aerosol indirect radiative forcing of the climate. Additional particles lead to clouds with larger droplet number concentrations and smaller droplets. The clouds have larger optical depths and albedos (Twomey 1974; Coakley et al. 1987). Albrecht (1989) suggested that clouds with smaller droplets have suppressed rates of droplet growth, thereby inhibiting precipitation. Examples of this suppression have been described by Radke et al. (1989), King et al. (1993), and Ferek et al. (2000). Albrecht also suggested that with their increased particle loading, polluted clouds would have larger liquid water amounts and longer lifetimes leading to greater cloud fractions. Studies of the response of marine stratocumulus to underlying ships, on the other hand, have revealed that the amount of liquid water in the polluted clouds is on average less than the amount found in the nearby unpolluted clouds (Platnick et al. 2000; Coakley and Walsh 2002; Segrin et al. 2007; Christensen et al. 2009). The smaller liquid water amounts led to the realization that some of the water thought to be missing had encroached on adjacent areas that may have originally been cloud-free as was suggested by results of large-eddy simulations (Ackerman et al. 2003). Still, in numerous extensively overcast regions, clouds polluted by underlying ships had less liquid water. The smaller liquid water amounts were thought to be caused by changes in the interaction of the polluted clouds with the environment. Smaller droplets increase the rate of entrainment between the cloud layer and the overlying free troposphere, bringing more dry air into the boundary layer and thereby leading to thinner clouds with less liquid water (Ackerman et al. 2004). While the increased evaporation rate led to smaller liquid water amounts for the polluted clouds, Christensen et al. (2009) using morning to afternoon trends showed that the rate of liquid water loss by the polluted clouds was smaller than that for the nearby unpolluted clouds. Evidently, the losses in the unpolluted clouds due to their heavier drizzle and slower rate of entrainment outpaced the losses in the polluted clouds due to their lighter drizzle and larger rate of entrainment. This finding confirmed, in part, Albrecht’s suggestion. More recently, Christensen and Stephens (2011) found that when ship tracks appeared in broken clouds, their cloud tops rose above the surrounding unpolluted clouds. Because they were thicker, the polluted clouds had greater liquid water amounts than did the nearby unpolluted clouds, also confirming, in part, Albrecht’s suggestion. Christensen and Stephens also noted that ship tracks found in broken clouds tended to occur with more moisture in the overlying free troposphere and smaller temperature jumps between the cloud layer and the overlying atmosphere than did the ship tracks found in extensive layers of marine stratus. These observations are generally consistent with the results of large-eddy simulations performed by Ackerman et al. (2004).

Aircraft observations of stratocumulus reveal that the ratio of the cloud droplet number concentrations to the number concentrations of particles beneath the clouds falls as the concentration of subcloud particles increases. Twomey (1959) explained the decrease of the ratio in terms of the concentration of cloud condensation nuclei (CCN) at a supersaturation of 1% near cloud base C, the concentration of cloud droplets within the cloud N, and the vertical velocity w. He demonstrated that the cloud droplet number concentration was approximately given by NC0.8w0.3, which suggests decreasing ratios with the increasing particle concentrations below cloud base. As a clear demonstration of this decrease, Martin et al. (1994) found that for low-level water clouds and small particle concentrations typical of clean environments, the cloud droplet number concentrations were linearly proportional to the below-cloud concentrations of aerosol particles with diameters ranging from approximately 0.1 to 3 μm. As the particle concentrations increased and approached those of polluted environments, the ratio of the droplet number concentrations to the aerosol particle concentrations fell. Martin et al. (1994) observed this relationship for both continental and marine air masses. In subsequent years, others have extended such measurements in “closure experiments” testing theoretical predictions with observations of subcloud particle properties and concentrations and in cloud droplet number concentrations (e.g., Snider and Brenguier 2000; Snider et al. 2003; Conant et al. 2004; Meskhidze et al. 2005; Fountoukis et al. 2007). The diminished increase in droplet concentrations with increasing subcloud aerosol concentrations arises from the drawdown of the supersaturation as the cloud droplet number concentration increases (Twomey 1959). Stevens and Feingold (2009) refer to this mechanism as a microphysical buffer on aerosol–cloud interactions.

The purpose of this study was to use ship tracks to investigate the effects of additional aerosol pollution on clouds that were already polluted. In particular, the goal was to find evidence for the diminished response of the column droplet number concentrations as the number concentration of the existing clouds increased. The column droplet number concentrations were deduced from retrievals of cloud visible optical depths and droplet effective radii. Satellite images reveal many incidents where one ship track intersects another. Crossings of ship tracks are regions that have been polluted twice by ships passing underneath. Since each crossing comprises two distinct ship tracks, the amount of aerosol loading in the crossing is expected to be approximately given by the sum of the aerosol loadings for the two ships.

Satellite retrievals of cloud properties from the Moderate Resolution Imaging Spectroradiometer (MODIS) 1-km pixels in a crossing were compared with those from the pixels of the two individual ship tracks and their nearby unpolluted control pixels. The properties of interest were the cloud droplet effective radius, visible optical depth, liquid water path, column droplet number concentration, pixel-scale fractional cloud cover, and temperature. The data collected from each individual crossing were compiled into an ensemble that was analyzed statistically to test for the changes in the response of the clouds as the particle loading increased. Since the marine stratocumulus in which the ship tracks were found exhibit a wide range of natural variability, a large sample of crossings had to be collected and objectively analyzed. An automated, but admittedly ad hoc, routine was developed to separately analyze the ship tracks, the nearby unpolluted control clouds, and the crossing of the two ship tracks. The collection of the crossings and the analysis methods are discussed in the next section. Since the response of the clouds to the added particles depends on the properties of the clouds, the ensemble of ship track crossings was first examined to assess the representativeness of the clouds and their responses based on those found in previous ship track and marine stratocumulus studies. Section 3 presents these results. The section also briefly describes the effects of partly cloudy pixels on the response of the clouds to pollution from the underlying ships. As partly cloudy pixels significantly alter the findings, this study was restricted to pixels identified as containing polluted clouds and to nearby pixels identified as containing uncontaminated clouds that were overcast. Section 4 presents the response of the clouds at the ship track crossings where the clouds had been previously polluted by an underlying ship. The section concludes with the change in column droplet number concentration as a function of the column droplet number concentration prior to pollution. The ship tracks showed behavior similar to that found in aircraft observations. Cloud susceptibility, the change in cloud albedo caused by an increase of 1 droplet per cubic centimeter, as introduced by Platnick and Twomey (1994), is also expected to be small for clouds with large column droplet number concentrations. Some of the cases analyzed in this study showed no significant changes in visible optical depths and, consequently, in cloud albedos. Nonetheless, the droplet radii for such clouds often decreased in response to the influence of the additional particles, thereby indicating an increase in column droplet number concentrations.

2. Data and methods

Segrin et al. (2007) used 1-km MODIS imagery to analyze the response of marine stratus to the particles from underlying ships. The first step in the analysis was to use sunlight reflected at near-infrared wavelengths to log by hand the locations of several thousand ship tracks. Here, the track positions logged by Segrin et al. were the starting point in the search for ship track crossings. First, each crossing identified by the hand-logged positions was inspected to verify that the crossing occurred in a region for which the retrievals of cloud properties were valid. Failed retrievals were caused by a lack of overcast pixels near the crossing, a lack of reliable cloud-free radiances for the region in which the retrievals were performed, or clouds that were too near the surface to reliably estimate the pixel-scale cloud fraction. Next, the crossing was inspected for the presence of cirrus or other upper-level clouds that could compromise the retrieved cloud properties. Crossings with overlying cirrus in the area were discarded. Then, the region containing the pair of ship tracks in the crossing was inspected for the effects of sun glint, which is sunlight specularly reflected by the ocean surface. Sun glint makes cloud-free reflectivities uncertain, thereby jeopardizing the retrievals of cloud properties for pixels not completely covered by clouds. If sun glint was identified and the pixels near the ship tracks or the crossing were not completely overcast, the crossing was excluded from further analysis. If a crossing survived the above screening, an automated procedure was used to identify the pixels of the individual ship tracks, the pixels containing unpolluted clouds on both sides of each ship track, and the pixels that constituted the crossing of the two ship tracks. Pixels in which the clouds were polluted by a ship are hereafter called “polluted pixels,” while the nearby pixels in which the clouds were uncontaminated are called “control pixels.”

Figure 1a shows an image of a ship track crossing created from 3.7-μm radiances. Ship tracks 10, 11, and 12 are numbered near their respective heads, the locations nearest the ships. The ship track numbers identify the individual tracks for which locations were logged. Tracks 10 and 11 cross near the center of the image. Tracks 10 and 12 cross near the lower right corner. Figure 1b shows the results of the automated identification of the ship (blue), control (red), and crossing (green) pixels for tracks 10 and 11. The automated routine started with the positions logged by Segrin et al. (2007) to determine the domains for each of the individual ship tracks. For the automated analysis, the domain of each track was assumed to stretch 20 pixels on both sides of the hand-logged track position. The procedure used 20-pixel segments along the track and a cross-track least squares fit within each segment to identify pixels with 2.1-μm reflectivities that exceeded estimates for the largest reflectivities in the nearby unpolluted pixels by at least three standard deviations. The 20-pixel segments were moved along the track in 10-pixel increments to avoid gaps in the identification of the polluted pixels. A description of the identification procedure for the ship and control pixels can be found in Segrin et al. (2007). The additional complication in this study was the identification of the pixels that constituted the crossing. The appendix gives a detailed description of the algorithm for identifying the crossing pixels along with other modifications to the Segrin et al. (2007) identification scheme.

Fig. 1.
Fig. 1.

Image of ship track crossings constructed from 3.7-μm radiances and the automated identification of the ship track pixels (blue), the control pixels (red), and the crossing pixels (green) for tracks 10 and 11. The observations were for the MODIS Aqua at 2155 UTC 29 Jul 2003, ranging from 39° to 41°N and 127° to 132°W. Tracks 10 and 11 cross near the center of the image, while tracks 10 and 12 cross near the right edge.

Citation: Journal of the Atmospheric Sciences 69, 6; 10.1175/JAS-D-11-0291.1

There were two parts to the analysis. First, the cloud properties were determined using the partly cloudy pixel scheme described by Coakley et al. (2005) and implemented for MODIS observations by Segrin et al. (2007). In the partly cloudy pixel retrieval scheme, the radiances are assumed to be linear functions of the cloud fraction within each pixel. The retrieved cloud properties are based on the contribution to the observed radiances contributed by the cloudy portions of the pixels. As such, the retrieved cloud properties for a pixel are taken to be the average properties for the fraction of the pixel that is cloud covered.

Second, the following steps were taken to analyze the properties of the clouds in the crossing and those away from the crossing and to determine how the already polluted clouds responded to additional particles. Figure 2 provides a schematic of a ship track crossing to illustrate how the analysis was implemented. The cloud properties were averaged in five-pixel along-track segments before and after the crossing. Separate averages were kept for the polluted and control pixels in each segment. Owing to the variable widths of ship tracks, the number of pixels averaged in each five-pixel segment ranged from 15 to 200. Figure 3 shows the average 3.7-μm derived cloud droplet effective radius retrieved for the overcast pixels within the five-pixel segments for tracks 10 and 11 shown in Fig. 1. The average droplet effective radius is shown as a function of the average along-track distance of the segment from the center of the crossing. Averages from the five-pixel segments were also collected for the visible optical depth, liquid water path (LWP), temperature, column droplet number concentration (CDNC), and fractional cloud cover. In this study, the LWP was taken to be given by
e1
where Re is the cloud droplet effective radius, τ is the cloud visible optical depth, and ρ = 1 g cm−3 is the density of liquid water. The CDNC was taken to be given by
e2
Fig. 2.
Fig. 2.

Schematic illustrating the procedures used to analyze the crossing of two ship tracks. The solid lines represent the boundaries of the ship tracks. The portions of the ship tracks analyzed in this study are rendered in light and dark blue. The ship tracks cross at the box near the center of the diagram rendered in green. The dashed lines indicate the extent of the pixels on both sides of the ship tracks that contain the unpolluted clouds that were used as controls. The control pixels analyzed in this study are those rendered in red. Each rectangle in the figure represents a five-pixel segment along the ship track. Properties for the clouds in the ship tracks and for the nearby unpolluted clouds were averaged separately within each five-pixel segment along the ship track. At the crossing, the properties of each ship track not affected by the second ship were estimated using linear least squares fits of the properties along the track for up to four of the five-pixel segments nearest the crossing on both sides. The estimated cloud properties were then compared with those of the pixels at the crossing to deduce the effects of the second ship. The completely colored five-pixel segments represent the 20-km segments used to infer the effects of the individual ships on the unpolluted clouds. These segments were at least 20 km from the pixels associated with the crossing.

Citation: Journal of the Atmospheric Sciences 69, 6; 10.1175/JAS-D-11-0291.1

Fig. 3.
Fig. 3.

Average droplet effective radius for five-pixel along-track segments derived using 3.7-μm radiances, and distance from the center of the crossing for the ship (○), control 1 (+), control 2 (×), and crossing (•) pixels of tracks 10 and 11 shown in Fig. 1. Control 1 and control 2 refer to the control pixels on opposite sides of the ship tracks. All averages were for overcast pixels. Least squares trend lines are shown for the averages of the ship track pixels within the first four five-pixel along-track segments on each side nearest the crossing that satisfied the conditions required for inclusion in the analysis as described in the text. Negative distances from the crossing are associated with the leg of the track that contains the head, the pixels nearest the polluting ship.

Citation: Journal of the Atmospheric Sciences 69, 6; 10.1175/JAS-D-11-0291.1

The primary objective of this study was to use ship track crossings to assess the degree to which the response of clouds to enhancements in particle loading was diminished because of effects caused by a previous pollution event. As the response of the clouds depends on their properties, the first objective was to survey the properties of the clouds in which the ship track crossings occurred. These properties were compared with those found in the Segrin et al. (2007) study of ship tracks and the Hayes et al. (2010) study of marine stratocumulus. For each track, all five-pixel segments that were at least 20 km from the crossing and had an along-track length of approximately 20 km were composited to provide representative averages for each segment. Segrin et al. (2007) found that the autocorrelation lengths of cloud properties for pixels overcast by marine stratus were between 5 and 10 km, depending on the property being examined. Furthermore, they found that the averages for 20-km segments behaved as if they were statistically independent. Consequently, the 20-km segments on either side of a crossing were taken to provide statistically independent observations. These segments are represented in Fig. 2 as the rectangles in which both the ship track and the control pixels are colored.

The purpose of the partly cloudy pixel retrieval scheme was to correct the retrieved cloud properties in the sizable fraction of imager pixels that were only partially cloud covered. As found by Hayes et al. (2010), the scheme appears to offer some improvement in the retrieved properties over those schemes in which all cloudy pixels are assumed to be overcast. But they also noted that the partly cloudy pixel retrievals exhibited some of the same trends in retrieved cloud optical depths and droplet effective radii exhibited by the other schemes. Retrieved droplet radii, fractional cloud cover, and cloud-top temperature tend to be overestimated while visible optical depths tend to be underestimated when partly cloudy pixels are assumed to be overcast (Coakley et al. 2005). Unfortunately, such trends give rise to false aerosol indirect forcing estimates (Matheson et al. 2006). Hayes et al. (2010) noted that these trends in retrieved cloud properties were prevalent in pixels found to be overcast within relatively small regions that also contained cloud-free and partly cloudy pixels. They attributed these trends to subpixel variability in the cloud liquid water. The droplet effective radius, LWP, and CDNC in this study were derived using 3.7-μm reflectivities, as opposed to the values commonly reported from the MODIS cloud product, which are derived from 2.1-μm reflectivities. Owing to the stronger absorption by liquid water at 3.7 μm, radiances at that wavelength are less susceptible to effects caused by subpixel spatial variability of the clouds than are the radiances at 2.1 μm (Hayes et al. 2010; Zhang and Platnick 2011). The following procedures were adopted to avoid some of the biases in the retrieved cloud properties arising from partly cloudy pixels. For a 20-km segment to be included in the analysis, at least 40 of its pixels had to be overcast. In addition, within each five-pixel segment that was part of the larger 20-km segment, overcast pixels had to represent at least 70% of all the pixels identified by the automated scheme as being part of the ship track. For the five-pixel segments in which some of the ship track pixels were partly cloudy, the average pixel-scale cloud cover of all polluted pixels had to be greater than 0.9. Retrieved cloud properties for partly cloudy pixels approach those of nearby overcast ship track pixels when the fractional cloud cover within the pixels approaches unity. When partly cloudy pixels were included in the average, the cloud properties for each pixel were weighted by the fractional cloud cover in the pixel. The same conditions were imposed for the pixels identified by the automated scheme as unpolluted controls on both sides of the ship track. Up to two sets of observations were thus collected from each ship track, one from the segment before the crossing and the other from the segment after the crossing. Thus, each pair of crossing ship tracks provided average cloud properties for up to four 20-km segments. The 20-km segments collected from all ship tracks were pooled and analyzed. The results are presented in section 3.

The crossing analysis was conducted by comparing the difference between a ship track and its unpolluted controls to that between the ship track and its intersection with another ship track. The 20-km ship track segments described above were used to establish the differences in cloud properties between the ship track and control pixels. To compute the differences at the crossing, a proxy for each ship track was developed for the crossing pixels. The proxy represented the properties that the clouds would have had in the absence of the other ship track. Gradients of the properties through the crossing were used to estimate the contribution of each ship track to the properties of the clouds in the crossing. A linear least squares fit was applied to the average values of the five-pixel along-track segments from one side of the crossing to the other. The ship track pixels in the five-pixel segments had to satisfy the same overcast and partly cloudy criteria used in the five-pixel segments that constituted the 20-km ship track segments described above. Both directions along the track were searched for the first four five-pixel segments nearest the crossing that satisfied the above criteria. In Fig. 2, these segments are represented by the four segments adjacent the ship track crossing on both sides of the crossing. Only the ship track pixels are colored for these segments in the figure. This search was restricted to a maximum distance of 50 km from the crossing. If the procedure failed to collect at least three five-pixel increments on both sides of the crossing, the crossing was discarded. Least squares fits were applied to the droplet radius, the optical depth, the LWP, the CDNC, and the cloud temperature of each ship track in the dataset. Figure 3 shows the least squares trend lines for the droplet radii of tracks 10 and 11 shown in Fig. 1.

At the ship track crossing, the results of the least squares fits were used as proxies for the properties of the clouds that would have been observed if the pixels had not been polluted by a second ship. The use of a gradient to construct these proxies approximated the large-scale variations in cloud properties along the length of the track. An image of a ship track is a snapshot in time. The pixels nearest the head were polluted most recently, while those farthest from the head of the track were polluted earlier. Gradients in the cloud properties of the ship track may have existed naturally prior to the ship transect. Gradients also arise from time-dependent processes such as diffusion, precipitation, and entrainment along the track. For example, the results in Fig. 3 indicate that there was a north–south gradient in droplet effective radius for the data associated with the images shown in Fig. 1. The gradient is most evident in the along-track trends for the ship and control pixels of track 10 and to a lesser extent in the differences between the controls in the tails (positive distances in Fig. 3) of both tracks. Differences between the retrieved values for the crossing pixels and the projected values for the individual ship tracks were collected for all cloud properties of interest. These differences were then compared with those between the ship track and control pixels calculated for the 20-km segments that were 20 km away from the crossing.

No two ship tracks involved in a crossing were alike. Often the properties of the clouds at the crossing closely resembled those of one of the ship tracks and not the other. Consequently, each pair of crossing tracks was separated into a “dominant ship track” and a “subordinate ship track.” Within a pair of crossing ship tracks, the track that had the larger average difference in the 3.7-μm derived droplet radius between the polluted and nearby control clouds was taken to be the dominant track. In the example shown in Figs. 1 and 3, track 11 is the dominant ship track. Typically, the dominant ship tracks were narrower at the crossing, suggesting that the polluting source was relatively strong and that the particles had not yet been dispersed to the extent of the pollution for the subordinate track. Generally, but not always, the distance between the head of the track and the crossing was shorter for the dominant track, suggesting that the ship that generated the subordinate track had reached the crossing first. Factors that alter these general rules are, of course, the relative strengths of the polluting sources, winds, entrainment, precipitation, and dispersion. Also of interest, as will be reported in section 4, the droplet effective radii retrieved at the crossing were less than those estimated for nearly all of the dominant and subordinate ship tracks. Ship track 11 shown in Figs. 1 and 3 was a rare exception.

3. Cloud properties for the ship track and neighboring control pixels

Daytime images created from MODIS Terra and Aqua level 1B radiances were inspected for the presence of ship track crossings. The overpasses stretched from 20° to 60°N and 110° to 150°W off the western coast of the United States. Observations for the 2001–04 summer months yielded 569 crossings, based on the hand-logged positions. Of these, 126 crossings were located in sun glint and had significant numbers of partly cloudy pixels. These crossings were eliminated from further analysis. An additional 60 crossings were underneath cirrus and eliminated. Also, 88 crossings fell in regions where the partly cloudy pixel retrievals failed. For 159 of the 295 remaining crossings, the signal-to-noise ratio was insufficient for the automated identification procedure to distinguish the ship track pixels from the unpolluted control pixels. Such failures typically arose from the lack of overcast pixels on one side of a ship track passing through a field of broken marine stratocumulus. Finally, 35 crossings were discarded because the two tracks forming the crossing intersected at an acute angle, thereby preventing the automated procedure from correctly identifying the individual ship tracks. The final dataset contained 101 crossings, comprising 202 ship tracks. Of these, 73 crossings came from Terra and 28 came from Aqua. Table 1 categorizes the breakdown of the crossings examined.

Table 1.

Breakdown of ship track crossings off the West Coast of the United States examined for the summer months of 2001–04.

Table 1.

For each ship track in a crossing, two segments of approximately 20 km length at a distance of 20 km from the crossing were collected as described in section 2. Ideally, the collection of two segments from each track would have provided a total of 404 segments. The tracks, however, were chosen based on the quality of the crossings and not on the retrievals of cloud properties. Hence, 71 of the 101 pairs of ship tracks satisfied the conditions for overcast and partly cloudy pixels described in section 2 with the dominant and subordinate ship tracks providing observations for the 20-km segment on at least one side of the crossing. The 71 crossing pairs produced 266 20-km ship track segments that satisfied the prescribed conditions.

Figure 4 shows the distributions of cloud droplet effective radius retrieved using the 3.7-μm channel, visible optical depth, liquid water path, and column droplet number concentration. Each of the 266 20-km segments provided composite averages for both the ship track and the control pixels on both sides of the ship track. Compared with the ship tracks analyzed by Segrin et al. (2007), the average visible optical depth, droplet effective radius, and LWP were significantly larger for the control pixels of the crossing ship tracks. The respective differences were 0.5 ± 0.4 μm, 1.3 ± 0.8, and 19 ± 8 g m−2 (mean ± 95% confidence interval). The differences probably arose because the crossing pairs generally included ship tracks that were farther from the coast than those analyzed in the Segrin et al. study. While ship track crossings are relatively frequent, more than half of all ship tracks have no crossing. Segrin et al. analyzed only isolated 20-km segments in which both the ship track and the nearby control pixels were overcast. In addition, they avoided regions in which ship tracks and their neighboring control pixels overlapped with other ship tracks. Consequently, they included in their analysis a sizable number of ship track segments that fell near the coast. In this study, the need for ship track crossings along with the avoidance of sun glint, which often fell on the side of the MODIS scan nearest the coast, pushed the analyzed segments farther out to sea. Owing in part to haze pollution in the marine environment near the coast and in part to the increase in the thickness of the stratus with distance from the coast, droplet effective radii generally increase with distance from the coast. Owing to the larger droplet effective radius and visible optical depth, the liquid water path was correspondingly larger in this study than in the Segrin et al. study.

Fig. 4.
Fig. 4.

(a) Average 3.7-μm derived cloud droplet effective radius Re for the ship track pixels (solid line) and the control pixels (dashed line), (b) visible optical depth, (c) liquid water path, and (d) column droplet number concentration. The averages are for the overcast pixels in the first 20-km along-track segments that fell on either side of a ship track crossing and were more than 20 km from the crossing. Number of samples, means, and 95% confidence intervals are shown.

Citation: Journal of the Atmospheric Sciences 69, 6; 10.1175/JAS-D-11-0291.1

Figure 5 shows distributions of the differences between composite averages for the 20-km segment, both between the ship track and control pixels and between the controls on opposite sides of the ship track. The droplet radius of the ship track pixels was on average 2.1 ± 0.2 μm smaller than that of the control pixels. Segrin et al. found that the droplet radius of the ship tracks was 2.4 ± 0.1 μm smaller. The difference is just at the 95% confidence interval for the two datasets. The visible optical depth of the clouds in the pixels contaminated by ships was significantly larger than that for the clouds in the nearby control pixels, implying that the polluted clouds had albedos that were larger than those of the nearby unpolluted clouds. The mean difference in the optical depth, 2.3, was nearly identical to the 2.1 difference reported by Segrin et al. Segrin et al. observed that the ship track pixels lost 10 g m−2 in LWP relative to the controls, suggesting that the loss of liquid water due to the entrainment of dry air dominated the suppression of precipitation that occurred as the cloud droplet size decreased (Ackerman et al. 2004). Here the loss of liquid water was half that found by Segrin et al., reflecting the smaller change in droplet radius. Again, the difference is just at the 95% confidence interval for the two datasets. The smaller response of the clouds in the ship track crossings, along with the larger visible optical depths of the controls is consistent with the responses found by Segrin et al. Clouds with large visible optical depths have small responses, regardless of the cloud droplet effective radii. As discussed in the next section, clouds with large visible optical depths also have small cloud susceptibility as defined by Platnick and Twomey (1994). For such clouds, changes in the visible reflectivities and consequently in the visible optical depths and albedos become difficult to detect.

Fig. 5.
Fig. 5.

As in Fig. 4, but for the average differences between the overcast ship track and control pixels (ship − control; solid line) and the overcast control pixels on either side of the ship track (control 1 − control 2; dashed line).

Citation: Journal of the Atmospheric Sciences 69, 6; 10.1175/JAS-D-11-0291.1

Not shown are cloud temperatures. As reported by others (Coakley and Walsh 2002; Segrin et al. 2007; Christensen et al. 2009; Christensen and Stephens 2011) there was no significant difference between the ship track and nearby control pixels, indicating that there were no measurable differences in cloud-top heights.

As noted by Segrin et al. the results in Fig. 5 illustrate an advantage of using ship tracks over other approaches to studying the response of clouds to aerosol particle loading. The lack of significant differences in droplet effective radius, visible optical depth, and LWP between the controls on opposite sides of the ship tracks indicates that the clouds were the same everywhere in the domain of the segment except in the pixels polluted by the underlying ship. In regional aerosol indirect effect studies (e.g., Quaas et al. 2008; Loeb and Schuster 2008; Kaufman et al. 2005), demonstrating that the unpolluted clouds are identical to the polluted clouds, except for the effects of the enhanced particle loading, entails a host of requirements that are difficult to achieve (Matheson et al. 2005).

The results presented here are based only on retrievals for MODIS pixels found to be overcast, or nearly so. The average pixel-scale cloud cover of the pixels identified as either polluted or uncontaminated controls within the five-pixel along-track segments that contained partly cloudy pixels had to be >0.9. In the 101 ship track crossings, approximately 5% of the ship track pixels within the set of all 20-km ship track segments were partly cloudy. The remaining 95% were overcast. Of course, the automated scheme that identifies the ship tracks seeks those with unusually large near-infrared reflectivities, thereby favoring overcast pixels. Of the control pixels, approximately 20% were classified as partly cloudy, while the remaining 80% were overcast. In the results shown in Figs. 4 and 5, only ship tracks from 71 crossings could be used because of the lack of overcast pixels, primarily among the controls. The use of partly cloudy pixels along with overcast pixels would have allowed 88 of the crossings to be used. Accounting for the cloud fractions in both the partly cloudy and overcast pixels, the cloud fraction was 0.99 for all ship track pixels and 0.95 for the control pixels. With the partly cloudy pixels included, however, the decrease in liquid water path for the polluted pixels became a slight increase, but not significantly different from zero at the 95% significance level. This reversal results from the clouds within partly cloudy pixels tending to be thinner than those within nearby overcast pixels (Coakley et al. 2005; Hayes et al. 2010). As a result, the partly cloudy pixels have smaller liquid water paths. Because the majority of partly cloudy pixels were in the controls, the average LWP of the controls was significantly smaller when partly cloudy pixels were included. The average LWP of the ship tracks, on the other hand, was only slightly smaller. Any loss of liquid water due to pollution of the clouds was compensated by the increased cloud cover in the ship track pixels and the corresponding increase in liquid water. In studies of aerosol–cloud interactions, the apparent increase in cloud fraction would have to be considered. Here, the objective was to analyze the response of the clouds, rather than the cloud-filling of cloud-free skies due to polluting events (Ackerman et al. 2003). The inclusion of partly cloudy pixels skewed the results because they disproportionately resided among the control pixels. Hence, the crossing analysis described in the next section was restricted to overcast pixels.

4. Changes in cloud properties for the crossing pixels

For each crossing, as described in section 2, linear least squares fits were used to establish projected values of the droplet radius, optical depth, LWP, CDNC, and cloud temperature in the crossing domain. These projected values acted as proxies for the cloud properties that would have existed had the pixels not been affected by a second ship. To test the reliability of this interpolation method, data from 164 50-km ship track segments were collected. Each 50-km segment was divided into two 20-km segments separated by a 10-km target segment. A least squares fit was applied to the averages of the five-pixel along-track segments within the two 20-km segments. The differences between the properties projected from the least squares trend lines and those retrieved for the 10-km target segments were averaged for each 50-km segment. Table 2 lists the average differences between the properties retrieved in the 10-km target segments and those projected from the gradients. With the exception of the droplet effective radius for which the 0.09 μm average error is just at the 95% confidence interval for detection, none of the differences departed significantly from zero. With regard to the error in the projected droplet effective radius, the changes at the crossing noted below were at least a factor of 4 larger. Owing to the lack of significant errors in the other parameters and the small magnitude of the error in the droplet effective radius, the trends in the ship track gradients were judged to adequately approximate the retrieved values.

Table 2.

Errors from the method used to project polluted cloud properties into the region of ship track crossings. Listed are means and 95% confidence intervals for the differences in the droplet effective radius ΔRe, visible optical depth Δτ, ΔLWP, and ΔCDNC between the retrieved cloud properties in 164 10-km ship track target segments and projected properties derived from linear gradients based on the retrievals in 20-km ship track segments on both sides of the target segments.

Table 2.

The number of five-pixel segments retrieved from the crossing domains ranged from 1 to 7 and was generally proportional to the size of the domain, which itself was determined by the widths of the intersecting ship tracks. For any ship track, differences were taken between each of the crossing retrievals and the value of the least squares fit at the appropriate distance from the center of the crossing domain. These differences were then averaged to obtain the change in a cloud property due to the pollution of one ship track by a second ship. Following this approach, each crossing yielded two values, one for the response of the subordinate ship track to pollution by the dominant ship; the second was the reverse. Aggregation over all crossings led to a dataset consisting of 142 measures of the response of each ship track to pollution by the other ship.

As the response to pollution by a ship depends on the properties of the clouds being polluted, differences in the dominant and subordinate controls for each ship track crossing pair were also examined. Table 3 lists the means and 95% confidence intervals for the differences between the control pixels associated with the dominant and subordinate ship tracks. Interestingly, both the droplet effective radius and the visible optical depth of the dominant controls were larger than those of the subordinate controls. While the difference in optical depth is not statistically different from zero at the 95% confidence level, that for the droplet radius is just significant. The result is that the liquid water paths of the dominant controls were significantly larger than those for the subordinate controls. Likewise, while not statistically different from zero at the 95% confidence level, the column number droplet concentrations for the dominant controls were smaller than those for the subordinate controls. Perhaps such differences promoted the larger changes observed for the dominant tracks.

Table 3.

Mean and 95% confidence intervals for differences among the controls for the dominant and subordinate (dominant − subordinate) ship tracks that constituted crossing pairs. The averages are for the differences in the 20-km ship track segments of the dominant and subordinate tracks taken from both sides of each crossing.

Table 3.

Figures 69 show the distributions of the differences between the ship track and control pixels and between the ship track and the crossing pixels. The differences are for the droplet effective radius, visible optical depth, LWP, and CDNC for both the dominant and subordinate ship tracks. As before, there were no changes in cloud temperatures; consequently, those changes are not shown. As indicated in Figs. 69, clouds already polluted clearly responded to the influences of both the dominant and subordinate ships, but compared with the responses of their controls, those of the already polluted clouds were subdued. Interestingly, the effects of the ships on the already polluted clouds led to further reductions in droplet radii and increases in the column droplet concentrations that were statistically significant at the 95% confidence level. Increases in the visible optical depth were statistically significant but those for the dominant ship track were just at the 95% confidence level. Only the losses of liquid water for both the dominant and subordinate ship at the crossing failed to be different from zero at that confidence level. Clearly, clouds with larger column droplet number concentrations responded less to the effects of the additional pollution.

Fig. 6.
Fig. 6.

Differences in average cloud droplet effective radius Re between the overcast pixels in the ship tracks and those in the nearby controls for 20-km ship track segments 20 km from the ship track crossing (solid line) and average differences for the overcast crossing pixels and the projected values at the crossing (dashed lines) for the (a) subordinate and (b) dominant ship track. The number of crossings, mean differences, and 95% confidence intervals for the differences are given.

Citation: Journal of the Atmospheric Sciences 69, 6; 10.1175/JAS-D-11-0291.1

Fig. 7.
Fig. 7.

As in Fig. 6, but for visible optical depth.

Citation: Journal of the Atmospheric Sciences 69, 6; 10.1175/JAS-D-11-0291.1

Fig. 8.
Fig. 8.

As in Fig. 6, but for liquid water path.

Citation: Journal of the Atmospheric Sciences 69, 6; 10.1175/JAS-D-11-0291.1

Fig. 9.
Fig. 9.

As in Fig. 6, but for column droplet number concentration.

Citation: Journal of the Atmospheric Sciences 69, 6; 10.1175/JAS-D-11-0291.1

As was noted in the introduction, low-level water clouds show a diminishing increase in droplet number concentrations with increasing particle concentrations in the subcloud layer (Martin et al. 1994). The same is found with ship tracks when crossing a region that is already polluted. Figure 10 shows the average changes in column droplet number concentrations for the dominant and subordinate ship tracks and their associated controls as well as the changes for the dominant ship when passing under the clouds perturbed by the subordinate ship and vice versa. Similar to the findings of Martin et al., the increase in the column droplet number concentration diminishes as the column concentration of the cloud prior to the pollution event increases. A linear least squares fit applied to the observations gave
e3
where CDNC has units of 105 cm−2. Owing to the vast domain of the region containing the ship track crossings and the correspondingly large range of cloud and ship track properties, there is considerable variability in the results. Not surprisingly, the linear fit explained less than 10% of the variance of the change in the column droplet concentration. Nonetheless, owing to the large number of independent samples, the probability that the correlation could have occurred by chance was less than 5%. Tests using Spearman’s nonparametric rank correlation also suggested that the changes in the column droplet number concentration were affected by the preexisting number concentration with the probability of no effect being less than 5%. The smaller increases in the column droplet number concentrations of clouds that had large concentrations prior to the addition of particles are clearly evident.
Fig. 10.
Fig. 10.

Average changes in column droplet number concentration due to pollution from the ship and the existing droplet number concentration for the overcast control pixels in 20-km ship track segments 20 km from the crossing for the dominant and subordinate ship tracks and for the crossing pixels. At the crossing, the changes for the dominant ship track are from the values projected at the crossing for the subordinate ship and the changes for the subordinate ship track are from the values projected for the dominant ship. The line shows the result of the least squares fit given by (3).

Citation: Journal of the Atmospheric Sciences 69, 6; 10.1175/JAS-D-11-0291.1

Another factor that affects changes in column droplet number concentrations deduced from satellite observations is cloud susceptibility (Platnick and Twomey 1994). Cloud susceptibility is the increase in the cloud albedo caused by an increase of one droplet per cubic centimeter. In the Eddington approximation, the change in the spherical albedo of a cloud ΔrC is given by
e4
where τ is the visible optical depth. When the cloud liquid water amount is constant the fractional change in the optical depth Δτ/τ = ⅓(ΔCDNC/CDNC). Thus, for a given fractional increase in the CDNC, the largest change in albedo occurs for clouds with albedos ~0.5, or in clouds that have visible optical depths ~8–10. When the CDNC in the clouds is small, so that the optical depth is small, the change in albedo becomes proportional to the fractional change in the CDNC. For thin marine stratus the droplet number concentration can be quite small, 20–50 cm−3. Even with the small number concentration, a 1 cm−3 increase is a small fraction and with the small albedo rC typical of such clouds, the change leads to a small change in albedo. As indicated by the occurrences of optical depths for the unpolluted clouds shown in Fig. 5b, no clouds in the current study were optically thin (τ < 5). Thin marine stratocumulus found near ship tracks are often broken so that many of the pixels identified as controls are partly cloudy. When the CDNC is large, so that the optical depth is large, the change in albedo is again small, partly because the transmittance (1 − rC) becomes small and partly because the droplet concentrations can be large (~103 cm−3). Consequently, for large CDNC, the fractional change in the CDNC can be small and the changes in the albedo and retrieved optical depths vanishingly small. Such cases were found in the current study, as indicated by the optical depth changes at the crossings for the dominant ship tracks being just detectable at the 95% confidence level (Fig. 7b). Indeed, for many crossings, the changes in the optical depths for the dominant ship tracks were negative. Such decreases along with the increases in droplet radii, like those indicated by the results in Fig. 6b and in the case of track 11 shown in Fig. 3, probably led to the decreases in column droplet number concentrations shown in Fig. 10. Still, the reductions in droplet radii for many of the ship tracks that had no accompanying increases in optical depths led to the increases in column droplet number concentrations (Fig. 9b). When changes in visible optical depths become difficult to detect, however, the changes in the column droplet concentrations are probably underestimated.

5. Conclusions

MODIS 1-km daytime imagery off the western coast of the United States for the summer months of 2001–04 was used to study the response of marine stratus already polluted by underlying ships to the additional pollution of a second ship. The properties of the clouds in the pixels at the crossing of two ship tracks were compared with those of the clouds in the separate ship tracks on both sides of the crossing. As done previously, the properties of the clouds in the ship track pixels were compared with nearby pixels containing uncontaminated marine stratus. The responses in column droplet number concentration (CDNC), droplet radius, visible optical depth, liquid water path, pixel-scale cloud fraction, and temperature to additional increases in CCN were determined for the clouds polluted by both ships separately and by the two ships together. A partly cloudy pixel retrieval scheme was used to determine the cloud properties (Coakley et al. 2005; Segrin et al. 2007). The ship track pixels proved to be largely overcast while some of the nearby control pixels were partly cloudy. Admittedly, the differences in the pixel-scale cloud cover fractions between the ship tracks and the nearby controls resulted in part from the automated scheme used to identify the polluted pixels. In any case, within single layers of marine stratocumulus, clouds found in overcast pixels generally have larger optical depths and liquid water paths than the clouds found in nearby partly cloudy pixels (Coakley et al. 2005; Hayes et al. 2010). Evidently, the clouds in the partly cloudy pixels are thinner. To avoid changes in cloud properties linked to pixel-scale cloud fractions, comparisons in this study were restricted to five-pixel along-track segments that satisfied the following conditions. If some of the pixels identified as containing polluted clouds within a segment were partly cloudy, then within the same segment at least 70% of all pixels containing polluted clouds had to be overcast and the average pixel-scale cloud fraction of these pixels had to be >0.9. The same conditions were applied to the pixels identified as containing the nearby unpolluted clouds within the segment.

On average, clouds in the polluted pixels had smaller droplet radii, larger optical depths, and larger CDNCs than their unpolluted counterparts. The polluted clouds also had smaller liquid water paths. No differences were observed in cloud temperatures, suggesting no changes in cloud-top altitudes. The droplet radii, optical depths, and liquid water paths in the unpolluted clouds associated with the crossing ship tracks were slightly but significantly larger than those of the ship tracks studied by Segrin et al. (2007). The clouds in the ship track crossings tended to be farther from the coast. As a result, they were thicker and had larger droplets than are typically found for the clouds near the coast. Owing to the larger optical depths, the clouds in this study were slightly less sensitive to the underlying ships than found in the Segrin et al. study.

For the four years of summer months analyzed, an ensemble of 71 ship track crossings satisfied the conditions of having overcast pixels for both the ship tracks and controls sufficient for the analysis of the separate ship tracks and the ship track crossing. Each ship track crossing had a dominant and a subordinate ship track. The dominant ship track had the greater average decrease in cloud droplet effective radius derived using 3.7-μm reflectivities when compared with the uncontaminated clouds in the nearby control pixels. The analysis was first conducted as if the subordinate ship track had formed before the dominant ship track and was then polluted by the dominant ship. The properties of the clouds in the dominant ship track were compared with those for the uncontaminated clouds and with the clouds that had already been polluted by the subordinate ship. Even though the clouds had already been polluted, significant reductions were found in the average droplet effective radius and significant increases were found in the average visible optical depth and CDNC (left-hand panels in Figs. 69). The average increase in CDNC, however, was only about 40% of that observed in going from the nearby uncontaminated clouds to the clouds in the dominant ship tracks. While the dominant track lost additional water when crossing the subordinate track, the loss was not statistically different from zero at the 95% confidence level. The analysis was repeated with the assumption that the dominant ship track had been laid down first and was then contaminated by the subordinate ship. The results were largely the same. Even for the clouds that had been polluted by the dominant ship track, significant reductions were found in the average droplet radius while significant increases were found for the visible optical depth and CDNC (right-hand panels of Figs. 69). Again, the subordinate track lost water when affected by the dominant ship track, but the loss was not statistically different from zero. As was the case for the dominant track, the responses for the subordinate track were much smaller when the clouds were already polluted. The average change for the CDNC was about one-third of that observed when the unpolluted clouds were polluted by the subordinate ship.

Martin et al. (1994) observed that for marine air masses the cloud droplet number concentrations increased linearly with the concentration of aerosol particles below cloud base when the particle concentrations were small. For below-cloud particle concentrations greater than about 200 cm−3, increases in cloud droplet number concentrations diminished with increasing particle concentrations. The actual concentration of aerosols was unknown in this study, but the response in the column droplet number concentration to further aerosol loading was clearly diminished as the aerosol concentration was increased. In addition, the dependence of the change in the column droplet number concentration for a given existing concentration mimicked the relationship suggested by Martin et al. for the dependence of the cloud droplet number concentration on the subcloud particle concentration.

The effect of aerosols on clouds and the consequent change in cloud radiative forcing is diminished in clouds that have large column droplet number concentrations. This reduction in response is in addition to the diminished change in albedo that arises from the small cloud susceptibilities for such clouds (Platnick and Twomey 1994). The small changes in cloud albedo and thus small changes in visible reflectivities lead to small changes in the retrieved visible optical depth. Small changes in the visible optical depth in turn probably give rise to underestimates of the changes in the column droplet number concentration. In fact, in this study when changes in the visible optical depths became largely undetectable, despite the increases in particle concentrations, decreases in the retrieved column droplet number concentrations sometimes occurred. Nonetheless, for many of the cases in which zero or negative changes had occurred in the visible optical depth, reductions in the cloud droplet effective radius were sufficient to indicate that the column droplet number concentration had increased.

Acknowledgments

This work was undertaken under NASA’s EOS Grant NNX08AG01G and Radiation Sciences Grant NNX08AK07G. We thank Mike King and Andy Ackerman for many comments on an earlier manuscript that greatly improved the presentation of these findings.

APPENDIX

Automated Algorithm for Ship Track Crossings

The identification of polluted and nearby uncontaminated control pixels uses the scheme described by Segrin et al. (2007) modified to allow for the existence of ship track crossings. Segrin et al. analyzed isolated ship track segments of about 20 km in length. With the crossing of two tracks, the problem of a second ship track interfering with the identification process had to be solved. The modifications to the Segrin et al. algorithm included 1) an exclusion mask that excluded all ship track pixels identified using the hand-logged positions of the tracks so that the automated routine would identify as polluted only pixels associated with the track being analyzed and 2) a scheme for predicting the near-infrared reflectivities of the pixels identified as polluted and the cross-track width of the domain occupied by the polluted pixels in the vicinity of the crossing. The reflectivities and track widths on either side of the crossing were used to estimate their values within the crossing domain. In addition to these modifications, which are described below, the automated scheme described by Segrin et al. (2007) was changed so that the 20-km along-track window used in the identification of polluted pixels was advanced along the track at 10-km intervals to ensure that all positions along the track were analyzed. In addition, instead of identifying polluted and nearby uncontaminated pixels for each 20-km track segment individually, as did Segrin et al., all of the polluted pixels for each pair of tracks that crossed were identified prior to the identification of the nearby uncontaminated control pixels adjacent the ship tracks.

The modifications to the Segrin et al. identification scheme are illustrated with the crossing shown in Fig. A1. The figure shows an image of 2.1-μm reflectivities captured by the Aqua MODIS on 26 July 2004. In Fig. A1b, the hand-logged positions of the tracks are indicated by dots. The dots are joined by straight lines. The lines indicate the approximate positions of the polluted pixels. To reduce clutter, the hand-logged positions of the other ship tracks in the figure are not shown. The scheme for identifying the polluted pixels, and subsequently the uncontaminated control pixels, used the hand-logged positions as a starting point for the automated identification.

Fig. A1.
Fig. A1.

(a) Image of 2.1-μm reflectivity from the Aqua MODIS off the coast of Northern California at 2140 UTC 26 Jul 2004, also shown with (b) the hand-logged track positions overlaid, (c) pixels identified as polluted by the automated scheme shown in color indicating the 2.1-μm derived droplet effective radius, and (d) pixels identified as uncontaminated control pixels on both sides of the polluted pixels shown in color. The tracks identified as 9 and 11 crossed at 40.5°N, 134.4°W at the time of the overpass. The × symbols in (c) and (d) indicate the central locations associated with the polluted pixels of the five-pixel segments along the track used to accumulate information on both the polluted and uncontaminated control pixels. To reduce clutter only the centers of five-pixel segments separated by 20 pixels along the track are shown.

Citation: Journal of the Atmospheric Sciences 69, 6; 10.1175/JAS-D-11-0291.1

The first modification to the Segrin et al. scheme was to use an exclusion mask based on the hand-logged position. The mask excluded from analysis all pixels within three MODIS pixels of the lines connecting the hand-logged track positions. When the polluted pixels of a given track were being identified, masked pixels from any other track could not be used except in regions where the tracks crossed, which was also where the exclusion masks crossed. As was described in Segrin et al., the automated identification was performed on 20-pixel segments along the direction of the track. The identification started from the head of the track, which is the position nearest the ship that produced the track. In the modified version of the Segrin et al. scheme these 20-pixel segments were advanced along the length of the track in 10-pixel increments. The overlap prevented gaps between pixels identified as being polluted.

As was noted in section 2, when two tracks cross, one track is dominant; the second is subordinate. The dominant track was taken to be the track with the larger changes in the 3.7-μm derived droplet effective radius for the polluted pixels when compared with the nearby uncontaminated pixels on either side. For most cases, the dominant track also had the larger changes in the 2.1-μm derived droplet radius and 2.1-μm reflectivities. At the crossing of the two tracks, the automated routine often lacked sufficient numbers of uncontaminated control pixels on one or both sides of the track needed to identify pixels in the subordinate track as being polluted. Consequently, as a second modification to the Segrin et al. algorithm, near-infrared reflectivities of the pixels identified as polluted and the widths of the ship tracks on either side of the crossing were used to predict the minimum reflectivities of polluted pixels and the ship track widths in the vicinity of the crossing. Figure A2 shows the 2.1-μm reflectivities of polluted pixels and Fig. A3 shows the cross-track widths of their domains in the vicinity of the ship track crossing. In this example, track 11 was the dominant track; track 9 was the subordinate track. The automated routine identified polluted pixels at nearly every position along track 11 but, because of the existence of track 11, for track 9 the routine failed to identify polluted pixels in the vicinity of the crossing. This example also illustrates a feature generally found for ship track crossings. The dominant track was typically the newer track, laid down after the subordinate track had been created. Distances from the ship heads to the crossings for the dominant tracks were usually smaller than those for the subordinate tracks.

Fig. A2.
Fig. A2.

The 2.1-μm reflectivity (%, small dots) of polluted pixels and along-track distances to the crossing for tracks 9 and 11 shown in Fig. A1. Negative distances are used for the leg of the track between the track head, the point nearest the ship, and the crossing. The smallest reflectivity for pixels identified as polluted are given by the means minus 1.5 times the standard deviations of the pixel-scale reflectivity accumulated in the five-pixel segments along the track and are indicated by large dots joined by lines. The dashed line is a least squares estimate of the smallest reflectivity associated with polluted pixels for each track in the vicinity of the crossing. It was used to predict the lower bound of the reflectivity for the polluted pixels at the crossing.

Citation: Journal of the Atmospheric Sciences 69, 6; 10.1175/JAS-D-11-0291.1

Fig. A3.
Fig. A3.

As in Fig. A2, but for the cross-track widths of the domain containing pixels identified as polluted using the automated scheme. The dots give the means of the widths and the error bars give the standard deviations for five-pixel along-track segments. Pixels that had sufficiently large 2.1-μm reflectivities and fell within the predicted cross-track width were identified as polluted.

Citation: Journal of the Atmospheric Sciences 69, 6; 10.1175/JAS-D-11-0291.1

To bridge the gap in track 9 shown in Figs. A2 and A3, linear least squares fits were applied to the smallest values of the 2.1-μm reflectivities for the pixels identified as polluted and the cross-track widths of the domains occupied by these pixels on either side of the crossing. The smallest values of the reflectivities for the pixels identified as polluted were taken to be the means of the reflectivities minus 1.5 times the standard deviations of the reflectivities for the polluted pixels within the first five five-pixel segments along the track on either side of the crossing. In Fig. A2 the smallest reflectivities are indicated by large dots connected by lines. In Fig. A3 the cross-track widths of the domains occupied by polluted pixels and the standard deviations of the widths are given. In the figures, the linear least squares fits for the reflectivities and widths are indicated by dashed lines. The fits were used to predict minimum reflectivities for polluted pixels and the widths of their domains in the vicinity of crossings. Pixels that had a 2.1-μm reflectivity greater than the minimum expected based on the fit and fell within the limits specified by the predicted width of the polluted pixel domain were identified as polluted. Finally, once all of the polluted pixels in the two tracks had been identified, the uncontaminated pixels on either side of the polluted tracks were identified following the procedures described by Segrin et al. Figure 1b illustrates the results of the identification of the ship tracks, associated controls, and crossing pixels.

As shown in Fig. A1d, the presence of the exclusion mask laid down according to the hand-logged positions of the ship tracks creates a gap in the uncontaminated control pixels for track 11. The gap occurs at the crossing of track 11 with track 10. In this study the emphasis was on the properties of the clouds in the vicinity of the crossings. Gaps that were sufficiently far from the crossing being analyzed (>40 km) had no impact on the outcome. In the analysis of the crossing for tracks 10 and 11, similar gaps appeared in the vicinity of the crossing for tracks 9 and 11, but not near the crossing of 10 and 11.

REFERENCES

  • Ackerman, A. S., , O. B. Toon, , D. E. Stevens, , and J. A. Coakley Jr., 2003: Enhancement of cloud cover and suppression of nocturnal drizzle in stratocumulus polluted by haze. Geophys. Res. Lett., 30, 1381, doi:10.1029/2002GL016634.

    • Search Google Scholar
    • Export Citation
  • Ackerman, A. S., , M. P. Kirkpatrick, , D. E. Stevens, , and O. B. Toon, 2004: The impact of humidity above stratiform clouds on indirect aerosol climate forcing. Nature, 432, 10141017.

    • Search Google Scholar
    • Export Citation
  • Albrecht, B. A., 1989: Aerosols, cloud microphysics, and fractional cloudiness. Science, 245, 12271230.

  • Christensen, M. W., , and G. Stephens, 2011: Microphysical and macrophysical responses of marine stratocumulus polluted by underlying ships: Evidence of cloud deepening. J. Geophys. Res., 116, D03201, doi:10.1029/2010JD014638.

    • Search Google Scholar
    • Export Citation
  • Christensen, M. W., , J. A. Coakley Jr., , and W. R. Tahnk, 2009: Morning-to-afternoon evolution of marine stratus polluted by underlying ships: Implications for the relative lifetimes of polluted and unpolluted clouds. J. Atmos. Sci., 66, 20972106.

    • Search Google Scholar
    • Export Citation
  • Coakley, J. A., Jr., , and C. D. Walsh, 2002: Limits to the aerosol indirect radiative effect derived from observations of ship tracks. J. Atmos. Sci., 59, 668680.

    • Search Google Scholar
    • Export Citation
  • Coakley, J. A., Jr., , R. L. Bernstein, , and P. A. Durkee, 1987: Effect of ship-stack effluents on cloud reflectivity. Science, 237, 10201022.

    • Search Google Scholar
    • Export Citation
  • Coakley, J. A., Jr., , M. A. Friedman, , and W. R. Tahnk, 2005: Retrievals of cloud properties for partly cloudy imager pixels. J. Atmos. Oceanic Technol., 22, 317.

    • Search Google Scholar
    • Export Citation
  • Conant, W. C., and Coauthors, 2004: Aerosol–cloud drop concentration closure in warm cumulus. J. Geophys. Res., 109, D13204, doi:10.1029/2003JD004324.

    • Search Google Scholar
    • Export Citation
  • Ferek, R. J., and Coauthors, 2000: Drizzle suppression of ship tracks. J. Atmos. Sci., 57, 27072728.

  • Fountoukis, C., and Coauthors, 2007: Aerosol–cloud drop concentration closure for clouds sampled during the International Consortium for Atmospheric Research on Transport and Transformation 2004 campaign. J. Geophys. Res., 112, D10S30, doi:10.1029/2006JD007272.

    • Search Google Scholar
    • Export Citation
  • Hayes, C. R., , J. A. Coakley Jr., , and W. R. Tahnk, 2010: Relationships among properties of marine stratocumulus derived from collocated CALIPSO and MODIS observations. J. Geophys. Res., 115, D00H17, doi:10.1029/2009JD012046.

    • Search Google Scholar
    • Export Citation
  • Kaufman, Y., , I. Koren, , A. Remer, , D. Rosenfeld, , and Y. Rudich, 2005: The effect of smoke, dust, and pollution aerosol on shallow cloud development over the Atlantic Ocean. Proc. Natl. Acad. Sci. USA, 102, 11 20711 212.

    • Search Google Scholar
    • Export Citation
  • King, M. D., , L. F. Radke, , and P. V. Hobbs, 1993: Optical properties of marine stratocumulus clouds modified by ships. J. Geophys. Res., 98, 27292739.

    • Search Google Scholar
    • Export Citation
  • Loeb, N. G., , and G. L. Schuster, 2008: An observational study of the relationship between cloud, aerosol and meteorology in broken low-level cloud conditions. J. Geophys. Res., 113, D14214, doi:10.1029/2007JD009763.

    • Search Google Scholar
    • Export Citation
  • Martin, G. M., , D. W. Johnson, , and A. Spice, 1994: The measurement and parameterization of effective radius of droplets in warm stratocumulus clouds. J. Atmos. Sci., 51, 18231842.

    • Search Google Scholar
    • Export Citation
  • Matheson, M. A., , J. A. Coakley Jr., , and W. R. Tahnk, 2005: Aerosol and cloud property relationships for summertime stratiform clouds in the northeastern Atlantic from Advanced Very High Resolution Radiometer observations. J. Geophys. Res., 110, D24204, doi:10.1029/2005JD006165.

    • Search Google Scholar
    • Export Citation
  • Matheson, M. A., , J. A. Coakley Jr., , and W. R. Tahnk, 2006: Effects of threshold retrievals on estimates of the aerosol indirect radiative forcing. Geophys. Res. Lett., 33, L07705, doi:10.1029/2005GL025614.

    • Search Google Scholar
    • Export Citation
  • Meskhidze, N., , A. Nenes, , W. C. Conant, , and J. H. Seinfeld, 2005: Evaluation of a new cloud droplet activation parameterization with in situ data from CRYSTAL-FACE and CSTRIPE. J. Geophys. Res., 110, D16202, doi:10.1029/2004JD005703.

    • Search Google Scholar
    • Export Citation
  • Platnick, S., , and S. Twomey, 1994: Determining the susceptibility of cloud albedo to changes in droplet concentration with the Advanced Very High Resolution Radiometer. J. Appl. Meteor., 33, 334347.

    • Search Google Scholar
    • Export Citation
  • Platnick, S., and Coauthors, 2000: The role of background cloud microphysics in the radiative formation of ship tracks. J. Atmos. Sci., 57, 26072624.

    • Search Google Scholar
    • Export Citation
  • Quaas, J., , O. Boucher, , N. Bellouin, , and S. Kinne, 2008: Satellite-based estimate of the direct and indirect aerosol climate forcing. J. Geophys. Res., 113, D05204, doi:10.1029/2007JD008962.

    • Search Google Scholar
    • Export Citation
  • Radke, L. F., , J. A. Coakley Jr., , and M. D. King, 1989: Direct and remote sensing observations of the effects of ships on clouds. Science, 246, 11461149.

    • Search Google Scholar
    • Export Citation
  • Segrin, M. S., , J. A. Coakley Jr., , and W. R. Tahnk, 2007: MODIS observations of ship tracks in summertime stratus off the west coast of the United States. J. Atmos. Sci., 64, 43304345.

    • Search Google Scholar
    • Export Citation
  • Snider, J. R., , and J.-L. Brenguier, 2000: Cloud condensation nuclei and cloud droplet measurements during ACE-2. Tellus, 52B, 828842.

    • Search Google Scholar
    • Export Citation
  • Snider, J. R., , S. Guibert, , J.-L. Brenguier, , and J.-P. Putaud, 2003: Aerosol activation in marine stratocumulus clouds: 2. Köhler and parcel theory closure studies. J. Geophys. Res., 108, 8629, doi:10.1029/2002JD002692.

    • Search Google Scholar
    • Export Citation
  • Stevens, B., , and G. Feingold, 2009: Untangling aerosol effects on clouds and precipitation in a buffered system. Nature, 461, 607613.

    • Search Google Scholar
    • Export Citation
  • Twomey, S., 1959: The nuclei of natural cloud formation. Part II: The supersaturation in natural clouds and the variation of the cloud droplet concentration. Geofis. Pura Appl., 43, 243249.

    • Search Google Scholar
    • Export Citation
  • Twomey, S., 1974: Pollution and the planetary albedo. Atmos. Environ., 8, 12511256.

  • Zhang, Z., , and S. Platnick, 2011: An assessment of differences between cloud effective particle radius retrievals for marine water clouds from three MODIS spectral bands. J. Geophys. Res., 116, D20215, doi:10.1029/2011JD016216.

    • Search Google Scholar
    • Export Citation
Save