Cloud detection in satellite imagery over the South Pole
Cloud detection in satellite imagery over the South Pole DAN LUBIN,
California Space Institute, University of California at San Diego, La Jolla, California 92093-0221 Department ofAstronomy, Boston University, Boston, Massachusetts 02215 D.A. HARPER, Yerkes Observatory, Williams Bay, Wisconsin 53191-0258
RICHARD CHAMBERLIN,
ecause of the remoteness of most of Antarctica, satellite B data have an important role to play in the study of climate at southern high latitudes. Our recent efforts to develop a cloud-detection algorithm for use with advanced very-highresolution radiometer (A\THRR) data over the high plateau are motivated by site evaluation requirements for the Center for Astrophysical Research in Antarctica (CARA) and by the potential that the large volumes of recently archived AVHRR data have for improving our understanding of the atmospheric radiation budget. The justification for deploying modern and sophisticated telescopes on the high plateau, as well as the planning of astronomical observing programs, can be furthered by a comprehensive climatology of cloud cover. Although such a climatology is available from the South Pole weather office for the immediate vicinity of Amundsen—Scott South Pole Station, future goals to deploy astronomical instruments at remote sites such as Dome A, Dome C, or Vostok can benefit from satellite cloud climatology. The atmospheric energy budget and surface temperature fields are known to be strongly influenced by cloud cover (Dolgin 1986), and given that general circulation models exhibit errors at high latitudes (Tzeng et al. 1994), it is important to take advantage of unique satellite data sets to refine our understanding of the physics of antarctic climate and perhaps to develop improvements to the general circulation models. The National Oceanic and Atmospheric Administration's polar orbiter overpasses used for this research were tracked by facilities set up at McMurdo and Palmer Stations by the Arctic and Antarctic Research Center (AARC) at the Scripps Institution of Oceanography. For this project, more than 400 AVHRR images of the high plateau were calibrated and Earth-located at full spatial resolution (1.1 kilometer at nadir) using the TeraScan software developed by SeaSpace, Inc. This data set encompassed most of 1992 and was divided into two parts. Images from the first 2 weeks of each month are used for algorithm development, whereas images from the second 2 weeks of each month are used for independent testing of the algorithm. From each image, a 32x32 pixel subset centered over the South Pole is extracted for application of the cloud-detection tests. The data being used to train the algorithm are the cloud-cover observations from the South Pole weather office. It is usually difficult to detect clouds in a polar satellite image, much less classify them or quantify their optical and radiative properties. Invisible wavelength imagery, often little or no radiance contrast can be detected between highly reflective snow and ice surfaces and the cloud tops. At middle infrared wavelengths, cloud-top brightness temperatures are very similar to surface brightness temperatures. Furthermore, low radiance levels in general over the poles can leave the satellite sensors operating near the lower limit of their perfor-
mance. Several useful techniques for polar cloud detection have been developed over the past decade, including multispectral thresholds (Yamanouchi and Kawaguchi 1992), mul tispectral methods supplemented by microwave data (Key and Barry 1989), pattern recognition (Ebert 1987), and the use of neural networks to recognize texture (Welch et al. 1992). Most of these methods have been developed over coastal regions, where the underlying surface is variable (e.g., open water, broken ice, solid ice cover, mountainous terrain). This variability in surface texture, contrasted with variability in the texture of cloud cover, provides scenes that lend themselves to successful classification using pattern recognition (Welch, Kuo, and Sengupta 1990). The high antarctic plateau, however, poses a more difficult problem due to the extreme homogeneity of both the surface and the stratiform cloud cover. Our initial efforts have focused on an algorithm using only the AVHRR channels 4 and 5 (10.3-11.3 microns and 11.3-12.3 microns, respectively), with the hope that we might have a useful tool for either the polar day or the polar night. Figure 1 shows an application of the brightness temperature difference between channels 4 and 5 to distinguish overcast from clear skies (scenes labeled using the nearest South Pole weather office observations). This bispectral technique takes advantage of the wavelength-dependent emissivity of optically thin clouds (Yamanouchi and Kawaguchi 1992) but can be inconclusive over optically thick clouds, sub-pixel-size clouds, or certain viewing angles where the variable emissivity of snow cover can influence the satellite signal (Dozier and Warren 1982). Figure 2 shows the application of one textural feature, the maximum entropy of the gray-level differences within the 32x32 pixel subset (e.g., Ebert 1987). A different classification threshold for clear and overcast scenes appears for warmer vs. colder scenes. Although textural features such as the example of figure 2 do not provide as consistent a discrimination as the brightness temperature difference (figure 1), they do add some accuracy when used in conjunction with the brightness temperature difference. The table shows an application of a preliminary clouddetection algorithm using the brightness temperature difference (AVHRR channels 4 and 5), supplemented by the maximum entropy of the gray-level differences. In this classification, "clear" refers to surface observations noting either clear skies or scattered clouds, whereas "cloudy" refers to surface observations noting either broken or overcast skies. Although, as of this writing, the full range of possible textural features has not been investigated, the algorithm is showing some promise. Results are approaching theoretical limits (Ebert 1987; Welch et al. 1992) during July and August, the best time of year for astronomical observing (due to the extremely low water-vapor opacity). One important differ-
ANTARCTIC JOURNAL - REVIEW 1994
307
Spectral Cloud Discriminator 6 Figure 1. The brightness temperature difference between AVHRR channels 4 and 5, as a function of brightness temperature in channel 4, applied to 32x32 pixel regions over the South Pole from the subset of the AVHRR data used for algorithm development (first 2 weeks of each month). Overcast vs. clear-sky scene identification is provided by the South Pole weather office data. The lower solid curve is a threshold for clear surface identification suggested by the radiative transfer study of Dozier and Warren (1982). The upper solid curve is a threshold for clear surface identification suggest by the satellite data analysis of Yamanouchi, Suzuki, and Kawaguchi (1987).
-F
+
It,
I-
C.)
+
overcast (+) clear (o)
+ +++
+ +#++ t ++ + + ++ + + ++, +f
C.) C Wi a) '4-
-1
o ++ ^0 0 0 - 0 + +
+
-80 -70 -60 -50 -40 -30 -20 Channel 4 (11 him) brightness temperature (C)
Gray-Level Difference Statistics 2.5 +
C.) Cn
/
0
+o-1 + sJ
< 1.5 CM > 0 1 4C
1
?05 -80
kq
+
o^
+ ++ 0 ++f 0 ++
to
-H+ + + +0 ++ ++
to
o0
00 00 00 0 0 0 00 t+0 00 0
+o
+
Figure 2. The maximum entropy of the gray-level difference statistics for AVHRR channel 4, plotted against the brightness temperature in channel 4, applied to 32x32 pixel regions over the South Pole from the subset of the AVHRR data used for algorithm development (first 2 weeks of each month). Overcast vs. clear-sky scene identification is provided by the South Pole weather office data.
-70 -60 -50 -40 -30 -20 Channel 4 (11 gm) brightness temperature (C) 1992 (Chamberlin and Bally 1994), and we have noticed a correlation between the AVHRR brightness temperature dif ference (between channels 4 and 5) and the 225-gigahertz water-vapor opacity. This suggests that we may be able to investigate the relationship between atmospheric watervapor burden and cloud formation. We anticipate that the combination of these two data sets, along with the South Pole weather office sky observations and rawinsonde data, will yield new insights into the atmospheric radiation balance over the high antarctic plateau. This work was supported by the Center for Astrophysical Research in Antarctica under National Science Foundation grant OPP 89-20223. We thank Robert H. Whritner of the Arctic and Antarctic Research Center at the Scripps Institution of
ence between this project and previously published clouddetection exercises is that this work will make use of totally independent sky observations (from the South Pole weather office) to train and validate the algorithm, as opposed to the use of training sets derived from visual inspection of some of the satellite images. This fact, combined with the extreme scene homogeneity over the antarctic plateau, would lead us to expect possibly poorer performance of the algorithm than previously published results. The table, however, shows reason for encouragement, and we plan to continue this research throughout 1994. We have also begun to co-locate these AVHRR images with water-vapor opacity measurements from a 225-gigahertz microwave atmospheric radiometer operated at the South Pole by CARA throughout
ANTARCTIC JOURNAL - REVIEW 1994
308
ments of the sky-brightness temperature. Applied
Application of preliminary cloud detection algorithm over ti te South Optics, 33(6), 1095-1099. Pole, to the subset of the AVHRR data used for validation (s econd 2 Dolgin, I.M. (Ed.). 1986. Climate of Antarctica. New Delhi: Oxonian Press. weeks of each month) Dozier, J., and S.G. Warren. 1982. Effect of viewing NOTE: For each month, the table lists the total number of observaa tO fl5,h e angle on the infrared brightness temperature of number of cloudy observations from the South Pole weather office, thi number snow. Water Resources Research, 18(5), 1424-1434. of cloudy observations in the AVHRR data, and the skill score of the a lgorithm. Ebert, E. 1987. A pattern recognition technique for The skill score is the percent of correctly identified scenes (using S uth Pole distinguishing surface and cloud types in the weather office data as ground truth) as determined using an error taA ile (Ebert polar regions. Journal of Climate and Applied 1987). Meteorology, 26(10), 1412-1427.
January 11 11 February 15 4 March 11 3 April 17 4 May 25 16 June 17 11 July 25 13 August 27 14 September 17 8 October 19 5 November 25 7 December 16 6
Key, J., and R.G. Barry. 1989. Cloud cover analysis with arctic AVHRR data: 1. Cloud detection. Jouri'itr.y nal of Geophysical Research, 94(D15), 18521-18535. Tzeng, R.-Y., D.H. Bromwich, T.R. Parish, and B. 18% Chen. 1994. NCAR CCM2 simulation of the mod73% ern antarctic climate. Journal of Geophysical 67% Research, 99(D11), 23131-23148. 71% Welch, R.M., K.-S. Kuo, and S.K. Sengupta. 1990. 56% Cloud and surface textural features in the polar 63% regions. IEEE Transactions on Geoscience and Remote Sensing, 28(4), 520-528. 86% Welch, R.M., S.K. Sengupta, A.K. Goroch, P. Rabindra, 85% N. Rangaraj, and M.S. Navar. 1992. Polar cloud 53% and surface classification using AVI-IRR imagery: 84% An intercomparison of methods. Journal of 72% Applied Meteorology, 31(5), 405-420. 81% Yamanouchi, T., and S. Kawaguchi. 1992. Cloud distribution in the Antarctic from A\THRR data and radiation measurements at the surface. International Journal of Remote Sensing, 13(1), 111-127. Yamanouchi, T., K. Suzuki, and S. Kawaguchi. 1987. Discrimination of polar clouds from AVHRR infrared multispectral data. In
2 2 0 1 5 5 16 18 4 2 2 3
Oceanography for ready access to the large volume of satellite data.
References
Report of the International Satellite Cloud Climatology Project (ISCCP) Workshop on Cloud Algorithms in the Polar Regions,
Chamberlin, R.A., and J. Bally. 1994. 225-GHz atmospheric opacity of the South Pole sky derived from continual radiometric measure-
World Meteorological Organization WCP-131, WMO/TD-No. 170, Appendix C.2. Geneva, Switzerland: World Meteorological Organization.
Cloud-mediated enhancement of cloud condensation nuclei at Palmer Station: A natural antigreenhouse mechanism V.K. SAXENA, North Carolina State University, Department of Marine, Earth, and Atmospheric Sciences, Raleigh, North Carolina 27695-8208
e present here a case study of cloud-mediated producWtion of cloud condensation nuclei (CCN) recorded by the Fukuta-Saxena CCN Spectrometer at Palmer Station (64°46'S 64°05'W), Antarctica, in January and February, 1994. Four instances of CCN bursts occurred (17, 19, and 20 January and 7 February 1994) when cloud base descended to the surface and dissipated under prevailing meteorological conditions. The most spectacular event occurred on 20 January when the CCN concentration was enhanced by a factor of four at 0.25 percent supersaturation (with respect to water) compared to the pre-event concentration. At 1.25 percent supersaturation, the corresponding enhancement was by a factor of seven. This indicated a larger production of aerosol
particles in smaller size ranges. The event lasted for over 15 hours. The CCN activity spectrum during the event resembled the ones that are typical of our previous measurements in the urban plume of St. Louis and Denver. The phenomenon is also significant in evaluating natural causes of antigreenhouse mechanisms. It is well understood that cloud droplet size spectrum close to the base is influenced by the activation spectrum of CCN in the cloud-forming air mass. The former determines the radiative characteristics of clouds that become more reflective if the pollution content of the cloud-forming air mass is higher (Saxena and Grovenstein 1994a). Enhanced CCN concentrations are also capable of reducing precipitation from clouds, thereby
ANTARCTIC JOURNAL - REVIEW 1994 309
California Space Institute, University of California at San Diego, La Jolla, California 92093-0221 Department ofAstronomy, Boston University, Boston, Massachusetts 02215 D.A. HARPER, Yerkes Observatory, Williams Bay, Wisconsin 53191-0258
RICHARD CHAMBERLIN,
ecause of the remoteness of most of Antarctica, satellite B data have an important role to play in the study of climate at southern high latitudes. Our recent efforts to develop a cloud-detection algorithm for use with advanced very-highresolution radiometer (A\THRR) data over the high plateau are motivated by site evaluation requirements for the Center for Astrophysical Research in Antarctica (CARA) and by the potential that the large volumes of recently archived AVHRR data have for improving our understanding of the atmospheric radiation budget. The justification for deploying modern and sophisticated telescopes on the high plateau, as well as the planning of astronomical observing programs, can be furthered by a comprehensive climatology of cloud cover. Although such a climatology is available from the South Pole weather office for the immediate vicinity of Amundsen—Scott South Pole Station, future goals to deploy astronomical instruments at remote sites such as Dome A, Dome C, or Vostok can benefit from satellite cloud climatology. The atmospheric energy budget and surface temperature fields are known to be strongly influenced by cloud cover (Dolgin 1986), and given that general circulation models exhibit errors at high latitudes (Tzeng et al. 1994), it is important to take advantage of unique satellite data sets to refine our understanding of the physics of antarctic climate and perhaps to develop improvements to the general circulation models. The National Oceanic and Atmospheric Administration's polar orbiter overpasses used for this research were tracked by facilities set up at McMurdo and Palmer Stations by the Arctic and Antarctic Research Center (AARC) at the Scripps Institution of Oceanography. For this project, more than 400 AVHRR images of the high plateau were calibrated and Earth-located at full spatial resolution (1.1 kilometer at nadir) using the TeraScan software developed by SeaSpace, Inc. This data set encompassed most of 1992 and was divided into two parts. Images from the first 2 weeks of each month are used for algorithm development, whereas images from the second 2 weeks of each month are used for independent testing of the algorithm. From each image, a 32x32 pixel subset centered over the South Pole is extracted for application of the cloud-detection tests. The data being used to train the algorithm are the cloud-cover observations from the South Pole weather office. It is usually difficult to detect clouds in a polar satellite image, much less classify them or quantify their optical and radiative properties. Invisible wavelength imagery, often little or no radiance contrast can be detected between highly reflective snow and ice surfaces and the cloud tops. At middle infrared wavelengths, cloud-top brightness temperatures are very similar to surface brightness temperatures. Furthermore, low radiance levels in general over the poles can leave the satellite sensors operating near the lower limit of their perfor-
mance. Several useful techniques for polar cloud detection have been developed over the past decade, including multispectral thresholds (Yamanouchi and Kawaguchi 1992), mul tispectral methods supplemented by microwave data (Key and Barry 1989), pattern recognition (Ebert 1987), and the use of neural networks to recognize texture (Welch et al. 1992). Most of these methods have been developed over coastal regions, where the underlying surface is variable (e.g., open water, broken ice, solid ice cover, mountainous terrain). This variability in surface texture, contrasted with variability in the texture of cloud cover, provides scenes that lend themselves to successful classification using pattern recognition (Welch, Kuo, and Sengupta 1990). The high antarctic plateau, however, poses a more difficult problem due to the extreme homogeneity of both the surface and the stratiform cloud cover. Our initial efforts have focused on an algorithm using only the AVHRR channels 4 and 5 (10.3-11.3 microns and 11.3-12.3 microns, respectively), with the hope that we might have a useful tool for either the polar day or the polar night. Figure 1 shows an application of the brightness temperature difference between channels 4 and 5 to distinguish overcast from clear skies (scenes labeled using the nearest South Pole weather office observations). This bispectral technique takes advantage of the wavelength-dependent emissivity of optically thin clouds (Yamanouchi and Kawaguchi 1992) but can be inconclusive over optically thick clouds, sub-pixel-size clouds, or certain viewing angles where the variable emissivity of snow cover can influence the satellite signal (Dozier and Warren 1982). Figure 2 shows the application of one textural feature, the maximum entropy of the gray-level differences within the 32x32 pixel subset (e.g., Ebert 1987). A different classification threshold for clear and overcast scenes appears for warmer vs. colder scenes. Although textural features such as the example of figure 2 do not provide as consistent a discrimination as the brightness temperature difference (figure 1), they do add some accuracy when used in conjunction with the brightness temperature difference. The table shows an application of a preliminary clouddetection algorithm using the brightness temperature difference (AVHRR channels 4 and 5), supplemented by the maximum entropy of the gray-level differences. In this classification, "clear" refers to surface observations noting either clear skies or scattered clouds, whereas "cloudy" refers to surface observations noting either broken or overcast skies. Although, as of this writing, the full range of possible textural features has not been investigated, the algorithm is showing some promise. Results are approaching theoretical limits (Ebert 1987; Welch et al. 1992) during July and August, the best time of year for astronomical observing (due to the extremely low water-vapor opacity). One important differ-
ANTARCTIC JOURNAL - REVIEW 1994
307
Spectral Cloud Discriminator 6 Figure 1. The brightness temperature difference between AVHRR channels 4 and 5, as a function of brightness temperature in channel 4, applied to 32x32 pixel regions over the South Pole from the subset of the AVHRR data used for algorithm development (first 2 weeks of each month). Overcast vs. clear-sky scene identification is provided by the South Pole weather office data. The lower solid curve is a threshold for clear surface identification suggested by the radiative transfer study of Dozier and Warren (1982). The upper solid curve is a threshold for clear surface identification suggest by the satellite data analysis of Yamanouchi, Suzuki, and Kawaguchi (1987).
-F
+
It,
I-
C.)
+
overcast (+) clear (o)
+ +++
+ +#++ t ++ + + ++ + + ++, +f
C.) C Wi a) '4-
-1
o ++ ^0 0 0 - 0 + +
+
-80 -70 -60 -50 -40 -30 -20 Channel 4 (11 him) brightness temperature (C)
Gray-Level Difference Statistics 2.5 +
C.) Cn
/
0
+o-1 + sJ
< 1.5 CM > 0 1 4C
1
?05 -80
kq
+
o^
+ ++ 0 ++f 0 ++
to
-H+ + + +0 ++ ++
to
o0
00 00 00 0 0 0 00 t+0 00 0
+o
+
Figure 2. The maximum entropy of the gray-level difference statistics for AVHRR channel 4, plotted against the brightness temperature in channel 4, applied to 32x32 pixel regions over the South Pole from the subset of the AVHRR data used for algorithm development (first 2 weeks of each month). Overcast vs. clear-sky scene identification is provided by the South Pole weather office data.
-70 -60 -50 -40 -30 -20 Channel 4 (11 gm) brightness temperature (C) 1992 (Chamberlin and Bally 1994), and we have noticed a correlation between the AVHRR brightness temperature dif ference (between channels 4 and 5) and the 225-gigahertz water-vapor opacity. This suggests that we may be able to investigate the relationship between atmospheric watervapor burden and cloud formation. We anticipate that the combination of these two data sets, along with the South Pole weather office sky observations and rawinsonde data, will yield new insights into the atmospheric radiation balance over the high antarctic plateau. This work was supported by the Center for Astrophysical Research in Antarctica under National Science Foundation grant OPP 89-20223. We thank Robert H. Whritner of the Arctic and Antarctic Research Center at the Scripps Institution of
ence between this project and previously published clouddetection exercises is that this work will make use of totally independent sky observations (from the South Pole weather office) to train and validate the algorithm, as opposed to the use of training sets derived from visual inspection of some of the satellite images. This fact, combined with the extreme scene homogeneity over the antarctic plateau, would lead us to expect possibly poorer performance of the algorithm than previously published results. The table, however, shows reason for encouragement, and we plan to continue this research throughout 1994. We have also begun to co-locate these AVHRR images with water-vapor opacity measurements from a 225-gigahertz microwave atmospheric radiometer operated at the South Pole by CARA throughout
ANTARCTIC JOURNAL - REVIEW 1994
308
ments of the sky-brightness temperature. Applied
Application of preliminary cloud detection algorithm over ti te South Optics, 33(6), 1095-1099. Pole, to the subset of the AVHRR data used for validation (s econd 2 Dolgin, I.M. (Ed.). 1986. Climate of Antarctica. New Delhi: Oxonian Press. weeks of each month) Dozier, J., and S.G. Warren. 1982. Effect of viewing NOTE: For each month, the table lists the total number of observaa tO fl5,h e angle on the infrared brightness temperature of number of cloudy observations from the South Pole weather office, thi number snow. Water Resources Research, 18(5), 1424-1434. of cloudy observations in the AVHRR data, and the skill score of the a lgorithm. Ebert, E. 1987. A pattern recognition technique for The skill score is the percent of correctly identified scenes (using S uth Pole distinguishing surface and cloud types in the weather office data as ground truth) as determined using an error taA ile (Ebert polar regions. Journal of Climate and Applied 1987). Meteorology, 26(10), 1412-1427.
January 11 11 February 15 4 March 11 3 April 17 4 May 25 16 June 17 11 July 25 13 August 27 14 September 17 8 October 19 5 November 25 7 December 16 6
Key, J., and R.G. Barry. 1989. Cloud cover analysis with arctic AVHRR data: 1. Cloud detection. Jouri'itr.y nal of Geophysical Research, 94(D15), 18521-18535. Tzeng, R.-Y., D.H. Bromwich, T.R. Parish, and B. 18% Chen. 1994. NCAR CCM2 simulation of the mod73% ern antarctic climate. Journal of Geophysical 67% Research, 99(D11), 23131-23148. 71% Welch, R.M., K.-S. Kuo, and S.K. Sengupta. 1990. 56% Cloud and surface textural features in the polar 63% regions. IEEE Transactions on Geoscience and Remote Sensing, 28(4), 520-528. 86% Welch, R.M., S.K. Sengupta, A.K. Goroch, P. Rabindra, 85% N. Rangaraj, and M.S. Navar. 1992. Polar cloud 53% and surface classification using AVI-IRR imagery: 84% An intercomparison of methods. Journal of 72% Applied Meteorology, 31(5), 405-420. 81% Yamanouchi, T., and S. Kawaguchi. 1992. Cloud distribution in the Antarctic from A\THRR data and radiation measurements at the surface. International Journal of Remote Sensing, 13(1), 111-127. Yamanouchi, T., K. Suzuki, and S. Kawaguchi. 1987. Discrimination of polar clouds from AVHRR infrared multispectral data. In
2 2 0 1 5 5 16 18 4 2 2 3
Oceanography for ready access to the large volume of satellite data.
References
Report of the International Satellite Cloud Climatology Project (ISCCP) Workshop on Cloud Algorithms in the Polar Regions,
Chamberlin, R.A., and J. Bally. 1994. 225-GHz atmospheric opacity of the South Pole sky derived from continual radiometric measure-
World Meteorological Organization WCP-131, WMO/TD-No. 170, Appendix C.2. Geneva, Switzerland: World Meteorological Organization.
Cloud-mediated enhancement of cloud condensation nuclei at Palmer Station: A natural antigreenhouse mechanism V.K. SAXENA, North Carolina State University, Department of Marine, Earth, and Atmospheric Sciences, Raleigh, North Carolina 27695-8208
e present here a case study of cloud-mediated producWtion of cloud condensation nuclei (CCN) recorded by the Fukuta-Saxena CCN Spectrometer at Palmer Station (64°46'S 64°05'W), Antarctica, in January and February, 1994. Four instances of CCN bursts occurred (17, 19, and 20 January and 7 February 1994) when cloud base descended to the surface and dissipated under prevailing meteorological conditions. The most spectacular event occurred on 20 January when the CCN concentration was enhanced by a factor of four at 0.25 percent supersaturation (with respect to water) compared to the pre-event concentration. At 1.25 percent supersaturation, the corresponding enhancement was by a factor of seven. This indicated a larger production of aerosol
particles in smaller size ranges. The event lasted for over 15 hours. The CCN activity spectrum during the event resembled the ones that are typical of our previous measurements in the urban plume of St. Louis and Denver. The phenomenon is also significant in evaluating natural causes of antigreenhouse mechanisms. It is well understood that cloud droplet size spectrum close to the base is influenced by the activation spectrum of CCN in the cloud-forming air mass. The former determines the radiative characteristics of clouds that become more reflective if the pollution content of the cloud-forming air mass is higher (Saxena and Grovenstein 1994a). Enhanced CCN concentrations are also capable of reducing precipitation from clouds, thereby
ANTARCTIC JOURNAL - REVIEW 1994 309
