During the field experiment phase of the Indian Ocean Experiment (INDOEX), linear cloud formations parallel to the West Indian coast and propagating westward have been observed. Meteosat-5 images are used for the description of the life cycle of these events. Single cloud bands, or main cloud bands followed by (up to 10) secondary parallel cloud lines with a wavelike pattern, have been observed daily during four periods in the dry season of 1999 (15 January, 16–19 February, 27 February to 7 March, and 1–3 April 1999). During these four periods, one single cloud band or a packet of cloud bands is generated every day at nighttime. Their length reaches several hundreds of kilometers, their width lies below 40 km, and their duration in some cases exceeds 24 hours. The smaller cloud lines observed behind the leading cloud line are narrower and are separated by a distance of 2–10 km. Cloud tops are about 3–8°C colder than the Arabian Sea, corresponding to an altitude between 500 and 1200 m during the night and below 2500 m during daytime. Cloud bands travel westward over the Arabian Sea at a speed around 13 m/s, greater than the wind speed measured in the surrounding area. The motion of the cloud band presents similarities with the wavelike propagation of atmospheric phenomena such as the Australian morning glories. Common elements to the different cases are the following: weak low-level winds close to the southwestern Indian coast when the cloud bands appear, winds with a northerly component in the vicinity of the northwestern Indian coast, the presence of a sea-land breeze circulation along the West Indian coast, and the presence of high concentrations of pollutants over the northeastern Arabian Sea. However, existing physical links between these elements (if any) still have to be investigated.
 The main objective of Indian Ocean Experiment (INDOEX) is to analyze the transport and evolution of aerosols and pollutants originating from the Indian subcontinent over the Indian Ocean and across the equator, and their direct and indirect effects on clouds, radiation and climate. Ship-borne, airborne and ground-based measurements have been carried out during the intensive field phase from 1 January 1999 to 31 March 1999. For this experiment, the Meteosat-5 geostationary satellite has been moved from its standby location at 10°W to a location over the Indian Ocean, at 63°E.
 This paper is focused on the detection and observation of westward propagating cloud bands expelled from the vicinity of the West Indian coast, using the full-resolution visible (VIS) and infrared (IR) images of Meteosat-5. Such a long cloud band can be observed on the VIS image of 6 March 1999 at 1230 LT, with LT for Indian local time, i.e., UTC + 5 h 30 min (Figure 1). The contrast of the image has been enhanced to show the large pollution plume on the eastern edge of the Arabian Sea and on the Bay of Bengal. The cloud band is located offshore the southern half of the West Indian coast whereas the pollution plume lies along the whole western coast. The exact nature of the pollution plume cannot be inferred from satellite images only: the observed VIS radiation may result from directly reflected or diffused sunlight, or be a mixture of the reflected light from cumulus smaller than the pixel size and from the sea surface. A few cumulus clouds along the southern part of the Western Ghats are the only clouds detectable over the Indian peninsula. Cloud bands are not observed off the eastern Indian coast on the Bay of Bengal.
 A similar phenomenon has been observed over the northern Arabian Sea from the space shuttle Columbia during April–May 1993 [Zheng et al., 1998a]. The altitude of the shuttle was near 260 km and the spatial resolution of the photograph was about 25 m. Cloud lines parallel to the Pakistan coast have been detected. The length of one leading cloud was around 250 km and the average distance between consecutive cloud lines was around 1 km. These cloud lines could be the signature of a solitary wave packet generated at 0430 LT and moving toward the south. Land breeze and katabatic flow are suggested as possible mechanisms for the generation of these wave packets [Zheng et al., 1998a, 1998b]. The North Australian morning glories appear in coastal areas in similar meteorological conditions: land breeze, katabatic flow, almost cloudless skies [Clarke, 1972; Smith, 1988]. The phenomenon observed several times during INDOEX shows common features with the northern Arabian Sea case: cloud bands parallel to the coast, moving in the offshore direction. Mountains are present behind the shoreline: the airflow comes from the dry Makran range culminating at about 2300 m in the Pakistan coast case and from the Western Ghats (culminating at 2694 m in the southern part) in the INDOEX case. Another related phenomenon, a coastal lee wave without outstanding leading cloud has also been observed over the Texas coast on Advanced Very High Resolution Radiometer (AVHRR) images provided by the NOAA 14 polar orbiting satellite [Li et al., 2001].
 The phenomenon described in this paper has been observed several times from a geostationary satellite. The spatial resolution of Meteosat-5 images (2.5 km for the VIS channel or 5 km for the IR channel) is much lower than the spatial resolution of the photographs taken from the shuttle (25 m). Thus the observation of small features such as the cloud lines with a wavelength of 1 km over the northern Arabian Sea is impossible at the coarser spatial resolution of Meteosat-5. On the other hand, the high temporal resolution (30 min between consecutive images) of the geostationary satellite radiometer enables the study of the time evolution and displacement of cloud bands that could not be investigated from the low-altitude orbiting shuttle. The shuttle passes only three times a day over an area (at times t − 93 min slightly to the east of the area, at time t over the area, at time t + 93 min slightly to the West) and then about 24 hours later [Capderou, 2003].
 This study presents a series of new observations of the phenomenon on Meteosat-5 VIS and IR images. Section 2 is devoted to the systematic detection of cloud bands parallel to the West Indian coast over the Arabian Sea, carried out every day on both VIS and IR images at 1230 LT during the extended period (1 January to 15 April) of INDOEX. Section 3 is focused on a “static” description (i.e., at 1230 LT) of the cloud bands detected in section 2 and their main characteristics (size, radiative temperature of the cloud tops, estimated altitude), whereas section 4 is devoted to the motion and time evolution of cloud bands. The comparison of analyzed winds with the velocity of the cloud band in section 5 will give elements for a limited interpretation of the observed phenomenon in the final discussion (section 6).
2. Occurrence of Cloud Bands Parallel to the West Indian Coast
2.1. Meteosat-5 Data
 Measurements of the radiances (W m−2 sr−1) by the radiometer are carried out in three channels. The VIS radiance is a measure of the radiation reflected and scattered by the observed surface (cloud, land, and/or water) between 0.4 and 1.1 μm. The thermal IR radiance is representative of the radiation emitted by this surface in the 10.5–12.5 μm band corresponding to a window for which the atmosphere presents a minimal absorption. The IR equivalent blackbody brightness temperatures (EBBTs) can be computed from the IR radiances with the help of calibration coefficients and radiance-temperature conversion tables provided by the EUMETSAT agency (European Organisation for the Exploitation of Meteorological Satellites). The real surface temperature is higher than the measured EBBT when water vapor is present in the low layers (due to important absorption in warm humid layers, but absorption is less important during the dry season). The EBBT is higher than the cloud top temperature (CTT) when pixels are only partially covered by clouds (such as small cumulus, or narrow cloud bands with a width below the pixel size) or are covered by semi-transparent clouds (such as high-level thin cirrus clouds). The third channel (WV), which measures the water vapor absorption between 5.7 and 7.1 μm, is useless for the detection of low clouds. The fact that the cloud bands detected in the VIS and IR channels cannot be observed in the WV channel indicates that these clouds cannot be located at a medium or high level.
 The temporal resolution of Meteosat-5 imagery of 30 min (providing 48 images per day) enables the tracking of cloud bands and the determination of their motion and lifecycle. Between 19 February 1999 and 14 April 1999, some of the 48 images were not available (from image 39 (0030 LT) to image 42 (0200 LT), depending on the day). The main reason for missing images is that the satellite enters the shadow of the Earth. During this period of eclipse, the imaging system is turned off for technical reasons: insufficient electrical power and spurious reflection of sunlight when the satellite enters into or exits from the Earth shadow.
 On the IR imagery, the contrast between low clouds and the sea surface is small because of minor differences between EBBTs. Nevertheless, IR images are required for the estimation of the altitude and the observation at night. The VIS channel is more sensitive than the IR channel to the presence of low clouds because the VIS brightness of low clouds largely exceeds that of the sea surface. The detection of cloud bands is better at midday because of the higher contrast of low clouds with the sea surface: high Sun elevation, reduced cloud shadows.
2.2. Systematic Detection of Cloud Bands
 IR and VIS pictures have been systematically checked every day at 1230 LT to detect cloud bands parallel to the West Indian coast over the central Arabian Sea. The observation of movie loops with previous and following images of the same day have confirmed the presence of the cloud bands.
 Results are presented in Table 1. Cloud bands can appear once a day. Sixteen cloud bands have been observed during four periods: 15 January, 16–19 February, 27 February to 7 March and 1–3 April 1999.
Table 1. Detection of Leading Cloud Lines Moving Westward Followed by Secondary Parallel Cloud Lines
Differences in equivalent blackbody brightness temperatures (EBBTs) between the leading cloud line top and the sea surface.
 The first cloud band (15 January) is very thin and cannot be tracked on the IR images because of its poor contrast. From 16 January to 15 February 1999, no cloud band parallel to the West Indian coast could be detected. During the second period (16–19 February), cloud bands become larger but no linear cloud formation is detectable on 17 February 1999 on both VIS and IR images. The end of February and beginning of March is the most favorable period for the occurrence of cloud bands with the longest series (nine consecutive days). At the beginning of April, cloud bands change shape; they become shorter and wider.
Table 1 lists the occurrences of the cloud bands we have detected on VIS and IR images. It should be noticed that cases with cloud lines narrower than the pixel size (5 km) cannot be detected. In other cases, when the contrast between the cloud and its surroundings is small, the cloud is barely detectable.
3. Selected Cases of Cloud Bands at 1230 LT
 Selected cloud bands are described with their surroundings. The earliest cloud band (15 January 1999) is shown on Figures 2a and 2b and a contrasted and long-living cloud band (6 March 1999) is shown on Figures 2c and 2d. The left panel shows the IR image and the right panel the VIS image, both at 1230 LT.
3.1. First Period (15 January 1999)
 On the VIS image (Figure 2b), the cloud band is well contrasted over its neighborhood. The leading line is located at 68.2°E, centered at 15°N, and its length reaches about 250 km. The minimum width of the most reflective part is one VIS pixel (2.5 km) and possibly less in the case of a partial coverage of the pixel by the leading cloud top. The reflectivity of the cloud top is high (VIS counts between 35 and 50).
 On the western side of the cloud band, the reflectivity is very low (VIS counts below 17), which is typical of the sea surface. The eastern side appears a little more reflective (VIS counts near 17) than the western side. This indicates that the air is clearer on the western side, possibly because of a subsidence, than on the eastern side where it contains a small quantity of aerosol particles. Another possible explanation is the presence of a higher number of small cumulus clouds (with partial pixel coverage) on the eastern side of the cloud band. In the vicinity of the cloud band and toward the south, the reflectivity slightly increases (VIS counts near 19).
 Three secondary cloud lines with an average separation of 2 to 3 VIS pixels (5–7.5 km) follow the cloud band. These lines are less reflective, narrower and shorter than the leading cloud line. The presence of a main cloud line followed by a series of secondary cloud lines may be the signature of a solitary wave packet.
 Because of the small difference of EBBT (1°C), it is almost impossible to distinguish between the leading cloud line and its surroundings in the IR channel (Figure 2a). The relatively high EBBT of the cloud band (top at 20°C) can be easily explained if it is composed of small cumulus partially covering the pixels. (This is also indicated by the smaller pixels of the VIS image.) The partial coverage of the IR pixels also makes the detection of trailing wavelets (i.e., secondary parallel cloud lines) almost impossible.
 The area on the east of the cloud band contains isolated cumulus clouds that have a higher top than the leading cloud line. On the IR image, the corresponding areas appear colder than the neighborhood by about 1°C. On the VIS image, these areas appear brighter (VIS counts near 50) than the neighborhood (VIS counts near 19). Thus fluctuations of the buoyancy in this region can lead locally to a stronger convection.
Figures 2a and 2b also shows the importance of the spatial resolution; the wave packet and its structure are observable on the VIS imagery with its 2.5-km resolution while they are practically not detected on the IR imagery because of the lower thermal contrast and the lower spatial resolution (5 km).
3.2. A Case of the Second Period (16 February 1999)
 During this period, cloud bands have been observed on 16, 18 and 19 February 1999. On 16 February, the central area of the leading cloud line is located around 68°E and 15°N, like the leading cloud line on 15 January. The differences between the western and the eastern sides of the cloud band both in EBBT and reflectivity become slightly higher than on 15 January 1999. Wavelets behind the leading cloud line are present and separated by 1–3 VIS pixels.
 The leading cloud line appears as a dashed line with more reflective and colder parts. No outstanding cloud band is detected on 17 February but on 18 February a continuous leading line is present again. Differences in the atmospheric circulation, i.e., between analyzed wind fields, are clearly observed between days with and without cloud bands (16 and 17 February respectively, see section 5).
3.3. A Case of the Third Period (6 March 1999)
 The 6 March case is one of the best observations, where the cloud band followed by smaller parallel cloud lines is outstanding with a cloudless area on its western side and low-level clouds mainly on its eastern side. On the IR image (Figure 2c, at 1230 LT), the leading cloud line is located at 67.5°E and between 7°N to 13°N, with its coldest part at about and 11°N and an extension of approximately 660 km. The width of the coldest part reaches 2 IR pixels (≈10 km); the corresponding cloud top temperature (CTT) reaches 18°C, and is 10°C colder than the sea surface (≈28°C) located on the western side. If the leading cloud line results from an ascending motion, several hypotheses may be done for the estimation of the cloud top level. If the sea surface temperature (SST) is the same as the air temperature, the dry adiabatic lapse rate is close to 10 K/km and the moist-adiabatic lapse rate in the cloud is close to 4 K/km (at 20°C and 850 hPa), then the cloud top height (CTH) can be estimated between 1000 and 2500 m. This is compatible with the thermal structure of trade-cumulus and with the INDOEX aircraft observations.
 On the VIS image (Figure 2d), the area on the western side of the leading cloud line is little reflective (VIS counts below 17). It corresponds to the sea surface, which has a higher EBBT in the IR channel (≈28°C) than on 15 January (≈22°C).
 On the eastern side of the cloud band, the area is warmer (25°C) and less reflective (VIS counts above 19) than the leading cloud line. However, compared to the sea surface, this eastern side is more reflective and colder. The lower temperature and the higher reflectivity indicate a higher particle content or partial cloud cover. Clear air could be subsiding ahead of the cloud band.
 Behind the leading cloud line, wavelets can be identified easily on the VIS image. They are less reflective, narrower and shorter than the leading cloud line, and the cloud tops are warmer and thus lower. The average distance between wavelets is about 1–2 IR pixels (5–10 km).
 Compared to the cloud band of 15 January, the width, length, and height of the 6 March cloud band have increased. The position of the cloud band has been displaced toward the South by about 4° in latitude (≈440 km) and the SST is warmer. Note also the presence of a smaller leading cloud band followed by secondary cloud lines located at the northwest of this large cloud band on the VIS image (Figure 2d).
3.4. Cases of the Fourth Period (April 1999)
 Cloud bands travel in an area further to the south and during spring, thus temperatures increase. SSTs are higher, and cloud bands move over areas with stronger convection and more cloudiness. In some cases, observation becomes impossible because cloud bands move under higher clouds.
3.5. Conclusion on the Systematic Observations at 1230 LT
 The pollution plume extends over the eastern part of the Arabian Sea while cloud bands are found over the southern half part of the West Indian coast (Figure 1). The average latitude of the central parts of the leading cloud lines (11°N according to Table 1) confirms this observation.
 The IR and VIS images at 1230 LT have shown that an area with clear air is often present on the western side of the cloud band, possibly due to subsidence. Among all cloud bands, the cloud band observed in January is the smallest one in size (height, width and length). Later, cloud bands are slightly displaced toward the south where a higher SST is also observed (due to the lower latitude and the higher Sun elevation).
 The leading cloud line is higher, longer, wider and more reflective than the following cloud lines. Secondary cloud lines are difficult to observe because of their small width and height.
 During the whole period, at 1230 LT, the central part of the leading cloud lines bands is located between 64°E and 69°E; this central position moves toward the southeast between January and April 1999, following a direction parallel to the West Indian coast.
4. Time Evolution of the Leading Cloud Lines
 The observation of the cloud bands on VIS images is limited to the period of the day with sufficient sunlight, i.e., between 0900 and 1600 LT. This does not allow the observation of the entire life cycle. For this purpose, the IR images are necessary, but a sufficient temperature difference between the top of the leading cloud line and the sea surface is required. Moreover, the use of IR data is limited by the absence of several images at night between 0000 and 0200 LT (developed in section 2.1), thus making the tracking impossible during that time.
4.1. Speed of the Leading Cloud Line on VIS Images
4.1.1. First Period (15 January 1999)
 The tracking is possible on the VIS images and the cloud band is shown at 0900, 1230 and 1600 LT on 15 January 1999 (Figures 3a–3c). Its central part covers the distance between 69.7°E and 67°E in 7 h at about 15°N; this corresponds to an average speed of 11 m/s.
 The aspect of the moving cloud band is modified during daytime because diurnal variations of the reflectivity of the cloud top occur. The observation of the cloud band is difficult on the IR images because the EBBT difference between the cloud top and the sea surface is too small (≈1°).
4.1.2. Third Period (6 March 1999)
Figures 3d–3f shows the cloud band on the VIS images at 0900, 1230 and 1600 LT on 6 March 1999. The leading cloud line is parallel to the Western Ghats and to the West Indian coast (Figure 1). The VIS count of the leading line is above 31 at 0900 LT, above 80 at 1230 and 1600 LT. Lower values (near 14, 16 and 14 respectively) are successively obtained for the sea surface on the western side of the leading cloud band. The area on the eastern side exhibits higher VIS brightness than the area of the western side. This area is brighter (VIS counts ≈19) at 1230 LT when the Sun elevation is higher.
 The cloud band moves westward in the offshore direction. The central part of the leading line is located at 69°E at 0900 LT and reaches 66°E at 1600 LT for a central latitude of 11°N. Thus the distance covered by the cloud band during 7 h is about 320 km; the average speed reaches 13 m/s.
 The observation of the leading cloud band on IR images is possible because the difference of the cloud EBBT relative to the sea surface and the resulting contrast on images are important. The tracking of the cloud band starts at 0230 LT and is continued after the last observation on a VIS image (1600 LT). A similar value of the velocity is obtained (13 m/s).
4.2. Leading Cloud Lines of Three Successive Days on IR Images
 The IR imagery is used to observe the cloud bands at 0230 LT. Figures 4a–4c shows the IR images for three days during the third period (5, 6 and 7 March 1999). The cloud band of the previous day is located on the western part of each panel at 62°E approximately, whereas the formation of the cloud band of the day starts at 72°E, close to the side of the land breeze cell where the air is suspected to rise (suggested by analyzed winds). The presence of a warm area on the western side of the cloud band is a significant element; this warm area stays in front of the moving cloud band during the first hours of its lifetime. On each panel, the distance between two cloud lines is about 10° in longitude at a latitude of 10°N. Assuming that the cloud bands are formed at the same time and same location, the separation of 1100 km corresponds to a displacement at a mean speed of 13 m/s.
 The IR images of 05 March 1999 (Figures 4d–4f) show that the temperature of the cloud top can also change significantly along the leading cloud line. The coldest part (23°C) of the cloud top is located between 12°N and 13°N at 0530 LT, between 11°N and 12°N at 0900 LT, and between 9°N and 10.5°N at 1230 LT. This cannot be interpreted as a southerly progression of the coldest part. It rather indicates small fluctuations or variations in the cloud top level, probably due to fluctuations of the buoyancy, convection or evaporation.
4.3. Trajectories of Clouds on IR Images
 Trajectories of different parts of the cloud band on 6 March 1999 have been computed with a technique [Szantai et al., 2001] derived from methods used to calculate cloud motion winds from satellite images [Schmetz et al., 1993; Désalmand et al., 1999]. Figure 5 shows the direct trajectories of clouds at different heights over the Arabian Sea retrieved on the IR imagery, on 6 March 1999 between 0900 and 2330 LT (this upper time limit is due to the eclipse, see section 2.1). In particular, three trajectories track small parts of the long leading cloud line.
 Practically, a complete trajectory is reconstructed by associating two trajectories starting at an instant when the cloud band was clearly identified (at 0900 LT in our case): one forward in time on a series of consecutive images, and one backward on the series of preceding images. The method follows a given square of pixels (32*32) by searching the highest correlation between two successive images at times t and t + 30 min for a forward (or direct) trajectory, and at times t and t-30 min for a backward trajectory. The tracking is possible as long as the contrast between the cloud and its surroundings is sufficient; it is interrupted when the tracked cloud undergoes important deformation, moves below other clouds or dissipates. This association of a direct and a backward trajectory is validated when the first vectors of the two trajectories have a consistent speed and direction. (The reconstruction of a trajectory-association of a direct and a backward trajectory-enables a better and longer tracking than the calculation of a single (direct) trajectory starting at the time when the cloud first appears and has a small size.)
 On Figure 6, the speed of the central part of the cloud band, measured along the reconstructed trajectory (between 0230 and 2330 LT, i.e., between two eclipse periods), is plotted versus time. The velocity first increases from 8 m/s (0230 LT) to 12.5 m/s (0830 LT), then remains stable during daytime (between 0830 and 2000 LT approximately), before increasing again, up to 15 m/s at 2330 LT. A similar value of velocity (around 12.5 m/s) has been measured from VIS images over a shorter period (limited to daytime).
4.4. Observation of the Entire Lifecycle of a Cloud Band
 The cloud band of 6 March could be tracked during almost its whole lifetime. The precise instant of formation could not be determined. On the last image before the eclipse period for the satellite (5 March, 2330 LT), no cloud line is visible in the vicinity of the West Indian coast on the IR image. After this period, the cloud band is identified (Figure 4b); on its western side, the sea surface reaches a temperature of 26°C. The cloud band then moves westward, away from the western edge of the land breeze cell. (The position of this cell can be crudely inferred from the IR image and from analyzed winds). With the assumptions of section 3, the cloud top level is limited to the range 500–1250 m, which is below the values at 1230 LT and compatible with the shallowness of land breeze cells. The cloud band then travels over the Arabian Sea, and is detected for the last time at 0430 LT on 7 March 1999. Thus its estimated lifetime is 26 h.
 The first position of the band at 0230 LT (6 March) is 72°E at 10°N, and the last position at 0430 LT (7 March) is 61°E at 10°N. During this time interval, the cloud band has covered a distance of approximately 1200 km, compatible with the average speed of 13 m/s.
 During the third period, the formation of a cloud band occurs once a day, at night, during or just after the period when the IR images are missing (2300–0230 LT). Figures 4a–4c also shows the presence of a strong thermal contrast at 0230 LT parallel to the West Indian coast.
5. Analyses of the ECMWF Fields of Temperatures and Winds
 European Centre for Medium-Range Weather Forecasts (ECMWF) wind fields are used for a very simple comparison of days with and without a cloud band. Because cloud tops have been estimated between 500 and 1200 m near the coast at night (probably just after the formation), we use wind fields at 925 hPa and 0530 LT (ECMWF winds are available four times a day). It is remembered that the spatial resolution of Meteosat-5 (5 km in the IR) and the width (5–10 km) of the observed structure are much smaller than the grid of the analyses (0.5°).
Figure 7 exhibits the wind fields, superimposed on the temperature fields in color for days with (left) and without (right) a cloud band: 16–17 February 1999, respectively with/without a cloud band (top), and 6–8 March 1999, respectively with/without a cloud band (bottom). A red line on Figure 7 represents the leading cloud lines.
 On 16 February 1999 (top left), a warm air mass is present over the Arabian Sea in the coastal area between 22°N and 16°N, with a core at 20°N. On 17 February 1999 (top, right) without a cloud band, the 22°C isotherm is closer to the coast. The atmospheric circulation is different in both cases. On 16 February 1999, very weak winds are observed between 20°N and 11°N along the West Indian coast. On 17 February 1999, without a cloud band, a relatively strong northerly wind at about 6 m/s blows over the eastern Arabian Sea. This could explain why the warm air covers a smaller area along the northern half part of the coast. Similar differences can be observed between 6 and 8 March 1999 (respectively bottom left and bottom right). Figure 7 also shows that winds are weak in the neighborhood of the cloud band (red line).
 From Figure 7, it seems that the presence of a warm air mass with calm winds is favorable to the formation of a cloud band. This warm air mass is slowly receding when a strong cold northerly outbreak comes close to the coast. The presence of a cloud band is not observed anymore in such conditions.
 Analyzed winds may help for the detection of sea and land breezes, although they have a coarse spatial resolution. At low levels (1000 and 925 hPa), one can notice that the wind in the coastal area is often weak and/or has a different direction from the open sea and/or inland airflow. The presence of the Ghats range behind the West Indian coastal plain may explain the change of strength or direction with the wind further inland. Another interpretation (explaining also the difference between offshore and coastal winds) is the presence of a sea breeze during midday and the afternoon, and the presence of a land breeze during the night. The nondetection of the return flow of the breezes on analyzed winds at a higher altitude may result from their limited horizontal and vertical resolution. For levels with a pressure of or below 850 hPa, the wind field is more regular and presents smaller fluctuations over the West Indian continental and coastal area (not shown). It has also been demonstrated that the land-sea breezes can be intensified below offshore winds and katabatic winds [Estoque, 1962]. The convergence of maritime and continental winds forces air to rise and cumulus clouds can form, as observed from ground and satellite radiometers [Simpson, 1994]. The cloud band always forms in the transition region between the continental and the maritime air masses. In our case, the presence of a band of cumulus clouds inland and parallel to the coast (observed on Meteosat images at 1230 LT, on Figure 1) is an indicator of the presence of a sea breeze.
 Suggested mechanisms for the formation of cloud bands involve interactions between the West Indian sea or land breeze and the easterly winds (possibly katabatic winds originating from the Ghats range) over the south of the Indian peninsula. Similarities with morning glories, i.e., long bands of clouds observed in Northern Australia and traveling with properties of a solitary wave [Smith, 1988], can be noticed.
6. Discussion and Conclusion
 A specific type of low-level cloud bands have been observed on several occasions during the INDOEX experiment and analyzed with the help of Meteosat images. In some cases, cloud bands are isolated structures. Otherwise, secondary parallel cloud bands that are smaller, narrower and less visible on images can follow a main cloud band. When the (leading) cloud band forms over the ocean at about 200–300 km off the coast at night, between 2300 LT and 0400 LT, it is parallel to the Western Ghats. On the western side of the cloud band, an area with a higher EBBT, probably due to the absence or a smaller coverage by cumulus clouds, can be noticed at the beginning of the lifetime of the cloud band. In one case, aircraft flights during daytime have shown that the cloud band is composed of cumulus clouds in higher density than in the surrounding area. It travels in the offshore direction over the Arabian Sea at higher speeds than the synoptic northerly flow (about 6 m/s), ranging between 10 and 16 m/s. The cloud top reaches a height of about 500–1250 m at night that can rise up to a height below 2500 m during daytime. The length of the band can be about 600–700 km, its width about 20–40 km (in one case, 2.5 km). Secondary cloud lines are separated by an average distance of 2–10 km.
 Favorable conditions for the formation of cloud bands are the presence of weak winds (≤10 m/s) off the southwestern Indian coast and the existence of winds with a northerly component close to the northwestern Indian coast (at 1000 and 925 hPa levels). Changes in the mesoscale and synoptic circulation may explain why cloud bands are present only on a few days of the observation period.
 The observed cloud bands have common characteristics with solitary wave phenomena such as morning glories: a long but narrow shape, a formation during nighttime and in presence of a land breeze, a faster propagation speed than the surrounding wind and the possible presence of secondary parallel cloud bands. However, the nature of the cloud bands may be different: uniform roll clouds for morning glories, cumulus clouds in higher density than in the surrounding area in the case of the cloud bands observed by aircraft during INDOEX.
 Cloud bands move in a strongly polluted environment (see Figure 1): increased concentrations of pollutants have been recorded in the Maldives Islands during or close to the four periods when cloud bands have been observed [Satheesh and Ramanathan, 2000, Figure 1]. A fraction of the pollutants is probably advected by northerly low-level winds from the northwestern part of India. Another portion may be expelled above the marine layer from the West Indian coastal plains by a mechanism involving sea-land breezes and influenced by the presence of coastal mountains, similar to the mechanism observed in the South Californian coastal area [Lu and Turco, 1994]. However, a relation between pollution events and the formation and evolution of cloud bands has not been established yet. Further measurements and observations under or in the vicinity of cloud bands, and complementary studies of the meteorological situation at synoptic scale are needed for a better comprehension of the processes triggering the formation and involved in the propagation of cloud bands. Aircraft and ship measurements as well as radiosoundings and constant level balloon soundings realized during INDOEX could help to complete this study.
 We acknowledge EUMETSAT for providing Meteosat-5 images and Jean-Louis Monge from Laboratoire de Météorologie Dynamique for the organization of the database ClimServ. This study has been supported by the French CNRS and University of Paris VI. We are grateful to Sethu Raman from the State Climate Office of North Carolina (North Carolina State University) for fruitful discussions and comments. We are grateful to the two reviewers for their useful comments to improve the paper.