During the Indian Ocean Experiment (INDOEX) Intensive Field Phase, 17 superpressure balloons drifting at lower tropospheric levels were launched from Goa to sample the motion and meteorological characteristics of polluted air masses originating from the Indian subcontinent. The mass flow, as expected, is attracted to the Intertropical Convergence Zone (ITCZ), which is reached in about 7 days. Successive trajectories show evidence of shifts in the flow regime, consistent with Meteosat-5 imagery and illustrative of the Madden–Julian oscillation. Balloons also show evidence of coastal anticyclonic circulations off the western coast of India, linked to both the orography of the Ghâts and the diurnal coastal wind mesoscale wind systems. These mesoscale circulations may be important in the transport of anthropogenic or mineral pollutants across the coast. The meteorological data gathered on board the balloons were systematically compared with European Centre for Medium-Range Weather Forecasts (ECMWF) analyses interpolated in time and space at the balloons' successive locations every half-hour. The agreement is generally good in terms of wind velocities, temperature, and humidity. However, above the marine boundary layer, balloon data seem to indicate a cold bias of analyzed temperatures and a northeasterly bias of horizontal winds. The underestimation of variability in the analyzed winds can be partly explained by mesh size smoothing in the analysis system.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The principal goal of Indian Ocean Experiment (INDOEX) is to study the transport and the evolution of natural and anthropogenic aerosols emitted from the Indian subcontinent, and their interaction with clouds, radiation and climate. In this study, precise knowledge is required for the transport and diffusion of pollution arising from the subcontinent. To contribute to this knowledge, a Drifting Balloons Experiment was conducted during the Intensive Field Phase by the Laboratoire de Météorologie Dynamique (LMD), with the support of the Centre National d'Etudes Spatiales (CNES, French Space Agency) and its Balloon Division, and with the collaboration of the Department of Physics, Goa University, and of the Indian Space Research Organisation for ground based support during launch operations. The Drifting Balloons Experiment was designed to provide direct information on the trajectories of polluted air masses originating from the Indian subcontinent, as well as in situ meteorological data within the polluted layers over sea; an additional objective was to help validate the set of INDOEX meteorological analyses and transport simulations.
 Drifting balloons, also called superpressure balloons or constant-level balloons have been used in meteorology for several decades. In the sixties and seventies, experiments such as GHOST (Global Horizontal Sounding Technique) [Lally et al., 1966], EOLE [Morel and Bandeen, 1973] or TWERLE (Tropical Wind, Energy Conversion, and Reference Level Experiment) [The Twerle Team, 1977] have demonstrated the value of tracking superpressure balloons from a satellite for studying synoptic-scale atmospheric motions. It is of course easiest to fly such balloons in the upper troposphere or in the lower stratosphere, where they are far from the surface and therefore protected from orographic hazards and humidity deposit. However, already in the seventies, boundary layer drifting balloons developed by the Laboratoire de Météorologie Dynamique (LMD) and the French Space Agency (CNES) were used successfully to sample the Southwest Monsoon airflow over the Indian Ocean during the Balsamine and Summer Monex experiments [Cadet and Ovarlez, 1976; Cadet and Olory-Togbé, 1978; Cadet et al., 1981]. To our knowledge, this kind of long duration, low-level superpressure balloons were not used since. We used them again during INDOEX to monitor the motion of the polluted air masses originating from the Indian subcontinent. Seventeen balloons were launched from Goa, on the west coast of India. The launch period extended from 15 January to 27 February 1999. The last balloon ended its flight on 8 March.
2. The INDOEX Constant-Level Balloons
 Superpressure balloons keep an approximately constant volume and therefore fly at quasi-constant density levels, acting both as semi-Lagrangian tracers of air parcels and as meteorological platforms. INDOEX balloons, produced by ZODIAC under CNES supervision, are inflated with helium. Their total mass, including instrumentation, energy, electronics and ballast, is around 9 kg. When fully inflated, the envelope has a quasi-spherical shape of 2.5 m in diameter (Figure 1); this size and weight allow the balloon to drift in layers comprised between the surface and approximately 800 hPa. The nominal superpressure (i.e., the pressure of helium minus outside pressure) is of the order of 100 hPa. As long as sufficient superpressure is maintained, the volume remains quasi-constant, except for slight thermal dilations or contractions producing low amplitude buoyancy oscillations mostly driven by the diurnal cycle. The dynamics of superpressure balloons has been studied by several authors [Hanna and Hoecker, 1971; Hoecker, 1981; Hirsh and Booker, 1966]. They are good tracers of horizontal motion close to the equilibrium density level (the level at which their mean density exactly matches that of the surrounding air). However, they may react to wind bursts by small high frequency vertical oscillations around this equilibrium level; they may also be sensitive to gravity waves. The most serious problem relating to flight level is water loading by rain, or by the occurrence of condensation when the temperature of the balloon envelope drops under dew point temperature; this latter case may happen when the balloon reaches saturated air, or whenever the helium gets sufficiently cooled by nocturnal radiation (it is not rare to find the helium cooler than the ambient air by about five degrees at night). Water retention by the envelope depends on its dimensions and characteristics: Mezeix et al. [1973a, 1973b] and Gorbinet  have shown that it may exceed one kilogram. Our own data on actual flights indicate that the mass of condensed water deposited on the envelope may often exceed five hundred grams, which makes the balloon drop by several hundred meters. When such large vertical oscillations happen, the balloon may get close to (or even reach) sea surface. In addition, balloons drifting close to sea level are more exposed to salt deposit. In any case, it is clear from our experience that the lower the balloons, the shorter their lifetime.
 When launching, the choice of nominal flight level and superpressure at flight level (usually 100 hPa), added to the knowledge of expected temperature at flight level and of the balloon volume, determine the helium mass; then the weight of the ballast is computed to ensure buoyancy equilibrium at flight level. Increasing the superpressure gives more resistance to the balloons in case of descent; however, a superpressure exceeding 160 hPa would lead to deterioration of the envelope, followed by rapid loss of the balloon. The balloon lifetime is limited by slow diffusion of the helium across the 100 μm thick envelope: the expected maximum lifetime is approximately one month. Figure 2 shows the exponential decrease of the superpressure of Balloon 1 due to helium diffusion with a timescale of about two weeks. Also visible in Figure 2 is the diurnal cycle associated with thermally induced buoyancy variations, producing both variation of helium pressure and variations of ambient air pressure following vertical excursions [Hamman, 1991]. As we shall see later, the larger amplitude of excursions or diurnal oscillations within the last week of the balloon life are attributable to moisture deposit.
 We now briefly describe the equipment and instruments on board the balloons. Power is provided by lithium batteries that ensure a three weeks lifetime for data acquisition and transmission. Instrumentation consists of:
one pressure sensor,
two temperature sensors,
a Global Positioning System (GPS) receiver,
data handling electronics,
an ARGOS transmitter.
 The electronics, the GPS receiver, the ARGOS antenna and the batteries are all located inside the envelope, therefore protected from salty water if the balloon hits, or gets close to the sea surface (Figure 3). The hygrometer and the two temperature sensors are fixed at the extremities of light arms 30 cm long, distributed at 120° of one another along the balloon's equator; the pressure sensor is located at the nadir pole. Additional instrumentation provides data for technological information: internal pressure and temperature, and the temperature of the hygrometer, necessary for correcting relative humidity values. Measurements are performed every 30 min and stored on board until they are sent through the ARGOS system. The balloons and their payload were prepared, tested and launched from the Goa University campus in Dona Paula, Goa.
3. Launching Strategy
 One constraint of the Drifting Balloons Experiment is that we had only one launch station. Therefore we were limited to investigating air masses leaving the Indian subcontinent in the vicinity of Goa. In addition, the low altitude of the balloons does not allow them to fly over land, especially across the western Ghâts. Because of that, and also because low-level balloons can be a hazard near airports for airplane take-off or landing, we had to be sure that from their very beginning, balloon trajectories would be directed over sea, another sampling limitation; this was checked before launch by performing trajectory forecasts. Thus we could not launch our balloons in some synoptic situations. In addition, we had to be careful of the sensitivity of the balloons to the land breeze and sea breeze systems: launching was systematically scheduled about two hours before the daily peak of land breeze, i.e., about 0400–0430 a.m. (local time, leading Greenwich meridian time by 5 hours and 30 min).
 Among the 17 balloons launched between 15 January and 27 February, the one with the shortest lifetime was Balloon 3 (a few hours). All balloons went over the Arabian Sea except one (Balloon 8) which went northeastward across India. Data from these two balloons (3 and 8) have been discarded. The mean lifetime was slightly over 7 days and a half. Figure 4 shows the histogram of the balloons lifetimes. Note that conditions were not always favorable for launch: from 27 January to 6 February for instance, all predicted trajectories did penetrate into land, so that not a single balloon was launched. Of all balloons, the one with the longest lifetime was Balloon 1 (19 days); this is most probably because that particular balloon was positioned at a relatively high flight level (around 805 hPa) and because its superpressure was larger than usual (120 hPa). The altitude of Balloon 1 was indeed adequate to fly in the aerosol plume. We were not able to repeat this configuration because of Air Traffic Control constraints, but this long flight is a good example of the capabilities of INDOEX balloons.
4. Balloon Motions and Air Motion Sampling
4.1. Vertical Oscillations
 Higher-level balloons were found more robust and long-lived than lower-level ones. Lower levels carry more humidity and chances of condensation occurring on the balloon envelope are greater. Due to their generally low level, the life of many balloons was indeed abridged by water deposition leading to sudden descent. In general, balloons descending close to the sea surface were unable to recover their nominal flight level afterward. Examples are given in Figure 5. Figure 5 right shows the altitude (or rather the pressure level) of Balloons 1, 9, 15 and 17 as functions of time; shaded vertical stripes identify successive nights at the balloon location, and sea level is indicated by plotting European Centre for Medium-Range Weather Forecasts (ECMWF) analyzed sea level pressure interpolated at the balloon position. Balloon 1 is drifting at a relatively high level (close to 805 hPa); its envelope remains dry almost all the time in its very long life (21 days). For balloons drifting in dry air, the amplitude of the diurnal oscillations in the vertical is around 5 hPa. Balloon 9 also keeps drifting in relatively dry air for almost 7 days; on the seventh night after launch, its envelope gets wet and it falls close to sea. At the next sunrise, the envelope starts drying and the balloon gets up again; however, either because it does not succeed in drying out during daytime, or because of the additional weight of salt, or because its rains in the afternoon, it reaches only half of its previous height, and the following night is fatal to it. Balloon 15 gets moisture on the third day at sunset, then oscillates in the marine boundary layer during 3 days until it dies. Balloon 17 remains dry for almost 9 days; it follows a slightly ascending motion (30 hPa in 9 days).
 The trajectories of Balloons 1, 9, 15 and 17 are shown in Figure 5 in parallel to the vertical oscillations. The entire set of trajectories, displayed in Figure 6, covers the whole Arabian Sea. Almost all trajectories are attracted to the Intertropical Convergence Zone (ITCZ) (located during that season around 5°S to 10°S), but none is actually crossing it. From this inability of the balloons to cross the ITCZ, we cannot actually infer that there is strictly no mixing between northern hemisphere and southern hemisphere air masses: indeed, balloons hardly resist moist conditions or strong uplifts and therefore they tend to die whenever they get close to strong convective systems. One conclusion we may draw is that in the mean, the time it takes for air masses coming from the Indian subcontinent to reach the ITCZ is of the order of one week. The balloon tracks depend on balloon levels, the tracks being more northerly at lower heights and more easterly at higher levels. Furthermore, the evolution of the balloon trajectories from January to March reflects gradual changes in the synoptic circulation and associated large-scale convective activity, linked in particular to the intraseasonal or Madden–Julian oscillation. During the second half of January, the flow over the eastern Arabian Sea slowly shifts from northeasterly to northerly (Figure 7): this evolution is consistent with the propagation of equatorial convective clusters from the south of the Arabian Sea to the south of the Bay of Bengal, quite clear on Meteosat-5 images (Figure 7). The rather impressive eastward motion of Balloon 1 along the ITCZ, from 29 January (45°E) to 4 February (87°E), illustrates a typical westerly wind burst. In mid-February, the trajectories are mainly southward, consistently with the location of convective activity near the Maldives Islands. In late February to early March, the airflow has evolved to a steady northeasterly regime, all balloons end up in the western Arabian sea or even on the African coast.
 The purpose of our Drifting Balloons Experiment was not so much to sample the low-level circulation over the Arabian sea, than to sample the outflow of polluted continental air. As mentioned above, the constraint of a single launching station at Goa was a limitation. Indeed, it has been shown that during the INDOEX Intensive Field Phase, pollution transport from Eastern India to the western Indian Ocean through the Bay of Bengal was abnormally high (M. Bonazzola et al., Interpretation of INDOEX atmospheric opacities with an Eulerian reverse transport method, submitted to EGS INDOEX Special Issue, 2000). Our experiment was certainly not fit to analyze those episodes. However, our trajectories must be fairly representative of the Bombay plume.
4.3. Local Circulations Near the West Coast of India
 The trajectories of all balloons launched after 21 January either exhibit a sudden change in direction at a distance of 200–300 km from the Indian coast (six cases), or even perform a complete anticyclonic loop within the same coastal zone (seven cases). This seems to be a signature of mesoscale coastal-orographic circulations, likely related to subsidence in the lee of the Ghâts during easterly flow regimes, as well as to the extension over sea of the diurnal sea breeze-land breeze systems. Indeed a diurnal signal seems to be rule in these local circulations: the sudden change in direction (usually from southeasterly to northeasterly) occurs typically during the first night after launch; similarly, the typical timescale of a loop is 24 hours (the only exception was Balloon 11, which remained trapped in the local circulation for 2 days). These coastal-orographic circulations and their 24 hour timescale may be important for aerosol transport and chemistry, due to the delay it implies for polluted continental air to reach open sea areas. To our knowledge, they had not been noticed before.
5. Comparisons of Balloons Data With ECMWF Analyses
 Balloon data may be particularly useful to validate operational analyses over the Arabian Sea and the Indian Ocean, where in situ data are scarce. Therefore we made systematic comparisons of our data with ECMWF meteorological analyses, given every 6 hours on a 0.5° × 0.5° regular longitude–latitude grid at the ECMWF model levels, distributed according to their hybrid pressure-sigma coordinate system [ECMWF, 1995]. In our comparison, we interpolate the analyzed data to the exact location of the balloon (time-longitude-latitude-pressure), using a four-dimensional gliding cubic-spline algorithm [Boor, 1978]. The interpolated analyzed data are then compared to the values given by the balloon instruments: temperature, humidity, and horizontal winds computed as centered differences between the following position and the previous one. The numerical simulation of balloon trajectories uses analyzed wind fields interpolated to our 15 mn Lagrangian model time step, to obtain the wind value at the exact location of each simulated balloon; then a second-order time marching technique is used to update the balloon position.
 Simulated air mass trajectories are increasingly used nowadays in all kinds of field experiments. In particular, INDOEX analyses heavily rely on backward trajectories providing information on the origin of polluted air masses measured from ships, aircraft or ground stations like the Kaashidhoo Climate Observatory. Drifting balloons offer the possibility to really check the reliability of these simulations. Indeed, Lagrangian simulations may suffer from a number of error sources, like the inability of the analysis to catch mesoscale structures, or phase errors when a structure has been detected. In addition, the errors in the wind analysis produce errors in the subsequent location of the simulated tracer, which feed back on the trajectory. In other words, trajectory errors arise from the combined effects of analysis errors and dispersion. Here we briefly discuss two cases, those of Balloon 2 and Balloon 15, illustrated in Figure 8.
Figure 8 compares the actual trajectory of Balloon 2 with our simulation using ECMWF analyses. We must note first that Balloon 2 was not affected by any coastal circulation regime. The two trajectories remain quite close to each other for several days, then dispersion takes over. This is a case where the wind analysis performs quite well in the analysis of air mass motion. The accuracy of the trajectory simulation of Balloon 2 is the same order as those found by the other authors [Stohl, 1998; Stohl and Koffi, 1998; Koffi et al., 1998; Bauman and Stohl, 1997]: about 17% after 24 hours, and 12% after 48 hours.
Figure 8 illustrates the case of Balloon 15, typical of the influence of a mesoscale coastal regime on the very start of the trajectory. In that case the resolution of the analysis is too coarse to catch the coastal circulation: the simulated balloon quickly leaves the coastal waters while the actual one performs a complete 24 hour loop, getting back to coast line on the following night before actually leaving the coastal waters. In this case the error after 1 day is already of the order of 500 km and increases everafter. We then tried to reconstruct coastal mesoscale circulations within the ECMWF analyses in the following way. Instead of using ECMWF analyses from Julian day 54, hour 0 (the approximate time of Balloon 15 launch), we interpolated the ECMWF analysis at Julian day 53, hour 12 in our LMD-Z atmospheric general circulation model, using a global grid of 192 × 144 points strongly zoomed along the west coast of India: resolution there was increased to 52 km by a factor 4 coordinate stretching, which allowed a more realistic description of the Ghâts. Then we integrated this model for several days and used the forecast to perform the Lagrangian simulation. In doing so, we succeeded in reconstructing a qualitatively acceptable local circulation at the time of balloon launching, the drawback being that we replaced analyzed fields by forecast fields, losing in this way part of what we gained in space resolution. Nevertheless the result is interesting. We actually simulate an anticyclonic loop (although too small and too far from the coast) in the balloon trajectory, and we decrease the error after 24 hours to approximately 200 km (Figures 9 and 10). We thus confirm the importance of local circulations in the transfer of polluted air from the Indian subcontinent to oceanic areas.
5.2. Temperature and Humidities
 Of our 17 balloons, eleven lasted at least 6 days. Three of them had hygrometry problems. We limited our analysis to the remaining eight balloons, of which we show here four representative samples.
Figure 9 compares observed and analyzed temperatures and humidities for Balloons 1, 9, 15 and 17. Negative biases of analyzed temperature are frequent; they appear here rather systematically on Balloons 1 and 17. For Balloon 9, the agreement between balloon data and analysis is very close; it is rather close also for Balloon 15, with the notable exception of day 3. On the whole set of balloons, we notice that the agreement of analyzed and measured temperature is always extremely close when the balloon oscillates in the marine boundary layer. This can be expected, as the analysis is strongly constrained in the lowermost by observed sea surface temperatures. This is seen clearly during the last days of Balloons 9 and 15. We must be aware that, when the balloon oscillates in the marine boundary layer, the balloon envelope is loaded with condensed water, and that may affect the temperature measurements. This does not seem to be the case, in view of the good agreement with the analyses obtained in those cases (In another experiment, we added on the same type of balloons a measurement of the temperature of the envelope from inside. We were then able to compare the skin temperature with the dew point temperature of the air. The difference was found exactly in phase with the occurrence of condensation, monitored by the vertical motion of the balloons. This tends to support the reliability of both temperature and humidity measurements, even in presence of condensed water). The temperatures measured by Balloon 17 agree with the analysis only at night, except during the two first days of flight; during day time, the analyzed temperatures are lower by three to four degrees.
 The measurements of relative humidity are always consistent with the vertical motion of the balloons. In general, there is also at least qualitative agreement between balloon measurements and ECMWF analyses. The transitions from dry to moist conditions are generally much faster in the balloon measurements, but this may often be due to mesh-scale smoothing inherent in the analyzed fields. Balloon 1 stays more than one week in very strong subsidence, with relative humidity dropping to values under 10%. Of the four moisture spells indicated by the analysis (days 21, 23, 25, 27), only one (day 25) is corroborated by our measurements. After 27 January, Balloon 1 gradually leaves the subsidence area in the northwest Indian Ocean to approach the northern side of the ITCZ and humidity increases. This moistening occurs about 2 days later than in the analysis. However, the large differences observed in the period from 23 to 28 January between measurements and analyses are not so significant, because the balloon spends all that time making several loops in a rather small area. The fact that both measurements and analyses show the occurrence of moist spells is a good point for the analysis. Nevertheless, looking at the whole moisture history, we may conclude that the analysis seems to overestimate the ratio of moist to dry areas. What we often observe (e.g., Balloons 15) is that the analysis underestimates levels of contrast and sharpness of transitions; but again, this may just be an artifact of coarse gridding.
 The comparison between winds retrieved from balloon data and ECMWF analyses is displayed in Figure 10 for Balloons 1, 9, 15 and 17. Concerning Balloon 1, the analysis obviously underestimates the variability of both zonal and meridional motion. For instance, the fast westward velocities attained during the night of 16–17 January and on 21 January are underestimated by about a factor two; the oscillations of both zonal and meridional velocities associated to the multiple loops off the Somali coast from 23 to 29 January are similarly underestimated; and finally, the eastward velocity corresponding to the westerly wind burst after 29 January is significantly faster than expected from the analysis. Similar underestimation of the variances is conspicuous on the three other balloons.
Figure 11 displays some statistics on the comparisons between balloon data and analyses. The close agreement of temperatures in layers nearest to sea surface can be explained from the SST constraint on the analysis; at the same time it confronts the quality of temperature measurements on board the balloons. Above the lowermost layer, the bias increases to about two degrees in daytime, and 1.5 degrees at night. The comparison of temperature variability is less conclusive; however, the analysis seems to underestimate it close to sea surface. This is perhaps an indication of a too strong SST constraint on the lowest layer; on the other hand, we expect more variability on local values than on grid quantities.
 Systematic biases are also observed on the wind components. The easterly bias close to 800 hPa is only produced by an underestimation of the westerly wind burst encountered by Balloon 1; it may not be significant of the whole period. Nevertheless, there seems to be systematic easterly and northerly biases of the order of 0.5–1 m s−1 that persist throughout the season. Again the underestimation of wind variability by the analyses may be partly explained by grid-scale averaging; but from individual histories of the balloon we note that synoptic-scale wind variability also tends to be underestimated.
 Following one reviewer's comment, we checked whether the absence of anticyclonic loops within the coastal zone in the analysis impacted the wind biases. Dropping these loops in our statistics does not qualitatively change our results (Figure 12). The same reviewer brought to our attention that the balloons were spending more time in light wind regions where the wind is not necessarily as light in the analyses: this could produce a bias toward stronger winds in the analyses. Again we dropped the light wind (<1 m s−1) cases in our statistics, and again this did not qualitatively change our results (Figure 12).
 During the INDOEX Intensive Field Phase, a set of superpressure balloons drifting at lower tropospheric levels, was used to sample the trajectories and meteorological properties of polluted air masses originating from the Indian subcontinent. The balloons were launched from Goa, from mid-January to the end of February 1999. They displayed some of the main characteristics of the pollution plume over the Arabian Sea and western Indian Ocean. The mass flow, as expected, is attracted to the ITCZ. The mean life time of the balloons (about 7 days) is also the mean time it takes for the air masses to reach the ITCZ. Balloons show evidence of shifts in the flow regime. In these the Madden–Julian oscillation appears prominent, with an eastward shift of equatorial convergence areas from south of the Arabian Sea to south of the Bay of Bengal, in the second half of January and beginning of February. This shift is corroborated by Meteosat-5 imagery. Around mid-February the flow is mainly southward; at the end of February and beginning of March, all trajectories are attracted to the western margins of the Arabian sea. Balloon trajectories also show evidence of anticyclonic circulations extending 200–300 km off the western coast of India. These circulations, linked both to orographic subsidence behind the Ghâts in easterly conditions, and to diurnal land breeze-sea breeze oscillations, must influence the transport of pollutants from continent to sea.
 The meteorological data gathered on board the balloons were systematically compared with ECMWF analyses interpolated in time and space at the balloons successive locations every half-hour. The agreement is generally good in terms of wind velocities, temperature and humidity. Above the marine boundary layer, the balloon data generally indicate a cold bias of one to two degrees Celsius on the analyzed temperatures, and a northeasterly bias reaching at most 1–2 m s−1 on the analyzed winds. The agreement of analyzed humidity with our measurements is qualitatively very good, though the analysis tends to underestimate the extent of dry areas. The latter bias, however, is partly due to smoothing associated with analysis truncation at the model mesh size. This smoothing also partly accounts for the underestimation of temperature and velocity variances by the analysis. Nevertheless, synoptic variations in wind velocities, associated with specific loops or oscillations in the balloon trajectories, seem also often underestimated.
 We thank A. P. Mitra from National Physical Laboratory in Delhi. We also thank E. Desa, L. V. G. Rao, C. K. Gopinathan, and many other scientists from the National Institute of Oceanography, Goa, for their kind and valuable help with communication logistics. Many thanks are also due to T. N. Krishnamurti from Florida State University and to the Director of the India Meteorological Department in Paniji for providing twice daily soundings. Finally, we are greatly indebted for the European Centre for Medium-Range Weather Forecasts, which provided us daily real-time analysis and forecast during the duration of the experiment. This project was funded by the Centre National d'Etudes Spatiales (CNES) and the Programme Atmosphère-Océan à Mésoéchelle (PATOM). Computations were done on the computing facilities of the Institut pour le Développement des Ressources en Informatique Scientifique (IDRIS).