The clouds are detected at a spatial resolution of 3 km along the orbit track horizontally and 480 m vertically after a linear interpolation of the signal along the vertical (the native vertical resolution of Aeolus is typically 1 km in the Tropical upper troposphere). Each single profile contains 40 vertical bins, and each 3 km × 480 m bin is either classified as “clear”, “cloudy” or “fully attenuated”. This last category corresponds to a bin where the laser is fully attenuated because the atmosphere above the bin has an optical thickness typically larger than 3 to 5. More details are given in Appendix A1.

We use this bin classification to discriminate between single profiles containing optically thin clouds (0.03 < τ < 3 to 5 ), such as in-situ or detrained cirrus clouds, from single profiles containing optically opaque clouds (τ > 3 to 5), such as convective clouds or the thickest part of anvil clouds, closer to the convective core (Fig. ). To do so, in each single profile containing at least one “cloudy” bin, we evaluate if all the bins between the ground and 1 km below the lowermost cloudy bin are fully attenuated. If this is the case, then the single profile is flagged as “opaque” (see profiles #4 and #5, Fig. ), otherwise, the single profile is flagged as “thin” (see profiles #2 and #3, Fig. ). Single profiles which contain no cloudy layer between the ground and 18 km of altitude are simply flagged as “clear” (see profile #1, Fig. )

Then, we consider all the single profiles falling within 2° latitude × 2° longitude boxes to build the gridded deep convective (cirrus) cloud fraction profiles, between 8 and 18 km of altitude. Those are three-dimensional quantities (2° longitude × 2° latitude × 480 m vertically), representing the number of 3 km × 480 m cloudy bins at a given altitude level among “opaque” (“thin”) profiles divided by the number of bins which are not “fully attenuated” at the same altitude level within the considered 2° longitude × 2° latitude grid box.

We further define diagnostics at high altitude, near the TEJ, between 14 and 17 km: the deep convective (cirrus) cloud fraction profiles at high altitude, which retain the values of deep convective (cirrus) cloud fractions but only between 14 and 17 km of altitude instead of 8–18 km.

In addition to the previous vertically resolved profiles, we define the deep convective (cirrus) cloud cover, which are two-dimensional (2° longitude × 2° latitude) quantities. They represent the number of opaque (thin) profiles containing at least one 3 km × 480 m cloudy bin between 14 and 17 km of altitude, divided by the total number of profiles within the considered 2° longitude × 2° latitude grid box.

Hereafter, we use Aeolus profiles of horizontal wind speed, noted uallsky in . Within a single profile, the value of the horizontal wind speed uallsky is given for each 480 m thick bin beside the cloud/clear flags (Sect. 2.1). We use the notation uallsky for the wind (instead of the notation vHLOS commonly used in the Aeolus literature) because the wind is observed both within clear and cloudy bins (but not within fully attenuated bins). We note positive (negative) uallsky values corresponding to eastward (westward) winds.

As the laser is pointed 35° off-nadir and perpendicular to the satellite track, away from the Sun, the Aeolus measurement is not the actual horizontal wind, but the horizontal projection of the wind component retrieved along the Line-of-Sight (LOS) of Aeolus. However, at the latitudes considered in this paper, the angle between the horizontal projection of the LOS and the West-East direction is between 8 and 10°. Consequently, Aeolus essentially retrieves the zonal component of the wind . The relative difference between the zonal component of the wind and the horizontal projection of the wind along the LOS is about 1 %. The restriction to the zonal component of the wind does not represent a limitation of our study, as the TEJ is also almost purely zonal .

We define three domains: the Indian Region (12–24° N and 53–98° E), the Arabian Sea (12–24° N and 53–75° E) and a third domain that encompasses the Bay of Bengal and a large part of India (12–24° N and 76–98° E). This analysis benefits from Aeolus sun-synchronous orbit configuration. Figure details an entire 7 d cycle of Aeolus orbits from 8 to 14 January 2020. On 8 January 2020, Aeolus crosses the middle of Arabian Sea domain at 06:00 LT, between 65° E at 24° N and 63° E at 12° N (dark blue orbit, Fig. ). On the same day, 12 h later, Aeolus ascending orbit overflies the Bay of Bengal domain at 18:00 LT on the very East, between 98° E at 12° N and 96° E at 24° N (dark blue orbit, Fig. ). On 9 January 2020, the orbits crossing each domain are shifted by about 300 km westward compared to the previous day orbits (lighter blue orbit, Fig. ). After 7 d, both domains are fully paved, and Aeolus follows exactly the same orbits with a 7 d periodicity. For example, Aeolus overflies the descending and ascending dark blue orbits (Fig. ) again on 15 January 2020 at 06:00 and 18:00 LT, respectively. On 3 d out of the 7 d cycle, the descending orbit of Aeolus over the Arabian Sea is located exactly 10.5°, or about 1150 km westward of the ascending orbit which overflew the Bay of Bengal or India 12 h before (see for example the distance between the yellow orbit over the Arabian Sea on the 14 January 2020, and the light green orbit over India on the 13 January, Fig. ). Consequently, an air parcel that is observed by Aeolus over the Bay of Bengal at 18:00 LT, and advected westward by the TEJ at about 25 m s⁻¹ will be located around the Aeolus orbit track 12 h later, over the Arabian Sea.

We compute the thermal wind at 06:00 LT along each Aeolus orbit in the Arabian Sea domain. To do so, we first compute the temperature differences ΔTNS(p) between the northern and southern borders of the Arabian Sea domain, at the mean longitude of the considered orbit, at each pressure levels (p) between 7 and 15 km of altitude, at 06:00 LT (more details are given in Fig. ): 1ΔTNS(p)=TN(p)-TS(p) Second, we compute the zonal component of the thermal wind between 7 and 15 km at 06:00 LT, Δug=ug(pz=15km)-ug(pz=7km): 2Δug=Rf∫pz=7kmpz=15kmΔTNSpΔydln⁡p, where ug is the geostrophic wind, R is the thermodynamic constant for dry air (287 J K⁻¹ kg⁻¹), f is the Coriolis parameter estimated at the latitude θ0= 18° N (the mean latitude of the Arabian Sea domain), pz= 7 km and pz= 15 km are the average pressures at the altitudes of 7 and 15 km respectively. The length Δy is given by: Δy=REΔθ, where RE is the Earth radius and Δθ= 15° is the latitude range over which the temperature difference is estimated.

We also estimate the longitudinal temperature differences ΔTEW between the eastern and western borders of the Arabian Sea domain, at each pressure level p between 7 and 15 km of altitude, at 06:00 LT: 3ΔTEW(p)=TE(p)-TW(p), where TE(p) is the mean temperature at the level p within the eastern border box defined as the area within 12 to 24° N and 75 to 79° E and TW(p) is the mean temperature at the pressure level p within the western border box defined as the area within 12 to 24° N and 49 to 53° E.

We compute the meridional component of the thermal wind between 7 and 15 km at 06:00 LT, Δvg=vg(pz=15km)-vg(pz=7km): 4Δvg=-Rf∫pz=7kmpz=15kmΔTEWpΔxdln⁡p The length Δx is given by: Δx=REcos⁡(θ0)Δλ, where Δλ= 26° is the longitude range over which the temperature difference is estimated.

We finally derive the projection of the thermal wind vector along the LOS of Aeolus: 5Δuthermal,ERA5=Δug2+Δvg2⋅cos⁡ϕ where ϕ is the angle formed between the thermal wind vector and the horizontal projection of the LOS of Aeolus. We note that because Δvg≪Δug, and because of the observation geometry, Δuthermal,ERA5 and Δug only differ by 1 m s⁻¹ at the highest.

The Indian Region is under the influence of the Sub-Tropical Jet (STJ) from January to May, with eastward winds exceeding 30 m s⁻¹ between 10 and 14 km of altitude (Fig. a). During the same period, both deep convective and cirrus cloud fractions observed by Aeolus, remain below 10 % (Fig. b, c). As the South-Asian Summer Monsoon (SASM) onsets in late May, a reversal of the wind (from eastward to westward) is observed: this is the TEJ. From mid-June to August, the TEJ is maintained and exceeds 30 m s⁻¹ between 14 and 17 km of altitude, before decaying in September (Fig. a). The wind remains westward until late October. During the active phase of SASM (late May–October), deep convective cloud fractions larger than 30 % and cirrus cloud fractions larger than 15 % are frequently observed over the Indian region between 14 and 17 km of altitude (Fig. b, c). Note that over the entire Indian region, profiles of deep convective and cirrus cloud fractions, estimated from the same orbits of Aeolus are well correlated. When SASM decays in late October, the Inter Tropical Convergence Zone (ITCZ) migrates southward and winds reverse again (from westward to eastward) as the STJ becomes the dominant zonal circulation (Fig. a). After the break of SASM, the deep convective and cirrus cloud fractions fall below 5 % between 14 and 17 km of altitude (Fig. b, c).

Figures a and b show the deep convective cloud cover observed during the active phase of SASM (June–October) at 18:00 and 06:00 LT respectively. The deep convective cloud cover is in general larger at 18:00 than at 06:00 LT. At 18:00 LT, the maxima of deep convective cloud cover between 14 and 17 km of altitude are located over the North-East of India (20° N, 80° E) and the South-West of the Bay of Bengal (12° N, 85° E), and reach 30 % (Fig. a). At 18:00 LT, the maximum over the North-East of India is expected as continental deep convection peaks in the late afternoon, around 18:00 LT, before decaying during the night to about 15 % at 06:00 LT (Fig. b). At the opposite, over the North-East of the Bay of Bengal (15° N, 90° E) that is more under maritime influence, the deep convective cloud cover is slightly below 20 %, at 18:00 LT (Fig. a) and increases to 25 %, at 06:00 LT (Fig. b). The diurnal contrast of deep convective cloud cover observed with Aeolus during the 2020 SASM over the Bay of Bengal is in good agreement with the one observed with passive imagery , in spite of the large inter-annual variations of the cloud patterns and precipitation over the Indian region and the specificity of the 2020 SASM . Over the Arabian Sea, we observe a diurnal contrast of the deep convective cloud cover, which is about 5 % larger at 18:00 than at 06:00 LT (Fig. c), consistent with . However, this diurnal contrast is weak compared to the one observed over the Indian continent.

The cirrus cloud cover also exhibits a diurnal contrast between 18:00 and 06:00 LT (Fig. f), although not as pronounced as that of deep convective clouds. At 18:00 LT, the cirrus cloud cover reaches values of above 20 % over continental India (Fig. d) and similar values above the entire Bay of Bengal. At 06:00 LT, the cirrus cloud cover remains at about 20 % over continental India (Fig. e), consistent with . One of the main reasons why the cirrus cloud cover diurnal contrast is much weaker than that of deep convective clouds is the longer persistence of cirrus clouds, as a part of the deep convective clouds at 18:00 LT are converted to cirrus clouds during the night . Over the Arabian Sea, characterized by a weaker deep convection at 18:00 LT, the cirrus cloud cover is only 2 % to 4 % larger at 18:00 than at 06:00 LT. The ratio between cirrus and deep convective cloud covers both measured over the Arabian Sea appears larger at 06:00 than at 18:00 LT, especially westward 65° E (Fig. i). We also observe that the zonal gradient of the cirrus cloud cover over the Arabian Sea at 18:00 LT is less pronounced than the zonal gradient of deep convective cloud cover at 18:00 LT (the cirrus cloud cover increases from 7 % to 20 % between 50 and 80° E, while the deep convective cloud cover increases from 5 % to 30 % between the same longitudes, Fig. i). This result is consistent with those of , who showed that the cirrus cloud cover has a larger spatial extent than the convective cloud cover, likely due to the spreading of anvil outflow effect. Finally from 06:00 to 18:00 LT, we observe a westward shift in both deep convective cloud cover and cirrus clover, which might be due to the TEJ.

The goal of this section is to examine if and how the convective patterns and locations influence the temporal variation of the mean TEJ speed (uallsky around 15 km of altitude) over the Arabian Sea during the SASM. In contrast to the studies of and , our study does not focus on the specific location of the TEJ core and its interannual variability, but instead on TEJ speed variation at a much shorter time scale (typically a few days).

To do so, we proceed in three steps. We first verify that at first order, the TEJ is in thermal wind balance during the SASM of 2020 (Sect. 4.2). Then, we make the hypothesis (noted H1) that short time scale fluctuations of the TEJ speed around the wind expected from thermal wind balance can be explained by the effects of deep convection patterns that act as perturbators and we test this hypothesis against observations (Sect. 4.3). Finally, we quantify the wind speed perturbations created by deep convection patterns and propose an explanation of the mechanism in play over the Indian region (Sect. 4.4).

All along this section we make use of the two following variables estimated over the Arabian Sea at 06:00 LT: the projection of the thermal wind vector along Aeolus LOS, noted Δuthermal,ERA5≈ug(z=15km)-ug(z=7km) (see Sect. 2.3) and the wind difference observed by Aeolus at the same altitudes as the thermal wind, noted Δuallsky=uallsky(z=15km)-uallsky(z=7km). We consider that the fluctuations of Δuallsky around Δuthermal,ERA5 are mostly due to uallsky(z=15km) rather than uallsky(z=7km). Indeed, we consider that uallsky(z=7km)≈ug(z=7km), and that the wind divergence and vorticity associated with deep convection (ageostrophic wind) are strongest at the top of convective towers . As a consequence, differences between Δuallsky and Δuthermal,ERA5 are due to the fluctuations of uallsky(z=15km) (i.e. the TEJ) around the geostrophic wind at the same altitude.

During the SASM of 2020, the median value of the difference Δuallsky-Δuthermal,ERA5 is equal to -2.1 m s⁻¹. This confirms that Δuallsky can be interpreted at the first order by the thermal wind set by the North-South temperature gradient, or, in other words, that the TEJ is in thermal wind balance (Δuallsky≈Δuthermal,ERA5). We define Δuthermal,ERA5′, the thermal wind corrected by the median of the difference Δuallsky-Δuthermal,ERA5: 6Δuthermal,ERA5′(t)=Δuthermal,ERA5(t)-2.1 Figure shows the time series of Δuallsky and Δuthermal,ERA5′ over the Arabian Sea at 06:00 LT between 1 June and 31 October 2020. As the SASM onsets in late May, both Δuthermal,ERA5′ and Δuallsky decrease at the same rate to reach values of about -35 m s⁻¹ in early August, coinciding with the date of the maximum intensity of the TEJ over the Arabian Sea, before increasing again at similar rates.

From June to October 2020, there are exactly 68 d where Δuallsky<Δuthermal,ERA5′, 69 d where Δuallsky>Δuthermal,ERA5′, and 16 d without Aeolus observations (Fig. ). We define uconv as the difference: 7uconv(t)=Δuallsky(t)-Δuthermal,ERA5′(t) As uallsky(z=7km)≈ug(z=7km), uconv can be seen as the convective perturbation of uallsky(z=15km).

To test hypothesis H1, we consider two ensembles of days in Fig. . The days where Δuallsky<Δuthermal,ERA5′ at 06:00 LT over the Arabian Sea are referred as “strengthened” wind days with respect to the thermal wind, and shaded in blue in Fig. . Oppositely, the days where Δuallsky>Δuthermal,ERA5′ at 06:00 LT over the Arabian Sea are referred as “weakened” wind days with respect to the thermal wind, and shaded in red in Fig. . The fluctuations of Δuallsky around Δuthermal,ERA5′ usually have a short time scale (of 1 d) but they can be as long as two weeks (in mid August 2020 for example). For each of these ensembles, we build the composite maps of deep convective cloud cover (between 14 and 17 km) observed at 18:00 LT the day before the winds are observed.

12 h before the strengthened wind days, we observe a maximum of deep convective cloud cover over the North-East of India (of about 40 %) and another maximum over the Bay of Bengal (of 29 %) (Fig. a). Over the South-East of the Arabian Sea, the deep convective cloud cover is about 20 % and on average 11 % over the entire Arabian Sea. In contrast, 12 h before the weakened wind days, the maximum of deep convective cloud cover is found over the centre and the West of India, while the deep convective cloud cover only reaches 23 % over the Bay of Bengal, about 30 % over the South-East of the Arabian Sea and 14 % on average over the entire Arabian Sea (Fig. b). We thus observe that the deep convective cloud cover at 18:00 LT over the Bay of Bengal is in phase opposition with that over the South-East of the Arabian Sea. The amplitude of the changes in the deep convective cloud cover between strengthened and weakened wind days is about 3 % over the Arabian Sea and 6 % over the Bay of Bengal (Fig. c). revealed the existence of a similar deep convection dipole at the same location, which exhibits a large interannual variability in July.

The deep convection dipole depicted in Fig. c is thus partly responsible for fluctuations of the wind downstream, particularly over the eastern part of the Arabian Sea. This is understandable having in mind that Deep Convective Systems (DCS) are associated at their top with a divergent outflow and a negative vorticity, which induce an eastward wind perturbation in the North-East and a westward wind perturbation in the South-West . described a similar phenomenon occurring over the Sahel region, which induces a westward (negative) perturbation of the TEJ of about -1.5 m s⁻¹ at the South-West of the DCS at around 200 hPa, and an eastward (positive) perturbation of the TEJ of about 1 m s⁻¹ at the North-East of the DCS at the same pressure level. further showed that the wind perturbation lasted up to 2 d after DCS development.

During strengthened wind days, 12 h after the deep convective cloud cover is maximum over the Bay of Bengal and North-East India, its divergent convective outflow, deflected in a clockwise motion by the Coriolis force, strengthens the westward wind at the South-West of the Bay of Bengal (Fig. c). This strengthening of the wind (a westward, thus negative wind perturbation uconv) is favoured by the weakening of the divergent convective outflow over the South-East Arabian Sea. Inversely, 12 h after the deep convective cloud cover is maximum over the South-East Arabian Sea, the divergent outflow at the North-East of this convection maximum induces an eastward, thus positive uconv perturbation over India, that is less counteracted by the weaker diverging outflow over the Bay of Bengal and North-East India (Fig. b).

We consider the variable Δyu which quantifies the difference of the derivative ∂∂y(uallsky) between strengthened and weakened wind days: 8Δyu=∂∂yuallskystrengthened-∂∂yuallskyweakened=∂∂yuallskystrengthened-uallskyweakened We estimate Δyu between the points AN and AS, respectively located in the North and South of the deep convective cloud pattern over the Arabian Sea (Fig. c) following Eq. (): 9ΔyuANAS=uallskystrengthenedAN-uallskystrengthenedAS-uallskyweakenedAN+uallskyweakenedASyAN-yAS where y(AN)-y(AS) is the distance between the points AN and AS. Similarly, we estimate Δyu(BN)-y(BS) between the points BN and BS, respectively located in the North and South of the deep convective cloud pattern over the Bay of Bengal (Fig. c). At 06:00 LT, we find a negative Δyu around the deep convective cloud pattern over the South-East of the Arabian Sea (ΔyuANAS=-1.04×10-6 s⁻¹), and a positive Δyu around the deep convective cloud pattern over the Bay of Bengal (ΔyuBNBS=1.69×10-6 s⁻¹). Although Δyu does not measure a vorticity difference between strengthened and weakened wind days (that would be impossible to estimate from Aeolus data as they do not contain any information on the meridional wind), these values are consistent with the existence of an anticyclonic circulation around deep convective patterns.

As the wind perturbations created by deep convection propagate during the night with the mean flow , we observe that on average over the whole Arabian Sea, uconv reaches -1.7 m s⁻¹ during strengthened wind days at 06:00 LT. On the other hand, over weakened wind days, uconv is equal to 1.3 m s⁻¹ over the Arabian Sea at 06:00 LT, so the total amplitude variation of the wind induced by the deep convective activity is about -3 m s⁻¹ on average over the Arabian Sea (statistically significant at a 90 % confidence level), and reaches up to -5 m s⁻¹ in its center (Fig. c). Note that when the deep convective cloud cover is maximum over the Bay of Bengal and North-East India, there is also a positive wind perturbation of up to 3 m s⁻¹, at 25° N, which is favored by the larger Coriolis parameter at this latitude (Fig. c).

Observational results of Sect. 3 suggest that deep convection is strongest over the Bay of Bengal and the Indian continent at 18:00 LT (Fig. a), providing a source of water in the upper troposphere. This water (vapour or ice crystals) may be advected westwards during 12 h by the TEJ, which shows the strongest winds over the Arabian Sea at 06:00 LT (Fig. h). As cirrus clouds found over the Arabian Sea are not always correlated with deep convective clouds found at the same place and at the same time (for example, cirrus clouds extend more towards the Western Arabian Sea than deep convective clouds observed at the same local time, see Fig. i), we can thus propose the following mechanism: the water injected over the Bay of Bengal and the Indian continent at 18:00 LT and advected by the TEJ during the night through cold temperatures is at the origin of the cirrus clouds found at 06:00 LT over the Arabian Sea (Fig. e). We attempt to establish this mechanism by testing the sensitivity of the cirrus cover over the Arabian Sea at 06:00 LT to: the wind (hypothesis H2, Sect. 5.2), the temperature (hypothesis H3, Sect. 5.3), the deep convective clouds above India and the Bay of Bengal at 18:00 LT (hypothesis H4, Sect. 5.4.). In Sect. 5.5, we draw a conclusion from our results and we study their implications in a warmer climate.

Figure shows the time series of uallsky averaged between 14 and 17 km of altitude over the Arabian Sea (AS) at 06:00 LT, noted uallsky(AS,06:00LT)‾. The bar denotes an average over all altitudes between 14 and 17 km. The median value of uallsky(AS,06:00LT)‾ is -23.6 m s⁻¹ between June and October. On the days where uallsky(AS,06:00LT)‾ is faster than its median value, noted “fast” wind days, the average wind speed is -29.9 ± 4.1 m s⁻¹. In contrast, on the days where uallsky(AS,06:00LT)‾ is slower than its median value, noted “slow” wind days, the average wind speed is -13.8 ± 7.7 m s⁻¹. Fast wind days correspond to the period of the TEJ peak, as they almost all occur between mid-June and the end of August.

We select days where uallsky(AS,06:00LT)‾ is faster (or slower) than its median value, and build the composite maps of the cirrus cloud cover observed at the same time (Fig. ). During fast wind days at 06:00 LT, the cirrus cloud cover reaches 25 % near the West coast of India (18° N, 70° E, Fig. a) and 14 % on average over the Arabian Sea. On the other hand, on slow wind days, the cirrus cloud cover is within the range of 2 %–15 %, and on average 6 % over the Arabian Sea at 06:00 LT (Fig. b). The cirrus cloud cover differences between fast and slow wind days (Fig. c) are the largest over the Arabian Sea in the North of 20° N, where they reach 12 %. The dependence of cirrus cloud cover on the wind speed, detailed by , is thus also observed in our study. We note that when spatially averaged over the Arabian Sea, the cirrus cloud cover increases by 8 % when the wind speed over the Arabian Sea doubles (from -13.8 to -29.9 m s⁻¹). In comparison, reported a relative increase in the cirrus cloud cover of about 20 % at 150 hPa close to the core of the TEJ when the wind speed doubles, and no increase at 70 hPa.

Figure a shows the PDFs of uallsky(AS,06:00LT) observed by Aeolus for each quartile of the cirrus cloud fraction over the Arabian Sea at 06:00 LT. Here, each occurrence of uallsky and cirrus cloud fraction corresponds to an average over the Arabian Sea on a given day for each 480 m vertical level between 14 and 17 km. The first quartile, median and third quartile of cirrus cloud fraction are delimited respectively at 0.4 %, 2.6 % and 8 %. We note that all PDFs intersect between -15 and -20 m s⁻¹. The first and second quartiles of cirrus cloud fraction (≤ 2.6 %) are dominated by winds slower than -15 m s⁻¹, while the third and fourth quartiles of cirrus cloud fraction (> 2.6 %) are dominated by winds faster than -20 m s⁻¹. Cirrus cloud fractions larger than 8 % are almost never observed over the Arabian Sea if uallsky(AS,06:00LT) is slower than -10 m s⁻¹. The occurrences of observations of cirrus cloud fractions larger than 8 % increase with wind speed and peak at -23 m s⁻¹, while the occurrences of having no thin clouds or cirrus cloud fractions lower than 0.4 % peaks at -16 m s⁻¹. The sensitivity of the cirrus cloud fraction over the Arabian Sea to the wind (H2) is thus well established.

To test hypothesis H3, we consider the ERA5 air temperatures co-located with Aeolus observations over the Arabian Sea from June to October 2020, during the SASM. We build the PDFs of ERA5 air temperatures (noted TERA5(AS,06:00LT)) for each quartile of the cirrus cloud fraction observed over the Arabian Sea at 06:00 LT in Fig. b. Each occurrence of TERA5(AS,06:00LT) here corresponds to an average over the Arabian Sea at 06:00 LT a given day and at a given 480 m vertical level between 14 and 17 km. The temperatures corresponding to the PDFs maxima are 197.6, 196.3, 199.2 and 196.7 K for the four quartiles (sorted by increasing values) of the cirrus cloud fraction. These temperature maxima are related to the altitude of the cirrus. The two lower quartiles (black and dark blue lines), are observed at the cold point tropopause (around 16 km) and above, where temperatures are very cold (below 195 K), which explains the large peaks of their PDFs centred on low temperatures. In contrast, the highest quartile (red line), with above 8 % cirrus cloud fraction, is observed between 14 km and the cold point tropopause, which explains that it is associated with both warm and cold temperatures. The second highest quartile (light blue), is typically observed below the cold point (15.2 km or less) and is thus in general associated with warmer temperatures than the three other quartiles. This shows that large cirrus cloud fractions are not necessarily associated with low values of temperatures averaged above the Arabian Sea, and conversely low values of cirrus cloud fractions are not associated with particularly high values of temperatures. Therefore we do not observe a significant dependency of cirrus cloud fractions on the mean temperatures experienced over the Arabian Sea between 14–17 km. Note however that these mean temperatures from the reanalyses could be affected by the presence of cirrus, that would have formed after having encountered colder temperatures. Indeed cirrus clouds can warm the upper troposphere locally by up to 4 K . In addition, these regionally averaged temperatures may hide small-scale processes. Cold temperature anomalies induced by gravity waves can form cirrus clouds but they are not captured in the regional mean temperatures. Thus, while we cannot thoroughly disprove H3 based on these results, we can conclude that the temperature is not the main or sole driver of cirrus clouds in this region.

In order to test our hypothesis H4, we show on Fig. c that for the four quartiles of deep convective cloud fractions observed over the Bay of Bengal at 18:00 LT (delimited by 6.9 %, 16.7 %, 26.2 %), the median values of cirrus cloud fractions observed over the Arabian Sea 12 h later at 06:00 LT are 3 %, 4.5 %, 5.5 % and 7 % respectively. Thus larger cirrus cloud fractions are associated with larger deep convective cloud fractions observed 12 h before at the same altitude, and the sensitivity of the cirrus cloud fraction over the Arabian Sea to deep convection (H4) is thus well established. This shows that deep convection is important for cirrus formation and persistence; deep convection provides a direct source of ice crystals in the upper troposphere and acts as a source for water vapour which is then advected by the TEJ to over the Arabian sea. Note however that comparing the Arabian Sea at 06:00 LT and the Bay of Bengal 12 h before, at 18:00 LT makes sense if the wind speed is sufficient to advect air parcels from the Bay of Bengal to the Arabian Sea.

We explore the relationship between deep convection and wind by categorizing each day into fast or slow wind days as described in Fig. . We build the composites of the deep convective cloud cover at 18:00 LT, depending on uallsky(AS,06:00LT), the wind being observed 12 h after the cloud cover is observed.

We note that 12 h prior to fast wind days, the deep convective cloud cover is 29 % on average over India and the Bay of Bengal (Fig. a), and maximum over North-East India at about 40 %; while 12 h prior to slow wind days, it is on average 22 % and the maximum of deep convective cloud cover of about 30 % is shifted southward (15° N, 78° E; Fig. b). The largest deep convective cloud cover differences between fast and slow wind days are located over the North of India and reach 20 % (Fig. c), while the differences are weak over the Arabian Sea, indicating that the additional sources of water are not located where the cirrus are formed.

These changes in the locations of the maxima of deep convective cloud cover depending on the wind intensity can be compared with the results of Sect. 4. The patterns of differences are correlated but are not identical, as the composites are built here depending on the absolute value of the TEJ, whereas in Sect. 4, they were built depending on the TEJ deviation from the thermal wind.

These results also show that the fast wind days associated with larger cirrus cloud covers are also associated with larger deep convective cloud covers over the North of India and the Bay of Bengal, making difficult to point out which factor (the wind or deep convection) is the controlling one for the cirrus formation over the Arabian Sea.

The sensitivity of the cirrus cloud cover to the wind, noted in Sect. 5.2, might be explained by the relationship between the cirrus cover and the deep convective cloud cover (Sect. 5.4), being itself larger when the wind is more intense (Sect. 5.4), in which case the wind would not play a direct role in the cirrus formation. Alternatively, higher occurrences of cirrus clouds over the Arabian Sea may be the result of a combination of two controlling factors: faster winds and larger sources of water injected in the upper troposphere over India and the Bay of Bengal.

To disentangle the roles of deep convection and upper tropospheric wind in the occurrence of cirrus clouds, we show in Fig. d four PDFs of uallsky(AS,06:00LT) for different values of α, with: 10α=cirrus cloud fractionAS,06:00LT(deep convective cloud fractionBoB,18:00LT+cirrus cloud fractionAS,06:00LT) The parameter α is between 0 and 1, and for any given value of deep convective cloud fraction observed at 18:00 LT over the Bay of Bengal, α increases when the cirrus cloud fraction over the Arabian Sea increases 12 h later at 06:00 LT. Furthermore, for any amount of cirrus cloud fraction, α increases when the deep convective cloud fraction decreases. In Fig. d, we only use the data coming from days where the orbit over the Arabian Sea at 06:00 LT is 1150 km westward of the orbit over the Bay of Bengal 12 h before, at 18:00 LT.

Figure d shows that at for given deep convective cloud fraction observed over India or the Bay of Bengal at 18:00 LT, faster westward winds lead to a larger cirrus cloud fraction (larger α) over the Arabian Sea 12 h later, at 06:00 LT at the same altitude. This increase in α is particularly visible when the wind increases from -20 to -25 m s⁻¹, and then seems to plateau at -25 m s⁻¹, consistently with Fig. a, when the contribution of the deep convective cloud fraction is ignored. Indeed, if cirrus are detrained from the deep convective clouds over the Bay of Bengal at 18:00 LT, and advected at about 25 m s⁻¹ westward, they are observed with Aeolus 12 h later over the Arabian Sea, which could explain why the mode of the PDF of uallsky(AS,06:00LT) for the largest quartile of α (α> 0.35) is located at about -25 m s⁻¹ (Fig. d). For slower westward uallsky(AS,06:00LT), cirrus clouds detrained from deep convection over the Bay of Bengal do not make it to the Arabian Sea before the pass of Aeolus, which could explain why α < 0.03 is dominant, especially for westward uallsky(AS,06:00LT) slower than 20 m s⁻¹. In contrast, for westward uallsky(AS,06:00LT) faster than 25 m s⁻¹, the cirrus clouds may have traveled beyond the Aeolus passover at 06:00 LT; however, we still see higher ice crystal concentrations likely due to the large spatial coverage/long lifetime of cirrus clouds. This could explain why values of α> 0.03 are more often observed than values of α< 0.03 for westward uallsky(AS,06:00LT) faster than 25 m s⁻¹.

Overall, these diagnostics validate the general mechanism proposed in Sect. 5.1, as they show that both faster westward winds over the Arabian Sea, and larger deep convective cloud fractions over the Bay of Bengal and North-East India contribute hand in hand to an increase of the cirrus cloud fraction over the Arabian Sea in the morning. A schematic is given in Fig. to summarize our results.

The TEJ shows a decreasing trend in the reanalyses during the period 1958–1998 and the central TEJ intensity is expected to weaken by 18 % over the Indian Ocean by the end of the 21st century . In addition, noticed over 40 years a decrease in the number of tropical cyclones over the Bay of Bengal, and by the end of our century, the CMIP simulations project a suppressed rainfall over the tropical eastern Indian Ocean . This simultaneous decrease in the TEJ and deep convection over the East of India is consistent with our findings of Sect. 4, as we find a similar behaviour on an intraseasonal scale. Based on our results of Sect. 5, we predict that the decrease in convective water sources and the reduction in water advection due to a weaker TEJ will lead to a decrease in cirrus cloud cover over the Arabian Sea in the future climate.