Effect of deep convection on the TTL composition over the Southwest Indian Ocean during austral summer

Balloon-borne measurements of CFH water vapor, ozone and temperature and water vapor lidar measurements 15 from the Maïdo Observatory at Réunion Island in the Southwest Indian Ocean (SWIO) were used to study tropical cyclones' influence on TTL composition. The balloon launches were specifically planned using a Lagrangian model and METEOSAT 7 infrared images to sample the convective outflow from Tropical Storm (TS) Corentin on 25 January 2016 and Tropical Cyclone (TC) Enawo on 3 March 2017. Comparing CFH profile to MLS monthly climatologies, water vapor anomalies were identified. Positive anomalies of water 20 vapor and temperature, and negative anomalies of ozone between 12 and 15 km in altitude (247 to 121hPa) originated from convectively active regions of TS Corentin and TC Enawo, one day before the planned balloon launches, according to the Lagrangian trajectories. Near the tropopause region, air masses on 25 January 2016 were anomalously dry around 100hPa and were traced back to TS Corentin active convective region where cirrus clouds and deep convective clouds may have dried the layer. An 25 anomalously wet layer around 68 hPa was traced back to the South East IO where a monthly water vapor anomaly of 0.5ppbv was observed. In contrast, no water vapor anomaly was found near or above the tropopause region on 3 March 2017 over Maïdo as the tropopause region was not downwind of TC Enawo. This study compares and contrasts the impact of two tropical cyclones on the humidification of the TTL over the Southwest Indian Ocean. https://doi.org/10.5194/acp-2019-1072 Preprint. Discussion started: 22 January 2020 c © Author(s) 2020. CC BY 4.0 License.

. Laser pulses are generated by two Quanta Ray Nd:Yag lasers, the geometry for transmitter and receiver is coaxial and the backscattered signal is collected by a Newtonian telescope with a primary mirror of 1200 mm diameter. 387 nm (N2) and 407 nm (H2O) Raman shifted wavelengths are used to retrieve the water vapor mixing ratio. Depending on the scientific investigations, specific filter points and integration times can be chosen. The raw vertical resolution is 15 m. Data are smoothed with a low-pass filter using a Blackman window. Based on the number of 120 points used for this filter to vertically average the data, the vertical resolutions are 100-200 m in the lowest layers, 500 m in the mid-troposphere, 600 m in the upper troposphere and 700-750 m in the lower stratosphere. In order to convert the backscattered radiation profiles into water vapor mixing ratio profiles, the calibration coefficient is calculated from water vapor column ancillary data: GNSS (Global Navigation Satellite System) IWV (Integrated Water Vapor). The description of the calibration method and the total uncertainty budget can be found in Vérèmes et al. (2019). 125 At the Maïdo Observatory, the lidar provides 4 to 8 water vapor profiles per month. The calibrated water vapor profiles of Lidar1200 database extends from November 2013 to December 2017. The time slot of routine operations is around 19:00 to 01:00 (+1) local time but there are intensive periods of observation during field campaigns that allowed longer measuring span. The Raman lidar water vapor observations were validated during the MORGANE intercomparison exercise in May 2015 (Vérèmes et al., 2019). During the MORGANE campaign, CFH radiosonde and Raman lidar profiles 130 showed mean differences smaller than 9 % up to 22 km asl.
Here we used the Raman lidar measurements for two nights when the CFH sondes were launched at the Observatory (25

Satellite data
The brightness temperatures of the infrared (IR) channel at 10.8 μm of the geostationary weather satellite  have been used to provide the regional characteristics of deep convection over the Indian Ocean. The satellite centered at 57.5°E provided images for the Indian Ocean from December 2005 to March 2017. 140 Aura Microwave Limb Sounder (MLS) v4.2 water vapor and ozone data were included in the study to compare with the in situ measurements and to evaluate the spatial extent of the convective air masses measured at the Observatory. In particular we have used water vapor from the Stratospheric Water and OzOne Satellite Homogenized (SWOOSH) data set (Davis et al., 2016). The SWOOSH dataset contains monthly mean stratospheric water vapor and ozone profiles from several satellite instruments for the period 1984 to present. The data are available on a 3D (longitude/latitude/pressure) grid. The SWOOSH input data for the period August 2004 to present day correspond to measurements from the Aura MLS satellite. The MLS water vapor data are available on a pressure grid with 12 levels per decade change in pressure between 1000 and 1 hPa (e.g. the vertical resolution is ranging from 1.3 to 3.6 km between 316 and 1 hPa). The estimated accuracy for MLS water vapor decreases from 20% at 216 hPa to 4% at 1 hPa and is ~ 10% in the TTL region (150-70 hPa).

Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) onboard Cloud Aerosol Lidar and Infrared Pathfinder Satellite 150
Observation (CALIPSO) makes backscatter measurements at 532 nm and 1064 nm since June 2006. We use the Total Attenuated Backscatter coefficients β'532 available from the CALIPSO V4.10 level 1 lidar data products. Following Vaughan et al. (2004), the attenuated scattering ratio SR532 (Equation 3 of Vaughan et al., 2004) profiles are computed as the ratio of β'532 corrected for molecular attenuation and ozone absorption and the molecular backscatter coefficient βm. βm is calculated using the number density of molecules from the GEOS 5 global model of the NASA Global Modeling and Assimilation 155 Office (GMAO), and the Rayleigh scattering cross section given in the CALIOP Algorithm Theoretical Basis Document (ATBD, cf. Equations 4.13a and 4.14).

Model
The origin of air masses measured at the Maïdo Observatory were assessed using the FLEXible PARTicle (FLEXPART) Lagrangian Particle Dispersion Model (Stohl et al., 2005). FLEXPART is a transport model that can be run either in forward 160 or backward mode in time. FLEXPART was driven by using ECMWF analysis (at 00, 12 UTC) and their hourly forecast fields from the operational European Centre for Medium Range Weather Forecasts -Integrated Forecast System (ECMWF-IFS). In March 2016, ECMWF introduced a new model cycle of the IFS into operations with a grid-spacing of 9 km roughly doubling the previous grid-spacing of 16 km used since January 2010. The ECMWF model has 137 vertical model levels with a top at 0.01 hPa since June 2013. To compute the FLEXPART trajectories, the ECMWF meteorological fields were 165 retrieved at 0.50° and 0.15° and on full model levels from the Meteorological Archival and Retrieval System (MARS) server at ECMWF. The 0.50° fields were used to drive the FLEXPART model over a large domain configured as a tropical channel, i.e., the domain is global in the zonal direction but bounded in the meridional direction (at latitudes ± 50°).
Furthermore, higher-resolution domains can be nested into a mother domain in a FLEXPART simulation. Thus, to have a better representation of convective transport associated with mesoscale convective systems or tropical cyclones with a 170 horizontal dimension on the order of a couple of hundred kilometers over the SWIO, we included a nest domain covering the SWIO region (cf. Figure 2). If a particle resides in the high-resolution nest, the ECMWF meteorological data at 0.15° from this nest are interpolated linearly to the particle position. If not, the 0.50x0.50° ECMWF meteorological data from the mother domain are used to compute the trajectories. Retrieving high-resolution ECMWF fields from the MARS server for FLEXPART consists in several steps which are: -retrieve the meteorological model data output from ECMWF (horizontal winds, temperature, humidity, surface fields) -compute total and convective precipitation rates, sensible and latent heat fluxes from the surface -calculate the vertical velocity from the continuity equation Therefore, the ECMWF high-resolution vertical velocity field already contains a convective mass flux component from the 180 Tiedtke scheme used in ECMWF. The convective scheme used in the ECMWF-IFS, originally described in Tiedtke (1989), has evolved over time. Changes made include a modified entrainment formulation leading to an improved representation of tropical variability of convection (Bechtold et al. 2008) and a modified CAPE closure leading to a significantly improved diurnal cycle of convection (Bechtold et al. 2014). Particles are transported both by the resolved winds and parameterized sub-grid motions, including a vertical deep convection scheme. FLEXPART uses the convective parameterization by 185 Emanuel and Zivkovic-Rothman (1999) to simulate the vertical displacement of particles due to convection. The results from model runs with and without cumulus scheme in FLEXPART have been compared to assess whether convective mass fluxes could be resolved in the higher-resolution nest domain. The results of FLEXPART runs with and without cumulus scheme look fairly similar (not shown) and thus here we will present only the model results with cumulus scheme turned off.
To determine the transport history of air masses sampled by balloon launches, a so-called retroplume was calculated 190 consisting of 10,000 back trajectory particles released from each 1 km layers of balloon launches used in this study, and advected backward in time. The initial positions of the 10,000 particles were distributed randomly within each 1-km vertical layer and a 0.10°x0.10 o longitude-latitude bin centered on the balloon location. The dispersion of a retroplume backward in time indicates the likely source regions of the air masses sampled by the in situ instruments.  Madagascar. It became a category 4 tropical cyclone on March 6 at 18:00 UTC, with 10-minute maximum sustained winds at 194 km/h. Enawo reached its peak intensity at 06:00 UTC on March 7, with ten-minute maximum sustained winds at 204 km/h and the central pressure at 932 hPa. TC Enawo reached Madagascar's northeastern coast on March 7 at around 9:30 220 UTC and was the third strongest tropical cyclone on record to strike the island. After March 8, TC Enawo gradually weakened to a tropical storm while moving southward over Madagascar.
The two balloon launches at the Observatory on 25 January 2016 and 3 March 2017 were specifically planned using FLEXPART Lagrangian trajectories and METEOSAT 7 infrared images. The goal was to sample the convective outflow from TS Corentin and TC Enawo as well as convection north of Madagascar on 24 January 2016. 225   Figure 3).

Climatological and monthly mean water vapor distributions.
Overall during El Niño conditions, water vapor mixing ratios at 215 hPa are enhanced over the SWIO west of 80°E. Ho et al. (2006) have studied the variations of TC activity in the South Indian Ocean in relationship to ENSO effects. During El Niño 245 periods TC genesis was shifted westward, enhancing the formation west of 75°E and reducing it east of 75°E. Therefore, on January 2016 the peak of water vapor west of 80°E at 215 hPa may be related to an increase in convection associated with strong El Niño conditions. The Quasi-Biennial Oscillation (QBO) also affects TTL temperatures and humidity (e.g. Zhou et al., 2001;Yuan et al., 2014;Davis et al., 2013). Thus, we computed the climatological water vapor concentrations at 100 hPa over the SWIO according 250 to the phase of the QBO. Following Davis et al. (2013), we defined a QBO index as the zonal mean (10°S-10°N) of the difference in the ERA-Interim zonal wind at 70 and 100 hPa. A positive QBO index (u70hPa -u100hPa > 0) corresponds to westerly shear conditions and the warm phase of the QBO (Baldwin et al., 2001).  data correspond to daytime measurements (balloon launches at ~11 UTC) while the CFH water vapor data correspond to nighttime measurements in order to coincide with water vapor lidar measurements at the Maïdo Observatory. Overall good agreement is seen between the lidar and CFH water vapor profiles over the whole troposphere. Note that the CFH water vapor profiles were not used to calibrate the lidar water vapor profiles as explained in section 2.2.
The altitude range 2-12 km on 25 January 2016 is moister by ~50% than the same altitude range on 3 March 2017 (mean 275 water vapor mixing ratio of 5076 ppmv and 4375 ppmv between 2 and 12 km on 25 January 2016 for the CFH and lidar respectively versus 3335 ppmv and 3398 ppmv on 3 March 2017 for the CFH and lidar respectively). The austral summer season, with warmer temperatures and greater cloudiness, reaches its peak in January/February and this could explain in part the higher humidity observed in January than March. In addition, January 2016 corresponded to a strong El Niño period and this could lead to higher tropospheric moistening associated with ENSO (Tian et al., 2019). On 3 March 2017, a moist layer 280 was observed between ~12 and 16 km in both CFH and lidar water vapor profiles with corresponding low ozone values. On 25 January 2016, a similar layer of moist air/low ozone is observed between ~ 9 and 14 km. The lidar smooths out the peak of water vapor at 10 km observed on 25 January 2016 but this could be due to the longer integration time used for that night observed on 4 February 2016 could be explained by the fact that TC Enawo was closer to the island (~902 km north of the island), was still intensifying and was a stronger system than TS Corentin. Above ~17 km the ozone profiles on 300 January/February 2016 and March 2017 are more similar to the climatological mean ozone profile, suggesting that deep convection did influence the upper troposphere but not the lower stratosphere. We will later show using FLEXPART that the moist/low ozone layers in Figure 5 are associated with the convective outflow of a mesoscale convective system north Madagascar on 23 January 2016, TS Corentin and TC Enawo. Figure 6 shows the CFH profiles of RHice (computed using the Goff-Gratch equation [Goff and Gratch, 1946]  distance between the Maïdo Observatory and the airport is ~20 km so while boundary layer temperature values will differ for 345 the two sites, free troposphere/TTL temperature distributions can be compared as they are less influenced by topography.

Relative humidity and temperature profiles 305
The seasonal mean CPT height is 17.31 km for the period December-March with a mean CPT temperature of 193.90 K (Table 1). The tropical tropopause is higher and colder during austral summer as a response to large-scale upwelling in the tropical stratosphere  and convection (Highwood and Hoskins 1998) A large positive temperature anomaly is observed on 25 January 2016 over a broad tropospheric region from 2 to 16 km (mean amplitude of +2.5 K) with a peak warming of +4.6 K at 10km. On 3 March 2017, a warm temperature anomaly is 355 mostly observed between 6 and 14 km (mean amplitude of +1.1 K) with a peak value of +3.1 K near 12 km. The stronger warming of the troposphere observed in January 2016 may be due to the strong 2015/2016 El Niño. The connection between interannual variations in tropical tropospheric temperature and ENSO is well established (e.g., Soden 2000). Using 13-year of temperature data from the tropospheric channel of the microwave sounding unit (MSU-2),  showed that a tropospheric warming occurs almost uniformly over the tropics and that the 360 magnitude of the warming is around 0.5-1°C for strong El Niño years. Chiang and Sobel (2002) updated the analysis of Yulaeva and Wallace to include the response to the strong 1997/98 El Niño (ONI of +2.2 K in DJF 1998) and indicated MSU-2 temperature anomaly of ~1.2 K in January 1998 (cf. Figure 1 of Chiang and Sobel, 2002). Note that the MSU-2 temperature data used in these studies provide a measure of the mean temperature of the 1000-200 mb layer (corresponding to the surface to ~ 11 km using a scale height of 7 km). Thus, part of the strong tropospheric warming ( The mean amplitude of the 16-19 km temperature anomaly is -1.6 K with a maximum cooling of -3.6 K at 18 km. A similar feature is observed on 3 March 2017, with a cooling between 14 and 17 km with a mean amplitude of -2 K and maximum cooling of -4.5 K at 15.1 km. The upper tropospheric warming and near tropopause cooling observed on both dates is consistent with a temperature response to deep convection (e.g. Sherwood et al., 2003;Holloway and Neelin, 2007;Paulik 375 and Birner, 2012). The cooling around the tropopause can be explained by either radiative cooling by cirrus clouds over the regions of deep convection (Hartmann et al., 2001) or diabatic cooling through convective detrainment (Sherwood et al., 2003;Kuang and Bretherton, 2004). CPT properties can also be modified by convectively driven waves (Zhou and Holton, 2002;Randel et al., 2003).
To assess the effects of deep convection on temperature in the upper troposphere and near the tropopause, we looked at the 380 distribution of deep convective clouds in the days preceding the soundings. The location of deep convective clouds can be https://doi.org/10.5194/acp-2019-1072 Preprint. Discussion started: 22 January 2020 c Author(s) 2020. CC BY 4.0 License. assessed by using maps of METEOSAT 7 infrared brightness temperature. Figure 8 shows convective cloud coverage for the 3-day period preceding the sonde launch date at the Maïdo Observatory. Convective cloud coverage was estimated using 3hourly METEOSAT 7 infrared brightness temperatures at 5 km resolution. A threshold of 230 K is used to detect deep convective clouds in the METEOSAT 7 brightness temperature data (i.e. pixels with brightness temperatures less than 230 K 385 correspond to convective clouds). This threshold has been previously used to identify convection on geostationary satellite infrared images (e.g. Tissier et al., 2016). This temperature corresponds to a height of about 11 km in the NDACC/SHADOZ climatological-mean summertime profile of temperature on They looked at temperature anomalies corresponding to low ozone anomalies between 12 and 18 km, thus temperature anomalies influenced by deep convection. A strong warming was observed near the level of main convective outflow at ~12 395 km and cooling was more pronounced above ~ 15 km and near the CPT at ~17 km. Thus, the upper tropospheric warm temperature anomalies as well as cold temperature above 15 km and near the tropopause on Figure 7 are coherent with a deep convective temperature signal. Paulik and Birner's study also showed that the amplitude of the temperature anomalies increases as convection strengthens with a warming of ~2K in the upper troposphere and a cooling of around -3K near 16 km (cf. Figure 5 of Paulik and Birner, 2012). Using CloudSat observations of deep convective clouds and COSMIC GPS 400 temperature profiles, they showed that the deep convective temperature signal (i.e. anomalously warm upper troposphere and an anomalously cold upper TTL) was only present for deep convective clouds above 15 km. Although the magnitude of the temperature anomalies decreases with increasing distance from convection, they observed a deep convective temperature signal during DJF ~3500 km away from the convective event. Within 1000 km of the deepest convection (deep convective clouds above 17 km), the convective temperature anomaly exceeds 0.75 K in the upper troposphere and ranges from -1 K to 405 -2.0 K near 16 km. In our case, the deepest convective clouds with cloud tops colder than 190 K are 1000 km away from the individual events while their deep convective temperature signal was estimated using 4 years of COSMIC data. Therefore, their estimates correspond to an average deep convective temperature signal; such a signal is likely larger when considering larger/more organized convective events such as tropical storms. 415

CFH and MLS comparisons
The CFH measurements analyzed in this study are further compared to coincident MLS profiles. The match criteria used are ±18h, ±500 km North-South distance (around ±5° latitude), ±1000 km East-West distance (around ±10° longitude). The To compare the high-resolution CFH water vapor profile to the MLS satellite data, we smooth the high resolution sonde 425 measurements to match the resolution of the satellite profiles using the MLS vertical averaging kernels, following the procedure described in Read et al. (2007) and Davis et al. (2016). The procedure for applying the MLS averaging kernels to a CFH profile requires an a priori profile as input; this is the same a priori profile used in the MLS retrieval. Figure  were prepared by the same operator and calibrated using the same recommended procedure. During these two flights, the 480 CFH data streams were transmitted to receiving equipment on the ground through the Intermet radiosonde. From an instrumental standpoint, there is nothing that might explain a CFH dry bias on 3 March 2017 compared to 25 January 2016.
Unfortunately, the CFH sondes are not recovered on the island after each flight as they land in the ocean and thus it was not possible to examine in more details the instrument after the flight on 3 March 2017. To our knowledge, the CFH instrument on that night has measured as well as it could in the stratosphere. Even though the CFH instrument launched on 3 March 485 2017 had a dry bias of 1 ppmv in the stratosphere, such bias does not affect the results of this paper found for TC Enawo.

FLEXPART Lagrangian analysis
The convective origin of air masses sampled in the upper troposphere and near the tropopause during the passage of TS Corentin and TC Enawo is evaluated using the FLEXPART Lagrangian model. Figures  According to FLEXPART, the 14-15 km layer measured above the Maïdo Observatory on 25 January 2016 ~18:30 UTC has two different origins. A day before, 69% of this air mass was below 10 km (with ~29% below 5 km) and ~1000 km northeast 510 of Réunion Island in a region with convective clouds with cold brightness temperatures less than 220 K (~12 km). Therefore, we can infer that the majority of the 14-15 km air mass was lifted by convection associated with TS Corentin a day prior to the launch. These trajectories are rather spread in the lower troposphere, suggesting that they experienced turbulent mixing and changes in wind direction in the lower troposphere. The rest of the trajectories are located higher in altitude, in the 10-15 and 15-17 km altitude ranges. They are also located above convective clouds, but are less scattered than the trajectories in 515 the lower troposphere, suggesting that these trajectories were less mixed with the surrounding upper troposphere.
2 days before (Figure 10b), 80% of the 14-15 km layer originated from the lower and middle troposphere (54% within the 0-5 km layer, 26% within the 5-10 km layer) over the northeastern convective region of TS Corentin, and 20% from the upper troposphere and near tropopause region (13% within 10-15 km, 6% within 15-17 km) above TS Corentin. The upper tropospheric branch had an anticlockwise rotation with an origin near TS Corentin, in agreement with the upper divergence 520 associated with TS Corentin. Hence, most of the 14-15 km air mass was located either in the lower troposphere or near the top of convective clouds 2 days before.
The 17-18 km layer measured at Maido on 25 January 2016 stayed in the upper troposphere and near the tropopause a day before before reaching Réunion Island. The trajectories followed an anticlockwise rotation associated with Corentin's dynamics. No trajectories that originate in the lower troposphere were found. On 24 January at 17 UTC (2 days before the 525 launch), the trajectories were located ~250 km north of the center of TS Corentin. Note that TS Corentin reached its peak intensity on 23 January 2016 at 06 UTC (pressure at the center of 975 hPa, ten-minute maximum sustained winds of 111 km/h). Hence, according to FLEXPART backtrajectories and the METEOSAT 7 infrared images, the origin of the 17-18 km layer was traced back to the active convective regions of TS Corentin and its upper divergence dynamics, but no trajectories originated from the lower troposphere. However, due to the 0.15° spatial resolution of the ECMWF winds used to drive 530 FLEXPART, the vertical updrafts of the deepest convective clouds that may reach the tropopause region/lower stratosphere  and 60°E are used to compute the mean MLS profile for that day. 575 On both dates, the CFH convolved profiles and the mean of MLS coincident profiles are drier than the MLS monthly averages at 316 and 261 hPa with relative differences ranging from -10 % to -70 %. The mean relative difference with the climatology for these two pressure levels is ~ -20% for the CFH convolved profiles and the means of MLS coincident profiles respectively.
On 25 January 2016, the mean of MLS coincident profiles and the CFH convolved profile show a peak of ~ 30% in the 580 relative difference with the MLS climatology but the pressure level of this peak differs in the two profiles with a peak at 178 hPa for the CFH convolved profile and 147 hPa for the mean of coincident MLS profiles. Overall, the region of moistening in the upper troposphere is broader by ~ 1.5 km in the mean of MLS coincident profiles compared to the CFH convolved profile. MLS water vapor data are retrieved on a grid having 12 levels per decade change in pressure, corresponding to ~1. They showed that tropical cyclones can hydrate a deep layer of the surrounding upper troposphere by ~30-50 ppmv or more 605 within 500 km of the eye compared to the surrounding average water vapor mixing ratios (cf. Figure 3 of Ray and Rosenlof, 2007). They also looked at the evolution of UT water vapor changes as a function of the storm intensity as measured by the peak wind speed (cf. Figure 5 of Ray and Rosenlof, 2007). In both the Atlantic and western Pacific basins, the average water vapor at 223 hPa around the storm center steadily increased from 4 to 5 days prior to peak cyclone intensity to 2 days following peak cyclone intensity. The average water vapor enhancement in the two ocean basins was from 5 to 20 ppmv 610 with an increase as high as 30-40 ppmv for some cyclones in the western Pacific. The CFH launch on 3 March 2017, 18 UTC occurred 3.5 days before Enawo reached its peak intensity on 7 March at 06 UTC (pressure at the center of 932 hPa, ten-minute maximum sustained winds of 204 km hr -1 ) and the storm center was ~ 700 km away from the island. Thus, deep convection associated with TC Enawo may have caused the strong increase in UT water vapor observed on 3 March 2017.
Ongoing work with MLS data to apply the methodology of Ray and Rosenlof (2007) to assess hydration of the UTLS by 615 tropical cyclones for the 2004-2017 cyclone seasons in the southwest Indian Ocean is under way. This will be the focus of a future study but preliminary results indicate water vapor differences of 35% to 48% at between 178 and 261 hPa for categories 2 to 4 hurricanes on the Saffir-Simpson scale. Ray and Rosenlof (2007) indicated that tropical cyclones hydrate a deep layer of the UT in the vicinity of the cyclones by up to 50% above monthly mean water vapor mixing ratios. Therefore, our estimate of UT water vapor increases of 20 to 100% using CFH&MLS data for TS Corentin (Category 1 hurricane at its 620 peak intensity) and TC Enawo (Category 4 hurricane at its peak intensity) are in broad agreement with our estimates based  Jensen et al. (1996) showed that ice clouds formed by large-scale vertical motions can result in depletion of water vapor mixing ratio by about 0.4 ppmv. Chae et al. (2011) investigated temperature and water vapor changes due to clouds in the TTL using MLS, CALIPSO and CloudSat datasets. They noted that generally 640 clouds humidify the environment near 16 km (~100 hPa) or lower but dehydrate the TTL above 16 km.
On 25 January 2016, CFH and MLS data are 10% (+0.4 ppmv) and 17% (+0.7 ppmv) moister than the climatological values at 68 hPa, above the tropopause. Observational and modeling studies have indicated that overshooting convection can moisten the lower stratosphere by injecting water vapor or ice crystals directly above the overshooting clouds (e.g. Danielsen, 1993;Corti et al., 2008;Dauhut et al., 2015;Frey et al., 2015;Allison et al., 2018). In our case, the observation 645 on 25 January 2016 was not made close to the deepest convective clouds that were ~1000 km north of the island (Figure 8 island.
It is difficult to conclude whether TC Enawo had a direct impact on water vapor in the lower stratosphere by using only the CFH observation on 3 March 2017. The FLEXPART analysis indicated that the CFH sounding did not sample the lower stratosphere downwind of Enawo. 655 Ongoing work with the mesoscale model Meso-NH at a 2-km resolution for TC Enawo for the period 2-7 March 2017 indicates that deep convective clouds within 500 km of the cyclone eye can inject ice crystals and moisten the lower stratosphere, resulting in an average anomaly of ~2ppmv within 500 km of the tropical cyclone eye. The strongest humidification in the lower stratosphere (17-19 km; ~88-66 hPa) was found after March 4 when the storm stalled over the ocean (while intensifying) and after March 6 when it reached its peak intensity. Thus, the CFH observation on 3 March 2017 660 was made before TC Enawo had influenced the lower stratosphere above 100 hPa. This is further confirmed by the fact CALIOP did not have a lower stratospheric signal on Figure 6.
Tropical cyclones are unique among tropical convective systems in that they persist for many days and thus could affect the UTLS more than other mesoscale convective systems. Clouds in tropical cyclones often reach to and sometimes beyond the tropopause (e.g., Romps and Kuang 2009). Allison et al. (2018) have investigated the vertical transport of water vapor by the 665 2013 tropical cyclone Ingrid in the North Atlantic. Results of their high-resolution numerical simulations indicated that hydration occurred between 17.5 and 21 km (83 to 56 hPa) due to the injection of ice crystals. As the exact role of deep convection, and tropical cyclones in particular, in hydrating the lower stratosphere is still under debate, additional TTL observations of water vapor and modeling work are needed to quantify the overall impact of convection on TTL and LS water vapor. High-resolution (2 km) numerical simulations of TC Enawo for the period 2-7 March 2017 are underway to 670 gain a closer look at the effect of TC convection on TTL temperature and water vapor. This work will be the subject of a subsequent study.

Summary
Two balloon launches from the Maido Observatory were specifically planned using the FLEXPART Lagrangian model and This study showed the impact of two tropical cyclones on the humidification of the TTL. It also demonstrates the need to develop balloon borne high precision observations in regions where TTL in-situ observations are sparse, such as the tropics and the SWIO in particular. High-resolution accurate observations of water vapor are needed to document the impact of tropical cyclones and deep convection in general on the TTL. The impact of tropical cyclones on the TTL water vapor budget will be analyzed in a more quantitative way using MLS data and tropical cyclones best tracks from 2004 to 2017 in a 700 subsequent paper. In addition, the impact of deep convection and overshooting clouds within TC Enawo on the water vapor budget of the TTL will be analyzed using high-resolution (2 km) mesoscale simulation of TC Enawo.
https://doi.org/10.5194/acp-2019-1072 Preprint. Discussion started: 22 January 2020 c Author(s) 2020. CC BY 4.0 License.        shown with respect to the METEOSAT7 cloud distribution at those times. The altitude range of the particles (e.g. 0-5km) and the percent of particles in that altitude range are indicated according to a color code shown on the bottom of each panel.