The vertical profiles of BC used in this study have been obtained using an Aethalometer (model AE42; Magee Scientific, USA) aboard high-altitude zero-pressure balloons launched from the National Balloon Facility at Hyderabad, (17.48∘ N, 78.47∘ E, 557 m a.m.s.l.) during three ascents made on 17 March 2010 and 8 January and 25 April 2011. The first and third flights corresponded to pre-monsoon conditions, while the second flight occurred in winter. The details of the experimental set-up, calibration, data collection and analysis are available in and only a brief account is given here. The 109 755 m3 zero-pressure balloon was made of 25 mm linear low-density polyethylene film and was capable of carrying ∼ 350 kg of payload up to a ceiling altitude of ∼ 35 km. The mass concentration of equivalent black carbon (EBC; ) was measured using an Aethalometer, which estimates the mass concentration of EBC by measuring the change in the transmittance of a quartz filter to 880 nm of light upon the deposition of aerosol. The value of the effective mass absorption cross section (MAC) used in these measurements is 16 m2 g-1 . The effective MAC includes the amplification of absorption due to multiple scattering on the filter fiber matrix and the decrease due to shadowing. The Aethalometer was configured for volumetric flow with an external pump providing a flow rate of 14 L per minute (LPM) at ground and operated at a time base of 5 s. The data from the Aethalometer was telemetered down along with the GPS coordinates. A few studies have reported uncertainties in Aethalometer-estimated EBC (for e.g. ). Following these suggestions from the reports, we have used an amplification factor of 1.9 and an R factor (shadowing effect) of 0.88 in this study. Nevertheless, when the aerosols are aged or mixed with other species, as they would be away from direct emissions, the shadowing effect would be negligible . In addition to BC, dust also absorbs shortwave radiation, but the mass absorption cross section (MAC) for dust is 2 to 3 orders of magnitude lower than that for BC at 880 nm . Hence if the mass of dust is substantially higher than EBC, then only the same optical absorption will be produced at 880 nm. Thus, under normal conditions, the effects of dust on measured EBC mass concentrations would be negligible.

In addition to the Aethalometer, each ascent also carried other payloads, such as a GPS receiver, meteorological sensors for temperature and RH, telemetry and telecommand systems. The entire payload was tested using a thermo-vacuum facility to ensure consistent operations up to ∼ 9 km of altitude (250 hPa pressure level and ∼ -40 ∘C). Beyond this altitude, the ambient pressure drops too low to provide sufficient flow to the Aethalometer. The telecommand system was used to switch off the instrument for the higher altitudes and switch it on again at this altitude during the descent phase. A gondola carrying the payloads was attached to the balloon using a parachute, which was deployed in the descent phase and enabled the safe recovery of the payloads for reuse. Ballast cans attached to the gondola ensured a slow and steady ascent of ∼ 2.6 m s-1, while the descent speed was controlled by the parachute, as the ballast cans were detached after the balloon rose to the free tropospheric altitude. The scientific data along with the housekeeping data were collected during the ascent and descent phases and were continuously telemetered to the ground in addition to being stored on-board. The total flight duration was about 3 to 4 h for each flight and the spatial ground spread of the flight path was within a 50 km radius of the launch site. The data were analysed (following details in ) and the profiles obtained were smoothed using a running mean filter. The same protocols were followed for all the flights.

We used the regional chemistry transport model, WRF-Chem, to simulate the observed vertical profiles of BC and to understand the causes behind the sharp layers of BC at elevated altitudes. The WRF-Chem model was employed with a horizontal grid spacing of 2 km with 70 vertical levels and 100 m of vertical resolution around the layers at 4–6, 8–9 and 10–11 km. Such a high vertical resolution was chosen to resolve the BC peaks accurately when simulated by the model. Of the remaining levels, 10 are located within the boundary layer (0–2 km). A relatively coarser resolution (500 m) is set for the remaining altitude bands, i.e. 2–4, 6–8 and 9–10 km. Beyond 11 km, the vertical resolution is set to be 1 km. The details about the vertical levels in the model simulations can be found in Table . The model domain (75.5–80.5∘ E, 14.5–20.5∘ N) centres at Hyderabad and spans around 330 km in each direction. The simulations were started 16 days prior to the balloon flight days (flight day 1 was 17 March 2010, flight 2 was 8 January 2011 and flight day 3 was 25 April 2011) and were run until the end of the flight day. In these simulations, the convective processes were not parameterised due to the fine grid spacing; cloud microphysical processes were parameterised using the Thompson scheme , boundary layer processes were parameterised using the YSU scheme and the surface processes were modelled using the RUC LSM scheme . The shortwave radiation processes were parameterised using the Dudhia scheme , while the long-wave processes were parameterised using the rapid radiative transfer model (RRTM) . Gas-phase chemistry in these simulations was handled by MOZART mechanisms , while the aerosol-phase chemistry was parameterised using the GOCART bulk aerosol scheme with the Fast-J scheme for photolysis .

The model takes into account the following aerosol species: BC1 (hydrophobic), BC2 (hydrophilic), OC1 (hydrophobic), OC2 (hydrophilic), dust (five bins with effective diameters from 0.5 to 8 µm), sea salts (four bins with effective diameters from 0.1 to 7.5 µm) and sulfate. The characteristic conversion e-folding lifetime for BC from hydrophobic to hydrophilic, i.e. BC1 to BC2, is considered to be 2.5 days. More details about the treatment of BC in WRF-Chem can be found . The model simulates aerosol transport processes like emissions, advection, diffusion and deposition (dry and wet). The aerosol direct effects are modelled by coupling the aerosol scheme with the radiation scheme. The NCEP FNL (Final) Operational Global Analysis data (interpolated to model resolution) were used for setting up the initial and boundary conditions (updated every 6 h) for meteorological variables in the model, while the global chemistry transport model MOZART-4 was used for the creation of initial and boundary conditions for chemistry variables. The near-surface emissions of gas-phase and aerosol-phase species are formulated using the standard emission pre-processor software PREP-CHEM-SRC (version 1; ). The RETRO database is used for various greenhouse and precursor gases, EDGAR is used for emissions of CO, NO, NH3 and VOCs and the GOCART database is used for the emissions OC, BC and SO2 over the region. The necessary modifications in BC emissions from the GOCART database over this region have been incorporated. We conducted two sets of simulations for each of the ascents. The first set included only surface emissions (NoACEM is a control run of simulations without the prescription of aircraft emissions of BC) and the second set (ACEM, simulations with the prescription of aircraft emissions of BC) included surface and elevated emissions. The details of elevated emissions (BC emissions from aircraft) are discussed in Sect. .

To understand the role of BC emissions from aircraft in the formation of sharp layers of BC at elevated altitudes, we prescribed aircraft BC emissions from the MACCity global anthropogenic emissions inventory in the WRF-Chem simulations. The inventory provides aircraft BC emissions at 0.5∘ × 0.5∘ grids and 23 vertical levels from 0.305 to 13.725 km with a vertical resolution of 610 m. The vertical profiles of BC emissions from aircraft averaged over a 1∘ × 1∘ grid box centring Hyderabad (Fig. , red line) and over another grid box (Fig. , blue line) with the same dimensions just 2∘ south-west of the Hyderabad grid box are shown in Fig. . The vertical locations of the emission peaks are seen to vary depending on the proximity to an airport (Hyderabad in this case). The near-surface peak could be related to landing and takeoff (LTO) emissions over the airport, while the upper level peaks could be due to the emissions from planes passing over the location without landing.

The basic information this inventory considers includes global aircraft movements, performance characteristics of the different aircraft types, the actual paths of aircraft during their journey and the emission factor for BC. Such a global inventory for aircraft BC emissions inherently has several uncertainties, especially over a region like India, mainly due to uncertainties associated with (a) the amount of total aircraft fuel used across the country by domestic and international fleets, (b) the quality of aircraft fuel, (c) the age of aircraft and hence the degradation in performance, (d) the exact number of civil aircraft flying over the region, (e) the number of military aircraft unaccounted for and (f) the exact routes followed by the aircraft over this region. Additionally, a major source of uncertainty in the estimates of BC emissions from aircraft is due to uncertainties in the estimations of the emission factor (also known as emission index) for BC “EI(BC)” for the particular aircraft fuel being used. EI(BC) is the amount of BC emitted (g) per kg of aircraft fuel burnt and it depends on engine type, load conditions and engine conditions. The measurement of EI(BC) by several studies indicates that EI(BC) spans around 4 orders of magnitude. discuss and quantify the underestimation of EI(BC) observed in currently available aircraft BC emissions inventories. In this regard, a previous study has stated: “The First-Order Approximation (FOA3) – currently the standard approach used to estimate particulate matter emissions from aircraft – is compared to measurements and it is shown that there are discrepancies greater than an order of magnitude for 40 % of cases for both organic carbon and black carbon emissions indices”.

Thus, to assess this inventory over Hyderabad (our study region), we estimated BC emissions from aircraft over this region using previously reported values for the fuel efficiency of different airline carriers , seating capacity of the carriers (websites of different airlines), density of aviation fuel used in the Indian region, BC emission index for the aviation fuel and actual data on air traffic over Hyderabad obtained from air traffic control . Depending on such an evaluation of the MACCity emissions database over the Hyderabad region, we have modified the inventory emissions by a corresponding scaling factor. Additionally, by acknowledging the coarse resolution of the MACCity inventory, the “line-source” nature of the freshly emitted aircraft trail and the finer resolution of our model simulations, we confine the aircraft BC emissions to a width of 2 km and a height of 100 m. The mass conservation of the emitted BC due to such confinement leads to an additional scaling factor. The total scaling factor becomes the product of these two scaling factors. The modified emissions are formulated by multiplying the original emissions by the aforementioned scaling factors.

The atmospheric heating rates due to the vertical profiles of BC (measured and simulated) are computed using a discrete ordinate radiative transfer model (Santa Barbara DISORT Atmospheric Radiative Transfer; SBDART; ). SBDART solves radiative transfer equations considering a plane-parallel atmosphere. Aerosols are specified in the model through total aerosol optical depth, single scattering albedo (SSA) and the Legendre moments of the scattering phase function for the assumed aerosol mixture at every vertical layer. To compute the heating rate associated with the observed BC profiles, the required layer-wise total AODs were computed using vertical profile total extinction coefficients from CALIOP (Cloud-Aerosol Lidar with Orthogonal Polarization) on-board the CALIPSO (Cloud-Aerosol Lidar Pathfinder Satellite Observation) satellite. The level 2 extinction coefficient data from CALIPSO were cloud-screened using the standard recommended cloud-screening algorithm (https://www-calipso.larc.nasa.gov/resources/calipso_users_guide/tools/index.php). This makes use of flags like AVD (atmospheric volume description), CAD score (cloud aerosol detection flag), extinction coefficient uncertainty and extinction coefficient quality control to separate aerosol from clouds. The final vertical profile of the extinction coefficient was the mean of all such cloud-screened vertical profiles which fall within ±1.5∘ of the balloon flight location in the x–y direction and ±20 days of the balloon flight date. Such an averaging was done for a mean picture of the aerosol loading over the balloon flight region. To obtain the required SSA we made use of the observed vertical profile of BC and the OPAC (Optical Properties of Aerosol and Clouds) model . OPAC provides optical properties like the extinction coefficient, absorption coefficient, scattering coefficient, single scattering albedo, asymmetry parameter and phase function for the prescribed mixture of aerosol species. In OPAC, we prescribed the measured vertical distribution of BC to get the corresponding absorption coefficients. These absorption coefficients, along with extinction coefficients from CALIPSO, were used to obtain the SSA. Finally, in order to compute the Legendre moments of the scattering phase function, we execute OPAC with one of its preset aerosol mixtures, “urban”, as it is best suited for our observational location. This aerosol mixture has water soluble species (28 000 particles cm-3), insoluble species (1.5 particles cm-3) and soot (130 000 particles cm-3). For the urban aerosol mixture, we derived the scattering phase function and computed eight Legendre moments of the scattering phase function. Along with the aforementioned primary inputs, SBDART requires some additional inputs, such as the mean state of the atmospheric variables temperature and pressure on every vertical level and the vertical profile of specific humidity and ozone. These inputs were taken from the SBDART database under the tropical category, which suits our observation location (Hyderabad). Additionally, spectrally varying surface albedo values for Hyderabad were taken from the MODIS surface reflectance product (MOD09CMG). This daily level 3 product provides surface reflectance for seven bands (469, 555, 645, 858.8, 1240, 1640 and 2130 nm) at 0.05∘ of resolution. With all these inputs, the SBDART model was executed with eight streams in the radiative transfer calculations with 7.5∘ of resolution for solar zenith angles varying from 0 to 180. The net radiative fluxes were computed at the top and bottom of every vertical layer starting from the surface to 100 km. A similar run of SBDART was conducted without the prescription of aerosols. Using radiative fluxes from these two runs, radiative forcing within each layer was estimated using ΔF=FluxNA-FluxA, where FluxNA is radiative flux in the absence of aerosols (W m-2) and FluxA is radiative flux in the presence of aerosols (W m-2).

The change in radiative fluxes in each layer due to aerosol is the amount of energy absorbed in the layer due to aerosols. Corresponding atmospheric heating rates were computed using : dT/dt=(g/Cp)(ΔF/Δp), where dT/ dt is the heating rate (K s-1), g is the acceleration due to gravity, Cp is the specific heat capacity of air at constant pressure (Cp = 1006 J kg-1 k-1) and p is the atmospheric pressure . The heating rates (K day-1) were then averaged for 24 h.

The vertical extents of the heating rate profiles from the measurements were limited due to the availability of extinction coefficient data from CALIPSO over the region of interest. Such a limitation does not exist for model data, and hence we have computed the atmospheric heating rates using the model results. To compute the atmospheric heating rates using model data, we follow exactly the same aforementioned procedure. The only differences are that we use extinction coefficient data from model simulations instead of the CALIPSO satellite and SSA from model simulations instead of deriving it from the observed BC.

The vertical profiles of BC derived from flight measurements are shown in Fig. a–c, in which each profile is the average of the ascent and descent profiles for that flight. All the profiles revealed sharp and confined peaks in BC concentration in the free troposphere above a near-steady mass concentration within the planetary boundary layer (∼ 2 km for pre-monsoon and ∼ 1 km for winter). These peaks are identified on the respective profiles in Fig. a–c. It is also seen very clearly that the sharp and most prominent peak occurred between 4 and 5 km in the first and last profiles (Fig. a and c), which incidentally correspond to the pre-monsoon flights; it occurred at a slightly lower altitude (between 2 and 3 km) in the winter profile. The highest altitude layer seen in the profiles was in the vicinity of 7–9 km, though the amplitude of it is much smaller than the layer seen at around 4.5 km. With the help of temperature data from the collocated meteorological payloads, we evaluated the environmental lapse rate profile for each of these profiles and the mean profile for each flight is shown in Fig. d–f following the same order as in Fig. a–c. It is clearly seen that in the vicinity of the prominent BC peaks (P1 to P5 in Fig. a–c) there is a large reduction in the environmental lapse rate (sometimes close to zero), e.g. at 4.5 km during March 2010 (Fig. d), around 2–3 and 7 km during January 2011 (Fig. e) and around 4–5 and 6–7 km during April 2011 (Fig. f). The lapse rate profile during April 2011 (Fig. f) shows multiple spikes, possibly due to a higher number of peaks in the corresponding BC profile (Fig. c). Since the raw temperature data are very noisy, we have smoothened the data using two fixed points and two moving points (average filter). Such reductions in temperature lapse rate would affect the local stability scenario. While the occurrence of high BC below 3 km can be explained by the boundary layer dynamics, high BC values at higher levels, especially at 7–9 km, remained a mystery. hypothesised that the occurrence of such BC peaks at high altitudes could be associated with local sources of BC at those altitudes, as 4–5 and 8–9 km are the preferred corridors for aircraft flying over Hyderabad. With a motivation to test this hypothesis and also to estimate the climatic implications of such BC layers, we investigate the causes for the occurrence of such sharp and confined BC peaks at higher altitudes using a regional online chemistry transport model. Crucial findings from this study are reported and possible ramifications are discussed.

Meteorological processes like advection, diffusion and deposition play significant roles in controlling the concentration of pollutants near the surface and their vertical distribution in the troposphere . Hence, before inspecting the model-simulated vertical profile of BC, we first examine the performance of the model in simulating meteorology over the region of interest. We show such comparisons for the March 2010 NoACEM simulations as representative. The model-simulated meteorological parameters are averaged over the balloon flight region and are then compared with the corresponding observations from the balloon flight. Three-point running mean smoothing is used to smooth the observational and simulated vertical profiles. Firstly, we have compared the model-simulated vertical profile of horizontal wind speed (blue line, Fig. a) and direction (blue line, Fig. b) with the corresponding observations (red line, Fig. a and red line, Fig. b) available from the balloon flight measurements (deduced from the GPS data on-board the balloon). The figures depict relatively weak (less than 10 m s-1), low-level, northerly–north-easterly winds up to around 5 km. Beyond this altitude, the winds change direction gradually and become westerlies above 8 km. The wind speeds also increase drastically beyond 8 km and reach a value of 20–25 m s-1 at a height of 9 km and beyond. While the model-simulated vertical variation in wind speed (blue line, Fig. a) and direction (blue line, Fig. b) agree broadly with the measurements, they differ in details and magnitudes. In model simulations, the change in wind direction starts occurring at an altitude ∼ 1 km lower than in the observations. The large increase in wind speed beyond 8 km of altitude is satisfactorily captured by the model. Thus, with some disagreement in the actual magnitudes, the model captures the broad features of horizontal advection strength (wind speed) and nature (wind direction) over the balloon flight domain vis-à-vis the measurements. We next examine the model simulations of vertical stability over the flight domain by comparing the model-simulated vertical profiles of potential temperature (θ) with the corresponding profiles from the balloon data. The observed profile (red line, Fig. c) shows a stable layer (θ increases with height) up to the first 2 km of the lower troposphere. Above this layer, there is a thick, well-mixed layer up to a height of 4 km in which θ remains almost height invariant. Above this height, the atmosphere remains largely stable with an increase in θ with height. The modelled altitude variation in θ (blue line, Fig. c) generally matches the measurements, barring a few discrepancies such as the extent of the stable and well-mixed layer in the lower atmosphere and the magnitude of θ in the vicinity of the primary BC maxima in observations (around 4–4.5 km; Fig. a). The model captures the stable layer lying up to around 1.5 km, which is similar to that seen in observations (up to 2 km; red line, Fig. c). A convectively unstable, well-mixed layer (θ constant) extending up to a height of 3.5 km occurs above this layer, which is akin to observations (up to 4 km; red line, Fig. c), with an underestimation in θ values. Beyond this height, the model does not show any sign of instability, while θ increases with height in agreement with the observations but with differences in the magnitudes. Thus, the model-simulated vertical profile of θ also appears to be broadly comparable to the observed profile with some differences. Examining the vertical thermal structure, we have also compared the model simulations of the vertical profile of temperature with the corresponding measurements from the flight. While generally showing a reduction in temperature with height similar to the observations (red line, Fig. d), the model (blue line, Fig. d) shows discrepancies in actual magnitudes vis-à-vis the observations (-1.7 to +3.4 K), as in the case of the other meteorological variables. Possibly owing to the existence of the primary BC peak between 4 and 5 km, the observations (red line, Fig. d) depict a near-steady temperature within the particular altitude layer. The model (blue line, Fig. d), on the other hand, fails to capture this feature. Thus, along with some differences in the details, the model simulations capture some of the large-scale features of the meteorology over the balloon flight region vis-à-vis the observations.

With the above broad agreement in the meteorological fields, we proceeded to examine and evaluate the model-simulated vertical profile of BC in the vicinity of the balloon flight domain (blue line, Fig. g–i) in the NoACEM configuration. For 17 March 2010, the model simulations (NoACEM) show considerable differences vis-à-vis the observed vertical profile of BC (Fig. a). The magnitudes of the simulated BC (blue line, Fig. g) show a relatively good comparison with the observations only over the lowermost altitudes, i.e. below 3.5 km. Above that altitude, the two profiles completely differ from each other; while the observations (Fig. a) show an increase in BC followed by a sharp peak at 4.5 km (BC > 12 µg m-3), the model (Fig. g) displays a rapid decrease in BC concentration with altitude, with no sign of the elevated (high-altitude) layers of high BC concentration seen in the measurements. Thus, though the model satisfactorily simulates the meteorology over the flight domain, it does not simulate the observed vertical profile of BC. The results for the other two flights are shown in Fig. h and i. During January 2011, the model simulations (NoACEM; blue line, Fig. h) show a rapid reduction in BC within the first 1 km. There is no sign of sharp and confined peaks beyond this height, unlike the corresponding observations (Fig. b). During April 2011, the model simulations (NoACEM; blue line, Fig. i) show a sharp reduction in BC within the first 1 km. Beyond this height, a sharp peak in BC is seen with a maximum near 2 km. A relatively gradual reduction in BC occurs beyond this height up to 4 km, followed by a flat steady BC profile with very low values above that height (blue line, Fig. i). Thus, none of the model simulations capture the sharp confined BC peaks (or the elevated BC layers) occurring at higher altitudes. This strongly suggests that the meteorological factors alone cannot be responsible for the existence of the elevated BC layers. To shed more light on this, we considered other hypotheses.

The meteorology (observed and simulated) being benign for all the cases and the possibility of any long-range transport of BC from other locations being an unlikely cause for such high concentrations , one has to look for a local injection of BC aerosols in the middle and upper troposphere because surface-based emissions would not lead to elevated BC layers at altitudes of 4–5 or 7–9 km. In this context, emissions from commercial air traffic (that overfly Hyderabad and land and take off from there) assume significance. In their first reporting of the elevated layers, hypothesised the role of such emissions and obtained an approximate estimate of air traffic over the study location; about 200 aircraft overfly Hyderabad in the corridor at 8–10 km and another 250–300 use the corridor at 4–5 km in the course of landing and taking off from the airport. We next evaluate elevated emissions due to aircraft as a likely candidate for these high concentrations.

We examine the outcome of the prescription of BC emissions from aircraft on the modelled vertical profile of BC in Fig. g–i in the vicinity of the balloon flight domain for the 3 balloon flight days. It is quite interesting to note that, upon the prescription of BC emissions from aircraft, the model simulations (ACEM) show sharp layers in the vertical profiles akin to the observed BC profiles (red line, Fig. g–i), which are not simulated otherwise (NoACEM control run). Though the actual altitudes and the magnitudes of the BC layers differ in comparison with the observations, the sharpness of the modelled BC layers makes them look similar to the observed BC layers. The two peaks in the BC profile during March (red line, Fig. g), the lower and upper level BC peaks during January (red line, Fig. h) and the clustered BC peaks during April (red line, Fig. i) are well simulated by the model only after the prescription of BC emissions from aircraft. Thus, even for the model with realistic meteorology, the high-altitude BC peaks and layers are captured only when the high-altitude sources of BC are prescribed. This clearly highlights the role played by aircraft emissions of BC in the creation of the high-altitude BC peaks, which remained as a hypothesis earlier. Additionally, as a consequence of such high-altitude BC emissions, the modelled vertical profile of temperature lapse rate over the balloon flight region shows a reduction in magnitude in the ACEM case compared to the NoACEM case, indicating warming due to BC (Fig. a). The corresponding differences in the temperature lapse rate (d(dT/dz) = (dT/dz)ACEM-(dT/dz)NoACEM) appear to be higher, especially at higher altitudes of 4 km and beyond 7 km (Fig. b). Such a reduction in the magnitude of the temperature lapse rate values results in a better match between the model simulations and the corresponding observations of temperature lapse rate, especially at higher altitudes (from 6 to 8 km and from 9 to 9.5 km; Fig. a).

We carried out one more model simulation, in which we prescribed the emissions of BC from biomass burning activities using the Fire INventory from NCAR (FINN) version 1.5 biomass burning data. The FINN provides high-resolution global emission estimates from such open burning activities. The temporal resolution of the inventory is 1 day, while the spatial resolution is 1 km2. In our simulations, we allowed such emissions from biomass burning to lift vertically following the online plume-rise module . However, the model could not simulate the observed sharp and confined BC peaks (figure not shown) when we switched off the prescription of BC emissions from aircraft. This clearly shows the important role of aircraft BC emissions in causing high-altitude BC peaks.

To test the robustness of the simulated high-altitude BC peaks, we have also examined the impact of near-surface emissions of BC on the elevated layers. For this, we have done a similar simulation with aircraft BC emissions, in which we turned off the near-surface fossil fuel emissions of BC over the model domain. We then examined the effect of this on the simulated vertical profile of BC in Fig. . It can be clearly seen from this figure that the vertical profiles simulated by the two runs (SE, the run with prescribed near-surface anthropogenic emissions of BC, and N-SE, the run with near-surface anthropogenic BC emissions turned off) differ only in the lower altitude region up to a height of about 4 km (probably the altitude up to which BC could be lofted by the convection). Beyond 4 km, the profiles are largely similar, implying that the elevated BC layers are insensitive to surface BC emissions. The correlation coefficient between the two BC profiles beyond 4 km comes out to be 0.97, which is 99.99 % significant. Moreover, the magnitudes of BC in the two profiles show good agreement with a difference limited to only 0.1 µg m-3. Thus, our model simulations indicate that the high-altitude peaks in BC are not a result of the convective lifting of near-surface BC, but they are indeed caused by BC emissions (injection of BC) at higher altitudes by aircraft.

We noticed that the model, WRF-Chem, produces the elevated sharp peaks of BC akin to the observations upon the prescription of aircraft BC emissions from the regionally fine-tuned MACCity inventory. We now proceed to examine the seasonal behaviour of such elevated sharp BC layers within the model simulations. For this, we carried out 1-month model simulations during each season of the year 2010, i.e. January, March, July and October, which are representative of winter and the pre-monsoon, monsoon and post-monsoon seasons. Keeping in mind the limited width and the dynamic behaviour of the horizontal location of the aircraft-emitted trail, we have computed the seasonal probability of the occurrence of a high-altitude BC peak at every point within the domain instead of plotting a monthly mean vertical profile of BC averaged over a region.

As a strong absorber of radiation over a wide spectral band, the elevated BC layers would absorb incoming solar radiation, which would then heat up the atmosphere locally and alter the vertical stability. In this section, we compute the atmospheric heating caused by the observed and modelled BC vertical profiles. The methodology that we follow to compute the atmospheric heating rates due to BC is similar to that followed by and .

To examine the convective lifting of high-altitude BC, we have carried out WRF-Chem simulations with and without the prescription of aircraft BC emissions (ACEM and NoACEM respectively) over the Indian region for a period of 1 week covering strong convective activity over the model domain during July 2010. July was selected for this purpose in view of the known prevalence of deep convection over the Indian region associated with the summer monsoon and the TTL being thinnest at this time of the year . The animation (Mov. 1 in the Supplement) shows height–longitude plots of the maximum value of BC for a latitude belt of 2.25∘ over Hyderabad (the X axis is longitude, and the Y axis is height in metres) in the ACEM case. The animation clearly reveals the capability of strong convection to lift BC to heights beyond 14 km. On a few occasions, these layers are seen to be transported even beyond 17 km across the tropopause. Similarly, we compute the difference in the upper tropospheric BC mass concentration values in ACEM simulations vis-à-vis the NoACEM case (ΔBC = BCACEM- BCNoACEM) averaged for the entire duration of the model simulations carried out during July 2010. The maximum of such time-averaged ΔBC values across the latitude belt of 2.25∘ centred over Hyderabad for every longitude of the model domain is shown in Fig. . The time-averaged increment in UTLS (upper troposphere and lower stratosphere) BC load due to aircraft BC emissions is seen to occur over the entire longitude belt, though it is more pronounced over a few longitude bands (79–80∘ E). Higher ΔBC values extend vertically even beyond 17.6 km, highlighting the mean vertical transport of aircraft-emitted BC in the UTLS region. This provides strong evidence for the intrusion of high-altitude aircraft-emitted BC into the upper tropospheric and lower stratospheric heights over the Indian region during strong convective periods. The intrusion of such tropospheric air into the stratosphere will be favoured under certain conditions like a thinner tropopause layer or unstable conditions within the tropopause layer. In the next section, we show observations supporting such intrusion.

There are several observational reports of the presence of BC at upper tropospheric and stratospheric altitudes (). One way to ascertain the occurrence of such layers in the stratosphere is to examine the CALIOP Lidar (on-board the CALIPSO satellite) extinction coefficient data (at 550 nm). We have examined all such vertical profiles for 3 consecutive years (from January 2010 to December 2012) over five specifically chosen regions (5∘ × 5∘) over and around India (Fig. ). Since we want to focus on the presence of aerosols at stratospheric altitudes, we have plotted the vertical profiles of extinction coefficients in the stratospheric altitudes (i.e. from altitudes of 20 to ∼ 30 km). The results are shown in Fig. .

To our surprise, we notice that the aerosol extinction coefficient values over the stratospheric altitudes are more than 0.02 km-1 over all these regions, and occasionally values greater than 0.06 km-1 are also noticeable (Fig. a). The corresponding appended time series of stratospheric (i.e. from altitudes of 20 to ∼ 30 km) aerosol optical depth over the regional boxes under consideration from 2010 to 2012 is plotted in Fig. b. The AODs are seen to be higher than 0.01 on most of the occasions with frequent spikes reaching as high as 0.16. We considered different scenarios to explain such high AOD. These included (i) volcanic perturbation to stratospheric AOD and (ii) high-altitude cirrus clouds at the TTL region. We first compared these extinction coefficient and AOD values with the corresponding area-averaged values for the entire tropical belt (20∘ S to 20∘ N; ). The maximum value of stratospheric AOD averaged for the entire tropical belt for the period 2010 to 2012 (which is considered to be the background tropical stratospheric AOD in this study) is marked with the dotted red line in Fig. b. It can be seen that the stratospheric AODs from our analysis over the five regional boxes are roughly 4–6 times higher than the background tropical stratospheric AOD (dotted red line, Fig. b); the same is the case for aerosol extinction coefficient values as well (comparison not shown). The large differences in the background tropical stratospheric AOD and the stratospheric AODs over the Indian region signify an enhancement in the stratospheric aerosol burden over the regions under consideration above the background stratospheric aerosol loading. One of the major controllers of stratospheric AOD is the influx of particulate, namely sulfates, organics and ash, and gases like SO2 from volcanic eruptions into the stratosphere . An examination of the occurrence of global volcanic eruptions from 2008 to 2012 suggests that of the 11 major volcanic eruptions that occurred during that period, 7 were centred away from the tropics. Also, the flux of SO2 in the stratosphere from the most intense of the 11 events was merely one-tenth that of the Mount Pinatubo eruption . This suggests that the enhancement in AOD and the extinction coefficient over the regions under consideration in this study may not be linked to volcanic eruptions. The monsoon season (June to September; denoted by the letter M in Fig. ) appears to be the favourite period for the lifted layers in the stratosphere (Fig. a). This looks to be linked to the vertical lifting associated with severe convection (as seen in our model simulations; Fig.  and Mov. 1 in the Supplement) and thin TTL as explained earlier . Additionally, these layers seem to be present during the other seasons (pre-monsoon and winter) as well. The Indian region experiences a large-scale ascending motion above 10 km during winter months . Such an ascending motion can cause the intrusion of tropospheric air (lying at 8–10 km) into the stratosphere . Thus, while the monsoonal convection and the related thermodynamics appear to be a cause of the summertime higher extinction coefficient values at stratospheric heights over the Indian region, the transport of tropospheric air mass into the stratosphere due to the large-scale ascent beyond 10 km looks to be responsible for the wintertime high values.

To shed light on the aerosol species responsible for such extinction coefficient values, we examined the particle depolarisation ratios (at 550 nm; henceforth PDRs) at these altitudes (Fig. c). The PDR is the ratio of the attenuated backscatter coefficient in the perpendicular polarisation to that in the parallel polarisation. We computed the PDR values from the corresponding backscatter coefficient values. The PDR values are seen to be higher than 0.3 on most occasions. This suggests the presence of non-spherical aerosol species at those altitudes. One of the major non-spherical aerosol species over this region is mineral dust. The source of mineral dust aerosol species in the stratosphere could be largely related to (a) volcanic eruptions, (b) meteoritic debris and (c) the convective transport of tropospheric dust. As mentioned previously, our study has been carried out for a period which is relatively volcanically quiescent. Thus, volcanically erupted dust may not have contributed much to the stratospheric aerosol burden during our study period. Meteoritic debris forms a minor part (5–10 %) of the stratospheric aerosol composition, especially at altitudes lower than 30 km , making it difficult to capture with a lidar. Moreover, meteoritic debris is reported to be coarse in size (i.e. having a radius larger than 1 µm; ) and hence would have large deposition rates and be largely spherical in shape with lower values of particle depolarisation ratio (PDR less than 0.1; ). The extinction coefficients associated with plumes of meteoritic debris are 3 orders of magnitude less than the values we notice over our study domain. These points together rule out the possibility of associating the observed values of extinction coefficients and PDRs with meteoritic dust. The convectively lifted dust and other air pollutants can move into the higher altitude regime , but their transport over the Indian region is limited to around 20 km , which is mainly governed by the height of convective towers . Moreover, dust aerosol is 2.6 times heavier than BC and also less solar absorbent; thus it is unlikely to be vertically lifted beyond 20 km by large-scale or self-lifting mechanisms . Hence we eliminate the possibility of tropospheric dust existing beyond 22 km of altitude over our domain. Additionally, the maximum cloud top altitude occurring over the Indian region during the monsoon season is 20 km . This suggests that the non-spherical ice crystals emanating from the anvils of the convective towers over the Indian region can be found only around 20 km and not above 21–22 km. This eliminates the possibility of the presence of non-spherical mineral dust particles and ice crystals at these heights. A few previous studies have noticed the presence of BC chain agglomerates at the upper tropospheric and lower stratospheric heights . Such transport of BC to the stratospheric heights could be the result of convective lifting (as seen in our model simulations; Fig.  and Mov. 1) and the self-lifting associated with it. Though pure and nascent BC has a relatively lower value of PDR, aged BC particles are porous and capable of forming long chain agglomerates that are non-spherical and would depict higher values of PDR. Such BC agglomerates, even when mixed with stratospheric sulfate, could still give rise to non-spherical shapes and thus higher values of PDR. This indicates the possible existence of BC at the stratospheric altitudes over the Indian region. Thus, the model-simulated convective transport of high-altitude aircraft-emitted BC layers to the UTLS region combined with favourable conditions for cross-tropopause transport of air mass over the Indian region during monsoonal months, large-scale wintertime ascending motion above 10 km over the Indian region and the presence of local BC emission sources (aircraft) at high altitudes (∼ 10 km) throughout the year together indicate the possibility of aircraft emissions being primarily responsible for the possible presence of BC in the stratosphere over the Indian region.