Here we focus on CO versus O3 and HCl versus O3 (Fig. 8) as well as on NOx versus O3 and NOx versus NOy (Fig. 9).

O3 mixing ratios in the TL exhibit a strong vertical gradient, increasing from the troposphere to the stratosphere. Given a chemical lifetime of the order of weeks, this reflects the degree of mixing between O3-poor UT air and O3-rich air from the lowermost stratosphere (Sprung and Zahn, 2010). That general picture might not hold in the ASMA though, where – depending on the availability of precursors – enhanced photochemical O3 production is superimposed on isentropic in-mixing from the stratosphere.

Enhanced CO is a tracer of boundary layer pollution and an O3 precursor in the troposphere. Oxidation with the hydroxyl radical (OH) prevails under stratospheric conditions, and CO mixing ratios decrease by about an order of magnitude across the tropopause (Hoor et al., 2002).

As a tracer for stratospheric air we use HCl (Marcy et al., 2004; Park et al., 2008), which in the UT has no significant photochemical sources but has a lifetime similar to O3. Wet scavenging in clouds effectively prevents convective transport of HCl to the UT, and no injections of HCl from volcanic activity affected the ESMVal flight from Male to Larnaca. Together this makes HCl a viable tracer of stratospheric O3 entrainments, until it is selectively removed by wet scavenging.

NOx (NO + NO2) is an O3 precursor and part of NOy. NOx primarily characterises fresh emissions. Only NO was measured during the HALO ESMVal campaign, but daytime NO is a good proxy for NOx in the UT (Seinfeld and Pandis, 2006). In the UTLS, enhanced NOy originates both from tropospheric and stratospheric sources. Thus it is not a viable tracer for stratospheric air on its own. In the lower troposphere odd nitrogen species are co-emitted with carbon monoxide in combustion processes. Lightning and aircraft emissions are sources in the UT. Photolysis of N2O peaks at about 30 km and is the principal source of NOx in the stratosphere, resulting in increasing mixing ratios of NOx and NOy above the TP (Seinfeld and Pandis, 2006).

In order to place the observed tracer–tracer relations into context, we plot the measured samples together with grid-cell samples from the EMAC simulation. Simulation output along the flight track is too sparse for a meaningful comparison (10 s resolution of measurements versus 12 min for the simulation). Therefore 5000 simulated samples per panel are chosen randomly, from the entire month of September 2012 and from throughout the ASMA region (Fig. 2: Tibetan and Iranian parts). Plotting all corresponding samples from the EMAC simulation would impair the visibility of clustering. Two different vertical ranges are chosen. The range from 50 above the TP to 100 hPa below it in the actual EMAC tropopause (Fig. 8a, d) provides a zoomed-out view of possible tropospheric and stratospheric tracer mixing ratios and tracer–tracer relations for the time of year and region of the measurements. Zooming-in to the altitude range of measurements, we choose tropospheric tracers from the pressure altitude range 200 to 100 hPa (Fig. 8b, e). The observations from the entire flight without ascent and descent are shown in (Fig. 8c, f).

Measurements south of the ASMA are marked by dark-blue dots in Fig. 8c, f and are clearly distinct from the measurements in the ASMA filament (orange boxes). The ranges covered by the measurements are also given in the corresponding panels with simulated data, but are adjusted for model biases there. Those biases were estimated according to visual comparisons of measured versus simulated trace gas mixing ratios along the flight track in the ASMA filament (shown in the accompanying paper). All measured ranges fit into the simulated monthly averages for September 2012 in the ASMA region; thus the simulation captures this aspect well and the measurements are unlikely to represent an exceptional situation. We also note that all measurements clearly fall into the tropospheric regions of the respective simulated tracer–tracer spaces. This is no surprise: all HALO ESMVal measurements considered here were taken well within the troposphere.

Simulated O3 and NOx increase in the stratosphere with a higher O3 / NOx ratio than in the troposphere (Fig. 9a). However, since the range covered by the HALO ESMVal measurements is just at the intersection of the stratospheric and tropospheric branches (orange box in Fig. 9a), this does not help to distinguish the stratospheric influence from tropospheric in situ production.

The colour code of the observations (Fig. 8c, f) corresponds to measurement time. Similar colours indicate spatial and temporal proximity, a prerequisite for mixing lines. Schematic lines L1–L5 (Fig. 8b, e) and their parallels indicate special, hypothetical cases for the evolution of air masses, which are discussed next.

Steep vertical gradients across the tropopause dominate O3 profiles in the monsoon regions (Fig. 5a, b), but the profiles also show temporal fluctuations on various timescales. Note that our lateral averaging regions are rather large, and smaller-scale structures get smoothed out, e.g. when an O3-poor interior is combined with an O3-rich fringe.

There is increased influx from the stratosphere in spring, enhancing O3 in the UT. This is in accordance with the study by Cristofanelli et al. (2010), but in contrast to their study there are non-negligible O3 enhancements connected to the stratosphere during the monsoon season (Fig. 5b, circled). In this respect our simulation is, however, consistent with trace gas budget considerations for the ASMA (Barret et al., 2016) and the TTL (Konopka et al., 2010). Entrainment from the TL rather than deep from within the stratosphere could reconcile the different findings. No other stratospheric contributions were found for the stratospherically influenced trace gas signatures in the HALO ESMVal ASMA observations.

Enhanced O3 is recognisable in the averaged profiles of the eastern ASMA part only from the tropopause to about 200 hPa below the tropopause, while the mid-troposphere in the Tibetan part is dominated by particularly O3-poor air during the monsoon season (Fig. 5b). The latter is consistent with the findings of Safieddine et al. (2016). O3 depletion in the mid-troposphere of the eastern part is contrasted by enhanced O3 in the mid-troposphere during the monsoon season in the western part (Fig. 5a, circled), marking the well-known summertime O3 maximum there.

This is reflected in the evolution of CO profiles in the ASMA region in 2012 (Fig. 5c, d). CO-poor air dominates in the UT during spring, consistent with the stratospheric influx indicated by O3.

CO-rich air rises throughout the troposphere of the eastern part during the monsoon season (Fig. 5d, circled). On the western side there is a conspicuous CO-depleted zone 300 to 500 hPa below the tropopause during the monsoon season (Fig. 5c, circled), while CO is episodically enhanced in the UT. Uplifted air with enhanced CO mixing ratios hardly reaches higher than 450 hPa below the tropopause in summer.

This difference between CO profiles in the Tibetan and the Iranian parts is consistent with the findings of Pan et al. (2016). Occasional horizontal transport in the UT from the eastern to the western part of the ASMA is a possible explanation for the spatio-temporal evolution of CO mixing ratios, indicating that trace gas signatures in the Iranian part are dominated by the UT outflow of the Tibetan part of the ASMA.

As expected for a stratospheric tracer, the simulated HCl profiles (Fig. 5e, f) show a strong anti-correlation with CO in the UT, with increased HCl in times of stratospheric influx (e.g. Fig. 5f, blue circle) and decreased HCl in the monsoon season. Stratosphere-to-troposphere exchange is pronounced during spring, consistent with the seasonality of the Brewer–Dobson circulation (Holton et al., 1995).

HCl plumes in the Iranian part rise to about 400 hPa below the tropopause in summer (Fig. 5e, circled), just like CO. Predominantly dry conditions in the western part prevent HCl from being washed out. However, CO and HCl are temporally anti-correlated in the mid-troposphere. Since HCl is emitted by the sea, we attribute this to alternating marine and continental origins in the uplifted air (Fig. 5c, e).

Some HCl descends from the tropopause into the mid-troposphere, as indicated by tilted patterns of enhanced HCl, which start at the tropopause and propagate downward (marked by an arrow in Fig. 5e). Similar telltale signs of descent is also recognisable in other species in the Iranian part during summer.

There is almost no HCl in the Tibetan part throughout the monsoon season (Fig. 5f, black circle), except for the UT. Convection and thunderstorms are frequent during the monsoon season in South Asia (Fig. 7d). Wet scavenging and subsequent washing out by precipitation does not affect CO, but effectively prevents transport of HCl to higher altitudes. Predominantly continental origins also contribute to an increased CO / HCl ratio in the rising plumes of the eastern part.

Simulated NOx and NOy profiles in the ASMA region from April to September differ to the rest of the year (Fig. 6a–d). E- and C-shaped NOy profiles dominate the Tibetan and Iranian parts, respectively (Fig. 6c, d; examples of individual profiles in the Supplement).

NOx and NOy from boundary layer sources are uplifted, accompanied by conversion of some NOx into NOy. Solvable NOy components (e.g. HNO3) become increasingly washed out (Fig. 6d), leading to a minimum of reactive nitrogen in the mid-troposphere in the Tibetan part in summer (Fig. 6b, d). Uplift from the lower troposphere is less important for UT NOx and NOy than LiNOx and stratospheric entrainments. The dependence of detrainment on altitude hardly affects trace gas profiles in the mid-troposphere, as indicated by the corresponding CO profiles (Fig. 5d, circled).

Both NOx and NOy gradually increase above the tropopause due to stratospheric photochemistry. Stratospheric influx contributes to increased NOx mixing ratios in the UT in spring (blue circles in Figs. 5f, 6b), but enhanced UT NOx during the monsoon (Fig. 6b, black circle) is more due to lightning NOx emissions in the Tibetan part than to stratospheric entrainments (Supplement).

NOy can rise to about 400 hPa below the tropopause in the heat low over the Arabian Peninsula (circled in Fig. 6c). Downward transport (as indicated by anti-clockwise tilted signals, one example marked by an arrow in Fig. 6c) dominates above that altitude, preventing further uplift. With little in situ production of lightning NOx over the Arabian Peninsula in summer (Figs. 7c and 10), UT NOy in the Iranian part is dominated by the outflow of the Tibetan part.

As a combination of the different processes affecting NOx and NOy, the NOx / NOy ratio maintains a broad maximum in the TL throughout the year (Fig. 6e, f). During the monsoon, the NOx / NOy ratio in the UTLS is larger in the western than in the eastern ASMA part (Fig. 6e, f, circles). This indicates preferential export of high-NOx air from the Tibetan part or is an artefact of the possible dominance of a single source of LiNOx in the Iranian region (Fig. 10).

In our EMAC simulations, LiNOx is released based on a parameterisation that links flash frequency to updraught velocity in – also parameterised – convection. It is difficult to pinpoint LiNOx emissions (Schumann and Huntrieser, 2007), and both parameterisations are a notorious source of uncertainty in global models. The parameterisations for convection (Tiedtke, 1989; Nordeng, 1994; Tost, 2006) and lightning (Grewe et al., 2001) used in our simulations have been tested in several studies (Tost et al., 2007; Grewe, 2009; Lopez, 2016) and appear to be state of the art. EMAC-simulated lightning activity matches the corresponding TRMM-LIS/OTD (Tropical Rainfall Measuring Mission Lightning Imaging Sensor/Optical Transient Detector) observations (Cecil, 2006) reasonably well, spatially and temporally (Supplement). Simulated and observed NO along the HALO ESMVal flight track agree remarkably well within the ASMA region (accompanying paper). The same is true for comparisons of IAGOS-CARIBIC measurements of NO and our simulation's output along the IAGOS-CARIBIC flight tracks, where the agreement is particularly noticeable for the monsoon season in the ASMA region (Supplement). A dedicated comparison to other models is still desirable, since NOx is complex to simulate and there might be compensating errors. For the current study, however, the above comparisons provide some confidence that LiNOx emissions have been captured well by the simulation.

Simulated LiNOx emission rate profiles for 2012 show prominent maxima for the eastern and western ASMA regions during spring (Fig. 7c, d). Overall, LiNOx emissions are much stronger in the Tibetan part. The emissions reach up to the tropopause throughout the year, implying that LiNOx is emitted at higher potential temperatures during the monsoon season (Fig. 4b). Despite higher emission rates in the laterally averaged profiles (Fig. 4d), lightning activity in the Tibetan part is more sporadic and localised in spring than in summer (Fig. 10d, e). During the monsoon season, LiNOx is constantly replenished in the ASMA throughout the region (Fig. 10g, h, i). Sensitivity simulations show that UT NOx (Fig. 6b) is mainly LiNOx during the monsoon season (Supplement).

The uplift of lower-tropospheric air to the UT is a well-known characteristic of the ASMA (Pan et al., 2016). Simulated CO profiles in the Tibetan region show episodes of such uplift, not only for 2012 (Fig. 5d), but for every monsoon season during 2010–2014 (Supplement). This is consistent with the HALO ESMVal measurements, since the trace gas gradients observed in the ASMA can be explained by mixing between lower-tropospheric air and stratospherically influenced air.

CO uplift to the UT and LiNOx emissions are both related to convection and it is remarkable that there is a much stronger correlation between the two in summer than in spring (Figs. 5d, 7d). We attribute this to three effects: (i) the large-scale uplift at the south-western flank of the Himalayas is only active during the monsoon season and is not exclusively driven by deep convection (see Sects. 1, 3). It has been suggested to be the main transport pathway of pollutants to the UT (Bergman et al., 2013; Pan et al., 2016), which is supported by our simulation (Fig. 11) and the location of maximum moist static energy (Boos and Hurley, 2013). Back-trajectory calculations in the accompanying paper identified this pathway as the source of enhanced CO in some of the HALO ESMVal measurements, despite the underlying reanalysis not accounting for (small scale) convection. (ii) In the UT the ASMA is a leaky transport barrier, allowing some accumulation of the uplifted pollutants (Pan et al., 2016). There is no such transport barrier in spring. (iii) The spatial and temporal matches of deep convection and increased CO at different altitudes reflect the potential for entrainment and subsequent convective transport of CO (Fig. 7g, h). It is clearly increased in summer. More detailed analyses (Supplement) show that convection is localised over the coastal regions of West Bengal and Bangladesh in April 2012. In contrast, during August 2012 convection is ubiquitous throughout the Tibetan region. It is most persistent at the south-western flank of the Himalayas and over the Tibetan plateau. This coincides with the highest CO mixing ratios, which accumulate there due to the prevalent south-westerly winds during the monsoon season. Consequently, more CO is transported through the troposphere in the Tibetan region during summer.

The net photochemical O3 production rate (Fig. 7a, b) is derived from the difference between EMAC-simulated diagnostic tracers ProdO3 and LossO3 (Jöckel et al., 2016). Here we take into account effective O3 production and loss terms following Crutzen and Schmailzl, (1983) and extended by Grewe et al. (2017; see their Supplement). There are known high-O3 biases in the simulation (Jöckel et al., 2016), and uncertainties in the chemical mechanism (Gottschaldt et al., 2013) also impose uncertainties onto O3 photochemistry. Nevertheless our simulated net O3 production rates in the ASMA (Fig. 7b) agree remarkably well with the independent estimate by Barret et al. (2016).

O3 photochemistry is dominated by catalytic cycles in the troposphere and affected by a variety of parameters, e.g. ambient mixing ratios of H2O, O3, CO and NOx (Ehhalt and Rohrer, 1994; Grooß et al., 1998; Jaeglé et al., 1998; Seinfeld and Pandis, 1998). We focus on NOx and CO for illustration (Fig. 12). Photochemical O3 production

Net O3 production in the UT is determined by ProdO3 rather than by LossO3 (Supplement, Figs. S6, S9, S17), so it is sufficient to analyse ProdO3 in this context.

The simulation shows the superposition of the above effects, among others within the full complexity of the chemical mechanism. As a result, a net O3-producing photochemistry prevails in the ASMA throughout the monsoon season (circled in Fig. 7a, b). This is accompanied by net O3 destruction during the monsoon season, 300 hPa below the tropopause and lower. At the tropopause and slightly above, there is a local minimum of net O3 production, followed by increased net O3 production in the stratosphere. Net O3 production is at maximum in the altitude range, where uplifted young air (enriched in CO and co-emitted volatile, organic O3 precursors) mixes with NOx-rich UT air (Figs. 5d, 6b, 7b).

ProdO3 per NOx shows a strong gradient at the altitude of maximum net O3 production (circled in Fig. 7b, f). This indicates the transition from the NOx-limited to the NOx-saturated regime (Fig. 12), but is superimposed by gradients of CO and other O3 precursors. The maximum corresponds to about 300 pmol mol-1 NOx (circled in Fig. 8h), and variations of NOx in that region have a relatively small effect on net O3 production (2 in Fig. 12).

Going down from the TP in the ASMA, NOx and O3 generally decrease, while CO and H2O increase (Figs. 5, 6; H2O not shown). It is a multi-dimensional problem. CO (among others) determines the curve in the NOx vs. ProdO3 diagram, and NOx determines the operating point on the curve. Considering typical ranges of CO and NOx in different parts of the ASMA, an area on the surface in ProdO3–NOx–CO space is termed “operating mode of the chemical system” in the following. Different chemical regimes (e.g. NOx-limited or NOx-saturated) are allowed within one operating mode. The non-linear dependence of ProdO3 on ambient trace gas mixing ratios leads to the simulated maximum within the opposite gradients of those trace gases (NOx decreases below the TP, while CO increases) in the UT ASMA. Going down from the TP, the chemical system goes from operating mode 3 to 2 to 1 (Fig. 12). In principle, all those operating modes could be in the NOx-limited regime and still lead to a maximum of net O3 production in the UT. However, our simulations also show the NOx-saturated regime (Supplement).

Periods of enhanced lightning NOx in spring (circled in Fig. 7d) correspond to increased net O3 production (Fig. 7b), but despite higher LiNOx emissions in spring, net O3 production is at maximum in the ASMA. The main differences between the seasons are that there is less CO in the UT during spring (Fig. 5d) and lightning NOx is available only locally (Sect. 6.1. and Supplement). More CO in the UTLS in summer increases ProdO3 and the maximum possible O3 production (Fig. 12). Too high NOx in spring does not help ProdO3 – or even pushes the system into the NOx-saturated regime (3 in Fig. 12). NOx close to maximum ProdO3 conditions (2 in Fig. 12) throughout the region in summer leads to higher ProdO3 in the lateral average.

Both NOx and other precursors are more abundant in the UT of the Tibetan part, resulting in larger photochemical O3 production than in the Iranian part (Fig. 7a, b). This is consistent with other studies (J. J. Liu et al., 2009; Barret et al., 2016), which have also found such an asymmetry. O3-depleting conditions prevail in the mid-troposphere over the Arabian Peninsula throughout the summer (Fig. 7a, circled). Thus increased O3 there (Fig. 5a) must be due to transport.

Confinement in the ASMA circulation allows the mixed air to age, i.e. to produce O3. CO is being depleted in the process of ageing, and NOx is transferred to NOy. The other source of aged air is entrainment from the TL, which is, however, enriched in HCl (at least within the photochemical lifetime of HCl). O3 is produced in the ASMA at a net rate of about 4 nmol mol-1 day-1 (Fig. 7b), and simulated O3 mixing ratios in the UT of the ASMA region vary by about 120 nmol mol-1 (Fig. 8b). It would take 30 days to cover that range by photochemical O3 production alone. The observed values in the ASMA filament cover a range of about 48 nmol mol-1 (Fig. 8c), corresponding to 12 days of photochemical O3 production. According to the trajectory calculations in the accompanying paper, this is about the time needed to circle the ASMA. The above O3 variability of course includes different amounts of O3 from the TL, and O3 productivity varies too. Neglecting these uncertainties, it takes one or two rotations of the ASMA to transform O3-depleted, freshly uplifted air into aged, O3-enhanced air. O3-depletion in the ASMA relative to the regional average can only be maintained by frequent replenishment of young air.

Here we focus on the less well-known entrainment of stratospheric or TL air into the free troposphere, which is supported by the unique thermodynamic conditions over the Tibetan plateau in summer (Fig. 2). The prevailing northerly winds (Kunze et al., 2010) of the eastern ASMA flank (Fig. 2) tend to transport high potential vorticity (stratospheric or TL, Fig. 4c) air along the isentropic surfaces into the troposphere (Ren et al., 2014; Kunz et al., 2015). This effect was also detected by Konopka et al. (2010) as enhanced horizontal transport of O3-rich air from the extratropics into the TTL. This transport from the TL or even the extratropical lower stratosphere into the free tropical troposphere may not leave a telltale sign of increased potential temperature in the corresponding air masses in the tropics. This includes the 350–370 K isentropes that were encountered during the HALO ESMVal campaign in the tropics (Fig. 4; θ estimates for the measurements are shown in the Supplement).

Additional stratospheric or TL contributions at the outer ASMA edge, other than at the eastern flank, are also plausible. The eastern Mediterranean and central Asian region is a global hotspot of tropopause-folding activity (Tyrlis et al., 2014), which is related to ASMA dynamics and generates enhanced O3 levels through stratosphere–troposphere exchange (Akritidis et al., 2016). If the ASMA circulation encompasses that tropopause folding hotspot, it may pick up stratospheric entrainments.

Stratospheric influence is manifested in our measurements by increased HCl mixing ratios in combination with other tracers (Sect. 4.3).

Did the HALO ESMVal campaign encounter an exceptional situation, or does TL entrainment at the eastern ASMA flank occur more often? The synoptic situation in question is characterised by a filament of enhanced HCl that is carried along the south-eastern ASMA flank, around a HCl-depleted ASMA interior. Here we analyse the evolution of simulated HCl mixing ratios at θ = 355 K, on a meridional (N–S) transect at 90∘ E, throughout the monsoon seasons of 2010 to 2014. The transition between eastward and westward winds indicates the location of the ASMA centre on the meridional transect, which wobbles around 30∘ N (white “ridgeline” in Fig. 3b). Enhanced tropospheric HCl mixing ratios south of the ridgeline serve as an indicator of TL entrainment (Fig. 3c). Caution with this interpretation is only needed at times when the ASMA is shifted to the west (compare Fig. 3a), because the transect in Fig. 3c may then be too far within the eastern ASMA flank. Episodes of increased HCl in the southern or eastern ASMA flank cover at least half of the time axis, showing that entrainment of TL air into the ASMA circulation is quite a common process. One horizontal slice from each analysed month is shown in the Supplement, indicating that filaments of enhanced HCl often protrude from a TP trough of a Rossby wave at the eastern ASMA flank. The association of this mixing process with planetary wave-breaking events is a topic of ongoing research (Lubis and Nakamura, 2017).

Deep convection from the lower troposphere discharges more towards the ASMA interior, as shown by studies that report relatively young air there (Li et al., 2005; Randel and Park, 2006; Park et al., 2008; Y. Liu et al., 2009; Kunze et al., 2010; Liu et al., 2011; Santee et al., 2017) and also by our simulation (Sect. 6.4). In contrast, trace gas signatures in a belt of outer streamlines are dominated by a combination of photochemically aged lower-tropospheric air and entrainments of UT air surrounding the ASMA. In this schematic of an undisturbed anticyclone, interior trace gas signatures are generally characterised by lower O3 mixing ratios than fringe signatures. This radial zoning in the ASMA is an expression of almost closed circulation and was observed in IAGOS-CARIBIC in situ data of flights between Chennai, India and Frankfurt, Germany (Baker et al., 2011; Rauthe-Schöch et al., 2016). Increased O3 mixing ratios were found in the northern part of the ASMA and decreased levels towards the southern end of the flights.

Radial stratification is counteracted by the general on–off nature of TL entrainment, upwellings from the lower troposphere and lightning. Still considering only undisturbed ASMA circulation, all those effects lead to patches of air with different trace gas signatures. Each of these patches might again receive contributions from any of the above sources. In principle all sorts of combinations are possible, generating heterogeneity. In contrast, mixing and photochemical ageing are homogenising effects. In combination with closed streamlines the preferential positions of the different sources might still show up as radial stratification in the ASMA, or show up in individual situations (Sect. 6.4).

However, neither the HALO ESMVal measurements nor sequences of simulated snapshots (Supplement) show clear stratification. The idealised picture that the ASMA circulation is dominated by stationary, closed streamlines is certainly not realistic – at least not on the timescales of the homogenising effects.

Transient streamlines, particularly eddy shedding or splitting of the ASMA, effectively overcome radial transport barriers. Whether stratified or patchy – any trace gas distribution in the ASMA might then be subject to effective stirring. There is an ongoing discussion about different dynamical modes of the ASMA (Nützel et al., 2016; Pan et al., 2016).

A splitting-up event occurred during the HALO ESMVal campaign, corresponding to the transition from a longitudinally elongated phase to a double centre phase in the nomenclature of Pan et al. (2016). A sequence of simulated streamlines and tracer distributions shows that parts of the fringe of the elongated anticyclone become part of the interiors of both resulting anticyclones after the splitting, and the interior parts are diverted into the fringes (Fig. 13). Even if not all possible cases are covered by the example, it is easily conceivable that the some or all of the fringing parts may stay in the fringes and some or all of the interior parts in the interiors. For a given location and timing of eddy shedding or splitting, the final trace gas distribution simply depends on the initial distribution of different patches. The redistribution of different parts of the anticyclone guarantees a high variability for outflow and interior, whenever the closed circulation breaks down.

How often does this happen? Splitting and eddy shedding mainly occur in a zonal (E–W) direction, whereby the transition from a northward wind in the west to a southward wind in the east indicates the centre of an anticyclone or eddy. We analyse the evolution of meridional winds at θ = 355 K, on a wide (15–35∘ N) zonal transect, throughout the monsoon seasons of 2010 to 2014 (Fig. 3a).

The pattern at the given altitude is consistent with one dominating anticyclone, centred at about 90∘ E. While the northerly winds east of 90∘ E are relatively persistent, episodes of entirely southerly winds in the western part of the one-piece ASMA alternate with episodes of smaller, secondary anticyclones. Smaller anticyclones also regularly occur east of the Tibetan anticyclone, corresponding to eddy shedding to the east. The splitting event that occurred during the HALO ESMVal campaign is clearly visible in Fig. 3a, too. Such instabilities occur approximately twice a month. This coincidentally corresponds to the timescale needed to photochemically erase O3-depleted signatures in young air masses.