In this study, we mainly focus on the analysis of data from a trajectory model to investigate the transport from the top of the PBL to the AMA. The trajectory model, which was used to calculate the backward trajectories starting in the monsoon region, was described by . This trajectory model propagates a set of trajectories, which are initialized by the user, using meteorological data, for example, from reanalysis data sets. As for the kinematic calculations presented by we have used a time step of 0.5 h and input data from 6-hourly ERA-Interim data with a horizontal grid spacing of 1.5∘×1.5∘ on 37 pressure levels from 1000 hPa (surface) to 1 hPa to calculate the trajectories.

For each day of the trajectory calculations, a set of trajectories with 1∘ horizontal grid spacing in the region 10–50∘ N ×0–150∘ E at 150 hPa was initialized at 00:00 UTC and calculated backwards for 90 d. Output (e.g. trajectory position and surface pressure below the trajectory) was produced every 6 h, and all analyses for the trajectory model data described here were performed offline on the output data. In the following, results from the trajectory model will also be indicated with the abbreviation TRJ (short for TRaJectory).

We chose the 150 hPa level to initialize the trajectories as it roughly corresponds to the 360 K from which trajectories tend to further ascend into the stratosphere . Moreover, the 150 hPa level is a level where there is strong anticyclonic circulation based on the maximum and minimum zonal wind speeds in the UT in the Asian monsoon region see, for example, Fig. 1 of. From the analysis shown in for the 100 and 200 hPa level, we expect that our qualitative results are not strongly dependent on the choice of the starting level.

We note here that there is a variety of approaches to calculate trajectories from or to the upper troposphere in the AMA region. For example, mainly focused on kinematic trajectories to investigate PBL-to-AMA transport. Similarly, used kinematic trajectories to calculate the transport from the PBL to the UT in the AMA region. Other studies employed kinematic and/or diabatic trajectories in combination with observed cloud top heights to investigate transport processes in the ASM region e.g. or hybrid diabatic trajectories e.g.. Based on Lagrangian transport model data from the CCM, we will also address the influence of diabatic versus kinematic trajectories.

In this study we also exploit Lagrangian model data from two CCM simulations described by , which incorporate the effect of parametrized Lagrangian convection. In these simulations, the CCM EMAC ECHAM/MESSy Atmospheric Chemistry;, was run together with the most recent version of the submodel ATTILA Atmospheric Tracer Transport In a LAgrangian model;, which calculated the Lagrangian transport of air parcels once with a diabatic and once with a kinematic vertical velocity scheme. For the diabatic scheme, the vertical velocity transitions from a mixed kinematic–diabatic velocity to a pure diabatic vertical velocity in the stratosphere seeand references therein. This mixed coordinate allows some of the problems to be overcome that are associated with pure diabatic trajectories in the troposphere mentioned by and by . The corresponding model results of the diabatic and kinematic simulation will be referred to as LG-D and LG-K, respectively.

Within these two EMAC–ATTILA simulations – which have the same grid point meteorology – about 1.16 million air parcels, which represent the global atmosphere, are initialized once at the beginning of the simulation and are consequently transported online with a model time step of 600 s according to the CCM's meteorological fields . Since its newest update, ATTILA can also be used with a Lagrangian convection parametrization, which is consistent with the grid point convection scheme: based on the mass fluxes of the grid point convection scheme – as provided by the host model – air parcels within a column have a probability to be vertically displaced due to convection such that there is no net vertical air parcel transport between grid boxes; i.e. the number of air parcels in each grid box remains unchanged see in particular their Section 2.2.4.

The underlying EMAC simulations have a grid point spacing of roughly ∼2.8∘×2.8∘, and the model top is located roughly at 0.01 hPa . The meteorology of the grid point model evolves freely ; i.e. it is not restrained by observed meteorology, and is hence described as free-running. The meteorological and Lagrangian data are available only every 10 h, a restriction owing to the large amount of data in the long-term CCM simulations. For further details regarding the simulation setups, see .

To analyse the transport from the top of the PBL to the AMA, we retrace the pathways of individual trajectories or air parcels during Northern Hemisphere (NH) summer (1 June to 31 August) for both the trajectory model and EMAC–ATTILA. This period covers the late ramp-up and the mature phase of the AMA . For both modelling approaches, the trajectories are followed up to 90 d backward in time. When the pressure at the trajectory position is larger than 0.85 times the surface pressure below the trajectory, we assume that the trajectory has encountered the PBL as described by . The first location where this happens backward in time will be referred to as the boundary layer source of the trajectory.

Figure  shows the definition of the PBL source regions used in this study: the TP (mainly the Tibetan Plateau) and IP (mainly the so-called Iranian Plateau) regions are defined as regions with a surface elevation of more than 2 and 0.5 km in the boxes 75–110∘ E ×25–45∘ N and 40–75∘ E ×25–40∘ N, respectively. The other source regions are named AF (mainly parts of Africa and the Arabian Peninsula), WIO (Western Indian Ocean), EIO (Eastern Indian Ocean), IND (mainly the Indian subcontinent), SEA (mainly consisting of Southeast Asia and parts of Southeast China) and the WP (West Pacific) region.

For our analyses the focus will lie on trajectories that start within the AMA, unless otherwise noted. We define the AMA boundary using a geopotential height anomaly (GPHA) criterion with respect to the 50∘ S–50∘ N mean as proposed by (; see details in the Appendix ). We emphasize that the GPHA criterion is only applied once at the starting point of the trajectories or air parcels to determine whether they are located within the AMA.

For the trajectory model data, the boundary of the AMA was determined via a GPHA threshold of 280 m using ERA-Interim data (see Appendix  for details). Consequently, all trajectories that show a GPHA of at least 280 m at their starting level of 150 hPa level were said to be located within the AMA. Sensitivity studies with a GPHA of 260 m for the trajectory model data showed that our qualitative results are not overly sensitive to the choice of the GPHA threshold.

For the EMAC–ATTILA analyses (Sect. ) a separate threshold (of 295 m) for the boundary of the AMA was determined (see Sect. ). Using a different threshold was necessary as the EMAC–ATTILA simulation is free-running (as noted before) and thus develops slightly different climatological states, for example, of the temperature . The trajectories in EMAC–ATTILA persist throughout the simulation and are thus distributed freely. Hence, they are hardly ever located at (numerically) exactly 150 hPa, and we had to use a pressure range (140–160 hPa) instead of a single pressure level (150 hPa for TRJ) to trace back air parcels from the AMA to their PBL origin. So, for each analysis time step, all air parcels which were located within 140–160 hPa on the NH in the region 60∘ W–180∘ E and fulfilled the GPHA threshold (295 m) were said to start within the AMA.

As the number of trajectories that start within the AMA varies from year to year in our analyses (both in the trajectory model and EMAC–ATTILA), we first calculate the respective distributions before producing the multiannual mean. Hence, each year contributes equally to the presented analyses.

ERA-Interim data at 1.5∘×1.5∘ horizontal grid spacing are used to calculate the TRJ data. Additionally, ERA-Interim data (partly also at different resolutions) are employed for the interpretation of the TRJ data (e.g. to provide corresponding meteorological fields, land–sea masks, and orography) and in complementing analyses.

First, we investigate the climatological properties of the transport pathways and the PBL sources of air masses from the AMA in the TRJ data, with additional notes on the interannual variability (Sect. ). For the analysis of the transport pathways, we will only consider trajectories that start within the AMA and reach the PBL within 90 d, whereas in the analyses of the PBL sources we also quantify the fraction of trajectories starting within the AMA that do not reach the PBL within 90 d (roughly 15 %; see Sect. ). Besides the strong interannual variability, the AMA is also known for its intraseasonal/subseasonal variability see, for example, Fig. 5 inshowing both interannual and intraseasonal variability. Hence, the intraseasonal variability will be discussed thereafter (Sect. ).

In Sect.  we have presented results regarding the transport from the PBL to the AMA based on our trajectory calculation (TRJ). We have found that the boundary layer source distribution (Figs.  and ) focuses over the ASM region (in particular over the Indian subcontinent and the TP). Further, these distributions support previous results regarding the PBL sources of the air masses of the AMA and its surroundings, for example, by and . Similarly, the boundary layer crossing distributions are in agreement with convective source maps of the AMA as presented by .

Moreover, we found similar regions of upward transport as , which are located on the south-eastern side of the AMA. However, we also complemented the view about the transport pathways, i.e. the conduit proposed by , by showing that air masses spread earlier in the AMA volume – in agreement with the transport pathways described by and . Combining our results with previous studies shows that the transport pathways as diagnosed by (i) a trajectory model including mixing effects , (ii) a trajectory model including the effect of observed convection , (iii) more puristic trajectory models and this study and (iv) forward trajectories (analysed backwards in time) from a Lagrangian model with parametrized convection driven by a free-running CCM (this study) are in agreement. Further, the transport pathway is also supported by (v) analyses of CO transport within a CCM and a chemistry transport model as shown by and and (vi) analyses of satellite data . In particular, our results also show that, although there is interannual and strong intraseasonal variability, the main transport characteristics are robust.

As noted by (; see end of their Sect. 3.1), there is a discrepancy between precipitation maps and source maps of the AMA air masses. Similarly, have discussed the relation of the position of strong vertical winds and their so-called conduit, i.e. the region of upward transport for trajectories that reach the AMA (see their Fig. 7 and Sect. 5). We note that precipitation maps from observations e.g.their Fig. 1 also do not directly correspond to high cloud distributions in the Asian monsoon region, as shown by . Further, it is noted by that orographic precipitation over west India is often related to low clouds. Based on these previous studies and our analyses, our understanding is as follows: low- to mid-level convection might be important for the precipitation patterns, but air parcels that are transported upwards in this convection need to find a region of onward transport to the AMA. Seemingly, for some of the regions with heavy precipitation, this rarely happens.

Regarding the source regions, our results are in agreement with some of the results found in previous studies, while keeping in mind that there are (sometimes subtle) differences in the study design: as an example, found that roughly 27 % of the all trajectories located in the AMA at 200 hPa come from the TP

Here we refer to the 1∘ ECMWF data results of , who find that about 35 % of the PBL crossing trajectories, which in turn correspond to roughly 78 % of all trajectories starting in the AMA, come from the TP in August 2011. This translates to an approximate contribution of the TP air masses to the AMA of about 27 %.

The representation of convective transport in the trajectory analyses forms the leading uncertainty in our results. This uncertainty can be addressed with two related questions: (1) how well is convective transport represented in trajectory analysis, which uses the resolved winds of analysis products? (2) What is the sensitivity of the calculations to the analysis products used? In particular, what is the influence of the relatively coarse spatial and temporal resolution of the ERA-Interim data employed in this study (here 1.5∘ and 6 h) on the presented results versus that of the newer generation reanalysis ERA5 at high horizontal resolution (∼0.25∘), provided in hourly intervals?

These questions are examined in a recent work by , in which convective transport timescales were quantitatively characterized using transit time distributions (TTDs), analogous to the age spectra, or distributions of the age of air, in stratospheric transport studies e.g. . The work uses a set of diagnostics to quantify the representation of convective transport in trajectory calculations, specifically by comparing TTDs from trajectory model results with the chemical-lifetime-based TTDs derived from airborne in situ measurements over the convection-dominated West Pacific. Four sets of wind products from commonly used operational analyses and reanalyses are examined in this study, including ERA-Interim and ERA5. The results of the study indicate that the trajectory-based TTD from ERA5 has a comparable mode and mean to that of the chemical-lifetime-based TTD. The ERA-Interim-based TTD, on the other hand, shows considerably slower transport, although it shows a qualitatively similar distribution in transport origins at the boundary layer. Using the TTD diagnostic, the ERA-Interim-based calculation misses approximately 30 % of the convective transport Table 2.

Based on this diagnosis, we expect that if the higher spatial and temporal resolution products from ERA5 were used, the result of this study would show enhanced convective transport, which should lead to a higher percentage of back-trajectories that reach the top of the PBL within the season. This assessment is also in agreement with the presented EMAC–ATTILA data, which contain the effect of parametrized convection and show a higher fraction of young (<90 d) air masses in the AMA than the TRJ data (Fig. ). Further, the EMAC–ATTILA data also support key characteristics of the transport pathways and the increasing contribution of the TP to AMA air masses over the course of the monsoon season. For the distribution of PBL source regions, although we expect changes in detail, the overall conclusions in the large-scale perspective are not expected to change. The latter is also supported by , who show similar source regions based on ERA5 (and ERA-Interim data) with an entirely different modelling approach (i.e. a combination of reanalysis and observational data).

In this study we investigated transport from the top of the PBL to the AMA; i.e. our analyses end at the top of the PBL. Convergence of surface winds at the southern flank of the TP their Fig. 8 might cause low-level transport of emissions from their source regions to the final exit and uplift region from the PBL to the AMA. As an example, emissions, for example, of CO are low over the TP their Figs. 9 and 10, respectively; nevertheless air masses transported from the PBL over the TP to the AMA can carry considerable CO signatures their Figs. 2b and 7.

Independently of potential limitations in the TRJ or EMAC–ATTILA data, the increase of TP air masses to the AMA composition is also backed up by ERA-Interim data, which is shown in Fig. : in May the core of the subtropical jet is located right above the TP. During the course of the monsoon season, the tropical easterly jet, which is located on the southern boundary of the AMA , strengthens. This indicates an increase of the anticyclonic circulation of the AMA. Further, the subtropical jet – which is located on the northern boundary of the AMA – as well as the zero wind line moves northward. Consequently, air masses that are transported upward from the TP are likely to be advected by the subtropical westerly jet during the early phase of the monsoon season (June), while they can feed into the core of the AMA during August.

Finally, as found a relatively large contribution of air masses from the TP to the AMA, they discuss their results in relation to other studies that either do or do not find important contributions of the TP to the air masses (or tracer fields) in the AMA or UTLS. While they correctly argue that the results strongly depend on the chosen analysis method, we want to add that the strong intraseasonal variability might be a reason for the differences in the assessment of the TP contribution: most of the studies that find strong contributions of the TP to the AMA or UTLS focus on August conditions, for example, , and . In contrast, investigated the source region contribution and transport budget of CO to the AMA and came to the conclusion that the TP has a relatively low impact on the CO maximum in the AMA region. For the source region contribution, i.e. the contribution of CO emitted from the TP, they showed that the lack of surface emissions from the TP leads to this minor impact. In a vertically resolved CO budget analysis for the TP region, they found that convection leads to a small maximum around 400 hPa, while advection leads to a negative tendency in the middle troposphere, and thus they argued that the TP does not play an important role in transport of CO to the AMA. The negative advection tendency found in their analysis is most likely related to the location of the subtropical jet over the TP in June 2005, which might have caused air masses to be transported out of the TP region. In our analyses, the contribution from the TP to air masses within the AMA increases as the subtropical jet shifts northwards from June to August, and we find that the transport of TP boundary layer air out of the AMA region decreases accordingly (see Fig. ). Further, put the importance of TP convection into perspective; however, they also showed that deep convective activity over the TP increases from June to August (see their Fig. 3). Similarly, from the convective upward mass flux in the EMAC–ATTILA data, we find that in July and August, the mass flux into the upper troposphere (above ∼350 hPa) over the TP is larger than in June (not shown).