Our study uses daily data from NASA's Modern Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) (Gelaro et al., 2017). This dataset is generated using the latest version of the Goddard Earth Observing System Model, Version 5 (GEOS-5), global data assimilation system, including the assimilation of aerosol optical depth (AOD) from MODerate resolution Imaging Spectroradiometer (MODIS) and Multi-angle Imaging Spectro-Radiometer (MISR) satellite retrievals. The MERRA-2 resolution is 0.5∘×0.625∘ latitude–longitude with 72 vertical levels (Molod et al., 2015). It provides 3-hourly global conventional meteorological data, i.e., temperature, winds, moisture, and precipitation, as well as the concentrations of chemical gases and various aerosol species. All the processes of aerosol transport, deposition, microphysics, and radiative forcing are included. MERRA-2 provides observation-based precipitation data; the product of precipitation has been assimilated and validated by both TRMM and GPCP (Reichle et al., 2017). Aerosol emissions from biomass burning and wildfires are derived from the satellite Quick Fire Emission Dataset (QFED; Darmenov and da Silva, 2013). The anthropogenic aerosol emission inventory is from the annual historical AeroCom Phase II (Diehl et al., 2012), up to the mid-2000s depending on the availability of emission data for various gases and aerosol species (Randles et al., 2017). Beyond that the anthropogenic aerosol emissions are not updated. As such, the direct effects due to changes in anthropogenic source emission cannot be assessed using MERRA-2. The implication of this for our results will be discussed in the Summary in Sect. 4.

In this study, we choose CO, CA that include BC and organic carbon (OC) and dust as tracers for diagnosing transport. Abundant quantities of CA and dust, found during the boreal summer season in the ASM region from local emissions and remote transport, could have strong impacts on the evolution of the Asian monsoon (Lau and Kim, 2006; Lau et al., 2006; Lau, 2014; Meehl et al., 2008; Park et al., 2009; Vinoj et al., 2014). CO is a representative pollution tracer commonly used in previous studies of UTLS transport (Pan et al., 2016; Santee et al., 2017). This chemical gas is mainly emitted from biomass burning and industrial pollution. Black carbon (BC) is a part of CA and is one of the main by-products emitted from anthropogenic sources, as well as from natural wildfire activities. OC, also a part of CA, derived mostly from biomass burning and wildfires, is more abundant than BC in ASM regions (Chin et al., 2002), and has been detected in the ATAL (Yu et al., 2015). CA are not evenly distributed in the atmosphere like CO and are subject to wet and dry deposition. Emission sources of CO and CA, such as from local biomass burning, can also be quenched by heavy monsoon rain (Lau, 2016; Lau et al., 2018). On the other hand, dust aerosols in ASM come from desert regions via long-range transport rather than from local emissions (Lau et al., 2008; Lau, 2014). This horizontal transport depends on the development of monsoon westerlies which extend from near the surface to the mid-troposphere (Gautam et al., 2009b; Lau et al., 2006; Zhang et al., 1996). While monsoon rain washout during the peak monsoon season (July–August) removes much of the coarse dust particles in and below clouds, ambient fine dust particles (< 0.2 µm) in and above clouds are lifted into the ATAL by penetrative deep convection anchored to the stem regions of the DSCC (Lau et al., 2018).

Figure 1a shows the climatological precipitation distribution and establishment of the AMA over the greater ASM region during the boreal summer monsoon season from July to August. The pronounced AMA with strong anticyclonic circulation (tropical easterlies and extratropical westerlies) develops in conjunction with heavy rainfall over the Western Ghats of India, the Indo-Gangetic Plain (IGP) of northern India, the Bay of Bengal, eastern China and the southeast Asian region (Fig. 1a). Additionally, the interannual variability of aerosols can be strongly affected by precipitation over the IGP region (Gautam et al., 2009a; Kim et al., 2016; Sanap and Pandithurai, 2015). In this study, we choose the domain (5–30∘ N, 70∘ N–95∘ E) to define strong vs. weak South Asian summer monsoon (SASM) years. This region is known to be subject to heavy monsoon precipitation and orographic forcing, which facilitates uplifting of water vapor and atmospheric constituents by penetrative deep convection to the UTLS region and above (Houze et al., 2007; Medina et al., 2010; Pan et al., 2016). The annual mean precipitation intensity for each year from 2001 to 2015 over the selected domain during the peak monsoon season (July–August) was calculated and used to represent the monsoon strength (Lau et al., 2000). Strong interannal variability and a robust increasing trend can be seen during this data period (Fig. 1b). This trend has been validated by observational data (Fig. S1 in the Supplement), and a similar increasing decadal trend of the SASM has been reported in previous studies (Jin and Wang, 2017). To focus on interannual variability, we first detrended the rainfall time series, and then defined strong vs. weak monsoon years based on the detrended time series (Fig. 1c). Strong (weak) monsoon years were selected when the mean rainfall was above (below) 1 standard deviation. Based on this procedure, four strong monsoon years (2007, 2010, 2011 and 2013; denoted as “SM”) and three weak monsoon years (2002, 2014 and 2015; denoted as “WM”) were identified. Composite mean distributions of monsoon meteorology, as well as aerosol loading transport, and ATAL variability were carried out for SM and WM, respectively, based on the detrended data. Henceforth, the term “anomaly” in the following parts refers to the difference between SM and WM composites (SM minus WM).

During SM, the AMA is stronger and more expansive than in WM, as evident in the corresponding 100 hPa geopotential height fields over the region (Fig. 2a). The AMA in SM is wavier over the extratropics and appears to have shifted poleward, indicating a stronger extratropical influence on the AMA compared to WM years. The enhanced AMA in SM occurs in conjunction with anomalous warming in the atmosphere above the TP and cooling in the lower stratosphere, as well as stronger anticyclonic circulation with anomalous westerlies at 35∘ N and easterlies at 20∘ N between 250 and 100 hPa, together with an elevated tropopause (Fig. 2b). Cooling found near the surface is due to increased precipitation and cloudiness during SM. These are well-known features of a strong SASM (Huang and Sun, 1992; Lau et al., 2018; Randel and Park, 2006; Rodwell and Hoskins, 1996; Wang, 2006; Wu et al., 2007).

Figure 3 shows spatial distributions of climatological and anomalous rainfall, AOD and low-level winds during July–August. Climatologically (Fig. 3a), heavy rain (> 6 mm day-1) is found over the Western Ghats, the Bay of Bengal and the southeast Asian region. AOD is high over northern Africa, the Middle East, and the Arabian Sea due to dust emissions from deserts and transport via the southwesterly monsoon flow to the Indian subcontinent. During SM years, enhanced precipitation is seen over the ASM land and adjoining oceanic regions of the Arabian Sea and the Western Indo-Pacific. The most pronounced increase is found over the Western Ghats of India and the HGP. Over east Asia, the presence of an elongated and southwest–northeast-oriented dipole-like precipitation anomaly, together with the increased anticyclonic low-level circulation, is indicative of a northward migration of the Mei-yu rain belt, associated with a strengthening of the subtropical high (Tao et al., 2001; Lau et al., 2000) (Fig. 3b). Stronger low-level anomalous westerlies and easterlies are found over the Arabian Sea and the equatorial western Pacific, respectively. During SM, AOD is overall lower over the Indian subcontinent and the tropical western Pacific due to stronger precipitation washout. Positive anomalous AOD is found over the Middle East and central Asia. The former is related to increase surface emission of dust, and the latter is likely due to increased biomass burning emissions (Fig. S2). Over east Asia, an increase in AOD is found, possibly due to increased CA from biomass burning (Figs. 3b, S1). Note that higher AOD and enhanced precipitation appear to coexist over northeastern China. This may be due to the aerosol swelling effect, which is related to relatively higher relative humidity induced by the enhanced Mei-yu rain belt during the moist summer monsoon season (Qu et al., 2016). Another possibility is that increased remote transport and uplifting above clouds by deep convection increased CA loading in the mid-troposphere to upper troposphere, even as CA in lower layers are removed by strong precipitation washout (Lau et al., 2018).

During SM, the 100 hPa geopotential height shows higher pressure over the subtropics and midlatitude regions (25–40∘ N), with centers over the eastern (east Asia) and western end (northern Africa) portions of the climatological AMA (see Fig. 2a). These high-pressure centers appear to be associated with a Rossby wave train pattern spanning the extratropics and the subtropics across Eurasia (Lau and Kim, 2012; Wang et al., 2008). Increased CO loading can be seen over three regions, i.e., northern Africa, the TP and central–northeastern China in an elongated “accumulation zone” along the southern flank of the expanded AMA (Fig. 4b). For CA, similar centers of action can be found, except that regions of enhanced CA loading are more expansive and cover large parts of the AMA. Stronger concentration of CA is also seen along the southern flank of the expanded AMA, consistent with stronger easterly wind transport during SM years (Figs. 4d, 2b) (Lau et al., 2018). Higher loading of CO and CA can be attributed not only to the deformation of the AMA, but also to the enhancement of surface emission during SM years. As shown in the next subsection, during SM years, higher loadings of both CO and CA in the UTLS are found near regions of enhanced emissions only when there is increased vertical motion from deep convection (Fig. S2). Similar to CA, more dust is also evident over the accumulation zone spanning northern Africa, the Middle East, the TP and east Asia during SM (Fig. 4f).

In this subsection, we examine the changes in the ATAL structure along the axis of the DSCC (25–35∘ N) during SM and WM years. We begin with the structural changes in the vertical motion field under the influence of the AMA (Fig. 5d). During SM years, overall enhanced anomalous ascending motions are found over the western sector (east of 85∘ E), while anomalous descending motions are found over the eastern sector (west of 85∘ E) of the AMA. In the western sector, two regions with strong vertical motion are found clustered over northern Africa and the Middle East (15–50∘ E) and over the foothills of the HGP (70–85∘ E) with an anomalous ascent extending above 100 hPa in both regions. Over the western sector and embedded within a large region of overall anomalous descent, an enhanced ascent is also found over east Asia around 105–115∘ E reaching above 100 hPa. As noted earlier (see Fig. 3c), during SM years, the Mei-yu rain belt is shifted northward, leaving behind mostly anomalous descending motions in this latitudinal zone. However, a moderately increased ascent is found over western central China (105–120∘ E) from the eastern foothills of the TP and the SB, collocating with the southern tip of the northward-shifted Mei-yu rain belt. These three regions of anomalous ascent play essential roles in the distribution of chemical gases and aerosols species in the ATAL.

During SM years, the CO concentration is generally increased in the ATAL (Fig. 5a), consistent with the enhanced advection by the strengthened easterlies at the southern flank of the AMA. Three centers of anomalous high CO concentration in the UTLS (200–100 hPa) over northern Africa, the TP and east Asia (identified in Fig. 4b) stand out. These centers appear to be connected via stems of high CO related to the aforementioned three regions of anomalous ascent. The large reduction in CO near the surface over east Asia may be related to the quenching of emission sources by increased precipitation over this region (Figs. 5a, S2b). For CA, the pattern of anomalies is similar to the pattern of CO, with overall increased loading in the UTLS, and three action centers connected by stems of high CA to the surface (Fig. 5b). The increase in near-surface CA over desert regions (east of 70∘ E) is consistent with increased surface emissions (Fig. S2d). The reduction in CA in the monsoon region (west of 70∘ E) is likely due to stronger precipitation washout during SM. Likewise, during SM, severely suppressed dust is found near the surface up to the mid-troposphere in the stem over the HGP (60–100∘ E), associated with washout by the increased precipitation (Figs. 5c, 3b). Similar to CO and CA, dust reduction can also be seen in the middle and lower troposphere over eastern China (105–135∘ E) because of the enhanced rainout process. Due to the increased near-surface wind, dust loading is increased over the Middle East (30–70∘ E) but decreased over northern Africa. Sources of dust contributing to the increased dust loading in the UTLS (above 200 hPa) seem to mainly come from the Middle East and west Asia, with some contribution from the eastern TP, abutting the SB region.

Two meridional cross sections (80–85 and 100–105∘ E), for the HGP (Fig. 6) and the SB (Fig. S3) regions, respectively, have been examined. Because of a similarity in patterns, only the HGP region (Fig. 6) is discussed here. Ascending motions during SM years over the HGP region near the foothills and top of the TP are enhanced and weakened locally in the vicinity of 20∘ N, associated with the enhancement and northward shifting of the AMA (Fig. 6d). Additional increased ascending motions are south of 20∘ N, likely with the increased precipitation over southern India and the northern Indian Ocean (see Fig. 3b). A dipole pattern featuring increased CO over the top of the TP from 500 to 70 hPa at the northern edge of the climatological CO maxima was coupled with reduced CO south of 20∘ N; Fig. 6a again indicates that more CO was lifted into the UTLS by the enhanced vertical motion associated with the northward shift of the AMA during SM years. The reduction of CO in the lower troposphere and near the surface in the extratropics (40–58∘ N) is likely related to the quenching of emission sources of biomass burning over the region (Fig. S2b). Similar to CO, more CA are transported and enter the UTLS via the HGP stem in SM years, and the increased loading is more expansive than CO spanning 25–60∘ N, from 500 to 50 hPa. This may be due to an increase in biomass burning emission sources over northern central Asia (Figs. 6b and S2d). Associated with the northward shifting of the AMA, CA concentrations below 100 hPa over the tropical region are substantially reduced. During SM years, dust is mostly reduced over the regions from the surface to the upper troposphere. Increase uplifting of dusts into the UTLS by anomalous ascending motions is found over the TP and the Taklamakan desert (35–42∘ N). The pronounced reduction in CA and dust loadings over the foothills of the TP and the Indian subcontinent is due to wet scavenging effect by the enhanced rainfall over the region. For the SB stem region (Fig. S3), the pattern of anomalous concentrations of CO, CA and dust in the ATAL is similar to the HGP region, reflecting the competing influences of lofting by deep convection, emission quenching (for CO) and removal by precipitation washout (for CA and dust).

To depict a long-term change in the ATAL we have computed time series of CO, CA and dust averaged in the 200–100 hPa layer, and over a large domain (60–120∘ E, 25–35∘ N), approximately bounding the AMA. For comparison, a time series representing the strength of the AMA, defined as the difference in zonal winds between northern (30–40∘ N) and southern (10–20∘ N) flanks of the AMA, has also been constructed (Fig. 7). Clearly, CO and CA in the ATAL show significant increasing trends during 2001–2015, at a rate of +7.8 % (p value =0.018) and +12.7 % per decade (p value =0.025), respectively. A similar trend of CO is also seen in the results from MLS observation, and the difference in a certain year can be attributed to bias from observations and the emission inventories used in simulation (Fig. S4). Both the CO and CA trends are consistent with a significant (p value =0.06) trend of AMA strength at a rate of +6.7 % per decade. Given that the AMA is an essential component of the SASM, this suggests that the trends of increased loading of ATAL CO and CA could be attributed to the strengthening of the SASM during 2001–2015. For dust, the positive trend is weak, with a rate of 1.6 % per decade, and not significant (p value =0.875) due to the large interannual variability. The weak ATAL dust trend may be due to the removal of a large fraction of dust particles by wet scavenging in and below raining clouds, outweighing the effects of lofting by deep convection (Chin et al., 2000; Lau et al., 2018). Additionally, the large interannual variability of ATAL dust transport is also likely a reflection of the influence of non-monsoon factors, such as extratropical westerlies that can strongly affect long-range dust transport at high elevations (Sun et al., 2001; Huang et al., 2007).

To better understand the physical processes underpinning the ATAL long-term trend signal, we have constructed the time–mean vertical profiles of ATAL constituents, vertical motions and rainfall along critical east–west cross sections spanning the AMA for the early period (EP; 2001–2006) and later period (LP; 2010–2015), respectively. The long-term change is defined as the difference between the two periods (LP minus EP). Figure 8a–d shows east–west cross sections of long-term changes in CO, CA, dust and vertical motions respectively, covering the same ASM region as in Fig. 5. During LP, enhanced ascending motions (relative to EP) that reach the ATAL are most pronounced over Pakistan and Northeast India and the HGP region (60–95∘ E) (Fig. 8d). A cluster of ascending motions are also found over greater SB regions of east Asia (100–130∘ E), in connection with the northward migration of the Mei-yu rainbelt (See Fig. 3b). A third region of enhanced ascent is found over northern Africa (15–30∘ E). During LP, overall, the CO concentration increases from the surface to the UTLS, with pockets of reduced CO near the surface due to biomass emission quenching by precipitation (Fig. 8a). Similarly, CA concentration at the UTLS is increased during LP (Fig. 8b), and appears to be connected to surface sources of increased CA over northern Africa and the Middle East and the west Asia region (Fig. S3) via the increased ascending motions over Pakistan and Northeast India and the HGP region (60–90∘ E). Strong reduction in CA from the surface to the mid-troposphere found over east Asia (100–130∘ E) is due to the removal by increased precipitation washout. Compared to CO and CA, the increase in ATAL dust is modest (Fig. 8c), and appears to follow a transport pathway from the surface to the UTLS similar to CA. The increase in surface dust over the Middle East and the west Asia region (40–70∘ E) may be related to a robust recent decadal warming trend over the Indian subcontinent and the Middle East (Jin and Wang, 2017). A hotter desert surface is likely to favor a deeper planetary boundary layer, enhanced dry convection and uplifting of dust from the surface (Gamo, 1996; Cuesta et al., 2009). During LP, an overall reduction in dust from the surface to the mid-troposphere over monsoon regions is due to removal by increased precipitation washout.

Next, we examine the competing influences of lofting by overshooting convection and precipitation washout in the DSCC stem regions (25–35∘ N, 65–115∘ E), including both the HGP and SB domains. The ATAL trend is found by examining the mean daily variations of monsoon precipitation and vertical profiles of CO, CA and dust over the region, during EP and LP, respectively. During LP, monsoon precipitation is enhanced compared to EP from June through August (Fig. 9d, h), consistent with the increased rainfall trend shown in Fig. 1b. CO concentrations from the surface to 200 hPa in LP are higher than in EP during the pre-monsoon period in May–mid-June (Fig. 9a, e), reflecting a hotter land surface and enhanced dry convection over the region before monsoon onset. The onset of the monsoon, as characterized by an abrupt rise in CO (region shaded by light yellow in Fig. 9a, e) to above 200 hPa reaching the ATAL, occurs earlier in LP (around 16 June) compared to EP (around 1 July). Thereafter, CO remains higher in LP, through the end of the monsoon season, maintaining a longer residence time in the ATAL, via the cumulative effect (multiyear mean) of lofting by deep convection. From the surface to the lower troposphere, CO concentration declines faster in LP, due to the quenching of emission by heavier monsoon rain. Likewise, for CA, features such as the earlier onset, the increased ATAL concentration (above 200 hPa) and the longer residence time during LP are also pronounced (Fig. 9b, f). The competing influences of convective lofting and wet removal can be seen in the more episodic increase in ATAL loading in both EP and LP, more so in the latter. During LP, the more efficient lofting of CA into the ATAL from the mid-troposphere during early July and late August coincides approximately with the time of maximum precipitation, when deeper and more overshooting convection tends to occur (Fig. 9f). During May in LP, a strong increase in CA from the surface to 200 hPa is noted. This could be related to a warming trend of the land surface over northern India and the desert regions to the west (Jin and Wang, 2017). A warmer and drier land surface before monsoon onset is likely to favor increased biomass burning emissions (van der Werf et al., 2006) (Fig. S5). In contrast to CO and CA, dust concentration in the ATAL varies little from EP years to LP, with a slight signal of increased convective lofting during mid-July to mid-August in LP. This is consistent with the weak positive, but statistically insignificant, dust trend shown in Fig. 7. A notable signal is the increase in dust loading from the surface to 300 hPa during May in LP compared to EP and a rapid decline due to removal by washout during June–August. A similar analysis has also been carried out separately for the HGP and SB regions. Results show that while both regions exhibit similar characteristic features regarding convective lofting and washout, the signal over the HGP is more pronounced than that over the SB region (Figs. S6 and S7). This may be because the Mei-yu rainfall system affecting the SB region possesses more transient and migratory features compared to the more landlocked convection over the HGP region (Ding and Chan, 2005; Lau and Weng, 2001).