Linkage of water vapor distribution in the lower stratosphere to organized Asian summer monsoon convection

Abstract

Accumulation of water vapor in the upper troposphere/lower stratosphere (UT/LS) over the Asian continent is a recognized feature during the boreal summer monsoon. While there has been a debate on the role of monsoon convective intensities on the UT/LS water vapor accumulations, there are ambiguities with regard to the effects of organized monsoon convection on the spatial distribution of water vapor. We provide insights into this aspect using high precision balloon measurements of water vapor from a high-elevation site Nainital (29.4° N, 79.5° E), India, located in the Himalayan foothills and satellite retrievals of water vapor from the Microwave Limb Sounder (MLS). We also use precipitation estimates from the Tropical Rainfall Measuring Mission (TRMM) satellite (i.e., merged product 3B42 and precipitation radar 3A25 estimates of rain rate and rain type viz convective/stratiform), reanalysis circulation data, as well as numerical model simulations. We first evaluate the MLS estimates of water vapor mixing ratios with in situ high precision hygrometer balloon observations over Nainital. It is seen from our analyses of the MLS data that the LS water vapor distribution is closely linked to the organization of the South Asian monsoon convection and its influence on the UT/LS circulation. This link between LS water vapor distribution and organized monsoon convection is also captured in the in situ observations on 3 August 2016. It is evidenced that periods of organized summer monsoon convective activity over the Indian subcontinent and Bay of Bengal promote divergence of water vapor flux in the UT/LS; additionally the Tibetan anticyclonic circulation causes widespread distribution of the UT/LS water vapor. In addition to the effects of Asian monsoon convection, we also note that global climate drivers such as El Niño-Southern Oscillation (ENSO), Brewer–Dobson circulation (BDC), and Quasi-Biennial Oscillation (QBO) can contribute to nearly 38% of the UT/LS water vapor variability over the Asian monsoon region. The main result of our study indicates that widespread spatial distribution and accumulation of water vapor in the LS (about 80% of total accumulation between May and August months) tend to co-occur with organized monsoon convection, intensified divergence of water vapor flux in the UT/LS and intensified Tibetan anticyclone. On the other hand, the circulation response and LS water vapor distribution to pre-monsoon localized deep convection tend to have a limited spatial scale confined to Southeast Asia. Results from model experiments suggest that the UT/LS circulation pattern to organized monsoon convection has resemblance to stationary Rossby waves forced by organized latent heating, with the westward extending response larger by about 15° longitudes as compared to that of the pre-monsoon localized deep convection.

1 Introduction

The distribution of atmospheric water vapor is highly variable in space and time and is known to exhibit strong gradients across the tropopause region (e.g., Sinha and Harries 1995; Forster and Shine 2002; Ravishankara 2012). Although water vapor abundance (or mixing ratios) in the tropical upper troposphere/lower stratosphere (UT/LS) is not phenomenal by amounts (see Johny et al. 2009), small quantities of water vapor entering the stratosphere still have marked impacts on radiative processes and stratospheric ozone chemistry, making it a critical element for radiative feedback processes (see Solomon 1988; Forster and Shine 1999; Dessler et al. 2013). The causal dynamics behind the distribution of water vapor in the UT/LS region is a challenging scientific problem due to its multi-scale variability ranging from daily to inter-annual time scales (Mote et al. 1995; Solomon et al. 2010; Fujiwara et al. 2010; Randel et al. 2015; Tao et al. 2019).

The hydration and dehydration mechanisms of the LS associated with the major monsoon systems are topics open to scientific research. Some studies have suggested the role of the seasonal northward movement of the Intertropical Convergence Zone (ITCZ) and convective activity over the Asian monsoon region on the seasonal cycle of water vapor in the tropical LS (e.g., Panwar et al. 2012; Jain et al. 2013). On the other hand, Uma et al. (2014) suggested that the high concentration of water vapor in the LS over the Asian summer monsoon (ASM) region may be linked to warmer temperatures in the tropopause, particularly during August. Randel et al. (2015) pointed out that the subseasonal variability of the LS water vapor is significantly higher over the ASM region in association with intense deep convective activities, as compared to its North American counterpart. Further, they noted that intense deep convection in the monsoon regions leads to colder subtropical temperatures and low amounts of LS water vapor, which is contrary to the expected increase through direct injection by deep overshooting convection. Fu et al. (2006) noted that elevated surface heating over the Tibetan Plateau during boreal summer generates warmer tropopause temperatures and less saturated ambient air, thereby allowing more water vapor to enter into the LS.

Seasonally persistent convection over South and Southeast Asia during the boreal summer monsoon is associated with a large-scale anticyclonic circulation in the UT/LS (also referred to as the Tibetan anticyclone; see Krishnamurti and Bhalme 1976; Krishnamurti et al. 2008). The Tibetan anticyclone in the UT/LS is a distinct quasi-permanent circulation feature during the boreal summer that spans zonally across the Eurasian continent, encompassing the tropical easterly jet stream and the subtropical westerly jet on the southern and northern flanks, respectively, and having vertical extension from 300 to 70 hPa (Mason and Anderson 1963; Hoskins and Rodwell 1995; Yanai and Wu 2006; Krishnamurti et al. 2008). The strength of the Tibetan anticyclone is generally related to variations in convective forcing and associated release of latent heating by the monsoon cloud systems (Krishnamurti and Bhalme 1976; Garny and Randel 2013; Siu and Bowman 2019). Eddy shedding dynamical instabilities (Popovic and Plumb 2001; Nützel et al. 2016), monsoon intra-seasonal oscillations (Goswami 2012), and mid-latitude wave interactions (Dethof et al. 1999) can significantly affect the variability of the Tibetan anticyclone. Observational and modeling investigations have highlighted the role of the Tibetan anticyclone in offering a dynamic pathway for transporting various greenhouse gas constituents and particulate matter from troposphere to stratosphere and vice versa (e.g., Gettelman et al. 2004; Park et al. 2007; Randel et al. 2010; Fadnavis et al. 2013; Brunamonti et al. 2018; Hanumanthu et al. 2020; Kumar et al., 2021). In particular, the water vapor mixing ratios in the UT/LS are substantiated through diabatic and adiabatic transport pathways, such as vertical injections of water vapor into the stratosphere resulting from tropopause (deep convective) overshooting (e.g., Corti et al. 2008), and quasi-horizontal isentropic (adiabatic) mixing along the subtropical tropopause (e.g., Dethof et al. 1999; Dessler and Sherwood 2004; Ploeger et al. 2013).

Monsoon convection, composed of stratiform-type precipitation from nimbostratus clouds together with embedded deep convective cells, is known to organize over large spatial scales (3000–4000 km) across South and Southeast Asia during the boreal summer monsoon season (e.g., Houze 2004; Choudhury and Krishnan 2011; Krishnan et al. 2011; Romatschke and Houze 2011; Choudhury et al. 2018). These organized systems contain an evolving pattern of newer (convective) and older (stratiform) precipitation types, where both types occur within the same complex of convection generated clouds (Houze 1997). Latent heat release by organized convection has a dominant control on the strength and spatial structure of the South Asian monsoon circulation (Choudhury and Krishnan 2011) and the large-scale tropical divergent circulations, in general (Schumacher et al. 2004). While taking note of the aforesaid studies, the role of organized monsoon convection on spatial distribution of water vapor in the UT/LS remains unclear. The present study aims to gain better understanding of the linkages among the LS water vapor distribution, the Tibetan anticyclone and organized monsoon convection. For this purpose, we use diagnostic analysis of satellite observations of water vapor and precipitation, meteorological reanalysis datasets, in situ balloon-borne water vapor observations and numerical experiments with a simplified global atmospheric model. Details of datasets, model setup and numerical simulations are described in the following section.

2 Datasets used in the study and model details

2.1 Balloon observations

In situ balloon-borne water vapor measurements from a high-elevation site Nainital (29.4° N, 79.5° E; elevation = 1793 m amsl) in the Uttarakhand state of India in the Central Himalayas (see Fig. 2 for location) are used together with satellite observations, reanalysis circulation products and atmospheric model experiments (Sect. 2.3) for the present study. Nainital is located in the southern slopes of the pristine Himalayan environment which is influenced by orographic forcing to the moist monsoonal circulation and precipitation processes (Vellore et al. 2016). The measurements were part of the StratoClim project (see Brunamonti et al. 2018) where 30 balloon soundings were conducted during the monsoon season (02–31 August, 2016). Meteorological latex balloons (TOTEX, Japan) filled with hydrogen gas and equipped with standard meteorological radiosondes RS41-SGP, Vaisala, Finland, along with the DigiCORA MW41 sounding system, were utilized for balloon soundings. Out of the 30 soundings, 27 were flown with the Cryogenic frost-point hygrometer (CFH, ENSCI, USA; see Vömel et al. 2007). The CFH works on the chilled-mirror principle and provides in situ estimate of water vapor mixing ratios with an uncertainty of typically less than 10% up to an altitude of 28 km (Vömel et al. 2007). More than half of the soundings were performed during the nighttime. CFH electronics failure was observed in one case, along with five instances of CFH contamination (supplementary Table S1), for which the data was discarded after comprehensive quality control procedures (for more details see Brunamonti et al. 2018; Jorge et al. 2020). The maximum burst altitude of these balloons was 34.4 km and more than 70% of the soundings reached at least 30 km. For the soundings with CFH instrument, valid data was available in the UT/LS for at least two third of cases where most of the flights went past 60 hPa (see supplementary Table S1). For additional details on meteorological conditions in the UT/LS over the Central Himalayan region, the reader is also referred to the analysis of high-frequency balloon soundings during the RAWEX-GVAX campaign by Naja et al. (2016).

2.2 Satellite-derived datasets and reanalysis products

The present study also utilizes the satellite-derived water vapor measurements from the Aura Microwave Limb Sounder (MLS) version 4.2 (v4.2) level 2 products (Livesey et al. 2018). The MLS measures thermal emissions from the limb of the Earth’s atmosphere in 5 spectral regions with near global coverage (82° S–82° N) by vertically scanning the limb in the orbit plane (Waters et al. 2006). The MLS instrument aboard the Aura spacecraft, since its launch in August 2004, has been providing global measurements of water vapor and many other chemical species in the lower and middle atmosphere. The MLS water vapor profiles are retrieved from the 190 GHz rotational line measurements where the vertical resolution varies from 1.3 to 3.6 km between 316 to 0.22 hPa, and the horizontal along-track resolution varies from 170 to 350 km (for pressure levels greater than 4.6 hPa) while the cross-track resolution is 7 km (for all pressure levels). The Aura-MLS provides about 3500 profiles per day and the reported accuracy (precision) of 8–15% (15–20%) with minimum of 0.1 ppmv is found in water vapor measurements. Compared with previous versions, MLS v4.2 applies better gas-phase retrievals in the presence of deep convective clouds, and uses improved data processing algorithms (Livesey et al. 2018). The present analysis is carried out using the daily MLS water vapor data for the period 2005–2016. We bin the full spatial domain covered by the MLS in rectangular boxes each having a resolution of 5° latitude × 10° longitude. A daily gridded data of the vertical profiles of water vapor is constructed by averaging the profiles within these boxes. It is to be noted that the daily mean here is obtained from twice a day observations, and does not refer to the diurnal mean. While processing individual water vapor profiles, we have followed all the recommended quality control criteria as documented by Livesey et al. (2018).

For precipitation estimates, we use the Tropical Rainfall Measuring Mission (TRMM) data. The daily accumulated TRMM precipitation used in the analysis, is a merged precipitation product based on the TRMM Multi-Satellite Precipitation Analysis TMPA (3B42) and precipitation gauge analysis archived at 0.25° × 0.25° resolution (Huffman et al. 2010). Additionally, we use the TRMM 3A25 precipitation radar products (Awaka et al. 1997) of rain rate and rain type (convective/stratiform) for our analysis. The NOAA outgoing longwave radiation (OLR) data (http://www.cdc.noaa.gov/) archived at a horizontal resolution of 2.5° × 2.5° is used for identification of monsoon convective activity. Three-dimensional reanalysis circulation (winds, geopotential heights, vertical motion) products used in this study are obtained from the European Centre for Medium-Range Weather Forecasts (ECMWF) Interim Reanalysis (ERA-Interim) (Dee et al. 2011) available at a 0.75° × 0.75° horizontal resolution and 37 pressure levels. The time-series of quasi-biennial oscillation (QBO) and total diabatic heating indices used in our study, are from NOAA Climate Prediction Center (http://www.cpc.ncep.noaa.gov/data/indices) and Modern Era Retrospective analysis for Research and Applications Version 2 (MERRA2; Gelaro et al. 2017), respectively. The MERRA2 based diabatic heating indices are used for the analysis of Brewer-Dobson circulation (BDC) impacts on the LS water vapor.

2.3 Model details

We conducted numerical simulation experiments using a multi-level (25 vertical levels) simplified dry atmospheric global spectral model with rhomboidal truncation at zonal wave-number 40, forced with observed heating. The dynamical core of the model is based on the original spectral formulation of Bourke (1974) and has been used in various studies that were focused on monsoon and tropical dynamics (eg., Krishnakumar et al. 1993; Krishnan and Kasture 1996; Sundaram et al. 2010; Choudhury and Krishnan 2011; Choudhury et al. 2018). The model is forced with observed latent heating estimated using surface rain estimates and convective/stratiform fractions from the TRMM PR 3A25 monthly rainfall dataset (see Choudhury and Krishnan 2011). Although the simplified atmospheric model does not simulate variations of water vapor in the UT/LS, the numerical experiments provide insights into the dynamical response to monsoonal heating which has direct consequences on the upper level monsoon anticyclone and the associated water vapor distribution in the UT/LS. The model experimental design is discussed in Sect. 4.

3 Diagnostics of LS water vapor during the Asian summer monsoon

In this section, we present a diagnostic analysis of LS water vapor, circulation and convection features during the ASM. Earlier studies, which compared MLS remote and CFH in situ water vapor measurements, suggested a generally good agreement between both the measurements within an areal distance of about 900 km of the satellite pass and a temporal difference of up to 12 h (see Vömel et al. 2007; Yan et al. 2016). First, we provide a brief analysis of the in situ measurements during the StratoClim campaign at Nainital during 02–31 August, 2016 and the satellite-derived MLS water vapor observations. It is noted that the mean distance between the Nainital measurements and the MLS satellite passes was about 500 km, with a time separation of 9 h. The balloon measurements over Nainital are also scientifically important given that the site location lies in the core of the Tibetan anticyclone, where the water vapor mixing ratios are larger by about 1 ppmv relative to the smaller concentration estimates observed over locations in southern India (e.g., Trivandrum [8.5° N, 76.9° E] and Hyderabad [17.47° N, 78.58° E]), as reported by Emmanuel et al. (2018). Furthermore, one of the observed case periods over Nainital during August 2016 co-occurred with organized monsoon convection, which will be later discussed in Sect. 3.4.

The vertical profiles of water vapor mixing ratios in the UT/LS region obtained from balloon-borne measurements and MLS retrievals during the monsoon period are shown in Fig. 1. The region between 100 and 70 hPa is usually considered for the analysis of LS water vapor distribution during the summer monsoon season (see Table 3 of Brunamonti et al. 2018). The in situ measurements during the monsoon season are based on the average and standard deviation of balloon soundings from 02 to 31 August 2016. It is clearly seen that significantly high water vapor mixing ratios (> 6 ppmv) are observed up to 85 hPa (Fig. 1). Further, it is noted that the difference between the MLS and in situ observations is relatively small (root-mean square error of 0.9 ppmv at 100 hPa). This comparative analysis between MLS estimates of water vapor mixing ratios and high precision hygrometer balloon observations shows that the MLS observations provide reliable estimates of water vapor variability inside the Asian summer monsoon anticyclone. The buildup of LS water vapor maximum over the Asian monsoon region starts typically from May and continues until August (see Randel et al. 2015). The May–August mean climatological water vapor mixing ratio at 100 hPa (Fig. 2a) shows a wide area of high water vapor (mixing ratio > 4.8 ppmv) that extends eastward from western Eurasia (~ 40° E) to the Western Pacific (~ 140° E) and meridionally between 20° N and 40° N. Further note that the widespread distribution of water vapor in the UT/LS closely follows the spatial extent of the Tibetan anticyclone, as evidenced from the climatological mean winds at 100 hPa (Fig. 2a). It is also verified that the spatial distribution of the MLS climatological mean water vapor at 82 hPa, is similar to that of 100 hPa with relatively smaller magnitudes (not shown).

Fig. 1

MLS and CFH observed profiles of water vapor mixing ratios (units in ppmv) during the monsoon season (August) of 2016. Solid lines (bars) indicate mean values (standard deviations)

Fig. 2

a May–August climatology of 100 hPa water vapor mixing ratios (shaded; units in ppmv) and winds (vector, maximum = 25 m s⁻¹) and for the period 2005–2016. The place of observation at Nainital is marked by a blue star. b Time series of box area-averaged [25° N–40° N, 70° E–105° E; shown as a blue rectangular box in (a)] water vapor mixing ratios at 100 hPa and 82 hPa for the period 2005–2016

3.1 Reconstructed time-series of LS water vapor

This section focuses on the analysis of LS water vapor using the MLS daily dataset for the period 2005–2016, over the core region of the ASM water vapor maximum (see Fu et al. 2006). The core ASM region is located over the Tibetan Plateau and its southern slope (highlighted with the blue rectangular box in Fig. 2a). The daily time-series of LS water vapor averaged over this region at 100 hPa and 82 hPa are shown for the period 2005–2016 in Fig. 2b. Note that the interannual variations of water vapor at these two levels are very similar. We have also confirmed that the water vapor variability over a broader region (black rectangular box in Fig. 2a) compares well with that over the core ASM region (Fig. S1). Furthermore, it is important to note the strong seasonal cycle of water vapor which starts increasing from the pre-monsoon month of May and reaches peak values during August and decreases thereafter, thus highlighting the irreversible seasonal increase in LS water vapor during May–August months (Fig. 2b).

The entry of water vapor in the LS is largely affected by the tropopause temperature and global scale climate drivers e.g. the El Niño-South Oscillation (ENSO), the Brewer-Dobson circulation (BDC), and the Quasi-Biennial Oscillation (QBO) (Dessler et al. 2014). Though convection can vary on shorter time scales, other processes mainly occur on seasonal or interannual time scales and the extent of coherency in these processes affects the water vapor mixing ratios in the LS (Avery et al. 2017; Das and Suneeth 2020). To bring out the influence of monsoon processes on the LS water vapor distribution, it is imperative to separate the effects of these major climate drivers. For this purpose, we employ a multivariate linear regression on the MLS water vapor data using BDC and QBO indices, and the tropical average 500 hPa temperature anomaly index (a better alternate of ENSO index) as regressors. The regression equation used in the analysis is similar to that of Dessler et al. (2014), given as following:

$${\text{H}}_{2} {\text{O}}_{{{\text{anm}}}} = {\text{a QBO}} + {\text{ b BDC}} + {\text{ c }}\Delta {\text{T}} + \varepsilon .$$

In the regression analysis, ΔT is used as it captures the most of variability related to convection and the ENSO. The standardized QBO index (prepared from equatorial 50 hPa zonal winds) is obtained directly from the NOAA Climate Prediction Center (http://www.cpc.ncep.noaa.gov/data/indices) where positive/negative indices denote westerly/easterly QBO phases. BDC index which indicates the strength of upwelling is prepared using the tropical 100 hPa total heating rate anomaly as a proxy (see Wright and Fueglistaler 2013; Dessler et al. 2014; Wang et al. 2019) where the positive/negative indices indicate stronger/weaker upwelling. We prepare ΔT and BDC indices by taking the tropical averages (30° S–30° N) of 500 hPa temperature anomaly and 100 hPa total heating rate anomaly. Also, anomalies in the regressors are normalized by their own standard deviation to obtain the standardized indices. The term ε denotes residual. During the analysis period (2005–2016), the three regressors are found to be uncorrelated. In the regression analysis, the QBO and ΔT regressors are lagged by 3 and 1 months respectively while no lag is used for the BDC regressor. It is done to account for the delay required for regressors to affect the 100 hPa water vapor. The coefficients a, b and c for QBO, BDC and ΔT are 0.08, 0.05 and 0.07 respectively. The coefficient of determination (R²) which summarizes the proportion of variance explained by the regression model is found to be 0.38. It shows that the regression explains about 38% of the total variance during the analysis period, while the dominant part of variability is linked to the monsoon processes. The results of this analysis are shown in Fig. 3.

Fig. 3

Time series (May-2005 to September-2016; the tick marks on time axis correspond to the May month) analysis of area-averaged [25°–40° N, 70°–105° E; shown as a blue rectangular box in Fig. 2a] 100 hPa water vapor mixing ratios (ppmv) denoting: a components of the multivariate regression of deseasonalized 100 hPa water vapor anomalies showing water vapor anomalies (black lines), water vapor anomaly due to each individual regressor i.e., ΔT at 500 hPa (red lines), the quasi‐biennial oscillation (QBO; blue lines), and the Brewer–Dobson circulation (BDC; pink lines), b the residual of deseasonalized water vapor mixing ratio after removing the ENSO, BDC and QBO effects, and c reconstructed time series

Figure 3a shows the deseasonalized water vapor time series and the regression fits (the explained water vapor vapor anomaly) as obtained by the regression analysis. It is seen that part of the total variability is caused by these processes where the faster variations result due to warmer tropopause and the BDC, while the QBO phases generate slower variations in the LS water vapor over ASM region. During the analysis period, ΔT regressor causes as high as 0.28 (− 0.15) ppmv increase (decrease) while the BDC regressor contributes to a maximum (minimum) of 0.12 (− 0.12) ppmv and the slower varying QBO signal causes close to a maximum (minimum) of 0.15 (− 0.2) ppmv changes in the LS water vapor. We use the residual water vapor obtained after removing these effects. In order to obtain the variability due to monsoon processes, we remove the effects of ENSO, QBO and BDC and the time-series of residual water vapor after removal of the effects is shown in Fig. 3b. Clearly, a large variability in LS water vapor is inherent to the monsoon system over the region. The daily mean water vapor values are reconstructed by adding the residual water vapor back to the daily time series. Further, we reconstruct the time-series of LS water vapor over the ASM region by adding the residual and mean values of daily water vapor. The reconstructed time series is shown in Fig. 3c. It is noted that the standard deviation during the monsoon period of the reconstructed time-series is 0.46 ppmv, which is substantially higher than the largest magnitudes of the effects due to ENSO (0.28 ppmv), QBO (0.2 ppmv) and BDC (0.12 ppmv) on the LS water vapor variability over the ASM region.

3.2 Frequency distribution of water vapor in the LS

We performed a frequency distribution analysis of the reconstructed time-series of daily LS water vapor (see Fig. 3c), which is discussed in this section. This analysis is performed both for the MJJA and JJA months, which is necessitated by the strong seasonality and monotonic rise of LS water vapor from May to August (MJJA), as well as the differences in the spatial scales of convection characteristics between the pre-monsoon and monsoon seasons. The corresponding frequency distribution curves for MJJA and JJA are shown in Fig. 4 by black and red lines, respectively. It can be seen that the frequency distribution analysis for the MJJA months (black line, Fig. 4) exhibits a bimodal-like appearance in the water vapor distribution at 100 hPa with peaks at 5 ppmv (referred to as the mode A, hereafter) and 4 ppmv (referred to as the mode B, hereafter) and a standard deviation of 0.46 ppmv. However, the bimodal appearance in this case is associated with the systematic transition in convective activity between pre-monsoon to monsoon season and it is equivalent to temporal separation between the two peaks. The separation between the two peaks is quantified using statistical tests and finite mixture modeling as a tool (see Appendix A, Table 2). It is also noted that the frequency distribution curve constrained for the JJA months (red line, Fig. 4) is largely unimodal. It is to be noted that the statistical separation of bimodality is a result on appearance and it is almost equivalent to separation of pre-monsoon and monsoon cases.

Fig. 4

Frequency distribution (binned at 0.1 ppmv interval) of the reconstructed water vapor time series for the months of May–June–July–August (MJJA) and June–July–August (JJA). The modes around the two prominent peaks (5 ppmv and 4 ppmv) during MJJA period are identified to as mode A and mode B by considering the cases falling within one standard deviation at either side of the peaks (0.23 ppmv both ways) of the distribution

To understand the UT/LS water vapor contributions from individual months, we analyzed the relative frequency distributions for May, June, July and August, separately (Fig. 5). It is clearly seen that progression from pre-monsoon to monsoon months leads to shift in the water vapor mixing ratio towards higher concentrations. During the pre-monsoon (May) month, water vapor concentration is mostly centered around 4 ppmv (Fig. 5a) and later shifts to around 4.5 ppmv in the month of June during the monsoon transition phase (Fig. 5b). As we progress to the core monsoon months, center of the frequency distribution is around 5.2 ppmv and 5.5 ppmv for July (Fig. 5c) and August (Fig. 5d), respectively. It is interesting to note that the width of the May distribution is narrower, with more frequent occurrences of lower concentrations (< 4 pmmv). The mean concentrations, as well as the range of distribution, increase during the monsoon months with the highest concentrations occurring during July and August. Clearly, the total accumulation is dominated by the monsoon months. The total water vapor accumulation during the analysis period (2005–2016) for each month separately was found to be 1440 ppmv for May, 1615 ppmv for June, 1937 ppmv for July and 2081 ppmv for August. It is noted that the cumulative water vapor accumulation during the JJA months accounts for about 80% of the total accumulation during MJJA. In other words, 80% of the total MJJA water vapor accumulation is observed during the JJA core summer monsoon season, while nearly 20% is observed during the pre-monsoon month of May.

Fig. 5

Frequency distribution of the reconstructed water vapor time series for each month separately: a May, b June, c July, and d August months

Using the frequency distribution for MJJA period, we identified the dates corresponding to the cases that occur within one standard deviation for the two peaks A and B (see Table 1). By so doing, we identified a total of 284 cases for the mode A and 233 cases for the mode B, which are given in Table 1. It is found that a majority of cases for the mode A occurred in the months of July and August during the peak summer monsoon season, whereas the events associated with the mode B were mostly observed during the late pre-monsoon (May) and early summer monsoon (June) months (see Table 1). Such distinction is expected as discussed in the frequency distribution of individual months already (Fig. 5). Based on the above analysis, the bimodal-like appearance of the frequency distribution for MJJA is attributed- to a systematic transition from the pre-monsoon localized convective activity to more organized convection during the summer monsoon months, rather than due to stochastic processes.

Table 1 List of cases in mode A and mode B

To gain further insights about the nature of convective activity during the core summer monsoon (JJA) season and the pre-monsoon (May) month, we examine the characteristics of precipitation activity based on the information about rain-type and rain-rate from the TRMM precipitation radar (3A25) (Romatschke and Houze 2011). During the core summer monsoon season (JJA), it is seen that stratiform precipitation tends to be substantial over the mountainous regions of the Himalayas, coastal areas of western India, central-north India, Myanmar, Bay of Bengal, Indo-China and Southeast Asia (Fig. 6a, c). On the other hand, the pre-monsoon month of May is mostly characterized by localized deep convective precipitation over the Myanmar coast, southeast Bay of Bengal, the Bengal plains, Indo-China and adjoining Southeast Asia (Fig. 6d). Further, it is seen that the climatological stratiform rain-rate is rather low over these areas during the pre-monsoon month of May (Fig. 6b). In short, it is inferred from Fig. 6 that organized stratiform precipitation over South and Southeast Asia is dominant during the core summer monsoon (JJA) season, whereas localized deep convection prevails over Myanmar coast and adjoining Southeast Asia during the late pre-monsoon month of May. It is important to note that convection is the primary source for these rains. Monsoon cumulus convection, composed of stratiform-type precipitation together with embedded deep convective cells, organizes over large spatial scales which contain an evolving pattern of newer (convective) and older (stratiform) precipitation where both occur within the same complex of convection generated cloud systems (Houze 1997). Furthermore, monsoon synoptic environments (e.g., lows and depressions) over south Asia have substantial contribution from stratiform precipitation, wherein updrafts are forced by large-scale vertical ascent of the monsoon disturbances (e.g., Choudhury and Krishnan 2011; Romatschke and Houze 2011).

Fig. 6

Climatological distribution of rain type and rain rate (source: TRMM-PR 3A25; units: mm day⁻¹): a June–July–August stratiform rain, b May stratiform rain, c difference between stratiform rain during June–July–August and May, and d May convective rain

3.3 Spatial distribution of water vapor and circulation

In the following analysis, we shall examine the connection among the UT/LS water vapor distribution, Tibetan anticyclone and monsoon convection, based on composite analysis of water vapor, circulation and precipitation patterns associated with the two modes. The composite maps are constructed based on the 284 cases for the mode A and 233 cases for the mode B shown in Table 1.

The MLS based composite maps of water vapor at 100 hPa for the mode A and mode B are shown in Fig. 7a, b, respectively. The pattern associated with mode A shows water vapor amounts in excess of 4.9 ppmv over large areas of the Asian continent and has resemblance with the climatological water vapor distribution (Fig. 2a). In contrast, the water vapor distribution for mode B is rather confined over the Southeast Asia. Clearly, zonally widespread and increased water vapor mixing ratios are seen in mode A as compared to mode B. The spatial distribution of water vapor for the two modes can be understood based on the composite maps of geopotential height and horizontal winds at 100 hPa (Fig. 8a, b). The circulation composite for the mode A shows a strong anticyclonic circulation centered around 70° E, 30° N and has a zonal extent (east–west) between 20° W and 170° E, while the meridional extent of the circulation spans from 15° N to 45° N. The 100 hPa wind composite for the mode A shows westerlies to the north of 35° N, and a strong tropical easterly jet over the low-latitudes and extending northward close to 30° N. In contrast the anticyclonic circulation for the mode B, which is centered around 85° E, 25° N, is weaker in intensity and has a smaller spatial extent compared to that of mode A. Note that the anticyclonic circulation for mode B is confined longitudinally between 30° E to 140° E and meridionally between 15° N to 35° N. Also note that the subtropical westerlies for the mode B extend to the south of 30° N over the Indian subcontinent, while they dip equatorwards to about 20° N to the west of the African continent and to the Far East. Additionally, the strength of the tropical easterlies at 100 hPa for mode B is relatively weaker, as compared to mode A. In short, the Tibetan anticyclonic circulation at 100 hPa associated with the mode A is stronger in intensity, has a significantly larger spatial extent and exhibits a distinct westward shift of the circulation centre by about 15°, as compared to that of the mode B.

Fig. 7

Composites of 100 hPa water vapor mixing ratios (ppmv) derived from the May–August period in 2005–2016 from MLS for a mode A and b mode B

Fig. 8

Similar to Fig. 6, composites of winds (vector, maximum = 50 m s⁻¹) and geopotential heights (solid contours; units in meters) for a mode A and b mode B

It is known from earlier studies that the large-scale tropical atmospheric response to off-equatorial monsoon-type heating produces a stationary Rossby response to the west of the heating (Gill 1980; Hoskins and Rodwell 1995). In this case, the forced large-scale circulation response in the upper troposphere manifests as an anticyclone that can be explained as a balance between the advection of potential vorticity and vorticity stretching—the so-called Sverdrup balance (Vallis 2017; Siu and Bowman 2019). In the following discussions, we will examine the prominent westward displacement of the Tibetan anticyclone for the mode A and its association with organized monsoon convection. A composite analysis of precipitation, outgoing longwave radiation (OLR) and vertical velocity fields associated with the two modes is presented in the Fig. 9. It may be noted that OLR serves as a proxy for tropical convection given that low values of OLR over the tropics and monsoon regions tend to be associated with deep moist convection, whereas high OLR values indicate scarcity of cloud cover (Krishnan et al. 2000). The precipitation composite for the mode A in Fig. 8a shows wide-spread monsoon precipitation extending from the west coast of India eastward into north-central India, Bay of Bengal (BoB), Myanmar and further into the Philippine region. It can be noticed that the region of high precipitation exceeding 25 mm day⁻¹ over BoB and Myanmar is associated with low OLR < 190 W m⁻² (Fig. 9a). In contrast, the precipitation pattern for the mode B is not as widespread and organized as compared to that of the mode A, instead it appears to be localized over smaller areas of northeast India and Southeast Asia (Fig. 9b). Low precipitation over central and northwest India for the mode B is associated with high OLR values (> 230 W m⁻²) indicating subdued convective activity over the region (Fig. 9b). The longitude-pressure cross-sections of ERA-Interim based vertical p-velocity (ω; Pa s⁻¹; multiplied by 100) averaged over 10° N–30° N for the two modes are shown in Fig. 9c, d, respectively. In the case of the mode A, significant organized ascending motion (ω < − 0.02 Pa s⁻¹, shaded region) can be seen extending from 60° E to 130° E across the South and Southeast Asian monsoon region (Fig. 9c). On the other hand, the vertical motions are relatively weaker and confined over smaller domains between 80° E–100° E and 120° E–130° E for the mode B (Fig. 9d).

Fig. 9

Composites of a, b TRMM precipitation (mm day⁻¹; shaded) and NOAA-OLR (W m⁻² blue contours), c, d longitude pressure cross section of vertical p-velocity (units in Pa s⁻¹; multiplied by 100; values less than − 0.02 Pa s⁻¹ are shaded) averaged over the latitude belt 10° N–30° N. The left and right panels refer to the mode A (a, c) and mode B (b, d) respectively during the May–August period averaged over 2005–2016

We examined the vertical profiles of geopotential height anomaly averaged over the region (15° N–25° N, 70° E–100° E) of the South Asian monsoon trough (see Choudhury and Krishnan 2011) for the two modes (Fig. 10). To highlight the vertical structure of the regional monsoon circulation, the zonal mean values have been removed from the total geopotential height field while constructing the vertical profiles shown in Fig. 10. The vertical profile for the mode A shows negative geopotential height anomalies from the lower troposphere up to nearly 350 hPa, indicating an intensified monsoon low (cyclonic circulation) in the lower and mid-tropospheric levels. It is also important to note that the mode A is associated with positive values of geopotential height anomaly from 350 to 70 hPa that correspond to an intensified high (anticyclonic circulation) in the UT/LS. This feature of strengthened low in the lower-to-mid troposphere over the monsoon trough region and an intensified anticyclone aloft is a typical feature during the peak phase of the Indian summer monsoon (e.g., Krishnamurti et al. 2008; Choudhury and Krishnan 2011; Choudhury et al. 2018) associated with organized monsoon convection (Sikka and Gadgil 1980; Schneider et al. 2014). On the other hand, the profile of geopotential height anomaly for the mode B shows a weaker and shallower atmospheric low (cyclonic) extending from surface to 400 hPa in the troposphere, as well as a weaker high (anticyclonic) in the UT/LS.

Fig. 10

Composite vertical structure of geopotential height anomalies (zonal mean removed) averaged over the monsoon trough region (15° N–25° N, 70° E–100° E) corresponding to periods associated with two modes

In short, it is seen that the dynamical features associated with the mode A correspond to a strong and spatially extended anticyclonic circulation in the UT/LS together with an intensified and vertically deep low (cyclonic anomaly) in the lower and mid-tropospheric levels and accompanied by widespread monsoon precipitation, organized convection and strong ascending motions across South and Southeast Asia. In contrast, the mode B is characterized by weak monsoon and subdued convection.

3.4 Case of organized monsoon convection during August 2016

To strengthen our argument on the role of organized convection on the UT/LS water vapor distribution, we present an example of a mode A event on 3 August 2016 for which in situ measurements are available from Nainital (see Table 1). The vertical profiles of water vapor from the CFH and MLS measurements on 3 August 2016 are shown in Fig. 11a. Although 30 soundings were conducted from Nainital during August 2016, we note from Table 1 that there are only 5 cases in August 2016 which were associated with the mode A. Out of the 5 cases, the water vapor profiles on 5 and 15 August had data discontinuities in the UT/LS (see supplementary Fig. S2), while soundings were not performed for the other two dates (i.e., 4 August and 13 August). The sounding on 3 August 2016 (Fig. 11a) shows water vapor mixing ratios exceeding the peak value of 5 ppmv of mode-A in the UT/LS region. Higher water vapor mixing ratios exceeding 7 ppmv can be seen between 100 and 80 hPa in the CFH measurements (Fig. 11a). The MLS estimates over Nainital on 3 August 2016 also indicated LS water vapor mixing ratios in excess of 5 ppmv.

Fig. 11

Observed patterns during August 3, 2016: a MLS (red line) and CFH (black line) observed profiles of water vapor mixing ratios (units in ppmv), b MLS water vapor mixing ratios at 100 hPa (units in ppmv), c 100 hPa winds (vector, maximum = 50 m s⁻¹) and geopotential heights (solid contours; units in meters), and d TRMM precipitation (mm day⁻¹; shaded) and NOAA-OLR (W m⁻² blue contours)

Spatial maps of water vapor, geopotential height and winds at 100 hPa on 3 August 2016 are shown in Fig. 11b, c. The corresponding maps of precipitation and OLR are shown in Fig. 11d. High water vapor mixing ratios can be seen covering a wide region of northern India and Tibetan Plateau extending eastward to southeast China (Fig. 11b). This zonally extended water vapor distribution in the LS is accompanied by a large-scale anticyclonic circulation with a wavy subtropical jet stream and intense tropical easterly jet stream along its northern and southern flanks, respectively (Fig. 11c). The precipitation pattern also displays a substantially larger zonal extent with rainfall magnitudes as high as 50 mm day⁻¹ over areas in South and Southeast Asia. The OLR field also shows large-scale organized monsoon convection with low OLR values (< 190 W m⁻²) between 65° and 110° E over the south and Southeast monsoon Asian region (Fig. 11d). This analysis highlights the linkage among organized monsoon convection, UT/LS anticyclone and water vapor distribution over the ASM region.

4 UT/LS circulation response to monsoon organized convection

We further supplement the diagnostic analysis by performing numerical simulation experiments using a simplified atmospheric model that enables us to better understand the circulation response of the UT/LS to monsoon convection associated with the two modes. We also analyze and discuss the large scale tropical conditions and transport of water vapor flux for the two modes as seen from the observations in this section.

4.1 GCM simulated response

Monsoon precipitation and organized convection are accompanied by substantial latent heating which is fundamental to the dynamics of the Asian monsoon circulation (e.g., Choudhury and Krishnan 2011; Houze et al. 2015; Shige and Kummerow 2016). In the numerical simulation experiments, we investigate the forced equilibrium response of the atmosphere to prescribed latent heating based on the TRMM satellite observations. The model uses linear damping of momentum and temperature in the form of Rayleigh friction and Newtonian cooling, respectively, both having an e-folding decay time-scale of 5 days (Krishnan and Kasture 1996; Sundaram et al. 2010). The model integration is performed without changing the heating and linear damping terms during the course of the integration, until it reaches a steady-state. It is seen that the simulation attains a near steady-state by the end of 100 days, which is taken as the equilibrium response.

In the experiments, the model is forced by three-dimensional latent heating due to precipitation which is constructed using surface rain estimates, stratiform/convective fractions from the TRMM PR 3A25 rainfall dataset and through a combination of vertical profiles of stratiform and convective precipitation based on Schumacher et al. (2004). The details of constructing the latent heating are given in Choudhury and Krishnan (2011). In the control experiment (CTRL), the model is forced by the climatological latent heating for the May–June–July–August (MJJA) season. The vertically integrated climatological mean heating shows maxima over the West coast of India, Bay of Bengal (BoB), Myanmar and a southeastward tilted band of heating extending from north and central India towards South China Sea, Philippines and tropical west Pacific (Fig. 12a). The vertical-zonal cross-section of latent heating averaged latitudinally between 10° N and 30° N, show that the level of maximum heating over South and Southeast Asia is located around 400 hPa (Fig. 12b). The maximum heating around 90° E over the BoB and adjoining areas is characterized by organized mesoscale convective systems with higher stratiform rain fractions which gives rise to a top-heavy vertical structure of latent heating (see Houze 1997; Romatschke and Houze 2011; Choudhury and Krishnan 2011). The mean circulation response at 100 hPa from the CTRL experiment shows a well-defined Tibetan anticyclonic (high) circulation that extends zonally across the Asian continent and further westward over Africa; and is flanked by tropical easterlies on the southern side and subtropical westerlies to the north of 30° N (Fig. 12c). The large-scale aspects of the mean UT/LS circulation response to latent heating in the CTRL experiment (Fig. 12c) are consistent with the observed features during the boreal summer monsoon season (Krishnamurti et al. 2008). For additional details of the forced response of the mean monsoon circulation in the lower and middle troposphere, the reader is referred to Choudhury and Krishnan (2011). Following the CTRL experiment, two sensitivity experiments were performed to understand the UT/LS circulation response to latent heating anomalies associated with the two modes. The latent heating for the two cases was constructed as follows. First the surface rain estimates for the two modes were obtained from the corresponding TRMM 3B42 rainfall anomalies, which were then superposed on the PR 3A25 climatological mean rainfall. In order to construct the vertical structure of latent heating for the two sensitivity experiments, it is assumed that the stratiform and convective fractions are the same as in the CTRL experiment. This implies that the vertical structure of latent heating remains the same for the CTRL and two sensitivity experiments, while the magnitude of the vertically integrated heating varies according to the corresponding surface rain estimate.

Fig. 12

Climatological (May–August) latent heating (K day⁻¹) distribution (source: TRMM-PR 3A25): a Column averaged (700–200 hPa), b longitude-pressure cross section (averaged over 10° N–30° N). c Simulated 100 hPa horizontal winds (vector, maximum = 15 m s⁻¹) and streamfunction (shaded; in units 10⁵ m² s⁻¹) in response to 3-dimensional LH shown in (a) and (b)

As seen from the previous analysis, mode B corresponds to pre-monsoon conditions wherein relatively high OLR values and minimal precipitation prevail over the south Asian region, with precipitation being mostly localized over northeast India and Southeast Asia (Fig. 9). The pre-monsoon convective precipitation mostly occurs from thunderstorm activity and the latent heating peaks close to the midtroposphere (see Choudhury and Krishnan 2011). The mode A which corresponds to organized monsoon conditions shows low OLR and wide-spread monsoon precipitation over south Asia and the Bay of Bengal (BoB), when the ITCZ moves northwards and reaches to highest latitudes over the Indian subcontinent. The precipitation during this time has significant contributions from the stratiform precipitation which is characterized by an elevated heating beyond the midtropospheric levels. Clearly the region of increased latent heating associated with the mode A shifts northwards. As the strength of the Tibetan anticyclone is significantly governed by the latent heat release during the monsoon cloud-precipitation processes, it is understandable that the zonal axis of anticyclonic circulation in Fig. 8 (water vapor distribution in Fig. 7) shifts northward for mode A. We next discuss the model UT/LS response in the two sensitivity experiments, corresponding to mode A and mode B.

Vertical cross-sections of the zonal and meridional components of circulation response in the UT/LS from the two sensitivity experiments are shown in Fig. 13. The latitude-pressure cross-sections for the zonal component of wind in Fig. 13a, c are based on the zonal wind averaged over the 60° E–120° E longitude belt. It can be seen that tropical easterlies in the UT/LS, as well as the subtropical and mid-latitude westerlies are substantially (nearly twice) stronger for the mode A (Fig. 13a) as compared to the mode B (Fig. 13c). It is also important to note that the meridional gradient of the zonal wind is stronger for the mode A, and the vertical placement of the core of the tropical easterlies is at a higher altitude for the mode A as compared to the mode B. Figure 13b, d show the longitude-pressure cross-sections of the meridional component of wind averaged between 10° N and 30° N. It can be seen that the meridional wind response for the mode A is stronger in magnitude and has a larger spatial scale, as compared to that of the mode B. In particular the spatial extent of the southerly winds in the UT/LS, on the westward side of the monsoon heating maximum around 90° E, is very prominent in the case of the mode A and is weaker for the mode B. Also the northerlies on the eastern side of the upper tropospheric anticyclone are stronger in magnitude for the mode A relative to the mode B. The aforementioned features of the zonal and meridional winds in the UT/LS have been noted from reanalysis datasets (Siu and Bowman 2019). The larger westward extent of the southerly wind response (Fig. 13b) for the mode A is particularly interesting. Our understanding suggests that this feature associated with the monsoon anticyclone in the UT/LS can be interpreted as dispersion of stationary Rossby waves to the west of the Asian monsoon heating (Hoskins and Rodwell 1995; Krishnan et al. 2009).

Fig. 13

Simulated vertical cross-sections of a, c zonal and b, d meridional winds (m s⁻¹) in the UT/LS region in association with mode A (top panel) and mode B (bottom panel). The cross sections are averaged over the longitude belt 60° E–120° E in (a, c), and over the latitude belt 10° N–30° N in (b, d)

Figure 14 shows the horizontal circulation simulated response at 100 hPa for the two modes. The streamfunction and horizontal winds for the mode A (Fig. 14a) shows a robust anticyclonic circulation centered over the Indian subcontinent, which extends westward over the African continent up to the 30° E longitude and up to 120° E on the eastern side. The circulation response for the mode B shows a weaker anticyclone and with the core of anticyclone located relatively eastwards (Fig. 14b). While the meridional scale of the simulated anticyclonic response in the UT/LS is also relatively larger in the case of the mode A relative to the mode B, it is important to highlight that the anticyclonic response extends far westward (about 15° longitudes) as compared to the mode B. While these features capture the effect of monsoon convection on the UT/LS circulation as seen through the model simulations, we also realize that there is a smaller northward shift in the zonal axis of the simulated anticyclone in mode A, relative to observations. Given that organized monsoon convection extends more westward and northward over the Indian subcontinent in mode A relative to mode B (Fig. 9), the model simulated response broadly captures the larger westward extent of the upper tropospheric anticyclonic circulation in mode A as compared to mode B, although there are differences between the simulated and observed circulation patterns (Fig. 8, 14). It is important to recognize that this analysis is based on a simplified atmospheric model forced only by latent heating, wherein the magnitude of heating varies in accordance with the surface rain estimates. However, accurate simulation of the UT/LS water vapor for the two modes requires incorporation of physical parameterization of interactive moist convective processes in the model, which is beyond the scope of this study.

Fig. 14

Simulated 100 hPa circulation response to latent heating for a mode A and b mode B. Shown are horizontal winds (vector, maximum = 15 m s⁻¹) and streamfunction (shaded; in units 10⁵ m² s⁻¹)

4.2 Water vapor flux divergence and transport

For the two modes A and B, we discussed the connection between convection (Fig. 9) and the UT/LS circulation (Figs. 8, 14) using observations and model simulations. A clear understanding of the physical link between convection and UT/LS water vapor distribution (Fig. 7) is important to explain the coexistence of the patterns of convection, circulation and water vapor distribution. To substantiate the influence of convection on the UT/LS water vapor transport for the two modes, we analyzed the observed patterns of divergence of water vapor flux vector for mode A and mode B, as well as their divergent and rotational components at 100 hPa. Figure 15 shows the 100 hPa divergence of water vapor flux for the two modes. It is seen that the magnitude of divergence of water vapor flux is more than twice stronger and spatially widespread for mode A as compared to mode B. Large positive values (> 0.5 ppmv day⁻¹) of divergence of water vapor flux for mode A can be seen extending to the west of 90° E over the Indian subcontinent and north of 45° N, while the divergence of water vapor flux for mode B is substantially weaker. Additionally, we analyzed the water vapor transports for the two modes by separating them into divergent and rotational components using Helmholtz decomposition (Krishnamurti 1971; Krishnan 1998; Behera et al. 1999; Trenberth et al. 2000). This approach makes it possible to express small-scale regional features of divergence/vorticity in terms of planetary-scale fields of velocity potential/streamfunction (Krishnamurti 1971).

Fig. 15

Divergence of water vapor flux (ppmv day⁻¹) at 100 hPa for a mode A and b mode B

The transport of water vapor by the rotational component and the corresponding streamfunction for modes A and B are shown in Fig. 16a, b, respectively. The divergent component of the water vapor transport and the corresponding velocity potential for modes A and B are shown in Fig. 16c, d, respectively. The divergent component shows stronger and widespread outflow of moisture for mode A, as compared to mode B, in the UT/LS (Fig. 16c, d). Additionally, the water vapor flux transport by rotational winds over the ASM region is substantially stronger with a much larger spatial extent for mode A as compared to mode B (Fig. 16a, b). Clearly, it can be seen that a strengthening of the divergence core of water vapor flux in response to organized monsoon convection results in a large-scale westward and northwestward transport of water vapor in the UT/LS (Fig. 16a, c).

Fig. 16.

100 hPa spatial patterns denoting a, b streamfunction of water vapor flux (shaded; in units 10⁶ ppmv m² s⁻¹) overlaid with water vapor flux by rotational wind component (vector, maximum = 140 ppmv m s⁻¹), and c, d velocity potential of water vapor flux (shaded; in units 10⁶ ppmv m² s⁻¹) overlaid with water vapor flux by divergent component (vector, maximum = 60 ppmv m s⁻¹). The left and right panels refer to the mode A (a, c) and mode B (b, d) respectively

5 Concluding remarks

In this study, we have analyzed the observed water vapor variability inside the Tibetan anticyclone using in situ balloon-borne measurements from a high-elevation site (Nainital) in northern India and MLS v4.2 based satellite-retrieved daily water vapor data for the period 2005–2016. Additionally, our analysis employs precipitation products from the TRMM satellite, including estimates of rain rate and rain type, reanalysis circulation products and numerical model simulations to investigate the links between organized monsoon convection and spatial distribution of water vapor in the UT/LS region of the ASM. First we evaluated the MLS derived vertical profiles of water vapor against in situ balloon-borne CFH measurements. We find that the MLS estimates of water vapor mixing ratios inside the Asian summer monsoon anticyclone compare reasonably well with those obtained from the high precision hygrometer balloon observations (root mean square difference of 0.9 ppmv at 100 hPa). Particularly, MLS measurements fall closer to CFH measurements as we move up in the stratosphere. During the sounding period instances of increased water vapor mixing ratios in the UT/LS are noted, which highlight the monsoonal influence. Motivated by good agreement between the MLS and in situ balloon observations, we analyzed the variability of MLS based UT/LS water vapor averaged over the core of water vapor maximum i.e. the Tibetan Plateau and the southern slope region (25°–40° N, 70°–105° E). We performed multivariate linear regression analysis to isolate and remove the effects of global scale climate drivers e.g. ENSO, BDC, and QBO from the LS water vapor variability over the ASM region. Our analysis shows that UT/LS water vapor variability over the ASM region is partly explained (~ 38%) by these global scale climate drivers, however much of the regional water vapor variability (~ 62%) is inherent to the subseasonal monsoon processes.

The findings from our study show that the LS water vapor distribution is closely linked to the spatial scale of organization of the South Asian monsoon convection and its influence on the UT/LS circulation. While a frequency distribution analysis of LS water vapor shows a bimodal-like appearance with peaks at 5 ppmv (mode A) and 4 ppmv (mode B), it is inferred that the two peaks mostly result due to systematic transition in convective activity between pre-monsoon to monsoon season and it is equivalent to temporal separation between the two peaks. Further, we note that 80% of the LS water vapor accumulation over the ASM region is contributed during the core monsoon months (JJA), while the remaining 20% contribution co-occurs with pre-monsoon localized deep convective activities.

It is revealed from our analysis that organized summer monsoon convective activities over the Indian subcontinent, Bay of Bengal and Southeast Asia promote widespread distribution of water vapor in the LS by enhancing the water vapor flux divergence in the UT/LS and associated large-scale transport by the Tibetan anticyclone with substantial zonal extent. On the other hand, the circulation response and LS water vapor distribution to pre-monsoon localized deep convection tend to have a limited spatial scale confined to Southeast Asia. The robustness of the water vapor flux divergence and transport by the Tibetan anticyclone in mode A as compared to mode B is also confirmed through an analysis of the observed UT/LS water vapor transport by the rotational and divergent components. Additionally, we performed numerical simulations to strengthen our results on the UT/LS circulation response to convection associated with modes A and B. It is seen that latent heating associated with the organized monsoon convection over the Indian subcontinent and Southeast Asia drives a stronger anticyclone in the UT/LS for the mode A as compared to that of mode B. Further, it is noted that the simulated anticyclonic response in the UT/LS extends far to the west of the monsoonal heating over the Indian subcontinent, with the westward extent enhanced by about 15° longitudes for the mode A relative to the mode B. This westward extending anticyclonic response in the UT/LS to organized monsoon convection has similarities to stationary Rossby waves forced by organized latent heating.

Appendix A: Statistics of distribution and interpretation of bimodal-like appearance

The statistics of the distribution and robustness of separation between the two peaks are quantified in this section using statistical tests and finite mixture modeling as a tool. We perform a ‘dip test’ through Hartigans' dip test which is a standard test for multimodality and describes the departure of any sample from unimodality (Hartigan and Hartigan 1985). Further, we use Gaussian finite mixture modeling for estimation of mixing density and it is found that the mixing probabilities of mode A and mode B are 0.54 and 0.46, respectively. The bimodality separation is found to be 0.65 (see also Zhang et al. 2003). In addition, we also check Ashman’s D statistic which provides a robust measure of differentiation between the two distributions (Ashman et al. 1994). For a given dataset, if the Ashman’s D statistic is above ~ 2, it is understood to have good separation. The Ashman’s D statistic for the present analysis is noted to be 2.3 which indicate significant separation between the two peaks. However, it should be noted that the bimodal-like appearance discussed here is associated with the systematic transition in convective activity between pre-monsoon to monsoon season and it is equivalent to temporal separation between the two peaks. Therefore the observed statistical separation is a result on appearance and it is almost equivalent to separation of pre-monsoon and monsoon cases. Statistical summary of the frequency distribution is given in the following Table 2.

Table 2 Statistics of bimodal-like appearance of LS water vapor over the Asian summer monsoon region