Stratospheric water vapor (SWV) is a potent greenhouse gas with a significant radiative forcing . In the tropical lower stratosphere, SWV values are determined primarily by the freeze-drying of moist tropospheric air entering the stratosphere at the cold-point tropopause (CPT) . The extent of ice injected into the stratosphere through deep, overshooting convection remains uncertain . Specifically, in the context of the Asian summer monsoon (ASM) region, recent in situ measurements suggest moistening due to convective activity above the local tropopause , while satellite observations indicate a broader-scale drying effect caused by convection .

Lagrangian studies commonly reconstruct SWV by tracking the minimum saturation mixing ratio of air at the Lagrangian dry point (LDP) . We combine airborne in situ measurements during the StratoClim campaign in Nepal with satellite observations of MLS and CALIPSO to investigate the representativeness of the moist air masses encountered above the CPT in order to reconstruct the large-scale SWV distribution. Using a microphysical model along forward trajectories, we address two main questions. First, how does the water vapor content of these air masses change during their ascent into the stratosphere? Second, how representative are these air masses in order to reconstruct the large-scale moisture budget of the lower stratosphere? Finally, we discuss the performance of LDP-based SWV reconstructions for these examples of moistening above the local CPT.

In this study, we utilize in situ data collected during all local Geophysica flights over Nepal in 2017, combined with merged back-and-forward trajectories driven by the ERA5 reanalysis (, Appendix ) to evaluate the influence of these air masses on SWV values. In Fig. a, two thick black lines define data points that are “sufficiently moist” (H2O > 7 ppm) and “sufficiently deep in the stratosphere” (O3>100 ppb). Our data set is reduced to 2315 data points by only considering those observed above the local CPT defined as the temperature minimum in the ERA5 temperature profiles interpolated to the Geophysica flight track. We denote the data with recent convective influence as type A (observed on 10 August) and with aged convective influence as type B (observed on 29 July). The data observed on 8 August (less than 5 %) contain mixed properties of types A and B and are labeled as type M (see Table ).

Figure b shows CO values for all three data sets as a function of vertical distance to the CPT. CO decreases with altitude in slowly ascending air within the ASM anticyclone due to its chemical lifetime of a few months . This decrease can be seen for type B and M data, extending up to 1.7 km above the tropopause. Type A data are closer to the CPT and show fresh convection signatures, with a spread of CO values between 30 and 95 ppb, a large spread of HDO/H2O ratios (Fig. c), and positive ice–CO correlations (Fig. d). In Fig. c, the distribution of the LDP ages derived from merged back-and-forward ERA5 trajectories within ±60 d is presented. Negative (positive) age denotes past (future) occurrence relative to the observation time. Consistent with our interpretation of recent and aged convection signatures, LDP ages range between -15 and -35 d for type B. For type A, more than 75 % of the LDP encounters are still expected to occur along the forward trajectories, with LDP ages of 0–3 d, despite these air parcels having been sampled above the local CPT.

We use forward trajectories starting from the locations of the observed values of water vapor and ice for all types of air (A and B). Along these trajectories, we apply a state-of-the-art microphysical box model, CLaMS-Ice (Appendix ), as well as a simple freeze-drying model (FDM) that instantaneously removes excess water vapor along the forward trajectories when the air becomes supersaturated with respect to ice. Figure a shows two exemplary forward trajectories (types A and B) that slowly ascend within the ASM anticyclone above the level of zero radiative heating with a rotation period of about 10 d . All trajectories of type A and more than 85 % of type B stay within the tropical band extending northward during the boreal summer up to ∼ 40∘ N. Only ∼ 15 % of the type B trajectories descend into the lowermost stratosphere (LMS) northward of ∼ 40∘ N (Fig. a) after being detached from the anticyclone. During their spiraling ascent, almost all trajectories repeatedly pass through regions with low temperatures, well below 195 K, mainly on the southeastern, southern, and southwestern flank of the anticyclone, where water condensation and ice formation can occur. CLaMS-Ice and a simple freeze-drying model (FDM) were used to simulate the trajectories and investigate dehydration scenarios. Figure c compares the models' results for one trajectory of type A with ice and water vapor observations from CALIPSO and MLS, respectively. CLaMS-Ice reproduces the CALIPSO ice observations fairly well, while FDM performs better in terms of water vapor comparison with MLS.

We extend the analysis to all types A and B forward trajectories initialized with in situ observations of H2O vapor and ice. Figure  displays the time-dependent frequency distribution at selected times for type A (left) and type B (right). Relative to the initial distribution, the distributions derived from CLaMS-Ice and FDM evolve over time by moving to significantly lower values of total H2O. The contribution of the dehydration driven by the parameterized gravity wave (GW) temperature fluctuations is very weak, as estimated from the small difference between the type B CLaMS-Ice distributions after 5 versus after 40 d (this difference vanishes if this parameterization is switched off). The 40 d distributions also show the impact of enhanced ice nucleation by using CLaMS-Ice-IN.

Despite the GW parameterization in CLaMS-Ice, the results of FDM are always drier than the results of CLaMS-Ice. This can be attributed to the different treatment of ice within the two models. In FDM, ice is removed instantaneously, while in CLaMS-Ice ice removal occurs through the interplay of condensation, evaporation, and sedimentation (see also Fig. c). A massive dehydration can be diagnosed for type A, affecting all air parcels, with mean (maximum) values after 40 d of 5.0 (11.3) ppmv (CLaMS-Ice) and 3.3 (4.1) ppmv (FDM). The degree of dehydration for type B is weaker, as there is less than 1 ppm of ice at the initialization time. The mean (maximum) values for type B after 40 d are 8.1 (9.8) ppmv (CLaMS-Ice) and 6.5 (7.9) ppmv (FDM). Only 14 % (CLaMS-Ice) and 1 % (FDM) of the initial observations did not experience any dehydration. The final positions of these non-dehydrated air parcels are within the LMS.

The dehydration scenarios for types A and B are consistent with the respective frequency distributions of LDP temperatures from back-and-forward trajectories (Fig. , the lower-right subpanels for both types A and B). The strong dehydration of type A air masses, detrained very recently from fresh convection, is mainly due to the LDPs being experienced in the forward direction rather than along the backward trajectories. On the other hand, most of the air masses of type B, detrained from convection several days to weeks before, have already experienced their lowest temperatures in the past (cf. ). However, even for type B air masses, a significant dehydration can still be expected in the future, well above the CPT, at the southern edge of the anticyclone during the upward spiraling motion of the forward trajectories, as seen from the shift in the frequency distributions to lower mixing ratios on the right-hand side of Fig. .

We compare the ice distribution calculated by CLaMS-Ice during the dehydration periods along the forward trajectories with CALIPSO observations, which infer ice mixing ratios larger than ∼ 0.1 ppm based on the parameterization of in situ data . Figure  displays the results, where the horizontal (a) and vertical (b) large-scale temperature distributions are given in grayscale and overlaid with the positions of the simulated and measured ice clouds, as well as the position of the ASM anticyclone (PV-based edge, with PV =3.7 PVU, and the mean easterlies and westerlies) in August 2017. The comparison shows that type A generates significantly more ice than type B and that type A agrees better with CALIPSO observations. Of the 442 type A ice events observed by CALIPSO, more than half are reproduced by CLaMS-Ice, while of the 132 CALIPSO ice events of type B, less than 6 % are simulated by CLaMS-Ice despite data set B being ∼5 times larger. The geographic positions of type A ice clouds are also better reproduced and are mainly found at 20∘ N between 17 and 18 km altitude, with the strongest signature over northern India. In contrast, type B ice simulated by CLaMS-Ice shows weaker signatures and a much larger horizontal spread, extending over the regions with coldest temperatures, mainly over Southeast Asia and the Maritime Continent. These signatures seem to be related to isentropic mixing driven by Rossby waves that are well characterized by bent PV isolines surrounding the anticyclone . There are a few weak CALIPSO signatures of ice in the LMS (type B) north of 35∘ N between 400 and 420 K, which are not resolved by CLaMS-Ice. However, the expected warm temperatures in this part of the atmosphere raise some doubts about the origin of these signatures.

We also validate the calculated water vapor with the MLS observations along the forward trajectories. The comparison of CLaMS-Ice and FDM with MLS data for the type A observations is quite good. However, there is a significant disagreement for the type B data, even when ice nucleation is enhanced in the model (CLaMS-Ice-IN). We also observe a similarly strong disagreement when using FDM. A weaker disagreement was found for the type M data, consistent with its mixed properties (30 % type A and 70 % type B). However, including dehydration also along the backward trajectories (FDM-full), i.e., at the LDP being in the past, as done in many previous studies , performs extremely well for all three data types (A, B, and M) particularly in the region above 390 K (see also Appendix ).

The mere existence of moist plumes over the CPT, which are also well above the lapse rate tropopause and therefore in the stratosphere, does not necessarily imply persistent stratospheric moistening , as significant dehydration events along the forward trajectories are still possible, particularly in the regionally confined anticyclonic Asian monsoon circulation. In other words, a 1-D view is misleading, as observation of a moist air mass above the CPT at a particular time does not imply that the moisture remains in the stratosphere. One aspect is simply time variability, but more important is the fact that the Lagrangian time history determines dehydration, which requires consideration of the 4 d temperature field (3 d space + time), not just the temperature field in a single profile (1 d space), or even the time-varying temperature field in a single profile (1 d space + time).

Based on our case study using StratoClim data, the importance of considering the full Lagrangian air mass pathway in both backward and forward direction is clear not only for type A air masses, where the absolute LDPs are still ahead, but also for type B air masses, where the absolute LDPs have already occurred a few weeks prior. Even for these cases, multiple and subsequent dehydration events at low temperatures, well below 195 K, can still occur during the upward spiraling in the ASM anticyclone. The highest ice concentrations are primarily located at the southern edge of the anticyclone, where the lowest ERA5 temperatures are observed (Fig. ). Our simulations with CLaMS-Ice reproduce CALIPSO ice observations well for type A air masses, but the agreement for type B is worse (Fig. a and b). Note that the absolute LDPs of 75 % of type A trajectories are in the future, with LDP ages ranging from 0 to 3 d. For type B trajectories, the LDPs have already occurred in the past, with ages ranging from -35 to -15 d. It is worth noting that our simplest approach, FDM, cannot reproduce any CALIPSO ice signatures, as ice is instantaneously formed and removed in this model setup.

To further evaluate the large-scale impact of the observed local hydration events, we compared different model results with MLS satellite observations. MLS has a significantly coarser vertical sampling resolution compared to CALIPSO, which can even resolve ice clouds extending over a few tens of meters. While the models exhibit good agreement with MLS observations for type A air masses, they consistently overestimate MLS, on average by 3–4 ppm for type B air masses when considering only forward trajectories (Fig. c2). Despite the significant dehydration events along the forward trajectories, the excess water vapor observed in situ persists and is not captured by MLS. This wet bias relative to MLS cannot be eliminated even by applying CLaMS-Ice-IN, which artificially enhances dehydration to its highest realistic limits. Only when assuming dehydration at the LDPs along backward trajectories, typically occurring 2–3 weeks prior to the observations, do the results of FDM-full align well with MLS water vapor measurements. Therefore, the small-scale moist plumes observed above the CPT, likely originating from convective overshoots, are not relevant for explaining the large-scale water vapor budget.

This significant result prompts critical remarks. Firstly, moisture-rich type B air masses were independently observed by the FISH and FLASH instruments, with differences below 10 %. Secondly, a potential warm temperature bias in the ERA5 data, estimated at approximately 1 K based on the differences with the observed temperatures along the flight track (Fig. b), may account for a maximum of 0.5 ppm of the diagnosed difference . Other factors such as the quality of the trajectories, the matching criteria used to identify MLS observations, or the precise definition of the bounding rectangle (Fig. a) are negligible (see Appendix ). Mixing may also play a role in smoothing out such moisture-rich structures. However, as no enhancement was diagnosed in the MLS observations, it would only support our statement that these structures are likely on small scales and not relevant for the large-scale water vapor budget. Finally, it is essential to note that our study is a single-case demonstration of the proposed mechanism's feasibility and seems particularly relevant for the Asian monsoon with its anticyclonic circulation regionally confining the ascending air masses. There are regions in the world, such as the American monsoon, where ice transport into the stratosphere could be more likely . Consequently, small-scale features may have a more substantial influence on the large-scale water vapor distribution in other regions. In such cases, additional case studies that follow our approach could provide further insights.

Despite these caveats, our results support some criticism related to the representativeness of local hydration events observed by in situ measurements. For example, this effect may influence the quantification of the SWV trends like those derived from the longest available record of the balloon-borne, in situ NOAA frost point hygrometer over Boulder . Our study shows that moist plumes can be sampled in the stratosphere that are not representative of the large-scale distributions of SWV. The fact that stratospheric satellite instruments capable of measuring SWV concentrations are approaching the end of their life time emphasizes the importance of setting up in situ observation networks (e.g., using stratospheric balloons) with regular and “statistically robust” sampling .

Finally, a few remarks are necessary regarding the performance of our most idealized modeling approach, reconstructing the SWV from the absolute LDP derived from full back-and-forward trajectories covering several weeks. The trajectory-based reconstruction propagates the minimum saturation mixing ratio encountered at the LDP into the full 3D space. As the quality of temperatures around the tropopause has improved over the last decades, particularly for ERA5 , the quality of the reconstructed SWV has also improved. Our results show that tropopause temperatures exert a dominant control over the tropical stratosphere in the ASM region , and this dominance seems to be more representative of global SWV values than sporadic observations of moist plumes in the stratosphere. However, we also found that the SWV reconstruction is not as effective for trajectories ending in the LMS (see Fig. d). In this region, the final values of SWV are controlled not only by LDPs but also by other processes such as mixing or downward transport of SWV affected by methane oxidation.