We use Microwave Limb Sounder MLS; version 5 (v5) gridded retrievals of water vapor , ozone , and carbon monoxide  at 2.5° ×2.5° spatial resolution from 2005–2021 covering the ASM region (30–130° E and 15–45° N). Daily gridded data are averaged into pentads (discrete 5 d periods) starting on 1 May and ending on 2 October, totaling 31 pentads per year (155 d). We focus on humidity in the monsoon tropopause layer between 147 and 68 hPa (approximately 370–440 K potential temperature).

In addition to MLS observations, we use model-level and pressure-level outputs from five atmospheric reanalyses. These include the fifth-generation European Centre for Medium-Range Weather Forecasts (ECMWF) atmospheric reanalysis ERA5;; the Japanese Reanalysis for Three Quarters of a Century JRA-3Q;; the Modern-Era Retrospective Analysis for Research and Applications, version 2 MERRA-2;; the MERRA-2 Stratospheric Composition Reanalysis of Aura Microwave Limb Sounder M2-SCREAM;; and the Copernicus Atmosphere Monitoring Service (CAMS) reanalysis of atmospheric composition . Among these, ERA5, JRA-3Q, and MERRA-2 are focused primarily on meteorological fields, while M2-SCREAM and CAMS are focused more on atmospheric composition. Accordingly, M2-SCREAM and CAMS feature more sophisticated chemistry and additional assimilated variables relative to the corresponding meteorological reanalyses MERRA-2 and ERA5, while the meteorological reanalyses provide a richer set of dynamical and thermodynamic diagnostics for mechanistic analysis. Key details of these five reanalysis systems, including model resolutions and a catalogue of how assimilated observations affect water vapor, ozone, and CO at these altitudes, have been provided by our companion paper .

We use these reanalysis products to support a deeper investigation into recurrent patterns of variability in LSWV, both to compare the consistency of internal variations to those observed by Aura MLS and to evaluate the potential mechanisms behind the observed variations. Three metrics are adopted to evaluate the effects of deep convection and thermodynamic conditions during transport. First, cold-point tropopause (CPT) temperatures are identified using instantaneous model-level temperature profiles from ERA5 and MERRA-2 and pressure-level temperature profiles from JRA-3Q as described by . Second, all-sky outgoing longwave radiation (OLR) fields from ERA5, MERRA-2, and JRA-3Q are used to indicate the prevalence of cold high clouds, in line with previous studies on variations in the overall frequency and spatial extent of deep convection in this region e.g.,. Variations in very deep convection are also evaluated using non-radiative diabatic heating, which primarily indicates the strength of convective ascent in the reanalysis model. Anomalous non-radiative heating at these levels is dominated by variations in the deepest simulated convective towers. Although such convection is rare, heating rates associated with these events are still approximately 1–2 orders of magnitude larger than contributions from parameterized shear-flow turbulence close to the tropopause . These heating rates are model products and are thus not directly constrained by data assimilation; however, they exert strong influences on composition and humidity in the reanalysis tropopause layer. Model-generated heating rates are only available for ERA5, MERRA-2, and JRA-3Q. Given the overlap in forecast models and, in the latter case, the use of the “replay” technique, heating rates for ERA5 may be taken as roughly representative of those in CAMS, and this is similar for MERRA-2 and M2-SCREAM.

Direct data assimilation constraints on humidity are weak at these levels . Different reanalyses deal with this lack of strong constraints in different ways. For example, the ECMWF reanalyses, ERA5 and CAMS, suppress direct assimilation increments in humidity at these levels by setting background error covariances to zero (although assimilation increments in winds and temperatures still influence the evolution of humidity fields at these altitudes). MERRA-2 only assimilates radiosonde humidities at pressures greater than 300 hPa, whereas JRA-3Q restricts radiosonde humidity assimilation to temperatures greater than -40 °C. Moreover, although satellite radiances can affect humidity at all altitudes in these two systems, small absolute concentrations mean that satellite radiances provide little information about water vapor at these altitudes. In addition, the deep vertical weighting functions of some assimilated satellite radiances may lead to data assimilation targeted at the troposphere producing correlated increments at and above the tropopause. This problem produced large biases in stratospheric humidity in JRA-55 , which persist in a less pronounced form in JRA-3Q . MERRA-2 sidesteps the issue by relaxing stratospheric humidity to a monthly-mean zonal-mean climatology ; however, this approach necessarily suppresses water vapor anomalies at these altitudes, especially those on longer timescales . M2-SCREAM is unique among the reanalyses considered here in that it directly assimilates Aura MLS retrievals of water vapor using the Goddard Earth Observing System (GEOS) Constituent Data Assimilation System CoDAS;. Accordingly, comparisons between M2-SCREAM and Aura MLS are not independent.

We also use several reanalysis-based variables to describe the dynamical state of the monsoon tropopause layer. First, we use the Montgomery streamfunction (MSF =cpT+Φ, where cp is the specific heat of dry air at constant pressure, T is temperature, and Φ is geopotential) on the 395 K isentropic surface to track the position and strength of the anticyclone . MSF, which is functionally identical to dry static energy, defines the geostrophic wind on an isentropic surface and is thus analogous to geopotential in pressure coordinates. The 395 K isentropic surface is typically located between 100 hPa and 83 hPa in the core ASM region south of the subtropical westerly jet. Second, we use horizontal winds, temperatures, ozone mass mixing ratios, and CO mass mixing ratios from each reanalysis to evaluate the dynamical transport environment within the ASM UTLS. Winds and temperatures are used to describe the structure of the anticyclone, while ozone and CO mass mixing ratios are used to qualitatively indicate anomalous in-mixing of stratospheric air and convectively detrained air, respectively e.g.,. Among the reanalyses, only CAMS and MERRA-2 provide estimates of CO and only CAMS includes any CO-related data assimilation . Finally, we use specific humidity tendencies due to parameterized physics, data assimilation, and dynamical transport from ERA5, JRA-3Q, and MERRA-2. These three meteorological reanalyses are used because they provide enough of the terms to close the vertically resolved water vapor budget. As with diabatic heating, tendencies from ERA5 may be considered roughly representative of those for CAMS; however, tendencies from MERRA-2 are not representative of those for M2-SCREAM because these two systems treat stratospheric water vapor very differently as noted above: MERRA-2 nudges to a climatology, while M2-SCREAM assimilates Aura MLS.

Motivated by the analysis of , who investigated the mechanisms behind water vapor variability at 100 hPa, we evaluate coupled horizontal and vertical variations in LSWV above the ASM in the 147–68 hPa layer, including observed and reanalysis water vapor mixing ratios. This layer, which contains the CPT, is a transitional zone where air detrained from convection spirals upward into the lower stratosphere e.g.,. We integrate variables in pressure from pbot (147 hPa) to ptop (68 hPa) and refer to the integrated variable as partial-column water vapor (PCWV) in units of mass per area: 1PCWV=-1g∫pbotptopqdp, where g is the gravitational acceleration, q is specific humidity, and dp denotes the pressure thickness of each layer between ptop (68 hPa) and pbot (147 hPa). Specific humidity from each reanalysis is interpolated from model levels (or pressure levels, in the case of JRA-3Q) to the MLS pressure levels, but no weighting functions are applied. Partial-column ozone (PCO3) and partial-column CO (PCCO) are calculated similarly from ozone and CO mass mixing ratios.

The vertically resolved water vapor budget is used to diagnose the contributions of parameterized physical processes, resolved dynamical transport, and data assimilation for determining the spatiotemporal variability in UTLS water vapor in the meteorological reanalyses: 2∂q∂t+∇⋅Vq+∂(ωq)∂p=Sphy+Sana+Sres, where p is pressure, V is the vector horizontal wind on an isobaric surface, and ω is the pressure vertical velocity. The first term on the left-hand side is the time rate of change, which is computed from assimilated water vapor fields at consecutive time steps. The second and third terms on the left-hand side are the horizontal and vertical moisture flux divergence, which are also computed from assimilated fields and referenced as the “dynamics” terms below. The terms on the right-hand side represent local sources or sinks. Sphy represents net sources and sinks due to parameterized physical processes in the forecast model (e.g., cloud microphysics, convection, and turbulent mixing), Sana is the source or sink due to data assimilation, and Sres is the residual. Subgrid-scale mixing is included in Sphy. The residual term collects high-frequency and/or small-scale covariance transports that are resolved by the reanalysis model but not by our calculation of moisture flux divergence . Additional details on the meaning and calculation of Sphy, Sdyn, and Sana have been provided by . A T42 spectral filter is applied to all budget terms to reduce noise in the spatial patterns. This filter simplifies the visualizations and does not change the results in any significant way.

For subsequent analyses, we average all variables into pentads (discrete 5 d means) starting on 1 May and ending on 2 October so that the first pentad of each warm season is centered on 3 May and the last is centered on 30 September. We calculate deseasonalized anomalies in each grid cell by subtracting the mean for the corresponding pentad over 2005–2021. We then apply empirical orthogonal function (EOF) analysis to the deseasonalized water vapor anomalies. The EOF analysis considers both horizontal and vertical variations at pentad resolution. Input data are weighted by area in the horizontal dimension but are weighted equally in the vertical dimension (i.e., the only weights applied to the input data are the horizontal grid area weights). We retain the three leading principal components (PCs) as indicated by Aura MLS. Eigenvalues associated with these three modes are mutually independent based on the North test .

Separate EOF analyses are then applied to water vapor volume mixing ratio anomalies from reanalysis products interpolated to the Aura MLS levels and downscaled to the MLS horizontal grid. The PCWV anomalies from each dataset are then regressed onto the corresponding PCs for intercomparison and evaluation of variability as represented by the reanalyses. Some PCs based on reanalysis fields must be reordered for consistency with the Aura MLS results; such cases are noted in the figures and related text. Deseasonalized anomalies in reanalysis fields are then regressed onto the PCs calculated from Aura MLS for a more detailed analysis of the underlying mechanisms. This two-pronged approach provides more stringent tests of the reanalysis products because it evaluates both the extent to which each reanalysis can capture observed spatial patterns of variability in LSWV (separate PCs) and the extent to which each reanalysis reproduces anomalies in the ASM and UTLS transport environment associated with a given mode (Aura MLS PCs). Combining the second step with the reanalysis-based water vapor budget expressed in Eq. () allows us to further assess process-level consistency across the reanalyses.

This section provides brief descriptions of the identified modes, which are analyzed in detail in the following sections. Figures  and show deseasonalized PCWV anomalies regressed onto leading principal components (PCs) of vertical and horizontal variability in tropopause-layer water vapor volume mixing ratios from Aura MLS, ERA5, CAMS, JRA-3Q, MERRA-2, and M2-SCREAM. The vertical structure of anomalies associated with each mode is discussed below. PC time series for all datasets are provided in Fig. . EOF results for reanalysis products are based on water vapor fields regridded to match the Aura MLS 2.5° × 2.5° grid and pressure levels but without applying the Aura MLS vertical weighting functions. The input data cover the ASM domain (30–130° E and 15–45°N) on the 147, 121, 100, 83, and 68 hPa isobaric levels at pentad temporal resolution.

We retain the three leading PCs from Aura MLS, which are mutually distinct at the 95 % confidence level according to the North test  and collectively account for 24 % of the variance in Aura MLS. The corresponding spatial patterns are identifiable in all reanalyses within the first four EOF modes (significant at 90 % or greater) and account for 43 % of the water vapor variance in ERA5, 45 % in CAMS, 48 % in JRA-3Q, 33 % in MERRA-2, and 44 % in M2-SCREAM. The first principal component (PC1; column a of Fig. ) accounts for about one-sixth to one-third of the variance in reanalysis water vapor in the ASM tropopause layer, with the smallest fraction of explained variance in MERRA-2 (16 %). PC1 primarily corresponds to interannual variability in the humidity of the lower stratosphere (Figs. S4, S6). The spatial pattern based on MERRA-2, with maximum variance in the southwestern quadrant of the anticyclone, is also considerably different from that identified in other reanalyses. MERRA-2 is a special case because it nudges stratospheric water vapor to a repeating annual cycle . Because the leading mode varies primarily on interannual and longer timescales (Fig. a), the nudging applied in the MERRA-2 lower stratosphere damps this mode in the upper part of our analysis domain.

Having no prior expectation of a substantial trend in water vapor in this region over this period, we did not detrend the input data. However, it is clear from Fig. a that PC1 comprises two main types of variability, a linear trend (PC1_TREND; Fig. b) and interannual variability (PC1_IAV; Fig. c), both of which project onto regional-scale wet or dry anomalies (Fig. ). We define the trend component as having a single value each year and constantly increasing from year to year (Fig. b). The spatial patterns associated with the trend component (Fig. b) represent the main discrepancy between water vapor variability over the ASM region as observed by Aura MLS versus that represented by the reanalyses during our analysis period. The slopes of linear trends in all of the PC1 time series are positive. However, whereas Aura MLS regressed onto the PC1_TREND time series indicates a drying trend over 2005–2021 centered in the southwest, the reanalyses show moistening centered in the southeast. This difference partially explains the larger signal and explained variance in PC1 based on M2-SCREAM, JRA-3Q, CAMS, and ERA5 relative to Aura MLS. We use version 5 (v5) of Aura MLS, in which the retrieval algorithm has been designed to mitigate a slow drift toward wet biases in the stratosphere in v4 . M2-SCREAM, which assimilated Aura MLS v4 throughout our analysis period, provides an indirect indication of the difference between v4 and v5 in this region . Consistency between M2-SCREAM and the other reanalysis products, which did not assimilate Aura MLS, thus raises the following question: could changes to the retrieval algorithm aimed at removing the spurious drift have also removed real regional trends? However, trends based on reanalysis products may not be physically meaningful and must be interpreted with caution e.g.,, especially over periods as short as this one, and recent work has shown that trends in stratospheric water vapor are difficult to distinguish from multi-decadal variability . Although the remarkable consistency across reanalysis products gives cause for careful consideration, we remain skeptical of the reanalysis-based trends shown in column (b) of Fig. . Further details are provided in Sect. .

The largest PCWV anomalies associated with the interannual mode (PC1_IAV) are located over East Asia and the area southwest of the Tibetan Plateau, with smaller anomalies over South Asia in the convective core of the monsoon (Fig. , right column). This spatial structure implies that interannual variability is stronger in the northeastern and southwestern sectors of the anticyclone i.e., the eddy shedding zones; and weak above and immediately downstream of the strongest convection. With the exception of MERRA-2 as discussed above, correlations between reanalysis-based PC1_IAV and that based on Aura MLS are high, with values ranging from 0.72 (CAMS) to 0.93 (M2-SCREAM) indicating that the other four reanalyses reproduce variations in this mode well. The phase of the interannual mode is typically set at the beginning of the monsoon season (Fig. c), suggesting the influence of external factors. In most years, the phase is then maintained through the monsoon season, suggesting either steady in-mixing of the anomaly from outside the anticyclone, maintenance by local feedbacks, strong confinement, or some combination of these. Likely candidates for sources of interannual variability outside the monsoon region include the El Niño–Southern Oscillation ENSO; and the Quasi-Biennial Oscillation QBO;. We evaluate these possibilities in Sect. .

The second principal component (PC2; column a of Fig. ) shows moistening or drying centered in the southwestern quadrant of the anticyclone, with the largest signal centered over the Persian Gulf. This mode explains 7 %–11 % of the variance in datasets other than MERRA-2, for which this mode corresponds to PC1 and explains 16 % of the variance (Fig. a_M). Correlations between PC2 from Aura MLS and the corresponding modes in each reanalysis range from 0.55 in MERRA-2 to 0.93 in M2-SCREAM (Fig. i), with JRA-3Q (r=0.78), CAMS (r=0.77), and ERA5 (r=0.83) all showing high correlations. Although this mode exhibits some variability at longer timescales, its fluctuations are mainly subseasonal.

The third principal component (PC3), which consists of a southwest–northeast dipole with opposing centers of water vapor anomalies over the Persian Gulf and the eastern Tibetan Plateau, explains slightly more than 3 % of the variance in Aura MLS (Fig. b). This mode corresponds to the fourth mode in M2-SCREAM (5 % of the variance; Fig. ) and the third mode in all other reanalyses (6 %–8 %). Variations in the corresponding PCs are mainly subseasonal and again show strong correlations with Aura MLS PC3, ranging from 0.59 in MERRA-2 to 0.80 in M2-SCREAM. PC2 and PC3 primarily reflect subseasonal variability below the tropopause (Figs. S5, S6). Spatial patterns show good correspondence overall, although the southwest-to-northeast tilt of the dipole is sharper in CAMS and ERA5 in comparison to the more east–west orientation in Aura MLS, MERRA-2, and M2-SCREAM. The dipole in JRA-3Q is broadly similar to the latter three but with a stronger signal on the equatorward side of the southwest pole.

Figure  shows diabatic heating anomalies along the 25° N transect and zonal-mean diabatic heating anomalies averaged across the ASM domain (105–130°E). Composite mean differences between the moist (PC ≥1) and dry (PC ≤-1) phases highlight changes in the intensity and locations of monsoon convection. Here, the “moist” phase is defined by the presence of positive convective heating and PCWV anomalies in the southwestern quadrant of the anticyclone, as this heating center is present in all three modes and all three reanalyses (Figs. S1–S3 in the Supplement). The moist phase of PC1_IAV is associated with enhanced deep convective heating near 70° E (near the Indus River delta), 90° E (over Bangladesh), and 130° E (western North Pacific) in ERA5 (Fig. a), with regions of reduced heating in between. These variations largely offset in the zonal mean, resulting in weak differences in zonally averaged heating (Fig. b). Comparison with results for MERRA-2 and JRA-3Q (Fig. S1) shows that increases in convective heating west of 70° E and near 130° E is robust in the three datasets but differences between 70 and 130° E are not. The largest discrepancies are over Bangladesh and the Bay of Bengal (around 85–95° E), where ERA5 shows stronger convection in the moist phase (Fig. S1a), MERRA-2 shows stronger convection in the dry phase (Fig. S1b), and JRA-3Q shows little difference between the phases. Figure b also shows uplift of the isentropic surfaces along the northern flank of the anticyclone in the moist phase relative to the dry phase, indicative of weaker UTLS baroclinicity, and a weaker tropical easterly jet equatorward of 25° N.

The moist phase of PC2 (Fig. c–d) is associated with enhanced convective heating in the west and reduced convective heating in the east along the 25° N transect, with zonal-mean heating enhanced south of this transect (15–20° N) and reduced to the north (25–30° N). These differences are consistent across all three reanalyses (Fig. S2), with the strongest signal in MERRA-2 and the weakest signal in JRA-3Q. Differences in convective heating along 25° N are located directly below strong differences in the meridional winds (Fig. c), with southerly anomalies over weakened convection and northerly anomalies over strengthened convection indicating an upper-level cyclonic anomaly during moist phases and an anticyclonic anomaly during dry phases. The zonal wind response (Fig. d) suggests a northward shift in the upper-level anticyclone.

Positive PC3 (Fig. e–f) is linked to enhanced deep convective heating over a broad region from 60 to 100° E along the 25° N transect, with stronger zonal-mean heating in the upper troposphere at all latitudes south of 30° N. Figure e also shows anomalies in meridional winds along the 25° N transect, which suggest a wave train along the subtropical westerly jet with a cyclonic center around 110° E and an anticyclonic center around 55° E. In the zonal mean, both the subtropical westerlies and the tropical easterlies are weaker in the positive phase and stronger in the negative phase (Fig. f). All three reanalyses indicate increases in the depth of convection along the southern flank of the anticyclone during the moist phase relative to the dry phase (Fig. S3). Distributions of anomalous heating and winds are likewise broadly consistent.

The following sections present detailed analyses of the interannual (PC1) and subseasonal (PC2 and PC3) modes, including changes in composition (based on CAMS) and the dynamical and thermodynamic environment (based on ERA5). Results for other reanalyses and inter-reanalysis differences are provided in the Supplement, with brief descriptions provided in the main text. For comparability, the following results are all based on principal components calculated based on Aura MLS.

The largest difference between Aura MLS and the reanalysis products in our results is in the linear trend component of PC1. Whereas Aura MLS suggests a drying trend during 2005–2021 centered in the southwest, the reanalyses all show regional moistening centered in the southeast (Figs. , a–b). Based on CAMS, this trend in the reanalysis is associated with increases in ozone below the tropopause and CO above the tropopause (Fig. d, f). Increases in water vapor (Fig. b) and CO below the tropopause are centered in the southeastern part of the domain, possibly suggesting an enhanced convective source to the tropopause layer in this region. However, reanalysis-based trends in water vapor do not match those in Aura MLS (Fig. a). Instead, the satellite observations suggest a negative trend in water vapor below the tropopause, especially in the southwestern quadrant of the anticyclone. Aura MLS further indicates that these decreases in the lower part of the tropopause layer are accompanied by decreases in CO and increases in ozone below the tropopause, while increases in lower-stratospheric water vapor are linked to decreases in ozone and increases in CO above the tropopause (Fig. a, c, e). Trends based on CAMS indicate a very different distribution of changes in CO (Fig. f). Ozone trends are in somewhat better agreement (Fig. d), but this is not a useful litmus test because CAMS, like most of the reanalyses, assimilates Aura MLS ozone .

The S-RIP Final Report  advises that trends based on reanalyses should be assessed with respect to three criteria: (1) consistency across reanalyses, (2) consistency with (ideally independent) observations, and (3) a clear and convincing physical explanation. The trend discussed here meets the first criterion but does not satisfy the second. On the one hand, this disagreement between the reanalyses and Aura MLS is qualified by the v5 water vapor retrieval for Aura MLS having been designed to remove a spurious increasing trend in water vapor , which raises the possibility that these adjustments could have also removed physically meaningful regional trends over the MLS record. On the other hand, except for ozone (for which CAMS assimilates retrievals from Aura MLS), trends in other variables based on Aura MLS also differ substantially from the patterns shown for CAMS (Fig. ).

Figure  addresses the third criterion, as it is important to understand the reanalysis-based trend even if this trend is spurious. Regressions shown in Fig.  are all calculated against PC1_TREND from Aura MLS, thus removing the effects of different magnitudes of the year-to-year increase in PC1 across the reanalyses. Based on ERA5, warm-season cloud ice water content and high cloud cover increased over southern China, northern India, and the southeastern Tibetan Plateau during 2005–2021 (Fig. a). Detrainment from this deeper convection, also supported by increases in CAMS CO in this sector (Fig. f), may explain local increases in PCWV (Fig. b_E–b_S), especially those centered in the lower part of the tropopause layer (Fig. b). Temperatures decreased in the lower stratosphere in this quadrant, although the magnitude of these decreases differs across the reanalyses (Fig. a, c–e). Increases in geopotential height and the Montgomery streamfunction (MSF; Fig. b) are consistent with expectations under global warming (see also the regional expression of greenhouse gas-driven tropospheric warming and stratospheric cooling in Fig. a). The local enhancements of MSF and geopotential height in the southeastern quadrant of the monsoon anticyclone imply additional tropospheric warming there, suggesting that the reduction in CPT temperature may result from an increase in CPT height. These discrepancies in the trend component between observations and reanalyses arise from different responses in the upper troposphere (Fig. b, d; Figs. S4, S7). The reanalyses exhibit better agreement with Aura MLS above the tropopause, where both reanalysis products and observations show a regional moistening trend (Figs. a, c; S4). The moistening trends below the tropopause are collocated with enhanced convection in the southeast quadrant (Fig. a), while the largest moistening trends above the tropopause based on Aura MLS are collocated with increases in CPT temperature (Fig. c–e; consistent with Fig. S8a, d, g). However, further analysis is needed to determine how these features contribute to trends in tropopause-layer water vapor.

Despite qualitative consistency in the signs of water vapor trends among the reanalyses, there are some notable differences in the spatial patterns. For example, whereas JRA-3Q (Fig. b_J) shows moistening mainly over southern China, CAMS (Fig. b_C) and ERA5 (Fig. b_E) show significant increases westward over northern India. This difference may be understood to leading order as differences in the trend of CPT temperatures between ERA5 and JRA-3Q (Fig. c–e). M2-SCREAM (Fig. b_S) and MERRA-2 (Fig. b_M) show even more widespread moistening spanning westward from the South China Sea (larger magnitudes in the MERRA-2 signal are largely attributable to the weaker slope of PC1_TREND in this reanalysis; Fig. b). These differences are more difficult to attribute to physical processes because M2-SCREAM assimilates Aura MLS v4 water vapor retrievals and MERRA-2 damps variability in lower-stratospheric water vapor by relaxing to a mean seasonal cycle. However, it is worth noting that MERRA-2 produces a pattern of cold-point temperature trends that lies roughly between that in ERA5 and that in JRA-3Q, with a similar spatial pattern (Fig. c–e). The robust nature of these CPT temperature changes is consistent with increased convection and cloud cover over the Tibetan Plateau and areas to its southeast. Although CPT temperatures are influenced by model physics and vertical resolution, they are also well constrained by observational data assimilation .

Trends in reanalysis variables may also result from jumps or drifts due to changes in assimilated observations . The budget decomposition articulated in Eq. () provides an opportunity to evaluate this possibility. Figure  shows year-to-year variations in specific humidities due to resolved transport (DYN; -∇⋅Vq-∂(ωq)∂p), parameterized physics (PHY; Sphy), and data assimilation (ASM; Sana) over 2005–2021 based on ERA5, JRA-3Q, and MERRA-2. We split the monsoon anticyclone domain into quadrants covering the southeastern (80–130° E, 15–30° N), southwestern (30–80° E, 15–30° N), northeastern (80–130° E, 30–45° N), and northwestern (30–80° E, 30–45° N) parts of our analysis domain. We focus especially on the southeastern quadrant, since this part of the anticyclone shows the largest increasing trends as represented by the reanalyses in UTLS water vapor over 2005–2021; time series for the northwestern and northeastern quadrants are provided in the Supplement (Fig. S10). Across the three reanalysis datasets, trends in resolved transport are small but positive both above (Fig. S11) and below the tropopause (Fig. S12) in the reanalyses. The increasing trend in 68–83 hPa water vapor in the reanalyses (Fig. a, c) is best explained by an increase in dynamical moistening in the southeast quadrant (Fig. S11). However, the increases in these transports are offset by a decreasing trend in the physical tendencies – and sometimes more than offset, as is the case in MERRA-2. The net changes in PHY + DYN trends are inconsistent across different products and are effectively offset by the data assimilation components, regardless of the altitude relative to the tropopause. Both dynamical and physical tendencies in the southwestern and southeastern quadrants based on ERA5 were relatively flat during 2005–2021, with negative values on average. By contrast, JRA-3Q shows an increasing trend with predominantly positive values, while MERRA-2 shows a decreasing trend with predominantly negative values (Fig. ). Regional moistening over the warm season shows no evidence of a trend in any of the reanalyses, indicating that the sources of the reanalysis-based increasing trends are outside the monsoon region or season and thus beyond the scope of this study. Such a trend is consistent with the trends based on both Aura MLS and the reanalyses in the lower stratosphere (Figs. a–b, a, c), satisfying the three criteria: (1) consistency across reanalyses, (2) consistency with independent observations, and (3) a plausible physical mechanism. Below the tropopause, there are two classes of explanations for the difference between Aura MLS and the reanalysis products: (1) Aura MLS did not observe an increase because it did not occur in reality or was limited to lower levels (i.e., an error in the reanalyses), or (2) Aura MLS did not observe an increase because of a sampling bias (e.g., it was limited mainly to a part of the diurnal cycle that MLS does not sample) or some other gap in the observations. Overall, although we find several reasons to doubt the reanalysis-based trends in water vapor below the tropopause, a more complete analysis of trends in UTLS composition and related fields in this region is warranted.

Figure a further illustrates how we divide interannual variability in PC1 based on Aura MLS into trend (PC1_TREND; Fig. b) and interannual (PC1_IAV; Fig. c) components. For PC1_IAV, we define moist years as those for which the 25th percentile of pentad-mean PC1_IAV was positive and dry years as those for which the 75th percentile was negative. Applying this definition to PC1_IAV from Aura MLS over 2005–2021, 6 years were anomalously moist, 6 years were anomalously dry, and 5 years were neutral (all within the last 6 years of the record; Fig. b). Remarkably, PC1_IAV maintained the same sign through all 31 pentads in 7 of the 17 years (dry: 2006, 2012, 2013; moist: 2007, 2010, 2011, 2017). These results reinforce the impression from Fig.  that the sign of PC1_IAV is set before the monsoon and maintained throughout the monsoon season.

Candidate mechanisms for controlling water vapor in the monsoon tropopause layer at interannual timescales include the El Niño–Southern Oscillation ENSO; and the Quasi-Biennial Oscillation QBO;. To represent ENSO variability, we use the Oceanic Niño Index, defined as 3-month rolling mean sea surface temperature anomalies in the Niño 3.4 region (190–240° E and 5° S–5° N). To represent QBO variability, we use the updated Berlin QBO time series at 50 and 70 hPa, which is based on radiosonde observations from Singapore (1.34° N, 103.89° E) during this period. Figure c shows lag correlations of these indices from the preceding August to the concurrent July against seasonal-mean PC1_IAV (grey diamonds in Fig. b). Correlations are significant for both QBO indices, with the largest correlations (r>0.7) found for the December (50 hPa) and March (70 hPa) preceding monsoon onset. These positive correlations are consistent with the moistening influence of warmer tropical tropopause temperatures as the QBO westerly phase descends toward the tropopause , particularly during Northern Hemisphere winter and spring . Correlations with ENSO are weakly positive and are not statistically significant over this period. Correlations between PC1_IAV and commonly used indices of the tropospheric monsoon intensity, including the All-India Rainfall Index produced by the India Meteorological Department and the tropospheric zonal wind shear index proposed by , are essentially zero (|r|≤0.05).

Figure  shows the vertical and horizontal structure of deseasonalized anomalies in water vapor and CO based on CAMS, along with temperature, cloud fields, MSF, geopotential height, and OLR based on ERA5. All variables are regressed onto Aura MLS PC1_IAV. Anomalies in water vapor and CO are reported as fractional changes given the sharp vertical gradients in these species around the tropopause. The horizontal structure of fractional water vapor anomalies based on CAMS (Fig. c) shows a regional-scale positive anomaly of about 5 %, with the largest anomalies arcing in a band from the Persian Gulf along the northern edge of the Tibetan Plateau to the East Asian monsoon region. This distribution is consistent with the absolute anomalies in PCWV shown in column (c) of Fig. . The southwestern segment of this band, spanning from the Persian Gulf northeastward over the western Tibetan Plateau, also shows positive anomalies in the high cloud fraction and negative anomalies in OLR (Fig. a). These indicators of enhanced convective activity correspond to negative anomalies in radiative heating on the 390 K potential temperature surface (Fig. S13a) and positive anomalies in non-radiative (mainly convective) heating in the lower part of the tropopause layer (Figs. a–b, S13c). Cloud ice, cloud cover, and diabatic heating anomalies thus show that the moist phase of PC1_IAV is associated with stronger convective activity over the Indus River valley in a band that extends northeastward from the Arabian Sea to the Karakoram Pass. The vertical structure along the southern flank of the anticyclone likewise shows large positive anomalies of convective heating penetrating into the tropopause layer in the western part of the domain and near-zero anomalies in the eastern part of the domain (Fig. a). In the zonal mean, the largest positive anomalies are located north of 25° N (Fig. b), consistent with the second center of enhanced water vapor located over the East Asian monsoon region (Figs. c, c). Positive water vapor anomalies are most widespread at 68 hPa; however, unlike anomalies below the tropopause, positive anomalies at 68 hPa are strongest in the southeastern quadrant. This difference is consistent with either in-mixing of moist air from the anomalously warm tropical lower stratosphere or clockwise ascent of water vapor anomalies over East Asia in the “upward-spiraling” carousel of the anticyclone e.g., and highlights the persistence of anomalies associated with this mode. Results based on other reanalyses are generally similar to those based on CAMS (Fig. S14). MERRA-2 is again the outlier, as its strong relaxation of stratospheric water vapor to a climatology eliminates almost all interannual variability at 68 hPa. Greater consistency in the magnitudes of relative anomalies compared to those of absolute anomalies indicates that reanalysis products with larger interannual variability also tend to have larger mean humidities at these levels.

Positive anomalies in MSF and geopotential height (Fig. b) indicate a warmer upper troposphere along the southern flank of the anticyclone, consistent with arrival of the QBO westerly phase at the tropopause. Weak warm anomalies at lower levels to the south of 25° N and weak cool anomalies to the north (Fig. a) imply westerly anomalies through most of the region (Fig. b) by thermal wind balance. Significant ozone anomalies are only found along the southern flank of the anticyclone in the lower stratosphere (Fig. S15). The distribution of ozone anomalies is consistent in five of the six datasets, with JRA-3Q as the exception (Fig. S15). JRA-3Q, which does not assimilate vertical ozone profiles , shows significant increases in ozone within the anticyclone associated with this mode, with large signals centered just below the tropopause (Fig. S15b). CO concentrations are significantly reduced over northern India and the Bay of Bengal and increased southeastward of the East Asian monsoon region. These anomalies are consistent among the CO products based on Aura MLS, MERRA-2, and CAMS (Fig. S16a–c).

Water vapor anomalies associated with PC1_IAV arc clockwise around the anticyclone (right column of Fig. ), with the minimum located over the Bay of Bengal and the maximum located over East Asia and the western Pacific. Figure  provides a framework for understanding this pattern. First, systematically deeper convection in the southwestern quadrant of the monsoon domain (Figs. a and a) injects larger amounts of water above the level of zero radiative heating (LZRH) . This water-rich air is advected northward around the inner western side of the anticyclone, where strong lower-stratospheric radiative heating rates relative to the monsoon core balance ascent across isentropic surfaces see Figs. 8.59–8.60 of. The moist phase of PC1 can thus be explained by anomalously deep convection delivering large amounts of water to the tropopause layer in the southwestern part of the domain and subsequent upward-spiraling transport. Conversely, the dry phase is linked to negative initial anomalies due to a colder tropical tropopause in the pre-monsoon, weaker convection in the southwest and stronger convection in the southeast and colder temperatures along the southern flank of the anticyclone that effectively throttle moist air detrained in the southeast. Additional factors may include variations in cross-tropopause transport over the East Asian summer monsoon e.g., or eastward eddy shedding e.g.,, both of which could contribute to the local maximum in water vapor anomalies in the northeastern part of the domain (Fig. c).

Figure  illustrates contributions to PC1_IAV based on Eq. () applied to ERA5, JRA-3Q, and MERRA-2. These distributions are based on the composite mean over the 6 moist years minus the composite mean over the 6 dry years. In all three reanalyses, the dominant balance associated with this mode is between moistening by moisture flux convergence and drying due to parameterized physical processes Fig. a_E–b_M; see also. The dynamical moistening and physical drying effects nearly offset each other, with the sum of these two terms smallest in MERRA-2 and largest in JRA-3Q (Fig. c_E–c_M). The three reanalyses show different distributions of net moisture sources to the tropopause layer, with ERA5 highlighting the Sichuan Basin, JRA-3Q highlighting northern India and the adjacent southern slope of the Himalayas, and MERRA-2 highlighting the coastal regions around the Bay of Bengal (Fig. e_E–e_M). However, all three reanalyses are consistent in producing sources of water vapor to the tropopause layer over the South China Sea. As with the trend, the sum of the dynamics and physics terms is again largely compensated by the effects of data assimilation in each system (Fig. d_E–d_M). After accounting for data assimilation effects, the net change in area-mean tropopause-layer PCWV is negative over the warm season in all three reanalyses, consistent with relaxation of the initial pre-monsoon anomaly. Several anomalies in the net tendencies are robust across the reanalyses. For example, robust positive tendencies during moist years are located over the western North Pacific and South China Sea, as well as the Indo-Gangetic Plain between the Bay of Bengal and the Indus River valley, while robust positive tendencies during dry years are concentrated mainly over the Yunnan Plateau, between the South China Sea and the Bay of Bengal, and the western coast of southern India.

Figure  shows the implied quadratic cycle of PC2 and PC3, which together describe at least 10 %–20 % of water vapor subseasonal variability in the Asian monsoon tropopause layer. Anomalies in OLR, geopotential height on the 100 hPa isobaric surface, and wave activity fluxes following are shown for the transition periods between different phases. Not all variability in PC2 and PC3 evolves through this coupled cycle, as discussed below: many different modes of convective variability share similar centers of action in this region e.g.,. However, the cycle serves as a useful framework for investigating the mechanisms behind these two modes. Fractions of events that propagated from the previous step in the cycle (Fig. a-i to d-i) or to the next step in the cycle (Fig. a-ii to d-ii) within 2 pentads (10 d) during 2005–2021 are marked in the upper right corners of the transition maps. The strongest links are between the moist phase of PC3 (PC3+), the moist phase of PC2 (PC2+), and the dry phase of PC3 (PC3-), clockwise from the top to the bottom of Fig.  (see also Fig. S17 in the Supplement).

Water vapor anomalies associated with PC3+ exhibit a horizontal dipole structure with a slight southwest-to-northeast tilt (Fig. a; see also Fig. b). Unlike the interannual mode, which is centered above the tropopause in the lowermost stratosphere, anomalies associated with this mode are mainly located in the lowest part of the tropopause layer. Positive anomalies centered in the southwest are linked to dry anomalies over the northeastern Tibetan Plateau, with amplitudes that decrease sharply with increasing height. Conversely, water vapor anomalies associated with PC3-, discussed in detail below, are associated with moist anomalies over the northern Tibetan Plateau and weak dry anomalies over the Arabian Peninsula and Persian Gulf (Fig. c). The largest moist anomalies associated with PC2+ are centered over Iran, Pakistan, and adjacent seas, similar to those in PC3+. However, in PC2+, these anomalies also connect through the ridgeline of the anticyclone to anomalies of the same sign northeast of the Tibetan Plateau above the East Asian monsoon rainband (Fig. b). Moist anomalies associated with PC2- (Fig. d) are centered over the southeastern Tibetan Plateau, bracketed by dry anomalies to the northwest and northeast and with amplitudes considerably weaker than those associated with the other phases. The moist anomaly over the southeastern Tibetan Plateau during PC2- produces an anomaly of the opposing sign when PCWV is regressed onto PC2 (Fig. a), even though this anomaly is not evident in the PC2+ composite (Fig. b).

It is important to emphasize that the distributions shown in Fig.  are based on composites for standardized principal components exceeding ±1, whereas those shown in Fig.  are regressed onto the principal components (see Figs. S18 and S19 for three-dimensional distributions regressed onto PC2 and PC3, respectively). The tropopause layer above the monsoon is very dry in absolute terms. Positive anomalies thus tend to be larger in magnitude than negative anomalies, as illustrated by Fig. . Accordingly, the two poles of the east–west dipole shown in Fig. b manifest more as individual monopoles in Fig. a and c, with the opposing dry anomalies almost disappearing. Water vapor anomalies associated with PC2 and PC3 are robust across the reanalyses and in good agreement with Aura MLS (see Figs. S18 and S19).

As noted above, the North test indicates that PC2 and PC3 are mutually independent. However, as illustrated by Fig. , lead–lag correlations suggest that these modes often occur in quadrature. PC3+ precedes PC2+ by 1 or 2 pentads 40 %–50 % of the time, with a strong area-weighted pattern correlation (r=0.91) between PCWV anomalies composited on PC2+ and average PCWV anomalies 2 pentads after PC3+ (Fig. S17a–c). PC3- then follows PC2+ within 2 pentads more than 50 % of the time. Accordingly, the pattern of PCWV anomalies composited on PC2+ is also highly correlated with average PCWV anomalies 2.5 pentads before PC3- (r=0.87; Fig. S17c–e). PC3- is often followed by strong wave breaking along the eastern flank of the anticyclone and enhanced precipitation over southern China (Fig. c-ii). These processes may complicate the transition from PC3- to PC2-, which only occurs about 30 % of the time within 10 d and shows weaker lead–lag pattern correlations (r=0.69; Fig. S17e–g). Evidence that PC2- is followed by PC3+ about 30 %–40 % of the time completes the quadratic cycle of water vapor subseasonal variability (Fig. d), although pattern correlations between PC2- and anomalies preceding PC3+ are again relatively weak (r=0.55; Fig. S17g, h, a). Figure e shows an example trajectory in the PC2–PC3 phase space, in this case calculated for 15 July–13 September 2013. The trajectory is based on bandpass-filtered 3-hourly CAMS water vapor anomalies projected onto PC2 and PC3 from Aura MLS. The primarily clockwise evolution of the trajectory matches the “typical” sequence illustrated in Fig. a–d.

The timescale of the PC3+-to-PC3- part of the cycle (15–25 d) slightly exceeds that corresponding to quasi-biweekly variations in monsoon circulation and precipitation fields , although this may arise in part from our use of pentad-mean anomalies. Indeed, the return periods indicated by Fig. e are somewhat shorter at 10–15 d. Monsoon quasi-biweekly waves have been shown to exert substantial influences on UTLS composition in this region e.g.,. Nonetheless, the relatively tenuous links to PC2- and the substantial percentages of PC2+ events that are neither preceded by PC3+ nor followed by PC3- emphasize that these modes also vary independently. Convective activity fluctuates on many timescales within the Asian monsoon, from diurnal pulses e.g., over the southeastern Tibetan Plateau and other high-altitude areas; to the synoptic scale e.g., monsoon depressions; upward to the quasi-biweekly and 30–50 d intraseasonal scales e.g., the boreal summer intraseasonal oscillation;. Variability in water vapor near the tropopause that projects onto PC2 and PC3 may emerge when these different modes of organized convective variability move into or through centers of action similar to those highlighted by the OLR anomalies in Fig. . Phase space trajectories for periods dominated by other convective modes evolve differently from those shown in Fig. e and thus may provide a means of distinguishing the imprints of these myriad modes of monsoon convective variability on water vapor near the tropopause.

Anomalies in OLR, geopotential height at the 100 hPa level, and wave activity fluxes provide insight into the mechanisms involved in this cycle. As the links are strongest between the two phases of PC3 and PC2+, we start from PC3+, at the top of the cycle in Fig. . To support the discussion, we also link these transitional anomalies to anomalous temperatures, upper-level cloud fields, OLR, geopotential height, winds, MSF, ozone, and CO regressed onto PC3 (Fig. ) and PC2 (Fig. ). An anomalous high west of the Tibetan Plateau emerges during the days leading up to PC3+ (Fig. a-i), with possible triggers including anomalous convection over northern India and tropopause folding upstream of the plateau. This intensified anticyclonic center in the west persists through PC3+, by which point its northern edge overlaps with the subtropical jet (Figs. a and b). Oscillations of the jet stream due to the restoring force on the poleward shift upstream of the Tibetan Plateau produce a related anomaly downstream over northern China. Each center spans approximately 20–25° longitude, separated by most of the width of the Tibetan Plateau (also ∼25° longitude). These distances indicate a zonal wavenumber of 8–9. With a background zonal wind speed of 25–30 m s⁻¹, the anomaly over northern China should appear about 5 d after the forcing anomaly west of the Tibetan Plateau, in line with the timescales shown in Fig. . Convection over the Indus River valley remains intense into PC2+ (Fig. a). The tail of enhanced high cloud cover and reduced OLR is suggestive of the “up-and-over” transport mechanism that often links lowland monsoon depressions to episodes of strong summer precipitation over the southwestern Tibetan Plateau .

During PC2+, moist anomalies in the southwest (30–70° E along 25° N) are associated with positive anomalies in cloud ice water content (Fig. a) and total diabatic heating (Fig. c). By contrast, cloud ice water content and total diabatic heating are reduced in the southeast (90–130° E) during PC2+. These anomalies are qualitatively robust between the reanalyses (Figs. S2, S18, and S20). The reduced diabatic heating in the southeast is situated under an upper-level cyclonic anomaly over the Bay of Bengal, as shown by negative anomalies in MSF on the 395 K isentropic surface (Fig. b). Temperature anomalies in this region are positive (Fig. a), indicating that the reduction in MSF results from a decrease in geopotential height (Fig. b). The decrease in geopotential height implies a cooler and shallower tropospheric column, consistent with reduced convective heating over Bangladesh and southern China. MSF anomalies associated with the pair of vortices in the north are also linked to anomalous geopotential heights (Fig. b), as both regions feature cold-temperature anomalies near the tropopause (Fig. a).

Warm-temperature anomalies span the entire southern flank of the anticyclone near the tropopause during PC2+ (Figs. a and a), with the largest warming centered between positive cloud ice anomalies in the southwest and negative cloud ice anomalies in the southeast (Fig. a). This location corresponds to the climatological location of the monsoon cold trap Fig. 2 of. The variations in high clouds and attendant changes in OLR (Fig. a) further emphasize the shift in convective activity from the southeast to the southwest in this phase. Radiative heating anomalies on the 390 K isentropic surface are opposite to those in temperature (Fig. S20a) and can therefore be interpreted via the “Newtonian cooling” approximation e.g.,, in which radiative heating rates are modeled as inversely proportional to the difference between ambient temperature and a fixed radiative equilibrium temperature.

Anomalies in ozone (Fig. c) and CO (Fig. d) reflect fluctuations in the upper-level anticyclone. Air with characteristics of the climatological anticyclone (low ozone and high CO) is concentrated in the pair of anticyclonic vortices to the northeast and northwest of the Tibetan Plateau during PC2+, implying an overall northward shift in the position of the anticyclone (see also regional-mean zonal wind anomalies along the eastern face of Fig. b and changes in the 2 PVU (potential vorticity unit) potential vorticity contour in Fig. b). Air along the southern flank exhibits more stratospheric characteristics (high ozone and low CO), especially around 90° E, where the upper-level cyclonic anomaly is centered. These changes indicate that anomalous descent and compression contribute to the warming along the southern flank, in line with reduced lower-stratospheric radiative heating (Fig. S20) and geopotential height (Fig. b). Large potential vorticity anomalies extend along the southern flank of the anticyclone downstream of the reduced convection over the Bay of Bengal (Fig. b). This distribution implies a positive diabatic potential vorticity tendency (i.e., weaker horizontal divergence) over the Bay of Bengal, with the resulting positive anomalies advected westward by the tropical easterly jet.

Variations in water vapor associated with PC2 can be explained to leading order by changes in convective activity and temperatures near the tropopause , mediated by the response of the upper-level anticyclone to the redistribution of convective heating . Area-mean CPT temperatures are much warmer than normal during the PC2+. Positive anomalies of 0.3–0.5 K over much of the region south of 30° N (Fig. a and Fig. S22e–f) translate to an increase of around 5 %–9 % in saturation mixing ratios at the CPT, sufficient to explain the positive fractional anomalies in water vapor anomalies west of this region (∼6 %–9 % Fig. b). The largest water vapor anomalies are in the southwestern quadrant, downstream of both the largest warm anomalies (Fig. a) and the enhanced deep convection north of the Arabian Sea (Figs. a, b-i). Stronger deep convection can inject water directly into the tropopause layer , while warmer CPT temperatures allow more of this injected water to remain. However, the anomalous cyclone to the east and the anomalous anticyclone to the north restrict the flow of anomalously humid air around the climatological anticyclone. The moist center in the northeast coincides with the northeastern anticyclonic anomaly, where ozone is reduced and CO is enhanced (Fig. c–d). These anomalies may thus be interpreted as an anticyclonic vortex displaced from the relatively humid southeastern quadrant into the drier northeast, possibly as a prelude to eastward eddy shedding e.g.,.

PC3- evokes the canonical “Tibetan Plateau” mode of the upper-level anticyclone , which has previously been linked to coherent anomalies in UTLS composition above the ASM e.g.,. The dry phase of PC3 follows a relatively quiescent period in terms of convective activity, with no notable negative anomalies in OLR within the analysis domain (Fig. b-ii, c-i). During this phase, the northeastern high from PC2+ propagates southward along the eastern flank of the anticyclone, as shown by enhanced southward wave activity fluxes over much of China east of the Tibetan Plateau (Fig. b-ii, c-i). Accordingly, the dry phase of this mode is associated with increased MSF in the east and decreased MSF in the west (Fig. b). Changes in MSF again arise mainly from changes in geopotential height rather than local temperature anomalies (Fig. a–b), with reduced MSF anomalies associated with warmer temperatures and vice versa. Changes in ozone (Fig. c) and CO (Fig. d) likewise show that the active (anticyclonic) pole contains air with more tropospheric characteristics, while the inactive (cyclonic) pole contains air with more stratospheric characteristics.

Unlike PC2, variations in local dehydration cannot provide a direct explanation for variations in water vapor in this mode because CPT temperatures are anomalously cold where water vapor is enhanced and warm where water vapor is reduced (cf. Figs. c, a). Concurrent variations in both isentropic ascent (anomalous upward tilt toward the active phase) and diabatic ascent (anomalous radiative heating where CPT temperatures are cold) may contribute, as may the changing boundaries of the anticyclone itself, under the assumption that relatively humid air is mainly confined to the active core. However, we note that deeper convective activity and more extensive high cloud cover are again located upstream of the water vapor maximum (Figs. a, S3, S24), though this arrangement is more pronounced in PC3+ (the “Iranian Plateau” mode) than in PC3- (the “Tibetan Plateau” mode). Over the eastern Tibetan Plateau, OLR anomalies are noticeably larger in ERA5 (Fig. a) than in JRA-3Q or MERRA-2 (Fig. S24), which may help to explain the sharper southwest-to-northeast tilt of this mode in the ECMWF reanalyses (i.e., CAMS and ERA5). Supporting this point, the inter-reanalysis discrepancy in the tilt of the dipole is found only in water vapor and not in ozone or CO (Figs. c, d, S16g-i and S23), suggesting that it may be explained by stronger convective injection in ERA5 and CAMS, which is then constrained by local dehydration in the cold anticyclonic vortex.

As in PC2, anomalies in temperature, radiative heating, MSF, ozone, and CO (Fig. ) are coherent in both poles of the PC3 east–west dipole, demonstrating the dynamical reorganization of the anticyclone associated with PC3. PC3+ is linked to anomalous convective activity over South Asia, where stronger heating induces a westward shift, while PC3- is linked to anomalous convective activity over the western Pacific, where stronger heating induces an eastward shift e.g., Fig. 15 of. However, our results do not rule out a role of convective injection, with the reanalyses showing enhanced deep convection over the Indus River valley and the Karakoram Pass when the anticyclone center is shifted westward and over the Sichuan Basin when the anticyclone center is shifted eastward (Figs. a, S24). Such shifts in deep convection may help to “fill up” the meridionally displaced anticyclone cores with freshly detrained air e.g.,, with the caveat that such moistening effects could be quickly damped by the negative-temperature anomalies near the tropopause (Fig. c).

The reanalysis products agree well with each other and with Aura MLS in their representations of anomalies in dynamical fields, temperature, cloud cover, ozone, and CO associated with PC2 and PC3 (Figs. S16, S20–S24). ERA5 shows the largest shifts and deepest penetration of anomalous deep convection along 25° N, with changes in MERRA-2 and JRA-3Q more confined to the upper troposphere, especially in PC2 (i.e., p>150 hPa; Fig. S3). Therefore, although all three reanalyses are consistent in the qualitative east–west shift in convective activity along this transect, we cannot confidently infer the implications for convective injection above the LZRH.

To emphasize these points, Fig.  shows anomalous water vapor tendencies due to resolved moisture flux convergence and parameterized physics (PCWVdyn+phy) during the transition phases between PC2 and PC3. Here, the transition phases are defined as the 3 pentads ending on the corresponding phase of PC2 or PC3 with no lead–lag requirement so that all events are sampled and each sampled event includes 15 d. The centers of water vapor enhancement for each phase, where fractional changes in PCWV exceed 10 %, are delineated by pink contours. Despite some differences in magnitude, anomalies in PCWVdyn+phy from these three reanalyses are broadly consistent, indicating that the reanalyses produce these modes primarily through internal processes. The one exception to this consistency is the western center of PC2 in JRA-3Q, where a weak dynamical term (Fig. S25) and a moderately strong physics term (Fig. S26) produce net tendencies that are largely out of phase with ERA5 and MERRA-2. Discrepancies between JRA-3Q and the other reanalyses in the southwestern quadrant are reduced by relatively large data assimilation increments there (Fig. S27). In all three reanalyses, the physics and dynamics terms mainly compensate each other. These individual terms have magnitudes that exceed their sum by more than a factor of 10, and their magnitudes and the detailed distributions are more varied across the reanalyses (Figs. S25 and S26). Assimilation increments are even more diverse (Fig. S27), with magnitudes only slightly less than the sum of the dynamics and physics terms PCWVdyn+phy.