Measurements of water vapor, ozone, and temperature from 2005–2021 are sourced from the Earth Observing System (EOS) MLS aboard the Aura satellite, version 5.0x. While MLS data continue to 2026, our analyses use data to 2021 to avoid impacts of the Hunga Tonga-Hunga Ha'apai eruption in early 2022 . Aura was launched in 2004 into a sun-synchronous satellite orbit with Equator crossing times of 01:30 and 13:30 local time as part of the NASA A-Train satellite constellation. MLS measures ∼3600 profiles per day with a 1.5° along-track separation between measurements . The retrieval range of MLS measurements varies by species; water vapor is observed from 316–0.001 hPa, ozone from 261–0.001 hPa and temperature from 261–0.00046 hPa, and the data are quality-controlled following the recommendations of . Retrievals are provided on a regular pressure grid, with data at 12 levels per decade of pressure (∼1.25 km vertical sampling) for pressures greater than 1 hPa. The precision, accuracy, horizontal resolution, and vertical resolution of each species vary by altitude and are described in detail in . In our region of focus in the UTLS, the effective vertical resolution is ∼3 km for water vapor and ozone. The data analyzed here are monthly, zonal averages of level 2 (orbital) data binned in 2° latitude bins, which contain on the order of 1000 measurements per month over midlatitudes.

We note that the previous version 4.x MLS data exhibited a spurious positive H₂O drift in the stratosphere that was identified by . further identified that there was an increasing drift associated with the 190 GHz sideband fraction in the MLS measurement, and this was subsequently addressed and partially ameliorated in version 5.x processing. However, the MLS instrument team suggests continued caution in studies evaluating H₂O trends using version 5 data .

We also analyze satellite water vapor observations from the ACE-FTS (hereafter ACE) instrument from 2005–2021, using the version 5.2 retrieval . ACE measures solar occultation spectra between 750 and 4400 cm⁻¹ at a spectral resolution of 0.02 cm⁻¹ to retrieve vertical profiles of temperature, pressure, and trace gas volume mixing ratios from the cloud tops to approximately 150 km. Altitude is the native vertical coordinate for these retrievals and the effective vertical resolution in the UTLS is about 3 km, oversampled at 1 km spacing. The instrument typically measures 30 profiles per day, requiring a three-month period to achieve near-global coverage (85° N–85° S). H₂O validation of the ACE-FTS instrument in the UTLS is discussed in depth by . The recommended ACE v5.2 quality-control flags described in are applied to remove scientifically unsuitable profiles prior to analysis. We perform an additional simple quality control measure to eliminate extreme and non-physical measurements (particularly near and below the tropopause) from the dataset by removing data points that fall outside of the 2nd and 98th percentile values for their specific height and latitudinal region. Because of the much sparser sampling of ACE compared to MLS, we use 3-month (DJF, MAM, etc.) averages in 10° latitude bins, which typically contain ∼20–100 measurements over midlatitudes. We require a minimum of 10 data points per 3-month 10° latitude bin for a valid data sample. Because of frequent loss of ACE-FTS retrievals below the tropopause (due to clouds or other problems) there are often missing data, and these regions are indicated as data voids in the figures below.

Numerical experiments were conducted using the Community Earth System Model version 2 with the Whole Atmosphere Community Climate Model version 6 as its atmospheric component CESM2/WACCM6;. WACCM6 is a state-of-the-art chemistry–climate model that extends from the Earth’s surface to approximately 140 km and includes comprehensive troposphere–stratosphere–mesosphere–lower thermosphere (TSMLT) chemistry. In this study, WACCM6 is run in a specified-dynamics configuration (WACCM6-SD), with prescribed sea surface temperatures and sea ice. The simulations provide daily data covering the period 1980–2022 and use a horizontal resolution of 0.9° latitude ×1.25° longitude with the finite-volume dynamical core . The model includes 70 vertical levels from the surface to ∼140 km, with a vertical resolution of approximately 500 m in the upper troposphere and lower stratosphere (UTLS). In the specified-dynamics configuration, temperature and winds (U, V) are nudged toward the Modern-Era Retrospective Analysis for Research and Applications Version 2 (MERRA-2) reanalysis using a 12 h relaxation timescale. This configuration has been shown to realistically reproduce key modes of stratosphere–troposphere coupling, including the quasi-biennial oscillation and stratospheric sudden warmings. It also provides a realistic representation of large-scale transport in the UTLS region and has been widely used in troposphere–stratosphere transport studies e.g.,. For a direct comparison with MLS, we restrict the analysis to 2005–2021, interpolate to the MLS vertical grid, and compute monthly, zonal averages in 2° latitude bins.

Studies of UTLS composition can benefit from analyses using alternative coordinate systems, both vertically and horizontally . Various vertical coordinate systems can be used to reduce the composition variability associated with changes in tropopause height to isolate other sources of variability, such as from stratosphere-troposphere exchange. For analysis of H₂O in the lowermost stratosphere, tropopause-relative altitude is a common vertical coordinate and has been frequently used in previous studies e.g.,.

In this study we conduct analyses in both geometric (e.g. pressure, altitude) and tropopause-relative coordinates. Tropopause-relative calculations are computed for daily observations before aggregating the results into monthly (MLS, WACCM) or 3-monthly (ACE) means. For ACE and MLS, a reference tropopause is computed from three-hourly MERRA-2 reanalysis data and interpolated in space and time to each profile location. The tropopause is also calculated daily in WACCM6 as the reference for the model profiles. Since the model is nudged to MERRA-2, these two tropopause datasets are nearly identical. The results shown here use the potential temperature gradient tropopause PTGT; as the reference tropopause definition. We also tested tropopause-relative calculations using the World Meteorological Organization WMO; lapse-rate tropopause definition, finding consistent results with the PTGT (not shown). For MLS and WACCM, data is placed into tropopause-relative logarithmic pressure bins with a resolution corresponding to 12 levels per decade of pressure (Δz∼1.25 km), while ACE observations are placed into 1 km tropopause-relative altitude bins. Monthly zonal means are then computed for MLS and WACCM data and three-monthly seasonal means are computed for ACE. The tropopause-relative vertical coordinates are transformed back into an estimated geometric coordinate using the climatological mean tropopause pressure, with the logarithmic pressure bins becoming a “pseudo-pressure” coordinate. Some of our results include (area-weighted) extratropical tropopause averages, including monthly/seasonal data points poleward of 20° N/S where the local tropopause pressure exceeds 150 hPa, indicative of the lower-tropopause (extratropical) environments poleward of the tropopause break . This requirement constrains the analysis to at least partially exclude summer monsoon anticyclones, which frequently exhibit tropopause pressures of <150 hPa .

An example of the H₂O variability compared in geometric vs. tropopause-relative coordinates is illustrated in Fig. a–b, showing time series of MLS, ACE, and WACCM data at one location near the midlatitude tropopause (45° N, 215 hPa, 12 km). In geometric coordinates (Fig. a) there is a large annual cycle of H₂O with maxima during boreal summer, which closely follows the well-known annual cycle in midlatitude tropopause pressure (Fig. c). Time series in tropopause-relative coordinates (Fig. b) show a much smaller annual cycle, with smaller maxima during boreal summer. The comparisons in Fig. a–b show larger H₂O values near the extratropical tropopause for WACCM compared to MLS and ACE, and this is a general behavior as shown and discussed further below.

Time series of the respective deseasonalized anomalies are shown in Fig. d–f. Tropopause-relative calculations (Fig. e) show a reduction in H₂O variance around the midlatitude tropopause compared to geometric coordinates by ∼30 % (Fig. 1d), due to removal of variability tied to the dynamic tropopause. This reduction in variance is consistent with the results of . Figure  shows global comparisons of deseasonalized H₂O variance in the two coordinate systems. In geometric coordinates (Fig. a) there are maxima concentrated around the ExTL region of both hemispheres concentrated in the ExTL region, primarily linked with variations in the tropopause itself. Calculations in tropopause-relative coordinates (Fig. b) substantially reduce the ExTL variability, as the variance associated with changes in the height of the tropopause is removed. Similar patterns are found when this analysis is conducted on the WACCM simulation (not shown).

In contrast to the reduction of variance near the ExTL, the use of tropopause-relative coordinates substantially increases H₂O variability throughout the tropical upper troposphere and below the subtropical jets (Fig. b – note maxima near 30° N and S and near the equator). This is an artifact of the tropopause-relative coordinate calculations. The tropical upper troposphere is characterized by strong vertical gradients in water vapor concentration, and small variations in tropopause altitude can lead to large changes in tropopause-relative water vapor. We find that much of the enhanced H₂O variance in this region is due to low frequency changes in tropopause height driven by ENSO and QBO variations . As a result, using tropopause-relative coordinates introduces artificial H₂O variability in the tropical upper troposphere, and is a poor coordinate choice for this region. Similarly, the increased variance in Antarctic tropopause-relative tropospheric H₂O stems from strong variations in the identified tropopause, where extreme stability makes tropopause identifications difficult. While the PTGT used here improves upon the frequent tropopause misidentifications by the LRT in polar winter, there are likely still cases where the diagnosed tropopause artificially increases the H₂O variability in the polar upper troposphere. However, this has minimal impact on our results focused on the extratropics.

Long-term average (2005–2021) cross sections of H₂O from MLS, ACE and WACCM are shown in Fig. , with results in both geometric and tropopause-relative coordinates. Patterns highlight the moist troposphere, dry stratosphere and strong gradients across the global tropopause. The tropopause-relative calculations highlight local vertical stratification in the extratropics compared to “fuzzy” structure in geometric coordinates. Careful comparison of the WACCM vs. MLS (or ACE) observations shows enhanced WACCM moisture in the extratropical lowermost stratosphere, and this is quantified in the (WACCM-MLS)/MLS percentage differences shown in Fig. . Compared with MLS, WACCM H₂O is more than 300 % higher in polar regions and more than 200 % higher in subtropical environments in the lower stratosphere of both hemispheres. Such moist biases in the LMS are a common problem for global models, as highlighted in . We note that this biased H₂O behavior is not simply due to circulation or temperature biases in WACCM, as our simulations use specified dynamics calculations with observed meteorological fields.

Seasonal averages in zonal mean tropopause-relative H₂O concentration and atmospheric temperature as observed by MLS are shown in Fig. . The tropopause acts as a transport barrier in the extratropics, where there is a sharp gradient in all seasons and in both hemispheres. In the deep tropics, tropopause temperatures vary from ∼192 K in December, January, and February (DJF) to more than 196 K in June, July, and August (JJA). This seasonal pattern is reflected by in-phase H₂O concentrations at the tropical tropopause (low H₂O during DJF), and this seasonal minimum subsequently propagates upwards in the tropics and spreads in latitude in the lower stratosphere over most of the globe, demonstrating the well-known “tape recorder” signal . The meridional transport of air via the shallow branch of the BDC is evident in the dry layers extending along isentropic layers to higher latitudes in MAM and all the way to the North Pole in JJA. Notably, this seasonally-varying dry layer does not extend all the way down to the tropopause in the extratropics but stays above ∼215 hPa in the pseudo-pressure coordinates. Air in between this dry layer and the extratropical tropopause is wetter and is a signature of the ExTL. The summer hemisphere ExTL is particularly enhanced relative to the background stratosphere.

The hemisphere-averaged annual cycle of lower stratospheric MLS water vapor concentrations in tropopause-following coordinates for both hemispheres is shown in Fig. a and c. These data represent area-weighted latitudinal and zonal averages poleward of 20° latitude, as described in Sect. , and are intended to represent mean hemispheric behaviors. As expected, there are strong vertical gradients across the tropopause. In both the Northern Hemisphere (NH) and Southern Hemisphere (SH) there is a clear annual cycle in upper tropospheric water vapor concentrations which peaks in hemispheric summer (labeled “B” in Fig. a and c), extending to the first layer above the tropopause (and weakly to the second layer). In the NH, water vapor concentration peaks at ∼16 ppmv in August in the first layer above the tropopause (∼1.25 km), while there is a peak of ∼7 ppmv in October that occurs ∼2.5 km above the tropopause. There is a similar summertime maximum in the SH for these lowest two stratospheric layers but with smaller amplitude. The dry layer associated with air originating near the tropical tropopause can be seen in Fig. a and c in both hemispheres (labeled “A” in Fig. a and c), occurring ∼3–4 km above the tropopause in February–August; this signal lags the seasonal H₂O minimum at the tropical tropopause by several months, and is coherent between hemispheres. This distinct seasonal cycle helps define the overworld influence on the water vapor concentrations in the extratropics. It follows that the height of this layer can be used as an upper boundary when defining the ExTL, providing a physically based definition to a layer that fluctuates in both space and time. Although not shown here, similar LMS H₂O seasonal cycles are observed in ACE data, but with limited resolution resulting from the 3-month temporal sampling.

Corresponding extratropical LMS seasonal variations for WACCM are shown in Fig. b and d. In WACCM, the ExTL has higher H₂O concentrations than MLS and extends ∼1–2 km deeper into the stratosphere in both hemispheres, consistent with the time average biases seen in Fig. . Hence, the time average WACCM LMS H₂O biases are reflected in a thicker seasonally-evolving ExTL.

One further minor feature identified in the SH statistics in Fig. c and d is a minimum in H₂O in the SH during August–October, ∼5–7 km above the tropopause (i.e. geometric altitudes near 15–17 km) (labeled “C” in Fig. c and d). The altitude and timing of this feature suggest a relation to seasonal dehydration in the Antarctic polar vortex, where effects can spread outside the polar vortex in springtime e.g., and may be reflected in hemispheric mean H₂O.

To assess interannual variability and structure, the average monthly seasonal cycle is removed from the MLS time series to create deseasonalized anomalies, with time series for both hemispheres shown in Fig. . There is a distinct separation of patterns in the vertical structure occurring around ∼2.5 km above the tropopause in both hemispheres. Above 2.5 km tropopause-relative altitude there are interannual H₂O anomalies that are coherent between hemispheres. This variability is closely tied to overworld transport from the tropical tropopause source region, such that air is transported poleward into the extratropical LMS with a time scale of ∼2–4 months e.g.,. In the ExTL below 2.5 km tropopause-relative altitude, the interannual patterns are different from above and variability is dominated by an apparent long-term decreasing trend in both hemispheres. This trend is further investigated in Sect. . Notably, even when the ExTL H₂O trend is removed, the vertical discontinuity in overworld and ExTL H₂O remains (see Fig. S1 in the Supplement).

A complementary perspective of the NH and SH H₂O anomalies in MLS data at several vertical levels is shown in the time series in Fig. . These show time series of overlaid NH and SH data to highlight correlated variability. Strong correlation (r=0.70) is found at the 3rd level (>2.5 km above the tropopause), with somewhat weaker correlation (r=0.41) over the 1.2–2.5 km layer. The first layer (0–1.25 km, within the ExTL) has weak interhemispheric correlation, aside from the strong trends.

Figure provides additional evidence regarding the origin and transport of extratropical overworld H₂O anomalies, showing spatial correlations between deseasonalized and detrended water vapor anomalies to a reference point located near the tropical tropopause (83 hPa), with results in both geometric and tropopause-relative coordinates. At zero-time lag there are positive correlations centered in the tropics and extending poleward in a hemispherically symmetric pattern. The tropopause-relative calculations show strong correlations extending downward through the tropical upper troposphere, but this is an artifact of the tropopause-relative coordinates that is not observed in geometric coordinates. At a 2-month time lag, which is the general timescale of poleward transport from the tropics to the extratropical LMS e.g.,, the regions of highest correlations spread upward and further poleward in both coordinate systems. The tropopause-relative analysis shows higher extratropical correlations (above 0.6) and more vertically confined branches of positive correlations in the extratropical and polar regions, with the highest correlations occurring a few kms (∼100 hPa) above the tropopause. This is collocated with the dry layer of tropical origin shown in Figs.  and a–b, and quantifies the vertical profile of overworld transport from the tropics that defines the effective top of the ExTL. Parallel analysis of the WACCM simulation is given in Fig. S2.

The observed distinction between overworld and ExTL, as identified by their vertical structure and temporal behavior, raises the question of whether similar behavior occurs for other trace gases. Identical to the analysis conducted for Fig. , deseasonalized ozone anomalies from MLS are shown in Fig. . In contrast to the H₂O results, there is a coherent vertical structure in each hemisphere in the ozone anomalies that extends from the overworld down to across the tropopause, with no evidence of discontinuity across the ExTL (although the background amounts of ozone changes significantly). Given that H₂O and ozone both exhibit large gradients across the tropopause, one might expect a similar behavior in the vertical continuity (or lack thereof) in the two trace gases. However, the exact vertical tropopause-relative locations and magnitudes of such gradients vary substantially between H₂O and ozone, such that H₂O experiences a far more extreme transition across the extratropical tropopause than ozone (see Fig. 11 in ). The distinct behavior for H₂O compared to ozone is therefore realistic due to the large reservoir of H₂O below the tropopause together with the source of overworld variability in the tropics. Additionally, there is little correlation in the ozone anomalies between the NH and SH because the location of the strongest ozone gradient exists well above the tropopause (∼2–7 km), where variability is more sensitive to extratropical downwelling which is mostly independent between hemispheres.

For comparison with MLS, we also examine deseasonalized water vapor anomalies in ACE and WACCM in Fig.  and the ACE and WACCM versions of Fig.  are given in Figs. S3 and S4. The broad, hemispherically coherent interannual variability seen in MLS above ∼2.5 km tropopause-relative altitude is also present in the ACE and WACCM datasets (Fig. a–b). ACE reproduces the timing and vertical extent of H₂O anomalies seen in MLS but with greater noise and reduced amplitude, likely related to the latitudinally-varying sampling across seasons. Despite this limitation, the sign and phase of anomalies generally match MLS, supporting consistency between observational datasets. WACCM also matches the overall structure and phase of the overworld anomalies seen in MLS, which is expected due to the prescribed tropical tropopause temperatures used in the specified dynamics run. However, the magnitude of the WACCM variability is substantially larger than MLS, perhaps due to the vertical smearing (∼3 km) inherent in the MLS data. While the overworld H₂O variations are consistent among MLS, ACE, and WACCM, the ExTL variations are markedly different between datasets as neither ACE nor WACCM displays the negative trend observed from MLS in the ExTL layer (Fig. c). These comparisons are explored further in Sect. . Importantly, the magnitude of the trend in the MLS ExTL makes investigation of dataset comparisons of other variability in this layer infeasible.

The vertical variation of the correlation between water vapor anomalies in the NH and SH in each dataset is explored in Fig. . The hemispheric correlations in MLS, ACE, and WACCM exhibit a similar vertical structure, with a minimum in the first or second layer above the tropopause, and a maximum that occurs around 4–5 km above the tropopause. Throughout the vertical profile, hemispheric correlation in ACE is consistently 0.1–0.3 smaller than MLS. This is likely a reflection of the sparser sampling of the ACE dataset. Comparisons of seasonal means between hemispheres can remain sensitive to the latitudinal distribution of contributing profiles, thereby adding additional sources of variability and reducing the correlation coefficient between hemispheres. The hemispheric correlation in WACCM above 5 km tropopause-relative altitude is nearly identical to MLS, which is expected due to specified dynamics and prescribed tropopause temperatures in the model accurately reflecting stratospheric overworld behavior. Below 5 km tropopause-relative altitude, however, the values of the correlation coefficients in WACCM are systematically smaller than MLS. This behavior may be a signature of the moist LMS bias and deeper extent of the ExTL in WACCM (e.g., Fig. ).

The commonality of higher hemispheric correlation above 4 km tropopause-relative altitude in all three datasets demonstrates that hemispheric symmetry is a robust signature of the stratospheric overworld, resulting from a common stratospheric entry point and similar transport times of the lower branch of the BDC e.g.,. In diagnosing the vertical structure of the lower stratosphere, hemispheric symmetry and correlation can provide a physical indicator of the separation between the stratospheric overworld, the ExTL, and the hybrid region in between (Fig. ).

The dominant variability in the ExTL in MLS H₂O data is a downward trend (Fig. c). However, this trend is not evident in either ACE observations or WACCM simulation (Fig. ). Here we examine large-scale variability and trends in the global data sets to identify coherent structures and systematically evaluate trends. We conduct a standard multivariate linear regression to the H₂O time series at each point and include terms representing a linear trend, ENSO, and the QBO (using two orthogonal time series) to isolate their respective contributions to observed variability e.g.,: 1y(t)=β0+β1t+β2ENSO(t)+β3QBO1(t)+β4QBO2(t)+ε(t) where β1t represents the trend term and ε(t) is the residual. The ENSO and QBO proxies are taken from the NOAA Physical Sciences Laboratory Multivariate ENSO Index (MEI) and the NASA Goddard Space Flight Center QBO datasets, respectively. Statistical significance of regression coefficients was assessed using a two-sided Student's t-test with a p<0.05 significance threshold. To account for serial autocorrelation in the regression residuals, an effective sample size for each dataset was calculated as: 2Neff=N1-r11+r1 where N is the number of observations and r1 is the lag-1 autocorrelation of the regression residuals. The effective sample size was then used when computing the standard errors and resulting p-values of the regression coefficients. Results are calculated for both geometric and tropopause-relative coordinates.

Results of the trend calculations are shown for all three data sets in Fig. . Trends in the overworld are generally positive in all data, reflecting a slight increase in cold point temperature over the data record 2005–2021. Similar positive stratospheric H₂O trends over this time period have been shown in , , , and . In contrast, H₂O trends in the troposphere and ExTL of both hemispheres are very different among the datasets. MLS exhibits negative trends in the global upper troposphere and ExTL regions, with patterns that generally follow the global tropopause structure. The trends are slightly more negative in tropopause-relative coordinates, reflecting the slight rise of the tropopause over this period e.g.,, which acts to reduce H₂O values in the tropopause region, all else being equal. However, these systematic, large-scale negative trends derived from MLS data are not evident in either ACE observations or WACCM simulation, and these differences question the reality of the MLS trend results in the troposphere and ExTL, as discussed further below.

Regression fits for ENSO variability are shown in Fig. . In general, H₂O is expected to follow temperature changes throughout the troposphere (from the Clausius-Clapeyron relationship), so that positive H₂O ENSO regression patterns are expected in the troposphere following the well-known positive zonal mean temperature changes for ENSO e.g.,. This positive ENSO signal is observed in MLS, ACE, and WACCM geometric coordinate results (top row in Fig. ), with patterns covering the tropical upper troposphere and extending across the tropopause to higher latitudes in both hemispheres. Hence, ENSO is a significant component of ExTL H₂O variability. However, the broad-scale ENSO signature is much less coherent (and insignificant) in tropopause-relative coordinates (bottom row of Fig. ), because the ENSO-related changes in tropopause height confuse the signal in the tropics.

The global QBO1 pattern in MLS H₂O data is shown in Fig. . The QBO1 index is negatively correlated with tropical cold point temperature e.g., and this produces a negative correlation with H₂O in the tropics and global lower stratosphere (i.e. overworld). Results in tropopause-relative coordinates show sharper and more vertically stratified patterns that delineate overworld influence and ExTL structure. QBO1 patterns in the troposphere in Fig. b (negative values over the equator and positive below the tropopause in extratropics) reflect the influence of the QBO on tropopause height, and are artifacts of the coordinate system. Similar QBO regression results are found in ACE and WACCM data and are not shown here. QBO2 regression coefficients have little impact on the CPT and hence have less impact on H₂O (not shown here).