Reanalyses aim to produce time-consistent datasets by integrating available observations through data assimilation to a “fixed” version of the Numerical Weather Prediction (NWP) model, thus providing the most complete data on past climate. The NWP model, its version, set of incorporated observations, and data assimilation methods depend on the assimilation centre and on the version of the analysis. In this work we use data from four global reanalyses: the Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA2) from the National Aeronautics and Space Administration (NASA) , the Japanese 55-year Reanalysis (JRA55) from the Japan Meteorological Agency (JMA) , the ERA-Interim and ERA5 reanalyses from the European Centre for Medium-Range Weather Forecasts (ECMWF). The ERA5 (European Centre for Medium-Range Weather forecasts Reanalysis v5) is the fifth generation ECMWF atmospheric reanalysis, covering the period from 1940 to the present. The reanalysis has 137 vertical levels extending from the surface to 0.01 hPa. In ERA5, the stratosphere is represented by approximately 45 levels. In all four reanalyses vertical resolution gets gradually coarser with altitude; in ERA5, the resolution is about 350 m in the lowermost stratosphere and 1500 m in the uppermost stratosphere. The ERA5 high-resolution data are produced as spectral coefficients with a triangular truncation of T639, which roughly corresponds to a horizontal resolution of 0.28°; the data are available at hourly intervals.

The ERA-Interim (Interim European Centre for Medium-Range Weather forecasts Reanalysis) is a predecessor of ERA5, it was discontinued in 2019 and covers the period from 1 January 1979 to 31 August 2019. ERA-Interim has 60 vertical levels spanning from the surface to 0.1 hPa. Approximately 20 of these levels are in the stratosphere, with a vertical resolution of about 1000 m in the lowest part of the stratosphere and roughly 3000 m at the top of the stratosphere. The spectral resolution of the reanalysis is T255, it corresponds to a horizontal resolution close to 0.70°; the data are available at 6 h intervals.

The JRA55 data are available from 1958 and were discontinued at the end of January 2024, the topmost level of JRA55 is 0.1 hPa, it also has 60 vertical levels about 20 of which are in the stratosphere. Vertical resolution in the stratosphere is similar to ERA-Interim. With spectral resolution of T319 the horizontal resolution of JRA55 is approximately twice as coarse as ERA5, at about 0.56°; the temporal resolution of the archived data is 6 h.

The MERRA2 reanalysis covers the period from 1980 to present. The reanalysis has 72 levels spanning from the surface to 0.01 hPa, about 25 of the levels are in the stratosphere. Vertical resolution of the reanalysis in the lowermost stratosphere is close to 1000 and to 2000 m at the top of the stratosphere. The native horizontal resolution of MERRA2 is 0.5° lat ×0.625° long. Three-dimensional products are available at 3 h intervals.

The main focus is on ERA5, which is the most recent reanalysis of these four. The enhanced horizontal and vertical resolutions of ERA5 compared to the other reanalyses are expected to result in improved ability to resolve smaller-scale waves . Other reanalyses are mostly used for comparison and to assess the robustness of the results.

To retain consistency of spatial and temporal resolution in the inter-comparison, all used reanalyses were resampled on a 1×1° latitude–longitude grid, interpolated vertically to the same 88 pressure levels and resampled in time to 6 h (00:00, 06:00, 12:00, and 18:00 UTC). The comparison between different reanalyses has been done on the common 38-year period from 1980 to 2017. The effect of resampling on ERA5 data is further explored in Sect. .

The pattern of the residual circulation in the stratosphere is shown in Fig. 1 based on ERA5 reanalysis (purple arrows in panel a). The figure indicates evidence for the two main branches of the overturning circulation: the deep branch transfers mass from the tropical tropopause to the middle and upper stratosphere and towards the winter pole, and the shallow branch acting in the lower part of the stratosphere, creates poleward mass flux in both hemispheres.

The analyses in this paper are based on the TEM framework utilizing the two independent estimates to diagnose and investigate the overturning circulation: (1) the “direct” residual circulation definition v‾*,w‾* and (2) an “indirect” momentum balance-based estimate of the residual circulation v‾D*,w‾D*, where the upwelling velocity is calculated from the wave drag based on the downward control principle (DWCP) e.g.. Although both approaches are based on the TEM framework, we will refer to the direct estimate of the residual circulation as the “TEM” approach and to the indirect momentum balance estimate as the “DWCP” approach, in the following. On the one hand, the direct TEM approach is conceptually simpler and more frequently used in studies of the stratospheric circulation. On the other hand, the DWCP approach allows to relate the residual circulation to the wave drag and in particular to estimate the relative contributions of individual waves to the driving of the residual circulation. Uncertainties and limitations such as those arising from numerical approximations, data assimilation procedures, and the lack of strict mass conservation in reanalyses affect both estimates in distinct way. Therefore, their comparison provides valuable insight into the robustness of our results.

The vertical and horizontal components of the TEM residual circulation in log-pressure coordinates are defined by 1v‾*=v‾-ρ0-1(ρ0v′θ′‾/θz‾)z,w‾*=w‾+(acos⁡(φ))-1(cos⁡(φ)v′θ′‾/θz‾)φ, where an overbar indicates the zonal mean, a prime the deviation from the mean, subscripts denote partial derivative, v is meridional velocity, ρ0 is basic density, θ is potential temperature, w is vertical velocity, and a is the mean radius of Earth.

Energy deposition from wave dissipation constitutes the driving force of the residual circulation. The wave dissipation results in eddy potential vorticity fluxes which, in turn, are approximately equal to the Eliassen–Palm flux (EP flux) . The latitudinal and vertical components of the EP flux are defined by 2F(φ)=ρ0acos⁡(φ)(uz‾v′θ′‾/θz‾-v′u′‾),F(z)=ρ0acos⁡(φ){[f-(acos⁡(φ))-1(u‾cos⁡(φ))φ]v′θ′‾/θz‾-w′u′‾}, where u is zonal wind velocity, f is the Coriolis parameter defined as 2Ωsin⁡(φ), where Ω is the angular velocity of Earth. The divergence of the EP flux (∇⋅F) represents the driving force in the zonal momentum balance equation 3∂u‾∂t+v‾*1acos⁡(φ)v∂u‾cos⁡(φ)∂φ-f+w‾*∂u‾∂z=∇⋅Fρ0acos⁡φ+X‾. Here, X includes contributions from parameterized waves and will not be further considered in this paper which focuses on the contribution from resolved waves.

The contributions of individual waves to the EP flux can be estimated by Fourier transformation of the three-dimensional EP flux data from longitude to wavenumber–frequency domain. The thus transformed components of the EP flux are defined by 4F(φ)(s)=ρ0acos⁡(φ)Re[uz‾v^(s)θ^*(s)/θz‾-u^(s)v^*(s)]/(ns2/2),F(z)(s)=ρ0acos⁡(φ)Re{[f-(acos⁡(φ))-1(u‾cos⁡(φ))φ]v^(s)θ^*(s)/θz‾-u^(s)w^*(s)}/(ns2/2), where Re denotes the real part of a complex number, asterisks denote complex conjugates, hats represent Fourier coefficients, s is the wavenumber index, and ns is the sample size (number of longitudinal grid points). To compute these terms, a fast Fourier transform (FFT) was applied to each fluctuation component, excluding the negative frequencies. This FFT results in a N/2 long set of individual wave contributions to the EP flux. Summation over all wave components in F(s) from Eq. () yields values that are almost identical to those of F in Eq. (), such that numerical uncertainties in the calculation are negligible.

Here, wavenumber represents the spatial scale of the wave, the number of waves per 360° of longitude, with wavenumber 1 (hereafter referred to as wave 1) being the largest possible zonal wavelength. In this work, the reanalysis data has been resampled to a 1 by 1° horizontal grid. The Nyquist–Shannon sampling theorem requires to have at least two data points per wavenumber to detect the wavenumber. Therefore we truncate the wave separation at wave 180, noting that even for this wavenumber the contribution could be underestimated e.g.. The effect of resampling ERA5 from 0.3 to 1° horizontal resolution on the contribution of smaller-scale waves is discussed in more detail in Sect.  The wave dissipation given by the ∇⋅F and the residual circulation are connected through the DWCP. The DWCP states that the mass flow across a given level is related to the momentum deposition from dissipating waves above that level. We use this relation to estimate the relative contributions of different waves to the driving of the residual circulation. Based on the DWCP, the following expressions can be derived for the residual mass stream function ψ‾D* and for the components of the residual circulation v‾D*,w‾D* 5aψ‾D*=∫z∞acos⁡(φ)∇⋅F[acos⁡(φ)(u‾+acos⁡(φ)Ω)]φdz,5bv‾D*=1ρ0cos⁡(φ)∂ψ‾D*∂z,5cw‾D*=1aρ0cos⁡(φ)∂ψ‾D*∂φ.

Strictly speaking, the integration in Eq. () should be performed along isolines of constant angular momentum. However, since outside the tropics the isolines of constant angular momentum are approximately lines of constant latitude, we perform the integration along constant latitudes cf.. Therefore, we estimate the mean residual circulation upwelling between 20° S and 20° N by integration of Eq. () 6w‾D*±=ψ‾D*--ψ‾D*+-aρ0S±, where the superscripts ± denote that the quantity is evaluated at the latitudes 20° N and 20° S, respectively. Hence, S± is the normalized area fraction of the latitude band between these latitudes and w‾D*± is the mean DWCP vertical velocity averaged over this region.

The turn-around latitudes (TAL) separate the area of upward residual velocity in the tropical stratosphere from the area of downward velocity in the extratropics e.g.. In this work, tropical upwelling is defined as the total mass flow between the TAL carried by w‾*, and tropical outflow is defined as the poleward mass flow across the TAL. The outflow at a given level represents the result of “gyroscopic pumping” , indicating the amount of air mass being pulled from the tropics to middle and high latitudes by the wave drag at that level. The tropical upwelling, on the other hand, represents the integrated effect of the wave drag at all levels above, as described by the downward control principle (see above). Hence, the total tropical upwelling is given by 7W=2πa2ρ0∫φ-φ+w‾*cos⁡(φ)dφ, and the outflow out of the tropics at a given level is the vertical derivative (top to bottom) of the upwelling 8V=-∂W∂z. The two major wave types which drive the overturning circulation are Rossby and gravity waves. Planetary- and synoptic-scale Rossby waves are large-scale and are related to the horizontal gradient of potential vorticity. These waves are influenced by Earth's rotation, the latitudinal gradient of the Coriolis parameter, and the mean zonal flow. The gravity waves are of smaller-scale, less dependent on the Earth's rotation and primarily result from the restoring force due to stable background stratification. The upward propagation of Rossby waves is controlled by the Charney–Drazin condition , which for stationary waves is 90<u‾<fϕk2+l2+(f0/2NH)2. Here, k and l are zonal and meridional wavenumbers, f0 is a reference value of the Coriolis parameter, and N is the buoyancy frequency. This condition implies that Rossby waves may propagate upwards into the stratosphere if the zonal flow is eastward and weaker than a critical value. This critical value is higher for larger-scale waves (smaller k, l) such that these waves may propagate deeper into the stratosphere compared to smaller-scale waves.

A common approach in BDC studies is a separation of wave-driving based on the zonal wavenumber of the involved atmospheric waves e.g.. In this study, the following scale distinction is adopted: waves 1 to 3 are categorized as planetary, waves 4 to 20 as synoptic-scale, and resolved waves with wavenumbers greater than 20 are classified as mesoscale waves. A combination of synoptic-scale and mesoscale waves will be referred to as smaller-scale waves. Accordingly, planetary waves are Rossby waves, synoptic-scale waves are dominated by Rossby waves with increasing secondary contributions due to gravity waves for higher wavenumbers., and mesoscale waves can be interpreted as gravity waves . Furthermore, planetary waves are typically forced by large-scale stationary features of Earth's surface, such as orography and land–sea contrasts, and therefore at the altitude range of interest are expected to be stationary or quasi-stationary . This interpretation is further supported by the consistency between Fig. b and the amplitudes of geopotential height variations associated with zonal wavenumber 1 in . In particular, the slowest normal mode with a 16 d period analysed by most closely resembles the structure of the Eliassen–Palm flux shown in panel b.

To address the research questions outlined in the Sect. , a Fourier transformation is applied to 6 h snapshots of the fluctuation components, and ∇⋅F(s) (with the zonal wavenumber domain replacing longitude domain) is calculated (see Eq. ). After that, the monthly mean is computed from the ∇⋅F(s) snapshots. Utilizing the DWCP framework (Eq. ), the monthly mean ∇⋅F(s) related to specific wavenumbers is used to reconstruct the circulation driven by individual waves or sets of waves. Subsequently, the associated upwelling and outflow are computed from the reconstructed circulation using Eqs. () and (). The analysis is then conducted on the resulting upwelling and outflow fields.

Figure shows the climatology of the residual mean mass circulation for the period 1980–2017 together with the resolved wave drag quantified in terms of ∇⋅F, which constitutes the circulation's driving force (see Eq. ). Note that ∇⋅F has been scaled by (aρ0cos⁡φ)-1 and 86400 s d⁻¹ to obtain the wave drag (m s⁻¹ d⁻¹). The well-known characteristics of the stratospheric BDC, with upwelling in the tropics, poleward motion and downwelling in the extratropics, and a stronger circulation cell in the winter hemisphere, are all evident in Fig.  (colours represent dissipation or “wave drag” by resolved waves).

The reduced wave drag in the middle and upper stratosphere in the summer hemisphere is related to the presence of westward winds which prohibit Rossby waves to propagate upwards, known as the Charney–Drazin criterion . The wave filtering resulting from this criterion prohibits the propagation of waves through a medium with an eastward wind speed above a certain threshold or any westward wind (see Eq. ). Consequently, as illustrated in the second and third columns of the figure, large-scale Rossby waves are more likely to satisfy the criterion and can generally propagate deeper into the stratosphere. Furthermore, the majority of small-scale Rossby waves with wavenumbers larger than 3 dissipate in the lower part of the stratosphere. Gravity waves remain unaffected by the filtering process, allowing them to propagate deeper into the atmosphere and to drive the mesospheric circulation.

The seasonality of the BDC causes seasonal movement of the TAL with significant displacement into the summer hemisphere (Fig. , dashed black lines).

The primary factor driving the tropical upwelling in the stratosphere is the wave drag near the TAL, as it is ultimately responsible for the uplift in the tropics .

The northern TAL in the lower stratosphere is typically located between 20 to 40° N. Figure  presents the time series of the wave drag and the horizontal component of the residual circulation, estimated using the TEM approach and DWCP, at 70 hPa within this latitude range. The figure shows a close relation between horizontal velocity v‾* and wave drag, as expected from Eq. (). Overall, the net outflow velocity calculated using the TEM approach agrees well with the momentum balance estimate of the velocity derived from the drag of all resolved waves. The discrepancy in residual circulation velocity between the TEM approach (green dashed line in Fig. ) and the DWCP (black dashed line) at 70 hPa is at least in part attributable to the fact that the DWCP only accounts for resolved wave drag (see Eq. ) while the full horizontal velocity based on the TEM definition is also influenced by small-scale parameterized gravity waves. Other potential contributors to the discrepancy include computational approximations and the known limitation that reanalyses do not strictly conserve mass. An effect of parameterized gravity wave drag on upwelling similar to the discrepancy between the approaches has been shown by . As a result, the DWCP estimate of the residual circulation driven by the drag of all resolved waves is weaker than the one obtained taking into account also the additional wave drag related to parameterizations, using the TEM approach. This deficit is shown in Fig.  and discussed in more detail in Sect. .

As might be expected, the wave drag at the considered level ∇⋅F (solid lines in Fig. ), and the horizontal component of the residual circulation estimated from the wave drag above using the DWCP v‾D* (dashed lines in the figure) are consistent and have similar variability. The wave driving at 70 hPa in the Fig.  shows that about three-fourths of the circulation at the level is being driven by smaller-scale waves. However despite driving only one-fourth of the circulation, the planetary waves influence the variability of the circulation almost as much as the smaller-scale waves. The variability of the circulation velocities and relation to wave drag are discussed in more detail in Sect. .

The wave filtering due to the Charney–Drazin criterion related to the pattern of zonal wind (Fig. ) limits the propagation of smaller-scale Rossby waves into the middle and upper stratosphere. As a result, the smaller-scale waves that do reach the stratosphere are expected to break and deposit their momentum at lower levels. Thus smaller-scale waves are expected to drive the shallow branch and planetary waves to drive the deep branch of the residual circulation .

Analysis of the resolved wave drag demonstrates that the large-scale waves (planetary- and synoptic-scale) generate most of the upwelling in the tropical lower stratosphere (Fig. a). The separation of horizontal outflow into contributions from different waves in Fig. b further shows that at upper levels planetary waves dominate, while at lower levels (below about 22 km) synoptic-scale waves dominate. This indicates that the deep branch is mainly a result of planetary waves drag, while the shallow branch is mostly driven by synoptic-scale waves. Furthermore, drag due to mesoscale waves almost exclusively contributes to the shallow branch. This drag induced by the breaking of mesoscale waves is significantly stronger in ERA5 than in the other three reanalyses (not shown).

We conduct a more detailed analysis of the specific wave driving of the deep and shallow circulation branches based on both a mechanistic and a statistical analysis. The mechanistic analysis is based on the DWCP and separates upwelling and outflow velocities into contributions of different waves. Since planetary and synoptic-scale waves account for the majority of the resolved wave drag, the contributions of the individual wavenumbers from 1 to 6 are separated in Fig. a. The vertical profiles of outflow presented in the figure clearly indicate that waves 1 and 2 drive the deep branch of the circulation at levels above about 22 km. On the other hand, wave 4 and smaller-scale waves dominate the upwelling below that level. The role of waves with wavenumber 3 and 4 is somewhat intermediate, with these waves driving a significant portion of the circulation at all levels. The other three reanalyses considered in this study exhibit similar vertical distributions of wave drag for the wavenumbers 1 to 6 (not shown). Regarding the relative outflow in Fig. b, waves with wavenumbers larger than 3 produce outflow maxima which are well-confined below about 25 km, while the outflow due to larger-scale planetary waves maximizes over broad regions above. The vertical location of the maxima of relative outflow indicates that wave 3 plays a more prominent role in driving the deep branch, while wave 4 is more strongly driving the shallow branch. The individual outflow induced by a single wavenumber up to wave 59 can be seen in Fig. , which shows that, resolved waves with scales smaller than wave 6 are almost exclusively driving the shallow branch of the stratospheric circulation.

The statistical analysis examines the correlation near the TAL between the outflow velocity and the ∇⋅F associated with different wave bands. Here, the outflow velocity is estimated as the horizontal component of the residual circulation from the TEM framework (v‾*) (multiplied by a minus sign in the Northern Hemisphere, such that a positive correlation results for increased poleward flow related to more negative ∇⋅F values). The correlation coefficient between the TEM outflow velocity and the ∇⋅F, derived from the total resolved wave drag, is approximately 0.8 to 0.9 – except in regions below 20 km altitude, where the correlation weakens (not shown). To further investigate whether waves 3 and 4 primarily drive the shallow or deep branch, the correlations of both large-scale wave drag with outflow velocity and that of smaller-scale wave drag with outflow velocity are calculated for the three cases of defining large-scale waves as 1–2, 1–3, and 1–4. Then the difference between the correlations for large-scale waves and for smaller-scale waves are calculated and profiles of these differences are shown in Fig. c, d. We expect the largest-scale waves to exhibit a stronger positive correlation within the deep branch region, while smaller-scale waves are anticipated to show a stronger positive correlation in the shallow branch region. Hence, the correlation difference profiles should be positive in the deep branch and negative in the shallow branch region, respectively. Including wave 4 into the large-scale waves results in a stronger correlation of the outflow velocity in the shallow branch region of the Southern Hemisphere with the large-scale waves compared to the smaller-scale waves (blue line in Fig. d). Additionally, incorporating wave 3 into the smaller-scale waves enhances the correlation between outflow velocity and smaller-scale wave drag at certain levels in the middle stratosphere, surpassing the correlation observed for the large-scale waves (red lines in Fig. c, d). Taking all these aspects into account, we conclude that the clearest separation between different circulation branches at upper and lower levels in terms of wave drag occurs between waves 3 and 4. Hence, we define the deep circulation branch of the stratospheric BDC to be driven by planetary waves 1–3 and the shallow circulation branch to be driven by smaller-scale waves (wavenumbers larger than 3). The following analyses of BDC structure, variability and trends, support our choice in separating the wave driving into these two classes of waves as separated by scale.

Vertical profiles of upwelling and outflow generated by planetary and smaller-scale waves were constructed for the different reanalyses and are presented in Fig. . Regardless of the reanalysis, planetary waves generate the majority of the outflow in the middle stratosphere and nearly all of the outflow in the upper stratosphere. Smaller-scale waves, on the other hand, are responsible for most of the outflow in the lower stratosphere. Compared to the other three reanalyses, ERA5 indicates a notably higher total outflow in the lower stratosphere between about 20 and 25 km. This difference is likely related to the effects of small-scale waves and the outflow they generate. Specifically, better spatial and temporal resolution of ERA5 allows for better representation of mesoscale (gravity) waves in ERA5 than in the other reanalyses e.g. . The drag caused by planetary waves is very similar between all four reanalyses. Hence, ERA5 is able to resolve more mesoscale waves and has a stronger related wave drag than other reanalyses. Consequently, in ERA5 these waves contribute significantly more strongly to the shallow branch.

Figure  further shows that the level which separates the two dynamical regimes with outflow mainly generated by planetary waves above and mainly by smaller-scale waves below emerges similarly for the different reanalyses. Therefore, this level represents a natural separation in terms of dynamical characteristics, i.e. wave driving, between the shallow and deep branches of the stratospheric circulation and is largely consistent across the different datasets considered.

The clear difference between the contributions of planetary and smaller-scale waves to the stratospheric residual circulation at upper and lower levels allows the robust definition of a separation level between a deep and shallow circulation branch in terms of wave drag. We define the separation level between the deep and shallow branches as the lowest altitude at which the contribution to outflow from planetary waves is equal to the contribution from smaller-scale waves. To eliminate sharp spikes of outflow in monthly mean data, which can cause unexpected variations in the separation level height, a rolling mean over five vertical levels is applied to the outflow profiles before determining the separation level. In cases where the monthly mean data shows equal contributions to the outflow from planetary and smaller-scale waves at multiple altitudes, we consider the separation level too vague for accurate estimation and exclude these months.

The seasonal variation of the separation level is shown in Fig. . The separation level shows a weak seasonal cycle with a somewhat lower separation level height in boreal summer months, similarly in all four reanalyses considered. The separation level height depends on the relative contributions of planetary waves and smaller-scale waves to tropical outflow. A stronger contribution from planetary waves raises the separation level, whereas a stronger contribution from smaller-scale waves lowers the separation level (Fig. b). Therefore, the seasonal cycle of the separation level is likely linked to the annual cycle of wave drag. In particular the weaker planetary wave activity during boreal summer is likely important for pushing the separation level downward.

The large variability of the separation level (standard deviation, shown by the whiskers in Fig. ) indicates a large inter-annual variation of the separation level height, which is of similar magnitude as the seasonal variation.

Furthermore, during most months the separation level is located at higher altitudes for ERA5 compared to other reanalyses (Fig. ). This difference is likely related to the ability of ERA5 to better resolve smaller-scale waves which shifts the separation level to higher altitudes compared to the other reanalyses. Given its higher resolution and ability to resolve larger parts of the mesoscale wave spectrum, we consider the estimate of the separation level height from ERA5 likely more accurate compared to the other reanalyses.

The focus of this study is on the effects of resolved waves, but we notice that contributions from smaller-scale, unresolved waves could modify our results. Since smaller-scale waves are responsible for driving the shallow branch and since the upwelling deficit increases at lower levels in the shallow branch region (Fig. ) it is likely that also the parameterized wave drag contributes significantly to the shallow branch. Therefore we expect the separation level to be located at somewhat higher altitudes for higher resolution data with a greater portion of the smaller-scale wave spectrum resolved or with parameterized wave drag included in the analysis.

Figure  shows the upwelling at 100 hPa which is close to the tropopause, at 43 hPa which is approximately the separation level between shallow and deep branch, and the difference between upwelling at 100 and 43 hPa. The upwelling at the tropopause represents the total mass flux into the stratosphere within the BDC and is related to the outflow at all levels above in the stratosphere (with a small portion of it even related to outflow at levels above the stratopause). Therefore, changes in upwelling at this level reflect changes in the overall overturning circulation. The annual mean upwelling at 100 hPa, generated by all resolved waves and estimated using the DWCP, is approximately 9.9×109 kg s⁻¹ in ERA5. In comparison, MERRA2 and ERA-Interim exhibit slightly weaker upwelling rates of about 9.5×109 kg s⁻¹, while JRA55 shows the weakest upwelling at this level, of about 9.2×109 kg s⁻¹. The upwelling variability at this level is substantial, with the standard deviations estimated from deseasonalized monthly mean upwelling ranging from 1.2×109 to 1.4×109 kg s⁻¹ across the reanalyses (Table ). At 100 hPa, the planetary waves and resolved smaller-scale waves contribute comparably to the annual mean upwelling, with approximately 53 % attributed to planetary waves and 47 % to smaller-scale waves. In contrast, the other reanalyses exhibit a slightly greater contribution from planetary waves ranging from 58 % for ERA-Interim to 61 % for MERRA2. The upwelling at 100 hPa caused by planetary wave drag has a standard deviation comparable to that caused by smaller-scale waves (Table ). Therefore, both planetary and smaller-scale waves contribute comparably to the forcing and variability of the whole overturning circulation.

divided the stratospheric circulation into a deep branch (above 30 hPa), a shallow branch (between 70 and 30 hPa) and a transitional branch (between 100 and 70 hPa). Although the lowest layer (100–70 hPa) is still influenced by tropospheric processes, our results show no clear indication of differences in wave driving for this layer compared to the layers above. Therefore, in the present study we aim at a dynamical separation of only the shallow and deep branches based on the circulation's wave driving.

The total outflow of the shallow branch can be estimated by subtracting the upwelling at the separation level (43 hPa) from the upwelling at the bottom of the stratosphere (100 hPa), as shown in Fig. b. In this 100–43 hPa layer, dissipation of the resolved waves generates an annual mean total outflow ranging from approximately 5.3×109 to 5.8×109 kg s⁻¹ depending on the reanalysis. Roughly two-thirds of this outflow can be attributed to smaller-scale waves, while only about one-third is attributed to planetary waves. However, the monthly mean standard deviations of the total outflow driven by planetary waves and smaller-scale waves indicate that both waves types contribute comparably to the variability of the total outflow in this layer, and hence to the variability of the shallow circulation branch.

The upwelling across the mean separation level is a result of the wave drag in the deep branch region above that level. The annual mean of this upwelling across the separation level is ranging from 3.9×109 to 4.1×109 kg s⁻¹ depending on the reanalysis, the majority of which is caused by planetary waves and less than 20 % by smaller-scale waves. The variability of the monthly mean upwelling across the separation level (43 hPa) is mostly controlled by planetary waves (see Fig. ). This is indicated by the fact that variability (here measured in terms of standard deviation) of upwelling induced by planetary waves is more than 2 times larger compared to the variability of upwelling induced by smaller-scale waves (Table ). Therefore, despite the clear separation in wave driving concerning the climatological mean circulation, both wave types contribute significantly to the circulation variability of the shallow branch, while the variability of the deep branch is mostly controlled by the planetary waves.

To further investigate the roles of planetary and smaller-scale waves in driving circulation variability, vertical profiles of standard deviation of tropical outflow attributed to planetary and smaller-scale waves were constructed, as well as vertical profiles of the correlation coefficients between the total outflow and the contributions by planetary or smaller-scale waves (Fig. ). Planetary waves are relevant to the variability of the circulation throughout the stratosphere and generally contribute more to the overall variability of the circulation compared to smaller-scale waves. In the middle and upper stratosphere these large-scale waves clearly dominate the circulation variability, consistent with a particularly high standard deviation of upwelling (Table ) At levels below about 25 km, the contribution of planetary waves to the variability of the circulation weakens, while the contribution of smaller-scale waves increases, such that both wave types contribute comparably (Fig. ). The generally higher impact of planetary waves on the variability of the circulation is possibly related to the fact that, the standard deviation of the upwelling at 100 hPa takes into account changes in the wave drag from that level and all levels above (see Sect. ). Filtering by the background wind according to the Charney–Drazin criterion (Eq. ) more strongly affects the smaller-scale waves, hinders their upward propagation and therefore dampens their contribution to circulation forcing at upper levels. Hence, the wave drag of the planetary waves is distributed over a larger altitude range in the stratosphere compared to the wave drag of the smaller-scale waves. In summary, planetary waves significantly contribute to the variability of the circulation at all levels while smaller-scale waves only affect the circulation variability in the shallow branch region.

Multi-decadal trends in the upwelling over the period 1980–2017 are presented in Fig. . Given the relatively short time period for trend calculation, upwelling trends are mostly not statistically significant at 95 % confidence level. It is known that BDC trends before and after about the year 2000 are affected by the opposite effects of ozone depletion and ozone recovery e.g.. Ozone depletion contributes to a BDC strengthening before 2000, strongest in the SH, while ozone recovery contributes to BDC weakening. However, dividing the analysis into the two sub-periods 1980–2000 and 2000–2017 did not enhance the significance of trends (not shown).

The trends of upwelling estimated from the TEM residual circulation velocity (black dotted lines in Fig. ) and from the DWCP (black solid lines in Fig. ) can differ significantly at different levels. While here, we mostly focus on the effects of resolved wave drag, most previous studies have concentrated on total upwelling derived from the TEM velocities e.g..

Although the upwelling trends differ strongly between different altitude ranges, several consistent patterns in the trend profiles emerge for the different reanalyses. Above 25 km, the upwelling trend estimated from the total wave drag is positive and S-shaped for all four reanalyses with a local trend maximum around 35 km. Strongest differences to the other reanalyses are found for MERRA2. In particular, at lower levels the shallow branch trend is mostly weakly positive for MERRA2 but negative for the other three reanalyses. For ERA5, ERA-Interim and JRA55 the deep branch trend is largely driven by planetary waves while for MERRA2 smaller-scale waves also contribute comparably to the trend. Furthermore, MERRA2 shows a generally more positive trend of upwelling driven by smaller-scale waves than that driven by planetary waves below about 30 km, which is opposite to the other three reanalyses. For most reanalyses and levels the agreement between trends of upwelling estimated from the total resolved wave drag and from the TEM approach is poor. Only for ERA5 does the trend from the resolved wave drag match the total TEM velocity-based trend well between about 23 and 35 km. Also, ERA5 is the only reanalysis where the trend in the shallow branch is almost exclusively controlled by the smaller-scale waves, and which shows a significant contribution of resolved mesoscale waves to the overall trend of the shallow branch.

The robust difference in the contributions of planetary and smaller-scale waves to dominate the upwelling trends above and below about 22 km (Fig. ) corroborates our wave drag-based estimate of the separation level between the shallow and deep branch layers. In particular, different trends in dynamics are apparent in different altitude regions, indicating a decoupling of the two circulation branches regarding their long-term evolution. These differences in trends between the shallow and deep branch are consistent, at least qualitatively, for the different reanalyses.

As noted before, the upwelling at a given level is driven by the wave drag at the level and all levels above it and hence the upwelling trend at a given level is not related unambiguously to wave drag changes only at that level. A clearer relation to wave drag changes at a given level holds for changes in meridional outflow at that level. Trend profiles for outflow are presented in Fig.  for the four different reanalyses and separated into contributions from smaller-scale and planetary waves. Given the relation to upwelling in Eq. (), trends in outflow are directly related to the vertical derivative of the trends in upwelling. Therefore, the outflow trends are positive in a shallow layer above the tropical tropopause. Above, outflow shows generally negative trends throughout a broad layer between about 18–35 km (somewhat depending on the reanalysis considered), and indicating weakening of transport out of the tropics in that altitude region. Above about 35 km, outflow increases throughout an about 5km thick layer, consistent with the negative vertical gradient in upwelling trends in Fig. . Whereas most of the reanalysis-based trends are statistically insignificant, we hypothesize that the robust vertical pattern, with outflow weakening below and strengthening above about 35 km, could be indicative for an upward shift of the deep branch of the stratospheric circulation. Another robust vertical pattern is the strengthening of outflow in the lower part of the shallow branch and weakening above, which could be indicative for a downward shift of the shallow branch of the stratospheric circulation. Overall, longer time series are needed to deduce significant long-term trends of the stratospheric circulation from reanalysis data.

Given the limited period of the analysed data and large inter-annual variability of the separation level height, the trend of the separation level for any of the reanalyses is smaller than 2 standard errors and therefore not significantly different from zero. However, three out of four reanalyses show negative trends of the separation level height, while only JRA55 has a slightly positive trend (Fig. ). The negative trend of the separation level can also be estimated from Fig. , which shows that the trend of large-scale waves is generally more positive than the trend of small-scale waves around the separation level. Hence the contribution of planetary waves around the separation level increases over time compared to the contribution of smaller-scale waves, and consequently pushes the separation level slightly downwards.