On the hiatus in the acceleration of tropical upwelling since the beginning of the 21st century

Chemistry–climate models predict an acceleration of the upwelling branch of the Brewer–Dobson circulation as a consequence of increasing global surface temperatures, resulting from elevated levels of atmospheric greenhouse gases. The observed decrease of ozone in the tropical lower stratosphere during the last decades of the 20th 5 century is consistent with the anticipated acceleration of upwelling. However, more recent satellite observations of ozone reveal that this decrease has unexpectedly stopped in the ﬁrst decade of the 21st century, challenging the implicit assumption of a continuous acceleration of tropical upwelling. In this study we use three decades of chemistry-transport-model simulations (1980–2013) to investigate this phenomenon and resolve 10 this apparent contradiction. Our model reproduces the observed tropical lower strato-sphere ozone record, showing a signiﬁcant decrease in the early period followed by a statistically robust trend-change after 2002. We demonstrate that this trend-change is correlated with corresponding changes in the vertical transport and conclude that a hiatus in the acceleration of tropical upwelling occurred during the last decade. 15


Introduction
The issue of whether the large-scale Brewer-Dobson Circulation (BDC) has strengthened in the recent past, as a result of anthropogenic activity, has been raised (Oman et al., 2009;Butchart et al., 2010;Randel and Jensen, 2013).Recent chemistry-climate model (CCM) simulations predict an increase of resolved wave activity and orographic gravity wave drag resulting from increasing sea surface temperatures (Garcia and Randel, 2008;Oman et al., 2009;Waugh et al., 2009;Butchart et al., 2010;Garny et al., 2011).This strengthens the upwelling branch of the BDC, commonly referred to as the tropical upwelling.In comparison, the behaviour of the observations is ambiguous.The long-term cooling of the tropical lower stratosphere (LS, about 17-21 km; Thompson and Solomon, 2005;Young et al., 2012)  spheric quasi-biennial oscillation (QBO; Kawatani and Hamilton, 2013) are consistent with the predicted increase of upwelling.On the other hand, the mean residence time of air parcels in the stratosphere (age of air) inferred from sulfur hexafluoride (SF 6 ) measurements is inconsistent with an overall acceleration of the BDC (Engel et al., 2009;Stiller et al., 2012).They indicate no significant changes or even deceleration of the vertical transport in the middle stratosphere.To reconcile the observed discrepancies it has been argued that the individual branches of the BDC are evolving differently, i.e. an increase in tropical upwelling does not necessarily imply an acceleration of the overall circulation (Bönisch et al., 2011;Diallo et al., 2012;Lin and Fu, 2013).Ozone (O 3 ) is a sensitive proxy for vertical transport in the tropical LS (Randel et al., 2006;Waugh et al., 2009;Randel and Thompson, 2011;Polvani and Solomon, 2012).Its local mixing ratio is considered to result from a stationary state involving production by oxygen (O 2 ) photo-dissociation and a steady influx of O 3 -poor tropospheric air from below (Avallone and Prather, 1996;Waugh et al., 2009;Meul et al., 2014).Meridional mixing from higher latitudes is a secondary effect that contributes to the seasonality in the O 3 mixing ratios (Ploeger et al., 2012).Several studies have reported a negative trend of O 3 in the tropical LS in the range of −(3-6) % per decade, consistent with the CCM predicted increase of tropical upwelling (Randel and Thompson, 2011;Sioris et al., 2014;Bourassa et al., 2014).In contrast, more recent O 3 observations from various satellite instruments indicate no statistically significant decrease of LS O 3 since the beginning of the 21st century (Kyrölä et al., 2013;Eckert et al., 2014;Gebhardt et al., 2014). 2 Data and analysis

Observations
For a quantitative analysis of tropical upwelling, we use combined O 3 observations from satellite instruments and sondes.The earlier decades  are covered by the ERBS/SAGE II instrument (McCormick et al., 1989), providing O 3 profiles based on solar occultation measurements.Due to its viewing geometry, the vertical resolution of the profiles is high (1 km, range 15-50 km), although the horizontal sampling is relatively sparse (global coverage in 1 month).Here we use version 7.0 of the data (Damadeo et al., 2013), screened for cloud and aerosol contaminated profiles as suggested by Wang et al. (2002).Two years of data after June 1991 has been omitted due to contamination by the eruption of Mt.Pinatubo.For the last decade (2002-2012), we use O 3 observations from ENVISAT/SCIAMACHY (Burrows et al., 1995) based on limb geometry (retrieval version 2.9; Sonkaew et al., 2009).The vertical resolution is about 3-4 km over an altitude range of 10-75 km; global coverage is achieved every 6 days.Data from both instruments has been binned into monthly samples on a uniform horizontal and vertical grid (15 • lon.× 5 • lat.× 1 km).To minimise sampling issues and taking into account the differences in horizontal and vertical resolution of the instruments, any further analysis is based on partial columns of O 3 between 17-21 km and 20 • N-20 • S, similar to the approach of Randel and Thompson (2011).
The satellite data is augmented by an ensemble of tropical sonde measurements from the Southern Hemisphere Additional Ozonesondes network (SHADOZ;1998-2013;Thompson et al., 2003Thompson et al., , 2012)).We use 10 sites located in the tropics with long and continuous records.The selected stations along with their temporal coverage and mean value are listed in Table 1.Typically there are 2-4 observations per month for each SHADOZ station, which provide O 3 profiles in a considerable higher vertical resolution (50-100 m) compared to the satellite instruments.As there is a high degree of longitudinal symmetry in the stratospheric ozone profiles (Thompson et al., 2003), we average the individual records to obtain a representative mean for the tropics.

Model
To obtain a consistent timeseries of LS O 3 of the last decades for direct comparison with observations, we conducted a 33-year simulation with the Bremen threedimensional chemistry-transport-model (B3DCTM; Sinnhuber et al., 2003;Aschmann et al., 2009;Aschmann and Sinnhuber, 2013).The current version of the model has a horizontal resolution of 3.75 • lon.× 2.5 • lat.and covers the vertical domain from the surface up to approximately 55 km using a hybrid σ − θ coordinate system (e.g., Chipperfield, 2006).The vertical resolution in the tropical LS is about 600 m.The model is driven by 6-hourly input of European Centre for Medium-range Weather Forecast (ECMWF) Era-Interim (EI; Dee et al., 2011) reanalysis data.Vertical transport in the purely isentropic domain (above ≈ 16 km in the tropics) is prescribed by EI all-sky heating rates.The B3DCTM incorporates a comprehensive chemistry scheme originally based on the chemistry part of the SLIMCAT model (Chipperfield, 1999), covering all relevant photochemical reactions for stratospheric O 3 chemistry.Reaction rates and absorption cross sections are taken from the Jet Propulsion Laboratory recommendations (Sander et al., 2011).To avoid initialisation artefacts, the model has been run with replicated input data to reach steady state before starting the actual integration from January 1979 to October 2013.

Regression
The multivariate regression analysis used throughout this study is based on Reinsel et al. (2002) with Y t as the monthly mean variable to be fitted: Here, μ is the baseline constant, S t a seasonal component, ω 1,2 are the trend coefficients with X 1,2t as linear trend functions and N t represents the unexplained noise.Introduction

Conclusions References
Tables Figures

Back Close
Full Note that in contrast to previous studies, which examined LS O 3 (Randel and Thompson, 2011;Sioris et al., 2014), our regression model assumes two linear components, which account for a possible change of trend.Here, ω 1 is the linear trend up to a specified inflexion date T 0 .After T 0 , the new linear trend ω comprises the sum of the earlier trend ω 1 and the trend-change component ω 2 .(Snow et al., 2014).
Assuming first order autocorrelation noise (AR(1) model), as commonly used in the regression of O 3 timeseries (e.g., Reinsel et al., 2002;Jones et al., 2009;Sioris et al., 2014), the corresponding errors for the trend components calculate as Here, σ N is the standard error of the fit residuals, n 0 , n 1 are the numbers of years of data before and after the trend-change, respectively, with n = n 0 + n 1 .φ represents the autocorrelation of the residuals with a time lag of 1 month.The choice of the inflexion year T 0 is essentially a free parameter in the regression analysis.Figure 1 illustrates the impact of the choice of T 0 on the regression of modelled LS O 3 columns and EI upward mass flux (as discussed below in Sect.3).A 2σsignificant trend-change (ω 2 ) is obtained for a wide range of possible inflexion years (marked by red circles).Therefore we use a χ 2 test based on the fit residuals, similar to the approach described by Jones et al. (2009), to identify the most probable inflexion year.We find a clear minimum in the χ 2 values close to 2002 and consequently select this year as the turning point in the trend analysis.8.2 % per decade (ω 2 ) is statistically significant within the 95 % confidence interval (i.e.ω 2 > 2σ ω ).

Lower stratosphere ozone column
To apply our analysis to the observational data we merge the available datasets (SAGE II-SCIAMACHY; SAGE II-SHADOZ).In either case, the correlation between LS O 3 partial columns exceeds 0.8 in the overlap period and the bias is generally lower than 2 %.Considering the good agreement between the observations it is reasonable to combine them into a continuous timeseries.We adopt the method described in Randel and Thompson (2011) and simply join the two individual timeseries and average the overlap period.When we apply the regression to the combined SAGE II-SCIAMACHY timeseries, we calculate a trend of −3.9 ± 0.5 % per decade (ω 1 ) for the pre-2002 period, consistent with the range of −(3-6) % per decade given by earlier studies (Fig. 3c  and d; Randel and Thompson, 2011;Sioris et al., 2014;Bourassa et al., 2014).The discrepancy between model and observations for the pre-2002 trend is likely caused by the O 3 high-bias between 1985-1990, mentioned above, and the possibly overestimated vertical transport velocity in the EI dataset (Ploeger et al., 2012;Diallo et al., 2012), which is discussed below in more detail.After 2002 the trend is 0.5 ± 1.5 % per decade (ω), yielding a statistically significant trend-change of 4.4 % per decade (ω 2 ).We obtain similar values (−3.6 ± 0.5, 0.4 ± 1.4 % per decade for ω 1 , ω) if we use the SHADOZ data instead of SCIAMACHY in the combined dataset.Consequently, both observational and model data show that the decrease of LS O 3 has effectively stopped since about 2002.This is in qualitative agreement with those studies, which focus solely on the most recent observational record of O 3 , although the differences in utilised regression models and timeseries length make a direct comparison difficult.Local chemical effects can be largely ruled out as explanation for the detected trendchange of LS O 3 .As stated above, O 3 abundance in the tropical LS is mainly determined by vertical transport and O 2 photolysis (Avallone and Prather, 1996;Waugh et al., 2009;Meul et al., 2014).O 3 -destroying catalytic species are scarce in the tropical LS, therefore the phase-out of ozone-depleting substances (ODS), and the associated recovery (e.g., World Meteorological Organization, 2011), has no direct impact on O 3 concentrations in this region.However, some studies point out a possible indirect relationship between ODS-related polar O 3 depletion and tropical LS O 3 by dynamical coupling (Waugh et al., 2009;Oman et al., 2009).Meul et al. (2014) predict an increase of photolytic O 3 production as a result from long-term changes in the overhead O 3 column.Furthermore, an increase in odd nitrogen (NO x ) might lead to additional O 3 production.However, neither process is sufficient to explain a short-term trend-change.Overall the most probable explanation of the observed behaviour is that changes in dynamics must be involved.

Tropical upwelling
Some studies point out that the increase of tropical upwelling may be compensated by an, as yet, unexplained weakening or shifting of tropical mixing barriers (Stiller et al., 2012;Eckert et al., 2014).However, it is also possible that the increase of tropical upwelling itself has ceased.To investigate this hypothesis, we analyse tropical upwelling in the EI reanalysis that drives our model.A typical representative quantity for the tropical upwelling is the upward mass flux at 70 hPa (≈ 18.5 km in the tropics; Butchart et al., 2010;Seviour et al., 2012).A recent study assessing the upward mass flux in EI found a negative trend of −5 % per decade for the years 1989-2009, based on EI kinematic vertical winds (Seviour et al., 2012).This is in contradiction with the results of current CCM, which predict an increase of upwelling of about 2.0 % per decade (ensemble mean; Butchart et al., 2010).The quality of stratospheric vertical transport in EI improves considerably, when diabatic heating rates are used instead of the kinematic wind.Although tending to overestimate the tropical ascent (Ploeger et al., 2012) diabatic representation of vertical transport yields more realistic estimates of stratospheric age of air in comparison to the kinematic approach (Diallo et al., 2012) and is also less dispersive (Ploeger et al., 2011).
Figure 4 shows the tropical LS EI all-sky heating rates ( 20• N-20 • S, 17-21 km; panel a), which are used to drive the vertical transport in our isentropic model, and the corresponding EI upward mass flux at 70 hPa (panel c).The upward mass flux is the integral of the residual vertical velocity w * between turnaround latitudes as described in Seviour et al. (2012).In turn, w * is calculated from the EI heating rates using the iterative algorithm described by Solomon et al. (1986).Applying the regression analysis to the upward mass flux yields a positive trend of 3.3±0.7 % per decade for the pre-2002 period (Fig. 4d).This value is consistent with the CCM results (2.0 % per decade) although somewhat high-biased, reflecting the overestimation of vertical transport mentioned above.After 2002, however, there is a statistically significant trend-change around 2002 leading to a negative trend of −2.3±2.5 % per decade mirroring the trend-change in the LS O 3 timeseries.We find significant anti-correlation between LS O 3 anomalies with either heating rates (−0.83), or upward mass flux anomalies (−0.55).Taking into account the known sensitivity of LS O 3 to vertical transport, we conclude that the observed trend-change in O 3 is primarily a consequence of the simultaneous trend-change in tropical upwelling.

Conclusions
In summary, we find a negative trend of tropical LS O 3 in observations and model before 2002, associated with a positive trend in tropical upwelling from the EI dataset based upon diabatic heating calculation.This finding is consistent with earlier studies (Butchart et al., 2010;Randel and Thompson, 2011).We also find an unexpected hiatus of the negative trend in LS O 3 during the last decade.We explain this behaviour by the change of tropical upwelling evident in the EI dataset.This change may be a consequence of the unexpected La-Niña-like cooling of the equatorial Eastern Pa-Introduction

Conclusions References
Tables Figures

Back Close
Full cific since the beginning of the 21st century (Meehl et al., 2011).The latter has a significant impact on global surface temperatures (Kosaka and Xie, 2013) and ultimately, by dynamical coupling, on tropical upwelling (Oman et al., 2009;Butchart et al., 2010;Garny et al., 2011).Recent studies describe the associated circulation changes (England et al., 2014) and their impact on tropospheric O 3 (Lin et al., 2014).In contrast to current unconstrained CCM, which generally do not predict this exceptional heat uptake by the equatorial Eastern Pacific (Kosaka and Xie, 2013;England et al., 2014), this feature can be clearly observed in the data-assimilated EI dataset (Fig. 5).In conclusion the accuracy of our predictions of future BDC development and its consequences for stratospheric O 3 critically depends on our understanding of the ocean-atmosphere interaction.Figures

Back Close
Full  Full and the observed weakening of the strato-Discussion Paper | Discussion Paper | Discussion Paper | Stimulated by the need to explain the unusual linear trends revealed from the vertical profile of O 3 retrieved from SCIAMACHY 1 we use three decades of O 3 observations and simulations to investigate this phenomenon.Section 2 describes the observations, model and regression analysis used in this study.The results are discussed in Sect.3Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

Figure 2
Figure 2 presents tropical LS O 3 column anomalies (20• N-20• S, 17-21 km) from measurements and the simulation.The agreement between model and observations is good, except for a small high-bias relative to the earlier SAGE II data(1985)(1986)(1987)(1988)(1989)(1990) of approximately 1 DU: correlation coefficients are 0.65 between modelled and observed datasets.A decline of O 3 is evident in the tropical LS during the first two decades, both in the observed and modelled timeseries.This is consistent with an increase of tropical upwelling during this period.However, this trend vanishes in the third decade(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013).Figure3a and billustrates the results from the regression analysis of the modelled timeseries showing the fit function and the corresponding residuals, respectively.The linear trend amounts to −8.1 ± 0.9 % per decade (ω 1 ) in the pre-2002 period and 0.1 ± 3.3 % per decade (ω) for the remaining years.The resulting trend-change of Gebhardt et al. (2014) compared several satellite instruments and report consistently positive trends of tropical O 3 between 17-21 km, ranging from about 2 (OSIRIS), 4 (SCIAMACHY) up to 14 % per decade (MLS), covering the years 2004-2012.Eckert et al. (2014) find a slightly positive trend of 0-1 % per decade in the same region in MIPAS observations Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Polvani, L. M. and Solomon, S.: The signature of ozone depletion on tropical temperature trends, as revealed by their seasonal cycle in model integrations with single forcings, J. Geophys.Res.-Atmos., 117, D17102, doi:10.1029/2012JD017719,2012.9953 Randel, W. J. and Jensen, E. J.: Physical processes in the tropical tropopause layer and their roles in a changing climate, Nat.Geosci., 6, 169-176, doi:10.1038/ngeo1733,2013.9952 doi:10.1002/joc.2336,2011.9956 World Meteorological Organization: Scientific Assessment of Ozone Depletion: 2010, Global Ozone Research and Monitoring Project-Report No. 52, Geneva, Switzerland, 2011.9959 Young, P. J., Rosenlof, K. H., Solomon, S., Sherwood, S. C., Fu, Q., and Lamarque, J.-F.: Changes in stratospheric temperatures and their implications for changes in the Brewer-Discussion Paper | Discussion Paper | Discussion Paper |

Fig. 1 .Fig. 2 .
Fig. 1.The dependence of the linear fit parameters ω 1 , ω 2 and ω (ω 1 + ω 2 ) on the inflexion year T 0 is shown for the regression of modelled tropical LS O 3 column (a) and EI upward mass flux at 70 hPa (b).Red circles denote the years where the trend-change (ω 2 ) exceeds the 95 % confidence threshold.The black lines are the normalised χ 2 values of the fit residuals.

Fig. 3 .
Fig. 3. Regression analysis of observed and simulated O 3 partial columns.Model and combined SAGE II/SCIAMACHY LS O 3 with regression function (a, c).Corresponding fit residuals excluding the linear terms (b, d).The dashed red lines depict the resulting linear trends before and after 2002.
The additional regression terms are QBO t for QBO, ENSO t for the El Niño Southern Oscillation (ENSO) and SC t for solar cycle.The QBO proxy consists of the QBO.U30 and QBO.U50 (zonal wind 30/50 hPa) from the NOAA Climate Prediction Center 2 , the ENSO proxy is represented by the Multivariate ENSO Index (MEI) from the NOAA Earth System Research Laboratory 3 (Wolter and Timlin, 2011) lagged by two months and the solar cycle by the Bremen composite Mg II index 4

Table 1 .
Geolocation, temporal coverage and average LS O 3 column of utilised SHADOZ sites.