Only sporadic information is available on the QBO before 1953. The first indirect indications of stratospheric wind variability relate to observations of volcanic plumes (Hamilton, 2012). The most famous example is the observation of the Krakatau volcanic plume in 1883, which circled the globe from east to west. The high-altitude winds (the stratosphere was not yet discovered) responsible for this transport became widely known as “Krakatau easterlies”.

Direct observations of equatorial stratospheric winds by means of balloons go back to 1908, when Berson, in an expedition to East Africa, reported unexpected westerly winds in the lower stratosphere (Süring, 1910). These westerlies were confirmed by van Bemmelen and Braak (1910), who performed observations of upper-level winds in Batavia from 1909 to 1918. Lower stratospheric westerlies were also confirmed by the observations of another volcanic eruption plume (Semeru, 15 November 1911), as reported by Hann and Süring (Hamilton, 2012). Reconciling Berson's westerlies with the expected easterly winds remained a challenge until the discovery of the QBO in the 1960s (Hastenrath, 2007).

Stratospheric wind observations were very sparse prior to the 1950s. The early results were summarized by Schove (1969) and Hamilton (1998). After the 1950s, when a global radiosonde network was built up, stratospheric winds were operationally observed in the equatorial region. It was in these data that Reed et al. (1961) and Veryard and Ebdon (1961) discovered the QBO. Based on radiosonde records from Canton Island (3∘ S, 172∘ W) from 1953 to 1967, Gan (Maldives, 1∘ S, 73∘ E) from 1967 to 1975 and Singapore (1∘ N, 104∘ E) since 1976, Naujokat (1986) and Marquardt and Naujokat (1997) were able to derive the QBO time series back to 1953, known as the FUB QBO. Since the advent of reanalysis data, the QBO is normally defined as the zonally averaged zonal wind in the stratosphere at the equator. We follow this definition.

In this paper we use the monthly reconstruction of the QBO (zonal mean zonal winds at the equator) from Brönnimann et al. (2007). This reconstruction is based on the surface signature of the QBO-modulated solar semidiurnal tide in hourly surface pressure observations from Batavia prior to 1945 as well as on historical upper-air wind profiles. For the reconstruction we first defined a perpetually repeating “ideal QBO cycle” from deseasonalized reanalysis data. Then we used the observational evidence to determine a time axis (i.e. timing of phases) and interpolated the ideal cycle onto this new time axis. Finally, we added back the annual cycle. We used historical total ozone data (which also show an imprint of the QBO) to assess the reconstruction and found generally good agreement, but the real QBO might be out of phase by up to 3 or 4 months.

The reconstruction is supported by historical upper-air observations mainly in the 1910s and in the 1940s, while the solar semidiurnal tide provides continuous information but stops in 1945; afterwards the reconstructions are entirely based on upper-air wind observations. The Freie Universität Berlin QBO starts in 1953. From September 1957 on, the QBO is taken from ERA-40 and after 1979 from ERA-Interim (Dee et al., 2011). The resulting 108 year QBO record is given in Fig. 1. We are currently in the 48th cycle since Berson's profile of 1908, which we take as a starting point of our work. The number of cycles thus allows robust statistics.

The data presented here are part of a set of 1.25 million upper-air profiles that were digitized in the framework of the ERA-CLIM project (Stickler et al., 2014a, b), adding to the 12.75 million upper-air wind profiles that were already available from the comprehensive historical upper-air data set prior to 1957 (Stickler et al., 2010). A plot of most of the equatorial stratospheric data comprised in the latter data set was already given in Labitzke et al. (2006) and they entered the reconstruction described above. For this paper we collected all additional (ERA-CLIM) data prior to 1950 from stations within 20∘ S to 20∘ N. In the following we highlight three particular records.

In 1908, the German meteorologist Arthur Berson organized an aerological expedition to East Africa with the aim of better understanding the monsoon system (Fig. S1 in the Supplement shows the launch of a registering balloon on Lake Victoria). Upper-level winds were observed with pilot balloons and registering balloons (briefly described in Brönnimann and Stickler, 2013). Only a few profiles reached the stratosphere. Surprisingly, some of them indicated westerly winds in the stratosphere. Figure S1 (right) shows the wind profiles that reached the stratosphere. Although all profiles except for two were taken during a 15-day interval, there is considerable scatter. It is very difficult to identify wind regimes from the raw data, although there are westerly winds in the stratosphere in several profiles.

The corresponding profiles from the reconstructions are also indicated. While there is good agreement with some of the profiles (westerlies between 18 and 20 km and easterlies above), others show relatively strong easterlies between 16 and 18 km (or even higher), where the reconstructions suggest zero zonal wind. Note that the reconstructions assumed westerly winds at 19 km altitude throughout the year 1908 based on the notion of Berson westerlies.

In 1909 the Dutch colonial secretary started aerological observations in Batavia (van Bemmelen, 1911), supported by leading aerologists such as Richard Assmann and Hugo Hergesell. Kites, registering balloons and pilot balloons were used. Many of the balloons reached high altitudes, and soon westerly winds were observed (van Bemmelen and Braak, 1910), thus confirming Berson's findings. Veryard and Ebdon (1961) and Ebdon (1963) analysed the Batavia winds from 1909 to 1918 (which they published in the form of monthly averaged wind directions for certain altitude bands) and found a clear QBO signature, including the downward phase propagation. Their published phases are interpretations, not raw data, and these phases were used to constrain our reconstructions. However, from the digitized data (Fig. S2 shows the earliest phase of measurements) it is difficult to discern clear wind regimes.

From 1925 to 1927, the German research vessel Meteor cruised the Atlantic and took aerological observations in addition to many oceanographical measurements. More than 1000 vertical profiles were retrieved on east–west transects across the tropical and South Atlantic (see Stickler et al., 2015, for details). Apart from kites and some registering balloon ascents (none of which reached the stratosphere), 801 pilot balloon ascents are available, of which the highest reached 20.5 km. In the tropical region (20∘ N to 20∘ S), however, only a few measurements higher than 16 km are available.

These measurements (as well as those from Berson, Batavia and all other measurements from the ERA-CLIM data set) are incorporated as circles into Fig. 1. Again it is difficult to find a coherent picture. Counting the agreement of the zonal wind between observations and reconstructions, we find a rate of 60 %. This is not a particularly good score for an evaluation, which is not even fully independent. The rate increases, though, if we only use observations above 19 or 20 km or exclude comparisons for which reconstructed winds are weak (i.e. close to phase change). Conversely, there is no systematic pattern of disagreement (no out-of-phase relation). We therefore continue and use our reconstructions for further analyses but note that the reconstruction remains to be further confirmed.

Several mechanisms responsible for QBO influence on tropospheric climate have been proposed. One pathway, known as the Holton–Tan effect (Holton and Tan, 1980; Baldwin et al., 2001), is through the extratropical stratosphere in boreal winter. This mechanism is understood to operate via changes in the extratropical planetary wave activity flux. An easterly QBO at 50 hPa leads to convergence of wave activity in the subtropical lower stratosphere and in the subpolar middle and upper stratosphere (e.g. Garfinkel et al., 2012). The waves deposit easterly momentum and decelerate the mean flow. The signal can propagate downward and eventually reach the Earth's surface, although the mechanism is still not fully understood (see Anstey and Shepherd, 2014; Kidston et al., 2015; a review of the proposed mechanisms is beyond the scope of this paper).

Compositing easterly minus westerly QBO in boreal winter in ERA-Interim since 1979 (Fig. 2) shows this classical response. The zonal mean zonal wind weakens, most strongly at around 30 km, 70∘ N. Cooling is found above and warming is found below. GPH exhibits positive anomalies poleward of 60∘ N in the lower stratosphere, indicative of a weak polar vortex. A corresponding cross section for 20CRv2c and ERA-20C (Fig. S4) shows that, while ERA-20C qualitatively reproduces the pattern found in ERA-Interim, 20CRv2c does not reproduce the pattern in the stratosphere, but all data sets agree in the troposphere.

Reconstructions and reanalyses both do not provide information for altitudes above around 10–15 km. The highest level we analyse here is 100 hPa (note that this might be too low to capture QBO effects but too high to be well reconstructed). In the Z100 index (Table 1) we find no significant effect in any of the subperiods. The early period even shows a negative difference (thus, opposite to what is expected from Fig. 2). A more consistent relation is found within the model simulations, which is highly significant for the pooled sample and is significant at the 95 % (90 %) level for two (three) out of the four simulations (not shown).

Compositing 100 hPa GPH and 200 hPa zonal wind for January to March gives similar results (Fig. 3). The analysis of 20CRv2c (1908–2012) and of the simulations (ensemble mean) show almost opposite patterns, but it should be noted that there is hardly any significance in the 20CRv2c composites (Fig. S5; no significance at all is found in 100 hPa GPH). In order to test whether uncertainties in 20CRv2c in the early times could be the cause of that, we compared the composite for 100 hPa GPH for the 1908–1957 period between 20CRv2c, ERA-20C and statistical reconstructions (Fig. S6). In fact, there are again some differences between the products. 20CRv2c exhibits a stronger negative signal over the polar region than the other data sets, but none shows the weakening of the vortex expected from the Holton–Tan effect.

The difference in the QBO imprint between the individual model simulations (Fig. S7) is smaller than between model and 20CRv2c analyses. Each of model simulations shows a significant signature over the polar region as well as between 35 and 45∘ N. The weakening of the zonal wind found in ERA-Interim in the recent period is reproduced qualitatively in the model simulations but not in 20CRv2c.

The lack of a consistent signal in two subperiods could point to the lack of a signal in general or to an intermittent behaviour of the QBO signature. For instance, the base state of atmospheric circulation might have changed (e.g. Vecchi and Soden, 2007), which would then modulate the QBO response through changing either the amount of upward propagating wave activity or its refraction. However, historical reanalyses are unsuitable for analysing changes in wave activity diagnostics.

The model results suggest that the QBO signal might be small (though significant), such that short periods may by chance show an opposite relation. To test this, east–west differences in 30-year moving windows are analysed (Fig. 4). In fact, the standardized differences for such periods vary considerably in the model. In the observation-based data, the difference was largest for the interval 1960 to 1989, i.e. close to the time window in which the Holton–Tan effect was originally discovered (although the behaviour changed during the 1977–1997 period; see Lu et al., 2008). Note, however, that 20CRv2c shows a weaker signal than the other data sets also in the last period.

In all, the reconstructed QBO and polar vortex strength at 100 hPa from reconstructions and reanalysis prior to 1957 do not show a relation. This could be due to inferior data quality of either or a too-low analysis level (note that ERA-Interim also does not show a significant response at that level (Table 1) but only at higher levels, Fig. 2). The model does show a significant relation as expected from the Holton–Tan effect, but the signal is rather small or transient. Results are consistent with ERA-Interim, considering decadal variability as found in the model.

To assess whether the QBO affects the surface the ERA-Interim analysis (Fig. 2) is again consulted. Zonal averages indeed show small surface effects (weaker zonal wind, higher SAT and pressure for easterly phases) but only poleward of 80∘ N. Very often, the NAO index is analysed as an indication for surface imprints of stratospheric perturbations. This index is well constrained in 20CR (Compo et al., 2011), and hence the NAO index is treated similarly as Z100. As expected, differences in the NAO have the opposite sign to those for Z100. However, none of the differences in 20CRv2c are significant. In the model simulations, differences are significant at the 90 % level in one out of four simulations (the p value for the pooled sample is exactly 0.05). The 30-year moving window composite of the NAO index shows that the decadal variability of the difference is large (both in observations and model simulations), but it is anti-correlated with that of Z100 (which is expected).

Perhaps the simple dipole-NAO index is not suitable for capturing the response. Therefore, we also analysed composites of SLP fields. We find negative anomalies at midlatitudes stretching from the eastern North Atlantic to central Eurasia in both 20CRv2c and model simulations. Indeed, the pattern is shifted southward as compared to a classical NAO pattern. In the North Atlantic–European region, the agreement between model and observation-based data is stronger at the surface than in the stratosphere (note that the pattern over North America, in contrast, is almost opposite in 20CRv2c and model simulations). In the model, the signature is consistent in all four simulations (Fig. S8). Thus, both the signatures in the stratosphere and in SLP are consistent with the Holton–Tan effect, but the variability is so large that even with very long records results remain near the limit of significance.

Baur (1927) analysed the 100-year record of Berlin SAT and found a very clear quasi-biennial cycle. Within our QBO reconstructions, we also find highly significant differences for Berlin SAT (we used observations rather than reanalyses) in winter (January–March) between eastern and western phases of the QBO (50 hPa, November–December). SAT is higher during easterly phases of the QBO. This is unexpected since Berlin SAT is positively correlated with the NAO and negatively with Z100. The effect might be real since the model simulations (grid point 15∘ E, 50.1∘ N) also show higher SAT during the easterly phase of the QBO as compared to the westerly phase, albeit not significantly.

Interestingly, the difference is significant only in the first period (which is when Baur, 1927, discovered 2.2-year cyclicities in Berlin SAT) and over the entire period, but not in the post-1957 period. In other words, the difference was significant in the period when no effect in Z100 and NAO was found. The 30-year moving window difference in the model simulations shows a similar behaviour. There are multidecadal periods when the QBO signature in Berlin SAT is positive while the NAO (Z100) signature is around zero, and there are periods when the NAO signature is negative and the QBO signature in Berlin SAT is around zero. As for the raw series, the 30-year moving window difference series of Berlin SAT, -1 × Z100 and NAO are positively correlated (numbers in Fig. 4). However, the curves do not scatter around the one-to-one line, as could be expected, but are slightly displaced towards the upper left quadrant (illustrated by their long-term average as dots). A possible explanation for this behaviour is a dipole-like variation that is induced by the QBO, but the dipole structure itself shifts with changes in the background climate. The locations or indices considered then do not capture the dipole structure well anymore and do not show a strong signature. Part of the decadal variability in the QBO–surface climate relation might thus arise from decadal climatic variability such as latitudinal shifts of circulation features.

We analysed the effect of a possible change in the base state using standard circulation indices (such as the NAO itself, the Atlantic Multidecadal Oscillation or the Pacific Decadal Oscillation, all filtered with the same moving window), but we did not find consistent results between model- and observation-based analyses. In the 1940s, the subtropical jet was in a relatively poleward position and then retracted equatorward (Brönnimann et al., 2015). This might be a possible explanation, but further evidence for this is required.

The composite field of SAT based on 20CRv2c (HadCRUT4v shows similar results) reveals that the warming for easterly phases stretches across much of Eurasia. It seems well reproduced in the model simulations, where it maximizes between the Caspian and Aral seas. However, there is quite a large discrepancy between individual simulations despite the fact that they are 405 years long (Fig. S9). For instance, the SAT signal over North America is totally different and results there are not robust.

Based on these results, we defined a new SAT index for the Caspian–Aral Sea region, which is the region with the strongest imprint in the model composites. Even in this optimized case, there are some (albeit few) 30-year periods in the 4×405-year model simulations that would exhibit a significantly negative relation when analysed in isolation. Interestingly, correlations between the 30-year moving window difference series of the new SAT index and those of NAO and -1 × Z100 are now predominantly negative.

Woeikof (1895) speculated that snow cover follows a biennial cycle. To test this, snow depth in March is analysed. The corresponding composites (see Fig. 3) were highly consistent with the results for SAT, but again they do not show a systematic effect. The high-resolution snow cover product from ERA-20C shows very similar results to 20CRv2c (see Fig. S7), i.e. the QBO east minus west differences for the two subperiods differ, and they both differ from the model simulations.

From these analyses there is no indication that snow cover in March is affected by the QBO in a significant way. Conversely, we can also not exclude intermittent effects. Peings et al. (2013) found an effect of the QBO on Siberian snow cover, but only after 1976 and not before. Hence, Woeikof (1895) might still have captured a QBO signal when finding differences in snow cover in Scandinavia between even and odd years, or (what is more likely) he was picking up random variability.

In 1992, Gray et al. (1992a, b) suggested an effect of the QBO on the ENSO system. Later publications addressed the effect of the QBO on tropical convection in observations (e.g. Collimore et al., 2003; Huang et al., 2012; Liess and Geller, 2012) or models (e.g. Giorgetta et al., 1999; Garfinkel and Hartmann, 2011; Nie and Sobel, 2015). Several mechanisms have been suggested as to how a link might proceed. Gray et al. (1992b) found that wind shear near the tropopause associated with the QBO phase in the lowermost stratosphere affects deep convection in the warm pool area. Giorgetta et al. (1999) and Huang et al. (2012) argued that the change in static stability due to the temperature QBO might play a more important role. However, the role of clouds and other feedbacks is not well understood (e.g. Garfinkel and Hartmann, 2011).

According to the wind shear mechanism, lower shear would favour deeper convection. Climatologically, easterlies dominate over the warm pool in the uppermost troposphere; hence, an easterly QBO phase at 70 hPa reduces the wind shear and enhances convection. With respect to the temperature, a westerly QBO phase at 70 hPa is associated with a warm layer below, leading to increased stability in the tropopause region and thus less convection. From both mechanisms we expect more convection during easterly phases of the QBO in the lower stratosphere. Therefore, stability and wind shear influences cannot easily be separated without more detailed diagnostics that are not available for our study.

The analysis in ERA-Interim (Fig. 5) for zonal averages over Indonesia and the Pacific warm pool (120 to 160∘ E) in boreal winter is consistent with the suggested mechanisms. While SAT and zonal wind do not show a tropospheric imprint, an increase in tropical convection is found for easterly phases. This increase is shifted towards the Northern Hemisphere relative to the climatological maximum in convection. It is statistically significant in the tropopause region.

As a first approach, we analysed the PWC index, which is well reconstructed and rather well constrained in 20CRv2c and ERA-20C (see Compo et al., 2011). We found a slight, insignificant increase in observation-based data during boreal winter. In summer, we found a significantly negative response (weakening Walker circulation during easterlies) during the ERA-Interim period in all data sets (and a response of the same sign – though not significant – in all other subperiods and data sets). In contrast, in the model the Walker circulation is stronger for easterly than for westerly QBO in both seasons (DJF and JAS). In some simulations it is even highly significant. The former is consistent with increased convection over the Pacific warm pool area and is consistent with observations (Fig. 5), whereas the latter is not consistent with increased convection and the sign is different from that found in the observations.

This imprint can be better understood when analysing fields rather than an index (although fields are less reliable). Composites of SST, vertical velocity and zonal winds at 10 m and at 200 hPa are shown in Fig. 6 (see Fig. S12 for significance). The most obvious signature is a slight eastward shift of the centre of convection over Indonesia during easterly QBO phases, resulting in the pattern seen in ERA-Interim. This is seen in both seasons (though stronger in boreal winter) and it is also seen in observations, making it robust. The response thus does not project well onto the Pacific Walker Circulation, and surface winds over the central Pacific remain unaffected. Note, however, that almost no areas show a significant response in 20CRv2c. Signatures in SST show a slight equatorial Pacific warming but a cooling in Indonesia. However, for these findings significance is also limited (see Figs. S10 and S11). The winter hemisphere subtropical jet moves poleward in 20CRv2c and to some extent also in the model simulations.

Finally, we also briefly analyse the relation between the QBO and Atlantic hurricanes or the Indian summer monsoon strength. In both cases, our results revealed no significant differences with the simple indices used. Note, however, that for the Indian summer monsoon the relation might be more complex (e.g. Claud and Terray, 2007). The existence of a tropospheric biennial oscillation in the summer monsoon has been claimed (Meehl et al., 2003), but this might arise from white noise and not from deterministic processes (Stuecker et al., 2015).