Quasi-biennial oscillation disrupted by abnormal Southern Hemisphere stratosphere

13 The quasi-biennial oscillation (QBO) is a repeating cycle of tropical stratosphere 14 winds reversing direction from eastward to westward roughly every 14 months 1 . Discov-15 ered independently by British 2 and American 3 scientists the QBO continued uninterrupted 16 for 27 cycles from 1953 until February 2016 when a westward jet unexpectedly formed 17 in the lower stratosphere during the eastward phase 4,5 . This disruption is attributed to 18 unusually high wave-momentum ﬂuxes from the Northern Hemisphere 5,6 . A second, sim-19 ilar, QBO disruption occurred during the 2019/2020 northern winter though wave ﬂuxes 20 from the Northern Hemisphere were weak. Here we show that this latest disruption to 21 the regular QBO cycling was stronger than that seen in 2016 and resulted from horizon-22 tal momentum transport from the Southern Hemisphere during abnormal winter con-23 ditions 7,8 . In both disruptions the normal downward progression of the QBO halts and 24 the eastward shear zone above the disruption moves upward assisted by stronger trop-25 ical upwelling during the boreal winter. The predictable signal associated with the QBO’s 26 quasi-regular phase progression is permanently lost during disruptions and the oscilla-27 tion reemerges after a few months signiﬁcantly shifted in phase from what would be ex-28 pected if the phase had progressed uninterrupted. We infer from an increased wave-momentum 29 ﬂux into equatorial latitudes seen in model climate projections supporting the latest In-30 tergovernmental Panel on Climate Change (IPCC) assessment that disruptions to the 31 QBO are likely to be more common in future. Consequently, we anticipate that in fu-32 ture the QBO will be a less reliable source of predictability on lead times extending out 33 to several years than it currently is. 34

manuscript submitted to Nature Geoscience

Abstract
The quasi-biennial oscillation (QBO) is a repeating cycle of tropical stratosphere winds reversing direction from eastward to westward roughly every 14 months 1 .Discovered independently by British 2 and American 3 scientists the QBO continued uninterrupted for 27 cycles from 1953 until February 2016 when a westward jet unexpectedly formed in the lower stratosphere during the eastward phase 4,5 .This disruption is attributed to unusually high wave-momentum fluxes from the Northern Hemisphere 5,6 .A second, similar, QBO disruption occurred during the 2019/2020 northern winter though wave fluxes from the Northern Hemisphere were weak.Here we show that this latest disruption to the regular QBO cycling was stronger than that seen in 2016 and resulted from horizontal momentum transport from the Southern Hemisphere during abnormal winter conditions 7,8 .In both disruptions the normal downward progression of the QBO halts and the eastward shear zone above the disruption moves upward assisted by stronger tropical upwelling during the boreal winter.The predictable signal associated with the QBO's quasi-regular phase progression is permanently lost during disruptions and the oscillation reemerges after a few months significantly shifted in phase from what would be expected if the phase had progressed uninterrupted.We infer from an increased wave-momentum flux into equatorial latitudes seen in model climate projections supporting the latest Intergovernmental Panel on Climate Change (IPCC) assessment that disruptions to the QBO are likely to be more common in future.Consequently, we anticipate that in future the QBO will be a less reliable source of predictability on lead times extending out to several years than it currently is.

Main
The QBO consists of alternating layers of eastward and westward wind that emerge above 40 km and gradually descend through the tropical stratosphere before dissipating near the tropopause (16 km) 9 .Observed periods of this irregular oscillation range from 22 to 35 months averaging around 28 months 10 .The QBO dominates stratospheric variability in the tropics while modulating variability in mid to high latitudes 11 and thereby provides a useful source of predictability on seasonal-to-decadal timescales 12 .The established QBO fluid dynamical mechanism, or canonical model, involves vertically propagating waves from the troposphere that accelerate the winds through interactions with the winds themselves 1 .This leads to the descending layers of winds of opposite sign (Fig. 1a).
Opposing the downward progression is tropical upwelling 13 from the Brewer-Dobson circulation 14 .It was thought, at least until recently, that horizontally propagating waves from the mid-latitudes into the tropics played only a minor part in the QBO's evolution 15 -explaining the QBO's remarkable cycle-to-cycle consistency, with predictability extending out to a few years 12 .Moreover, the oscillation's deep vertical structure allows 90% of the month-to-month variability of the QBO to be described by just two vertical structure modes (Empirical Orthogonal Functions, or EOFs 16 ; Methods), and the highly predictable QBO signal is associated with the time evolution of the amplitude of these two modes.
Conventional wisdom was challenged in February 2016 when the usual QBO cycling was disrupted [4][5][6]17 for the first time since its discovery in the early 1960s 2,3 . A ertically thin layer of westward winds appeared at 40 hPa, within a decaying eastward QBO phase (Fig. 1a).Anomalous westward acceleration resulted from unusually large horizontal fluxes of wave-momentum from the Northern Hemisphere (NH) 5 , linked to the occurrence of a very large El Niño event 17,18 . Conitions in the subtropics contributed to focusing the wave activity into the QBO jet 19,20 .Failures by models to predict the disruption 5 are consistent with it originating in the extratropics since predictability timescales are shorter there than in the tropics.The abnormal westward winds at 40 hPa subsequently strengthened, descended, and the QBO returned to its usual cycling by early 2017 (Fig. 1a).
A second QBO disruption began in December 2019, only four years after the previous event and without a strong El Niño being present.Here we compare the recent event to the previous one and examine their causes.We use future projections from climate models to assess whether QBO disruptions are an emerging signal of climate change, and consider the implications of these events when using the QBO as a source of predictability for lead times extending to 3-4 years.

Disruptions to regular QBO cycling
The characteristic QBO descending eastward and westward wind pattern disintegrated in 2019/20 with unexpected westward winds appearing near 40 hPa along with an atypical ascending layer of eastward winds (Fig. 1a).The small vertical scale of the ascending eastward layer is unique in the QBO record.A decomposition of the QBO winds manuscript submitted to Nature Geoscience into EOFs (Methods) quantifies this unusual vertical structure (Fig. 1b).The first two EOFs (encompassing the largest scale downward propagating structure of the QBO) typically explain over 90% of the vertical structure variance but their values drop drastically to ∼20% by May 2020 as the higher order, smaller scale, EOFs 3 and 4 grow in amplitude.This extreme 2019/20 decrease in the variance explained by EOFs 1 and 2 greatly exceeds the decrease to 60% associated with the 2015/16 disruption.
The overall rate of phase change of the QBO had been remarkably stable before the 2015/16 disruption (Fig. 1c).Constant QBO phase progression represented by the upper red line in Fig. 1c provides a reasonably accurate representation of the true phase from 1976 until the 2015/16 disruption with a standard deviation of the phase until then of ∼45 • .The 2015/16 disruption resulted in a retrogressed phase shift of ∼135 • , well outside this standard deviation.The lower red line denotes the post-disruption constant phase progression prediction, but this again failed in early 2020 when the QBO phase rapidly increased by ∼135 • , returning close to the original phase that would be expected based on the historical phase progression (upper red line).Since there are only a few months where the phase has returned near the original prediction line, it remains to be seen exactly where the QBO phase will settle after this most recent disruption.

Canonical model vs. meridional wave fluxes
For both disruptions, strong wave-forcing by meridional momentum transport (meridional EP-flux; Methods) instigated an eastward-to-westward transition of the zonal-mean zonal winds in the lower stratosphere, around 40 hPa (Fig. 2).The canonical model of the QBO explains the oscillation as resulting from a feedback between the zonal-mean zonal wind and vertical momentum transport 21,22 .Momentum deposition by upwardpropagating waves causes wind vertical shear zones to descend even as the Brewer-Dobson circulation moves the entire tropical stratosphere upward 13 .Beginning in June and lasting until September 2019, westward forcing by meridional momentum transport at 50 hPa was large compared to the net forcing from the canonical QBO model (Fig. 2a).In the context of the 1979-2020 ERA5 record this forcing was extremely large over all QBO altitudes below the descending westward shear zone (i.e. between 70 and 20 hPa; cf., red line and grey shading in Fig. 2c).In the canonical QBO model, waves deposit momentum in the zonal-mean flow over narrow altitude ranges, where they encounter strong vertical shears.However during July-September 2019 strong deposition occurred over all altitudes of the QBO eastward phase, including those well below the descending westward shear zone (Fig. 2c).
Similar features are evident for the 2015/16 disruption (Fig. 2b,d), but with different timing.Strong forcing by meridional momentum transport at 50 hPa began in November 2015 and persisted through early February 2016 when westward winds emerged near 40 hPa (Fig. 2b).Again this forcing was extremely large in the context of the 1979-2020 record and it occurred within a deep eastward QBO phase (Fig. 2d).A shallow layer of westward wind shear centered at 50 hPa, that appeared in November 2015 and strengthened over the next 3 months, is clearly visible in the December-February average vertical profile (Fig. 2d, black line).In contrast, in July-September 2019 the beginning of a shear anomaly is only just discernible as an inflection point in the wind profile near 50 hPa (Fig. 2c, black line).The westward shear anomaly then gradually strengthened from September onward (Fig. S2a and −13 m s −1 , respectively) though was spread over a longer period (Fig. S3).As with the EOF analysis, this suggests that the more recent disruption was the stronger of the two.
-5-manuscript submitted to Nature Geoscience Role of Southern Hemisphere in 2019/20 disruption Rossby waves propagate upward and equatorward from their extratropical source regions, but the tropical stratosphere is usually shielded from their incursions by a region of westward or weak eastward zonal wind in the subtropical stratosphere.This was the case near 20 hPa in July-September 2019 when the wind near 20 • S was very strongly westward compared to the other years between 1979-2020 (Fig. 3a).In contrast, near 70 hPa the SH subtropical winds were very strongly eastward compared to other years (Fig. 3b).Consequently, this allowed for an exceptionally large northward wave-momentum flux (meridional EP-flux) from 20 • S to the equator at 50 hPa (Fig. 3c).Since Rossby waves cannot propagate into the westward summer hemisphere winds, this caused large westward momentum flux convergence at the equator and corresponding westward tendency (Fig. 2a,c).
Similarly during the previous disruption equatorward wave propagation was inhibited during December-February 2015/16 by a westward NH subtropical barrier at higher altitudes (Fig. 3d) but favoured by NH subtropical winds at lower altitudes that were very strongly eastward compared to other years (Fig. 3e), allowing an anomalously large equatorward wave-momentum flux at 50 hPa (Fig. 3f).In contrast the equatorward flux from the NH in December-February 2019/20 was unremarkable, close to the median value for 1979-2020 (Fig. 3f), thereby confirming the importance of SH forcing for the 2019/20 event.Both disruptions occurred when subtropical winds favoured equatorward Rossby wave propagation at the lowermost altitudes of the QBO but not at higher altitudes.This explains why meridional momentum flux convergence did not occur at higher altitudes in the shear zones and accelerate their downward progression.
The SH winter of 2019 was unusual in that a rare minor sudden stratospheric warming (SSW) occurred, beginning in late August 7,8 .The timing of the warming roughly coincided with the peak westward forcing by meridional EP-flux in August and September (Fig. 2a).Concurrently, a large increase in tropical upwelling occurred, most likely due to the anomalous meridional overturning circulation associated with the SSW 23 , leading to increased vertical advection at the equator near 40 hPa in September 2019 that contributed to the deceleration of the eastward QBO phase at that level (Fig. S2c).The displacement of the SH polar vortex during the minor warming may also have contributed to a subtropical corridor of eastward winds at 40-50 hPa over South America enabling synoptic-scale wave propagation toward the equator in late August / early September, in a manner similar to that documented for the 2015/16 disruption 24 .However, further investigation will be required to determine whether or not the occurrence of the minor warming was essential to the 2019/20 QBO disruption.Nonetheless the common feature of both disruptions was the large equatorward meridional momentum fluxes, whatever their proximate causes.

Climate change
While the 2015/16 disruption could reasonably be judged as a "once in 50-year event" a second disruption suggests possible climate-change connections.In a warming climate the quasi-regular QBO cycling breaks down in some model projections but, in general, uncertainties in the representation of small scale gravity waves in models leads to a wide spread in QBO projections 25 and hence any projected changes in occurrences of disruptions are unreliable.On the other hand, in all multi-model QBO projections, there is an overall weakening of the oscillation in the lower stratosphere [25][26][27] , attributed 26 to the well established speeding-up of the Brewer-Dobson circulation (tropical upwelling) in models in response to climate change 14 .A weaker QBO with eastward phase persisting longer in the lower stratosphere due to a faster Brewer-Dobson circulation is more likely to be vulnerable to the effects of extra-tropical wave fluxes penetrating the equatorial latitudes.
In the climate model projections used for the latest IPCC assessment 28 this flux increases in winter and spring between the historical period 1961-2000 and 2061-2100 under Shared Socioeconomic Pathways 29 (SSP) 5-8.5 and 3-7.0 scenarios (Fig. 4).Interannual variability (standard deviation) of the monthly mean fluxes also increased, on average, and combined with the increase in the mean this implies a greater proportion of winters are likely to have sufficiently anomalous fluxes to disrupt the QBO.Robustness of the response for the mean is confirmed by remarkably similar results (not shown) obtained from idealised double and quadruple CO 2 simulations preformed for the SPARC (Stratospheric-Tropospheric Processes and their Role in Climate) QBO-initiative 30,31 .Using this novel approach of examining the more reliable response to climate change of the wave fluxes rather than the simulated QBOs per se, plus the already established speeding up of the Brewer-Dobson circulation 14 and weakening of QBO amplitudes [25][26][27] , enables us to infer with some confidence that QBO disruptions are likely to become more common due a changing climate.

Discussion and Outlook
The quasi-biennial oscillation has been disrupted again for only the second time since its discovery.Both disruptions occurred near 40 hPa and were initiated by historically large forcing from extratropical waves.The 2019/20 event differs in that the dominant wave disturbances originated from the SH rather than the NH, no strong El Niño event was present, and an eastward jet subsequently emerged above the shallow westward layer.
When this predictability disintegrates, as occurred in the 2015/16 and 2019/20 disruptions, the accuracy -and hence value to society -of such forecasts may be reduced.Following both disruptions the normal QBO cycling resumed, manifesting in 2020 as an eastward jet emerging above the shallow westward layer, consistent with the standard QBO paradigm (see Fig. S4) and auguring a return to the high predictability of the QBO until meteorological conditions once again favour disruption.As of September 2020 the QBO has returned to a typical eastward pattern.
Whether disruptions themselves can be predicted more than ∼ 1 month in advance remains an open question 20 .The 2015/16 disruption was not predicted by operational seasonal forecasting systems 5 and early indications are that the same is true of the 2019/20 disruption, although models may perform better at predicting the evolution of the disruption once it has begun 20 .Predicting the full "life cycle" of QBO disruptions could provide a stringent test of models.Such work will be aided by the availability of new Aeolus satellite wind observations that will monitor the evolution of the QBO over the whole tropical belt 36 .Inherently shorter predictability of disruptions (as contrasted with the usual QBO) is consistent with their extratropical origins, since the extratropics are less predictable than the tropics.
Under climate change, Rossby wave propagation into the low-latitude stratosphere is expected to increase 37 and we have shown this occurs in model climate projections supporting the latest Intergovernmental Panel on Climate Change (IPCC) assessment.Under increasing influence from the extratropics, tropical stratospheric winds will likely become less predictable, leading to less skilful seasonal forecasts.Combined with an increas-manuscript submitted to Nature Geoscience ing Brewer-Dobson circulation 14 and weakening QBO amplitudes [25][26][27] , the prospect of QBO disruptions is likely to increase in a changed climate.

Methods
Our characterization of the QBO disruption is based on a tropical rawinsonde station (Singapore, 1 • N, 104 • E) and global gridded analysis fields (ERA5 reanalysis 38 ).
Daily and monthly averages of the zonal wind component were constructed from the twice daily meteorological Singapore soundings 39     they reach ∼ 70 hPa, any barrier to upward-propagating waves with eastward phase speeds is removed, and eastward wave forcing becomes available aloft.Following 2015/16 disruption this led to the resumption of eastward phase descent in April 2016, forced by the canonical QBO model forcing terms (Fig. 2b).The aftermath of the 2019/20 disruption unfolded similarly, but with a distinct twist: an eastward jet emerged in May 2020 from the remnants of the stalled westward phase near 25 hPa.This was forced by the vertical EP-flux component (Fig. S4), of which ∼50% is due to large-scale waves (k = 1-10; not shown), consistent with radiative damping of Kelvin waves as they encounter eastward wind shear.The rapid (sub-monthly) variations seen in Singapore observations in this region (Fig. 1a) are also consistent with Kelvin wave activity 43 .manuscript submitted to Nature Geoscience ), until westward winds emerged near 40 hPa in late December 2019.The 2019/20 QBO disruption thus resembles the 2015/16 event in that meridional momentum fluxes, neglected in the canonical QBO model, became anomalously strong and weakened the QBO eastward phase in the lowermost stratosphere, leading approximately three months later to the emergence of a shallow westward layer near 40 hPa.In both events this forcing occurred near the bottom of the eastward QBO phase, well below the descending westward phase above.The peak wind speed reached in the shallow westward layer was similar in both cases, being −21 m s −1 in 2019/20 and −19 m s −1 in 2015/16 (Fig. 2a,b).The two events differed in the timing of the strongest forcing by meridionally propagating waves: during Southern Hemisphere (SH) winter for the 2019/20 disruption, and during NH winter for the 2015/16 disruption.Forcing strengths also differed: peak forcing was stronger in 2015/16 (Fig. 2a and 2b, red lines) but concentrated over a shorter period from when the QBO eastward phase began its decay to when the 40 hPa westward layer emerged.At 50 hPa the time-integrated forcing from June to December 2019 was roughly 30% larger than that from October 2015 to February 2016 (−17 m s −1

Figure 1 .
Figure 1.(a) Twice daily radiosonde zonal wind observed at Singapore (1.4 • N, 104 • E, station id 48698).Lapse-rate tropopause determined from the radiosonde temperatures is shown as a black line.Missing radiosonde data are filled in with MERRA-2 interpolated to the location of Singapore.(b) The percent variance explained by principal components (PCs) 1 and 2 combined (black curve) and PCs 3 (red curve) and 4 (blue curve) as a function of time based on the monthly averaged Singapore zonal wind profiles (1976-2020) from 100-10 hPa.The EOF calculation was based on monthly averaged winds limited to 1976-2014 to avoid the two disruptions.The red vertical lines bracket September 2019.(c) Singapore QBO phase as a function of time in units scaled so that each 2π is one QBO cycle.The upper red line is fitted to the the phase from

Figure 2 .Figure 3 .
Figure 2. (a,b) Time series of ERA5 equatorial forcing tendencies due to meridional EP-flux convergence (red) and the sum of vertical EP-flux convergence and vertical advection (grey) superimposed on the altitude vs. time progression of zonal-mean zonal wind (5 m s −1 contours, eastward red, westward blue) for (a) the 2019/20 disruption and (b) the 2015/16 disruption.(c,d) Vertical profiles of meridional EP-flux convergence (red) and zonal-mean zonal wind (black) averaged over (c) July 2019 to September 2019 and (d) December 2015 to February 2016; the averaging periods are bracketed by vertical dashed green lines in (a,b).Grey shading shows the 5%-95% range (median dashed) of meridional EP-flux convergence over the 1979-2020 period for (c) July-September (JAS) and (d) December-February (DJF).All panels use ERA5 daily data, 4 • S-4 • N average, smoothed in (a,b) with a 5-day (31-day) running mean for wind (tendencies).(Forthe the full momentum budget, see Fig.S2.)

Figure 4 .
Figure 4. Solid lines: Multi-model mean vertical profiles of northward wave momentum flux averaged over 10-15 • N for NH winter and spring (December-May, left side) and 10-15 • S for SH winter and spring (June-November, right side) for ten models that provided wave flux diagnostics for the historical and SSP5-8.5 scenario simulations for phase six of the Coupled Model Intercomparison Project 28 .Dashed lines: Multi-model means for the subset of six models that also provided data for the SSP3-7.0scenario.Only one realization was used per model.Circles indicate data levels and are filled when the difference between the historical and SSP5-8.5 model ensembles are significant at 95% (large circles) and 90% (small circles), based on a Students ttest.Models that provided data were: CanESM5, CESM2, CESM2-WACCM, GFDL-CM4, GFDL-ESM4, HadGEM3-GC31-LL, INM-CM4-8, MIROC6, MRI-ESM2-0, UKESM1-0-LL, with names in bold indicating the model provided data for both scenarios.

Fig. S2 compares
Fig.S2compares the evolution of ERA5 zonal-mean zonal wind and its vertical shear in the 2015/16 and 2019/20 disruption events (Fig.S2a,b) and details of the zonal-mean zonal momentum budget at 41 hPa and 50 hPa (Fig.S2c-f).More precisely the pressure levels shown are 40.53 and 49.60 hPa, the closest ERA5 levels to 40 and 50 hPa.The resolved forcing terms are the meridional EP-flux component, vertical EP-flux component, meridional advection, and vertical advection.Forcings are smoothed with a 31day running mean as in Fig.2a,b.The canonical model forcing shown in Fig.2a,b(grey line) is the sum of the vertical EP-flux and vertical advection at 50 hPa as shown in Fig.S2e,f.The time axis in Fig.S2places the "central date" of each disruption at the same relative position on the axis for both disruptions.This is the time at which ERA5 zonal-mean zonal winds turned westward at 40 hPa, determined as 21 December 2019 and 1 February 2016 for the two events.For the 2019/20 disruption, ERA5 data are extended slightly in time using ECMWF operational analysis41,42  .

Figure S4 .
Figure S4.(a) Zonal-mean zonal wind (black contours; zero thick, westward dashed, 5 m s −1 spacing) and wind tendency due to vertical EP-flux component (filled contours).(b) Zonal-mean zonal wind tendency due to eddy momentum transports and advection, and zonal-mean zonal wind (thick black line), at the altitude of emerging eastward winds.Vertical green dashed line marks the time when westward winds first emerge near 40 hPa.Horizontal green dashed line in (a) indicates the altitude shown in (b).All panels use daily ERA5 daily, 4 • S-4 • N average, smoothed with (a) 11-day running mean, (b) 31-day running mean.ERA5 data are extended in time using ECMWF operational analysis (transition marked by grey vertical line in panels a, b).