Quasi-Biennial Oscillation and Sudden Stratospheric Warmings during the Last Glacial Maximum

The quasi-biennial oscillation (QBO) and sudden stratospheric warmings (SSWs) during the Last Glacial Maximum (LGM) are investigated in simulations using the Whole Atmosphere Community Climate Model version 6 (WACCM6). We find that the period of QBO, which is 27 months in the preindustrial and modern climate simulations, was 33 months in the LGM simulation using the proxy sea surface temperatures (SSTs) and 41 months using the model-based LGM SSTs. We show that the longer QBO period in the LGM is due to weaker wave forcing. The WACCM6 simulations of the LGM, preindustrial, and modern climates do not support previous modeling work that suggests that the QBO amplitude is smaller (larger) in a warmer (colder) climate. We find that SSWs in the LGM occurred later in the year, as compared to the preindustrial and modern climate, but that time of the final warming was similar. The difference in SSW frequency is inconclusive.


Introduction
The stratosphere during the Last Glacial Maximum (LGM) is expected to be considerably different than in the preindustrial and modern climate because of reduced CO2, CH4, and N2O concentrations; the presence of large ice sheets in the Northern Hemisphere (NH); and lower sea surface temperature (SST) [1][2][3]. Modeling studies show a slower Brewer-Dobson circulation (BDC) during the LGM than in the modern climate [4][5][6], which is consistent with model-projected BDC strengthening in response to greenhouse gas-induced warming [7][8][9][10][11][12][13][14]. The latter is supported by satellite observations [15,16]. Fu et al. [5] and Wang et al. [17] used the Whole Atmosphere Community Climate Model version 6 (WACCM6) [18] to examine the BDC and stratospheric O3 during the LGM. They showed that the stratospheric O3 changes in the LGM as compared to the preindustrial climate were caused by chemical processes in the upper stratosphere associated with a temperature increase and OH and NOx decreases, a slow-down of the BDC in the lower stratosphere, and a lowering of the tropopause in the extratropics [17]. In this study, we examine two aspects of stratospheric variability in the LGM-the quasi-biennial oscillation (QBO) and sudden stratospheric warmings (SSWs)-both with known links to the troposphere [19,20], by analyzing the same simulations described in Fu et al. [5] and Wang et al. [17].
The QBO is the primary mode of interannual variability in the tropical stratosphere, characterized by the downward propagating easterly and westerly wind regimes with an average period of ~28 months [21]. The model study of the QBO changes in a warming climate by Giorgetta and Doege [22] suggested a shorter QBO period at two times pre-industrial CO2 (2 × CO2) due to an increased gravity wave sources, while Kawatani et al. [23] and Watanabe and Kawatani [24] indicated a longer QBO period but weaker QBO amplitude in the warming climate simulations due to increased tropical upwelling. By analyzing global warming scenarios in the Coupled Model Intercomparison Project phase 5 (CMIP5) models that are capable of simulating QBO-like variability, Kawatani and Hamilton [25] found a long-term reduction in the QBO amplitude in the lower stratosphere but their results concerning QBO period changes were inconclusive. They also showed a QBO amplitude reduction at 70 hPa in observations from 1953 to 2012 but no significant trends in QBO period. Butchart et al. [26] examined 10 CMIP6 models with realistic QBO and found a weakening of the QBO in all 10 CMIP6 models and shortened QBO periods in 7 of the 10 models under warming scenarios. Richter et al. [19] examined the response of the QBO to a warming climate in eleven atmosphere general circulation models. No consistency was found among the models for the QBO period response, with the period decreasing by 8 months in some models and lengthening by up to 13 months in others in the 2 × CO2 simulations. In contrast, all the models projected a decrease in the QBO amplitude for 2 × CO2.
On the other hand, the SSW is a wintertime event in which the polar stratospheric temperature rises by several tens °C over the course of a few days and the stratospheric polar vortex reverses direction completely, with a subsequent recovery after the event [27]. SSWs occur about six times per decade in the NH [28]. SSWs can also occur in the Southern Hemisphere, but they are rare because of the small planetary wave amplitudes there [29], largely due to a lack of large-scale mountain ranges [30][31][32]. In this paper, we study only SSWs in the NH. The response of SSW frequency to increasing CO2 concentrations was investigated in several studies, but there is no consensus on changes in the number of SSWs [20,[33][34][35][36][37][38]. For example, Bell et al. [36] showed that at four times pre-industrial CO2 (4 × CO2), the frequency of SSW events per year is doubled with respect to the control simulation. By analyzing the Coupled Model Intercomparison Project Phase 6 (CMIP6) simulations, Ayarzaguena et al. [20] found that there is a large disagreement across the models as to the sign in changes of SSW frequency under 4 × CO2 forcing, with some models showing the rate to be significantly halved and others showing it to be significantly doubled.
Herein, we analyze the LGM simulations that used the prescribed SST from both model simulation and proxy data, and compare them with results of pre-industrial and modern climate simulations that used the observed SST and sea ice. We find that the period of QBO increases from 27 months in the preindustrial and modern climate to 33-41 months in LGM, while the SSWs in the LGM are found to occur later in the year. We analyze simulations of the LGM and preindustrial and modern climate constrained by observed SSTs to help broaden our perspective regarding the stratospheric response to anthropogenically induced climate changes in the future.

Model Experiments
The WACCM6 has a realistic, internally generated QBO in the stratosphere and simulates a realistic frequency of SSWs [18,39]. We analyzed five simulations from Fu et al. [5] and Wang et al. [17], who used the WACCM6 with a horizontal resolution of 0.9° latitude and 1.25° longitude and 70 vertical levels. They consist of (1) a LGM simulation with prescribed SSTs based on models (LGMPMIP3), (2) a LGM simulation with SSTs based on proxy data (LGMPROXY), (3) a preindustrial climate simulation (PI), (4) a modern climate simulation (MC), and (5) a modern climate simulation without ozone-depleting substances (ODS) (MCno_ODS). Each simulation was 60 years long, and only the last 50 years were analyzed. For the PI (MC and MCno_ODS) simulations, the SSTs and sea ice were the observed climatology of 1870-1890 (1980-2000) with a seasonal evolution [40]. For LGMPMIP3, the SSTs and sea ice data were obtained as the model differences between multi-model mean SST and sea ice for the LGM from PMIP3 [41] and the corresponding modern climate simulation plus the modern observed SSTs and sea ice. For LGMPROXY, the SSTs were from the MARGO proxy data [42]. The simulations from LGMPMIP3 and LGMPROXY provide information on the sensitivity of results to the SSTs used for the LGM. In the LGM simulations, the WACCM6 was coupled to the CLM4.0 LGM land boundary conditions [43] with the ice sheet topography from Abe-Ouchi et al. [44]. The greenhouse gas concentrations and orbital parameters corresponding to the LGM, preindustrial, and modern climate were used [5,17]. The ozone depletion substances (ODSs) in MC were taken from the year 2000. Figure 1 shows the deseasonalized monthly time series of the zonal-mean zonal wind averaged over 5° S-5° N at 30 hPa for MC and LGMPMIP3, which suggests a substantial increase in the QBO period during the LGM. Here, we define the QBO cycle with the transition time method that identifies zero wind transitions from easterly to westerly winds [19,45]. We use the zonal-mean zonal wind monthly anomaly time series averaged over 5° N/S, smoothed using a five-month centered running mean, e.g., as in Richter et al. [19]. The period and amplitude for each QBO cycle between individual transition points is defined, and the mean and standard deviation of the periods and amplitudes can then be derived. The QBO mean periods for MC, PI, LGMPROXY, and LGMPMIP3 are about 27, 27, 33, and 41 months, respectively, at both 10 and 30 hPa (Table 1), and are found to be independent of the altitude (Figure 2a). The mean periods of WACCM6-simulated QBO in the MC and PI agree well with observations [21]. The mean QBO periods for LGMPROXY and LGMPMIP3 are about 6 and 14 months longer, respectively, than in MC (Table 1 and Figure 2a). The differences in the QBO period between LGMPMIP3 and MC, LGMPROXY and MC, and LGMPMIP3 and LGMPROXY are all statistically significant at the 95% confidence level.

The QBO in the LGM
.  Table 1. Globally-averaged surface air temperature (SATG) (°C); globally-averaged sea surface temperature (SSTG); sea surface temperature averaged over 10° N/S (SST10S-10N); and quasi-biennial oscillation (QBO) mean period (P) (month) and mean amplitude (A) (m s -1 ) at 10 and 30 hPa from the MC, PI, LGMPROXY, and LGMPMIP3 simulations. Estimates of the QBO mean period and amplitude based on the transition time method are denoted by PTT and ATT, respectively, and the standard deviation follows the ± sign. The mean amplitude derived following Dunkerton and Delisi (1985) is denoted as ADD.    Figure 3 shows the mean QBO periods versus SST averaged over 10° S-10° N, indicating an increase in the QBO period with a decrease in SSTs over the deep tropics.    Figure 4 that the zonal wind tendency (i.e., the time derivative of the zonalmean zonal wind) (1st row) is generally smaller in LGM than in MC. The advection (2nd row) is due to the residual mean meridional circualtion and is dominated by the near-equatorial vertical advection (not shown). The advection tends to damp the zonal wind pertaurbation in the QBO, especially during the descent of the easterly shear zones, consistent with the observations [46]. The damping by the advection is weaker in LGM, as is the upward BDC [5], which opposes the downward propagation of successive easterly and westerly phases of the QBO. The adjusted tendency (3rd row) is the zonal wind tendency minus advection, which is the zonal wind tendency that remains to be explained by the wave forcing associated with the Eliassen-Palm (EP) flux divergence (i.e., the resolved wave forcing) (4th row) and the parameterized gravity wave drag (GWD) (5th row). The adjusted tendency is appreciably smaller in LGM, indicating a smaller wave forcing at that time. The EP flux divergence is considerably stronger in the westerly shear zones than that in the easterly shear zones. This is because the QBO eastward forcing is mainly due to planetary-scale Kelvin waves, while its westward forcing is dominated by gravity waves [47]. WACCM6 with its limited horizontal and vertical resolutions is not capable of resolving the gravity waves, which must be parameterized. It is evident from Figure 4 that the parameterized GWD dominates the wave forcing in both MC and LGM simulations, especially during the descent of the easterly shear zones. Both resolved wave forcing and parameterized GWD are smaller in LGM than their counterparts in MC ( Figure 4). The weaker resolved wave forcing in LGM could be attributed to weaker wave activity and/or weaker shear zones in LGM. The latter can be interpreted as a result of similar QBO amplitudes in LGM and MC ( Figure  2) but a longer QBO period in LGM. The weaker parameterized GWD in the LGM reflects the smaller latent heat release associated with weaker convection over colder tropical oceans.

It is evident from
It is suggested that the QBO period in the LGM from the WACCM6 simulations is longer than in the MC, because the wave forcings over both westerly and easterly shear zones are weaker ( Figure  4). The result is that it takes longer for the QBO to be forced from one phase to another in the LGM, although the weakening of the BDC partly cancels the wave forcing effects.
Previous studies have suggested that a QBO period increase results from a decrease of wave sources and/or an increase in tropical upwelling [19,[22][23][24]48]. Richter et al. [19] indicated that changes in the gravity wave momentum flux entering the stratosphere has a stronger influence on the QBO period than the tropical upwelling changes. In view of the weaker tropical upwelling in the LGM relative to MC [5], the weaker wave fluxes must be mainly responsible for the longer period ( Figure 4). In the models considered by Richter et al. [19], vertical advection and the resolved wave forcing did not play a large role in driving the QBO period changes in a warming climate: the change was mainly due to parameterized momentum flux due to gravity wave entering the stratosphere from below. Figure 4 suggests that the gravity wave forcing associated with cold climate was indeed smaller. That the QBO period was longer in the cold (LGM) climate simulated by the WACCM6 is consistent with the shorter period in the warmer climate simulated by WACCM5 [19]. Now, let us consider the change in QBO amplitude from LGM to MC. Following [19], we define QBO amplitude using the transition time method in which the amplitude for each cycle is estimated as the difference between the maximum and minimum zonal winds divided by two. The mean QBO amplitude is also estimated following Dunkerton and Delisi [49] as √2 , where is the standard deviation of the deseasonalized time series of the monthly mean winds with 5-month running mean. The mean QBO amplitudes from the two methods are similar (Table 1 and Figure 2b,c). The differences between the QBO amplitudes estimated by comparing PI, LGMPROXY and LGMPMIP3 with MC are 9.0% (0.6%), −5.0% (−6.9%), and 4.1% (−2.3%), at 30 (10) hPa based on the transition time method. From Figure 2b,c, the QBO amplitude for PI (red lines) is larger than MC (black line), consistent with Butchart et al. [26] and Richter et al. [19] who showed a robust and systematic decrease in the QBO amplitude with increasing SSTs and CO2 concentration. However, the QBO amplitude for LGMPROXY (green lines) is smallest among the five simulations, while the QBO amplitude for LGMPMIP3 (blue lines) is larger than MC (black lines) but generally smaller than PI. Thus, the WACCM6 simulations of the LGM and preindustrial and modern climate do not support the notion that the QBO amplitude is smaller (larger) in warmer (colder) climate. The latter is shown by Butchart et al. [26] and Richter et al. [19] for a projected warming world from models including the WACCM5 and 6. One key difference between this study and Butchart et al. [26] and Richter et al. [19] is that we focus on the past and modern climate with the simulations constrained by observed SSTs. It is interesting to note that the amplitude differences between LGMPMIP3 and LGPPROXY and between PI and MC at 30 hPa are both statistically significant.
The QBO period and amplitude from MCno_ODS are very similar to those from MC (not shown). We also examined the QBO periods and amplitudes from all five simulations using the actual zonal wind instead of the zonal wind after removing the seasonal cycle. The results (not shown) are similar. One advantage of using the zonal wind monthly anomaly is that the QBO period can still be derived between 70 hPa and 50 hPa, where the QBO amplitude is largely underestimated by the models [45].
The QBO is largely driven and maintained by momentum deposition from upward propagating Kelvin waves, mixed Rossby-gravity waves, and gravity waves, but damped by vertical advection. Large-scale Kelvin and mixed Rossby-gravity waves in GCMs are generated largely by parametrized convections, and gravity wave parametrizations are tuned to obtain the present-day QBO period and amplitude, which constitute a large source of uncertainty in modelling of the QBO [19]. Thus, the LGM QBO from the WACCM6 might be model-dependent. Hence, it would be useful to be able to compare the WACCM results with those of other models.

The SSWs in the LGM
We identify SSW events following Cao et al. [50] based on zonal-mean zonal wind as follows. The day on which the zonal-mean zonal wind at 60° N and 10 hPa turns from westerly winds to easterly winds, with easterly winds consequently lasting for more than five days, is defined here as the zeroth day of a sudden warming event. No new SSW events will be defined again during the 20 consecutive days following a prior SSW event. Stratospheric final warmings are excluded because the easterly winds persist until the next winter. Figure 5 shows the daily time series of the zonal-mean zonal wind at 60° N and 10 hPa from November to May for each year of the simulations (50 thin color lines) along with the 50-year average (black thick line) for (a) MCno_ODS and (b) LGMPMIP3. The results from MC and LGMPROXY are similar (not shown). We see the larger mean westerlies during the LGMPMIP3 than in MCno_ODS by comparing black thick solid and dashed lines in Figure 5b. This is consistent with the weaker wave drag but stronger zonal winds in the stratosphere over the NH mid and high latitudes in the winter (Figure 6b,c). Note that positive change (Figure 6c) in the negative wave drag regions is indicative of a smaller absolute wave drag. While lower greenhouse concentrations during the LGM favor a warming of the stratosphere, the reduced warming in the upper stratosphere and even the cooling in the lower stratosphere over the NH high latitudes almost certainly reflect a dynamical cooling signal (Figure 6a). This dynamical cooling in the cold climate is consistent with the previous modeling studies showing a boreal polar stratospheric dynamic warming in the warming world due to increasing CO2 [33,51]. Note that for the MC including ODS, the stratospheric temperature change is not only caused by the changes in greenhouse gas concentrations and wave activities but also by the anthropogenic O3 depletion. In this case, the dynamic signature in the temperature fields is not obvious because wave and O3 effects have an opposite sign (not shown). Since the ozone depletion impact on the SSWs is not the purpose of the paper, we only show the comparison between LGM and MCno_ODS in Figure 6. Figure 5 shows that the episodes of easterlies associated with SSW events tend to occur later in the year during the LGMPMIP3 than MCno_ODS. The first SSW event appears in November for both MCno_ODS and MC, December for PI, and January for both LGMPROXY and LGMPMIP3 (Table 2). However, a similar mean final warming day between day 115 to day 119 (i.e., April 25 to April 29) is seen in all the five simulations (Table 2).    (Table 2). In the MC, the ODSs were taken from the year 2000 when the ODSs were highest, which leads to the largest O3 depletion [52]. The average SSW frequency of MCno_ODS and MC, which is 0.57 per year, agrees well with observations [27,28]. The similarity of the SSW frequencies in PI, LGMPROXY and LGMPMIP3 (Table 2) is intriguing: is the lack of response in LGM because of countering influences from colder temperatures (fewer SSWs) but potentially more wave forcing from the ice sheet (more SSWs)? Table 2 also indicates that ozone depletion might reduce SSW frequency. Is this because the enhanced stratospheric cooling caused by the ozone depletion over the NH high latitudes leads to a stronger stratospheric zonal wind and thus it is harder to hit the 0 m s −1 threshold? In addition, is the difference in SSW frequency between PI and MCno_ODS (or MC) expected? Note that the significant test following Charlton et al. [53] shows that the SSW frequency difference between any two simulations is not statistically significant based on the 50 years simulations. We will examine these issues in our follow-up study by extending our simulations for a longer period of time and carrying out an additional LGM experiment by excluding the ice sheet.

Conclusions
The QBO and SSW during the LGM are examined in simulations using the WACCM6. It is found that the period of QBO, which is 27 months in the preindustrial and modern climates, was 33-41 months in LGM. The magnitude of the decrease in QBO period from then to now is sensitive to the SST over the deep tropics (10° S-10° N) prescribed in the LGM simulations. It is shown that both resolved wave forcing and parameterized GWD became smaller in LGM than those in MC from the WACCM6 simulations, leading to a longer QBO period despite a weaker advection in the LGM. The WACCM6 simulations of the LGM and preindustrial and modern climate do not support the notion that the QBO amplitude is smaller (larger) in warmer (colder) climate. It is shown that SSW events typically start to occur in January for the LGM but in November for the modern climate, but that the timings of the final warmings in the LGM and modern climate are similar. The similarity of the SSW frequencies in PI, LGMPROXY and LGMPMIP3 suggests potentially countering influences from colder temperatures and wave forcing associated with the ice sheet, which will be investigated in a followup study. Considering the past climate in our study helps to broaden our perspective regarding how the climate of the stratosphere may change in the future in response to human activities.