Impact of Increased Vertical Resolution in WACCM on the Climatology of Major Sudden Stratospheric Warmings

Abstract

Sudden stratospheric warmings (SSWs) are a major mode of variability of the winter stratosphere. In recent years, climate models have improved their ability to simulate SSWs. However, the representation of the frequency and temporal distribution of SSWs in models depends on many factors and remains challenging. The vertical resolution of a model might be one such factor. Therefore, here we analyse the impact of increased vertical resolution on the simulation of major sudden stratospheric warmings (SSWs) in the Whole Atmosphere Community Climate Model (WACCM). We compare two versions of the model, WACCM3.5 and WACCM4. We find that the frequency of occurrence of SSWs is improved in the newer version and closer to that obtained using reanalysis. Furthermore, simulations with a coupled ocean best reproduce the behaviour of temperature during these events. Increasing vertical resolution increases the number of occurrences; however, it does not produce significantly different results than standard resolution. WACCM4 also does not reproduce vortex split events well, generating far fewer of these than observed. Finally, the ratio between polar vortex splits and displacement events in the model is slightly better for non-ocean-coupled simulations. We conclude that, at least for WACCM4, the use of the high vertical resolution configuration is not cost-effective for the study of SSWs.

1. Introduction

Over the last several decades, a good understanding of the stratosphere has been seen as crucial for interpreting a multitude of climatic and meteorological processes [1,2,3,4]. These are related to ozone [5], coupling with the troposphere [6,7,8,9], climate change [10,11] and others, such as sudden stratospheric warmings (SSWs), with a large impact on the seasonal and sub-seasonal meteorological predictability.

The stratospheric polar vortex is the strong westerly wind system that dominates the stratospheric circulation during winter. This circulation can be disturbed by upward propagating waves (primarily planetary scale zonal wave-number 1–2 quasi-stationary waves) from the troposphere that dissipate in the stratosphere [8,12,13]. Sufficient wave forcing of the mean flow by these waves, that is, the transfer of westward momentum to the back-ground flow, can result in an SSW, with the breakdown of the polar vortex and replacement of westerly winds by easterlies [13], which is accompanied by rapid warming of the polar stratosphere (30–40 K in a few days); the effects extend to the Earth’s surface, as well as through the mesosphere and beyond [8]. This warming can be observed from the polar to the subtropical stratosphere and is accompanied by cooling in the tropics [8,13,14]. During SSWs, stratospheric winds in the polar vortex change direction, accompanied by a vortex displacement or split. We denote these two types of SSW by ($\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}$) and ($\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$), respectively. $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}$ and $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$ are generally associated with the vertical propagation of Rossby waves with wavenumber 1 (WN1) and 2 (WN2), respectively [12,15,16,17,18,19,20]. Significant differences have been found in the strength of surface and ocean responses for $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$ and $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}$. For example, a composite of split SSWs displays strong anomalies in the implied Ekman heat flux and net atmosphere-surface flux, modifying the mixed layer heat budget [7].

In the northern hemisphere (NH), intra- and inter-annual variability is mainly dominated by occurrences of SSWs, with a frequency of between 0.60–0.68 events per year, (following the criterion of the reversal of zonal wind at 60${}^{\circ }$ N and 10 hPa) [17,21,22,23]. This frequency varies significantly depending on the definition used for detection, however [22,24,25,26].

The general circulation climate models (GCMs) are an essential tool for understanding the causes and behaviour of the different stratospheric phenomena. However, the reproduction of SSWs in GCMs faces biases that are dynamical in nature. Some of them are related to dynamical cores. For example, as shown by other studies [27,28], dynamical cores can underestimate the occurrences of SSWs. It is known that this shortcoming affects CAM-FV, a broadly used dynamical core, which is the dynamical core of the model that we use in this work. However, advances in the simulation of SSWs using the Whole Atmosphere Community Climate Model (WACCM) [29] have greatly improved since earlier works. Older versions, such as WACCM1b, featured an average annual frequency of SSWs of 0.10 for the years 1958–2002, which is well below the value of 0.60 obtained in the reanalysis [30]. Later, WACCM3 included all physical parameterisations of the Community Atmosphere Model (CAM) version 3 [31]. In this case, an increase in the annual frequency was observed (0.30) in the version of the model with a horizontal grid of 4${}^{\circ }$× 5${}^{\circ }$ for the years 1958–2000, but this was still poor compared to reanalysis for the same period (0.60) [32].

It is essential to understand extent to which the increase in the number of vertical levels in a model can help reproduce the atmosphere’s dynamical features better. Given the impact of gravity waves (GWs) on SSWs [18,33,34,35], it could be hypothesized that increasing the number of levels could improve the vertical propagation of waves, hence changing the simulation of SSWs. Indeed, increasing the grid resolution (horizontal or vertical) of a model is one of the critical aspects of dealing with a process as resolved or parameterized. Since the beginning of its development, increasing the number of levels in WACCM has proved to be a successful technique to improve the representation of physical processes. An example is how increasing the vertical resolution for the same model version improved the representation of the upper troposphere–lowermost stratosphere [36,37]. Therefore, understanding the potential benefits and drawbacks of increasing vertical resolution in a model is of the utmost relevance.

WACCM3.5 is based on the Community Atmosphere Model, version 3.5 (CAM3.5) with the vertical model domain extended to 145 km. The horizontal resolution for WACCM3.5 runs presented here is 2.58${}^{\circ }$ × 1.98${}^{\circ }$ (longitude × latitude), resulting in a further increase in the predicted occurrence of SSWs (0.60). This is caused by the addition of surface stress due to unresolved orography, which was modelled as turbulent stress in mountainous regions [38]. Further in-depth analysis of the climatology and characteristics of SSWs was achieved with this version [17]. Such study confirmed a closer agreement between the annual frequencies of SSWs from the reanalysis data and those obtained by the model (0.57) for the period 1954–2005. It was shown that the model could reproduce the correct form of the two types of SSWs. In other previous studies, WACCM was used to analyse the roles of planetary and gravity waves during a major SSW [18,38].

WACCM4 is based on the Community Atmosphere Model, version 4 (CAM4 [39]). WACCM4 uses the finite-volume dynamical core [40] with 66 vertical levels, with variable vertical resolution. The horizontal resolution is the same as that for WACCM3.5, 2.58${}^{\circ }$ × 1.98${}^{\circ }$ (longitude × latitude); vertical resolution varies with altitude, from 1.1 to 1.4 km in the troposphere (above the boundary layer) and lower stratosphere, to 1.75 km in the upper stratosphere and 3.5 km in the upper mesosphere and lower thermosphere. The upper boundary is located at a geometric altitude of about 140 km. WACCM4 presents several improvements compared to previous versions, including changes to the parameterization of the effects of the orography and gravity waves, and also in terms of the forcing of the quasi-biennial oscillation (QBO) [41]. These improvements allow a better representation of SSWs. The model now also works with the latest version of the parameterization of gravity waves [42]. Furthermore, in using WACCM4, it has been shown that the Brewer–Dobson circulation accelerates 15 days before the wind reversal occurs, before undergoing a deceleration that lasts up to 60 days after the central date of the SSW. This deceleration is explained by the cessation of wave forcing after this date [43].

Several studies have documented the relevance of the increase in the vertical resolution in GCMs to correctly reproduce different atmospheric factors related to water vapour dynamics [44,45,46], gravity waves [47,48,49] or stratified turbulence based on shear instability [48,50]. This higher resolution has also been shown to be critical in reproducing the QBO disruption of 2016 [51,52]. WACCM has already been used at high resolution in the mesosphere to analyse measurements of sodium obtained by lidar (with 88 levels and $\overline{dz}$ = 3.5 km in mesospheric altitudes) [53]. It has also been used to assess the temperature variability in the tropopause and the tropopause inversion layer (with 103 vertical levels and about 300 m vertical resolution in the upper troposphere-lowermost stratosphere (UTLS)) [54], as well as the climatology of the tropical tropopause layer [55]. In the SPARC Chemistry-Climate Model Validation Activity 2 (CCMVal2) report [36], a version of WACCM with 300 m vertical resolution in the UTLS was shown to improve the vertical profile of temperature for the UTLS significantly [36,56]. Previous studies showed that greater vertical resolution could lead to a decrease in the temperature in the winter pole, which is accompanied by a strengthening of the vortex and a reduction in the propagation of waves in the vertical [57].

In this study, we analyse SSWs in the simulations with the same WACCM version (WACCM4) used for the Chemistry-Climate Model Initiative (CCMI) [58,59], with simulations using standard (66 levels) and high vertical resolution configurations (132 levels). Our 132-level distribution has the highest number of vertical levels from the surface to the top among all the models used to study SSWs to date (See

Table A1. Comparison between mean of the vertical resolution ($\overline{dz}$) of the Models used for the study of the SSW during the period 2007–2020 and the Model WACCM with high vertical resolution (WACCM_hvr) for the same model top.

			$\overline{\mathit{d}\mathit{z}}$ (km)
	Models	References	Model	WACCM_hvr	Model Top
1	ACCESS1.0	[25,75]	1.03	0.49	39 km
2	ACCESS1.3	[25,75]	1.03	0.49	39 km
3	ACCESS CCM	[26]	1.40	0.75	84 km
4	AMTRAC	[76]	1.58	0.74	0.02 hPa
5	CanESM2	[25,75,77]	1.98	0.56	0.5 hPa
6	CanESM5	[78]	0.99	0.53	1 hPa
7	CCSM4	[25,75,79]	1.59	0.51	2.194 hPa
8	CCSRNIES-MIROC 3.2	[26]	2.33	0.73	0.012 hPa
9	CESM2	[78]	1.25	0.50	40 km
10	CESM2-WACCM	[78]	2.14	1.14	150 km
11	CMAM	[26,80]	1.41	0.85	0.0006 hPa
12	CMCC-CESM	[25,75,77,81]	1.82	0.67	0.04 hPa
13	CMCC-CM	[25,75]	1.04	0.49	10 hPa
14	CMCC-CMS	[25,75,77]	0.85	0.73	0.01 hPa
15	CNRM-CCM	[26]	1.40	0.75	84 km
16	CNRM-CM5	[25,75,79]	1.04	0.67	10 hPa
17	CNRM-ESM2-1	[78]	0.89	0.73	0.01 hPa
18	EMAC-L47	[26]	1.71	0.73	0.01 hPa
19	EMAC-L90	[26]	0.89	0.73	0.01 hPa
20	FVGCM	[30]	1.46	0.73	0.01 hPa
21	GEOS-CCM	[26]	1.12	0.73	0.01 hPa
22	GFDL-CM3	[25,75,77,79,81]	1.68	0.73	0.01 hPa
23	GFDL-CM4	[78]	1.47	0.53	1 hPa
24	GISS-E2-R	[79]	1.61	0.63	0.1 hPa
25	GISS-E2-H	[79]	1.61	0.63	0.1 hPa
26	GFDL-CM4	[78]	0.90	0.80	0.002 hPa
27	GISSL53	[30]	1.73	0.80	0.002 hPa
28	HadCM3	[75,79]	1.70	0.49	10 hPa
29	HadGEM2-CC	[25,75,77,79,81]	1.40	0.75	84 km
30	HadGEM2-ES	[79]	1.05	0.50	40 km
31	HadGEM3-ES	[26,82]	1.00	0.76	85 km
32	HadGEM3-GC31-LL	[78]	1.00	0.76	85 km
33	INM-CM5-0	[78]	0.82	0.60	0.20 hPa
34	IPSL-CM5A-LR	[25,75,77,79,81,83]	1.82	0.67	0.04 hPa
35	IPSL-CM5A-MR	[25,75,77,79]	1.82	0.67	0.04 hPa
36	IPSL-CM5B-LR	[25,75,77]	1.82	0.67	0.04 hPa
37	IPSL-CM6A-LR	[78]	1.01	0.64	80 km
38	IPSL-LMDZ-REPROBUS	[26]	1.79	0.67	70 km
39	MAECHAM	[30]	2.07	0.73	0.01 hPa
40	MIROC3.2	[84]	2.37	0.73	0.01 hPa
41	MIROC5	[25,75,79]	1.02	0.49	3 hPa
42	MIROC6	[78]	1.08	0.77	0.004 hPa
43	MIROC-ESM	[25,75,79]	1.10	0.78	0.0036 hPa
44	MIROC-ESM-CHEM	[25,75,77,79,81]	1.10	0.78	0.0036 hPa
45	MPI-ESM-LR	[25,75,77,79]	1.71	0.73	0.01 hPa
46	MPI-ESM-MR	[25,75,77,81]	0.85	0.73	0.01 hPa
47	MRI-CGCM3	[25,75,77,79,81]	1.68	0.73	0.01 hPa
48	MRI-ESM1	[75,77]	1.68	0.73	0.01 hPa
49	MRI-ESM1r1	[26]	1.01	0.73	0.01 hPa
50	MRI-ESM2-0	[78]	0.81	0.73	0.01 hPa
51	MRIJMA	[30]	1.79	0.73	0.01 hPa
52	NIWA-UKCA	[26]	1.12	0.55	0.07 hPa
53	SOCOL3	[26]	2.01	0.73	0.01 hPa
54	TIME-GCM	[85]	0.72	0.49	10 hP
55	UKESM1-0-LL	[78]	0.76	0.73	0.01 hPa
56	WACCM	[17,18,30,38,43,57,86,87,88]	2.20	1.10	145 km

). The goal of the present study is to compare the results of these simulations, to identify potential improvements, and to understand the differences by validating the output using reanalysis data. We also assessed the ability of the model to reproduce SSWs in terms of duration and frequency of occurrence.

2. Dataset and Methodology

Here we use WACCM4, a general circulation model including chemistry, radiation and dynamics that is based on the Community Atmosphere Model (CAM4) [60]. This, in turn, is one of the atmospheric components of the Community Earth System Model (CESM1.1) [61]. The vertical domain of the model spans from the surface of the Earth up to ≈140 km, with a horizontal resolution of $1.9^{\circ }\times 2.5^{\circ }$ (latitude×longitude).

We used two reference simulations created for the CCMI (REFC1 and REFC2) [62]. For each simulation, we produced an ensemble of three free-run members. These are free-run simulations with prescribed initial conditions (see [62]) and high-vertical resolution (hvr) of “132 levels”, and another set of three free-run members was provided by the USA National Center for Atmospheric Research (NCAR) with a standard resolution of “66 levels”. REFC1 is a free-running atmospheric simulation with boundary conditions that can be specified by observations, covering the period 1955 to 2014, while REFC2 works with an interactive coupled ocean and covers the period from 1955 to 2099. It uses the RCP6.0 scenario of stabilization in which it is considered that in 2100 the total radiative forcing will be 6.0 ${\mathrm{W}/\mathrm{m}}^2$ [63]. The vertical resolution does not imply a constant difference between levels. Instead, the resolution decreases with height. The troposphere and most of the stratosphere features a dz = 0.5 km. It then increases slowly after reaching 40 km altitude, until reaching dz = 2 km at around 100 km altitude, maintaining that resolution up to the highest altitude of the model (see Figure 1). It is hypothetized that the increase in the vertical resolution in the troposphere and the stratosphere of the model will support an adequate reproduction of the stratospheric dynamics that favour the generation of SSWs.

In order to validate the results, we used daily averages of wind, temperature and geopotential height obtained from the JRA55 reanalysis [64,65]. We also performed our analyses with MERRA [66] and ERA-Interim [67] reanalyses (not shown) and found similar results. JRA55 has a vertical resolution of 60 levels from the surface to 0.1 hPa with a horizontal resolution T319 (60 × 60 km).

The detection of SSWs was achieved using criteria described by Charlton and Polvani [21]: a major SSW occurs when the zonal-mean zonal wind at 60${}^{\circ }$ N and 10 hPa becomes easterly during the Northern Hemisphere winter. The date of this reversal is taken as the central date of the event, which ends when westerly winds are re-established. To ensure that detected events are distinct, there must be a difference of at least 20 days between successive events. Furthermore, to rule out final warmings, we require that the winds remain westerly for at least ten days before 30 April. The duration of each event is computed as the number of consecutive days of easterly zonal-mean zonal winds at ${60}^{\circ }$ N and 10 hPa after the central date.

To determine whether the event is a displacement $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}$ or a split $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$, we plot the geopotential height at 10 hPa in the polar vortex and subjectively check the type of event corresponding to each central date detected.

We compared the model results with reanalysis in terms of the average annual frequency, the relationship $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}$/$\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$, the duration of the SSWs, and the temperature in the polar vortex, using the same statistical tests as described elsewhere [17,26,30]. All analyses were made using the historical period in common between the model and the reanalyses (1958 to 2014).

Subsequently, the monthly zonal wind climatology, the monthly climatological Eliassen-Palm flux divergence and the mesoscale gravity waves (parameterized) in the model are analyzed, seeking to explain the variations between the different simulations and their respective resolutions.

3. Basic SSW Characteristics

Table 1. SSWs climatology in WACCM4. Total number of major SSW events; Frequency of SSWs per winter (standard errors); $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}/\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$ ratio for winter. The * indicates no significant differences at p $<0.05$ with respect to JRA55 and † indicates no significant differences at p $<0.05$ compared to the high vertical resolution version.

Dataset	Total SSWs	SSWs/Winter	$\mathbf{S}\mathbf{S}{\mathbf{W}}_{\mathbf{D}}$/$\mathbf{S}\mathbf{S}{\mathbf{W}}_{\mathbf{S}}$
REFC1.1_hvr	48	0.84 (0.07)	2.69
REFC1.2_hvr	43	0.75 * (0.07)	1.87
REFC1.3_hvr	40	0.70 * (0.08)	1.85
REFC1_hvr	131	0.77 * (0.04)	2.12
REFC1.1	40	0.70 * (0.07)	2.33
REFC1.2	34	0.60 * (0.07)	2.78
REFC1.3	41	0.72 * (0.08)	1.93
REFC1	115	0.67 †* (0.04)	2.29 †
REFC2.1_hvr	40	0.70 * (0.09)	7.00
REFC2.2_hvr	42	0.74 * (0.09)	3.67
REFC2.3_hvr	36	0.63 * (0.06)	3.50
REFC2_hvr	118	0.69 * (0.04)	4.36
REFC2.1	44	0.77 * (0.09)	3.40
REFC2.2	38	0.67 * (0.07)	3.75
REFC2.3	33	0.58 * (0.06)	3.71
REFC2	115	0.67 †* (0.04)	3.60 †
JRA55	37	0.65	0.68

presents the results referring to the detection and classification of SSWs. The mean frequencies of occurrence of SSWs in the model (0.67–0.77) did not turn out to be significantly different from the reanalysis (0.65) at the 0.05 confidence level in all cases. Although the average for the model and the reanalysis is very similar, it is notable from the ratio $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}$/$\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$ that WACCM4 has a much larger ratio of displacement to split SSWs than WACCM3.5 ((2.12–4.36) vs. (1.13–2.18) [17]). For the simulations, the ratio $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}/\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$ is significantly different from the reanalysis at the 0.10 level in most of the cases. The ratio $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}/\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$ is significantly larger in the REFC2 (coupled ocean) than in REFC1 (specified sea surface temperature (SST)) at the two resolutions used. This points to the importance of the SST to reproduce SSWs in the model [68,69].

The duration was computed as the number of consecutive days of easterly zonal mean zonal winds at 60${}^{\circ }$ N and 10 hPa after the central date. A box plot showing the mean durations of each type of SSW for all the simulations is shown in Figure 2. By typical values, we refer to those that lie within the range of the error bars. It shows that the duration of the $\mathrm{S}\mathrm{S}\mathrm{W}\mathrm{s}$ is not significantly different from the reanalysis, presenting very similar dispersion values between them with variations from 1 to 41 days. The vertical resolution of the simulations has no apparent impact on the duration of the events, and the differences between REFC1 and REFC2 are similarly insignificant.

4. Analysis of the Monthly Distribution of SSWs

Figure 3 shows the monthly frequency of occurrence of the SSWs for each of the members of the two ensembles compared with the reanalysis JRA55 (red lines).

The results obtained for the reanalysis in Figure 3 agree with those in previous studies [17,21,23,24]. They show that the model generates a higher occurrence of SSWs compared to the reanalysis during November, December and March, while the opposite appears to be the case in January and February. From December to March, the frequency is between 0.13–0.22 events per year. November features the minimum values, close to 0.02. Thus, in REFC1, it seems that the high vertical resolution favours the occurrence of SSWs, except for March. While in REFC2 simulations, the impact is not homogeneous.

5. Polar Cap Temperature

Figure 4 shows a composite of the changes in SSWs in averages of temperature anomalies (calculated as the deviation of the daily climatology having as reference the central day) in the polar cap (60${}^{\circ }$ N to 90${}^{\circ }$ N) in the middle and lower stratosphere (10 hPa (${\overline{T^{\prime }}}_{10}$) and 100 hPa (${\overline{T^{\prime }}}_{100}$)). It spans from 50 days before, to 90 days after the central date of the events of interest. ${\overline{T^{\prime }}}_{10}$ serves as an indicator of the amplitude of the SSWs in each of the cases. While ${\overline{T^{\prime }}}_{100}$ indicates the strength of the downward extent of temperature anomalies in the stratosphere in each case.

During SSWs, ${\overline{T^{\prime }}}_{10}$ is cooler after the event, especially in JRA-55, although the values are within the interquartile range. Before this, ${\overline{T^{\prime }}}_{10}$ increases, reaching a maximum with an average of approximately 5–10 K shortly after the central date. Next, it decreases to negative values of as much as −5 K. The behaviour of the average temperature during SSWs is close between the reanalysis and the model, with a maximum difference of 2–3 K for peak temperature near the central date. From 30–22 days before and 20–90 days after the central date, ${\overline{T^{\prime }}}_{10}$ in the reanalysis is lower than the simulations. ${\overline{T^{\prime }}}_{100}$ is very similar in all cases, with differences between the model and reanalysis not greater than 0.5 K, and with all the models and reanalyses lying in the same inter-quartile range. ${\overline{T^{\prime }}}_{100}$ begins to increase up to 20 days before the central date, reaching ≈3 K. Next, it decreases, and stabilizes approximately 60 days after the central date.

6. Differences in Monthly Variability of Winds

Figure 5 shows the differences in the zonal average of the u component of the wind between reanalysis and the model.

The polar vortex produced by the model is relative weaker than the reanalysis in the NH in the stratosphere for November to January. In November and December, the differences are of up to −25 m/s and −15 m/s, respectively, and in December, up to −15 m/s. In January, the winds are weaker in the model than in the reanalysis in the NH stratosphere, with differences up to −10 m/s. February sees an observed strengthening of the polar vortex compared to reanalysis in the NH, and with differences up to 10 m/s and 5 m/s, respectively. Finally, in March, it strengthens compared to the reanalysis, with differences of up to 5 m/s. Increasing the vertical resolution in the model causes some minor improvements in some months, mostly for the REFC1 cases over the polar cap. Taking Figure 3 and Figure 5 from November to January together, the greater differences in winds are negatively poleward of ${50}^{\circ }$ N. These differences correspond to a greater number of occurrences of SSWs in all models with respect to the reanalysis. In these same latitudes during February, the greater differences in winds are positive, corresponding to a lower number of occurrences of the SSWs in the model compared to the reanalysis, except for REFC1_hvr where the positive differences are lower. For March, the same reasoning does not work, and the result is not conclusive.

7. Eliassen-Palm Flux and Mesoscale Gravity Wave Drag (GWD)

Given the non-significant differences in the frequency of SSWs observed between the standard and high-vertical resolution versions of the model in the previous analyses, and considering the well-known importance of the wave–mean flow interaction for the SSW dynamics [12], we hypothesize that the dynamical forcing in both model versions has to be similar.

To verify this, Figure 6 shows a monthly climatological Eliassen–Palm flux divergence (EPFD) [70] as a proxy for the resolved wave forcing for the REFC2_hvr simulation (upper plots) and the difference between the simulations in the NH Winter. Supporting our hypothesis, we found largely insignificant differences in EPFD between the simulations in the middle to upper stratosphere, with only locally significant differences in the UTLS and mesosphere. The significant differences between the high and standard vertical resolution (last two rows of Figure 6) show a very similar spatial distribution for both versions of the model (interactive and specified ocean).

In the UTLS, on the poleward side of the subtropical jet (around 40–50${}^{\circ }$ N), the dissipation of resolved waves is weaker in REFC1_hvr and REFC2_hvr than in REFC1 and REFC2 during the whole period of analysis. Equatorward from 30${}^{\circ }$ N and poleward from 60${}^{\circ }$ N, the situation is reversed. In the REFC2_hvr EPFD climatology, we can even see small areas of EP flux divergence in this region. These patterns point to the differences in GWD between the simulations, as the GWD can alter the propagation of resolved waves in this region [71].

Figure 7 shows that this is indeed the case. In the high-vertical resolution versions of the model, the GWD is significantly stronger at the upper flank of the UTLS jet and weaker further above. At this point we can only speculate why this is the case. However, given the efficient interaction (but, in the wave mode, very sensitive to the actual meteorological conditions) between GWD and leading planetary wave modes [72] that dominate the resolved wave field in the stratosphere, the different role played by GWD could be the reason for larger differences in $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}/\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$ between the REFC1_hvr and REFC2_hvr simulations than between REFC1 and REFC2 (see Table 1).

To provide additional insight on the role of GWD on the propagation of planetary waves, we performed an analysis separating the contribution of WN1 and WN2, using a Fast Fourier Transform on the EPFD in high-vertical resolution REFC1 and REFC2 simulations (see Figure 8). From the plots, it is clear that WN1 dominates over WN2, which is according to the results that we previously obtained, that show a prevalence of $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}$ (see Table 1). Furthermore, the plot for WN1 shows wave convergence in the region where they can contribute to $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}$. For the case of the differences between coupled and non-coupled ocean simulations, they are only significant in small areas and with values that depend on the month.

8. Conclusions

We have assessed the climatology and the variability of SSWs in the NH as obtained using WACCM4. The simulations were first obtained using a standard vertical resolution, and then using a higher resolution, to determine the impact of the increased resolution on the model results. Comparing all the results, we can conclude that the model is effective in reproducing the frequency of occurrence of the SSWs. Compared to the previous version WACCM3.5, the difference in the number of SSWs/year between WACCM (standard vertical resolution) and reanalysis has improved. In the previous version, in the most favourable comparison, the absolute difference in SSW frequency between model and reanalysis was 0.03, while for WACCM4 it was 0.02, although these differences could not be significant. However, while it again may not be significant, the case of the ratio in high vertical resolution is now worse ($\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}$/$\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$ (0.57)) [17]. We speculate that the cause for this result is a change in the parameterization of mesoscale GWs in WACCM4 [42] that can impact the propagation of zonal WN1 and WN2 from the troposphere and in the stratosphere [72]. The results show an increase in SSW frequency when the vertical resolution of the model is increased, although these differences are not significant. In analyzing the frequency of the monthly occurrence of the SSWs, we can observe that the differences in the zonal wind climatology reflect differences in the frequency of SSW. However, the correspondence breaks down in March.

Although WACCM4 can reproduce $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$ and $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}$, the ratio $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}/\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$ is larger (≥1.85) compared to the reanalyses (0.68). This is due to the poor ability of the model to generate $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$. It is relevant to notice that this relation does not show any systematic change in the high vertical resolution configuration, but shows better values when the model is run with a coupled ocean (REFC2) than without it (REFC1). The relationship is significantly different from the reanalysis, at least at the 0.10 level, in all the ocean-coupled scenarios. Although the total number of SSWs is about right (0.65 in reanalysis vs. 0.67–0.77 in WACCM), the balance between $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}$ and $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$ is not as observed. The model produces only half of the $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$ that it should have compared to the reanalysis. Therefore, this implies that the number of $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}$ can be up to double that of the reanalysis. Some recent works [73,74] have studied the impact of lower tropospheric wave events (LTWEs) on the morphology of SSWs, finding that such events are less connected to $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$ than $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}$. Indeed, de la Cámara et al. [74] studied the same WACCM version that we use here. However, their results also point out that only a tiny percentage of deceleration events are connected with LTWEs. This is an open question, and hopefully, in the future, models with improved resolution or wave representation will provide better insight into it.

The different scenarios demonstrate adequate simulation of the amplitude and vertical distribution of the temperature anomalies during SSWs (see Figure 4), and also for both types of events (not shown), with the exception of the period prior to the central date, although the values are in the interquartile range. During this period it is observed that the anomalies of temperature are greater, reaching values of up to 5 K. Finally, the results reveal no significant differences between the use of a standard vertical resolution and those produced by a high vertical resolution in the model for the analysis of SSWs.

The results show that, although there are significant differences between resolved and GWD forcing in both resolutions in the UTLS and upper stratosphere-lower mesosphere, these do not generally explain the monthly differences in the frequency of SSWs. Instead, these seem to be dominated mainly by the climatology of the zonal wind. Finally, we speculate that the different GWD distribution between the standard vertical and high vertical resolution can influence the difference in the $\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{D}}/\mathrm{S}\mathrm{S}{\mathrm{W}}_{\mathrm{S}}$ ratio between simulations with different ocean configurations. In view of the additional computational cost, simulations using a higher vertical resolution cannot be recommended for the analysis of SSWs. However, we must consider that the turbulent mountain stress parameterization was not updated between WACCM3.5 and WACCM4, or adapted to a high-vertical resolution configuration. This should be the goal of further research. Nevertheless, these results are relevant for the understanding of dynamical processes in the stratosphere, vertical transport analysis, and their representation in GCMs.