## 1. Introduction

Gravity waves (GW) play an important role in atmospheric dynamics. They are excited mostly in the troposphere—for example, by flow over orography, convection, and jets and front systems. In the course of their propagation, they affect the momentum and energy balance in the atmosphere everywhere up to the thermosphere (see, e.g., Kim et al. 2003). The direct impact of GWs on the large-scale circulation is largest in the middle atmosphere; however, they also affect tropospheric weather and climate significantly (e.g., Scaife et al. 2005, 2012).

In general circulation models (GCMs) and numerical weather prediction (NWP) models, effects of GWs must be parameterized, given the wide spatial and temporal spectrum they act on, part of it being far below the effective resolution of global model applications. Wentzel–Kramer–Brillouin (WKB) theory (Bretherton 1966; Grimshaw 1975; Achatz et al. 2017) is the basis of most GW parameterizations (GWP) in climate simulations and weather predictions (Lindzen 1981; Medvedev and Klaassen 1995; Warner and McIntyre 1996; Hines 1997a,b; Lott and Miller 1997; Alexander and Dunkerton 1999; Scinocca 2003; Orr et al. 2010; Lott and Guez 2013). There is, however, an increasing appreciation that the present handling of this technique needs improvements: a simplification typically used is the neglect of 1) horizontal GW propagation (single-column approximation) and 2) transient effects such as nondissipative GW–mean-flow interactions (steady-state approximation). The former has been shown to be an important weakness of state-of-the-art parameterizations, by, for example, Sato et al. (2009), Ribstein et al. (2015), Ribstein and Achatz (2016), and Ehard et al. (2017); Bölöni et al. (2016), Muraschko et al. (2015), and Wilhelm et al. (2018) propose improvements with regard to the latter aspect. Another drawback of GWPs in current climate and weather codes is that their applicability outside of the tropics (where Coriolis effects are nonnegligible) relies on the assumption of balanced (hydrostatic, geostrophic) resolved flows, which might not be valid with the increasing spatial resolutions applied nowadays. If, however, the resolved flow is not balanced, additional forcing terms due to the GW dynamics appear both in the momentum and the entropy equation representing, for example, elastic effects (Achatz et al. 2017; Wei et al. 2019). Potential triad wave–wave interactions in the atmosphere are also not taken into account in current GWPs, although their neglect has never been justified explicitly, to the best of our knowledge. In addition to the propagation issues listed above, faithful representation of GW sources is a key to success, and is another area where one finds room for improvement: theory and applications for orographic (Palmer et al. 1986; Bacmeister et al. 1994; Lott and Miller 1997) and convective GW sources (Beres et al. 2005; Song and Chun 2005) are relatively well developed, but the representation of GW emissions by jets and fronts—despite the efforts of Charron and Manzini (2002), Richter et al. (2010), and de la Cámara and Lott (2015)—remains difficult.

This paper focuses on the issues of GW propagation. In a novel framework, transient effects are incorporated by removing the steady-state approximation. This work is an extension of the study by Bölöni et al. (2016), in which effects of the transient, nondissipative GW–mean-flow interactions have been assessed in an idealized set-up, whereas here the same is done in a more complex framework in which the proposed transient GWP has been implemented into a state-of-the-art GCM/NWP model. The single-column approximation has been kept for the sake of simplicity, with the intention to give it up in a later step of our developments.

Section 2 motivates the implementation of a transient GWP to a state-of-the-art GCM and recalls the necessary theoretical background for the rest of the paper. This is followed by the actual implementation details in section 3 and by the presentation of the GCM-simulation results in section 4. A summary of the most important findings is given in section 5.

## 2. Theory

In the following we outline the theoretical basis of the proposed transient GWP called the Multiscale Gravity Wave Model (MS-GWaM). In section 2a we do so for locally monochromatic GWs together with the simplifying assumptions applied and a comparison to standard parameterization approaches. In section 2b the monochromatic perspective is generalized to full GW spectra.

### a. Locally monochromatic GW fields

In this section we first sketch the general WKB theory on which MS-GWaM is built [section 2a(1)] and then describe the simplifying pseudomomentum-flux approach and single-column approximation that are used in the current study [sections 2a(2) and 2a(3)]. Then, our transient formulation is compared with the one with the steady-state approximation on which present-day GW parameterizations are based [section 2a(4)].

#### 1) General WKB

**k**(

**x**,

*t*) =

**e**

_{x}

*k*+

**e**

_{y}

*l*+

**e**

_{z}

*m*and local frequency

*ω*(

**x**,

*t*), while its amplitude can be deduced from its wave-action density

**x**=

**e**

_{x}

*x*+

**e**

_{y}

*y*+

**e**

_{z}

*z*and time. Frequency and wavenumber are connected by the dispersion relation

**U**(

**x**,

*t*) is the local horizontal wind of the large-scale flow,

*N*

^{2}(

*z*) is the squared Brunt–Väisälä frequency,

*f*is the Coriolis frequency, and

*K*= (

*k*

^{2}+

*l*

^{2}+

*m*

^{2})

^{1/2}and

*K*

_{h}= (

*k*

^{2}+

*l*

^{2})

^{1/2}are the magnitudes of the total and horizontal wavenumbers, respectively. Wave-action density

*E*

_{gw}and the intrinsic frequency

**v**and

**u**denote the full wind vector and the horizontal wind vector, respectively, and ∇

_{h}stands for the horizontal components of ∇. The GW momentum fluxes

**k**and

#### 2) Pseudomomentum approximation

**k**

_{h}is the horizontal part of the wavenumber vector,

*f*. In future work it is intended to drop both the pseudomomentum approximation and the single-column approximation.

#### 3) Single-column approximation

*c*

_{gz}is the vertical group velocity. This approximation neglects all effects of horizontal GW propagation. Note that the pseudomomentum-flux convergence in the right-hand side of Eq. (9) can be written as (e.g., Achatz et al. 2017)

#### 4) Steady-state approximation and its implications

*S*on the right-hand side of the wave-action-density equation. This is decisive. One sees that in the steady-state approximation the horizontal wavenumber is a constant so that without any source or sink there would be no GW forcing of the mean flow in Eq. (9). Hence, in this approximation, GW dissipation—for example, by GW breaking or close to critical layers—is indispensable for a GW effect on the mean flow, while the explicit description of GW transience as in Eqs. (7) and (8) also allows GW impacts on the mean flow via nondissipative wave–mean-flow interactions.

Consequences of applying the steady-state approximation instead of the transient GW-model Eqs. (7)–(9)—and thus neglecting nondissipative GW–mean-flow interactions—have been studied by Bölöni et al. (2016) in a highly idealized setup using wave-resolving simulations as a reference. They achieved a reliable evolution of the GW energy and the mean flow only using the transient model. In case of using the steady-state equations, important features of the GW–mean-flow interactions were not captured: the GW packet propagated way too fast until static instability set in and its induced mean flow did not agree with the results from wave-resolving simulations. Using a Fourier-ray model (Broutman et al. 2006) and high-resolution WRF (Skamarock et al. 2019) simulations, Kruse and Smith (2018) found that, in the interaction of mountain waves with the mean flow, both dissipative and nondissipative forcings of the mean flow seem to play an important role. The natural question of how important are nondissipative GW–mean-flow interactions in the context of global dynamics has motivated the present study.

### b. Spectral treatment of transient GW distributions

**is a three-dimensional index field and each combination of**

*β***k**

_{β}satisfies Eqs. (3) and (4) or Eqs. (7) and (8) separately. If the corresponding wave amplitudes are weak enough, a superposition of these solutions is a WKB solution of the basic dynamical equations as well, assuming that the required scale separation between the various spectral components and the large-scale mean flow still holds, and one can derive the prognostic equation

**c**

_{g}(

**x**,

**k**,

*t*) = ∇

_{k}Ω is again the group velocity defined in Eq. (2) for wavenumber

**k**, and

**k**as it appears on the right-hand side of Eq. (3);

*D*

_{r}/

*Dt*is a material derivative along trajectories in phase space, so-called rays, tangential to the phase-space velocity

*d*

^{3}

*k*=

*dk*

*dl*

*dm*. Similar expressions can be formulated also without the pseudomomentum approximation (Wei et al. 2019). We note in passing that in the absence of background winds, Eq. (14) agrees with the radiative transfer equation without wave–wave interactions that has been used in the oceanic context for GWPs (Olbers et al. 2019, and references therein). There, however, the shape of the GW spectrum is prescribed, whereas in our implementation it develops without constraints.

In a steady-state approximation, one again neglects the time derivatives in the wave-action density Eq. (17) and in the ray Eqs. (19) and (20). Hence **k**_{h} is again a constant and Eq. (18) yields together with the steady-state version of Eq. (21) the nonacceleration result ∂**U**/∂*t* = 0; that is, the mean flow is unaffected by GWs, unless Eq. (21) is supplemented by sources or sinks.

## 3. Implementation in a high-top atmosphere model

Our single-column pseudomomentum-approximation subgrid-scale GW model applying the transient Eqs. (17)–(20), extended by a saturation scheme, has been named MS-GWaM. It has been implemented into the Icosahedral Nonhydrostatic (ICON) model (Zängl et al. 2015) in its upper-atmosphere configuration UA-ICON (Borchert et al. 2019), allowing numerical studies over a wide altitude range from Earth’s surface to the lower thermosphere. For the sake of simplicity and clear traceability of causes and consequences, the current orographic GWP in UA-ICON, based on Lott and Miller (1997), has been left untouched, and MS-GWaM only replaces the nonorographic GWP there, based on Orr et al. (2010).

As a reference and a representative of currently available GWP schemes, in addition to the transient implementation, two steady-state versions of MS-GWaM have also been implemented to UA-ICON. The first one excludes nondissipative GW–mean-flow interactions through the steady-state approximation but shares all other parameterization components with the transient MS-GWaM, such as GW sources and the saturation scheme. The other one differs from MS-GWaM in its saturation scheme as well, i.e., instead of an integrated treatment of the GW breaking criterion, it applies a monochromatic approach (see the details in section 3b). Throughout the paper, the implementation of the transient MS-GWaM into UA-ICON will be referred to as TR, while the two steady-state implementations will be called ST and STMO, respectively.

### a. Transient scheme

In a global implementation the interaction equations would have to be rewritten in spherical coordinates. The single-column approximation, however, eliminates any horizontal changes of the GW field and all metric terms, which amounts to treating the parameterization equations in local Cartesian coordinates on an *f* plane.

#### 1) GW propagation and interaction with the mean flow

Following Muraschko et al. (2015), we define Lagrangian ray volumes as carriers of the GW fields’ wave-action density and simply trace their positions in phase space. Because of Eq. (17), their spectral wave-action density is conserved, unless wave dissipation is active. Each ray volume is six-dimensional, and its horizontal cross section is given by that of the corresponding ICON column. In the single-column approximation, it does not change so that we suppress it in the following notation. Likewise, horizontal wavenumbers do not change either, but because of the source formulation below we keep track of the ray-volume extent in the corresponding directions.

As illustrated in Figs. 1a and 1b, each ray volume has an extent Δ*z* in *z* direction and extents Δ*k*, Δ*l*, Δ*m* in the three-dimensional wavenumber space. They move, expand, or shrink in the *z* and *m* directions. From *V*_{p} = Δ*z*Δ*m*, do not change. To achieve this, first, changes in the vertical extent of ray volumes are calculated via Eq. (19) as *z* = *z*_{t}) and the bottom (*z* = *z*_{b}) of the ray volume, respectively. Second, the displacement of the ray-volume center point is calculated via Eq. (19) as *m* direction. For the latter, the resolved dynamical fields *N*, ∂_{z}**U**, and ∂_{z}*N* are interpolated to the ray-volume center point *z* = *z*_{c}. Last, the ray-volume extent in the *m* direction is updated by Δ*m* = *V*_{p}/Δ*z*.

*z*=

*z*

_{i+(1/2)}, the integral on the right-hand side of Eq. (18) is approximated as

*N*

_{i}is the number of ray volumes overlapping the vertical layer

*z*

_{i+1}<

*z*

*z*

_{i}with a thickness of

*δz*

_{i}and

*j*is the index over those ray volumes. As can be seen in Fig. 1a, the ratio

*j*th ray volume contributing to the layer centered at

*z*=

*z*

_{i+(1/2)}. After the calculation of the pseudomomentum fluxes at half levels, the mean-flow tendency

*z*=

*z*

_{i}is calculated via Eq. (18) by centered finite differences. The resolved horizontal wind is updated as

*δt*= 60 s is used. To ensure that the ray volumes do not jump over strong, shallow shear layers—and thus describe reflection and critical-layer filtering properly—a fourth-order Runge–Kutta sub–time stepping is used for the integration of Eqs. (19) and (20) with a time step of

*δt*

_{s}= 30 s. Note that the development of

*z*and

*m*via Eqs. (19) and (20) depends on the stratification and the wind shear, i.e., on

#### 2) GW breaking

*m*

^{2}|

*B*|

^{2}>

*N*

^{4}, where

*δz*

_{i}centered at

*z*=

*z*

_{i+(1/2)}—is

*j*-index denote known properties of ray volumes overlapping the layer, and those without are resolved variables at

*z*=

*z*

_{i+(1/2)}. Whenever static instability is diagnosed via Eq. (24), the saturation scheme is called, and the turbulent diffusivity and viscosity are determined so that they exactly counteract the amplitude increase of the contributing GWs that would cause Eqs. (23) or (24) to be satisfied. In the case of buoyancy, for example, the turbulence effect is then captured via

*t*is

#### 3) GW source representation

*ϕ*∈ [0, 2

*π*) so that

*k*=

*K*

_{h}cos

*ϕ*,

*l*=

*K*

_{h}sin

*ϕ*, is

*m*< 0, implying

*m*

_{*}= 2

*π*/(2 km) is used, while the value of

*p*is set to 5/3 on the basis of Warner and McIntyre (1996) and Fritts and Lu (1993). The factor

*C*is a tuning parameter enabling to set a desired launch-level pseudomomentum-flux magnitude

*M*, so that

*ϕ*. To account for the seasonal variability of nonorographic GW sources emitted by jets and fronts, time and latitudinal dependence of

*M*has been introduced as

*φ*is the latitude in degrees,

*β*(

*t*) = {1 + cos[2

*π*(

*t*−

*t*

_{0})]}/2 is a time-dependent function with

*t*

_{0}being 0000 UTC 22 December of the given year, and

*M*

_{bw}and

*M*

_{bs}are respectively the boreal winter and summer flux-magnitude profiles given as

*s*= 11°. The resulting meridional launch flux profile is plotted in Fig. 2 for the northern summer and winter solstices with the choice of (

*M*

_{min},

*M*

_{max}) = (1.5, 2.5) mPa, which are the actual values chosen for the implementation.

*k*,

*l*,

*m*) as needed by MS-GWaM, the

*N*at launch level, the launch spectrum spans

*λ*

_{z}∈ [0.8, 8] km with an increasing resolution in

*m*toward large vertical wave lengths (corresponding to large group velocities) and

*λ*

_{x,y}∈ [46, 1036] km with an increasing resolution in

*k*,

*l*toward large horizontal wave scales. Because

*z*=

*z*

_{l}(=300 hPa in this study) is initialized as

*j*= 1, …,

*N*

_{l}, is the ray-volume index with

*N*

_{l}= 4

*n*

_{c}

*n*

_{ω}being the total number of spectral elements launched at a time in the four azimuthal directions and

*m*

^{j},

*k*

^{j},

*l*

^{j}denoting the wavenumber values at the ray-volume centers. The spectral extent of the

*j*th ray volume in the

*m*direction is calculated as

*k*and

*l*directions are defined as

*ϕ*=

*π*/2 is the azimuthal angle difference between propagation directions.

In the transient framework discussed in this section, the GW emission by nonorographic sources is implemented as a lower boundary condition for Eqs. (17)–(20). This requires that the GW ray volumes are emitted continuously at the launch level for the whole spectrum, so that the total pseudomomentum flux *z*_{l}, a ghost layer is defined with a thickness *δz*_{g}, and at launch time *t* = *t*_{0} the ray volumes are initialized via Eq. (36) with their top matching the bottom of the ghost layer at *z* = *z*_{l} − *δz*_{g} (see Fig. 4a). The five ray volumes sketched in the figure are located in adjacent spectral positions in *m*. At time *t* = *t*_{0} + Δ*t* all ray volumes have propagated vertically, but to different extents (see Fig. 4b). To preserve the shape of the spectrum until the launch level is reached, all ray volumes with a center point below *z* = *z*_{l} propagate without refraction, that is, with *z* = *z*_{l} − *δz*_{g}), a new ray volume is launched so that its top matches the bottom of the previous ray volume in the same spectral position. In Fig. 4b, this happens at the spectral positions 4 and 5, where the “old” ray volumes launched at *t* = *t*_{0} are denoted by a black center point and boundaries while the “new” ray volumes launched at *t* = *t*_{0} + Δ*t* have red center points and boundaries. Two additional features are demonstrated at the spectral position 5: 1) here, the old ray volume has traveled so much within Δ*t*, that even the new ray volume’s bottom ends up at *z* = *z*_{l} − *δz*_{g}, which allows the emission of a second new ray volume right away at the same launch time *t* = *t*_{0} + Δ*t*, and 2) the old ray volume’s center point has passed *z* = *z*_{l} so that the full set of GW–mean-flow interaction Eqs. (17)–(20) begins to act, leading to refractions (displacement of the center point in the *m* direction) and deformations (changes of Δ*z* and Δ*m*). It is repeated here that the above-described GW source is kept simple and nonintermittent on purpose in order to allow a clear separation of transient propagation effects from those implied by intermittent sources.

### b. Steady-state schemes

In this section the steady-state implementations of the single-column GW–mean-flow interaction Eqs. (17)–(20) are presented. In the steady-state context it is assumed that the GWs propagate instantaneously from any source to model bottom and model top and that they instantaneously assume an equilibrium with the resolved mean flow and the source distribution. This equilibrium remains unchanged until source or resolved flow change, when the GW distribution again adjusts instantaneously. As a consequence, GWs cannot influence the resolved flow, unless wave dissipation is active. The mean-flow acceleration by GWs is hence realized exclusively via GW breaking and critical-layer filtering, that is, by diagnosing at what height the equilibrium breaks down as a result of dissipative processes, leading to corresponding pseudomomentum-flux convergences. The next few sections describe the steady-state implementations of MS-GWaM in detail.

#### 1) GW source representation

The spectral characteristics and the magnitude of the nonorographic GW sources are identical to the transient implementation presented in section 3a(3); that is, the GW launch-level pseudomomentum flux is distributed among monochromatic spectral elements characterized in the very same way in spectral space as in the transient case (*λ*_{x,y} ∈ [47, 1036] km, *λ*_{z} ∈ [0.8, 8] km), with the very same values *z* = *z*_{l} can be calculated as *i* = 1, …, *N*_{l} is the spectral-element index with *N*_{l} = 4*n*_{c}*n*_{ω}.

#### 2) Equilibrium profile

*D*

_{r}Ω/

*Dt*= ∂Ω/∂

*t*=

**k**⋅ (∂

**U**/∂

*t*) that one can derive directly from the definitions

**c**

_{g}= ∇

_{k}Ω and

*ω*

^{i}(

*z*

_{l}) are unchanged along a ray. This means that after diagnosing

*ω*

^{i}(

*z*

_{l}) at the launch level, the intrinsic frequency profiles can be calculated based on the known wind profile as

#### 3) Critical-layer filtering and reflection

At critical layers, the intrinsic frequency approaches *f* and the vertical wavenumber diverges, see, for example Eqs. (1) or (38). With decreasing vertical wavelength, a GW eventually becomes unstable and dissipates. In the steady-state picture, critical layers are diagnosed at the lowest altitude *z* = *z*_{c} where *i*, we set the pseudomomentum-flux profile PMF^{i}(*z*) to zero at *z* ≥ *z*_{c}.

When wave reflection occurs, the intrinsic frequency approaches *N* and *m* changes sign so that the group velocity is reverted. In the steady-state versions of MS-GWaM this is taken into account by diagnosing the height of potential reflection by finding the lowest altitude *z* = *z*_{r} where *z* = *z*_{r} for a spectral element, its corresponding pseudomomentum-flux profile PMF^{i}(*z*) is set to zero above *z*_{r}. Unless GW breaking has changed the equilibrium profile below the reflection layer, the pseudomomentum-flux profile PMF^{i}(*z*) vanishes also for *z* < *z*_{r} as well, because under reflection the pseudomomentum-flux changes sign so that the contributions from the upward- and downward-propagating components cancel each other when the mean flow is in a steady state. However, if GW breaking takes place (see the next section) at any altitude *z* = *z*_{b} < *z*_{r}, the pseudomomentum fluxes carried by the upward- and downward-propagating spectral elements do not completely cancel; thus, at altitudes *z* < *z*_{b} a residual

#### 4) GW breaking

In the steady-state setups of MS-GWaM the instability criterion Eq. (23) is used as well. The two steady-state implementations (ST and STMO) differ, however, in the way this is done and how the GW amplitudes are adjusted whenever wave breaking is diagnosed.

*z*=

*z*

_{b}if

*z*=

*z*

_{b}if Eq. (23) is fulfilled there, with the integral taken over all spectral components, so that

*c*

_{gz}in Eqs. (42) and (43) [cf. Eqs. (26) and (27)] so that, for a given vertical distance and diffusivity, slowly propagating waves tend to dissipate more than those propagating fast. This implementation is called steady-state MS-GWaM, or ST. Given that it shares the treatment of the GW sources and GW breaking with the transient implementation, the only difference between ST and TR is how the propagation is accounted for, that is, GW transience.

*z*

_{i+(1/2)}.

As in the transient implementation, the present study does not take corresponding effects on frictional heating and GW energy deposition into account.

#### 5) Mean-flow forcing

### c. Stability measures and computational aspects

To facilitate GW studies in a large altitude range, our model top within UA-ICON has been set to 150 km. In UA-ICON and ICON in general a sponge layer prevents spurious wave reflections from the model top, based on a Rayleigh damping applied to the vertical wind (Zängl et al. 2015). In the setup used here the bottom of the sponge layer is at 110 km. Several measures had to be taken in MS-GWaM to prevent numerical instabilities in the sponge, because of excessive mean-flow accelerations by insufficiently controlled GW pseudomomentum fluxes.

#### 1) Molecular viscosity

*η*=

*η*[

*T*(

*z*,

*t*)] based on Sutherland’s viscosity law (see, e.g., Atkins and Escudier 2013). In the steady-state implementations the same prescription is used, but proceeding from layer to layer, and with

#### 2) Scale height correction

The WKB theory applied by Achatz et al. (2017) predicts that in case of a clear scale separation between GWs and a resolved flow, the former obey the Boussinesq GW dispersion relation in Eq. (1). In the numerical implementation (i.e., TR, ST, and STMO), however, the scale separation does not always hold. Vertical GW wavelengths can grow by refraction and eventually reach values similar to the scales of vertical variations of the resolved mean flow. An ideal treatment of such a situation would be to somehow “transfer” the large-scale GW to the resolved flow and stop treating it as a subgrid-scale wave. A theory for such a procedure, however, is not known to us, and the problem is complicated further by the possibility of such a wave still being unresolved in the horizontal.

*H*

_{ρ})] − (1/

*H*

_{θ}) is the pseudoincompressible scale height with

*H*

_{θ}= [(1/

*θ*)(

*dθ*/

*dz*)]

^{−1}being the density scale height and the potential temperature scale height, respectively. By assuming that the atmosphere is locally isothermal, the pseudoincompressible scale height in TR, ST, STMO is further simplified to Γ = [1/(2

*H*

_{ρ})][(1/2) − (

*R*/

*c*

_{p})] with

*R*being the ideal gas constant and

*c*

_{p}being the specific heat capacity at constant pressure. From the modified dispersion relation, one obtains for the vertical group velocity

*m*becomes

*m*

^{2}becomes small enough so that it is only an order of magnitude larger than Γ, that is, when 0.1 ×

*m*

^{2}~ Γ

^{2}. By substituting

*m*= 2

*π*/

*λ*

_{z},

*H*

_{ρ}≈ 5–8 km, and

*R*/

*c*

_{p}≈ 2/7, one arrives at a vertical wavelength of

*λ*

_{zcorr}≈ 45–75 km. Although, in the launch spectrum used in this paper, the vertical wavelength is not larger than

*λ*

_{z}= 8 km, the transient implementation has been observed to lead to vertical GW scales as large as

*λ*

_{zcorr}. Thus, to be on the safe side, and for being fully consistent between the transient and steady-state schemes, all implementations of MS-GWaM have been based on the discretized versions of Eqs. (48)–(50) to describe the evolution of the GW field.

#### 3) Pseudomomentum-flux smoothing

In the TR implementation, because of unavoidable local undersampling of ray volumes, pseudomomentum-flux profiles can get noisy so that the GW impact on the resolved flow can exhibit undesired spikes. Thus, a crucial numerical aspect to stabilize TR simulations has been to apply a vertical smoothing on the pseudomomentum fluxes after the projection via Eq. (22) and before calculating the resolved wind tendencies. The smoothing is using the zeroth-order filter of Shapiro (1975), which removes noise with length scales of 2*δz* but leaves larger-scale structures mostly unaffected.

#### 4) Controlling the total number of ray volumes

To prevent excessive computational costs, the total number *N*_{c} of ray volumes per column is limited to a value *N*_{cmax}. It has been found that in terms of the time-averaged zonal-mean circulation, a numerical convergence of the TR simulations has been achieved by using *N*_{cmax} = 2500 if using a source with *N*_{l} = 4 × *n*_{c} × *n*_{ω} = 4 × 6 × 2 = 48 [see section 4b(4)]. The practical implementation is simple: each time step before the call to the ray-volume emission at the launch level, it is checked columnwise whether *N*_{c} > *N*_{cmax}. If this is the case in a column, *N*_{c} − *N*_{cmax} of lowest-energy ray volumes are removed.

#### 5) Computational costs

Table 1 shows the computational costs of TR, ST, STMO, and the operational GW drag scheme used in ICON for NWP purposes (Orr et al. 2010). The computational costs are presented in terms of 1) *t*_{tot}, that is, total run times of 1-month simulations with UA-ICON using the different GW schemes (see the Table 1 caption for the grid spacing) and 2) *t*_{av}, that is, average time spent on a single call of the subroutines corresponding to the different parameterizations. The TR scheme is ~5 times as expensive as ST in terms of *t*_{av}, which leads to about a factor ~2.5 of overhead costs in terms of *t*_{tot}. This is what transience costs. If the wave breaking scheme is monochromatic (STMO), and as such simpler, accelerations by a factor ~2.3 in terms of *t*_{av}, and by a factor ~1.3 in terms of *t*_{tot} can be achieved. There is a further acceleration by factors ~4.5 and ~1.2 between STMO and the operationally used (Orr et al. 2010) scheme in terms of *t*_{av} and *t*_{tot}, respectively. Hence TR is ~50 and ~4.1 times more costly than the operational scheme in terms of *t*_{av} and *t*_{tot}, respectively.

Computational costs of the different GW parameterizations coupled to UA-ICON on 960 CPUs with a horizontal grid spacing of ~160 km (R2B4 grid) and with 120 vertical levels up to 150 km with the same distribution as described by Borchert et al. (2019).

With regard to the costs in memory, TR simulations use 2% more memory than ST simulations, where 100% stands for the memory cost of the ST simulations. This is not negligible but is relatively small.

## 4. Results

### a. Experimental setup

The first step in order to validate the implementation of MS-GWaM was to reproduce the idealized one-dimensional cases of Bölöni et al. (2016) in UA-ICON. This technical step has been followed by global simulations using TR, ST, STMO, with a horizontal grid spacing of ~160 km (R2B4 grid^{1}). A stretched vertical grid has been used with layer thicknesses gradually increasing with height, with a typical thickness of a few tens of meters in the boundary layer, 700–1500 m in the stratosphere, and a maximum of ~4 km in the lower thermosphere. Similar to Borchert et al. (2019), a model top at ~150 km has been used with a sponge layer acting above 110 km. As initial condition, operational Integrated Forecasting System (IFS) of the European Centre for Medium-Range Weather Forecasts (ECMWF) (https://www.ecmwf.int/en/publications/ifs-documentation) analyses have been used. They have been interpolated to the ICON grid at altitudes covered by IFS/ECMWF and extrapolated toward a simple climatology above. The first few weeks’ simulations after each initialization have been discarded from any scientific analysis to make sure that the adjustment process from the climatology toward the actual realization of the circulation at higher altitudes is excluded.

### b. Mean circulation

The first proof of concept for MS-GWaM in a global modeling framework was to validate the zonal-mean circulation it generates as coupled to UA-ICON. For this validation the *UARS* Reference Atmosphere Project (URAP) data (Swinbank and Ortland 2003) have been used as a reference, because this dataset involves zonal-mean climatologies up to rather high altitudes, i.e., 85 km for temperature and 110 km for zonal wind.

#### 1) Zonal-mean wind and temperature

Simulations with TR, ST, and STMO have been run for the 8 URAP years (1991–98) for December and June (initialized on the 1 November and 1 May, respectively). Figures 5 and 6 respectively show the time-averaged zonal-mean zonal winds and temperatures from the TR and ST simulations as well as the reference URAP data. In general, both the TR and ST simulations produce a very similar zonal-mean circulation (results from the STMO simulations are not shown due to their similarity to ST), which compares reasonably well with URAP. Both models capture the reversal of the summer mesospheric jet although somewhat too low in altitude both in December and June and too weak in June. The corresponding summer mesopause is too cold by ~20–30 K, which might be explained by the fact that the thermal effects of energy deposition (e.g., Becker 2017) of GWs are ignored for the time being. The polar night jet is reasonably well placed, but its magnitude is overestimated in both the TR and ST simulations in both months, especially in June. The stratospheric easterly jet cores are placed too much equatorward in both models and both months. Based on this qualitative comparison, the similarity between the TR and ST suggests that transience does not play a very important role in terms of seasonal-mean and zonal-mean circulation. This does not come as a surprise, as indeed, spatial and time averaging should hide local, short-lived transient effects and eventually reflect a quasi-steady-state circulation of the respective months. At a second look, a sharp eye will spot already nonnegligible differences between TR and ST simulations in Figs. 5 and 6. For instance, the magnitude of the June polar night jet is overestimated to a larger extent, while the June lower-thermospheric jet magnitude in Northern Hemisphere (NH) is underestimated to a larger extent in the ST simulation.

#### 2) Residual circulation and zonal-mean GW drag

The residual circulation of UA-ICON with MS-GWaM is presented in Fig. 7 by plotting the residual-mean mass streamfunction along with the corresponding meridional velocity *υ** in the transformed Eulerian mean (TEM) equations [Andrews and McIntyre 1978; Hardiman et al. 2010, their Eq. (19)]. Both TR and ST simulations result in a qualitatively similar circulation as presented by, for example, Smith (2012) and Becker (2017). It appears that ST simulations lead to a stronger *υ** in the upper mesosphere in comparison with TR simulations, implying that the vertical branch of the residual circulation near the poles is also stronger in ST simulations. This is in line with the somewhat colder temperatures at the summer mesopause regions in ST simulations as compared to TR simulations, because a stronger residual circulation corresponds to stronger cooling in the summer upper mesosphere and heating in the winter lower mesosphere. The difference in the residual circulations of TR and ST can be explained by the zonal-mean GW drag (Fig. 8) from both simulations. The structure of GW drag is well matched with that of *υ** in the mesosphere, demonstrating the impact of GWs on the residual circulation. The GW drag of ST in the MLT is larger than that of TR by ~80 m s^{−1} day^{−1} in December and by ~160 m s^{−1} day^{−1} in June. This corresponds well to the differences found in the strength of the residual circulation between TR and ST and reflects that adding transience to a GWP has important implications on the mean circulation and the heat budget.

#### 3) Perpetual runs

A more comprehensive appreciation of the differences between the simulations with the transient and the steady-state GW schemes has been enabled by running long perpetual December simulations with TR, ST, and STMO. The perpetual runs have been achieved by imposing a constant radiative and surface forcing, corresponding to 22 December 1992, including a diurnal cycle. The simulations have been run for 24 months of which the last 12 months have been used for comparison. Mean wind differences between the TR and ST simulations (Fig. 9a) are larger in magnitude, and more statistically significant, than those between the two steady-state simulations ST and STMO (Fig. 9b). This shows that the impact of GW transience is somewhat larger than that of the change in the saturation scheme between ST and STMO [see section 3b(4)], even in the context of the time averaged zonal-mean circulation.

#### 4) Numerical convergence

As a validation of the employed maximum number of ray volumes per column *N*_{cmax} = 2500, we show in Fig. 9c the mean-wind difference between perpetual December TR simulations using *N*_{cmax} = 2500 and *N*_{cmax} = 5000. These differences are clearly lower in magnitude and less statistically significant than those between the ST and TR. This demonstrates that the effect of transience is much larger than the effect of doubling the amount of ray volumes in the TR simulations. It confirms both that the TR simulations using *N*_{cmax} = 2500 are numerically converged and that the difference due to transient GW propagation (ST − TR) is robust; that is, it reflects a physical feature and not a numerical uncertainty.

### c. GW pseudomomentum fluxes

Apart from the time averaged zonal-mean circulation, temporal and spatial variability of the GW pseudomomentum fluxes is of interest. As will be shown, the modulation of the GW spectrum through transient propagation leads to fundamentally different pseudomomentum-flux magnitudes and spatial structures as compared to the steady-state GW schemes.

#### 1) Intermittency and variability

A simple quantification of GW intermittency is the histogram of pseudomomentum fluxes, i.e., the probability of occurrence of various pseudomomentum-flux values at given geographical locations. Following Hertzog et al. (2012), histograms of GW absolute zonal pseudomomentum fluxes have been plotted for TR, ST (Fig. 10), and STMO (not shown but similar to ST) with a similar spatial and temporal sampling as in the above-mentioned paper (see figure captions). The difference between TR and ST is obvious, showing a much better fit of the TR simulations to the observed histograms based on the Vorcore superpressure balloons and the HIRDLS satellite (see Fig. 2 in Hertzog et al. 2012). The low intermittency of the ST simulations is not surprising, since steady-state schemes with a nonintermittent source—such as used here—are known to underestimate the occurrence of high pseudomomentum fluxes. Due to the fact that in the steady-state approximation only dissipative effects—due to wave breaking or close to critical layers—can lead to pseudomomentum-flux variations, no higher values can occur than the launch-level pseudomomentum-flux magnitudes. With the GW source used in this study, the launch-level absolute zonal pseudomomentum-flux magnitude in October is ~4 mPa. Indeed, in the ST simulations no higher values occur than that. In contrast, in the TR simulations at *z* ≈ 20 km, pseudomomentum fluxes of 60 mPa occur with a nonzero probability, which means that fluxes happen to grow by a factor of 15 at this altitude with respect to their launch values. Figure 10 also shows that, up to flux values of ~30 mPa, the probability of large fluxes decreases with altitude, which is in line with the findings of, for example, de la Cámara et al. (2016) in this respect. The probability of occurrence for flux values larger than ~30 mPa shows a vertical dependence that has never been found in steady-state GWPs: it is increasing with altitude between *z* ≈ 20 km and *z* ≈ 40 km, and then it drops down significantly above.

To understand the vertical dependence of GW intermittency in the TR simulations and to further illustrate the large difference between the TR and ST simulations, Hovmöller diagrams of absolute pseudomomentum fluxes are shown in Fig. 11. Obviously, in the ST simulation pseudomomentum-flux magnitudes decrease monotonically with altitude, while in the TR simulation slanted stripes of increased values with time and altitude demonstrate that GW packets gain pseudomomentum flux in a nondissipative manner in the course of their propagation up to the altitude of 50–70 km and then they dissipate due to saturation. The only way the nondissipative increase can happen—**k**_{h} being constant—is via variation of *U*/∂z > 0 between the launch level and *z* ≈ 55–60 km. Then, given *m* < 0, this shear tends to shift westward-propagating (eastward propagating) GWs toward larger (smaller) vertical wavelengths and thus larger (smaller) vertical group velocities in this altitude range. Hence the effect of local vertical group-velocity increase can only play a role for the westward-propagating GWs. Separate Hovmöller diagrams for the westward and eastward pseudomomentum fluxes (not shown), however, reflect similar levels of intermittency as for the absolute values (Fig. 11). Thus, the variation of

#### 2) Zonal and time mean

An interesting consequence of nondissipative GW–mean-flow interactions is that the nondissipative pseudomomentum-flux convergence is reflected not only locally and for short periods, but also in the time averaged zonal mean. This is illustrated in Figs. 12a–f, where monthly-mean (Octobers of 1991–98) zonal-mean pseudomomentum fluxes from TR simulations turn out to be larger than those obtained from ST simulations everywhere below *z* ~ 40 km. This is the mean effect of the transient flux changes shown in Fig. 11, which—as explained above—should be due to local variations of

The pseudomomentum-flux differences between the TR and ST simulations can be put in the context of the missing drag—a general underestimation of the GW forcing at about 60°S by GCMs (McLandress et al. 2012). In particular, Jewtoukoff et al. (2015) showed that the relatively high-resolution operational IFS/ECMWF analyses are underestimating the GW momentum fluxes by a factor of 5 over the Southern Ocean at about 20-km altitude, as compared with superpressure balloon observations. In addition, de la Cámara et al. (2016) showed that the parameterized GW fluxes in the Laboratoire de Météorologie Dynamique Zoom (LMDz) model agree with those resolved by the operational IFS/ECMWF to a good degree, indicating that some state-of-the-art GCMs suffer from an underestimation of GW pseudomomentum fluxes by about a factor of 5. Several studies suggested that part of this underestimation originates from the lack of orographic drag due to small islands not represented in the topographic databases of GCMs (McLandress et al. 2012; Alexander et al. 2009; Alexander and Grimsdell 2013; Garfinkel and Oman 2018) and others proposed that some of the underestimation is due to the lack of horizontal GW propagation in GWPs (Sato et al. 2009; Ehard et al. 2017) or the misrepresentation of nonorographic sources (Hendricks et al. 2014; de la Cámara et al. 2016). It appears, however, that the lack of transience in present-day GWPs might also be responsible for a small but nonnegligible fraction of the missing drag. This is demonstrated in Figs. 12g–i, where horizontal maps of absolute pseudomomentum fluxes are plotted at *z* ≈ 20 km above the Southern Ocean from the TR (Fig. 12g) and the ST (Fig. 12h) simulations. As expected from the cross sections in Figs. 12c and 12f, the absolute pseudomomentum-flux values from the ST simulations are smaller than those from TR, and as shown in Fig. 12i, if ST fluxes are multiplied by a factor of 1.3, a relatively close match with TR is achieved. Transience thus brings an increase of 30% in terms of absolute momentum fluxes, which is, however, still very far from the missing 500% reported by, e.g., Jewtoukoff et al. (2015). The difference between TR and ST simulations can also be expressed in terms of the zonal GW drag. The drag averaged over *φ* ∈ [−65°, −55°], *z* ∈ [20, 50] km and over the Octobers of 1991–98 is −0.291 m s^{−1} day^{−1} from ST and −0.476 m s^{−1} day^{−1} from TR simulations. Hence transience seems to increase the drag by about 60%. The nonnegligible effect discussed above shows that the transience does matter even over monthly time scales. It is also recalled that all differences presented between ST and TR simulations in this paper are due to the nonorographic fluxes only, in a completely nonintermittent GW source setup.

### d. Contribution of different wavelengths to the GW signal

Given that MS-GWaM is a spectral scheme, a decomposition of the GW momentum fluxes and drag into the contributions from different wavelengths is straightforward. Such a decomposition could be of interest for validation purposes against observations if GW sources were realistically taken into account. This is yet not the case here; however, even with the simple GW source used in this study, a decomposition by scales is useful to get a simple first guess about the required horizontal and vertical resolutions for GW resolving simulations. The decomposition is based on the TR implementation of MS-GWaM given its additional realism as compared to ST, i.e., given transience and the prognostic treatment of the vertical wavenumber spectrum. The contribution of GWs with different spatial scales to the pseudomomentum fluxes has been diagnosed by calculating Eq. (22) for a subset of the ray volumes *j* = 1, …, *N*_{i} for which certain conditions hold with respect to their horizontal (*λ*_{h}) or vertical wavelengths (*λ*_{z}). The corresponding drag contribution has been calculated via Eq. (18) (its discretized form) just like for the full drag. These diagnostics have all been achieved in an offline mode, meaning that the resolved flow has been forced with the total drag imposed by the total pseudomomentum fluxes.

#### 1) Decomposition results

The contribution of GWs with different spatial scales to the total absolute flux and drag is shown in Fig. 13. Figure 13a shows the zonal-mean total absolute pseudomomentum flux and drag averaged over Junes of 1991–98. Figures 13b–e suggest that excluding horizontal wavelengths smaller than 50, 100, 200, and 250 km leads to signal losses of ~20%, 50%, 75%, and 75%–85%, respectively, both with respect to fluxes and the drag. The contribution of GWs with different vertical scales can be seen by comparing the total signal with Figs. 13f–i, where vertical wavelengths smaller than 1, 2, 5, and 10 km are excluded, respectively. Here the drag signal is much less affected; namely, no loss can be seen if having contributions from waves with *λ*_{z} > 5 km, and only ~25% is lost if waves with *λ*_{z} < 10 km are excluded. In terms of fluxes, however, the loss of signal below *z* ≈ 40 km is larger than ~50% if GWs with *λ*_{z} < 2 km are excluded, which increases to a loss of ~75%–80% if GWs with *λ*_{z} < 10 km do not contribute.

The contribution of GWs with different scales to the total intermittency has been examined as well (not shown). It turns out that at *z* = 20 km, occurrence of large momentum fluxes (≳10 mPa) is completely lost if waves with horizontal scales smaller than 100 km are excluded, leading to similarly unrealistic intermittency curves as obtained from ST simulations. If only small-scale GWs with horizontal scales *λ*_{h} < 50 km are left out, most of the total intermittency is reproduced, leaving us with a loss of at most ~30% for all flux values. Excluding the smallest vertical scales (*λ*_{z} < 2 km) does not affect intermittency, but leaving out even larger-scale waves (*λ*_{z} < 5 or 10 km) reduces the occurrence of fluxes between 5 and 40 mPa significantly.

#### 2) Consequences for GW resolving simulations

Explicitly resolving GWs instead of parameterizing them is recently of increasing interest even in global simulations. In the light of the above, a simple estimate of the required spatial resolution can be given: to get most of the GW signal one needs to resolve horizontal scales of 50 km or smaller and vertical scales of 2 km or smaller. Because in NWP models and GCMs the effective resolution of a given spatial scale *λ* requires 7–10 grid points per *λ*, the necessary horizontal and vertical grid spacings to be used for GW resolving simulations can be estimated as Δ*x* < 5 km and Δ*z* < 200 m, respectively. This estimate has to be treated with caution because it does not take into account that resolving the generation (GW source mechanisms), and dissipation of GWs might require even higher spatial resolution than is suggested by the scale decomposition applied here.

## 5. Summary and conclusions

This paper describes the first implementation—to the best of our knowledge—of a transient subgrid-scale GW parameterization into a state-of-the-art GCM and NWP model. This parameterization is called Multiscale Gravity Wave Model (MS-GWaM). It does not rely on the steady-state approximation and therefore enables both dissipative and nondissipative GW–mean-flow interactions, whereas standard GW parameterizations assume an instantaneous equilibrium between GWs, mean flow, and sources, thereby leaving room only for dissipative forcing. For an estimate of the GW-transience impact, a steady-state version of MS-GWaM (ST), using exactly the same GW saturation scheme, and coupled to the same GW source, has been implemented and used as reference for the transient GW parameterization (TR). The TR implementation of MS-GWaM differs in several respects from other GWPs in the literature that use ray tracing. Song and Chun (2008) as well as Amemiya and Sato (2016) have implemented somewhat similar GWPs into state-of-the-art GCMs. They have, however, kept the steady-state assumption in the prediction of the wave amplitudes via wave-action conservation. As compared with earlier transient implementations (Senf and Achatz 2011; Ribstein et al. 2015), one main difference is that TR MS-GWaM allows a feedback from the resolved mean flow to the subgrid-scale GW field through the ray equations, which is especially not the case for Senf and Achatz (2011). Also, MS-GWaM applies the phase-space representation (section 2b), which, so far, is the only viable solution to avoid numerical problems that arise as a result of caustics. Ribstein and Achatz (2016) already used a fully coupled ray tracer including the phase-space approach, however, not in a GCM but in a more-simple tidal model, similar to Senf and Achatz (2011). Last, but not least, the wave breaking scheme of TR MS-GWaM is also a point that makes an important difference with respect to other GWPs, in that the saturation is diagnosed with a contribution from the full GW spectrum represented by the parameterization at a given altitude at a given time.

The time averaged zonal-mean circulation turned out to be broadly similar in TR and ST simulations, both of them agreeing reasonably well with observations (URAP data by Swinbank and Ortland 2003). Closer inspection shows, however, that in some aspects TR yields slightly better results than ST. By excluding interannual variability via perpetual runs, it has also been shown that the effect of transience is larger than that of varying the saturation scheme in the steady-state implementation, especially in the mesosphere and lower thermosphere. That the summer mesopauses are too cold both in TR and ST simulations is likely a consequence of ignoring thermal effects of energy deposition by GWs. Having a leading-order thermal effect in the MLT (e.g., Becker 2017), this process will have to be included into MS-GWaM. Another finding in the same context is that temperature errors at summer mesopauses are smaller in TR simulations than in ST simulations, which is explained by the weaker residual circulation driven by weaker zonal-mean net GW drag in the MLT region. This is a sign that transient effects do not average out completely and may have important implications on the mean zonal and meridional circulations.

Even more evident differences between TR and ST simulations are found in terms of GW pseudomomentum-flux variability. As expected from earlier studies (e.g., de la Cámara et al. 2016), ST simulations strongly underestimate the intermittency of GW pseudomomentum fluxes (occasional occurrence of large values), while TR simulations lead to considerably more realistic results. The reason for this is that the steady-state assumption only allows dissipative effects to change GW-pseudomomentum fluxes, and hence only allows them to decrease as compared to the source, while nondissipative GW–mean-flow interactions can also lead to an increase of these fluxes. This effect is not only visible locally and over short time scales, but it also affects monthly averages of zonal means: mean pseudomomentum fluxes in the lower stratosphere are ~30% larger in TR simulations than in ST. In the Southern Hemisphere, this is where a missing GW drag has been diagnosed by several studies (McLandress et al. 2012; Jewtoukoff et al. 2015). Hence the neglect of transient GW–mean-flow interactions in standard GW parameterizations might contribute to this issue in a modest extent, beside the lack of lateral propagation (Sato et al. 2009; Ehard et al. 2017), the misrepresentation of nonorographic sources (Hendricks et al. 2014; de la Cámara et al. 2016) or the lack of orographic drag due to missing islands in the insufficiently detailed model topographies (Alexander et al. 2009; McLandress et al. 2012; Alexander and Grimsdell 2013; Garfinkel and Oman 2018).

Increasing the realism of GW parameterizations by including transient wave–mean-flow interactions is seen by us as only a first step. Lateral GW propagation will have to be included as well, which—on the basis of Senf and Achatz (2011); Kalisch et al. (2014); Ribstein et al. (2015); Amemiya and Sato (2016)—changes several aspects of the GW distribution and its impact on the mean flow. Corresponding work has just begun, after a six-dimensional (6D)^{2} version of MS-GWaM has been successfully implemented into the same *f*-plane pseudoincompressible flow solver as used by Bölöni et al. (2016) and Wei et al. (2019). More realistic source schemes are also an issue. In Part II of this study (Kim et al. 2021), we report on the effects of coupling MS-GWaM in ICON to a convective GW-source scheme, and more improvements with regard to mountain waves and GWs due to jets and fronts will have to follow. As pointed out by Plougonven et al. (2020), one should always be aware that a realistically looking large-scale circulation is no proof that the parameterization is correct. Instead, the parameterized processes will have to be studied by measurements and wave-resolving simulations as well, and it will have to be made sure that all parts of the parameterization reproduce the properties identified therein. Only then can we have a guarantee that the GWP will be reliable even in a changing climate.

The reader might wonder whether the computational cost of a Lagrangian ray-tracing approach as suggested here is not too overwhelming. As summarized in Table 1 and in section 3c(5), according to the strictest measure (*t*_{av}), including transient effects increases the computational costs by a factor of ~5 with respect to the ST implementation of MS-GWaM, and by a factor of ~50 with respect to Orr et al. (2010)—the current operational scheme used in the NWP configuration of ICON. The discrepancy in computational costs by a factor of ~10 between the steady-state scheme ST and Orr et al. (2010)—which should perform calculations of similar complexity—suggests that MS-GWaM’s efficiency in general (both TR and ST) could probably be improved by means of code optimization. On the basis of this assumption, an optimized transient MS-GWaM should be about a factor of ~5 more costly than state-of-the-art GW schemes. For the time being it cannot be excluded that lateral GW propagation might increase the costs further, although there is no reason to expect that more ray volumes will be needed per column than are already used in the present MS-GWaM implementation. Keeping in mind other potential overhead costs, such as the MPI communication of ray volumes, a safe estimate for a 6D version of MS-GWaM is approximately a factor-of-10 increase of computational costs, as compared with standard steady-state GW parameterizations. This might be seen as a large increase in costs; however, relating it to costs of other alternatives—such as GW-resolving simulations—might quickly change one’s perspective. As also suggested in section 4d, GW resolving simulations would require a horizontal grid spacing of 5 km (or smaller, e.g., 1 km) and a vertical grid spacing of 200 m. If this requirement were to be satisfied with respect to the horizontal resolution alone, the computational costs (in terms of *t*_{tot}) would increase by a factor of ~30 000 (for 5 km) or ~5 million (for 1 km). The vertical resolution increase to 200 m everywhere above the troposphere would lead to a cost increase of a further factor of ~8, ending up with something between a factor of 240 000 and a factor of 40 million. Therefore, already in its present state, ICON/MS-GWaM can be a useful tool for research purposes, allowing much less costly simulations than those resolving GWs globally and more realistic than achievable by standard GCM resolutions with classic steady-state parameterizations. Once flow-dependent sources for GWs from orography and jet–frontal systems have been implemented, it will be ready, for example, to accompany field campaigns and help in interpreting their results. The long-term goal of eventually using ICON/MS-GWaM in climate simulations and weather forecasting, however, is also not to be left out of sight.

## Acknowledgments

The authors thank the German Research Foundation (DFG) for partial support through the research unit Multiscale Dynamics of Gravity Waves (MS-GWaves) and through Grants AC 71/8-2, AC 71/9-2, AC 71/10-2, AC 71/11-2, AC 71/12-2, BO 5071/2-2, BO 5071/1-2, and ZA 268/10-2. Calculations for this research were conducted on the supercomputer facilities of the Center for Scientific Computing of the Goethe University Frankfurt. This work also used resources of the Deutsches Klimarechenzentrum (DKRZ) granted by its Scientific Steering Committee (WLA) under Project bb1097.

## Data availability statement

The ICON software is freely available to the scientific community for noncommercial research purposes under a license from DWD and MPI-M. Potential users who would like to obtain ICON can contact icon@dwd.de. The MS-GWaM code and its module for an implementation in ICON have been developed at Goethe-Universität Frankfurt am Main. Please contact Prof. Ulrich Achatz (achatz@iau.uni-frankfurt.de) for these. The URAP wind and temperature data are available online (https://www.sparc-climate.org/data-centre/data-access/reference-climatology/urap/), as are the ERA5 reanalysis data (https://cds.climate.copernicus.eu).

## REFERENCES

Achatz, U., B. Ribstein, F. Senf, and R. Klein, 2017: The interaction between synoptic-scale balanced flow and a finite-amplitude mesoscale wave field throughout all atmospheric layers: Weak and moderately strong stratification.

,*Quart. J. Roy. Meteor. Soc.***143**, 342–361, https://doi.org/10.1002/qj.2926.Alexander, M. J., and T. J. Dunkerton, 1999: A spectral parameterization of mean-flow forcing due to breaking gravity waves.

,*J. Atmos. Sci.***56**, 4167–4182, https://doi.org/10.1175/1520-0469(1999)056<4167:ASPOMF>2.0.CO;2.Alexander, M. J., and A. W. Grimsdell, 2013: Seasonal cycle of orographic gravity wave occurrence above small islands in the Southern Hemisphere: Implications for effects on the general circulation.

,*J. Geophys. Res. Atmos.***118**, 11 589–11 599, https://doi.org/10.1002/2013JD020526.Alexander, M. J., S. D. Eckermann, D. Broutman, and J. Ma, 2009: Momentum flux estimates for South Georgia Island mountain waves in the stratosphere observed via satellite.

,*Geophys. Res. Lett.***36**, L12816, https://doi.org/10.1029/2009GL038587.Amemiya, A., and K. Sato, 2016: A new gravity wave parameterization including three-dimensional propagation.

,*J. Meteor. Soc. Japan***94**, 237–256, https://doi.org/10.2151/jmsj.2016-013.Andrews, D. G., and M. E. McIntyre, 1976: Planetary waves in horizontal and vertical shear: The generalized Eliassen-Palm relation and the mean zonal acceleration.

,*J. Atmos. Sci.***33**, 2031–2048, https://doi.org/10.1175/1520-0469(1976)033<2031:PWIHAV>2.0.CO;2.Andrews, D. G., and M. E. McIntyre, 1978 a: An exact theory of nonlinear waves on a Lagrangian-mean flow.

,*J. Fluid Mech.***89**, 609–646, https://doi.org/10.1017/S0022112078002773.Andrews, D. G., and M. E. McIntyre, 1978b: Generalized Eliassen-Palm and Charney-Drazin theorems for waves on axisymmetric mean flows in compressible atmospheres.

,*J. Atmos. Sci.***35**, 175–185, https://doi.org/10.1175/1520-0469(1978)035<0175:GEPACD>2.0.CO;2.Atkins, T., and M. Escudier, 2013:

*A Dictionary of Mechanical Engineering*. Oxford University Press, 448 pp., https://doi.org/10.1093/acref/9780199587438.001.0001.Bacmeister, J., P. Newman, B. Gary, and K. Chan, 1994: An algorithm for forecasting mountain wave–related turbulence in the stratosphere.

,*Wea. Forecasting***9**, 241–253, https://doi.org/10.1175/1520-0434(1994)009<0241:AAFFMW>2.0.CO;2.Becker, E., 2017: Mean-flow effects of thermal tides in the mesosphere and lower thermosphere.

,*J. Atmos. Sci.***74**, 2043–2063, https://doi.org/10.1175/JAS-D-16-0194.1.Beres, J. H., R. R. Garcia, B. A. Boville, and F. Sassi, 2005: Implementation of a gravity wave source spectrum parameterization dependent on the properties of convection in the Whole Atmosphere Community Climate Model (WACCM).

,*J. Geophys. Res.***110**, D10108, https://doi.org/10.1029/2004JD005504.Bölöni, G., B. Ribstein, J. Muraschko, C. Sgoff, J. Wei, and U. Achatz, 2016: The interaction between atmospheric gravity waves and large-scale flows: An efficient description beyond the nonacceleration paradigm.

,*J. Atmos. Sci.***73**, 4833–4852, https://doi.org/10.1175/JAS-D-16-0069.1.Borchert, S., G. Zhou, M. Baldauf, H. Schmidt, G. Zängl, and D. Reinert, 2019: The upper-atmosphere extension of the ICON general circulation model (version: ua-icon-1.0).

,*Geosci. Model Dev.***12**, 3541–3569, https://doi.org/10.5194/gmd-12-3541-2019.Bretherton, F. P., 1966: The propagation of groups of internal gravity waves in a shear flow.

,*Quart. J. Roy. Meteor. Soc.***92**, 466–480, https://doi.org/10.1002/qj.49709239403.Broutman, D., J. Ma, S. D. Eckermann, and J. Lindeman, 2006: Fourier-ray modeling of transient trapped lee waves.

,*Mon. Wea. Rev.***134**, 2849–2856, https://doi.org/10.1175/MWR3232.1.Charron, M., and E. Manzini, 2002: Gravity waves from fronts: Parameterization and middle atmosphere response in a general circulation model.

,*J. Atmos. Sci.***59**, 923–941, https://doi.org/10.1175/1520-0469(2002)059<0923:GWFFPA>2.0.CO;2.de la Cámara, A., and F. Lott, 2015: A parameterization of gravity waves emitted by fronts and jets.

,*Geophys. Res. Lett.***42**, 2071–2078, https://doi.org/10.1002/2015GL063298.de la Cámara, A., F. Lott, V. Jewtoukoff, R. Plougonven, and A. Hertzog, 2016: On the gravity wave forcing during the southern stratospheric final warming in LMDZ.

,*J. Atmos. Sci.***73**, 3213–3226, https://doi.org/10.1175/JAS-D-15-0377.1.Ehard, B., and Coauthors, 2017: Horizontal propagation of large-amplitude mountain waves into the polar night jet.

,*J. Geophys. Res. Atmos.***122**, 1423–1436, https://doi.org/10.1002/2016JD025621.Fritts, D., and T. J. Dunkerton, 1985: Fluxes of heat and constituents due to convectively unstable gravity waves.

,*J. Atmos. Sci.***42**, 549–556, https://doi.org/10.1175/1520-0469(1985)042<0549:FOHACD>2.0.CO;2.Fritts, D., and T. E. VanZandt, 1993: Spectral estimates of gravity wave energy and momentum fluxes. Part I: Energy dissipation, acceleration and constraints.

,*J. Atmos. Sci.***50**, 3685–3694, https://doi.org/10.1175/1520-0469(1993)050<3685:SEOGWE>2.0.CO;2.Fritts, D., and W. Lu, 1993: Spectral estimates of gravity wave energy and momentum fluxes. Part II: Parameterization of wave forcing and variability.

,*J. Atmos. Sci.***50**, 3695–3713, https://doi.org/10.1175/1520-0469(1993)050<3695:SEOGWE>2.0.CO;2.Garfinkel, C. I., and L. Oman, 2018: Effect of gravity waves from small islands in the Southern Ocean on the Southern Hemisphere atmospheric circulation.

,*J. Geophys. Res. Atmos.***123**, 1552–1561, https://doi.org/10.1002/2017JD027576.Grimshaw, R., 1975: Nonlinear internal gravity waves in a rotating fluid.

,*J. Fluid Mech.***71**, 497–512, https://doi.org/10.1017/S0022112075002704.Hardiman, S. C., D. G. Andrews, A. A. White, N. Butchart, and I. Edmond, 2010: Using different formulations of the transformed Eulerian mean equations and Eliassen–Palm diagnostics in general circulation models.

,*J. Atmos. Sci.***67**, 1983–1995, https://doi.org/10.1175/2010JAS3355.1.Hendricks, E., J. Doyle, S. Eckermann, Q. Jiang, and P. Reinecke, 2014: What is the source of the stratospheric gravity wave belt in austral winter?

,*J. Atmos. Sci.***71**, 1583–1592, https://doi.org/10.1175/JAS-D-13-0332.1.Hertzog, A., C. Souprayen, and A. Hauchecorne, 2002: Eikonal simulations for the formation and the maintenance of atmospheric gravity wave spectra.

,*J. Geophys. Res.***107**, 4145, https://doi.org/10.1029/2001JD000815.Hertzog, A., J. Alexander, and R. Plougonven, 2012: On the intermittency of gravity wave momentum flux in the stratosphere.

,*J. Atmos. Sci.***69**, 3433–3448, https://doi.org/10.1175/JAS-D-12-09.1.Hines, C. O., 1997a: Doppler-spread parametrization of gravity-wave momentum deposition in the middle atmosphere. Part 1: Basic formulation.

,*J. Atmos. Sol.-Terr. Phys.***59**, 371–386, https://doi.org/10.1016/S1364-6826(96)00079-X.Hines, C. O., 1997b: Doppler-spread parametrization of gravity-wave momentum deposition in the middle atmosphere. Part 2: Broad and quasi monochromatic spectra, and implementation.

,*J. Atmos. Sol.-Terr. Phys.***59**, 387–400, https://doi.org/10.1016/S1364-6826(96)00080-6.Jewtoukoff, V., A. Hertzog, R. Plougonven, A. de la Cámara, and F. Lott, 2015: Comparison of gravity waves in the Southern Hemisphere derived from balloon observations and the ECMWF analyses.

,*J. Atmos. Sci.***72**, 3449–3468, https://doi.org/10.1175/JAS-D-14-0324.1.Kalisch, S., P. Preusse, M. Ern, D. S. Eckermann, and M. Reise, 2014: Differences in gravity wave drag between realistic oblique and assumed vertical propagation.

,*J. Geophys. Res. Atmos.***119**, 10 081–10 099, https://doi.org/10.1002/2014JD021779.Kim, Y.-H., G. Bölöni, S. Borchert, H.-Y. Chun, and U. Achatz, 2021: Toward transient subgrid-scale gravity wave representation in atmospheric models. Part II: Wave intermittency simulated with convective sources.

,*J. Atmos. Sci.***78**, 1339–1357, https://doi.org/10.1175/JAS-D-20-0066.1.Kim, Y.-J., D. S. Eckermann, and H. Chun, 2003: An overview of past, present and future gravity-wave drag parametrization for numerical climate and weather prediction models.

,*Atmos.–Ocean***41**, 65–98, https://doi.org/10.3137/ao.410105.Kruse, G. C., and R. B. Smith, 2018: Nondissipative and dissipative momentum deposition by mountain wave events in sheared environments.

,*J. Atmos. Sci.***75**, 2721–2740, https://doi.org/10.1175/JAS-D-17-0350.1.Lindzen, R. S., 1981: Turbulence and stress owing to gravity wave and tidal breakdown.

,*J. Geophys. Res.***86**, 9707–9714, https://doi.org/10.1029/JC086iC10p09707.Lott, F., and M. Miller, 1997: A new subgrid-scale orographic drag parametrization: Its formulation and testing.

,*Quart. J. Roy. Meteor. Soc.***123**, 101–127, https://doi.org/10.1002/qj.49712353704.Lott, F., and L. Guez, 2013: A stochastic parameterization of the gravity waves due to convection and its impact on the equatorial stratosphere.

,*J. Geophys. Res. Atmos.***118**, 8897–8909, https://doi.org/10.1002/jgrd.50705.McLandress, C., T. G. Sheperd, S. Polavarapu, and S. R. Beagley, 2012: Is missing orographic gravity wave drag near 60°S the cause of the stratospheric zonal wind biases in chemistry–climate models?

,*J. Atmos. Sci.***69**, 802–818, https://doi.org/10.1175/JAS-D-11-0159.1.Medvedev, A. S., and G. P. Klaassen, 1995: Vertical evolution of gravity wave spectra and the parametrization of associated wave drag.

,*J. Geophys. Res.***100**, 25 841–25 854, https://doi.org/10.1029/95JD02533.Muraschko, J., M. Fruman, U. Achatz, S. Hickel, and Y. Toledo, 2015: On the application of Wentzel–Kramer–Brillouin theory for the simulation of the weakly nonlinear dynamics of gravity waves.

,*Quart. J. Roy. Meteor. Soc.***141**, 676–697, https://doi.org/10.1002/qj.2381.Olbers, D., C. Eden, E. Becker, F. Pollmann, and J. Jungclaus, 2019: The IDEMIX model: Parameterization of internal gravity waves for circulation models of ocean and atmosphere.

*Energy Transfers in Atmosphere and Ocean*, Springer, 87–125, https://doi.org/10.1007/978-3-030-05704-6_3.Orr, A., P. Bechtold, J. F. Scinocca, M. Ern, and M. Janiskova, 2010: Improved middle atmosphere climate and forecasts in the ECMWF model through a nonorographic gravity wave drag parameterization.

,*J. Climate***23**, 5905–5926, https://doi.org/10.1175/2010JCLI3490.1.Palmer, T. N., G. J. Shutts, and R. Swinbank, 1986: Alleviation of a systematic westerly bias in general circulation and numerical weather-prediction models through an orographic gravity wave drag parametrization.

,*Quart. J. Roy. Meteor. Soc.***112**, 1001–1039, https://doi.org/10.1002/qj.49711247406.Plougonven, R., A. de la Cámara, A. Hertzog, and F. Lott, 2020: How does knowledge of atmospheric gravity waves guide their parameterizations?

,*Quart. J. Roy. Meteor. Soc.***146**, 1529–1543, https://doi.org/10.1002/qj.3732.Ribstein, B., and U. Achatz, 2016: The interaction between gravity waves and solar tides in a linear tidal model with a 4D ray-tracing gravity-wave parameterization.

,*J. Geophys. Res. Space Phys.***121**, 8936–8950, https://doi.org/10.1002/2016JA022478.Ribstein, B., U. Achatz, and F. Senf, 2015: The interaction between gravity waves and solar tides: Results from 4D ray tracing coupled to a linear tidal model.

,*J. Geophys. Res. Space Phys.***120**, 6795–6817, https://doi.org/10.1002/2015JA021349.Richter, J. H., F. Sassi, and R. R. Garcia, 2010: Toward a physically based gravity wave source parameterization in a general circulation model.

,*J. Atmos. Sci.***67**, 136–156, https://doi.org/10.1175/2009JAS3112.1.Sato, K., S. Watanabe, Y. Kawatani, Y. Tomikawa, K. Miyazaki, and M. Takahashi, 2009: On the origin of mesospheric gravity waves.

,*Geophys. Res. Lett.***36**, L19801, https://doi.org/10.1029/2009GL039908.Scaife, A. A., J. R. Knight, G. K. Vallis, and C. K. Folland, 2005: A stratospheric influence on the winter NAO and North Atlantic surface climate.

,*Geophys. Res. Lett.***32**, L18715, https://doi.org/10.1029/2005GL023226.Scaife, A. A., and Coauthors, 2012: Climate change projections and stratosphere–troposphere interaction.

,*Climate Dyn.***38**, 2089–2097, https://doi.org/10.1007/s00382-011-1080-7.Scinocca, J. F., 2003: An accurate spectral nonorographic gravity wave drag parameterization for general circulation models.

,*J. Atmos. Sci.***60**, 667–682, https://doi.org/10.1175/1520-0469(2003)060<0667:AASNGW>2.0.CO;2.Senf, F., and U. Achatz, 2011: On the impact of middle-atmosphere thermal tides on the propagation and dissipation of gravity waves.

,*J. Geophys. Res.***116**, D24110, https://doi.org/10.1029/2011JD015794.Shapiro, R., 1975: Linear filtering.

,*Math. Comput.***29**, 1094–1097, https://doi.org/10.1090/S0025-5718-1975-0389356-X.Skamarock, W. C., and Coauthors, 2019: A description of the Advanced Research WRF version 4. National Center for Atmospheric Research Tech. Note NCAR/TN-556+STR, 145 pp., https://doi.org/10.5065/1dfh-6p97.

Smith, A. K., 2012: Global dynamics of the MLT.

,*Surv. Geophys.***33**, 1177–1230, https://doi.org/10.1007/s10712-012-9196-9.Song, I.-S., and H.-Y. Chun, 2005: Momentum flux spectrum of convectively forced internal gravity waves and its application to gravity wave drag parameterization. Part I: Theory.

,*J. Atmos. Sci.***62**, 107–124, https://doi.org/10.1175/JAS-3363.1.Song, I.-S., and H.-Y. Chun, 2008: A Lagrangian spectral parameterization of gravity wave drag induced by cumulus convection.

,*J. Atmos. Sci.***65**, 1204–1224, https://doi.org/10.1175/2007JAS2369.1.Swinbank, R., and D. A. Ortland, 2003: Compilation of wind data for the Upper Atmosphere Research Satellite (UARS) Reference Atmosphere Project.

,*J. Geophys. Res.***108**, 4615, https://doi.org/10.1029/2002JD003135.VanZandt, T. E., 1982: A universal spectrum of buoyancy waves in the atmosphere.

,*Geophys. Res. Lett.***9**, 575–578, https://doi.org/10.1029/GL009i005p00575.Warner, C. D., and M. E. McIntyre, 1996: On the propagation and dissipation of gravity wave spectra through a realistic middle atmosphere.

,*J. Atmos. Sci.***53**, 3213–3235, https://doi.org/10.1175/1520-0469(1996)053<3213:OTPADO>2.0.CO;2.Wei, J., G. Bölöni, and U. Achatz, 2019: Efficient modelling of the interaction of mesoscale gravity waves with unbalanced large-scale flows: Pseudomomentum-flux convergence versus direct approach.

,*J. Atmos. Sci.***76**, 2715–2738, https://doi.org/10.1175/JAS-D-18-0337.1.Wilhelm, J., T. Akylas, G. Bölöni, J. Wei, B. Ribstein, R. Klein, and U. Achatz, 2018: Interactions between mesoscale and submesoscale gravity waves and their efficient representation in mesoscale-resolving models.

,*J. Atmos. Sci.***75**, 2257–2280, https://doi.org/10.1175/JAS-D-17-0289.1.Zängl, G., D. Reinert, P. Ripodas, and M. Baldauf, 2015: The ICON (Icosahedral Non-hydrostatic) modelling framework of DWD and MPI-M: Description of the non-hydrostatic dynamical core.

,*Quart. J. Roy. Meteor. Soc.***141**, 563–579, https://doi.org/10.1002/qj.2378.