Medvedev 2000

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 19

Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033

www.elsevier.nl/locate/jastp

Parameterization of gravity wave momentum deposition based


on nonlinear wave interactions: basic formulation and
sensitivity tests
A.S. Medvedev, G.P. Klaassen ∗
Department of Earth and Atmospheric Science, York University, 4700 Keele St. Toronto, Ont., Canada M3J 1P3
Received 5 November 1999; accepted 27 April 2000

Abstract
We present a practical method for parameterization of gravity wave drag based on the Medvedev and Klaassen (1995.
Journal of Geophysical Research 100, 25,841–25,853) theory of gravity wave spectral evolution and saturation. The only
tuning necessary for the scheme involves assumptions about the nature of the source spectrum of subgrid-scale gravity waves,
i.e. the wavenumbers, launch heights and amplitudes of the spectral components. In this paper we employ a column model with
representative distributions of mean wind and temperature to examine the sensitivity of the parameterized wave drag to the
source spectra. For the range of anticipated variability of source spectra in the troposphere the scheme produces plausible results
consistent with observations and with theoretical estimates. Computationally, the scheme is as ecient as a multiple-wave
Lindzen scheme, and suitable for use in general circulation models (GCMs) of the atmosphere. c 2000 Elsevier Science Ltd.
All rights reserved.

Keywords: Gravity waves; Parameterization; Middle atmosphere modelling

1. Introduction is directly associated with wave damping. The spectra in the


tail can be approximated by S(m) ≈ AN 2 =m3 , where N is
Gravity wave (GW) drag deposited by waves propagat- the Brunt–Vaisala frequency, m is the vertical wavenumber,
ing from below substantially a ects the circulation of the and A is the constant. It is this “saturation” of wave ampli-
middle atmosphere. Parameterization of these subgrid-scale tudes with height, despite the diminishing gas density, which
processes in general circulation models requires the spec- implies the existence of some physical damping process.
i cation of gravity wave sources as well as an absorption Various spectral theories of GW drag have been devel-
mechanism which causes damping of waves and deposition oped for use in atmospheric models. One approach is to
of their momentum to the mean ow. assume waves propagate independently, with each compo-
A manifestation of systematic wave damping processes in nent of the spectrum attaining saturation after exceeding
the middle atmosphere is the apparent “universality” of ob- the convective=dynamic instability threshold proposed by
served gravity wave spectra (Smith et al., 1987). In this re- Lindzen (1981, hereafter L81). Although this approach has
spect, perhaps the most signi cant feature of universal spec- been employed with some success in a number of numerical
tra is the power-law behavior of the high vertical wavenum- models (Holton, 1982; Garcia and Solomon, 1985), it tends
ber “tails” of their power-spectral densities (PSD), S, which to generate step-function GW drag pro les where the mo-
mentum deposition onsets suddenly at the breaking level.
∗ Corresponding author. Tel.: 1-416-736-2100; fax: 1-416- More recently, Hamilton (1997) and Norton and Thuburn
(1997) found that such multiple-wave Lindzen-type GWD
736-5817.
E-mail addresses: medvedev@nimbus.yorku.ca (A.S. parameterizations tend to produce drag in GCMs at alti-
Medvedev), gklaass@yorku.ca (G.P. Klaassen). tudes that are too high, if moderate launch amplitudes are

1364-6826/00/$ - see front matter c 2000 Elsevier Science Ltd. All rights reserved.
PII: S 1 3 6 4 - 6 8 2 6 ( 0 0 ) 0 0 0 6 7 - 5
1016 A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033

assumed. Stronger launch amplitudes of course induce mo- ory and produces the correct power law at high vertical
mentum deposition at lower altitudes, but in this case the wavenumber, Doppler spreading is a fundamentally conser-
drag is too strong. Such linear schemes therefore must em- vative process in which waves are not damped but rather
ploy arti cial “eciency factors” which reduce the drag, have their intrinsic frequency (and therefore their vertical
under the assumption that the wave activity is intermittent. wavenumber) transformed. Wave breaking must therefore
Alexander and Dunkerton (1999) have recently proposed a be imposed by chopping the spectrum a maximum “cuto ”
broadband spectral parameterization based on the saturation vertical wavenumber. Waves shifted beyond this prescribed
mechanism proposed by Lindzen; this scheme also invokes high-wavenumber cuto are assumed to be obliterated, and
an eciency parameter based on intermittent wave forcing. their momentum transferred the mean ow. The H97 scheme
Fritts and Lu (1993) formulated a semi-empirical spec- has been tested in various middle atmosphere models by
tral parameterization which circumvents some of the prob- Mengel et al. (1995), Manzini et al. (1997) and McFarlane
lems associated with the Lindzen scheme. The Fritts and et al. (1997).
Lu scheme calculates the momentum deposition by ap- In Medvedev and Klaassen (1995, hereafter referred to
plying saturation criteria which conform to the average as MK95) we employed a modi cation of Weinstock’s
observed evolution of GW spectra with height. One ex- (1982,1990,1993) assumptions and technique, including
pects this approach to successfully reproduce wave drag the incorporation of nonlinear Doppler e ects, to develop a
pro les when GW spectra are close to average conditions. somewhat di erent representation of the vertical evolution
However, observed instantaneous spectra usually deviate of GW spectra in the atmosphere. Although we make use of
signi cantly from the averaged “universal” shape, and the nonlinear di usive spectral interactions in the spirit of We-
amplitudes of averaged spectra have been shown to vary instock (1990), our approach considers only the damping
under di erent ow conditions (Whiteway, 1999). Thus, it that spectral components with smaller vertical scale (higher
remains to be established whether the Fritts and Lu scheme m) and longer period (lower !) exert on a given wave com-
can reasonably represent the complex feedback processes ponent. This portion of the spectrum represents a “ragged”
between wave drag and the mean ow. For example, Hecht background which contains slowly varying, small-scale
et al. (1998) have compared observed tides with those in vertical variations. As a wave possessing similar or larger
the TIMEGCM including the Fritts and Lu GWD scheme. scale vertical variations propagates through this “ragged”
McLandress (1997) has also tested the Fritts and Lu pa- background, nonlinear interactions with the spectrum cause
rameterization in a linearized primitive equation model and increased damping by forcing the given wave closer to an
reported that this particular scheme strongly damps tides in overturned state. If the damping associated with these in-
the lower thermosphere. Warner and McIntyre (1999) have teractions is moderate, the wave may achieve saturation. If,
recently proposed a related gravity wave drag parameteri- on the other hand, the interactions are suciently strong (as
zation which couples a linear conservative model of gravity in a critical level approach) the wave may be obliterated
wave propagation with saturation criteria similar to those (often before reaching the critical level). MK95 veri ed
invoked by Fritts and Lu. with numerical tests that their nonlinear wave damping
Other theories attribute saturation to nonlinear interac- formula also tends to produce saturated spectra with m−3
tions between components of the broad spectrum of gravity dependence at high values of m.
waves. In Weinstock’s (1990) approach, wave components In MK95, the RMS background wind associated with the
with larger m and higher frequency ! are considered as a higher-m=lower-! portion of the spectrum also produces
source of enhanced scale-dependent di usion which can af- Doppler shifting toward higher m. However, it should be
fect the vertical evolution of waves with smaller m and lower recognized that the MK95 theory accounts for nonlinear
!. Weinstock (1990) was able to show that under saturation interactions in a fundamentally di erent way than Hines’
conditions and in the absence of wind shear, his mechanism Doppler-spread theory. In fact, we have found that the
of wave damping produced a power-law spectrum with m−3 Doppler shifting associated with high-m=low-! portion
dependence at high wavenumber [but note the typographical of the spectrum is small enough to be neglected for the
error in his Eq. (18)]. practical purpose of wave drag parameterization. Calcu-
The Doppler-spread parameterization (Hines, 1997, here- lations justifying the neglect this e ect will be presented
after H97) treats the irregular uctuations of the wave eld in the current paper. This nding is partially at odds with
as an additional “RMS” background wind which preferen- Hines’ Doppler spreading. Since Hines assumes all vertical
tially shifts wave components from the lower m part of the wavenumbers contribute to the spectrum-induced Doppler
spectrum to the high-m portion. Hines himself has noted shifting of any speci c wave component his treatment must
that the use of Doppler-spreading to represent all nonlin- necessarily produce stronger nonlinear shifts than those
ear interactions is an approximation (Hines, 1996,1999). obtained with the restricted spectrum used in MK95. How-
One of the chief shortcomings of the Doppler Spread the- ever, as pointed out by MK95, those wave components
ory is that it does not produce a power-law spectrum at having higher frequency than the wave to be shifted bear
high wavenumbers (Hines, 1993). Whereas wave dissipa- little similarity to a steady background wind and should not
tion emerges as a natural consequence of Weinstock’s the- be treated as such. Recently published studies by Ecker-
A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033 1017

mann (1997) and Broutman et al. (1997) have con rmed of column-model tests based on representative distributions
that Hines’ assumption which “freezes” the entire RMS of mean elds from CIRA-86. In Section 7, we compare
wave background may to some extent overestimate wave our parameterization to a multiple Lindzen-type-scheme.
damping. Evidently, spectrum-induced critical levels, un- A novel spectral decomposition of GW momentum deposi-
like those associated with a truly steady background wind, tion and RMS wind is presented in Section 8. Summary and
are imperfect wave absorbers. It should be noted that the conclusions are given in Section 9.
parameterization scheme described in this paper is capable
of reproducing realistic middle atmospheric circulations in
the absence of spectrum-induced Doppler shifts (Medvedev 2. Outline of MK95 scheme
et al., 1998).
MK95 demonstrated that in the absence of background According to MK95, the evolution of gravity wave spectra
wind shear, wave damping due to nonlinear interactions with height z is given by the following equation:
tends to be a gradual process which forms a saturated “uni-  
dS(mR ; z) 0z mRz
versal” m−3 tail in agreement with observations. However, = − + − ÿ S(mR ; z); (1)
dz 0 mR
the MK95 theory also includes the possibility of violent
breaking and rapid elimination of a wave component as it where S is the power-spectral density (PSD) of horizon-
approaches the convective instability threshold or a critical tal wind associated with gravity waves at altitude z, 0
level. In this paper we will demonstrate that for the case of a is the mean density, mR is the real part of the nonlinear
saturating monochromatic wave, the MK95 wave damping Doppler-shifted vertical wavenumber, and ÿ is the coe-
formula reduces to a form similar to, but more general than, cient of nonlinear damping due to interactions of the com-
the simple saturation formula proposed by Lindzen (1981). ponent mR with other waves in the spectrum. The subscript
Even in this monochromatic case, MK95 includes subover- z in Eq. (1) indicates a vertical derivative. For eciency of
turning and nonlinear self-interactions not found in L81. notation, we will suppress the z dependence in all dependent
Consequently, saturation and wave drag predicted by the variables, writing for example S(mR ).
MK95 theory generally onset more smoothly and at lower The vertical wavenumber mR includes an additional non-
altitudes than for L81. This characteristic of MK95 carries linear Doppler shift produced by waves of higher vertical
over to broad spectra, where again wave damping due to wavenumber and is related to the “linear” vertical wavenum-
nonlinear interactions tends to be a gradual process which ber m by the expression
leads to the formation of a saturated “universal” m−3 tail. √
mR = m  exp(− 2 ) er ( ): (2)
In this paper we outline a practical gravity wave drag pa-
rameterization based on the MK95 theory, which is suitable In Eq. (2) the following notation for the error function of
for use in large-scale circulation models. Parameterizations an imaginary argument is used:
usually imply further simpli cations of the theories they are Z
2
based on as well as an introduction of some tunable param- er ( ) = √ exp(x2 ) d x;
 0
eters. The scheme we present retains all physically impor-
tant features of the theory outlined in MK95, and introduces and the dimensionless parameter is de ned by
no new tunable parameters. Although the scheme has been Z ∞
N c−u
shown to produce realistic middle atmosphere circulations = √ = √ ; 2 = S(m0 ) dm0 ; (3)
2m 2 mR
in both a mechanistic version of the NCAR Middle Atmo-
sphere CCM2 (Medvedev et al., 1997), and the Canadian where N is the buoyancy frequency, and 2 is the horizon-
Middle Atmosphere Model (Medvedev et al., 1998), neither tal wind variance created by all waves in the spectrum with
an implementation guide nor a complete examination of the vertical wavenumbers larger than the given mR . As will be
various tuning options has been presented to date. These demonstrated in Section 4, is proportional to the square
systematic sensitivity tests and comparisons to other GWD root of an e ective “spectral” Richardson number Ri for the
schemes will provide a better understanding of the mecha- component with wavenumber m. Small values of  corre-
nisms and behaviour of the MK95 scheme, and assist those spond to large Richardson number. Hereafter, we will refer
interested in implementing it in other middle atmosphere to as the “instability parameter”.
models. Eq. (2) also makes use of the hydrostatic nonrotating
In Section 2, we present a concise formulation of the dispersion relation
scheme. A numerical implementation is given in Section 3.
m = N=(c − u); (4)
In Section 4, we discuss the physical properties of the
scheme, its relation to other well-known GWD parameteri- for “linear” waves with intrinsic frequencies !ˆ which
zations, namely those of Lindzen (L81) and Hines (H97), satisfy f ! N
ˆ . Eq. (4), which is also known as the mid-
and approximations that are recommended in the interest of frequency dispersion relation, speci es the variation of the
eciency. The range of variability of input source spectra is apparent “linear” vertical wavenumber m for a wave with
discussed in Section 5. In Section 6, we present the results a given constant observed (extrinsic) phase velocity (and
1018 A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033

varying intrinsic phase velocity) as it propagates through re ections, a process in which the intrinsic frequency ex-
vertically inhomogeneous mean wind and background ceeds N . Clearly, wave groups which are re ected down-
temperature elds. ward are unlikely to contribute to momentum deposition in
The MK95 theory yields an expression for the nonlinear the middle atmosphere, so they may be excluded from the
damping coecient ÿ that may be conveniently approxi- launch spectrum in the interest of computational eciency.
mated as Note that our choice of a xed characteristic value for kh
√ carries with it the implication that waves of higher vertical
ÿ = 2N−1 exp(− 2 ): (5) wavenumber have lower intrinsic frequencies. This is dic-
The detailed behaviour of the vertical damping rate ÿ will be tated by the hyperbolic form of the relation between m and
considered in Section 4. Brie y, ÿ decreases with increas- !; i.e. for the mid-frequency dispersion relation, the intrin-
ing Richardson number Ri (increasing ), becoming negli- sic frequency !ˆ = Nkh =m.
gible for waves of suciently small amplitude. On the other It should also be noted that in Eqs. (1) – (6) we have
hand, as the amplitude of a particular component and that of adopted a convenient (but nonstandard) convention, in
the higher-wavenumber portion of the spectrum grow with which the sign of m does not re ect the sign of the vertical
√ phase speed, but rather the sign of the intrinsic horizontal
height, the instability parameter → 1= 2 (Ri → 1) from
above and the damping rate increases dramatically. phase speed c − u.  This convention is permitted by the
The set of equations (1) – (5) provides a closed system midfrequency dispersion relation, i.e. it follows from Eq.
for calculating the evolution of the PSD S with height. Once (4) that “extrawaves” with c − u ¿ 0 have m ¿ 0, while
the power-spectral density S(mR ) and the coecient of non- “intrawaves” with c − u ¡ 0 are characterized by m ¡ 0. To
linear damping ÿ(mR ) are known for a particular wave, the ensure that the vertical group velocity is directed upward,
spectral density of the momentum deposition associated with i.e. cgz = −Nkh =m2R ¿ 0 (the only case which produces
this wave can be found from momentum deposition in the middle atmosphere), we also
adopt the convention that kh ¡ 0 throughout this paper. It
ÿ(mR )kh S(mR ) is seen from Eq. (6) that the sign of the drag then coincides
ah (mR ) = − ; (6)
mR with the sign of (c − u).
 This particular sign convention is
where kh = |kh | is a characteristic horizontal wavenumber. 1 convenient for coding purposes, but of course Eqs. (1) – (6)
The total wave drag is calculated by integrating the contri- can be readily adapted to other sign conventions.
butions from all components of the spectrum. As we shall MK95 have shown that gravity wave saturation in the at-
demonstrate, our scheme permits the simultaneous use of mosphere typically occurs in the range ≈ 1 to 3. Together
reasonable launch amplitudes and horizontal wavenumbers, with Eq. (2) this implies mR ≈ (1:1 to 1:3)m. Therefore, for
and does not require kh to be multiplied by an “eciency fac- purposes of wave drag parameterization, the systematic non-
tor”. In order to produce the necessary drag, Lindzen-style linear Doppler shift created by spectral components with
schemes require eciency factors that are small compared higher vertical wavenumber can be neglected, and mR can
to unity (Hamilton, 1997). be replaced by the “linear” vertical wavenumber m every-
Note that the vertical evolution of wave spectra is entirely where in Eqs. (1) – (6). This matter will be discussed further
described by Eqs. (1) – (5) in terms of the vertical wavenum- in Section 4. Numerical calculations which quantify the con-
ber mR alone (a simpli cation allowed by the midfrequency sequences incurred by neglecting nonlinear Doppler shifts
dispersion relation), while the intensity of wave drag ah (mR ) in the calculation of wave drag are given in Section 6.
is proportional to kh . Although the MK95 theory is su-
ciently general to permit di ering values of kh for individual
3. Numerical implementation
wave components, we have found that the constant kh case is
capable of producing very realistic middle atmosphere circu-
For numerical purposes, we convert the continuous input
lations. For the present purposes, it is therefore sucient to
source PSD S to a discrete spectrum consisting of M har-
consider the simplest case where kh enters only as a scaling
monics 2 according to
factor for the momentum deposition. This assumption re-
stricts the number of tunable parameters associated with the hUj2 i = Sj mj ;
initial spectrum, and, with a suitable choice of kh , allows one
where mj = mj+1 − mj de nes the discretization of the
to exclude from the launch spectrum high-frequency waves
spectrum, and j is the index of a given harmonic. A suitable
that might be re ected downward. A nonhydrostatic disper-
choice for mj will be given in Section 6. Once the initial
sion relation such as that used by Alexander the Dunkerton
mj has been assigned for the harmonic, it may be uniquely
(1999) would have to be invoked to properly predict wave
identi ed by its observed horizontal phase speed c, which

1 The horizontal wavevector k =k x̂+k ŷ is used in recognition 2 We will use the word “harmonic” to denote a discrete compo-
h x y
of the possibility that waves may not necessarily be aligned in the nent with a particular observed vertical phase speed, which is not
x̂ (i.e. zonal) direction. necessarily constrained to integer multiples of a fundamental mode.
A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033 1019

is conserved during vertical propagation under assumptions spectra with height and the corresponding wave momen-
of horizontally homogeneous steady ow. The horizontal tum deposition. The procedure starts from the lowest level
velocity uctuations of the jth harmonic are where the source spectral distribution hUj2 (zs )i is speci ed.
At each vertical step, the component GW velocity variances
Uj = Re{Û j ei }; (7)
j2 , parameters j and nonlinear damping rates ÿj are calcu-
where (x; t) is the wave phase, the latter being a func- lated using Eqs. (8) and (5) for every harmonic, beginning
tion of time t and the usual Cartesian coordinate vector with the highest mj , i.e. with j = M . Then the vertical
x = (x; y; z). In accordance with the convention adopted in wavenumbers mj , averaged squared wave amplitudes hUj2 i
MK95, the angle brackets introduced above R 2denote twice and wave drag Ah can be found at the next vertical level zi
the average over , namely hf()i ≡ 1= 0 f d. Thus from Eqs. (11), (10) and (12), respectively, MK95 showed

hUj2 i = Û j Û j = |Û j |2 , the squared amplitude of the jth har- that the instability threshold for a wave harmonic in the
3
monic. (The asterisk denotes a complex conjugate.) Then, presence of a broad spectrum is j2 6 1=2. Harmonics
the total horizontal velocity variance 2 and the parameter meeting this criterion, which is equivalent to cj − u = j ,
in Eq. (3) are calculated as follows: break and deposit their momentum. This e ectively re-
moves harmonics slightly before they reach the “linear”
X
M
N
j2 = hUn2 i; j = √ ; (8) critical levels cj − u = 0 of Lindzen-type schemes. Further
n=j 2mj j comparisons with parameterizations based on convectively
unstable saturation are given in Section 7.
where j=1 and M are harmonics with the lowest and highest
vertical wavenumbers in the spectrum, respectively.
The solution of the discretized counterpart of Eq. (1) has
the form (cf. Eq. (27) in MK95) 4. Physical properties
 Z z 
2 2 0s mj 0 0 In this section, we discuss the physical properties of the
hUj (z)i = hUj (zs )i exp − ÿj (z ) d z ; (9)
0 mjs zs MK95 theory for nonlinear interactions in broad gravity
where subscript s denotes the values at the source level zs . wave spectra, as well as approximations we have applied in
If the vertical coordinate is discretized, and if the source the construction of a practical GWD parameterization. We
spectrum hUj2 i is known at some height, Eq. (9) can be also discuss the relationship of MK95 to the Lindzen (L81)
integrated upward step-by-step as follows: and Hines (H97) schemes.
Consider a test wave of vertical wavenumber mj propa-
0 (zi−1)mj (zi) gating in concert with a broad spectrum of waves. In the
hUj2 (zi)i=hUj2 (zi−1)i exp[−ÿj (zi−1=2)zi ];
0 (zi)mj (zi−1) MK95 theory, this test wave interacts with waves of similar
(10) and smaller vertical scale, that are also evolving on similar
where zi is the grid interval between the higher level zi and slower time scales. 4 As a matter of convenience, our
and lower level zi−1 , and zi−1=2 denotes the half-level be- parameterization is formulated for waves of xed horizon-
tween zi−1 and zi . In Eq. (10) the apparent “linear” verti- tal wavenumber kh , so that the relevant portion of the back-
cal wavenumber mj of the harmonic with a given kh and ground is composed of waves with m ¿ mj , i.e. a “ragged
observed phase velocity c varies with height according to background” of similar and smaller scale waves possessing
Eq. (4), following variations of the background wind u and similar and lower frequencies. The net e ects of nonlinear
Brunt–Vaisala frequency N , interactions with the “ragged background” are damping of
the test wave and the shifting of its frequency. We will pro-
N (zi )
mj (zi ) = : (11) vide arguments that the nonlinear (i.e. spectrum-induced)
c − u(zi ) frequency shifts may be neglected for the practical purpose
The “nonlinear” spectrum-induced Doppler shift of the ver- of calculating wave drag.
tical wavenumber can be calculated from Eq. (2), although In Lindzen-type schemes, individual harmonics begin to
in practice it may be neglected. decay when they overturn, i.e. when they induce supera-
Integration and discretization of Eq. (6) yields an expres- diabatic temperature lapse rates. The criterion for over-
sion for total wave drag: turning instability of a monochromatic wave of amplitude
Z ∞ Û in a constant background ow u may be expressed as
XM
ÿ j kh 2
Ah = ah (m0 ) dm0 = − hUj i: (12) |Û | ¿ |c − u|
 (Fritts, 1984) or equivalently, in terms of the
0 j=1
mj

Together, Eqs. (8), (10), (11), (5) and (12) represent a 4 It is unclear how to treat interactions with more rapidly os-
closed set of equations for calculating the evolution of
cillating harmonics, but presumably these are transient and per-
haps less signi cant than interactions with slowly varying waves.
3 Note that in other conventions the variance is sometimes de- The latter may be e ectively considered as an enhancement of the
ned as hUj2 i = 12 |Û j |2 . steady background ow.
1020 A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033

Fig. 1. (a) Conceptual illustration of a test wave (solid curve) with phase speed c = 16 m s−1 propagating through a “ragged” small-scale
wave background that is slowly varying in time (dotted curve). (b) The RMS wind for the test wave alone (solid curve) and the test wave
plus small-scale background (dotted curve). Both the instantaneous (oscillatory) wave elds and the corresponding  pro les (exponential
“envelopes”) are shown. For simplicity, the mean wind u = 0. According to L81, the test wave would independently achieve convective
instability and saturate at z = 70 km, where the wave amplitude matches its phase speed (vertical dash– dot line). According to the MK95
theory, the wave would be obliterated by strong instabilities if the total  (test wave plus background) → c, as it does near z = 63 km.
Nonlinear damping in the MK95 theory is usually strong enough to cause saturation of the test wave in the range c=4 6  6 c=2, indicated
by the thick dashed bar between z ∼ 45 and 55 km.

monochromatic Froude number small-scale shear deforms parcels and parcel paths, lead-
ing to gradual dissipation of the coherent wave motion. It
|mÛ | Û
F1 ≡ ≡ ; (13) should also be recognized that when m exceeds N , one
N (c − u)
expects patches of convective instability to form where the
as F1 ¿ 1. Owing to the polarization relations for a small-scale buoyancy uctuations locally overcome the sta-
monochromatic gravity wave, the threshold m = N corre- ble background strati cation and overturn the uid.
sponds to vertical gradients of the wave temperature eld In the basic L81 scheme, the damping rate is imposed ac-
that are suciently strong to locally cancel the stable back- cording to the assumption that the wave amplitude saturates,
ground strati cation. Recall that the threshold amplitudes thereby o setting the e ects of background strati cation and
for convective and dynamic Kelvin–Helmholtz instabilities mean wind shear. In contrast, the MK95 theory provides an
of internal gravity waves are virtually identical if the e ects explicit formula for the damping rate ÿ which is a function
of rotation can be neglected (Fritts, 1984). of the wave eld itself (the spectral variance 2 ) as well as
In the MK95 theory, the corresponding criterion is  → background ow and strati cation parameters. Furthermore,
 where the spectral variance 2 de ned in Eqs. (3)
|c − u|, damping due to nonlinear wave–wave interactions is found
and (8) includes the test wave amplitude Û j itself (as the to be non-zero for suboverturning wave amplitudes and in-
rst component). As illustrated conceptually in Fig. 1, this creases as  → |c − u|, i.e. as the spectral Froude number
superposition of the test wave and the “ragged” smaller-scale Fr → 1, or equivalently as the instability parameter
background leads to locally enhanced wave amplitudes and c−u 1 1
wave shear. This in turn drives the spectral Froude number = √ =√ → √ :
2 2Fr 2
|m|  The approximate vertical damping rate ÿ for the kinetic en-
Fr ≡ ≡ (14)
N |c − u| ergy of a test wave of vertical wavenumber m is given in
toward unity (i.e. toward “overturning”), and leads to in- the MK95 parameterization by Eq. (5). Inverting Eq. (3) to
creased damping of the test wave. Since m is a statisti- obtain m as a function of allows us to rewrite Eq. (5) as
cal measure of the velocity shear associated with the high √
ÿ=m = 2  exp(− 2 ); (15)
wavenumber portion of the spectrum, one would expect on
physical grounds that increasing m would enhance instabil- where the quantity ÿ=m corresponds to the vertical damping
ity and hence momentum deposition. In e ect, this complex rate normalizd by the vertical wavenumber. Fig. 2 shows
A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033 1021

damping rate, namely


 
ÿ 1 √
= √ = 2e−1=2 ≈ 1:51;
m 2
is very strong. In the absence of Doppler shifting or density
changes, the PSD will decay as S(z0 + z) ∼ S(z0 )e−ÿz .
Over a distance of one vertical wavelength  = 2=m, a PSD
subjected to ÿ=m = 1:51 will decay to
S(z0 + ) = S(z0 )e−2ÿ=m ≈ S(z0 )e−9:48 ≈ S(z0 )=13 100:
This√corresponds to the wave amplitude dropping by a factor
of 13 100 ≈ 115 over one vertical wavelength. Clearly,
the di erence between the exact and approximate formulae

may be e ectively minimized by testing for 6 1= 2 and,
if satis ed, depositing the wave’s momentum to the mean
ow. This is in fact what is done in our implementation of
the scheme. We note that this e ectively removes waves
from the spectrum somewhat before the altitude at which
Fig. 2. Comparison of the exact (ÿexact =m: solid curve) and approx-
imate (ÿ=m: dashed curve) vertical damping rates for the MK95 a linear critical level corresponding to = 0 would occur.
theory of nonlinear gravity-wave interactions. The vertical dotted By using the approximate formula (5), and depositing √ the
line marks the value = 2−1=2 . momentum associated with harmonics having 6 1 2, we
may avoid the evaluation of the computationally expensive
error function in Eq. (16), while retaining good agreement
√ with the behaviour of the exact MK95 theory.
that ÿ=m achieves a maximum value for = 1= 2, which In principle, the approximate theory permits → 0, which
corresponds to the overtuning value for a single harmonic. A corresponds to a linear critical layer approach for the wave,
standard Lindzen scheme, on the other hand, imposes a nite where c − u → 0 and m → ∞. It may appear from Fig.
background-related
√ damping once the harmonic achieves 2 that the damping rate ÿ for the approximate theory van-
6 1= 2. ishes in such a classic critical layer approach. However, one
It is instructive to compare the approximate formula (15) must remember that for a critical layer approach, the verti-
with the exact MK95 theory, which gives cal wavenumber of the test wave m → ∞. The behaviour
of ÿ in the limit → 0 can be elucidated by introducing the
er ( ) characteristic wavenumber
ÿexact = ÿ : (16)
[er 2 ( ) − 1]1=2
mc = N=: (18)
It can be seen from Fig. 2 that the approximate ÿ=m rapidly The generalized spectral Richardson number (neglecting
approaches the exact value for signi cantly larger than contributions from the wave buoyancy) may then be written
unity.
√ The principal di erence is that ÿexact =m → ∞ at = as
1= 2, while√the approximate ÿ=m remains nite. Thus, the
value = 1= 2 corresponds in the exact theory to a violent Ri ≡ N 2 =2 m2 = (mc =m)2 = 2 2 : (19)
instability which immediately dissipates the wave. We note Note that Ri is equivalent to the squared inverse of the spec-
that the exact theory also includes a nonlinear Doppler shift, tral Froude number Fr introduced earlier in Eq. (14). From
given by Eq. (19) it is clear that mc is the wavenumber for√which the
spectral Richardson number Ri = 1 (and = 1= 2), corre-
√ 2
mexact
R = m  e− [er 2 ( ) − 1]1=2 ; (17) sponding to waves which undergo strong nonlinear interac-
tions within the spectrum. For Ri ¡ 1, buoyancy uctuations

which yields mR → 0 as → 1= 2. Since the damping rate in the spectrum are suciently strong to cause local regions
ÿexact =m → ∞ at the same time, waves are subjected to large of overturned uid. (Note that the critical Ri for convective
dissipation rates and deposit the major portion of their mo- instability would be zero if gradients of wave temperature
mentum before their vertical wavelength becomes in nite. uctuations were included in its de nition.)
At larger values of , where waves may persist due to the With these de nitions we may cast Eq. (15) into the form
smaller damping rates, the corresponding nonlinear Doppler √
ÿ=mc = 2 exp(− 12 Ri): (20)
shifts are small enough to produce negligible contributions
to wave drag. This will be demonstrated later in Section 6. If the test wavenumber m mc , then Ri1 and the wave
Although the approximate
√ formula gives a nite vertical will not experience signi cant damping. On the other hand,
damping rate at = 1= 2, it should be recognized that this for a wave which is Doppler shifted so that m → mc , the
1022 A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033

to immediate obliteration. The ratio in Eq. (22) may be re-


garded as the minimum Richardson number in the spectrum
(i.e. the Richardson number associated with the maximum
vertical wavenumber). For stratospheric cases, a value of
1=2 = 2:4 yields a local spectral slope of approximately
−2:7 at vertical wavenumbers near mmax , while a value of
11.5 yields a local spectral slope of −2:3. Mesospheric cases
tend to yield shallower slopes (Hines, 1993). H97 therefore
recommends a value of 1=2 lying in the range
mDSP
c 1
= = 2:5 to 10;
mmax 2
with a particular value of 1=2 ≈ 3 found to be suitable
in the modelling study of Mengel et al. (1995). Owing to
the di erent de nition of the spectral amplitude DSP , it is
not possible to place a speci c line representing the cuto
wavenumber for Hines DSP in Fig. 3. It is possible to say
Fig. 3. Vertical damping rate ÿ normalized by the “characteris- that the smaller value for mDSP
c in Hines DSP would tend
tic” wavenumber mc = N=. Note that the horizontal coordinate to move the DSP cuto line toward the left. Thus, Hines
mc =m = Ri1=2 = 21=2 . preferred value of 2 ≈ 0:3 would fall somewhat to the
left of the line mc =m = 0:3 in Fig. 3. The DSP damping
rate is in nite to the left of that line, while waves on the
Richardson number Ri → 1, and nonlinear interactions will right propagate conservatively. By way of comparison, the
lead to strong damping of this particular harmonic. During Lindzen parameterization speci es a nite damping rate ÿ =
a critical level approach m → ∞, Ri → 0 and → 0. As 1=H (for a windless √case) once the wave has overturned (i.e.
shown in Fig. 3, ÿ=mc → 2:51 in this limit. From our previ- at mc =m = 1 or = 1 2), and ÿ = ∞ at a linear critical level
ous discussion it is clear that this damping rate is extremely (where mc =m = 0).
large, and in fact causes wave amplitude to decay by a fac- Further insight into the MK95 theory may be gained
tor of about 2400 over a vertical distance of 2=mc . Thus, by assuming that the wave spectrum consists of a general
for practical purposes, harmonics √ may be considered to be Desaubies spectrum of the form
immediately damped once 6 1= 2. (m=m∗ )s N2
The particular form of ÿ in Eq. (20) also serves to clar- S(m) = S0 ; S0 = A ; (23)
1 + (m=m∗ )s+t m3∗
ify the relationship with the Doppler spread parameteriza-
tion (DSP) of Hines (H97). DSP stipulates the form of the where m∗ is the characteristic wavenumber which describes
low-wavenumber portion of the spectrum below the “knee” the transition between spectral forms at large and small m
(i.e. m ¡ m∗ ) to be S(m) ˙ ms , with the recommended (the “knee”), and the parameter t de nes the power law for
form having s = 1. At each altitude, a high wavenumber the high-wavenumber tail (i.e. for m ¿ m∗ the PSD asymp-
“cuto ” is computed and that portion of the spectrum with totes to m−t behaviour). Observations of saturated spectra
m ¿ mmax = 2 (N=DSP ) where dictate the tail value t = 3 and, for the purposes of this sec-
Z mmax tion, we assume the modi ed Desaubies form with s = 1.
2
DSP = S(m0 ) dm0 (21) The constant A is determined from observational data to be
0 A ≈ 1=6 by Smith et al. (1987) and A ≈ 1=8 by Fritts and
is immediately obliterated and its momentum deposited. The VanZandt (1993). 5 Given Eq. (23) with s = 1 and t = 3, the
constant 2 represents a tuning parameter which re ects un- spectral variance j2 contained in the vertical wavenumber
certainty in the value of the cuto . It is important to recog- interval [mj ; ∞] is
nize that the characteristic wavenumber has a signi cantly Z ∞ Z
AN 2 ∞ m0 =m∗
di erent formulation in DSP: namely mDSP = N=DSP . Here j2 = S(m0 ) dm0 = 3 dm0 (24)
c
mj m∗ mj 1 + (m0 =m∗ )4
DSP is calculated from the variance over the entire spec-
trum, which includes large amplitude, low-m waves that "  2 #
are excluded from  and mc as de ned in MK95. Thus, in N2  mj
=A 2 − arctan : (25)
practice mDSP
c is signi cantly smaller than the correspond- 2m∗ 2 m∗
ing MK95 value of mc . In the context of Fig. 3, Hines’ DSP
would subject all waves falling to the left of a vertical line
5 Note that Fritts and VanZandt (1993) specify a slightly di erent
representing
normalization equivalent to A = 4=10 = 0:127 ≈ 1=8 (see their
mDSP
c =mmax = 1=2 (22) Eqs. (1), (2) and (22)).
A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033 1023

Then, for the harmonic with wavenumber mj , the instability


parameter

N 1= A
j = √ = : (26)
2mj  (mj =m∗ ) [=2 − arctan(mj =m∗ )2 ]1=2

For the high-wavenumber tail in which√ mj =m∗ 1, this for-


mula asymptotes to the value j → 1= A. Thus, the asymp-
totic instability parameter for high-m waves is related to
the overall spectral amplitude. Over the observed range of
1=10 ¡ A ¡ 1=2, the asymptotic values of j range from 3.2
to 1.4, corresponding to

j ∼ (0:22 to 0:5)|c − u|:

This implies that a broad spectrum saturates at am-


plitudes far below the usual overturning criterion for
non-interacting monochromatic waves. The correspond-
ing damping rates lay in the range 0:00041 ¡ ÿ=m ¡ 0:7 Fig. 4. The instability parameter j for the jth harmonic mj in a
(see Fig. 2). Thus, a rather broad range of MK95 damp- Desaubies spectrum with s = 1, t = 3 and amplitude A = 1=6. The
ing rates are consistent with “saturated” Desaubies spec- vertical wavenumber mj is normalized by the Desaubies wavenum-
tral tails having m−3 form. For the favoured middle ber m∗ .
atmosphere value A = 1=6, we obtain the asymptotic
value j → 2:45 which corresponds to a damping rate
of ÿ=m ≈ 0:022. For waves at the knee of a spectrum
√ 5. Input source spectra
(mj = m∗ ) with A = 1=6, we obtain ∗ = 2= A ≈
2:76 and ÿ=m ≈ 0:0048, indicating that such waves To apply the parameterization procedure described above,
are damped much more weakly than those in the tail. vertical wavenumber “source” spectra must be speci ed at
For example, waves subject to the knee damping rate some reference altitude. Since the troposphere is believed to
ÿ=m = 0:0048 lose about 1.5% of their amplitude be a region of strong gravity wave generation, source spec-
over a distance of one vertical wavelength, whereas tra are most conveniently speci ed at some level above the
those subject to the “tail” value ÿ=m = 0:022 lose tropopause. Observational studies have provided a number
about 7%. of estimates of amplitudes and spectral shapes as well as
As shown in Fig. 4, j grows rapidly as the wavenumber their variability. However, these studies were mainly con-
mj =m∗ → 0, indicating that the low-wavenumber por- cerned with the behaviour of the high-m tail of spectra which
tion of the spectrum (below the “knee”) will continue to is a ected by saturation processes, whereas harmonics grow-
grow in amplitude as it propagates upward. Waves with ing with height in the unsaturated portion of spectrum are
wavenumbers slightly lower than m∗ therefore experience most important for GW momentum deposition at higher al-
gradually increasing values of damping as they propagate titudes.
upward. Eventually, an equilibrium is reached, and their The problem of estimating vertical wavenumber source
amplitudes saturate. At this point, they become part of spectra from observations is nontrivial because these spec-
the “tail” spectrum and m∗ moves to lower values. This tra characterize wave elds over a range of altitudes, and
process was demonstrated in MK95 (see Fig. 3), and is hence they cannot be attributed to a particular height zs .
considered√ further in Section 8. If, on the other hand, In the presence of vertically inhomogeneous gravity wave
j → 1= 2 ≈ 0:71; corresponding to rapid instabilities sources in the lower atmosphere, instantaneous m-spectra are
associated with overturning and other processes, the wave expected to be highly variable in time, space and wavenum-
will experience damping sucient to cause obliteration as ber. A model spectrum which approximates the typical am-
it propagates a fraction of its vertical wavelength. Thus, plitude and shape of observed PSDs is given by Eq. (23)
we see that the simple formulae that constitute the MK95 (see Fritts and VanZandt, 1993 for further discussion). The
theory contain the essential physics of wave damping, lower m portion of the spectrum is believed to vary gener-
namely Doppler shifting by the background wind, critical ally from s = 0 (“Desaubies” spectrum) to s = 1 (“Modi ed
level approaches, rapid dissipation associated with wave Desaubies”). It should be noted that the uncertainty in this
overturning, and enhanced dissipation due to nonlinear portion of the spectrum ultimately derives from the vertical
interactions within a broad wave spectrum. As demon- variation of background elds, e.g. the density scale height
strated in MK95, it is capable of reproducing the spectral is on the same order as the vertical wavelengths associated
form and altitude dependence of observed gravity wave with waves in this portion of the spectrum. The charac-
spectra. teristic wavenumber m∗ increases in altitude, re ecting the
1024 A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033

entrance of waves with increasingly lower-m into the pro-


cesses of dissipation and saturation.
Constraints for source spectra can be obtained by equat-
ing the observed asymptotic form of Eq. (23) for mm∗ ;
namely S(m) → S0 (m∗ =m)t to the observed form
S(m) = N 2 =(2 to 10)m3 : (27)
The form of Eq. (27) automatically constrains the parameter
t to the value 3, which we adopt for the remainder of the
paper. The spectral amplitude S0 is then constrained to the
range
N2
S0 = : (28)
(2 to 10)m3∗
The values of m∗ observed in the troposphere typically vary
from m∗ = 2=2:5 rad km−1 to 2=900 rad m−1 (Allen and
Vincent, 1995; Fritts and Chou, 1987; Tsuda et al., 1991). Fig. 5. Desaubies (s = 0) and modi ed Desaubies (s = 1) model
An example of two such model spectra with m∗ = 2=∗ = vertical wavenumber spectra for m∗ = 2= = 0:006 rad m−1 and
0:006 rad m−1 ; s = 0; 1 are shown in Fig. 5. However, we S0 = 50 m3 s−2 . The tail exponent t = 3 in both cases.
must keep in mind that the Desaubies form represents aver-
aged observations, and that individual spectra observed in
the lower atmosphere generally have a more complicated
form. There are also indications that the lower-m portion
consists of several peaks representing a systematic “global”
wave source in the lower atmosphere (Weinstock, 1996).
Although such features are undeniably important for the
purposes of wave drag parameterization, unfortunately the
details of source spectra variability are not well known.
However, the model distributions presented in Fig. 5 are
suciently representative for the purpose of examining
the sensitivity to the observed range of source spectra. In
the following section we will examine the sensitivity of
the drag pro le to the choices of s = 0; 1, as well as to S0 .

6. Calculations for typical wind proÿles and source


spectra

In this section we examine the response of the proposed


parameterization scheme to di erent mean wind pro les and
variations of the source spectrum. We have chosen two ◦
mean wind and mean temperature pro les from the CIRA-86 Fig. 6. Mean zonal wind for January at 50 N (solid line) and

50 S (dashed line) latitudes from the CIRA-86 model used in the
model (50◦ N and 50◦ S in January), which may be con-
parameterization tests.
sidered as representative of winter and summer solstices.
The temperature is used to calculate the density in Eq. (10)
and the buoyancy frequency N . The distribution of zonal models. Wave source spectra of the form described above
mean wind u is shown in Fig. 6. The winter pro le exhibits were speci ed in the troposphere at z = 10 km. All results
two eastward jets, one in the troposphere (with maximum presented below were obtained by assuming a characteristic
∼20 m s−1 ) and the other in the stratosphere (∼ 45 m s−1 ), horizontal wavenumber kh =−2=300 rad km−1 and source
and a reversal in the mesosphere. The summer pro le shows spectra spanning the interval from m = 2=900 rad m−1 to
the eastward tropospheric jet (∼ 20 m s−1 ), the westward 2=19 rad km−1 . Both characteristics are typical of obser-
stratospheric jet with peak value ∼ 68 m s−1 and a further vations. In the calculations we consider waves propagating
reversal near 90 km. in both directions with respect to the local wind.
If not noted otherwise, all calculations in the one-dimen- The computational time required for the MK95 scheme
sional model were carried out using a vertical gridstep is essentially proportional to the number of harmonics re-
z = 2 km. This vertical resolution is typical for the mid- tained in the spectrum. We therefore begin by examining
dle atmosphere in many contemporary general circulation the sensitivity of the resulting wave drag to spectral reso-
A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033 1025

Fig. 8. Distribution of horizontal velocity wave amplitudes


as a function of phase speed for the source spectrum with
S0 = 50 m3 s−2 , s = 1, m∗ = 0:006 rad m−1 , and M = 15 harmon-
ics. Note that c = N=m.


Fig. 7. GW drag pro les calculated for “winter” conditions (50 N
January) utilizing M = 200, 30, 15 and 5 harmonics in the model This discretization gives equal spacing in log m and can be
spectrum. A modi ed Desaubies source spectrum with s = 1, eciently formulated to yield reasonable drag pro les in the
m∗ =0:006 rad m−1 , and amplitude S0 =50 m3 s−2 was employed. upper mesosphere and lower thermosphere. The resulting
wave-amplitude source spectrum for S0 = 50 m3 s−2 ; s = 1;
lution. For these tests, we discretize the same source spec- m∗ = 0:006 rad m−1 , and M = 15, is shown in Fig. 8 as a
trum by employing uniform m-grids with varying numbers function of phase velocity, 6 c. Note that we have mapped
of harmonics. Fig. 7 exhibits wave drag pro les obtained the spectral amplitudes onto observed (i.e. Earth-frame)
for the “modi ed Desaubies” source spectrum (23) with phase velocities using the relation c=N=m at the source level.
s = 1; m∗ = 0:006 rad m−1 , and amplitude S0 = 50 m3 s−2 . A discrete wave-amplitude source spectrum of the form pre-
For the purposes of these tests, we employ isotropic source sented in Fig. 8 (with M = 15 harmonics) is used in all the
spectra comprising a total of 2M harmonics, equally dis- tests reported below, unless stated otherwise. Note that in
tributed in the positive and negative horizontal directions. these tests, the parameters S0 , s and m∗ will be varied.
The number of harmonics is varied from M = 200, which We next consider the behaviour of the parameterization
we regard as an “exact” representation, to M = 5. It is seen scheme when various source strengths S0 are speci ed.
from the gure that the resulting wave drag is almost indis- Fig. 9a presents pro les of gravity wave drag for a “win-
tinguishable for M = 200; 30, and 15 harmonics, while the ter” distribution of mean wind employing the “modi ed
M = 5 case still provides reasonable accuracy. As a point Desaubies” source spectra (23) with s = 1; t = 3 and
of comparison we note that multiple-Lindzen schemes typi- m∗ = 0:006 rad m−1 approximated by M = 15 harmonics.
cally employ on the order of 10 harmonics. Since the MK95 Pro les are shown for S0 = 10; 50, and 100 m3 s−2 . It is
scheme does not incur signi cantly larger overhead than a seen from the gure that the factor of ten increase in the
multiple-Lindzen scheme, it is clearly competitive in terms overall spectral amplitude increases the wave drag peak
of computational eciency. value from ∼ − 50 to ∼ − 180 m s−1 day−1 , although we
Most of the gravity wave momentum deposited in the note that the response is not linear. The heights of drag
mesosphere and lower thermosphere originates in the maxima decrease from ∼ 87 km for S0 = 10 to ∼ 80 km for
lower-m=higher-c portion of the source spectrum. In order S0 = 100 m3 s−2 re ecting the obvious fact that waves with
to provide sucient spectral resolution in this key portion, weaker amplitudes generally saturate at higher altitude.
while not incurring an excessive amount of computation The RMS horizontal wind variance  associated with the
for the higher-m=lower-c portion (which only deposits a parameterized wave spectra can also be compared qualita-
modest amount of momentum at lower altitudes), we rec- tively to that deduced from observations. In Fig. 9b we have
ommend the use of variable m-grids with a discretization plotted  as a function of altitude. At the lower boundary
given by ln mj+1 = ln mj + ln r. The ratio between adjacent
wavenumbers, r = mj+1 =mj , is determined by 6 In fact, it is more convenient to label harmonics according to

ln r = (ln mmax − ln mmin )=(M − 1) or; equivalently; their horizontal phase speed c, since the latter quantity is conserved
during vertical wave propagation through a horizontally homoge-
r M −1 = [mmax =mmin ]: neous medium.
1026 A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033


Fig. 9. Pro les of (a) wave drag and (b) horizontal RMS wind calculated for “winter” conditions (50 N, January) using the modi ed
Desaubies input spectrum (s = 1) with S0 = 10, 50, and 100 m3 s−2 . Other parameters are as in Fig. 8.

(10 km),  associated with the input spectra varies between


0.3 and 1 m s−1 , which is in good accordance with observa-
tions (e.g., Sato, 1994), and with values recommended for
use in gravity wave drag models (e.g., Fritts and Lu, 1993).
In the middle atmosphere the calculated pro les exhibit
RMS velocities up to a few tens of m s−1 , which are also
consistent with observations (Tsuda et al., 1994). However,
the RMS wind reaches a maximum near z ∼ 75 km (slightly
below the peaks of momentum deposition) and diminishes
at higher altitudes. Observations, on the other hand, indicate
that  should increase slowly above 60 –70 km (Thorsen
and Franke, 1998). The explanation for this di erence may
be found in the limited range of our spectrum. To produce
stronger parameterized RMS winds at higher altitudes, one
can decrease mmin for the input spectrum. It is natural to an-
ticipate that upon doing so, the peaks in -velocity pro les
will atten, and pro les will vary more slowly with height.
However, we have chosen not to explore this further, owing
to the large uncertainty in the form of gravity wave spectra
at lower m.
In the summer hemisphere, strong stratospheric easterlies Fig. 10. Horizontal RMS wind calculated for “summer” conditions

lter out westward travelling gravity waves almost entirely, 50 S using the modi ed Desaubies input spectrum (s = 1) with
while the tropospheric jet lters out all eastward directed S0 = 10, 50, and 100 m3 s−2 . Other parameters are as in Fig. 8
harmonics with phase velocities below ∼ 20–30 m s−1 . As
a result, only fast harmonics (with low m) can participate for the remaining large-c portion of the spectrum. There-
in the momentum deposition. In our calculations the satura- fore, we expect wave saturation and momentum deposition
tion of these waves generally occurs above 90 km. Fig. 10 in the summer hemisphere to occur somewhat higher than
presents pro les of RMS wind for the “summer” wind pro- in the winter hemisphere.
le calculated with the input spectra described above. It is Figs. 11a and b demonstrate the sensitivity of gravity
seen from the gure, that  associated with the spectrum of wave drag to variations in the low-m source spectrum slope
propagating gravity waves does not exceed ∼ 4 m s−1 up to s and to variations in the characteristic vertical wavenum-
∼ 75 km due to the ltering e ect of the mean wind. Only ber m∗ of the source spectrum. Both calculations are for
above ∼ 90 km does  become comparable with (c − u)  “winter” conditions, using M = 15 spectral harmonics and
A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033 1027

Fig. 11. Wave drag pro les calculated for (a) the input source spectra with (s = 1) and (s = 0) with m∗ = 0:006 rad m−1 , and (b) for the
modi ed Desaubies spectrum (s = 1) with m∗ = 0:006 (solid line) and m∗ = 0:0025 (dashed line). Other parameters are as in Fig. 8.

a source amplitude S0 = 50 m3 s−2 . Fig. 11a presents wave


drag pro les obtained for the slopes s = 0 and 1 with
m∗ = 0:006 rad m−1 . As anticipated, the spectrum contain-
ing more energy at lower m (i.e. s = 0) saturates at lower
altitudes and produces more momentum deposition at the
same height than the spectrum with s = 1, for which the am-
plitudes decrease as m → 0. We also note that the additional
energy present at low wavenumbers leads to dramatically
stronger momentum deposition in the lower thermosphere.
Fig. 11b displays wave drag pro les calculated for the
s = 1 source spectra with m∗ = 0:006 and 0:0025 rad m−1 .
The wavenumber m∗ characterizes the location of the
spectral “knee”, i.e. the largest vertical wavenum-
ber una ected by wave damping processes. Fig. 11b
clearly demonstrates that source spectra with higher m∗
produce maxima of wave drag at lower altitudes.
All the calculations described above have neglected the
nonlinear Doppler shift produced by the spectrum, i.e.
that speci ed by Eq. (2). To justify this simpli cation, we
directly compare results obtained with and without this
Doppler shift. Fig. 12 demonstrates wave drag pro les

calculated at 50 N for M = 15 harmonics using source
spectrum with S0 = 50 m3 s−2 and m∗ = 0:006 rad m−1 .
It is seen from the gure that the neglect of nonlinear Fig. 12. Pro les of wave drag calculated with (dashed line) and
Doppler shifting results in errors of only a few percent, and without (solid line) accounting for spectrum induced Doppler

is comparable with errors introduced by approximating the frequency shift. Pro les have been calculated at 50 N for the
source spectrum with the following parameters: S0 = 50 m3 s−2 ,
continuous spectrum by a discrete one (see Fig. 7).
m∗ = 0:006 rad m−1 , M = 15 harmonics, as in Fig. 8.

7. Comparison with the Lindzen scheme note that the approximation of a broad randomly phased
spectrum has been made in the derivation of MK95, and
If the broad spectrum is replaced by a single harmonic, our strictly speaking, the theory does not apply to the case of a
GW drag parameterization produces formulae which cor- single monochromatic harmonic. Nevertheless, taking this
respond to generalizations of Lindzen (1981). We should limit is instructive and yields some important insights re-
1028 A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033

garding the relation of the nonlinear MK95 scheme to linear


Lindzen-type convective saturation schemes.
The rst step in reducing the MK95 formulae to standard
Lindzen form is to introduce a spectrum consisting of a
single harmonic, so that
S(m) = U02 (m0 )i(m-m0 ) and
Z ∞
2 = S(m) dm = hU02 i; (29)
m0

where m0 is the vertical wavenumber, U0 = Û 0 ei is the hor-


izontal velocity eld and  is the wave phase. Multiplying
both sides of Eq. (1) by (m0 =N )2 , we may rewrite it in the
form
   
d m20 hU02 i m20 hU02 i 0z 2Nz 3m0z
= − − + − ÿ ; Fig. 13. Wave amplitude |Û 0 | (left) and the associated wave drag
dz N2 N2 0 N m0
(30) (right) for a single harmonic with c=25 m s−1 and 2=kh =300 km
propagating through the summer wind pro le shown in Fig.
where the ratio within square brackets is equal to (1=2 2 ). A 6: MK95 scheme (solid line) and Lindzen-type parameteriza-
signi cant property of our nonlinear theory of wave spectra tion with breaking thresholds (i) break = 2−1=2 , (ii) break = 1,
is that under saturation conditions, the parameter becomes (iii) break = 21=2 , and (iv) break = 2.
a very slow function of the intrinsic vertical wavenumber
m0 and mean wind shear and typically lies between 1 and 3
(see Section 4). Thus, if we assume = sat = constant, the We must emphasize that the fully nonlinear broad-spectrum
term on the left-hand side of Eq. (30) must vanish. Using MK95 scheme can yield appreciably di erent behaviour
the relation mz =m ≈ Nz =N + mu z =N and assuming Nz =N = 0 from a Lindzen scheme owing to interactions with the
as Lindzen did, we then obtain rest of the spectrum. Even the monochromatic version of
  MK95 includes nonlinear self-interactions 7 which allow
0z 3m0 u z wave damping and saturation to occur over a broad range
ÿsat = − + :
0 N of . As we shall see, unlike L81 wherein the wave break-
ing onsets abruptly, our scheme produces more gradual
Substitution of this “saturated” value of ÿ together with the
wave damping which onsets below L81 breaking levels.
“saturated” horizontal velocity variance hU02 i = N 2 = sat
2
2m20 ,
These fundamental di erences are illustrated in Fig. 13,
into Eq. (12) yields
  which compares pro les of wave amplitude and associated
kh N 2 0z 3m0 u z drag given for a single monochromatic GW harmonic by
AL81 = − 2 − + : (31)
sat 2m30 0 N the nonlinear MK95 and a generalized version of the L81
parameterization employing sat = 1 and various break-
This drag formula reduces to standard Lindzen form in-
ing criteria ( break ). The example shown is for a harmonic
cluding wind shear (e.g. Holton, 1982) if sat = 1. Thus, if
with phase velocity c = 25 m s−1 , horizontal wavelength
the spectrum consists of one harmonic, and sat is xed at
2=kh = 300 km, propagating through the CIRA mean
some value (i.e. all nonlinear self-interactions which require
wind and temperature distributions for January at latitude
variable are excluded), our parameterization reduces to a ◦
50 S, calculated with vertical resolution z = 200 m. As
straightforward generalization of Lindzen’s (1981) scheme.
expected, Fig. 13a shows that as the parameter break is
The choice of di erent sat corresponds either to di erent sat-
decreased (e ectively increasing the overturning threshold
uration amplitudes, or, alternatively, to di erent “eciency”
amplitude), wave growth with height ceases at progressively
factors. We note, however, that the usual convective over-
higher altitudes. 8 The pro le calculated with the overturn-√
turning threshold invoked in a Lindzen-type scheme may be
ing instability threshold |Û 0 | = |c − u|
 [i.e. break = 1= 2,
written as |Û 0 | = |c − u|,
 which corresponds to
denoted by (i)] corresponds to the original L81 scheme.
√ √
break = (c − u)= 2|Û 0 | = 1= 2: Owing to nonlinear self-interactions and variable , the
2 2
Equating sat to break , expression (31) then yields twice 7 These “self-interactions” may also be regarded as representing
the drag derived by L81. Consequently, this linearized
interactions between waves in the band mj ± mj =2.
monochromatic version of MK95 cannot simultaneously 8 Strict saturation yielding constant amplitude with height does
match the breaking altitude and the drag derived by L81. not occur in this case due to interaction with the mean wind through
Since the drag formula of L81 is generally applied with the second term in Eq. (1); i.e., as the critical level is approached, m
an eciency factor anyway, this discrepancy is of little and the associated wave damping both increase. The linear critical
practical importance. level for this harmonic occurs at an altitude of about 95 km.
A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033 1029

Fig. 14. Horizontal RMS wind (a) and wave drag (b) for the GW spectrum of Fig. 8 propagating through the winter pro le shown in
Fig. 6: nonlinear MK95 scheme (solid line), noninteracting MK95 scheme (MK95 SIO), and multiple L81 scheme with breaking thre-
2
sholds break 2
= 1=2 (labelled just by L81) and break = 2.

monochromatic MK95 wave amplitude (solid curve) ceases the wave drag pro les grow more smoothly with height. In
growth at a lower altitude than the L81 case. By choosing principle, even a rather small amplitude harmonic could ex-
a value of break corresponding√to the range which occurs perience signi cant damping in the presence of other waves
in MK95 (i.e. approximately 2 ¡ break ¡ 2, or Û 0 ≈ in the spectrum which contribute to the RMS wind j .
[0.5 to 0.35]|c − u|),
 we can obtain generalized L81 wave Fig. 14 presents pro les of RMS wind and GW drag calcu-
amplitude pro les [curves (iii) and (iv)] which correspond lated for “winter” CIRA conditions (see Fig. 6) using a mod-
more closely to the MK95 wave amplitude. However, we i ed Desaubies source spectrum of GW (s=1) at the bottom
note that all of the wave drag pro les for the generalized (z = 10 km) with S0 = 50 m3 s−2 and m∗ = 0:006 rad m−1 ,
L81 schemes (Fig. 13b) are signi cantly di erent from that approximated by M = 15 harmonics (illustrated in Fig. 8).
of the monochromatic MK95 drag pro le (solid curve). The The two pro les for the multiple-Lindzen case have been
nonlinearity of the MK95 scheme ensures a smooth onset calculated as the sum of contributions from all harmon-
of wave drag with height. In contrast, wave drag onsets ics; each harmonic propagates and saturates indepen-
suddenly (with maximum value) at the overturning altitude dently. The pro le labelled “L81” corresponds √ to the
in each of the L81 cases. Such a sudden onset of strong original Lindzen breaking threshold break = 1= 2, while
drag can cause instability problems in atmospheric models that labelled “L81 ∧ 2 = 2”, corresponds √ to a revised
(B. Boville, pers. comm.). In order to diminish this sudden Lindzen breaking threshold break = 2. [In both cases
onset of drag, Lindzen (1988) has considered the possi- we set sat = 1 in Eq. (31)]. The other two pro les have
bility of supersaturation of vertically propagating gravity been obtained with the fully nonlinear MK95 scheme, and
waves, a process which is associated with the maintenance a version of MK95 which neglects interactions between
of convective instability. Supersaturation would result in a harmonics with di erent m. The latter can be considered as
smoother onset of wave drag, but of course cannot spread intermediate between a multiple-Lindzen scheme and the
the drag to lower altitudes. full MK95 parameterization. In this latter scheme (denoted
In multiple GW drag parameterizations employed in many as MK95 SIO — “self-interactions only”), the instability
previous studies (Holton, 1983; Garcia and Solomon, 1985) parameter for each harmonic √ depends only on its own am-
momentum deposition is assumed to be created by a set of plitude, i.e. = (c − u)= 2|Û 0 |, thereby excluding the
wave harmonics with di erent phase velocities, each har- higher-m spectral components from the RMS wind . Fig. 14
monic propagating and saturating independently according demonstrates that the altitude at which wave growth ceases
to L81 criteria. In MK95 we systematically account for non- is lowest for the fully nonlinear MK95. Since GW ampli-
linear interactions between harmonics within the spectrum. tudes are lower at lower heights, the associated wave drag is
Since our instability parameter j depends on the RMS wind weaker for MK95.
√ The wave drag given by the L81 scheme
j associated with the higher-m=lower-c portion of the spec- with sat = 2 is very similar to that of the non-interacting
trum (including the amplitude of the “wave” itself), the al- MK95 scheme, although the multiple-Lindzen schemes
titudes at which damping and saturation occur are generally both exhibit jagged GWD pro les owing to the sudden
lower than in the case of waves which propagate and over- onset of breaking for individual harmonics. Generally, the
turn independently. In addition, as we have demonstrated, nonlinear MK95 schemes produce vertical pro les of GW
1030 A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033

Fig. 15. GW drag spectral density multiplied by phase velocity, Fig. 16. PSD, S(c) (in m2 s−2 =m s−1 ), as a function of height
ah (c)|c| (in m s−1 day−1 ), as a function of height calculated for calculated for windless atmosphere.
windless atmosphere.

drag which vary more gradually, with somewhat weaker der to keep the dimensions of the plotted quantity equal
peaks located at lower altitudes than those given by L81. to m s−1 day−1 . This enhances contributions from large
c while maintaining the sign of ah (c). Numerically, the
quantity |c|Ah (c)=c is plotted here, where c is the dis-
8. Spectral decomposition of wave drag cretization in phase velocity space. The source spectrum
was taken to be in modi ed Desaubies form (23) with
In previous sections we have considered the net or inte- s = 1; S0 = 50 m3 s−2 , and m∗ = 0:006 rad m−1 , as in
grated momentum deposition and RMS wind produced by many calculations shown in this paper. Assuming, N =
the entire spectrum. GW drag pro les strongly depend on 0:02 s−1 , the part of the source spectrum with phase ve-
both the details of the source spectra and the vertical distri- locities |c| ¿ 3:3 m s−1 has power-spectral density S ˙ m.
bution of the background wind. Generally, details regarding In this portion the vertical momentum ux u0 w0 (which is
the spectral composition of GW sources in the lower atmo- proportional to Sm−1 ) is uniformly distributed among the
sphere are not well known. Therefore, in this section we harmonics. Waves with negative phase velocities deposit
consider the contributions of individual wave harmonics in negative momentum, and waves with c ¿ 0 deposit east-
creating GW drag. It is more convenient to present spectral ward momentum. If the spectrum is isotropic and there is
quantities in terms of horizontal phase velocity c rather than no background wind, as in the case shown, the wave drag
in terms of the vertical wavenumber m, particularly since produced by harmonics propagating in opposite directions
the former parameter is invariant with height if horizontal cancels, giving nil drag.
inhomogeneities are neglected as in the present treatment. Fig. 15 shows that the phase velocity of the harmonic
For all the plots presented in this section, spectral values at contributing the strongest wave drag increases with altitude.
small c are not shown because we have employed GW spec- As seen in Fig. 16, the phase speed of the largest amplitude
tra limited by mmax = 2=900 rad m−1 or cmin ≈ 3 m s−1 . waves also increases with height. The harmonic with max-
(This is done since many models have separate orographic imum amplitude at a given altitude represents the “knee”
GWD schemes.) The fastest harmonic in our source spec- of the modi ed Desaubies spectrum, and divides the low c
trum corresponds to cmax = N=mmin ≈ 60 m s−1 , where “saturated” portion from the high c “unsaturated” part. The
mmin = 2=19 rad km−1 is the minimum vertical wavenum- RMS velocity uctuations scale with the PSD at m = m∗
ber. In order to provide adequate resolution of the spectral or c = c∗ , growing with height as exp(z=HE ), where HE ≈
decomposition presented in this section, we discretize the 18 km according to Fig. 16. This energy scale height is in
spectrum with constant c and 100 harmonics (50 each in good agreement with estimates from observations (Smith
the eastward and westward directions). et al., 1987) giving HE ∼ 14–21 km, with the most likely
Phase velocity vs. altitude cross-sections of wave drag estimate HE ≈ 2:3H , where H is the density scale height.
spectral density multiplied by the absolute value of phase Our calculations using data from Fig. 16 also show that
velocity, |c|ah (c), and PSD S(c) for a windless atmosphere, m∗ = N=c∗ ˙ exp(−z=H∗ ) with a characteristic scale height
are shown in Figs. 15 and 16, respectively. Since ah is a H∗ ≈ 25 km. Smith et al. (1987) give H∗ = 2 HE , with a
spectral density, we present |c|ah (c) instead of ah (c) in or- corresponding range of HE ∼ 28– 42 km.
A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033 1031

Fig. 17. Phase velocity–altitude cross-section of GW drag multi- Fig. 18. Ratio R = =(c − u)  as function of phase velocity c

plied by phase velocity, ah (c) × |c| (in m s−1 day−1 ), calculated and altitude, calculated for “winter” conditions (50 N, January,

for “winter” conditions (50 N, January, CIRA-86). CIRA-86).

Fig. 17 shows phase velocity-altitude cross-sections of closeness of the RMS wind associated with GW; , to the
wave drag, cah (c), calculated for “winter” conditions (Jan- intrinsic phase velocity of √the harmonic,√|c − u|.
 As a re-

uary, 50 N, CIRA-86) with the same source spectrum used minder, R = 1, or break = ( 2R)−1 = 1= 2 corresponds to
in the previous “windless” case. The background wind sig- the convective breaking criteria utilized by Lindzen (1981).
ni cantly alters GW momentum deposition, by gradually l- As seen from the gure, GW with moderate eastward phase
tering out harmonics with c ¿ u as the spectrum propagates speeds break at altitudes below 60 km via critical layer inter-
upward. The eastward (positive) momentum deposited by actions. However, the momentum deposited by these break-
these harmonics is very weak since these waves are ltered ing waves is very small due to the smallness of the wave
out below ≈ 60 km, where their amplitudes are still rela- amplitudes at such low altitudes. The gap near c = 0 rep-
tively small. Above this height, harmonics with c ¡ u satu- resents an absence of waves in the spectrum. Most of the
rate, transferring westward momentum to the mean ow. As total wave drag above 60 km is caused by GW harmonics at
seen from Fig. 17, only slow harmonics with absolute phase R between 0.3 and 0.5, which translates into ≈ 2:35 and
velocity less than approximately 10 m s−1 contribute sig- 1.4, correspondingly. This suggests that the main contribu-
ni cantly to the main peak of wave drag near z=80 km. The tion to the total wave drag is made by GW harmonics with
second (weaker) maximum of wave drag near z ≈ 95 km amplitudes below the convective breaking criteria R = 1. It
is created by faster westward-propagating waves. Compar- is also seen from Figs. 17 and 18 that the contribution of
ison with the phase velocity-altitude cross-section of wave waves with R ¿ 0 is negligible above ≈ 60 km.

drag for the “summer” wind distribution at 50 S (not shown
here) reveals strong asymmetry in spectral momentum depo-
sition between the two hemispheres. The “summer” strato- 9. Summary and conclusions
spheric jet entirely lters out harmonics with c ¡ 0, while
the tropospheric westerlies lter out “slow” eastward mov- We have presented a spectral parameterization scheme for
ing waves with c less than approximately 20 m s−1 . As a calculating gravity wave momentum deposition in the mid-
result only “fast” GW penetrate to the upper stratosphere dle atmosphere. The scheme is based on the MK95 theory
and mesosphere, where they deposit signi cant momentum of the vertical evolution and saturation of GW spectra.
owing to their large amplitudes. It is worth noting that the The parametrization has been tested using a column model
total wave drag produced by eastward travelling wave har- for representative CIRA86 distributions of mean wind and
monics near the “summer” mesopause is on the same order temperature. Once the vertical wave number spectrum of the
of magnitude as the total drag exerted by the c ¿ 0 portion GW source has been speci ed, the proposed parameteriza-
of the GW spectrum near the main peak at ≈ 80 km in the tion has only one tunable parameter, namely the character-
winter hemisphere. istic horizontal wavenumber of subgrid-scale gravity waves
Some additional information on the behaviour of satu- kh . As in other schemes, this parameter serves as an “e-
rated GW spectra can be obtained from Fig. 18, where the ciency” scaling factor for wave drag; however, in contrast
spectral-altitude cross-section of the ratio R = =(c − u)
 is to Lindzen schemes which generally require low values of
shown for “winter” conditions. This ratio characterizes the kh , values characteristic of the dominant portion of the GW
1032 A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033

spectrum may be assigned to kh in our GWD scheme. Of lytically integrable source spectrum, and obliterates a sub-
course, further tuning can be accomplished by varying the stantial portion of the tail spectrum in order to provide the
source spectra at the lower boundary within observational necessary wave drag. The algorithm proposed here for the
limits, e.g. by imposing latitudinal or temporal dependen- MK95 parameterization requires fewer adjustable parame-
cies. This implicitly characterizes the degree of physical ters, makes only minor approximations relative to the orig-
and theoretical completeness of the MK95 scheme, and em- inal theoretical framework, and has the advantage of track-
phasizes the need for a more extensive knowledge of GW ing individual harmonics in the spectrum. This has allowed
sources in the lower atmosphere. us to compare the resulting GW spectrum and RMS wind
For the range of anticipated variability of source spectra to observations, as well as to consider spectral composition
parameters in the troposphere, the scheme produces wave of the net deposited GW momentum. For example, our re-
drag pro les as well as RMS horizontal winds due to GW sults show that “slow” GW harmonics with observed phase
that are consistent with observations and with theoretical velocities c less than ≈10 m s−1 contribute the most to the
estimates. Sensitivity tests have shown that for purposes total wave drag in the mesosphere, while faster harmonics
of GW drag parameterization, the e ects of nonlinear are important at higher altitudes.
spectrum-induced Doppler shifting derived by MK95 can As with any other successful GW parameterization, pre-
be neglected compared to the e ects of nonlinear damping cise tuning of the scheme requires a more detailed speci-
and “linear” Doppler shifting by the background wind. cation of source spectra in the lower atmosphere. MK95
We have demonstrated that our scheme can be viewed has the advantage that all tuning is e ectively con ned to
as a nonlinear generalization of a multiple-Lindzen-type the speci cation of the source spectrum. The information
(L81) parameterization. For a broad spectrum of waves, our regarding the spectral composition of the net drag provided
scheme di ers from the multiple-Lindzen scheme by ac- by the MK95 scheme can greatly simplify this tuning pro-
counting for nonlinear interactions between harmonics. The cess, and we suggest that other spectral gravity wave drag
wave damping rate for the MK95 scheme varies √ continu- schemes may bene t from a similar analysis.
ously with the instability parameter =|c− u|=
 2, where c
is the phase speed, u is the mean wind, and 2 is the spectral
wind variance corresponding to harmonics with similar and Acknowledgements
higher vertical wavenumber. In Lindzen’s approach
√ (L81),
wave breaking occurs abruptly at xed =1= 2, and damp- This research was partially supported by a group Strate-
ing occurs at a rate associated with background ow param- gic Grant from the Natural Sciences and Engineering Re-
eters. As an example of the exibility of the MK95 scheme, search Council of Canada and by the Canadian Climate Re-
it can, with only a couple of minor coding changes, be con- search Network of the Atmospheric Environment Service
verted to a standard multiple-Lindzen scheme. Of course, the of Canada. We gratefully acknowledge helpful discussions
inherent nonlinearity of the MK95 scheme ensures gradual with Len Sonmor and Byron Boville. Suggestions from two
wave damping and associated wave drag at heights where anonymous reviewers helped to improve the presentation.
wave amplitudes are below the breaking criterion.
Our tests have shown that the total number of harmonics
representing the spectrum can be reduced to 10 –30 without
References
a signi cant e ect on the resulting wave drag produced by
the MK95 scheme. Computationally, then, MK95 is as ef-
Allen, S.J., Vincent, R.A., 1995. Gravity wave activity in the
cient as a multiple-wave Lindzen scheme, and is suitable lower atmosphere: seasonal and latitudinal variations. Journal of
for use in GCMs. Preliminary results from implementations Geophysical Research 100, 1327–1350.
in a mechanistic version of the Middle Atmosphere version Alexander, M.J., Dunkerton, T.J., 1999. A spectral parameterization
of CCM2 and in the Canadian Middle Atmosphere model of mean- ow forcing due to breaking gravity waves. Journal of
have been reported in Medvedev et al. (1997,1998), respec- the Atmospheric Sciences 56, 4167–4182.
tively. As a guide to those who wish to implement the MK95 Broutman, D., Macaskill, C., McIntyre, M.E., Rottman, J.W., 1997.
scheme, we have successfully employed source amplitudes On Doppler-spreading models of internal waves. Geophysical
of S0 ∼ 200 m3 s−2 and kh ∼2=300 rad km−1 for isotropic Research Letters 24, 2813–2816.
sources and S0 ∼450 m3 s−2 and kh ∼2=100 rad km−1 for Eckermann, S.D., 1997. In uence of wave propagation on the
anisotropic sources. Source spectra are launched at p = Doppler spreading of atmospheric gravity waves. Journal of the
Atmospheric Sciences 54, 2554–2573.
165 hPa and represented by a modi ed Desaubies spectrum
Fritts, D.C., 1984. Gravity wave saturation in the middle
with mmin = 2=19 rad km−1 ; mmax = 2=900 rad m−1 , and
atmosphere: a review of theory and observations. Reviews of
m∗ = 0:006 rad m−1 (see Section 6). Geophysics and Space Physics 22, 275–308.
Hines (H97) introduced several “fudge factors” (his own Fritts, D.C., Chou, H.-G., 1987. An investigation of the vertical
terminology for 1 ; 2 , etc.) and other approximations to wavenumber and frequency spectra of gravity wave motions in
adapt the Doppler-spread theory into an ecient parame- the lower stratosphere. Journal of the Atmospheric Sciences 44,
terization. In addition, H97 recommends the use of an ana- 3610–3624.
A.S. Medvedev, G.P. Klaassen / Journal of Atmospheric and Solar-Terrestrial Physics 62 (2000) 1015–1033 1033

Fritts, D.C., Lu, W., 1993. Spectral estimates of gravity wave (Ed.), Gravity Wave Processes: Their Parameterization in Global
energy and momentum uxes. Part II: parameterization of wave Climate Models. Springer, Berlin, pp. 244–274.
forcing and variability. Journal of the Atmospheric Sciences 50, Medvedev, A.S., Klaassen, G.P., 1995. Vertical evolution of gravity
3695–3713. wave spectra and the parameterization of associated wave drag.
Fritts, D.C., VanZandt, T.E., 1993. Spectral estimates of gravity Journal of Geophysical Research 100, 25,841–25,853.
wave energy and momentum uxes. Part I: energy dissipation, Medvedev, A.S., Klaassen, G.P., Boville, B.A., 1997. The
acceleration, and constraints. Journal of the Atmospheric parameterization of gravity wave drag based on the nonlinear
Sciences 50, 3685–3694. di usion of wave spectra. In: Hamilton, K. (Ed.), Gravity Wave
Garcia, R.R., Solomon, S., 1985. The e ect of breaking gravity Processes: Their Parameterization in Global Climate Models.
waves on the dynamical and chemical composition of the Springer, Berlin, pp. 309–325.
mesosphere and lower thermosphere. Journal of Geophysical Medvedev, A.S., Klaassen, G.P., Beagley, S.R., 1998. On the
Research 90, 3850–3868. role of an anisotropic gravity wave spectrum in maintaining
Hamilton, K.P., 1997. The role of parameterized drag in a the circulation of the middle atmosphere. Geophysical Research
troposphere–stratosphere–mesosphere general circulation model. Letters 25, 509–512.
In: Hamilton, K. (Ed.), Gravity Wave Processes: Their Mengel, J.G., Mayr, H.G., Chan, K.L., Hines, C.O., Reddy, C.A.,
Parameterization in Global Climate Models. Springer, Berlin, Arnold, N.F., Porter, H.S., 1995. Equatorial oscillations in the
pp. 337–350. middle atmosphere generated by small scale gravity waves.
Hecht, J.H., et al., 1998. A comparison of atmospheric tides inferred Geophysical Research Letters 22, 3027–3030.
from observations at the mesopause during ALOHA-93 with the Norton, W.A., Thuburn, J., 1997. The Mesosphere in the Extended
model predictions of the TIME-GCM. Journal of Geophysical UGAMP GCM. In: Hamilton, K. (Ed.), Gravity Wave Processes:
Research 103, 6307–6321. Their Parameterization in Global Climate Models. Springer,
Hines, C.O., 1993. The saturation of gravity waves in the middle Berlin, pp. 384–401.
atmosphere. Part IV: cuto of the incident wave spectrum. Sato, K., 1994. A statistical study of the structure, saturation
Journal of the Atmospheric Sciences 50, 3045–3060. and sources of inertio-gravity waves in the lower stratosphere
Hines, C.O., 1996. Nonlinearity of gravity wave saturated spectra observed with the MU radar. Journal of Atmospheric and
in the middle atmosphere. Geophysical Research Letters 23, Terrestrial Physics 56, 755–774.
3309–3312. Smith, S.A., Fritts, D.C., VanZandt, T.E., 1987. Evidence for a
Hines, C.O., 1997. Doppler-spread parameterization of gravity- saturated spectrum of atmospheric gravity waves. Journal of the
wave momentum deposition in the middle atmosphere. Part Atmospheric Sciences 44, 1404–1410.
1: basic formulation. Journal of Atmospheric and Terrestrial Thorsen, D., Franke, S.J., 1998. Climatology of mesospheric
Physics 59, 371–386. gravity wave activity over Urbana, Illinois (40◦ N, 88◦ W).
Hines, C.O., 1999. Comments on “In uence of wave propagation Journal of Geophysical Research 103, 3767–3780.
on the Doppler spreading of atmospheric gravity waves”. Journal Tsuda, T., Murayama, Y., Nakamura, T., Vincent, R.A., Manson,
of the Atmospheric Sciences 56, 1094–1098. A.H., Meek, C.E., Wilson, R.L., 1994. Variations of the
Holton, J.R., 1982. The role of gravity wave induced drag and gravity wave characteristics with height, season and latitude
di usion in the momentum budget of the mesosphere. Journal revealed by comparative observations. Journal of Atmospheric
of the Atmospheric Sciences 39, 791–799. and Terrestrial Physics 56, 555–568.
Holton, J.R., 1983. The in uence of gravity wave breaking on Tsuda, T., VanZandt, T.E., Mizumoto, M., Kato, S., Fukao,
the general circulation of the middle atmosphere. Journal of the S., 1991. Spectral analysis of temperature and Brunt–Vaisala
Atmospheric Sciences 40, 2497–2507. frequency uctuations observed by radiosondes. Journal of
Lindzen, R.S., 1981. Turbulence and stress owing to gravity wave Geophysical Research 96, 17,265–17,278.
and tidal breakdown. Journal of Geophysical Research 86, Warner, C.M., McIntyre, M.E., 1999. Toward an ultra-simple
9707–9714. spectral gravity wave parameterization for General Circulation
Lindzen, R.S., 1988. Supersaturation of vertically propagating Models. Earth, Planets and Space 51, 475–484.
gravity waves. Journal of the Atmospheric Sciences 45, Weinstock, J., 1982. Nonlinear theory of gravity waves: momentum
705–711. deposition, generalized Rayleigh friction, and di usion. Journal
Manzini, E., McFarlane, N.A., McLandress, C., 1997. Impact of the Atmospheric Sciences 39, 1698–1710.
of the Doppler-spread parameterization on the simulation of Weinstock, J., 1990. Saturated and unsaturated spectra of gravity
the middle atmosphere circulation using the MA=ECHAM4 waves and scale-dependent di usion. Journal of the Atmospheric
general circulation model. Journal of Geophysical Research 102, Sciences 47, 2211–2225.
25,751–25,762. Weinstock, J., 1993. Lagrangian coordinates and their application
McFarlane, N.A., McLandress, C., Beagley, S.R., 1997. Seasonal to gravity wave spectra. In: Thrane, E.V., Bix, T.A., Fritts, D.C.
simulations with the Canadian Middle Atmosphere Model: (Eds.), Coupling Processes in the Lower and Middle Atmos-
sensitivity to a combination of orographic and Doppler-spread phere. Kluwer Academic Publishers, London, pp. 241–260.
parameterizations. In: Hamilton, K. (Ed.), Gravity Wave Weinstock, J., 1996. Spectra and a global source of gravity waves
Processes: Their Parameterization in Global Climate Models. for the middle atmosphere. Advances in Space Research 17,
Springer, Berlin, pp. 351–366. (11)67–(11)76.
McLandress, C., 1997. Sensitivity studies using the Hines and Whiteway, J.A., 1999. Enhanced and inhibited gravity wave
Fritts gravity wave drag parameterizations. In: Hamilton, K. spectra. Journal of the Atmospheric Sciences 56, 1344–1352.

You might also like