You are on page 1of 25


4, PAGES 851-$75, NOVEMBER 1982

Development of a Turbulence Closure Model
for Geophysical Fluid Problems
Unlver#ry, Princelon, Ne"' i t rsey O&J4()
Los Alamos National Loboratory, Los Alamos, Ntw Mtxlco 87545
Application! of SCC<!ndmoment closun: hypothcscs lO gcophysical Ouid problems bavc
dcvclopcd raptdly smce 1973, wben genuinc predictivc skill in coping with thc ciTccts ofstratilication
Wtts The porpose bere is to and organize material that has appeared in a
numbcr or anocles and add ncw =fui material that a complete (and improved) dcscription of a
model from conptJon to apphcatJOn 1s condenscd in a s:inglc anitlc. Jt is ho,xd tbat ths
wdl be a u.s.eful reference to users of IM model for applicalion to e.i1her atmospheric or oceanic
boundary layers.
This article is a summary of our experiencc at Princeton
University in secondmoment modeling where we have tried
to trace model development from the detcrmination o r model
the applications of the model to geophysical
ftutd dynamoc problems. lt includcs, in summary fashion tbe
basic modeling assumptions first put forth by Mellor 'and
Herring (1973) , Mcllor [19731. and Mellor and Yamada
11974] and references numerous other application papers by
us and others who have used the model. New material is also
included in this paper.
Ou r first cxperience with second-moment turbulence mod-
eling began at the time of lhe 1968 Stanford Conference on
Boundary Layer Computation (Kiine et a/., 1968)
With a stmple Prandtl type model based on the solution ofthe
turbulent kinetic encrgy cquation. AI thal time, this model, a
model by Bradshaw, and our older eddy viscosity model did
very well in predicting ali of tbe data compiled for tbe
conference. Thus it was not elear that more complicated
models based on hypotheses by Rotta 11951) and Kolmo-
gorov 11941) and requiring consideration of ali components
of the Keynolds stress tensor could be justi6cd on the hasis
of improving predietions.
The first results from the second-moment calculalions of
and [1968) (using Rotta's cnergy
hypothesos but not the Kolmogorov isotropic
dosstpauon hypothcsis) when applied to wakes and jets and
subscquent studies by Haflia/c and Ltmnder [1972] were
encouraging. Tbe newer models, bowever, involved more
empirical constants than did the older models. whieh. on this
basis aJone, would facmtate agreement with data.
Our interest in thc Roua-Kolmogorov model [Me/for and
Herrmg, 1973) was greatly enhanced by data obtained by So
and [1973, 1975) wbich demonstrated the etfect of
wall curvature on turbulent tlow. On stable, concave walls
the Reynolds stress was vinually extinguished in the inner
Cepyrigbt 1982 by thc Amcrican OcophysicaJ Uoion.
Paper numbcr 2R0808.
0034-6853/821002R .()8()8$ I S. 00
portion. WiJh no empiricaJ adjustment invo1ving curvature,
the model quantitative1y predicted this rather dramatie oc-
currence ISo, 1915; Mellor, 1975]. Tbe same model, when
in (it now seems) a rather straightforward manner,
prcdtcted thc observcd stabilizing and dcstabilizing cffcct of
density gradients in a gravity ficld (Me/lor, 1973] according
to data by Businger [1971). Ali empirical constants
were obtained from neutral fl ows and were directly related
to data for the s pecial case where turhulenl energy produc-
tion is in balance with dissipation. This process of selecting
constants is reviewed in lhis paper aod refined somewbat in
relation to the eartier papers.
The model was then extended and applied to nows of
practical interest in gcophysical ftuid dynamics. Wc wiJI
illustrate some of thesc applications and note referenccs to
other investigators who have made use of lhe model.
The model (or modcls, since tbere are various degrces of
approximation aod simplification) is not fundamentally dif-
ferent from models by L'wel/en and Tesk' (1973],
ti a/. [1976}, IA under 11975), and Zeman and Lumley (1976)
in that hypotbeses by Roua and Kolmogorov are their more
imponant elements togelher with the fac! thaJ various turbu-
Jent 1ength scales, which reside in tbe hypotheses, are a1J
assumed to be proponional . There are dilferences. For
example, we use a relatively low order version of Rotta's
'cnergy redislribution' bypothesis. Other aulhors claim
some benefit in adding more nonlinear terms: bowever. our
perception is that the benelits are marginal and may even
create errors in one application relative to the one in which
t.he additional, requisite constanJs were obtained. We also
use a fairly simple turbulent dilfusioo modcl in oontra5t, for
example, to the more complex dilfusion model of et
a/. l 1916a. b). lt is probab1y lrue that in many cases,
turbulent diffusion is not modeled accurately by us and
others. The limiting, free-convection case (a shearless
bouodary layer heated from below and bounded from above
by a stable, inversion layer) is cited by Lum/ey et a/. I 1978)
as an application peculiarty in need of higher-order diffusion
modeling. They state that in models using grodient transpon,
'the rise of the inversion base is very poorly predicted, while
lhe vertical distribution of turbulent energy is wildly in
852 MeLLOR ANo Y AMADA: Tva:tauL..e.Nce CLOSuRB Mooe.1..

error .' As will be seen bclow, our own experience contra-
dicts lhis statement. As a matter of fact, practically any
model, including lhe so-called 'convective adjustment' mod
el, wiU produce accurate prediction of the rise of tbe
inversion base. and. further, our model seems to do well in
predicting turbulcnt energy and other turbulent quanlities, as
did the modcl by Lewe/len et ai. [1976). (See also Lewellen
[1977] for a concise review of the ' ARAP" model. A review
by Harsha [1977) in lhe same monograph further reviews a
generally simpler class of models.)
The major weakness o f ali the models probably relates 10
lhe turbulent master length scale (or turbulent macroscale,
or turbulent inertial scale), and, most important, to the fact
that one sets ali process scales proportional to a single se ale.
As we have tried to do here, it is possible first lo test the
proportionality idea exclusive of the length scale per se, and
ali seems 10 work not perfectly. but well.
The master length scale equaLion we use is quite empirical.
11 acknowledges that eddies and their scales advecL; it
undoubtedly intcrpolates well between t he few well-docu-
mented laboratory ftows on which il is based and does seem
to perform correctly for geophysical flows ; however, it may
be deficient if extrapolaled far from its data base. Neverthe-
less, on conceptual grounds we believe Lhe equation we use
is much to be preferred relative to lhe transpor! equat ion for
dissipation adopted by many to provide lhe needed length
scale; our reasons will bc discussed below.
Allhough there are differences in the various models, lhe
overall observation may be made that in less than a decade
we bave progressed from an emphasis on modeling mean
flow properties per se to a concern for lhe accuracy of
modeling turbuleot variances and covariances. To be more
specitic, it would appear that in comparison with laboratory
data, maximum errors for these latter quantities are of tbe
order of 30-50%, whereas errors in mean properties are
much less.
On the basis of the ralher broad spectrum of laboratory
flows simulated by tbe models , it may not be too optimistic
to believe thaL they should perform realistically in more
complicated geophysical situalions. Recent experience
where the t urbulence models have been incorporaled inlo a
large-scale geophysical numerical model contributes to this
2. TliE BAste MooEL
The C/osure Assumptions
The equations for the ensemble mean velocity U
, pres-
sure P, and potential Lemperature e are
ap a
- +- (pU;) = O
qt ax;
oe a
p- ~ - ( - p(u,e))
DI ax.
where D( )/DI '" U,a( )lx, + ( )dt; gJ is tbe gravity
vector, and ft is t he Coriolis vector. Uppercase letters
represent ensemble mean variables, whereas lowercase let
ters are the turbulent variables. Angle brackets represent
ensemble means o f turbulence variables. An exception is the
mean densit y p. 1t should be noLed t hat t he Boussinesq
approximation (densily a constant except in the buoyancy
l.erm) necd not be invoked for the mean equations o f motion.
However, in what follows, the Boussinesq approximation
for t he turbulence equations is invoked in t hat density
llucLuations are neglected except in t he buoyancy term,
where, indeed. they exert a profound influence on the
Lurbulence. In the ocean the mean density is related through
the equation of state LO the mean potentialtemperature and
salinity. The equation for salinity is identical to {3). Analo-
gously, in l he atmosphere, mean density is related to mean
potential temperature and water vapor. In the following
discussion. density fluctuat ions are written f30. where f3 is
the coefficient of thermal expansion. lt is a simple extension
to include Lhe elfecLs of salinity and water vapor fluctuation.
l nclus ion of liquid water in the atmosphere is more compli-
caled and is postponed until section 8.
In general. a radiative tlux divc rgence term should appear
on the right side of (3); howe vcr, in this paper. radiativc
effects will not be considered.
lf one writes the s ingle-point equations for the moments of
velocity and temperature, unknown higher-moment terms
appear. The model we adopted [Mellor and 1/erring. 1973;
Mellor, 1973) (see these papers for tbe mclhodology by
which the specific tensorial algebra is established) to deter-
mine these unknowns is based on thc energy redistribution
hypolhes is of Rona in the form
- - +- = - - (ll;lt;) - - q
P (au, au
)) q ( ao )
p i!x
+ c,<( - +-
..2 (iJU
Xj d.t1
and the Kolmogorov bypothesis of local, small-scale isotro-
PY sucb tbaL the dissipation is modeled according to
iJu1 i!u
) 2 <I
2v --- =-- ao
iJx, dx, 3 A,
In the above, u
is the turbulent fluctuation velocity. whereas
r/ "' (u?}; p is t he flucl uation pressure; v is tbe kinematic
viscosity; /
and A
are length scales; and C
is a nondimen
sional constant. The above model was then extended (Mel
lor. 1973)to include temperature (or any scalar) such tbat
/ p !!...) = - .!!.... (u;ll) (6)
\p ax
{a + v) -- =O
ax. ax.
wbere ais tbe tbermal di.ffusivit y. The temperature variance
dissipation is
iJIJ i)IJ ) q
2a - - = 2 - ( li-)
ax. ax. Az
wbere (li) is lhe temperature variance and /
and 11
length scales.
To complete tbe model. we must add closure expressions
for (u.u
u;), (pu
), (u
11), and (plf). The choice is ambiguous,
as discussed by Mellor (md Herring [1973) and Mellor [ 1973).

a( I!)'
(u,li) = - lq s8 -
to model what we shall call the turbulent velocity diffusion
terms. s . s . and S, are dimensionless numbers which
might be absolute constants or functions of invariant param-
Consistent with the above, one might reasonably specify
(pl!) = O, since there is no zero-order tensor involving a
gradicnt of a scaler, and also spccify (pu,) = lq Sq' a</'lax,.
The problem is that the relative roles of velocity diffusion
versus pressure diffusion do not seem to be understood
expcrimentaUy. However, within approximations made in
section 3, the important part of (9) is seen to be (u u.') = /q
s. a</'lax . Thus thc pressure and velocity diffusion tenns are
hardly distinguishable; that is, the present model would
probably not discriminate between the two typcs of diffusion
for ali of the cases discussed in this papcr. To reduce
nomenclature, we will here formally set Sq' = O, but it must
be stated that we do not know how the model (or, apparent-
ly, real data) divides the total dift"usion in to its two separate
pans. There will be more discussion on this poiot in section 7
relative to the free convection problem.
Despite the uncertaioty connected with model equations
for turbulent diffusion, it is probablc that the concomitant
error in predicting mean propcrtics is not large.
I! is fundamental to current second-moment models (and,
pcrhaps, their greatest weakness) that aU length scales be
evcrywhere proportional to each other. Thercfore we set
where I is the master lurbulent length scale. The conslants
. B 1, A
, 8
, and C
musl be determined from data. This
can be accomplished wilhout rcsort to a trial and e.rror
process (somelimes termed cornpuler optimization') by
appcaling to data where turbulent energy production and
dissipation are balanced. This will be discussed later. as well
as means to determine /. The remaining unknowns are s., ,
S.,., and s., although we note here that up to the present time
they have been set equal to each other.
Higher-order tcnns (see definition of ' order' in section 3)
can be added to the above closure approximations [Launder
er ai., 1975; Launder, 1975; Wyngaard, 1975): for example,
to equation (4) one could add
C,(Pikiif + P1m - (21!;/J)PIIdrJ) + C,(P,,.. + PJtik
- (2l!;j3)Pwt) + C.{J(gl,uJf11 + g
{u;d) - (26j3)gt(u,IJ))
where Pu>m "' -(u,u
} iJUwiJx,; to equation (6) one could add
{{Jf!} + C6(0u,) iJU,IdxJ + C,{ Ou,) iJUjiJx
In a rotating flow, other terms containing the rotation vector
can be added along with other higher-order terms ILum/ey
and Khajeh-Nour, 1974). AdditionaUy, in lhe vicinity of
walls. tenns involving a unil vector "X
normal to a waU could
be included [Monin, 1965]. However, wc have resisted
added complexity for the following reasons: (I) Mellor [1973]
found that lhe above model met with immediate success in
predicting the very dramatic, stabilizing or destabilizing
ell'ects of density stratification in a gravity field. Similar
effects offlow curvature were equally well predicled. (2) The
data base is not sufficient or accurate enough to determine
many constants. (3) We are motivated to minimize complex-
ily and lhe number of empirical constants. (4) The weakesl
link in our model (and ali other aclive models) is probably
the length scalc equation (scction 5) ralher than the closure
assumptions (4)-(11).
Thus the model rcpresented by (4)-(12) is relatively sim-
pie. Our expectation of thc model is that it will accurately
predict mean velocily and tempcrature fields and do a
reasonable job o f estimating turbulent variances and covari-
The Leve/4 Model
N ow if the closure assumptions are inserted into the mean,
turbulent moment equations (contained in many of the
references ciled in this paper), the model which we had
labeled lhe levei 4 model [Mellor and Yamada. 19741 is as
D(u,u) . iJ [' (iJ{u;11j) a(u
u.) a{up>)]
-'-:-'--''- - - ~ lq s. + +
Dt ilx. iJx, Xj ax,
=- - (u,uj} - -<l + c,c; - +- -- -l!u
q ( 1Ju ) (au, au,) 2 t/
3/1 3 dXj ~ X { 3 A,
In the above equations we distinguish 8. from 8 where the
former always appears in combination with {J, thus repre-
senting a density fluc.luation. However, for most of the
discussion in this papcr we let e. = o. in which case the
above equation set is closed. Later, however, we wiU
consider atmospheric problems where water vapor and
liquid must be included in the equalion of state and oceanic
problems where salinity is an important factor; then o. and O
will acquire scparate identities.
.. - . --

The Leve/ 3 Model
Although we havc used the complete model in the numeri-
cal solution of problems [Mel/or and Yamada, 1974, Briggs
et a/., 1976], it is generally too complicated for practical
application to most geophysical fluid dynamics problems. As
discussed I ater. the model must be extended to include other
scaJar quantities besides temperature (in the atmosphere,
water vapor and liquid water along with olher chemical
constituents; in the oceaos, salinity and other chemical
constituents), and the numerical effort can quickly get out of
A process of simplilication has been described by Mel/or
and Yamada (1974] in some detail. BrieRy, it iovolves
scaling ali terms in the model equations as a product of e/IA
and powers of a, whcre /1 = 0 (/1
) and a' = O(a/); aij is the
nondimensional measure of anisotropy in tbe cxpression
) = ( + "") q
Similar paramet.ers are introduced for the temperaturc varia-
bles. We then evaluated terms in (13), (14), and (15) in
powers of a and eliminated terms of order '. Ali of this is
suggested by thc kinetic theory of gases, wherein a is related
to the Knudson number and is generally a very small
number. For turbulent flows, a/ 0.15 and is not overly
small . Nevertheless. the procedure contributes some disci-
pline to the process of simptification and provides a self-
consistent model. What we have caUed a 'levei 3' model
[Mellor and Yamada. 1974) is as follows:
Del a [ aq']
- - - lq s. - = U.P, + Ph - e)
Dt ax, ax,
+ fJgJ.O"II) + h e;., (u,ll)]
- - lq s8- = - 2(u,l1) -- - (li}
D<lf> a [ , ae 2q
Dt ax, ax, ax, A2
P, = aU/x;
is the shear produclion of turbulent energy,
P "' - {Jgt<,u
-. -
is lhe buoyant production, and
e "' q
is the model dissipation.
(A mistake was made in Lhe paper by Mellor and Yamada
[1974). Tbe last term in their equation (21) and subsequent
terms labeled D
in (52 a, b, c) and Df in (55 a, b) and (56)
should be purged, since they are of order a' . The mistake is
embarrassing, since the original purpose of lhe ordering
analysis was to etiminate terrns.)
11re Leve/ Model
Subsequcnt to the Mel/or and Yamada (1974) paper a
further modificati on has been made to lhe levei 3 model, the
result of which is hcreby labeled lhe levei '2!' model.
The modification is to neglect the material derivative and
dilfusion terms in (15) or (19). This modification immediately
removes the need to solve one dilferential equation but also
is consistently applied to extensions where, say, water vapor
(or salinity in the case of oceanic appl ication) is added to the
list of variables. Without the modification one would need to
solve differential equations for temperature variance and
water vapor variance and ao equation for the cross correla-
tion of water vapor and temperature. The situation quickly
gets out of hand if more scalars are added to the list of
corilputed variables. This change is not generally justified by
our orderiog analysis. However, a close reexamination of
the ordering analysis indicates that the modification can be
justit\ed for ali stable ftows and slightly unstable flows; error
is more likely as one approaches the free-convect ion limit.
Nevertheless, our practical experieoce indicates thal lhe
modification should exert little effect on computed results.
We therefore replace (19) by
(fi!) = - /12 (u,IJ)
q i!x,
The Leve/ 2 Model
This model ncglects ali material derivative and diffusion
terrns. Thus (16) is replaced by P, + Pb = e and, further,
consists of ( 17), (18) (which can be further simplified by the
production-dissipation balance), and (21).
We note that a levei I model was also identifi ed by Mellor
arrd Yamada (1974). However, it fails to reproduce observa-
tional data (as in Figures 5a and 5b) and does not olfer
significant, compensatory mathcmatical simplilication.
Tloe Boundary Layu Approximation: Levei
For ali subsequent versions of the model we will negleet
the Coriotis term in the turbulence moment equation. This
appears to be justified in boundary layers wbere /flu, is
small. (However. the issue probably should be investigated
more thoroughly.) Note that if tbe Coriolis terms were
retained, algebraic simplification of the type to be presented
in this section does not appear possible; a large matrix
inversion is required to obtain the ftuxes for insertion into
tbe mcan equations of motion, (2) and (3).
Now, i.f one makes tbe boundary layer approximation, the
vertical component of the momentum equation becomes
hydrostatic, and in the other equations, ali components of
the tensor, aujax,, may be neglected except for autaz and
.- ------. .. - . - ---- --' ..
avtaz. Then, if we set g
= (0, O, -g), (2) and (3) simplify to I f we further define
o a
p - U + - {l,uw) = -Pix + p/V
Dt az
D a
p - V + - r/,vw) = - aP/ily - pfU
Dt z
O = -aPi az - pg
(D9/Dt) + ((w6)/z) = O
Although ali modcl leveis are simplified by virtue of the
boundary layer approximation, leveis and 2 are mos!
advantageously afTccted. Here we proceed with discussion
of the levei 2! model. Thus (16), after the boundary layer
approxi mation. is
E_ (1) -..!_ [lqS = P, + Pb- e (24)
Dt 2 z z 2
whcre now
au av
P, -(wu) - - (wv) -
ilz z
Pb = {3g(wll).
e = f/IA
Equations (16), (17), and (18) may now be written as
(ul) = - + - - 4(wu) - + 2(wv) --2PB
1 1, [ au av J
3 q az az
(vl) = - +- 2(wu) - - 4(wv)-- 2P
ti 1, [ au av ]
3 q az az
(,.?) = - +- 2(wu) - + 2(wv)-+ 4P
ti ' [ au av J
3 q az az
(uv) - - (uw) - - (vw) -
[ av au]
q az az
(wu) - - ((w
) - C
q") - + {3g(u8.)
3/, [ au J
q az
(wv) - - ((w") - C
q"} - + {3g(vOJ
3/t [ iJU ]
q az
= - -(uw) - - (w6) -
31, [ ae au]
q az a,
(v6) = - - (vw)- - (w6)-
31, [ ae av]
q az az
(w6) - - (w
) - + {3g(99.)
31, [ ae J
q az
A, ae
(8') = - - (w6) -
q az
-(uw) K., aU/az
- (vw) = K., V/az
- ( Ow) = Ku il61z
Ku lq Su
G,. a- - + - P [(au)' (av)']
az az
P ae.
Gu - - {3g-
ti az
then (26), (27), (28), and (29) afler considerable algebra
reduce to
+ SHf J - 3A., B,Gu- 12A. ,A.:rGul = A., (34)
S,.[t +

9A.,A.zGul - SHfi 2A.,

Gu + 9A.t A.,Gul
= A. tO - 3C
) (35)
which are readily solved (however, sce discussion in section
6) for S., and S
as functions of G., and G
From the
definitions (2So)-(25c) and equutions (30)-(33) il may bc
s hown that
(P, + P&)ft o B,(SMGM + Stt<JH) (36)
Looking ahead somewhat, contour plots of S.,, SH, and
(P, + P)le versus G,.. and G
are exbibited in Figure 3 after
the emprica! constaots were determined.
Expressioos for !Utbulence variances may be obtained.
For example, the venical vclocity varianc:e is given by
(W')Iq' =! - .. + 4A ,S
t<Ju (37)
lt is useful for !ater discussion to also write S., and S
1\Jnctions ofG
and (P, + Pb)le. Thus using (36) to eliminate
G"' from (34) and (35), we obtain
- (3A.,S, + J8A.,A. z)G
1 = A.
[1 - 6A, p:...':..---=.+P.::. ]
B, e
= A. , I - 3C, --
P,+ P]
Boundory Layer A.pproximotion: Leve/ 2
A lower-order model is lhe levei 2 model of Mellor ond
Yamtlda [1974], wherein l he material derivative and di.lfu-
sion terms in (16) and (19) are ncglected so that production is
balanced by dissipation. Tbe levei 2 model may be applied
(to fuirly good approximation it would appear) to an entire
boundary layer but also applics rigorously in the limitas the
s urface is approached (and buoyancy elfects vanishl. Bound-

o th-10 1

1 1'}>10
v nJ>.o
, I Vo _,


o o
00 lf!Joo V v
Fig. I. Longitudinal turbulent imensity in a pipe as obtained by
Perry llnd Abel/ (19751. The horizontalline is the estimated outer
asymptote of the inner function (law of the wall coordinates). The
Rcynolds number is based on thc pipe diameter.
ary conditions for the higher order models are obtained from
lhe levei 2 model.
lf the boundary layer approximation is made and applied
to ( 17), ( 18). and (21) after further simplificalion by virtue of
lhe production-dissipation balance, then it is possible to
obtain algebraic relations for S"' and S
as functions of
eithcr Aux Richardson number (see Figure 4)
R1 - Pi/P,
or gradient Richardson number
R. ,. = _ _ G_H = s_._, R,
' (iJUiilz)
+ (iJV/iJzl' G.w S11
The resulting relations are
'YI:..._..,; (..:.Y:.. -: +:-n:..::..) R.!.. r
SH = 3A
I - R1
y, - ! - (2A ,18 ,)
n "'(82/B,) + (6A,/B,)
As seen in (41a) and (4lb), a criticai Richardson number,
where S., = S
= O, is obtained when R
= ')'
+ y,).
After evaluation o f lhe constants A, 8 , and 8
we wiU find
that lhe criticai Richardson number is 0.19.
Other quantities such as (u
and (,J)/Ql, may be ob
tained from (29)-{32) as funclions of Richardson number.
We note t hal (41a) and (41b) may also be obtained from
(33b), (38). and (39). First, sei (P, + PbJie a I in (38) and
(39). Then eliminate q
in (33b) using, again, <f a A
e c
(P, + P.).
As stated previously, lhe constants A
, 8
, A
, 8
, and C1
can be determined wilhout resorl 10 a trial and error process
by appealing to data where turbulent energy production and
dissipation are balanccd. 1bis occurs in the overlap (law of
the wall) region near walls and in homogeneous s hcar Aow
data where diffusion is zero and wherc at some downstream
point in the ftow it happens that aitax c o.
Meltor [1973) had previously deterrnioed that (A
, 8
, A
, C
) = (0.78. 15.0, 0.79, 8.0. 0.056). In this paper we have
exerted more elfort in collecting and interpreting data;
hopcfully, these results will also be useful to other modelers.
Interpretation of wall data to extract large Reynolds
number resuhs requires some care. We wish to correctly
identify the outer (inviscid) asymptote ofthe ' inncr,' viscous
wall functions (law of the wall coordinates) which dcscribe
thc various turbulent ftow properties and which will match
with the inner asymplotes of the 'outer functi ons. lt is lhe
outer functions whicb our model is supposed to s imulate.
The large Reynolds number asymptotic behavior of turbu
lenl wall Oows has been disc ussed by Yajnik (1970) and
Mellor [1972). Perry and Abel/ (1975) have provided a nice
experimental illustration of the matter, which we rcpeat here
in Figure I. lt is secn that we need to determine a quanlity

where y =- yul v == 60 and where
-(wu} u.' = const. In general, data are not as well
resolved in lhe near-wall regions as in Figure l. Therefore
there is likely error in interpretation. However, it is believed
that this kind of error is considerably smaller than lhe
variations among lhe ditferenl data sets rela1ed to mcasure
ment error.
The values for turbulent velocity variances are collected in
Table 1, and those for turbulent tbermal variances in Table 2.
In Table I , aside from Reynolds number and u,IU
, l he
independent data may be lhought to be u'/u, v' lu., and w'/
u,; the remaining two quantities are derived. In Table 2,
- (w(f)I U
0.6, 6'10.6. and P, are independent, whercas the
remaining two variables are derived.
Wc now simplify tbe model equations to lhe conditions
governing the data in Tables I and 2, namely, that the ftow is
two dimensional, production equals diss ipalion, - (uw) "' u.'
= const, - ( Ow) H = const. and buoyancy effects are
negligible. We next stipulatc that
iJUiat. = u,/1
which constitutes a partia! definition of I; that is, any further
prescription of l(z) must be constrained so lha! I = near
solid surfaces if the model is to reproduce the logari1hmic
law of the wall. Thus l he ' master length scale' I is defined so
thal it is coincident with Prandtl's mixing length near solid
s urfaces, although it is here presumed 10 play a much
broader role. A definition of turbulent Prandtl number P, =
K..,!KH aod (30a), (31), (32a) and (35b) yield
aetilz = (Htu,)(P , ![) (43h)
Note lhat the turbulent Prandtl number will vary witb
stratification; in tbis discussion, only lhe neutra) value is
considered. Setting I = where 1< = 0.4, and integrating
(43a) and (43b) will yield laws of lhe wall equations for
velocity and tcmperature, respectively.
Flows wherc production equals dissipation necessariJy
include ftows near solid surfaces. Then q
/ A
= - (uw) iJUI
z, and from (43a) and (12),
<f = B1u/
. - ----

TABLE I. Obscrved Values ofTurbulenl Velocily Variables Whcre Production Js Balanccd by Dissipation
UoXIv 10- u,/Uo v' (uw}fu'w'
Lal(fer (1954) 25.0 0.0346 2.2 1.70 1.00 2.95 0.45
Perry and Aell (1975] 12.8 0.0370 2.12 1.03 0.46
Bremhorst and Bullock (1973] 3.5 0.0398 1.9 1.26 0.42
Laufer (1950] 3. 1 0.0377 1.88 1.19 1.03 2.45 0.52
Laufer I 1950] 6.2 0.0367 1.74 1.14 1.03 2.32 0..55
ComuBellot (1965) 2.3 1.30 1.03 2.8 0.42
Boundary l.Ayer
Klebanoff (l955] 7.4 0.0377 2.02 1.41 1.03 2.66 0.48
So and Mtllor [1973) 2.0 0.042 1.70 1.18 1.00 2.30 0.58
Young e/ a/. (1973)
surface 4.6 0.039 2.25 1.15 0.39
Wave surface 5. 1 0.042 2.4 1.3 /0.32
Wave sulface 8.8 0.049 2.4 1.3 0.32
Homogeneous Shear Flow
Rose (1966) 0.0066 1.66 1.35 1.26 2.48 0.48
Champagne et a/. (1970] 0.0105 1.70 1.29 1.20 2.45 0.49
Primes respresent rms values and u
{uw}. U
is either the centerfine velc.>City or the mainstream velcx:ity. X represents pipe radius R,
half channel width h, or boundary layer thjcknes-s 8.
Under the present constraints, equations (261 reduce to
) a (l - 2y
(.f} : (wl) = "'M2
where y
was previously defioed in (42a) and
I 113
c,=,, --s,
8 2 = - _:..,..:,.:...
Pn H
We must choose values of B., y., Pn. and 8
from Tables I
and 2. It is not a simple choice, aod perhaps lhe principal
value of the tables is to show that some uncertainty exists; if
we bias our choice toward one set of data, that se.t will be
predicted well. but another set may not bc. For example,
there is an obvious diffcrencc bctwecn the wall data and tbe
homogeneous shear flow data.
The firsl choices we have made are 8
= 16.6 and -y
0.222. From (42a) and (44d) we obtain A
= 0.92 aod C
0.08. From (44a}-(44c) !bis yields q!u, = 2.55, u'!u.,. = 1.9,
and u'!u, = w'lu, = 1.2. The fac! that u' and w' are equal is
not supported by lhe data, and as discussed earlier, the
modcl could bc complicated 10 pcrmit u' f- w'. 'fhe choice of
additional terms is not clear, however, and the further choice
of lhe additional constants they would inlroduce would be
less clear.
The choice ofrurbulent PrandU number is quite ambiguous
from Table 2. This quantity has been mcasured by others
besides those listed in Table 2. A J%1 survey [Kestin and
Richardson, 1%1) infers that near a smooth wall, 0.74 < Pn
< 0.92. Gowen and Smith (1%8) obtain 0.8 < Prt < 1.0 for
smooth pipes and 1.0 < P, < I .2 for rough pipes a! a radius
Reynolds number of 20,000. (For large enough Reynolds
number, Pn should not depend on roughness.) Atmospheric
boundary layer dam. to be discussed bclow. indicate p n =
0.74 in the neutra! case. Here we choose P" = 0.80, so that
(44e) yields A, = 0.74. Finally, we choose 8
= 10. 1. Using
(44[), this yields u,
= 3.1, which may be com
pared with the corresponding data in Table 2. Pcrhaps it is
bctter to coosider the correlalions ( - (uw))l(u' w') and
( -(w9))/(w' 9') which can be calculated from the above
constants and equations. We obtain 0.44 and 0.42 for com
TABLE 2. Observcd Values of Turbulenl Thermal Variables Where Produclion is Balanccd by
UoX!v 10-
(wfi)!UoA6 6'/tJ.e prl (w({)/w' 8' u,'( li')/( w/IJ'
Bremhost and
Butlock I 1973] 3.5 0.0020 0.102 0.41 3.0
Boundary Layer
Yuung et ai . (19731
Fiai surface 4.6 0.0015 0.075 0.95 0.46 3.8
Wave surface 5.1 0.0032 0.12 0.55 0.48 2.48
Wave surface 8.8 0.0035 0.11 0.55 2.37
Primes represe.nt rms values. 6.9 is eitber the centerline or the mainstream temperature minus the
walJ temperature. X represents either pipe radius R or boundary layer thickness .
- -
.. , ..

2.010-4 .---,-----.---.-----.---.-----,

' o.
' lliCH(S
Fig. 2a. Data by [1957) for a homogeneous anisotropic
ftow. The dashed lines are faired through thc data points, whereas
lhe solid tines are modcl generated.
parison with data. lt should also be noted that similar-
ity considerations yield B
= 213 for dccaying. bomoge-
oeous temperature and velocity fields [Hinze, 1975. p. 300],
which apparently agrees with the measuremeots of Gibson
and Sclnvarrz [ 1965] . However, these measurements cover a
rather s hort decay history.
Summariziog, we find now that
(A" A2, B2, C,) = (0.92, 16.6, 0.74, 10.1 , 0.08) (45)
which dilfer just a bit from values c ited previously which
were evaluated with less information.
Uberoi' s Experiment
Ali of the results above were obtained independeo! of a
prescription for /(x). Proceeding in the same vein, we tum to
experiments performed by Uberoi [1956, 1957] wbereby
near-homogeneous, a nisotropic turbulence was created in a
wind t unoel by the combination of a turbulcnce-producing
P .lu ..




- 1.0
o o

1.5,1,;--+- --';- ---,,_-+--!;---!;---- +-
-a.o -.a ... 6 .4 .2 o .
Fig. 2b. The sold straght lne is obtained from (4) and (20c) and
the slope is - B
) 3.00 as obtained from (45). The open circlcs
and squarcs are from Uberoi {19571. wbereas lhe crosses are from
Uberoi ( 1956].
grid foUowed by a tunoel cross-sectional area conlraction
resulting in an intal condition where (w
) = (J) > (u
); x
and (tl) are the coordinate and turbulence energy compo-
nent in lhe tunnel flow directon. The governing equation for
(tl), (v
}, (w
) is
a(u.')l2 ( au ) e
u __;_;:....:__ = p _!!_ - -
ax ilx. 3
a == x, y. t
We use Greek subscripts to denote suspension of the sum-
mation convention. Note lhat we assume that the dissipalion
is isotropic; but that is the only assumption necessary 10
reduce the data so that it is very nearly a direc1 1est of
Rotta's bypothesis. Measurements of (u
)(x ) and (v
)(x) from Uberoi [ 1957] are plotted in Figure 2a. Values,.of
and (v
) = (w
) are obtained at each data point localion
but from the smoo1hed curve passed t hrough the data by
Uberoi (dashed line}. From these data and the above equa-
tion it is possible to obtain values of (p auli!x), (p i!vli!x), and
e as functions ofx. Thus in Figure 2a we plot the ratios (p i!ul
ax)/( e/3) and (p iJvliJx)l(t/3) (to eliminare length scales) as a
func1on of (u
- q
13) and (u
- q
/3), res pec-
tively, to compare wit h
dU0 )/(F}J) = _ .!!L (u.l) - <f/3
p iix. 6A, q
which may be obtained from (4) and (5). The results are
presented in Figure 2b as circles and squares along with the
straight-line plot of (46} where, usiog the constants previous-
ly obtained in (45), we determine that B
) = 3.00.
We oow note that lhe solid line in Figure 2a is a model
predictioo which embodies (46) but also requires a length
scale prescription. While in this se,ctioo we do not yet wish
to emphasize the latter, t he solid curveis a useful reference
to emphasize that small error in data as plotted in Figure 2b
would re,sult in apparent large error in Figure 2a. particularly
near the origin of Figure 2b; this region of small anisotropy is
where we would most likely expect Rotta's hypothesis to be
On the other hand, the readcr wiU observe data (crosses)
for larger anisot ropy which departs significantly from Roua'
hypothesis. The departures seem overly abrupt. Neverthe-
less, the abscissa values. - 1.0 and 0.5. in Figure 2a repre-
sent the limits for two-dimensional, axisymmetric turbulence
where (J) = O and (u
) = (w'). AI these limits one should
anticipate that Rotta' s hypothesis might fail. In fact. it
appears from Figure 2a that it fails wbcn tbe smallesl
component is (J)/(q') 0.12.
The grid Reynolds number for the Uberoi experimcnts
was 12.000 and tberefore not very large. Corroborative
experiments are not available. Nevertheless , the princ.ipal
results suppon Rotta' s hypothesis, suppon the choice of
emprica! constants (obtained from cntirely independeo! data
where production and dissipation balance), and place a limit
on the range of applicability of Rotta's hypothesis.
The model has not yet been described completely, but
enough has becn dctermined to demonstrate the surprising
capabilily of t he model to predict the stabilization of t urbu-
lent lields by densit y stratification and curvature. Tbe data
obtained by So mui Mellor [ 1973] werc a c lear demonstration
Fig. 3. The stability functions S.,(GH, G,.) and SH(GH, G..,). Thc heavy solid I ines are contOU<$ of S..,, whereas the
dashed tines are contours of Su. Thc lighter solid lines are comours of(P., + P,)Jr. . One could also draw Unes ofconstant
R;= G,/GM, which are mdiallines on this dagram. The shaded portion is where ( w
:s 0. 12 .
of the fact that stabilizing curvature could literally extinguish
turbulence. The effect oi' curvature as predicted by lhe
model has been described by So [1975] and Me/lor [ l975] and
was irnportant corroboration that the model could predic-
tively extend far beyond the neutra! data on which it is
based. However. lhis paper is meant to emphasize geophysi
cal tluid problems. for which we will shonly tum to tbe near-
surface, atmospheric boundary layer data of Businger et ai.
Using lhe constants in (45), the levci2J functions, SM(G..,,
G") and S,(O,.., Off), are plotted in Figure 3 from (34) and
(35). Contours of(P, + are aJso plotted. As the limit G"
= 0.0338 is approached, we fi nd tbat SH -> "' In principie
this means that gradients such as 1!6/az will be mixed out
wilh an indefinitcly large value of KH such that GH cannot
exceed lhe limiting value. On this graph onc can identify
probable limits to the model's validity. The upper right
portion of Figure 3 is a region where (w'}/q
< 0.12,
representing (approximately) a region where Rotta's hypoth
esis is probably invalid. Dickey and Mellor [1980] have found
experimcntally that for stratified homogencous, decaying
turbulcnce the point G.., = O, - 0.4 is rcached where
the initial turbulcnce is converted into an eosemble of
internai waves which decay very slowly (or, probably, oot at
ali in tbe limit of largc Rcynolds number). We therefore
know that (5) is no longer valid, and this is probably true for
the olher modeling assumptions. Thi s may not be a serious
deficiency, since the model s hould produce very low turbu
lence leveis and linle mixing for the aforementioned values
of G.., and 0

Figure 4 (correspooding to the cross section given by (Pb
+ P.)t. = I of Figure 3) is a plot of lhe levei 2 functions
S.!.Rf) and S,(R
) from (41) and (42) . A criticai Richardson
number R1"" R; = 0.19 is obtained beyond which S.,= S,-
For lhe purpose of direct comparison of data with model
prcdiction we apply the levei 2 model where ali material
derivativo aod diffusion terms in lhe turbulent moment
equations have been neglected. This should be a valid
simplification near surfaces, at least for neutra! and stable
flows. For near-surface ftows we also assume I = kt . The
data are cast in Monin-Obukhov similarity forrn. <I>M(t),
<J>,(t), where
KZ au 3 -114
</>., = --= [B,(I - R
)S.,] (47a)
Ur dz
t<ZU, <l6 4 - 114
tf>H =- - = [B,(l - Rt)SH IS..,] (47b)
H <l t
t .. u/I(Kg{JH) = q,..,R1 (47c)
and where u.' = ((uwf + (vw)
n and H = - (w8) near the
Figures 5a and Sb compare the model results with the data
of Businger et ai. I 1971). Note that the point R r= Riu = O. 19
in Figure 4 is mapped ioto the straightlines in Figures 5a and
5b as t --> "' Note also that lhe prediction of R
r, does not
depend on l(z).


R f
Fig. 4. Tbe stability functions SM and SH as a function of flux
Richardson number corresponding tO the condition (P$ + Pb)IE = I
of Figure 3.


.. .,

o 7
0-'-' o.e

0.4 .


f .
-0.10 -o.oo o o.os 0.10

, .......

- 2.S - 21) I.S 1.0 0.5 O 0.5 1.0 1.5 2.0

Fig. 5a. Comparison of lhe veloeily profile data of Busnger et
a/. [19711 with predicted values (solid curve). lnsert is de1ail near t

The fact that the basic model has the predictive capability
of extrapolating neutra! data into stratified regimes was
probably the most important finding [Mel/or, 1973) in the
development of the model.
Lewellen and Teske [1973) have also shown a favorable
comparison of their model with these data. They included
the di.ffusion term and did gct better agreement than we did
for q,., in the unstable region where diffusion is liable to be
importao!. However, they had to inseri a spccific Ricbard
son number parameterization into their model to obtain the
correct criticai Richardson number. Gambo [1978) also
obtained improvement in q,., in the unstable region by
including buoyancy terrns in (4) in a maoner described at tbe
end of section 2; bowever, agreement witb lhe stable data for
botb q,.., and <i>H was not as good as tbat sbown here.
It s hould be noted that the Businger data were also
evaluated to yield a von Karman constao! K of 0.35. some-
what less than the more commonl)' accepted value of 0.40.
Wieringa [1980) discusses the issue, reevaluates l he data
taking into account tower interference, and argues in favor
of the value 0.40. The later value is used throughout this
We havc postponcd considcration of the equation for the
master length scale I so that the other elements of the model
were first justified on the basis of neutra! data and a direct
test of the model's predictive power, that is, the prediction
of turbulent stabilization in a density-stratified fl ow. The
factor 5
in (24) must also now bc determined.
Following Rotta (1951), we use the integral of the two-
point correlation function to supply an equation for tbe
master length scale. The closure assumptions are complicat-
cd, and we consider the result less convincing than the
previous assumptions and more likely to be amended in the
future. Tbe version we bave used for some time (see Mellor
and Herring 11973] for a general discussion) is
.!!._(qll) - !_[q/S1 _!... (qll)] = lE,[P, + Pb)
-f. {1 +

For neutra! , homogcncous, dccaying grid turbulcnce where I
is much smaller than L, (48) along with (26) predicts the
initial pcriod dccay Jaw, q
o: t-
, E
, and E
are empirical
constants to be determined. P mjght be preceded by anoth-
er constao! if data can be found to unambiguously support a
value othcr than unity.
L is supposed to be a measure of the distance awa)' from
tbe wall as is spccified according to
C '(r) = _1 ff dA(ro)
2rr [r - rol'
which is similar to ide as offercd by Shir [ 1973) and Launder
et a/. [1975]. Here ris any point in lhe flui d domain bouoded
by solid wall at r
; dA(r
) is an elemental wall area. For a
boundary layer flow near an infinite plane wall , L z; for
channcl flow, L-
+ (2h - z)-
, where 2h is thc
distance separating the channel walls. lt can, by the way, be
sbown that a third term on tbe rigbt of (48) , bere represented





0.2 5
-O. IO 0.05 o
0 .10 .




-2.5 - 2.0 I.S 1.0 -0.5 O 1.0 2.0
Fig. 5b. Compar ison of the temperature profile da<a of Businger
er ttl. [19711 with predicted values (solid curve). Inseri is detail near

by (c/181)&; (I/KL)
, is absolutely necessary, but the one
cbosen bere is one o f severa! altemativcs IN g and Spa/ding,
1972; IVo/fstein, 1970; Mellor and Hcrring, 1973; Lewelle11 et
a/., 1976: Roua, 1973). Ali one can say in the present case is
that, as we will see, it works well.
While onc cannot assen great conftdence in (48), we prefer
it ralher than the difl'erential equation for dissipation [Da/y
and Harlow, 1970: Ha,Yalic and Laundu, 1972: Lumley and
KJrajehNouri, 1974). The dissipation transpon equation is
an equation for lhe curvature of lhe two-point , velocity
correlation function as the scparation dist.ance approaches
zero (for large Reynolds number, the transpon tenns, per se.
are negligible, as sbown by Tennekes and Lunoley I 1972)).
Ahematively, it is an equation for the integral oflhe spectral
density fuoction afer multiplication by lhe square of t he
wave number. tbus weighting the integral so thatlarge wave
number and smallscale turbulenee are emphasized. Thus it
seems fundamentally wrong to us to use an cquation which
describes the smaU-scale turbulence to determine the re-
quired turbulent macroscale. Operationally, however, afler
some terms are modeled, the dissipation transpor! equation
is a special case of a more general lcngth scale equation
(Mellor and Hurng, 1973; Lewelle11 et a/., 1976].
In subsequent discuss ions we wiU mention some calcula-
tions using the levei 2 model and an algebraic expression of
the form
I a lo __::::__
kz + /
r lzJq da
lo ...
q dz
in place of(48). For boundary layers this works well, but it is
limited to boundary layers , and the emprica! constao! ,.
would depend on the type of layer, for example. a boundary
layer as in Figure 7 or an Ekman lnyer. Some studies actually
simplify (50) further so that I /
This will not produce a
logarilhmic velocity behavior near surfaces. Howcvcr, for
lhe case of an ocean surface mixed layer lhe additional
simplilication does not seem to impact mixed layer deepen-
ing or temperature.
Boundary Conditions
We have post.poned stipulation of boundary conditions
until this time when the complete set of model diO'erential
equations available.
For the mean velocity and temperature boundary condi
tion near a surface at z l<J, one either specilies stress and
heat flux ,
- (wu,)(_., , y, l<J) 'bl - (qiS., aU/az)
- (w8)(x, y, l<J) H - (qiSu iJEl/iJz)
(5 I}
or. near solid surfaces at rest, the numerical solution is
matched to
Tw (z - l<l )
U/..x, y, z) --In
KUr t. MS
z-+ lo> (S3a)
HP.., (z - l<l)
e(x, y, z) - 9(x, y, to) - - In
KU, tns


- .OI
/ t ,
.2 .4 .6 .8 1.0
Fig. 6. Channel ftow, Comparison of prcdicled mean velocity
and tuobulcnt intensity (solid tines) with data by Lat4frr [19SO] and
by Hanjalic as rccorded by Launder <1 a/. [197S). The circle s are
Lauffer's data.
wbere u/ = (To/)
, i = x, y, and and zus are lhe
roughness parameters; for s mooth surfaces, exp
( -4.9K), whereas z
sfzMs is a function of number.
Asymptotieally close to surfaces, ali equations for the
turbulence variances reduce to the levei 2 algebraic equa-
tions. Furthermore, the buoyancy production tenns vanish.
Thus for the levei model we have from (44a)
q'(x, y, l<J) = 8
l/J u.' (54)
to whicb we add
q"l(x, = o (55)
on solid surfaces. For free-stream conditions the mean
velocity and temperarure are specified. l f the free-s tream
turbulence is known, that is specified. Otherwise. a very
s maU value is used. Solutions are quite insensitive to frec-
stream values of q
For tbe levei 4 and levei 3 models, additional boundary
conditions are obtained from (26H29) after replacing P

O, U/iJz = -(uw)I(KU.,i.), and iJV/iJz - (vw)I(Ku,z); some
rcsults are given in (44b) and (44c).
Newral Channel Flow and Boundary Layer Flow
Calculations were made for channcl flow and a constant
pressurc boundary layer to assess s . S
and E
, E
became apparent that one s hould set s, = s.: otherwisc thc
behavior of I in the center ofthe c hannel or at the edge o f the
boundary layer was unrealistic. To insure that I - Kl as z -+
O, it may bc shown that E
= K'B
+ E, - I. The values of
--- -



.8 0,..0
. 6

.2 '




..... -------
.01 .L
........ o

2 4 .8 lO
Fig. 7. Boundary layer. Comparison ofthe prcdictcd mean vcloci
ty and turbulcnt intcnsity with data by Kleba,off[l955).
s. and E
are then chosen to optimize agreement belween
model results and data at lhe center and outer edgc of lhe
channel and boundary layer, respeclively. An agreeable
result was that lhe values s. = 0.2 and E
= 1.8 were
optimum for botb llows. Thus we obtain (S., S
, E
, E
) =
(0.2, 0.2, 1.8, 1.33) and the predictions shown in Figures 6
and 7. In boi h cases there is near-perfect agrecment with
measured values of - (uw)(z) and q(z). For the channel flow
case lhe observed and calculaled Reynolds stress distribu
tions are linear and are not shown.
The outer freestream turbulence levei for the boundary
layer ftow has been set at q/u, Q 0.07 as cstimated from
Klebanoff [1955}, but this may be a bit low. Two boundary
conditions for q
1 have been specilied and correspond to the
calculated solid and dashed !ines in Figure 7.
The separate components u', v', and w' are not shown,
since they do not enler in to the determination of E, . Agree
mcnt with data is quite good in this respect except that we
predict u' ; w' as discussed earlier. Away from walls this is
- ---
in agreement with lhe data, but near waUs there is some
disagreement in accordance with Table L
In unpublished work we have also calculated circular duct
flow after (26), (47), and (48) are cast in cylindrical coordi
nate form, and these calculations compared favorably witb
lhe duct flow data of Lat4 [1954) .
This is perhaps an appropriate placc to record the fact that
we have had occasional difficulty with the levei 2! model; for
some model simulations a discontinuity in velocity could
develop and persist. We will not detail lhe nature of lhe
problem here except to note that its occurreoce depends on
lhe specifics of linite dilferencing. Thus it occurs when
Reynolds stresses and mean velocities are staggered with
respcct to each orhcr but not when these variables are
located at tbe same grid points. It further occurs when (P, +
P)le attains large values (unrealistica!Jy greater than 2) as,
for example, when wind stress is impulsively applied tO an
ocean surface ioilia!Jy at rest with initially zero layer thick-
oess. A modilied levei model [Worth.em and Mellor, 1980)
obviated the problem but required a complicated ileration .
More reccntly, however, we have reverted to the original
schemc (cquations (26H29)) after we learned that the prob-
lem can be avoided by constraining the domain of depen-
dence of the S,;(GM, GH) functions so as to exclude the
regions where (w
:s 0. 12 (lhe s haded region in Figure 3)
and where, according to the discussion in section 4, Uberoi' s
data present reason to believe that Rolta' s assumption fails.
Of course, one might accordingly revise Rona' s hypothesis,
but that step might better await further corroborative data. It
would also complicate lhe model and, it is belicved, unncc-
essarily so for most practical problems.
The constraint we use on (34) and (35) is G n :s 0.033 and
G., :s 0.825 - 25.0 Gn.
The models as described thus far, mainly in lhe context of
one-dimensional simulations, have now matured sufficiently
so that these form the basis for other one-dimensional
investigations to compare model predictions with laboratory
ftows and to generate new information oo boundary layer
responses. The closure models have also been incorporated
in to larger. three-dimensional atmospheric and oceano-
graphic numerical models. The following discussion will
highlight some of these applications.
Free Convection
With no alteration in l he model, levei 2! calculations are
performed to compare with the free-convection experiment
of Wills and Deardorff [1974} wherein a beat flux was
imposed at the bollom surface of a lank of water after a
linear temperature gradient had been established in the tank.
Since there is no shear production, the Richardson number
changes abruptly from +<>o to -oo at z a thc inversion
Tbe calculated tempcrature and heat ftux very nearly
overlie the data in Figure 8 and havc not been ploued. A
possible exception is that the smaU, ncgative ftux oversboot
near tbe inversion beigh.t, z = z
, is underestimated by a
factor of 3 or 4. Note that the experimental heat tlux was
determined by integrating the lemperature tendency and
might be subject to error.
The predictions of (w
) and ( '-) i.n Figures 9 and lO agree
well witb the data, whereas (u
) does oot. Furthermore, free-
Meu..oR ANO YAMAOA: TuRBULENCE CLosuu Moou 863

60 13



'fj o
o 6 '
o 0.2 ...


Fig. 8. Mean temperature and heat ftux profiles from the laboratory cxperimcnt of Willis and Deardorfffl9'74].
CaJculatcd profilcs wcre nearly concident wich tbcse data. Opcn data poinlS are atmospheric aircrafl measurements in
conditions thought to be similar to those of the laboratory cxperiment.
convection scaling 'laws' wbere (wl} (gfJHdl) and (IP}
(g{JHz)- VJ as z-+ O are seen to prevail for these quantilies
but not for as z --+ O. ln other words, both Lhe scaling
laws and the model fail to represent (u
) well near the
surface. As discussed by Sun and Ogura (1980], a rational-
ization for this finding is that (u
) is dependent on both t and
z, thc ovcrall height of the boundary laycr.
lt should be noted that the sharp minimum in (8'} near t =
would undoubtedly be modificd by a levei 3 model
calculation which includes diffus ion of ( 8').
Readers might wish to compare lhe results of Figur es 8. 9,
and lO with similar results obtained by wel/en e/ al. [1976],
Zeman ond Lumlq [1976), Sun and Oguro [ 1980], and Andre
ti a/. [1976b] . The lauer solved tbe aforementioned difficulty
wilh lhe (u
) problem by setting their length scale to a
constant proponion to t
and by matching their lower
boundary condit ions to the data.
1t should be noted that Willis ond Deardorff [ 1974] also
obtained data for (wq
)/2. If (pw) were zero, one could
compare t hose data witb the diffusion represented by ql S/
(qln.)laz in (24) (even though we formally neglected pressure
diffusion. it must, as discussed in seclion 2. be considered a
pan or lhe diffusion term (24) if observations indicate lha! pw



02 06


03 OA

'I' 0). in which case tbe modcl appears to underestirnate lhe
measured (wt/}.
The observational determination of net turbulent diffusion
imponantly depends on lhe measured behavior of the dissi-
pation. Since the buoyancy production must decrease linear-
ly, or nearly so, with height, it may bc inferred tbat net
difl'usion is small if dissipation behaves s imilarly.
Unschow [1970] and Kuklwrfls and Tsvang [1977) ob-
served dissipation rale profiles that were almost uniform
with height in lhe middle regions of the convective mixed
layer. Also, tbe explicit turbulent simulation model of Deor
datff[l974) produced pressure diffusion that was small and
velocity difl'usion tbat was comparable to dissipation. On tbe
other hand, nearly linear dissipation rate profiles wcre
observed by Yokoyama era/. [ 19771 over land and by Penne/
and LeMone [1974) over the tropical ocean.
Also. Caughey and Wyngaard (1979) have, in an atmo-
spheric convection layer, measured buoyancy production.
dissipation, and turbulent velocity diffusion and by differ-
ence determined that pressure diffusion is of opposite sign to
velocity diffusion and, in fact, lhey tend to caocel.
When tbe observational record i.s clarilied. lhe model
might be suitably modified.


Fia. 9. Horizontal and verticalturbulent energy compon<nts (solid symbols) by Willi.J and Dnrdorff[l974]. Open
data symbols are airc,raft measureme:ots, solid tines are calculated, and w.


1.0 -------------------------- ------ --- --

o .

O. A ~


0.1 lO
Fg. 10. Temperature variance (solid symbols) by Willis and
Deardorff (1974]. Open data symbols are aircraft measurements,
solid lincs are calculated. and e. fllw .
It is reasonablc to cxpcct tbat s. in (26) is Richardsoo
number dependeo! as are, analogously, SM and SH We
have, in fact, computed the Wills and Deardorjf [1974)
experiment with s. = const = 0.2 and with s. o: SM (where
lhe constao! of proportionality is adjusled so that s. = 0.2 in
lhe neutra! flow limit). The difference in the calculated
results were not large.

l--Dor 33--+----Dor 34-- -4-IDoy 35--1
Forced Convection
The experiment whereby a shear stress was impulsively
applied to the top surface of stable, salinity-s tratilied water,
tbcreby mixing Lhe top layer. has been perfonned by Karo
arrd Phillps [1969]. Qualitatively, the mixing process is
inhibited by Lhe stratification such that an abrupl density
change occurs across the interface separating strongly lurbu-
lent and quiescenl fluid. Using the simpler, level2 version of
lhe model (ao algebraic length scale recipe aod neglect of the
turbulent kinetic eoergy tendency and diffusion terms) ,
Mellor and Durbin [1975) predicled these data quite weU,
and we expect that prediction will prevail when this latest
versioo of the model is applied. Other examples of mixed
layer dynamics are explored and a favorable comparison
with ocean observations in the North Pacific is included in
lhe paper by Mellor and Durbin. The levei 2! model has also
been favorably compared to the two-level experiment of
Kamha era/. [1977) by Mellor and Strub [1980].
The Wangara Dma Set
The present model has been compared with atmospheric
boundary layer data of C/arke er ai. [1971), whicb are called
the 'Wangara data.' The temperature variables in the previ
ous equations must now be interpreted as virtual potential
Comparison of simulations and observations by Yami:zda
and Mellor [1975) are shown in Figures 11- 13. The calcula-
' I
: .
. /1
i '
. ~
i /j
I f
I ' I
I I '
1!2 I 18
/ ..
' .3 !
rl i
' .,
I . ;
, . ~ . ...
I :
' .:
' J
/ i
' I
' .
: I
i :
i. I
j> .
' .,.,..- 11 !

o 5 lO
15 20 5 10 15 20
O 5 10 15 2C
0900, doy 33 to 0300. doy 3A
5 lO 15 20
0900. dor 34 to 0300, doy 35
Fig. I I. Observed and calcutated atmosphcric boundary la.ycr and vertical and 1emporal variations of mean virtual
potential tcmpcraturc - 273K. Units are degrccs Kclvin.
- -

12 18 o 6 12 18
1-----Doy 3-0----;-Doy --1

' . I
J \
, I !
{ \ /
\ \ /
J 1
I "
' I I
I '
I . ' I I
(t. l,
10 ..1

I . i (
\ i
I .
, : I
I \ .
, I
/ ..
f ), "
<"'; ,.- i.
,, I
.. '
(m UI C
1100, doy 33to 0600, doy 3A
} I

i i\
, I \ \ .
A .
f i
' .
' .
..__ ...... -. --=;-
15 10 ..1 o
(m s.c'
't> i
I )\

j., I

"!......,_.,_ \,
-....... .
1S 10 ..1 O
lmsoc I)
1200, doy :).0 lo 0600, doy 35
Fig. 12. Observed and ealculated atmospllerie boundary layer and vert.ieal and lempc>ral varatioos of Lhe <lllilwarcl
mean wind component. Units are meters pe:r second.
tions shown in tbese figures used the algebraic lcngth scale
equation (50), eod of section 6, but we expect the use of (48)
would produce little change in lhe resuh. The calculatioos
IIS$Urne horizontal homogeneity so that altitude and time are
thc only independent variables.
A feature o f the velocity ficld prediction is the appearance
of ' the nocmrnal wind maximum' around midnight and near
z - 200 m. The m<\ior effcct of tbe diurna! surface heating
cyclc can bcst be seen in Figure 14. where we show contours
of calculated turbulent kinetic energy. Further details are
provided by Yamada and (19751.
Pollutant Dispusion
Model equations devcloped for temperature may also be
applied to any scalar. such as a chemically inert pollutant.
Using thc horizontally homogencous wind field generated
for the Wangara simulation. Yamada (1977) has made a
three-dimensional calculation o f the dispersion of a poUutant
point source located at various distances from lhe ground.
Figure 15 illustratcs the dispersion ofa source located at z =
40 m during lhe earty moming and afternoon hours. Tbe
moming, low-level inversion confines pollutants to ncar
surface a.ltitudes, whereas vertical mixing in thc aftemoon is
vigorous. as would be deduced from Figure 14. Olher
calculations, which include an assessment of grouod levei
conccntration as influenced by source beight. will bc found
in Yamada's paper.
One finding of inlcresl is that the effect of lateral diffusion
terms in lhe mean concentration equation is negligible ex-
cept very close to lhe source. Lateral dispersion is dominat
ed by vertical variability of mean wind specd and direction.
Horirontal mean advective dispersion creates mean vertical
concentration gradients which are subsequenUy mixed
through vertical ditfusion.
Dobmy [1979] has also constructed a twodimensional
(vertical plane) model for pollutant transport based on the
levei and 3 models.
Tw&- and ThreeDimensional Flow With Orograplry
As a firsl step toward I'Clllist ic treatment of teiTIIin elfects,
the airflow over single and double Gaussian mountains was
investigated [Yamada, t978b] . The governing equntions are
transformed in1o a temlin-following coordinate system in
order to simplify lhe surface houndary conditions. Figure
16a shows the horizontal wind veclors 1000 m ahove the
surface ofa Gaussian mountain 500 m high. Accelcration and
dcceleration of the horizontal wind speeds are seen on lhe
lee and upwind sides, respectively. Additionally, lhe airflow
diverges strongly as it approaches thc mountain and con
vergcs in lhe lee of the mountain. The covergence and
divergence in the horizontal wind fields result in vertical
motions computed from the continuity equation. A maxi
mum downward motion of2 m .- is obtained approximately
3.5 km above the downwind slope of tbe mountain;
Ih r)

l O
( l



.s o

.. ,



lO .s
,.,, \

' . \
. \
' I
" ' '
, l'li

..... -., \

.to .5 o
l} (msec .. J
f--Ooy 331- -l----Oa y Jl - ---1-0ay 35--l 1200, clav 33 to OOO, dar Jl 1200, day J.4 10 OOO, clay 35
Fia. 13. Ot>served and calculated atmOSjlhaiC boundary laytr and vertical and ternpo<al variations of tbe northward
mcan wind componen1.
maximum upward motion is only 0.2 m s - o, a.s upward
motion occ urs over a grca1cr volume. The polentiallempera
lure in a venical plane lhrough the diagonnl AB in Figure
16a is shown in Figure 16b. The p01en1ial temperature is
increased by S"C in the Ice of lhe peak to subsidcnce
occurring on lhe 1ee slope or lhe moumain. See Yamada
[1978bl for further discussion. Recen1ly. more rea1istic to-
pography is included in lhe 1wo-dimensional [Yamada,
. : .. : .::;:::
. : ,_: "":'
,_......Day 33- -+---Day Jl - - - 1-IDay 35-t
Fia. 14. Time and space varialion of compuled q' Uwice tbe
turbu_lent kinetic eoergy); unit.s are squnrc meters per second. The
stipplcd IIKOS indicate rcgions wberc 10 > < q' < 10> m
$ "

1982b) and three-dimensional [Yamoda, 1981] models. and
lhe rcsults are compared wilh obscrvations. Tree canopies
are paramcterized in a relalively simple manner, but effccts
of bmh solar radialion and drag force are properly consid-
ered [Yamado, 1982a) .
f}'ecu of Water Vapor
Exactly the same equations derived in the previous sec-
tions are applicable for a moist atmosphere as longas phase
cbanges do no1 occur and provided lhat \irlual potental
temperature . defincd as
. = (I + 0.61 Q.,} (56)
is cverywhere substituted for . In (56), Q. is thc mixing of water vapor, and a tilde indica1es an inslantaneous
value. In ali previous equalions lhe mcao and the fluc!Uation
of the virtual potential temperature, e. and o should
replace the mean and ft uctuation of potential temperalure. O
and 8. In addilion. conservation equations of lhe mean and
turbulence fluxes of waler vapor can be derived assuming
closure assumptions identical to lhose for the polen!lal
temperature are needed 10 oblain absolute tempera!Ure or
potentialtcmpcraturc. The resu1ting equations for lhe waler
vapor are also identical in form to those for the polcntial
lemperature. Thus , for example, replacing e ande by Q. and
q., respectively, in (23), (28), (29), and (31) yields lhe
required equations.
Burle I 1977) used the levei 3 mode1 to study temporal

Fig. 15. (a ) Simulations o(pollutant dispenion from a source at
40 m above ground leveJ. The mean wiod, temperature. and t u r b u ~
lence field are thc same as in Figures 11-14 at 1500 hours on day 33.
The box represcnts a 40 km by 40 km by 1200 m <lomain. (b) Samc"'
Figure ISa except the time is 0600 on day 34.
variations in lhe moist atmaspheric boondary layer an<l to
investigate mechanisms to explain observations of layers
which are well mixed thoroughly but far from well-mixed in
terrns of moisture.
!f phase changcs occur. however, it becomes rather diffi.
cult to solve the cquations for the virtual potential tempera-
ture and watcr vapor. since thcy are not conserved and
suitable stipulations of sourec (sink) terms are not known.
Tberefore we bave taken a different approach. One obvious-
ly conserved quantity, even when phase chaoges occur, is
lhe total water content Q defined as
(57 a)
wbere Q
is lhe liquid water. Another conserved quaotity
uscd here is the liquid water potential temperature [Betts,
1973] ,, defined as
. er...
e, ---Q,
T C,
where L. and C, are lhe latent heat of evaporation and tbe
specific heat of dry air at constant pressure, rcspectively.
Therefore the conservation equations for the potential tem
perature as in (23) are replaced by identical equations for the
liquid water potenlial temperature and for total water.
In order to recover the watcr vapor and absolute tempera-
ture fields, whlch are nccessary for determining buoyancy
/// // ./// /
Fia. 16a. Hori.zootal wind vectors at 1000 m above th<: surface
of a mountain in Figure 25. Tenain is contoured by dashe<l line.s with
an increment of L 50 m. 'rhe lowcs1 contour ls z. = 20 m.
effccts on the turbulence and for computing long-wavc
radiation cooling of the atmosphere, we foUow Sonrmuia
and Deardorjf (19n] nnd Mellor (1977) and assume a joint,
Gaussian probability distribution for
and Qw. First, how-
ever, we will derive equations for the various turbulence
momentS. The derivation of equations in detail has been
already given by Yamoda (1 978a) or Yamada Ofld Mellor
(1979]. Most of the equations for the turbulent moments are
similar to those discussed in the prcvious sections. For
example, (26) and (27) rcmain the same. Equation (28) is
virtually the same except that 8 is replaced by 8
so tbat
(u8,) E 312/q r - (uw) ii9/iit - (w8,) Wlzl
(uS,) e 3/,!q ( -(IIW) IJe/ iiz - (wS,) iiV/ii z) (58)
------ ---1 Jl6
.--------1 J12
- - ~ - - - - - ; J08
---- ----1 304
-----1 296
[)tS T ANC.r AL[)NG " - B (I< !.I)
Fig. t6b. Distribution of the potcntial tempcrature (kelvins) in a
vertical plane through the diagonal AB in Fi$ure 16a.
- - --
and (29) becomes
A, ae,
(11,) = - - (w8,)-
q az
ln addit.ion, equations for lhe fluxes and variancc of thc total
watcr may be derived, resulting in
(uqw) 3/,Jq l- (uw) i!Q..Jiiz - (wqw) i!Uiiizl
(uqw) - 3/,lq ( - (uw) iiQ.Iiiz - (wqw) W/iiz) (60)
(wq . ) - 31:/q [ -(w
} iiQJaz + fJg (8.,qw)]
" A: iiQw
(qw-1 - - (wqw)-
q az
Finally, the equacion for the cross correlation becwecn 111 and
q. is
iiQw ae, (!W.J
O -(wlll) - - (wqw) - - 2q (62)
ii z iiz A,
By using (59), (61) , and (62) we reduce to thc levei
model. Rowevcr, lhe following eondensation physics can be
used in tbe levei 4 or 3 models. In particular, Burk [ 1980,
1982( has applied the levei 3 model with condensation
physics in a s tudy of lhe turbulence structure parametcrs
which provides useful information in interpreting acoustic
optical or microwave propagation measuremcnts.
Unlike thc previous cases where phase- changes do not
oecur, lhe revised set of equations h ave more unknown
terms than the number of equations, bccause of various
eorrclations involving thc fluctuation of virtual potcntial
tempcraturc. Thcrcforc additional cxpressions to relate
thcsc terrns (uJO,,). (11. 11,), and (ll.,q...) to known terrns are
necded. Rcaders are again referred to ramada [1978a) or
Yamada and Mllor [1979) for details of derivations. Only a
brief discussion and the final results are given here. Tbe
virtual temperature relation which includcs liquid water is,
instead of (56), . e (I + 0.61 Qw - 1.61 (!
}. Togethcr
with (57b) we obtain
. - o + o.61 (! .. - 1.61 '>(, + (63>
T C,
from which the mean is readi1y extracted to give
e. - ( I + o.61 Q., - t.61 Q,>( e, + t Q) <64>
lffrom (63) and (64) we extract tbe ftuctuating part and take
moments, we obtain
/Kli;O.) a fJr(u,BIJ + f!w(Uftw) + /Muj(I)
P<lllll.) ; /J-r(llll) + f!w(!W.J + (M&,q,}
p(qw8J {J-r(q.,8,} + fJw(q., ') + /M,Q .t/t)
wherc higherorder moments are neglected and
fJ-r '" fJ(I + 0.61 Q., - 1.61 Q1)
fJ1 .. fJ (I + 0.61 Qw- 3.22 Q,)-- - 1.61 9,
e Lu ]
fJ. . o .61 e, + - - Q,

e L. )
T Cp
- - ----
We are tberefore rcquired to determine (uj(/
), (9/Q,), and
(QwQt) to close the equation set. For the latter two quantities,
as detailed by Sommnia and [ 1977) and
[ 1977], one way is to assume a binorrnal disuibutioo for 8,
and Qw. which we signify by G(O,, Qw): we also assume 'fast'
eondensation pbysics aceording to Q
(Q. - Q,)H(Q., -
0.,), where H is the Heaviside opcrator and (!, is the
saturation spccific bumidity value.
Moments are then forrned such that
R ; J: rw H(()_., - Q,)G((I1, q.,) d91 dq.,
is lhe cloud fract ion and
Q, e [ [ {Qw - Q,)H(Qw - QJG(IJ,, Qw) dlll dq.,
is lhe mean liquid watcr. For the turbulent momeots,
(4>Q,} = f_ r_ 4l<lt H(Q., - Q.)G(B,, Qw) de, dq.,
where = 9, or q.,. H is zero when Qw < Q, and unity when
Q., > Q,. in which case Q
; Q., - Q, . In the lattcr case we
expand (!, in a Taylor scries so that it is equated to
saturation specific humidily at thc mcan liquid potential
temperature Q,, plus terrns proporcional to 0
and Q
result is that the integrais may be worked oul and yield
where = 6
or q.,. The parameters a, b, and u, are
a "' I + Q,,T Cp
b o(TIII)Q,,,T
; (a'(q.,
) - 2ab(q,.ll,) + b
.r is obtai.ned from the Clasius-Ciapcyron equation and is
thc derivative of lhe saturation s pecific humidity with re-
spect to temperature and evalunted at the mean liquid
By assuming a trinorrnal probability distriboton whcb
includes as arguments u
as well as 8
and q.,, it may be
shown that (69) is also valid wherc tb u

With the use of the flux equations and (59), (61), and (62),
equation (70c) may be written in lhe following, computation-
ally convenient form:
- ---- -- - -=

Now successively inserting f/> ; u
, 0
and qw into (69)
provides necessary moments for insertion into (65aH65c).
The results for (u
9.), (9
0.), and (q O.) are thcn used in (58),
(60). and (62). The expression for ( wO.,) for use in (25b)
(3(w6,) = (/3r - {J,R' b){w6,) + ({3,. + {J,R' a)(wq) (72)
Finally, these results may be cas t in the forrn of (30a). (30b),
and (31) where, however, we write (0
w) and e, in place of
( Ow) and e and add a flux equation for (qw),
- (wqw} = Ku QJi!z (13}
Equations (33a), (34), (35), and (36) remain unehaoged
excepl for lhe fact that G" rnust be redefined so that
Cu = -(f'lq
) {3g'
where t is defined by
were discussed in detail by Yamada and Mt llor [1979}. The
BOMEX was cooducted over and in the tropical ocean near
the island of Barbados. The vertical profiles of computed
and observed mean wind speeds, water vapor, and virtual
tcmperature agreed reasonably well and are not sbown. The
results of the model-computed profiles of some turbulence
variables, the mean and variance of liquid water, and cloud
volume are shown in Figure 17. No data wcre availablc for
quantitativo comparison of these variables. Satellite photo-
graphs, bowever, indicare lhat the computed cloud volumes
were slightly overestimated. One intcresting resull is seen in
the vertical profiles of the cornputcd turbulcnce energy and
eddy viscosit y in Figure 17. These variables exhibit severaJ
muima in the layers 1- 3 km above tbe surface. lncreascs of
the computed t urbulencc encrgy and eddy viscosit y are
closely related with the increases or the liquid water. The
condensation process releases latent beat whicb produces
Jocally unstable layers, resulting in larger turbulence.
(741>) Cooling Pond Simulation
, ae, , aQ .
C ({Jr - {3,R b) - + (fJw + {3,R a) -
Z l
lt is uscful to note that as u, -> O, the above formulism
reverts to a simple condensation model invol ving onl y mean
quaotitics. Sincc it may be shown that a(Q., - Q,
) Qw -
Q, . we obtain R = H(Qw - Q,) and Q
; (Qw - Q, JH(Q. -
Q,) as u, -+ O, where, once again, H is tbe Heaviside
Cloud Simulation
The application of lhe cloud model is still very much in a
research stage, and de veloprnent is handicapped by lack of
data. However, a onc-dimensional version ofthe present sel
of equalons including condensalion processes was used to
simulat.e the Barbados Oceanographic and MeteorologicaJ
Experiment (BOMEX) data [Holland and Rasmusson,
1973). Model results and cornparison with available data
. ,
.......... ___ ';!


q2/2 (m2 ,-2)
.. ~ ....



100 200
(m2 s"' )
A threc-dimen.sional version of the model developed here
has beco uscd by Yamada [ 1978a, 1979) loevaluate quantita-
lively the effec1s of a large cooling pond on the sum,unding
cnvironment. The time depcndent equations were integrated
to obtain a nearly stationary state of tbe circulation over a 2
km by 2 km square pond. Typical values observed for lhe
wind specd, ternperature, relative humidit y in the surface
layer, and water surface temperat ure were used to construct
initial values a.nd boundary coodit.ions. The water and land
temperaturcs were kept constanl lhrougbout the integration
at 1sc and -IOOC, respectively. Figure 18a shows the
computed horizontal surface wind vectors at 0.2 m above the
surface. Wind speeds increased from 1.4 m s- at the i nllow
boundary to 3.8 m s- over tbe cooling pond. This accelera-
tion of wind speed is caused by the dccrease of roughness
length over land (3 x 10-
m is assumed) to t.hat over water
(d) {e)
o 0.4 o 0.4 o
(o ko"')
( q 2 ) ~
(g kg"') R
Fig. 17. BOMEX cloud simulation. (a) Simulaled turbulenee enc:JV, (b) viscosily coef!icients, (<) mixing ratio of
liquid water, (ti) rms of lhe variance of liquid woler, and () cloud fraction. Solid I ines are for 0600 hours and dashed
lines for ISOO hours on June 24.
' 4 -1
, , / / / , / / / ;
, / / / / / / , / '
/ /. / / / :/ /
J-/ //1. / /
/,/// l.
. .
/ .. :/ .. } /
, ,
, ,
/ / / / /
, , ,
2 3

Fia. 18a. Horizon1al wind vectors at 0.2 m above the surface. Thc
boundary of the cooling pond is indcated by dashed !ines.
(S x to-s m, a typical value computcd from 0. 16u,
/g, wbere
u, is a friction velocity) and by the air circulation induced by
lhe temperalure diffcrcocc between the land and the pond.
Convcrgence and divergence in the horizontal wind distribu
tons rcsuh in lhe vertical wind distribution (Figure 181>) in a
vertical plane. A maximum upward motion of J.S em s-
computed over the downstream edge of lhe cooling pond.
Global Atmosphuic Simularions
The levei 2i mndel, wilh lhe algcbraic length scale cquation
described previously, has now beeo incorporated into the
general cireulation models of the National Oceanographic
and Atmospheric Administration's Geopbysical Fluid
Dynamics Laboratory. Currently, onc such mndel repre
sents the global at mosphere with horizontal resolution of
about 4 tatitude and longitude and 18 vertical leveis; the first
tive leveis are assi8ned to lhe lowcr 2 km.
Tbesc calculations produce an enormous amount of num-
bers. Calculacions extracted from a paper by Miyakoda a11d
Sirutis 11977] are shown in Figure 19. They are zonally
averaged plots of temperature, zona! velocity, and KM "' /q

"' ;;;

o 4 5 6
F"13. 18b. Oislribution of W in a venicaJ plane lhrouah 11>e
d;.,onal AB shown ill r,gure 18a. UniU are centimeters pc:r
SM. Synoptic detail is l herefore averaged out of thc plots.
Nevcnheless, one can identify the troposphere, tropopause,
and strnt.osphere in Figure 19a; in Figure 19b lhejet weams
are evidenl.
Ouanographic Simularioi&S
The models we have described have been applied to
oceanograpbic problems. The effect of salinity can be incor
poratcd in. say, the levei 2! mndel by simply setting {3(wO. )
= {3,(wd) + {3.(ws}. wherc (w6} is the potential temperature
llux and (ws) is the salinity Oux; fJ. and {3, are the corre
sponding coeflicients or lhcrmal expansion. 1t can be shown
that KH obtained in (32b) applies to both potential tempera
t ure and salinity. Funhermore, (34) and (35) are unchanged
so long as {3 aejaz {3
ae l ilz + {J, aSiilt in (33b).
The levei 2 mndel was applied by Mellor alld Durbin
[1975) to occan mixed layers and seemed to perform weU
wilh respect to labor.ltory simulations and also seemed to
simulate data obtained at station PAPA in the North Pacific.
1t has been applied by Wearher/y and Martin [1978) to the
study of ocean bo1tom layers and by Adams and Weatlrerly
[1981] to tbe study of bouom sediment tmnspurt. Simcms
[1980] has used the model to llindcast che seasonal variation
of horizontaUy averaged temperature profiles in Lake Ontar-
io using measured winds and storagederived swface heat
flux. The calc.ulated temperarure profiles were in close
agrcemcnt with observation. A similar study was made by
Klein 11980], and agreement was obtained with wind-driven
mixed layer deepening data obtained in the Mediterranean.
Klein [ 19801 also performed severa] useful sensitivity srudies
conceming tbe etrect of wind-foreing frequencies. Mixcd
layers when forced with the resonant, inertial frequency will
deepen much faster than when forced with higher or lower
A paper by Clancy and Martn [1981] presages lhe role of
turbulent closure modcls in operationally forcasting tbe
properties of ocean surface mixcd Iayers. 1'hey also used the
simplest levei 2 model. Ooe of their eonclusions is that
forccast errors are more associated with initialization fields
(combining climatology and expendable bathylhermograph
drops where available) and surfacc boundary conditions
(derived from lhe Navy's meteorologjcal model) than wi th
inadequacy of model physics. Further discussion of the
needs and potential for operational ocean forecasting can be
found in the work by Elsberry and Garwood [ 1980]. Gar-
wood 11979] has also made available a general review of
mixed layer dynamics a nd mixed layer models.
More recently. the levei 2! mndel h as been applcd to
tbree-dimcnsional ocean simulation by Blumhtrg and Mtllor
[1980). However, some two-dimeosional (x, z) caleulations
were also included in thcir paper. Thus Figures 20 and 21
illustrale the results of an impulsively imposed alongshore
wind stress. Figure 20 is a homogeneous upwelling event,
whereas Figure 21 is a density-srratiRed upwelling event.
The role of stratification in confining mixing to surface and
bonom layers is readily apparem. Also. in Figure 21. one
will observe the formation of a nearshore (x a 2 km) front
and baroclinic jet. Foo [ 1981 I has presented a more compre-
hensivc s tudy of nenrshore upwelling a nd frontogenesis
using a levei 24 mndel.
1t should be that in applicalion tbe masler lenglh
scale equation (48) works quite well. lt is integrated through.
out lhe entire computational domain and yiclds both surface

..__ ________

c ->00 _)

(b) 20-
Fig. 19. Zonally averaged vnrinblc contours from lhe global a1mospheric simulation by Mlyakodn a11d Sinlllt [1977]
in March. (a) Virtual potentjoJ tempcrature. Units are deg:rees Kelvin. (b) Eastward velocily contours. Units meters
per second. (<) Loe
or K.,i(cm
layer and bottom layer master lengths appropriate to each
Figure 22 sbows comparison between observed tempera-
ture data and calcularion for rhe annual cycle in upper layers.
The observed data are the climatologjcal temperature pro-
files in lhe Gulf or Mexico. Data from hydrographic surveys
are averaged by month. The data are tben area-averaged
throughout the basin. Salinity profiles are also obtained but
play a minor dynamical role in the Gulf. The model is dl'iven
by winds obtaincd from sbip repons and averaged in a
similar fashion. Drag cocfficients very close to tbose recom-
mended by Bunker [1976) were used to obtain surfacc stress
statistics i ., i
, and 1-rl. where the overbar bere represents
lhe average of ali data in a given monlh and on an average.
The actual imposed wind slress oscillatcs (with a 4-day
period) in magnitude and dircction so thatlhe resulting f .. fy,
and ltl match the data. The imposed surface heat Aux was
simply extrac1ed from the data so that the heat storage
(vertical in1egral of temperature) of lhe calculalion and data
agree. (Nore tbal for the same heat storage many profiles are
possiblc including, at onc exlreme, very large summenime
temperarure confincd to a vcry thin surface layer.) The final
resull is that observationally predicted temperature profiles
compare quile weU. Further details and an account ofa fully
tbree-dimensional model applied to the Gulf of Mexico are
found in a report by 8/umbug and Mellor [1981).

Distaoce Otishore (km)
Fig. 20. A hom011eneous event induced by an along
sbore wind slress of2.0 dynlcm directed into the plane ofthe paper.
The wind stress h8$ been impo>Md for 6 hours . The onsbore (U
negative) and offshorc (U poltive) isotactos are dcpicted in the
upper portion ofthe figure, while lhe alongshore (V positivc into tbe
plane ohbe pape r, V nogative out ofthe plane oftbe paper) isotacbs
are depioted in the lowcr panion.
A turbulence model has been developed which is n:lative-
ly simple and which can be applied to a wide variety of
engineering and gcopbysical nows.

--- --- - li.J- .... ------- ---- --- - ;
5 r

- --- --17.$- - - --- -- --
DISllru Oll>tac (km)
2 4 6 10 12
We separare the sludy of the model into (I) the group of
closun: assumptions propOsed by Rona, extended to include
temperature (or any other scalar) and den.sity stratification,
and (2) the master lcngth scale.
The rules of lhe game we an: playing, at least until lhe
pn:sent, an: to obtain ali empirical constants from neutra!
data and then see if the modcl can predict stabilization or
destabilization ofturbulent lields dueto dcnsity stratification
in a gravity field and, in sepa.rate studics, due to ftow
curvature and olhe r body elfects. The constants in
(12), one of which is not indcpendent , are unambiguously
related to simple neutra! now data, and computer solutions
are not required lo identify these constants. The remaining
constants are von Karrnan's coostant " and the three con
stants in (48), one of wbich is not independent. Trial and
error computer solutions have determined these last two
independent constanls.
Tbe model and the fixed sct of constants seem to perform
weU in predicting diverse neutra! flows. The samc model.
with no alteration, appears 10 predict density-stratified flows
in a manner which far exceeds expcctation prior ro 1973.
The various models. leveis 4. 3, aod 2, n:present
decrcasing complcxity and decreasiog requrements for com
puter time and storage. For one (vertical) and pcrhaps two
dimensional model simulations. aoy vcrsion is aO'ordable.
However. for large, three-dimcnsional. atmospheric or
oceanographic numerical modcls the levei 4, cenainly, and
the levei 3, probably, must yicld to eithcr the level2i or lhe
levei 2 closure model on tbe basis of computational econom
The levei 2 model has received the greatest exposure in
the literature, partly because of computer cconomy and
partly because it is relatively easy to program. For most of
our own work we use the levei 2i model. This requires lhe
solution of equations for</ and </1. over and above the usual
prognostic equations for mcan velocity, temperature, and
Distanc:e Ofbhlre (\m)
Fig. 21. A stntified upwelling evcnt induced by an. atonvbore wind stress of I .O dynlcm directed into the plane or
tbe paper. This wind 5tress has been imposcd for 12 hours. The direclion ofthe iso1achs is tM samc as in Figure20. The
in.itial temperature distribution is denoted as T

.. .. ---_ .. --
' r"


22<1. The climato!oskal, V11riation
of tempera<ure in lhe GuJf of Mexico.
humidity for lhe atmosphere or salinity for tbe ocean. lt has
a grealer prediclive range than the levei 2 model, and the
length cquation, although the most empirical element
of lhe complete modcl. does seem 10 perform in a more
satisfactory manncr than the simplc algebraic equation asso-
ciated with the levei 2 model. For example, in ocean
simulations, ali equat ions may be integrated top 10 bottom
and generally include separate, surface and bottom mixed
layers bounding nonturbulem How. The protolype example
of a singte boundary layer bounded by nonrurbulcnt Oow is
lhe laboratory How in Figure 7. On the other hand. in
shallow water. where surface and bottom layers merge. the
lenglh scale equation performs well, as was seen in the
prototypical merged layer case of cbannel ftow in Figure 6.
1t is important to realize that tbe simpler models cannot, in
principie, account for some How bebavior. For example.
Uberoi's experimenl requires the fulJ levei 4 model to

rmi -10
- li)
. !OI)

Fia. 22b. 1\todel imulalion ofthe time-depth varialion oflempera-
cure in the Oulf of Mex.ico.
accouot for the retum to isotropy for a homogeneous,
inilially anisotropic flow field where shear and buoyancy
production are nil. The leveis 3 and 2j models will correctly
account for the decay of lurbulent kinetic energy, but ali
energy components will be declared equal. Finally, lhe levei
2 model, for t he same flow. will yic.ld 1.cro turbulent kinetic
energy. Since spatial difl'usion of energy is lacking, it will
also yield zero turbulent kinelic energy at the center tine of
channcl ftow, conlrary lo observalion. Nevertheless, in
many geophysicaJ applications lhe levei 2 model works quite
well. since mixing events are generally dominated by the
shear or buoyancy produclion tcrms in the 1urbulent kinet ic
energy equation.
llcla!owldgmmu. Tbis research was supported by lhe Air
Force Office of Scientilic Rcsearch under granl AFOSR 79.0118. by
lhe National Occanic aod Atmosphcric Admins1r01ion under grant
04-7-022-44017 (G.L. I>l. ), aod by lhe U.S. Departmenl of Energy' s
ASCOT program ( f. Y .).
Adams, C. E., Jr . aod G. L. Wealherly. Some elfeccs ofsuspended
sediment strarification on an oeeanic bouom boundary layer, J.
Oeoflhys. R.s .. 86, 41 61-4172. 1981.
Andre. J. C .. G. DeMoor. 1'. Lacanere. and R. Ou Vachal,
Turbulence approximalion for inhomogencous llows, I. Tbe clip-
ping approximation. J. Atmos. Sei .. JJ. 476-481, 1976a.
Andre. J. C., . Del\toor, P. Lacorrcoc. and R. Ou Vachat,
Turbulence approximation for inhomogcneous flows, ll, 'fbe
numerical slmuJation of a penetrdtivc convcction expcrimenc J.
lltmos. Sei .. 33. 482-491. 1916b.
Bens, A. K., Nonprecipitating c;;umulus convec.tion and its param
eterizalion, Q. J. R. Mtuorol. Soe., 99. 178-179, 1973.
Blumberg, A. F .. and G. L. Mcllor. A coastal occan numerical
modcl, in PrQCee.dings o f thr Symposlum on Mathematico/ Mod
tlling t>[Estuarin<> Physics, Springer-Verlaa. New York, 1980.
Olumberg. A. F., and G. L. MeUor. A nuoncrical cakulacion or lhe
circulation in the Gulf of Mexico, Rtp. 66. 153 pp., Oyoalsis of
Prince1on, Princeton. N. J., 1931.
Bremhom, K .. and K. J. BuUock. Spec1ral measurements of
turbulenl heat and momeotum tntnsfer in fully developed pipe
now. lnr. J. Htot Mass Trmufu, 16, 2141-2154, 1973.
Brisas, M., G. Mellor, and T. V amada. A .ecood moment <urbu-
lence modcl applied to fully scparated 6ow. in Proettdings: 1976
SQUID Worksloof/ on T11rbulen<'e in fnttrnal Flaws. pp. 249-276.
Hemisphere, Washington. D. C .. 1976.
Bunkcr, A. F . . Computation of surfucc cnergy ftux and alrsea
intcrncrion cycJes of thc North Atlantic Ocean, Mon. WtathtT
Rev .. 11)4, 1122- 1139, 1976.
Burk. S. 0., The moist boundary laycr wit.h a hiaJ>er order lurbu
lente closurc model, J. Atmos. Sei .. )4, 629-638. 1977.
Burk. S. 0 ., Refractive index structure parMnelcrs: Tifne.depen-
dent calculations using a numcriaJ boundarylayer model. J.
App/. Mneorof., /9, 562-516, 1980.
Burk. S . O . Comparisoo of structure parame1er scalin& expression
wilh lurbulence closure model ptedic:tions, J. Atmos. Sei., J8.
751-761, 1931.
Busingcr. J. A . J. C. Wynpacd. Y. lzumi. and 1!. F. Bradley, AW<
profile rdaJionships in the atmospheric surf'ace layer, J . Armo.s.
Sei .. 18, 181-189. 1971.
Cauahey, S. J .. and J. C. Wyngaard. Tbe tutbulence kinetic energy
budgct in convecti\'e conditions, Q. J. R. Soe., JOJ,
231-219, 1979.
Champagne, F. H., V. G. Harris. aod S. Comin. Expcriments on
nearly homogcneous turbulent shear ftow, J. Fl11id, 4/, 81-
141 . 1970.
Clnncy, R. M., and P. J. Martin, Synoptic forccosling of the occanic
mixed layer using the Navy's opcrationnl cnvironmental data
base: Present capabilities and future applicntions, Buli. Am.
Meteoro/. Soe. , 62, 110- 784, 1981.
Clarke. R. W .. A. Oyer, R. R. Brouk. D. G. Reid. andA. J. Troup,
The Wangara experiment: Boundary layer data. Tteh. Pop. 19,

Di v. of Mete<>rol. Pllys . Commonwealth Sei. and Jnd. RC$.
Organ. , Melboqrne, Australia, 1971.
Comte-Bellot, G. , Ecoulement turbulent entre deu parois pvd-
leles. Publ. Sei. Tech. 419, de I'Air, Paris, 1965.
Daly, B. J. , and F. li. liarlow, Transport equations in turbulcnce,
Phys. Fluids, IJ, 2634-2649, 1970.
Oeardorfr, J. W . Three dimensionaJ numerical s1udy of turbulence
in an eotraining miJCed layer, Boulldtlry Loyu Mettorol., 7. 199-
226. 1974.
!Mkey. T. 0., aod G. L. MeUor. Deeaying turbulenee in neutr.ol and
stratilied ftuids, J. Fluld Muh., 99, 13-31, 1980.
Dobo3y, R .. Dispersion of atmosphcric pollutants in ftow over the
shorelines of a large body o r water, J. Appl. Mrttorol .. 18, 117-
132. 1979.
Donaldson. C. D .. and li . Rosenbaum, Calculation of the turbulent
shea.r Rows tbrough closure of the Reynolds cquruionli by invari ..
ant mndeling, ARAP Rtp. 127. Aerooaut. Res. A>>lOC. of Prince
ton. Princeton, N. J .. 1968.
Elsbtny, R. L .. and R. W. Oarwood. Numerieal ocean predi<bon
modei.......Goal for the 1980's, Buli. Am. Mtttorol. Soe .. 61. 1556-
1566. 1980.
Foo. E.C., A tw<>-dimensional diabatic isopycnal model simulating
the coastal upwelling front. J . Phys. Ouanogr., 1/, 604-626.
Gambo. K., Notes on lhe turbulcnce closure model for atmospheric
bouodary layer, J. Mtttorol. Soe. Jpn .. S6. 1978.
Garwood. R. w .. Jr., Airsu intcractions and dynamics of lhe
surfo<:e rruxed layer. Rv. Gtophys. Spaa Phys .. 17, 1507- 1524,
Gibson. C. H., and W. H. Schwartz. The universal spcctra of
lurbulent velocity and scnlar fields, J. Fluid Mtch., 16. 365-384,
Gowcn. K. A., and J. W. Smilh. Turbulent heat transfcr from
smooth and rough surfaces. lnt. J. Heat Mass Tran.ifer, li, 1657-
1673, 1968.
Har\inlc, K., and 8. E. Launder. FuJJy dev<lopcd asyrnmetric ftow
in a plane cbannel. J. Fluid Mteh., .l1, 689. 1972.
H:u-sha, P. T. , Kinet.ic energy mctbods. in Handbook o[Torbulncr,
edited in W. Frost and T. li. Moulden, pp. 187- 232. Plenum, Ncw
York. 1977.
Hinu, J. O. , Turbulenco, 7'lQ pp., McGraw-Hill , New York, 1975.
Hulland, J . z .. and E. M. Rasmusson, Meas urements of the
atmospheric, mass. eneray and momentum budget over a 500-
kilomcLer square of tropical cx::can, Mon. Weothtr 101. 44-
45. 1973.
Kantha. L. H. , O. M. Phillips, and R. S. Azad, On tbe turbulent
entn.inment a1 a stablc density llterface, J. Fluld Mech., 79. 753-
758. 1977. .. andO. M. PhiUips, On lhe peoetratioo oh tutllulentlayer
into a stratilied ftuid. J . Fluid Mech .. 37, 643-655, 1969.
Kcstin. J. , and P. O. Richardson, Heat transter across turbulent
incompressible boundary layers. in Calloque lnt ernmtonal sur la
Mc:mrique de la Marseille. Editions du Centre
Ntional de la Rechcrche Scientifiquc. Paris. 1961.
Klebanolf. P. S. , Characteriuics of turbulence in a boundary laycr
with zero pressure gnldient. Rep. 1247, Adv. Comm. for
Aeronaut .. Washington. D. C .. 1955.
Klein, P .. A simulation of the airsea transrer variability on lhe
>tructure of marine uppcr layers. J. PhY Oaanogr., 10. 1824-
1841 , 1980.
Kllne, S. J., M. V. Morkovin, O. Sovran, and D. J . Cockrell, in
Proceedngs o/ the AFOSH Suuiford onferentt on Computation
Boundary I.Aytrs. Thcrmoscicnces Division. Depa.rt
ment of Mechanical Engineering. Stanford Universily, Stanford.
Calif .. 1968.
Kolmogorov, A. N., The loc&l structure of turbuleoce in incom-
pressible viscous Huid for very large Re)nolds number (in Rus-
sian). Dold. Alwd. Nauk SSSR. 10, 301. 1941. (EnjJlish transla
tion, S. K. friedla.nder and L. Topper, Turb11ft nct, lnletscience,
Ncw York, 1961.)
Kukharcts, V. and L. R. Tsvang, The turbulent energy dissipa
tion rate in an unstable Stnuitled atmospheric boundary Jayer, IZl'.
Arad. Sei. USSR Atmos. O<tanic Phys . Engl. Transl., /1, 419-
42A. 1977.
Laufer, J., lnvestigation or turbulent ftow in a two-dimensional
cbannd, Tech. Nott 1111, Natl. Adv. Comm. for Aeronaut.,
Washington, O. C .. 1950.
Laufcr. J .. The structure of turbulcnce in fully developed pipe llow,
Rep. 1174, Natl. Adv. Comm. for Aeronaut .. Washington, O. C ..
Launder, 8. E . , On the efl'ect of a gravitational field on the turbulent
lransport ofheat momentum. J. Fluid M.c/1.. 67, 569-581. 1975.
Launder, 8 . E .. G. J. Reeoe. and W. Rodi, Progress in lhe
developmen1 of a Reynoldsstress turtulence closurc, J. Fluid
Muh .. 68. 537-566, 1975.
Lenschow, O. J., Airplane rntasuremen1s of planetary bouodary
layer structwe, J. Jlpp/. Meteoro/ .. 9, 874-834, 1970.
Lewellen, W. S .. Use or invariant modelling, in llandbook of
Turb11lenu. edited by W. Feasl and T. H. Moulden. pp. 237- 2n,
P1cnum, New York. 1977.
l..ewellen, W. S. , and M. E. Tcske, Prediction of lhe Monin-
Obukhov similarity funcLions from an invariant model of turbu-
lenee. J. Atmos. Sei., JO, 1340-1345, 1973.
Lewellen. W. S .. M. E. Teske. and C. O. Donaldson. Variable
densily ftows computed by a sccond-order closure description of
turbulenee. AlAA J .. 14, 382-387, 1976.
Lumley, J. L., and 8 . Khajeh-Nouri, Modeling homoseneous
deformation of turbulence, Ad. G<opilys., 18A. 162-192, 1974.
Lumley, J. L., O. Zeman. and J. Seiss, The inftuencc of buoyancy
on turbulent transport, J. Fl11id M.cil., 84, 581-597, 1978.
Mellor. G. L .. The large Reynolds number asymptotic theory of
turbulenl boundary layers . Int. J. Eng. Sei .. li, 851-873. 1972.
Mellor, G. L .. AnalyliC predction of tbe propcnies of stratified
planetary surface layen, J. Atmos. Sei., JO, 1061- 1069. 1973.
Mellor. G. L ., A comparative study of curved ftow and density,
stnuified llow. J . Atmos. Sei .. J2, 1278-1282, 1975.
Mellor. O. L .. Tbe Gaussnn cloud model relatio n, J. Atmos. Sei.,
34, 356-358. 1977. (Corriacnda. J. Atmos. Sei ., 34, 1483-1434,
Mellor, G. L., aod P. A. Durbin. The structure nnd dynamics of lhe
ocean surfaee mixed laycr, / . Phys. Oaanogr .. S, 718- 728, 1975.
Mellor, O. L., and H. J. Herrina. A suoey of lhe mean turbulent
field closure models, AIAA J ., li, 1973.
Mcllor. O. L .. and P. T. Strub. Similarity solutions for lhe stratilied
turbulent Rayleigh problem, J . Plrys. Oc'anogr., 10. 4SS-460,
Mcllor. O. L., and T. Yamada. A hierarchy of turbulence c105Ure
models for planetary bounda.ry layers, J. Atmos. Sei .. 3/, 1791-
1806, 1974. (Corrigenda, J. Atmos. Sei .. J4, 1482, 1977.)
Miyo.koda. K., and J. Sinnh;, Comparative integra1ions of global
models with various parunu:teriz.ed processes of sub-grid scale
vertical traosports, Contrib. Atmos. Phys .. SO. 445-441, 1977.
Monin. A. S. , On lhe symmetry properties of lurbulence in the
surface layers of ait. l z' Acad. Sei. USSR Atmos.
Phys., EnjJl. Traosl. , / , 45-$4. 1965.
Ng. K. E .. aod O. B. Spalding. Turbulence model for boundary
layers near walls, Phys. Fluids, I S. 20-30, 1972.
Pennel, W. T .. and M. A. LeMone, An experimental study of
turbulence structure in the fair-weather trade wind boundary
layer. J. Atmos. Sei .. Jl. 1308-1323, 1974.
Peny, A. E .. and C. J. Abel!. Scaling laws for pipe ftow tu.rbulenee,
J. Fluid Mech., 67, 257- 271, 1975.
Rose. W. G., Results of an attempt to ge.neratc a homogenous
turbulont sbear llow, J. Fluid .. 67, 2$7-271, 1966.
Roua. J. C .. Statistische Theorie nichthomogener Turbulenz. I , Z.
Phys. 129, 547- 512, l951a.
Rotta. J. C . Statistische ' O.eorie ruththomogener Turbulenz. 2. Z.
Phys., 131,51-77, 1951h.
Roua, J. C .. Turbulent shear laycr prediction on the basis of the
transport equations for che Rtynolds stresses, in Proctedlngs o!
the /)th /nternationa/ Congrns on Thrortrical and A.pp/itd
Mt chanics, Moscow Unlersity, pp. 295-308, Springcr-Verlag,
New York, 1973.
Shir. C. C .. A preliminary study of atmospbcric turbulent Oows in
lhe idealiud plaoetary boundary layer, J. Atmos. Sei., JO, 1327-
1339. 1973.
Simons, T. J., Vcri.fie:ation of seasonal stratification models. rcport,
lnst. of Metcorol . and Ocennogr., U niv. of Utrcchl, Netherlands,
So, R. M. C., A turbulent velocity scale for curved sbear ftows . J.
Fluid Mch .. 70. 37-57, 1975.
So, R. M. C .. and G. L. Mellor. Experiments on con eurvatwe
etrects in turbulent boundary layers, J. Fluid Meclr .. 60, 43-63.
M_eu..oa ANo V AMADA: T uu uLENCE CLOSuae MoDEL

So, R. M. C., nnd G. L. Mellor. Experiments on turbulent boundary
layers on concave wall. A"i>nllMt. Q. XXVI. 25-40, 1975.
Sommeria, 8., and J. W. Deanlot1f, Subgrid scale condcnsation in
d oud mod<:ls, J. Atmos. S<i .. 34. 344-lSS. 1971.
Sun. W. Y .. and Y. Ogura, Modeling the evolution oflbe convec.tive
planetary boundary layer, J . Atmos. Sei .. J 7, 1558- 1572, 1980.
Tennekes. H .. and J. L. Lumley. A First C:omse in Turbultnr:e. 293
pp., MIT Prus, Cambrid&c. Mau .. 1972.
Uberoi . M. S .. Eft'ect of wind-tunnel tontractioo on ft<e stream
turbulenee, J. Aeronout. Sei .. 2J. 754-764. 19S6.
Uberoi , M. S., Equipartition of energy and local isotropy in turbu-
lent ftows, J . Appl. Phys.. /8, 11 65- 11 70 , 1957.
Wcatberly, G. L., and P. J. Martin. On tb< s tructure and dynamics
of tbe ocoan bottom boundary Layer, J. Phys. Oceanogr., 8. 551-
510, 1978.
Wieringa. J. . A reevaluation or the Kansas Masl i_nOuence on
measurements of srrcss and anemometer overspccdin,g. Ooundary
I.Ayer Meteoro/ ., /8, 411-4)0, 1980.
Willis, G. E .. and J. W. Deardot1f, A 1abor.uory mode1 of Ih<
un5table p1an<tary boundary l>)'<r, J. Atmos. Sei .. Jl. 1297- 1307,
Wolfstein, M .. On the length<ea le of turbulence equation, /sr. J.
Tethn<>l .. 8. 87-99, 1970.
Worthem, S., and G. L. Mellor, A turbukonce of closure mode1
pplied to the upper tropical oeean. Dup S.a R-.., Gate Suppl. I.
16. 231- 212. 1980.
Wynsaard. J. C .. Modeling the planetlll'y boundary Layer- Exten
s ion to thc stable case. 8()11ndllry I.Ayu Meteoro/ ., ~ . 44 1-41,
Yllinik. K. S . Asymptotic tbeory of turbulent shear ftow. J. Fluid
Muh . 41, 11-427, 1970.
Yamada, T., A numerical experime-nt on pollut.anr dispersion in a
horizonta lly-ho mogeneous-atmosphetic boundary laycr, Armos.
Environ., 1/, 1015-1024, 1977.
V amada. T., A threc:-dimcnsional. secondordc:r closure numerical
model cl mesoscak circuladon in lbe lower atmosphere, Topicol
Rtp. ANL/RER-78-1, 61 pp. , Radial. and Environ. Res. Div. ,
Argonne Natl. Lab . Araonne . 111 .. 1978a.
Yamada, T .. A three-dimensional numerical s1udy of complex
atmospheric cireulations produced by temn, in Procf*f*dings:
Coriferenct on s;nra Ntvada Meleorology, pp. 61- 67, American
Mctcorologieal Society, Boston, Mass .. 1978b.
Yamada, T., Ao application of a three-dimensional, simplifled
second momcnt closure numerical model to stucly atmospheric
eft'ects of a large coolina-pond, Atmos. Environ. , /J, 693-704,
Yamada, T., A numerical 5imulation ofnoctumal drainagc now, J.
Meteoro/. Soe. Jpn .. 59, 108- 122. 1981.
V amada. T .. A numeticaJ mode l study of turbulent airflow in aod
above a forest eaoopy, J. Me1corol. Soe. Jpn .. 60. 439-454,
Yamada, T .. Simulations or noctumal dminage ftows by a tll
turbulencc closurc modcl. J. Atmos. Sd .. in prcss. 1982b.
Ya mada, T .. and G. L. Me llor. A !imulation oi tbe Wanpr
atmospberic bouodary layer data. J . Atmos. Sei .. 12, 2309-2329.
Ya mada, T .. and G. L. Mellor. A numcrical simu1ation of the
BOMEX data using a tut'bulcnce closure model coupled with
ensemble c loud re1ations, Q. J. R. Me,.orol. Soe .. /OS. 915- 944,
Yokoyama. O .. M. Gammo. and S. Yamamoto, Oo tbe turbuleoce
quantitics in he atmospheric mixin.g layer, J. Meuorol. Soe. Jpn.
55. 182-192, 1971.
Young, M. 9 . O., E. Y. Hsu. and R. L. Strcct, Airwnter interac-
tion: 'Thc. nacure of turbulent he-at mau a.nd momtntum tran.sfcr
mechanisms in tbe air boundary layer, Tuh. Rcp. /6J, Dep. of
Civ. Eng . Stanrord Univ .. Stanford. Calif .. 1973.
Zcman. O .. and J. L. Lumley. Modeling buoyaoey driven mixed
layers, J. Atmos. Sei. , JJ, 1974- 1988, 1976.
(Reccivcd February 9. 1981:
acccpted May 10, 1982.)
- . . -