Course 12.

812, General Circulation of the Earth's Atmosphere

Prof. Peter Stone

Section 8: Lorenz Energy Cycle

Energy Forms: As we saw in our discussion of the heat budget, the energy content of the
atmosphere per unit mass can be expressed as the sum of four terms:
E = I + + LH + K
I = C V T = internal energy

= gz = potential energy

) (

1 2
1
u + v 2 + w 2 u 2 + v 2 = kinetic energy
2
2
LH L V q = latent heat.
K=

We assume CV and LV are constants.

If we assume hydrostatic equilibrium, then for the total internal energy in the column of
the atmosphere we have:
ps (z=0)

C T
Idz
=
C
Tdz
=

gV dp; for the potential energy we have

0
0
0

dz =
gzdz =
0

ps (z = 0)

ps

=
pdz =

zdp =

ps (z=0)

d(zp)

pdz

RT

dp ; (we choose z = 0 to be a level below the lowest surface,

RTdz =
g

zs min, and for convenience, in the presence of topography, we take p = ps for z < zs. We
will see later why. Note that the zero point for potential energy is arbitrary.)
ps (z=0)
ps (z=0)

CV + R
CP T
Thus I + dz =
dp =
dp.
g
g
0
0
0
Therefore in a column of the atmosphere the sum of I.E. and P.E. equals the enthalpy,
and all three are equal within a constant factor. This is also sometimes called the total
potential energy = TPE. This can be re-written in terms of the potential temperature,

R
2

:
= T 0 where =

7
CP
p

p
1
Tdp = dp =
dp+1;

1+ p0
p0

dp+1 = d p+1 p+1d;

 integrating by parts we find for the total potential energy:

ps (z =0)

p=ps ,z =0

ps (z=0)
CP
C
1
+1
p+1

Tdp = P

p
d

0
g
g 1+ p0

p=0

What are the limits on ? We now adopt the convention that below the surface, z < z s ,
< S = surface , and p = ps = constant, and 0 at z = 0. (We implicitly assume that

dz

> 0, i.e. the system is always stably stratified.)

p+1

ps
0

= 0 and

CP

g 1+ p0

+1

d = TPE

However, we note that in a stably stratified atmosphere like ours much of the TPE can
never be released or converted to KE. In particular, consider a typical arrangement of the
isentropic surfaces as shown in the first diagram below.

Figure by MIT OCW.

6 > 5 ... > 1. As we see there are horizontal gradients and because of this PE can be
released by the motions in spite of the stable stratification. Consider for example a
displacement like that shown below.

Figure by MIT OCW.

For this displacement high air is being introduced into lower surroundings, so

' > 0, but w ' > 0 also, i.e., w '' > 0 , warm air is rising, and PE is being converted into
KE This is of course the mechanism of baroclinic instability.
Now suppose we re-arrange the isentropes adiabatically (so no new energy is added) and
while conserving mass, in such a way as to eliminate any horizontal gradients. Since in
H.E. p is just the amount of mass above any given level, conserving mass means the
integrated or average p above any surface is conserved. The average p is
1
p! = p x, y, dA and using this definition, one can now construct the state with no
A
horizontal gradients, as shown below.

()

Figure by MIT OCW.

In this new state, since

> 0, there are no displacements which can release PE,

because now for any displacement, w '' < 0 . Thus the PE of this state is unavailable.
This led Lorenz (1955) to define the available potential energy as the difference between
the TPE in the actual atmosphere and that in the adjusted state just described. If we call
this P, then for a given system or region,

P=

CP

g 1+ p0

dxdy d ( p

+1

p! +1

Since p! is already integrated over the area, we can rewrite the integral as

(
)
!
> 0, ( p ) > ( p" )

+1
A d p!
p" +1 . This is the so-called exact formula. Note that, since
0

+1

+1

, i.e. P is according to our definition always positive.

Lorenz derived an approximation for P that is commonly used. Let p = p! + p' , and
similarly for other variables:

+1 p' +1 p'2

p = p!
= p!
1+
+
+ 2 + ...

!
p
2
p!

+1 p#
'2
"
+1
+1
1+ 0 +
+ ...

p = p!
2

2
!
p

+1

+1

+1

p'
1+
p!

+1

C P A +1

2g 1+ p0

#2
C
+1 p '
!
p p! 2 d = 2gpP
0
0

dxdy p!

p'2 d

As a test of the accuracy of this approximation, we can take from Oort (1971) the
following numbers for = 300 K in the annual mean: p equator = 980mb ,

p 70N = 435mb, p! ~ 708mb, p ' ~ 273mb, and the ratio of the 1st neglected term in
the series to the first retained term is
1 p' 0.4 273
=
= 0.05
3
p!
3 708

Carrying out an analysis in coordinates is not very convenient, because we would have
to interpolate from pressure coordinates, which would introduce error. Lorenz
introduced another approximation. This is based on the approximation that both
isentropes and pressure surfaces are quasi-horizontal with very small slopes:

40
z
y
104 km ~ 103 and
~
=

5
y
z
km
p
4
z
y u 10 10
~
~
~ 104.
=
p
g
10

y p
z

Consider for example an isentrope on which p has an average p! ; now consider a point
A on = constant. At this point, there will be a pressure deviation p' .

Figure by MIT OCW.

Now consider the variations on p! at the same latitude, . Since the variations are
primarily vertical, or equivalently, vary primarily with pressure, we can write:
p '
p!
'
=

< 0, (Note that as drawn, p',

' < 0 ).

.
The minus sign is necessary since
p!

Substituting this into our integrand for P, and changing coordinates from to p! :
, before integrating over x,y we have that the integrand
d = dp!
p!

CP

2gp0

p! 1 '2

p!
! Note that ' is now the variation of along a constant p surface,
dp;

p
since = T 0 .

' T'

= ; Also + =
d (use H.E.) where d = g .
CP
T
P
d p

p!

p
d

p!
=
, and = T 0 ;

=
p!
d

Substituting this, the integrand simplifies to

CP
2 gp0

p! 1

p! 2 2
T' dp!
d T 2

1
p! p0 T'2
1 p!
=
dp!
2 p d p! T

1
T'2
! Now we substitute into the integration and absorb the minus sign by
dp;
2 T d

reversing the limits of integration:

PS

1
P =
2

T'2
.

dxdy dp! T

Finally Lorenz assumed that the coefficient of T'2 in the integrand varied slowly
compared to T'2 , and replaced T and by their area weighted means T! and ! . (i.e., 3rd
order quantities are again being neglected.) This is not as good an approximation as the
earlier ones. (From Oorts data I estimate ~10% error, primarily from T.) Once this
assumption is made we can write (noting that now p! is a dummy variable)

P=

1
2

pS

dp T!

dxdyT'

Note that
2
1
1
dxdyT'2 = dxdy T T!

A
A
1
= dxdy T 2 2TT! + T! 2
A
p
1
1 S
dp
2
2
!
= dxdy T T ; P =
A
2 0 T! d !

dxdy (T

T! 2

The first term can be expanded in the space domain: T = T + T * ,

T 2 =
T2 + T *2

we can decompose P! into mean and eddy components; the mean is

1
PM =
2

pS

dp
!
d

T!
0

dxdyT

1
(You may verify that PM =
2

T! 2 .

T'

dxdy
dp

T! ! .)
A
0
d
pS

We have included the T! term in the mean available potential energy because it has no
zonal variations. The eddy component is the remaining part,
2

PE =

1
2

pS

dp
!
d

T!
0

T * dydx.
2

Now let us compare P to TPE:

1
2

pS

dp T!

CP
g

1
!
d

dxdyT'

pS

dp dxdyT

1 g
T'2

! !
2 C P TT
d

! g = ; T' ~ T;
T ~ T;
d
CP
P
1

~
TPE 2

(T)

1
; T ~ 30 , T ~ 250 ; ! ~ d

2
!
T 2 1

30

P
1

~
~
; Hence only ~1 to 2% of the TPE is available for conversion to
TPE

250

70

kinetic energy.
Balance Equations for Kinetic Energy:

We have already derived this for the eddy kinetic energy, integrated over the whole

2
2
1
atmosphere, (remember, K E = u + v ), i.e.

Z Z
!
!
!

K Edydp = g
dydp +
v H F

v H

F dydp

p
t
p

+ u uv + udydp + v v 2 + vdydp.
p
p
y

Let us expand these terms in the space domain, u = u + u*, etc.

Consider the first term: expanding, and also substituting for H.E.,
Z
1

= = ; we have dydp = * *dydp.

p
g
g
The term
1
C PE , K E * *dydp
g
represents a conversion of eddy potential energy to eddy kinetic energy, i.e. in pressure
coordinates in zonal planes, if warm air is rising * > 0,* < 0 and cold air is sinking

1
is necessary to make C have
g
units of energy per unit time. Also, recall that this is total potential energy, i.e. internal
energy plus conventional potential energy, not just the latter.

(* < 0,* > 0) , then C ( P , K ) > 0 , etc.

E

The factor

Similarly, the second term when expanded becomes simply

u * Fx * +v * Fy * dydp = gD K E ,

(
)

!
where D K E represents the dissipation of eddy kinetic energy due to friction. Since v
!
and F are negatively correlated, the dissipation, D(KE) is positive.

( )

Now consider the third term, and expand it:

u y uv + p udydp

u v + u * v * + u + u * * dydp.
= u

p
y

The first and third terms can be combined and rewritten: expanding the derivatives, we
have:


u
u
2 v
2
u + u v + u + u




y
y
p
p

! 1 2
1 2 1 2

= v
u +
u = 2 v 2 u ,
y 2
p 2
2

because of continuity. Now we have a pure divergence, and can integrate by Gauss
!
theorem, and applying the B.C. v n = 0 , we get no contribution from these terms. Only
the eddy terms survive. The same thing happens with the fourth term in the equation, and
these terms are combined to define another conversion, this time a conversion from eddy
kinetic energy to mean kinetic energy:

2
1

C K E , K M = u u * v * + u u * * + v v * + v v * *dydp
g y
p
y
p

Note that in the Q-G approximation, the vertical flux terms would disappear and the v
term would be small, of order Ro .
Thus when there is convergence of eddy momentum flux positively correlated with the
momentum itself, either zonal or meridional, then C K E , K M > 0 . (Peixoto and Oort

(1992) give the conversions in the more complicated spherical coordinates.) Thus we can
now write for our eddy kinetic energy equation:
1
K Edydp = C PE , K E C K E , K M D K E
t g

( )

Also, we now need merely to substitute this into our earlier result for the sum of K E and
2
1 2
K M to get an equation for K M = u + v :

K Mdydp =

! !

g
v

dydp
+

p
H Fdydp t

dydp;

( )

Rewriting the term as before, and substituting for C PE , K E and D K E we get

!
!

!
!

K M dydp =
dydp + * *dydp +
v H F
dydp
v H *

F *dydp + gC K E , K M

t
!
!

dydp +
v H
Fdydp + gC K E , K M .

Define

C PM , K M =

dydp;
g

i.e. now warm air rising and cold air sinking in meridional planes converts zonal mean
total potential energy to zonal mean kinetic energy; and define

( )

D KM =

1
t g

1 ! !
v H Fdydp
g

= dissipation of the zonal mean flow, and we have

) (

( )

dydp = C K E , K M + C PM , K M D K M

Balance equation for mean available potential energy:

2
PS
T'
1

! ' = ! ,
; T' = T T;
We need an equation for PM = dxdy dp
!
!
2
T

0
d

T! = T!

( ) , etc.

We start by writing the heat balance equation in p coordinates with as

the dependent variable:

p
dp
R

dT 1 dp !
CP

= H; = ; = T
0 ; =

dt
CP
dt dt

p
T

p +
p
p C P

p
RT

p
pC

Substitute and divide through by C P = C

P

p0

p H
"
+u +v + = 0
= + v.
t
x
y
p p C P t

10

()

"
! =H
! '+ H
Substitute = '+ ! p ; also H

()

!"
"
'+ H
p H

0
!
!
!
'+ + v H '+ +
' +
=
.
t
p
p
CP
p

!
Because of continuity, v H ! +
! = . Substituting this into the equation and
p
p
!
take the x,y average, ~ ; using the B.C., v n = 0 , we have

( )

( )

( )

()

K
!#
! p H
!
"
0
! = 0 because of continuity.
!
;
+0+
' +
= 0
t
p
p p C P

( )

Note that this is precisely the equation on which global mean radiative-convective models
are based (cf. Hantel, 1976). Also, this equation in effect tells us that in equilibrium, the
! . Now
mean diabatic heating is balanced by the mean eddy vertical heat flux, '
subtract this equation from the '+ ! equation:

p H
$'
'

!
" + # =

+ v H '+
' '

t
p
p p C P

Take the zonal mean of this equation, and multiply by G ' where
2

p
1
G =
;
p0 T! d !

1 2
! + G ' " = G p0 H
#

' '
G
' + G ' v' + G '


''
t 2
y
p
p C P p

II

III

IV

VI

Now consider what will happens to each of these terms when integrated over y,p. (The
integration over x is implicit.) In I, in order to obtain the desired term, PM t , we have
to take G inside the time derivative, i.e., we have to neglect the time variations in T! and
! in this term. Since these are comparable in magnitude to the already neglected y
variations, it is no worse an assumption. If we are doing a global analysis, it is a much
better approximation because the seasonal changes in the two hemispheres counteract
2

each other. (See Oort & Peixoto, 1983, Fig. 47). Substituting for G and ' the I term
becomes:
2
2
1
p
2
P
p
2
0

T' dydp = M .

t p0 T! d ! p
t

Also note that this approximation does not affect the calculation of the mean balance
from this equation.
11

Terms II and III can be combined and simplified by integrating the perfect derivatives, as
follows:

dydpG ' y v' + p '

=

' v' G ' ' G '

y
p

v ' + v * '* G ' + ' + * '* G '

y
p

dydp y v' G ' + p G '

= dydp

The first and third terms can be rewritten as

2
2
2
v 1 G ' + 1 G ' = v! 1 G ' ,




2 2
y 2
p 2
2
because of continuity, and this integrates to zero. Noting that * = '* , we now have

II + III = dydpv * *
G ' + * *
G ' and substituting for ' , *
y
p

and G, this becomes

T'
* T * p T'

dydpv * T *
+
= C PM , PE .

y T! d !
p T! d
p

) (

This represents the conversion from mean to eddy available potential energy. Thus if
there are negative correlations between the eddy transports and the gradients of T' ,
(actually in the * T * case it is not the gradient of T' , but the coefficient of T'

does not vary strongly in p), i.e. if the transports are down gradient, then C PM , PE > 0
and PM is converted to PE (see diagram below). In effect, longitudinal T gradients are
being created from the meridional and vertical gradients.
Note that in the quasi geostrophic approximation, the vertical eddy flux term would not
appear.

12

Figure by MIT OCW.

Next consider the term IV:

dydpG '

( )

!
'
p

!' and G are independent of y, !' = 0 by definition, and ' integrates

= 0, because

to zero over y.
Now consider the term V:

! ! d !
!

=
. We find
V = dydpG ' ; substitute for G, ' and
p
d
p
p
V=

1
g

dydp' = +C ( P

,KM

Note that ' and are interchangeable in the integrand because the integral of is
0.
The factor of

1
comes in because we are integrating in pressure coordinates.
g

Note that =

1 RT
RT'
=
; ' =
; if warm air rises and cold air sinks, and '

p
p

in meridional planes are negatively correlated, and PM is converted to K M . Also note

that we can replace ' by because ! integrates to zero.
Finally, the VI term, upon substituting for G and ' , becomes:
K
G p0 !
VI =

H '' dydp
C P p

13

!
H
'T '
G PM
"

dydp C T"
P

( )

! fluctuation
mean available potential energy is generated if the zonal mean T and H
are correlated, i.e. if we have heating where T is high and cooling where T is low. Note
! 'T" = 0 .
that T' can again be replaced by T because #
H

Now we put it all together to get the PM equation:

PM
t

( ) (

) (

= G PM C PM, PE C PM, K M .

Finally, a PE equation can be derived in similar fashion by deriving an equation for

T *2 from the same thermodynamic equation. The already defined conversions for

C PE , K E and C PM , PE will appear again, and the only new term is a generation term

( )

exactly analogous to G PM , i.e.

( )

G PE =

dydp

!
H

* T *
"
C P T)
d

Thus our equation is

PE
= G PE + C PM, PE C PE , K E .
t

( ) (

) (

Observational Analyses: We shall follow Oort & Peixoto (1983). Oort & Rasmussen
(1992) cite Oort and Peixoto (1983) as their source, but there are minor discrepancies.
We will only look at the Northern Hemisphere analyses, because of the better data base.
Note however that if we do, then in general there are boundary fluxes across the equator
which must be added to the equations above. Oort & Peixoto (1974) calculated these and
found that they were generally not important except for the flux of PM in the seasonal

( )

extremes, B PM , associated with the strong seasonal winter Hadley Cells and the term
2
v 1 G ' . Thus we add B P to the P equation, PM = B P + ... .


M
M
M
2
t

( )

( )

The analyses are based on the 10 years in the MIT general circulation library, and they
calculated the four energies, C PM , K M , C PM , PE , and C K E , K M . Note that in

) (

evaluating the first two conversions they assumed Ro << 1 , and only retained the leading
terms. G PM and G PE cannot be calculated because of lack of data on diabatic

( )

( )

14

heating, C PE , K E cannot be calculated because of the lack of data on * * , and

( )
( )
G ( P ) , C ( P , K ) and D ( K ) were known, all the other unknowns can be calculated as
D K M and D K E because of lack of data on friction. However if any one of the three
E

residuals, assuming a stationary state (see Fig.1, top). Oort & Peixoto (1983) assumed
values for G PE and calculated the others for the annual mean and for the December-

( )

January-February and June-July-August seasons. They did not explain the basis for their
assumed G PE values in any of their papers. We can regard them as educated guesses.

( )

They assumed positive values, which could be explained as being due to a positive
! * and T! * (i.e. in zonal planes) due to baroclinic eddies releasing
correlation between H
! in the energy equations), i.e. warm air moving
latent heat (which is contained in H
poleward rises, leading to condensational heating.

15

Figure 1

16

Figure 2

17

Fig. 1 (bottom) shows the annual mean Northern Hemisphere energy cycle, and Fig. 2
shows the Northern Hemisphere seasonal cycles. (Note apparent small errors in the
residual calculations in the annual mean.) Looking first at the annual mean, we note:
1. The main input is due to G PM , caused by heating in warmer low latitudes and

( )

2.
3.
4.
5.

6.
7.

cooling in cooler high latitudes. The radiative fluxes cause this both directly and
indirectly by driving low latitude moist convection and condensation.
From PM energy flows mainly to PE , because of baroclinic eddies i.e., their
downgradient heat transport.
PE energy flows mainly to K E , i.e. K E is generated by upward eddy heat fluxes,
i.e. * T * > 0 .
This flow is mainly balanced by dissipation, i.e., mainly by friction near the
surface.
There is however some flow of energy from K E to K M , i.e., there is an
upgradient transport of momentum by the eddies, on balance. In fact this is the
main source of K M .
There is some flow of energy from K M to PM , i.e., on balance T < 0 , i.e. on
balance this conversion is dominated by the indirect Ferrel Cell.
D K E and G PE are relatively small.

( )

( )

Fig. 2 shows the seasonal changes. Note:

1. In winter the dominant energy source is B PM , i.e. PM is mainly being

( )

generated in the Southern Hemisphere, but a lot of it is being transported

across the equator to the Northern Hemisphere.
2. G PM is greater in June-July-August, presumably because tropical moist

( )

convection and the associated LH release is greater then.

3. The eddy convergences are all much weaker in summer and thus the flow
of energy is weaker in summer, and all the energy forms are less.
4. The Ferrel Cell does not dominate C PM , K M in December-January-

February, because it has to contend with the very strong winter Hadley
cell.
Because of the difficulty in calculating the energy cycle (calculating G repeatedly for
example) there have been very few attempts to calculate it from models. The only one I
know of was with an early version of the GISS GCM (Somerville et al, 1974). It used
4 lat x5 long resolution with 9 vertical levels, and a single January was simulated.

( ) (

No boundary fluxes were calculated; the Northern Hemisphere simulated energy cycle is
shown in Fig. 12a in Somerville et al (1974). Comparing with the Peixoto and Oort
winter results, we note
1. The energy cycles are qualitatively similar.
2. Most of the models energies are too high (except for K E )
18

3. Most of the conversions, generation and dissipation terms are too weak, except for
C PM , PE

( )

4. Gratifyingly, G PE is 0.6, close to Oort and Peixotos assumed value.

Other features for the GISS simulation indicate that the differences are not due to the
resolution or short time base. The large K M may be in part because the tropospheric jet
is not closed.
Note that the results for the Energy Cycle tell us that the EP theorem is not at all satisfied
in the trosposhere, because there is a net exchange of energy between the eddies and the
mean flow. C PM , PE is about four times C K E , K M

19