The Thermodynamic Energy Equation and WRF

The governing equations for various vertical

coordinates

In any coordinate system, the total derivative operator may be written as

D
=
Dt

+ v +

(1)

where represents the choice of coordinate system, =

D

D
t

and v = (u, v) is the two-dimensional

cartesian wind vector (Laprise 1992). The Advanced Research WRF (ARW) equations use a terrainfollowing hydrostatic-pressure vertical coordinate called defined by
(2)
ph
ph pht
= p p = pht
hs
ht

where ph is the hydrostatic pressure, pht is the hydrostatic pressure at the top of the model, phs is
the hydrostatic pressure at the model surface, and
= phs pht

(3)

is the mass per unit area within the column in the model domain (x,y). Using this coordinate
system, we can rewrite the total derivative operator defined in (1) as

Dt

D
=

+ v +

(4)

WRF flux-form dry thermodynamic equation

Following Ooyama (1990), we can write prognostic equations in terms of variables that have conservation properties. For the thermodynamic equation, this is
D()
= F
(5)
Dt
where F represents all sources and sinks and represents the mass theta flux. Using (1) we can
expand the above as
() + v +
= F
(6)

or more

concisely
()
+ u = F
(7)
t

where u = (u, v, w) is the three-dimensional cartesian wind vector. Letting

1

U = u
=
=
we can then write (7)

According to vector identity,

(8)
(9)
(10)

+ U = F .

(11)

(aA ) = a A + A a
where here we let a = and A =
U

(12)

so that
(U

) = U + U .

(13)

By mass conservation, = 0 leaving

(U

)=U

(14)

Substituting the above into (11) the flux-form thermodymamic equation becomes

r
+ U

= F.

(15)

The equation of state is

p = RdT.

(16)

where p is the dry air pressure, is the density of air, Rd is the dry air gas constant, and T is the
actual temperature. Letting = 1 the above can be written as

p = RdT

(17)

which solving for p gives

Rd
.
p= T

The definition of potential temperature, is

=T

po

Rd/cp

(18)

(19)

where po is the pressure at 1000 hPa and cp is the specific heat capacity at constant presssure.
Solving the above for temperature, T:
T =

p
po

Substituting the above into (18):

p=
or rearranging

Rd /
cp

Rd/cp

= pRd/cp po

Rd

(20)

pRd/cp p
(21)

Rd/cp
o

Rd

ppRd/cp =
1

R
cpd

Rd
=

Rd/cp

o
Rd/cp

(22)
(23)

Noting that cp = cv + R, then we can note that

1
Rd

and

p
c
Rd
=
p
c

cp
=
Rd
cp

cv
= c =

cp
cv

cv
cp
=
cp
cv

p
c

(24)

(25)

(26)

(27)

cp
1

(28)

where = p
v
c and cv is the specific heat capacity at constant volume. Subsituting (24) and
into (23) gives

(28)

R
p = d po .

(29)

Raising both sides to the power gives

1

Rd

rp
=
p =

(30)

po

d
1
R p

(31)

p = po

d .
R

(32)

po

WRF moist thermodynamic equation

When including moisture, the vertical coordinate from (2) can be written as
pdh pdht
=
=
pdhs pdht

pdh

(33)

pdht
d

where the subscript d now represents that it is for the dry air component and d represents the
mass of dry air in the column. The coupling of the variables to dry mass is retained and written as

U = d u
= d
= d .

(34)
(35)
(36)

In this method, the thermodynamic equation (15) is not directly changed, but instead an
additional conservation equation is added to the governing equations to include the mass mixing
ratios of water in all of its phases:
Qm
= FQm
(37)
+ rU

t
qm
where
Qm = d q m
(38)
qm = qv, qc, qi, qr, qs, etc

and

(39)

are the mass mixing ratios of water vapor, cloud liquid water, cloud ice, rain water, snow, and
any additional mixing ratio class such as graupel for instance. Further, pressure is adjusted to
include moisture via the moist potential temperature (not to be confused with virtual potential
temperature) by re-writing (32) as
p = po

where

m ""

d m
R

(40)

d o

1+

v
R qv = (1 + 1.61qv ).
Rd

(41)

Map factors

The WRF model uses map projections such as the Lambert conformal, Polar strereographic, and
Mercator. This means that the calculations produced on the model plane must be projected onto
the Earths sphere. In general, if we let
be a vector representing the cartesian distance on the
X
models plane, then to obtain the true distance x on the sphere we must make some
transformation by using some scale, say m, which is called the map factor. Using this
notation, we can write changes along the cartesian plane in terms of changes on the Earths
surface as
=

md x

(42)

X
dX
dx=

(43)

This map factor varies but remains close to unity where it is exactly unity at the true latitude of
the map projection. The true velocity vector is
Substituting the above into (34):

u=

(44)

dx
dt

dx
U = d u = d
dt
which using (43)
gives

(45)

d u
U =

(46)
m
and similarly
d
=
.
(47)
m
Applying this to the thermodynamic equation (15), the flux-form moist thermodynamic equation
with map factors m is

2
(U)
+m
t

(V ) + m ()
y

= F .

(48)

Perturbation equations

The WRF equations are modified so that they are written as perturbations from a
hydrostatically- balanced reference state. Perturbations are added to the pressure p, geopotential
, inverse density , and mass . Thus the thermodynamic equation is left unchanged and is
given by (48).

Model integration

6.1 Overview
A time-split time integration scheme is used. The low frequency modes are integrated using
a Runge-Kutta third order scheme while the fast (acoustic) modes are integrated over smaller
timesteps. The Runge-Kutta three step time integration process from t to t+t for potential
temperature is
t
t
R()
= +
(49)

3
t
t
R( )
= +
(50)
2
t+t = t + tR( ).
(51)
(52)
For notational simplicity, we will rewrite the thermodynamic equation (48) as
t + m2 [x(U) + y (V )] + m () = F

(53)

where the subscripts t, x, y and are the respective partial derivatives. The forcing terms, F include potential temperature tendencies from radiation, cumulus parametrization, planetary
bound- ary layer (pbl) schemes, mixing, and diffusion. We can write that as
F = F,radiation + F,cumulus + F,pbl + F,mixing/dif fusion.

(54)

The above is calculated only on the first RK3 step. The microphysics contributions to are calculated in a non-timesplit technique and are therefore calculated outside the Runge-Kutta
timestep- ping loop.
Within the large timestepping Runge-Kutta loop, there is a smaller timestepping acoustic loop
so that the meteorologically insignificant acoustic modes do not limit the RK3 accuracy. This
acoustic timestepping is performed in a correction type manner. Thus to increase the accuracy, a
perturbation form of the governing equations are integrated using smaller acoustic time steps by
defining small time step variables that are deviations from the most recent RK3 predictor
(denoted by the superscript t*). For instance, the thermodynamic equation uses the following:

= t

(55)

t
= V V
= t.

(56)
(57)

Let the right hand side of the tendency from the large timestep which is fixed throughout the
acoustic integration be Rt
Large, . Then from (53):

Rt

Large,

[x(U) + y (V )] m () + F

(58)

which is calculated in each of the RK3 substeps, but NOT within the acoustic integration. The
variables in (58) are used as a beginning reference to which the perturbations in (55) - (57) are
initially built upon entry into the acoustic integration and are not themselves updated within the
acoustic loop. Substituting (55)-(57) into (53) and using (58) the acoustic thermodynamic
equation is
l
t

t
+ m2 x(U t) + y (V t) +
Large,
+

m (
where
+
.
(60)
=

Just as with the right hand side tendency from the large timestep, we can write the right hand side
tendency from the small timestep using (59) as
l
2

=
t ) + y
t ) m
t ).

(61)
Small,

+
R

(U

(V
x

m
Substituting the above into (59) the acoustic tendency is
= R

Small,

+ RLarge,

(62)

which using (60) can be written as

+ = + R

Small,

+ RLarge, l

(63)

in which at the end of the acoustic loop + = t+t. The default number of acoustic
timesteps for RK3 is 4. After the scalars in (37) are updated, the Runge-Kutta loop is then
complete. After the coupled potential temperature is calculated from the above, then it is
adjusted for the microphysics tendencies. If we let the coupled potential temperature
tendencies from the microphysics be t,mp then the final result is
t+t = t+t + t t,mp.

(64)

)=R

(59)

To summarize, the WRF (ARW) calculation flow procedure for potential temperature is as follows:
1. Begin time step
(a) Begin Runge-Kutta three step (RK3) time integration loop (steps 1, 2, 3) using (49)-(51).
i. If RK3 step = 1, calculate time-split physics:
F = t,radiation + t,cumulus + t,pbl + t,mixing/dif fusion.
ii.

Calculate
Rt

iii.

(65)

(= Advection + F) using (58).

Large,

Begin acoustic time step loop (1 n) using (63).

A. If RK3 step = 1, if the number of acoustic timesteps
is 4 (ns = 4), then the
number of iterations is n = 1. After this well have .
B. If RK3 step = 2, if the number
of acoustic timesteps is 4 (n = 4), then the
number of iterations is n = ns = 2. After this well have . s
2
C. If RK3 step = 3, if the number of acoustic timesteps is 4 (ns = 4), thenthe
number of iterations is n = ns = 4. After this well have t+t.

iv.

End acoustic loop.

(b) Update scalar equation using (37).

(c) End (RK3) loop.
(d) Adjust for the non time-split schemes (only microphysics) using (64).
2. End time step. Return to step 1.

6.2 Some notes on calculating tendencies

To calculate the tendenices in WRF from (54), one must use extreme caution due to the coupling
of the tendencies with the total dry air mass. For instance, the radiation component, F,radiation
may be expanded as
F,radiation = F,lw + F,sw
(66)
where F,lw and F,sw are the longwave and shortwave radiation components respectively.
WRF calculates the individual components F,lw and F,sw within the radiation_driver
module, but then only carries their sum F,radiation into the model integration. The
calculations for radiation are not coupled with total dry mass when passing through this
subroutine. However, the total radiation term must be coupled with total dry mass in order
to be integrated. Therefore, if you add the total radiation tendency to the WRF Registry
(RTHRATEN), you must decouple it from the total dry air mass. The dry air mass in terms of
WRF variables is
mut = mu + mub

(67)

where mu is the perturbation dry air mass and mub is the background dry air mass. However, the
shortwave and longwave radiation components are NEVER coupled with dry air mass, so they
are not coupled when written out of WRF.
The WRF ARW technically carries the microphysics tendency term when integrating
through the timesteps, but removes it in the end. This is done to prevent the excitation of
acoustic waves

WRF Variable
name
RTHRATEN
RTHRATENLW
RTHRATENSW
RTHBLTEN
RTHCUTEN
H_DIABATIC
T_TENDF

Description
t from total radiation in K/s
t from longwave radiation in K/s
t from shortwave radiation in K/s
t from planetary boundary layer scheme in K/s
t from cumulus scheme in K/s
t from microphysics scheme in K/s
t from physics in K/s

Coupled with
MUT?
Yes
No
No
Yes
Yes
No
Yes

Table 1: Summary of potential temperature tendency variable names in WRF ARW model. Since
the ARW integration involves equations in mass-flux form, the tendencies that are computed in
the time stepping themselves are coupled with the total dry air mass while variables that are only
calculated individually within physics or microphysics schemes are not coupled.

while retaining the balances produced by the physics at the end. The inclusion of the diabatic
forcing tendency estimate used in the RK3 integration is from the previous timestep. It is
necessary to include it here to relieve the acoustic noise that would be produced otherwise. At the
same time, it is important that the physics be balanced at the end of the timestep. Thus this
microphysics term can be thought of in relation to F as rewritting (58) as
[x(U) + y (V )] m () + F,physics + F,mp
(68)
Large, =
t
2
Rm

where F,physics, which is called T_TENDF in WRF, is given by (54) and F,mp is the microphysics
tendency given by
F,mp = d

t+t

t
(69)
t
where d is the total dry air mass. The above term, when uncoupled is carried in WRF as variF
able H_DIABATIC or H_DIABATIC = d,mp . However, after the timestep, and before updating the
microphysics, WRF subtracts off this tendency with
t+t = t+t ns t d H_DIABATIC

(70)

where ns is the number of timesteps.

Table 1 gives a summmary of the WRF ARW potential temperature tendency terms that can
be written out when added to the Registry. The only adjustments that need to be accounted for
are the couplings with the total dry air mass. No timestep adjustments are necessary, as they are
already divided by the model timestep when written out.

Stability and the CFL condition

7.1 RK3 timestep

Limitations to the timestep used in the RK3 integration arise through the advective Courant
number
( ut ), which depends on the advection order chosen by the user. Currently options for
x
the advection
scheme are from 2nd order through 6th order. The suggested scheme is the 5th
order scheme, of which it was shown by Wicker and Skamarock (2002) that the maximum stable
8

Courant number for 1D advection (while using the RK3 timestep scheme) is 1.42. Since the
spatial dimension is 3 in the model, we must therefore scale the Courant number from the 1D
case by a factor of 1 . That is,
3
Cr
ut
(71)
x =

3
where Cr is the Courant number as found by Wicker and Skamarock (2002) for 1D advection.
Solving the above for the timestep t:
tmax

Cr x
/
ma
(3) x

(72)

u
1.42 x

3 umax
x
0.8198
.
umax

(73)
(74)

A general guideline for a maximum timestep is:

t6 x
"" 1000

(75)

where x is the grid spacing in meters. This will allow stability for up to a 136 m/s jet.

7.2 Acoustic timestep

The maximum Courant number for the acoustic timestep is
Cr <

cs
1
<
x
2

(76)

where cs = 300ms1 is the speed of sound. To get a more conservative estimate of the maximum
acoustic timestep, we can replace 1 with 1 to get an acoustic timestep constraint of
(2)

< 2

<

(77)

cs

150

where x is in meters. Generally, the acoustic timestep should be approximately 1/4 of the RK3
timestep. Therefore the default number of acoustic timesteps should be 4.

7.3 Radiation timesteps

The recommended timestep is 1 minute per km dx. For example, if horizontal grid resolution is 30
km, then the radiation timestep should be 30 minutes.

7.4 Cumulus and planetary boundary layer scheme timesteps

Generally set to zero minutes, which means it is updated with the model large (RK3) timestep
from Section 7.1.
9