1 Numerical Weather Prediction Parameterization of diabatic processes Convection III The ECMWF convection scheme Peter Bechtold and Christian Jakob
Mar 27, 2015
1
Numerical Weather Prediction Parameterization of diabatic processes
Convection IIIThe ECMWF convection scheme
Peter Bechtold and Christian Jakob
2
A bulk mass flux scheme:What needs to be considered
Entrainment/Detrainment
Downdraughts
Link to cloud parameterization
Cloud base mass flux - Closure
Type of convection shallow/deep
Where does convection occur
Generation and fallout of precipitation
3
Basic Features
• Bulk mass-flux scheme• Entraining/detraining plume cloud model• 3 types of convection: deep, shallow and mid-level - mutually
exclusive• saturated downdraughts• simple microphysics scheme• closure dependent on type of convection
– deep: CAPE adjustment– shallow: PBL equilibrium
• strong link to cloud parameterization - convection provides source for cloud condensate
4
Large-scale budget equations: M=ρw; Mu>0; Md<0
)()()( FMLeecLsMMsMsMp
gt
sfsubclddududduu
cu
Mass-flux transport in up- and downdraughts
condensation in updraughts
Heat (dry static energy): Prec. evaporation in downdraughts
Prec. evaporation below cloud base
Melting of precipitation
Freezing of condensate in updraughts
Humidity:
subclddududduucu
eecqMMqMqMp
gt
q
)(
5
Large-scale budget equations
Cloud condensate:
u ucu
lD l
t
uMMuMuMp
gt
ududduu
cu
)(
Momentum:
vMMvMvMp
gt
vdudduu
cu
)(
uucu
lDat
a)1(
Cloud fraction:
(supposing fraction 1-a of environment is cloud free)
6
Large-scale budget equations
Nota: These tendency equations have been written in flux form which by definition is conservative. It can be solved either explicitly (just apply vertical discretisation) or implicitly (see later).
Other forms of this equation can be obtained by explicitly using the derivatives (given on Page 10), so that entrainment/detrainment terms appear. The following form is particular suitable if one wants to solve the mass flux equations with a Semi-Lagrangian scheme;
note that this equation is valid for all variables T, q, u, v, and that all source terms (apart from melting of precipitation term) have cancelled out )()()(
dduuducu
DDp
MMgt
7
Occurrence of convection:make a first-guess parcel ascent
Updraft Source Layer
LCL
ETL
CTL
1) Test for shallow convection: add T and q perturbation based on turbulence theory to surface parcel. Do ascent with w-equation and strong entrainment, check for LCL, continue ascent until w<0. If w(LCL)>0 and P(CTL)-P(LCL)<200 hPa : shallow convection
2) Now test for deep convection with similar procedure. Start close to surface, form a 30hPa mixed-layer, lift to LCL, do cloud ascent with small entrainment+water fallout. Deep convection when P(LCL)-P(CTL)>200 hPa. If not …. test subsequent mixed-layer, lift to LCL etc. … and so on until 700 hPa
3) If neither shallow nor deep convection is found a third type of convection – “midlevel” – is activated, originating from any model level above 500 m if large-scale ascent and RH>80%.
8
Cloud model equations – updraughtsE and D are positive by definition
uuu DEp
Mg
uuuuuu LcsDsE
p
sMg
uuuuuu cqDqE
p
qMg
uPuuuuuu GclDlE
p
lMg ,
uuuuu uDuE
p
uMg
uuuuu vDvE
p
vMg
2,1
(1 )2 , (1 ) 2
v u vu u ud u u
u v
T TK E wC K g K
z M f T
Kinetic Energy (vertical velocity) – use height coordinates
Momentum
Liquid Water/Ice
Heat Humidity
Mass (Continuity)
9
Downdraughts
1. Find level of free sinking (LFS)
highest model level for which an equal saturated mixture of cloud and environmental air becomes negatively buoyant
2. Closure, , 0.3d LFS u bM M
3. Entrainment/Detrainment
turbulent and organized part similar to updraughts (but simpler)
10
Cloud model equations – downdraughtsE and D are defined positive
ddd DEp
Mg
dddddd LesDsE
p
sMg
dddddd eqDqE
p
qMg
ddddd uDuE
p
uMg
ddddd vDvE
p
vMg
Mass
Heat
Humidity
Momentum
11
Entrainment/Detrainment (1)
u u uu u org turb org
M M Mg E D
p
1 1 2
00
3
3 1 4 11 2
1.3 1.3 ;
1.75 10 ; 0.75 10 ;
turbs
buoybuoy
s
sbase
qc F c RH F c
q
qc m c m F
q
ε and δ are generally given in units (m-1) since (Simpson 1971) defined entrainment in plume with radius R as ε=0.2/R ; for convective clouds R is of order 500-1000 m for deep and R=50-100 m for shallow
Scaling function to mimick a cloud ensemble
Constants
NB: This is a simple 1-RH or saturation deficit formulation for the organised entrainment, but any other formulation using buoyancy or also works
,e es
turb
org
12
Entrainment/Detrainment (2)
Organized detrainment:
zzK
zK
zzM
zM
u
u
u
u
)(
Only when negative buoyancy (K decreases with height), compute mass flux at level z+Δz with following relation:
with
2
2u
u
wK
anduuD
u
uu Bf
KCM
E
z
K
)1(
12)1(
13
Precipitation
2
10,
crit
u
l
l
uu
uuP elw
cMG
Generation of precipitation in updraughts
Simple representation of Bergeron process included in c0 and lcrit
Liquid+solid precipitation fluxes:
gdpMelteeGpP
gdpMelteeGpP
P
Ptop
snowsubcld
snowdown
snowsnow
P
Ptop
rainsubcld
raindown
rainrain
/)()(
/)()(
Where Prain and Psnow are the fluxes of precip in form of rain and snow at pressure level p. Grain and Gsnow are the conversion rates from cloud water into rain and cloud ice into snow. The evaporation of precip in the downdraughts edown, and below cloud base esubcld, has been split
further into water and ice components. Melt denotes melting of snow.
14
Precipitation
uu
precufallout r
zw
VMS
Fallout of precipitation from updraughts
2.0, 32.5 urainprec rV 2.0
, 66.2 uiceprec rV
Evaporation of precipitation
1. Precipitation evaporates to keep downdraughts saturated
2. Precipitation evaporates below cloud base
3
12
, assume a cloud fraction 0.05surf
subcld s
p p Pe RHq q
15
Closure - Deep convection
Convection counteracts destabilization of the atmosphere by large-scale processes and radiation - Stability measure used: CAPE
assume that convection reduces CAPE to 0 over a given timescale, i.e.,
CAPECAPE
t
CAPE
cu
0
• Originally proposed by Fritsch and Chappel, 1980, JAS
• Implemented at ECMWF in December 1997 by Gregory (Gregory et al., 2000, QJRMS), using a constant time-scale that varies only as function of model resolution (720s T799, 1h T159)
• The time-scale is a very important quantity and has been changed in Nov. 2007 to be
equivalent to the convective turnover time-scale which is defined by the cloud thickness divided by the cloud average vertical velocity, and further scaled by a factor depending linearly on horizontal model resolution (it is typically of order 1.3 for T799 and 2.6 for T159)
The quantity required by the parameterization is the cloud base mass-flux. How can we get this?
H
u nH W
16
Closure - Deep convection
cv v
vcloud
CAPE g dz
dzttgt
CAPE
cloudv
vcv
cv
v
cu
2
Assume:
1 and cloud, statesteady i.e., ,0
v
cv
cv
t
dzt
gt
CAPE
cu
v
cloudvcu
1
17
Closure - Deep convection
z
M
tvc
cu
v
i.e., ignore detrainment
CAPE
dzz
Mg
t
CAPE v
cloudv
c
cu
dtdubuduc MMMMM ,, utd MM ,
1
, ,11 1
nu b u b
nv vu d
v vcloud cloud
CAPE CAPE
M Mg dz g M dz
z z
where Mn-1 are the mass fluxes from a previous first guess updraft/downdraft computation
18
Closure - Shallow convection
Based on PBL equilibrium : what goes in must go out - including downdraughts
0
0cbase
conv
turb dyn rad
w h h h hdz
z t t t
0
0cbase h
dzt
With
,,
(1 ) ; / ;u b u d u dcbaseconv cbasew h M h h h M M and
00, convhw
0
,(1 )
cbase
turb dyn rad
u b
u d cbase
h h hdz
t t tM
h h h
19
Closure - Midlevel convection
Roots of clouds originate outside PBL
assume midlevel convection exists if there is large-scale ascent, RH>80% and there is a convectively unstable layer
Closure:bbu wM ,
20
Vertical Discretisation
k
k+1/2
k-1/2
(Mulu)k+1/2
(Mulu)k-1/2
Eul Eul
Dulu Dulu
cu
GP,u
(Mul)k-1/2(Mul)k-1/2
(Mul)k+1/2 (Mul)k+1/2
Fluxes on half-levels, state variable and tendencies on full levels
21
Numerics: solving Tendency = advection equation explicit solution
( ) ;u u
conv
g M St p
Pr
pS
1/ 2 1/ 2 1/ 2 1/ 2 1/ 2 1/ 2u u u u u ukk k k k k k
conv
gM M M M S
t p
k k -1ψ ψ
2/12/1 kk PPp
if ψ = T,q
In order to obtain a better and more stable “upstream” solution (“compensating subsidence”, use shifted half-level values to obtain:
k -1/2 k -1ψ ψ
Use vertical discretisation with fluxes on half levels (k+1/2), and tendencies on full levels k, so that
1/ 2 1/ 2 1/ 2 1/ 2 1/ 2 1/ 2u u u u u ukk k k k k k k
conv
gM M M M S
t p
k+1/2 k -1/2ψ ψ
22
Numerics: implicit advection
( ) ;u u
conv
g M St p
Prp
S
1/2 1/2 1/2 1/2 1/2 1/2
1/2 1/2 1/2 1/2 1/2 1/2
n u u u u u u nk k k k k k k k
u u n+1 u u u u nk k k -1 k k k k k
tψ -ψ g M M M M tS
p
t(1+M )ψ -M ψ g M M tS
p
n 1 n 1 n 1k k k -1
n 1k
ψ ψ
2/12/1 kk PPp
if ψ = T,q
=> Only bi-diagonal linear system, and tendency is obtained
as
For “upstream” discretisation as before one obtains:
k -1/2 k -1ψ ψ
Use temporal discretisation with on RHS taken at future time and not at current time
n1n
1n nk k k
convt t
23
Numerics: Semi Lagrangien advection
;)( SMp
gt
uu
Pr
pS
if ψ = T,q
;)(
uuu D
p
g
pgM
tdt
d
Advection velocity
( ) ;dep u u
conv
gD
t t p
)( tgMPdep u
27
Tracer transport experiments(1) Single-column simulations: Stability
Surface precipitation; continental convection during ARM
28
Tracer transport experiments(1) Stability in implicit and explicit advection
instabilities
• Implicit solution is stable.
• If mass fluxes increases, mass flux scheme behaves like a diffusion scheme: well-mixed tracer in short time
29
Tracer transport experiments(2) Single-column against CRM
Surface precipitation; tropical oceanic convection during TOGA-COARE
30
Tracer transport experiments(2) IFS Single-column and global model against CRM
Boundary-layer Tracer
• Boundary-layer tracer is quickly transported up to tropopause
• Forced SCM and CRM simulations compare reasonably well
• In GCM tropopause higher, normal, as forcing in other runs had errors in upper troposphere
31
Tracer transport experiments(2) IFS Single-column and global model against CRM
Mid-tropospheric Tracer
• Mid-tropospheric tracer is transported upward by convective draughts, but also slowly subsides due to cumulus induced environmental subsidence
• IFS SCM (convection parameterization) diffuses tracer somewhat more than CRM
IFS small aqua planet configuration: to see how far we can go with parametrised vs explicit global convection
Example Vis5D: explicit simulation of deep convective system at T159 (with R/40 ~ Δx<3 km)
• Without parametrised convection the IFS results are quite sensitive to the numerics and to the tuning of the microphysics.
• Want to exploit convection globally in the 1-16 km resolution range using T159 (125 km) and T1279 (16 km) truncation with R/10-> dx=125 – 1.5 km range