ecmwf numerics of physical parametrizations slide 1 numerics of physical parametrizations by nils...
TRANSCRIPT
![Page 1: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/1.jpg)
ECMWFNumerics of Physical Parametrizations Slide 1
Numerics of physical parametrizationsby Nils Wedi (room 007; ext. 2657)
![Page 2: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/2.jpg)
ECMWFNumerics of Physical Parametrizations Slide 2
Overview
Introduction to ‘physics’ and ‘dynamics’ in a NWP model from a
numerical point of view
Potential numerical problems in the ‘physics’ with large time
steps
Coupling Interface within the ‘physics’ and between ‘physics’ and
‘dynamics’
Concluding Remarks
References
![Page 3: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/3.jpg)
ECMWFNumerics of Physical Parametrizations Slide 3
Introduction...
‘physics’, parametrization: “the mathematical procedure describing the statistical effect of subgrid-scale processes on the mean flow expressed in terms of large scale parameters”, processes are typically: vertical diffusion, orography, cloud processes, convection, radiation
‘dynamics’: computation of all the other terms of the Navier-Stokes equations (eg. in IFS: semi-Lagrangian advection)
![Page 4: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/4.jpg)
ECMWFNumerics of Physical Parametrizations Slide 4
Different scales involved
![Page 5: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/5.jpg)
ECMWFNumerics of Physical Parametrizations Slide 5
Introduction...
Goal: reasonably large time-step to save CPU-time without loss of accuracy
Goal: numerical stability
achieved by…treating part of the “physics” implicitly
achieved by… splitting “physics” and “dynamics” (also conceptual advantage)
![Page 6: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/6.jpg)
ECMWFNumerics of Physical Parametrizations Slide 6
Introduction...
Increase in CPU time
substantial if the time step
is reduced for the ‘physics’
only.
Iterating twice over a time
step almost doubles the
cost.
![Page 7: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/7.jpg)
ECMWFNumerics of Physical Parametrizations Slide 7
Introduction...
Increase in time-step in the ‘dynamics’ has prompted the question of accurate and stable numerics in the ‘physics’.
![Page 8: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/8.jpg)
ECMWFNumerics of Physical Parametrizations Slide 8
Potential problems...
Misinterpretation of truncation error as missing physicsChoice of space and time discretizationStiffnessOscillating solutions at boundariesNon-linear termscorrect equilibrium if ‘dynamics’ and ‘physics’ are splitted ?
![Page 9: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/9.jpg)
ECMWFNumerics of Physical Parametrizations Slide 9
Equatorial nutrient trapping – example of ambiguity of (missing) “physics” or “numerics”
Effect of insufficient vertical resolution and the choice of the numerical (advection) scheme
A. Oschlies (2000)
observation
Model A Model B
![Page 10: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/10.jpg)
ECMWFNumerics of Physical Parametrizations Slide 10
Discretization
Choose:
),(1
,duMuFzt
u
MMM
with
MuuMMMuF
jj
dudu
11
, ,)(),(
z
uuM
t
uu nj
nj
nj
nj
211
1
e.g. Tiedke mass-flux scheme
![Page 11: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/11.jpg)
ECMWFNumerics of Physical Parametrizations Slide 11
Discretization
Stability analysis, choose a solution: )(
0knzjiknn
j euu
)(sin1
21
22 zk
ee
z
tM
ztM
zikzik
1 Always unstable as above solution diverges !!!
![Page 12: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/12.jpg)
ECMWFNumerics of Physical Parametrizations Slide 12
Stiffness
hrthrKt
tF
tFeFutu
dt
tdFtFtuK
dt
tdu
Kt
in ; 100 ; 5.010
)(
:choose
)()0()0()(
:solutionexact
0K 0,u(0); )(
)()()(
1
slowfast
e.g. on-off processes, decay within few timesteps, etc.
![Page 13: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/13.jpg)
ECMWFNumerics of Physical Parametrizations Slide 13
Stiffness
11
n
1
stability analysis of homogeneous part:
1 1ˆu(n t)
1
conditionally stable, explicit ( 0) 2
unconditionally stable, implicit ( 1) 1
stable, bu
n nn n n nu u
K u u KF Ft
tKu
tK
K t
t oscillating, trapezoidal ( 0.5) 1
![Page 14: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/14.jpg)
ECMWFNumerics of Physical Parametrizations Slide 14
Pic of stiffness here
Stiffness: k = 100 hr-1, dt = 0.5 hr
5.0
1
![Page 15: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/15.jpg)
ECMWFNumerics of Physical Parametrizations Slide 15
Stiffness
No explicit scheme will be stable for stiff systems.
Choice of implicit schemes which are stable and non-oscillatory is extremely restricted.
![Page 16: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/16.jpg)
ECMWFNumerics of Physical Parametrizations Slide 16
Boundaries
2
2
0
( , ) ( , ) ; 0
boundary conditions:
(0, ) 0, ( , ) 0
initial condition:
( ,0) ,0
T z t T z t
t z
T t T Z t
T z T z Z
Classical diffusion equation, e.g. tongue of warm aircooled from above and below.
![Page 17: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/17.jpg)
ECMWFNumerics of Physical Parametrizations Slide 17
Boundaries
11 2 2
2 2
21 1
2 2
(1 )
2
n nn n
i i i
i
T T T T
t z z
T T TT
z z
![Page 18: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/18.jpg)
ECMWFNumerics of Physical Parametrizations Slide 18
Pic of boundaries here
sdt 3600,1 sdt 3600,5.0
T0=160C , z=1600m, =20m2/s, dz=100m, cfl=*dt/dz2
![Page 19: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/19.jpg)
ECMWFNumerics of Physical Parametrizations Slide 19
Non-linear terms
1( )( ) ( )
temperature difference between ground and air
exchange coefficient, 10 3
( ) 1 sin(2 / 24) , diurnal cycle
P
P
T tKT t D t
tT
KT K ,P
D t n t
e.g. surface temperature evolution
![Page 20: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/20.jpg)
ECMWFNumerics of Physical Parametrizations Slide 20
Non-linear terms
nnPnn
nPnn
DTTKt
TT
DTTKt
-TT
11
correctorpredictor
~
~)(
~
![Page 21: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/21.jpg)
ECMWFNumerics of Physical Parametrizations Slide 21
Non-linear terms
11
0
( ) (1 )
linearized stability analysis:
1 ( 1 ) ,
1
1 , unconditionally stable, over-implicit
n nn P n n nT T
K T T T Dt
PKT t
P
![Page 22: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/22.jpg)
ECMWFNumerics of Physical Parametrizations Slide 22
Pic of non-linear terms here
predic-corr
4
1
non-linear terms: K=10, P=3
![Page 23: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/23.jpg)
ECMWFNumerics of Physical Parametrizations Slide 23
Negative tracer concentration – Vertical diffusion
Negative tracer concentrations noticed despite a quasi-monotone advection scheme
(Anton Beljaars)
![Page 24: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/24.jpg)
ECMWFNumerics of Physical Parametrizations Slide 24
Physics-Dynamics couplingVertical diffusion
Single-layerproblem
(Kalnay and Kanamitsu, 1988)
dynamics positive definite
![Page 25: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/25.jpg)
ECMWFNumerics of Physical Parametrizations Slide 25
Physics-Dynamics couplingVertical diffusion
Two-layerproblem
Not positive definite depends on !!!
dynamics positive definite
![Page 26: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/26.jpg)
50 55 60Model level
-1e-14
0
1e-14
2e-14
3e-14
4e-14
Aer
osol
con
cent
ratio
n
old time levelafter dynamics only new time level
72.3N/2.5E
50 55 60Model level
-1e-14
0
1e-14
2e-14
3e-14
4e-14
Aer
osol
con
cent
ratio
n
old time levelafter dynamics only new time level
72.3N/2.5E
(D+P)t+t
Dt+t
(D+P)t
= 1.5
= 1
Anton Beljaars
Negative tracer concentrationwith over-implicit formulation
![Page 27: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/27.jpg)
ECMWFNumerics of Physical Parametrizations Slide 27
Wrong equilibrium ?
, ( ) , .
correct steady state solution:
TD P P T gT g const
t
DT
g
![Page 28: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/28.jpg)
ECMWFNumerics of Physical Parametrizations Slide 28
Compute D+P(T) independant
1
11
1.
2. (1 )
add 1. 2. together and seek steady state solution:
explicit 0 :
implicit ( 1 : (1 ),
n nn
n nn n
n
n
T T TD D
t t
T T TgT g T T
t t
D(γ ) T
g
Dγ ) T g t
g
wrong!
![Page 29: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/29.jpg)
ECMWFNumerics of Physical Parametrizations Slide 29
Compute P(D,T)
1
11
1.
2. (1 )
seek steady state solution of 2. :
explicit 0 :
implicit ( 1 : ,
n nn
n nn n n
n
n
T T TD D
t t
T T TD gT D g T T
t t
D(γ ) T
g
Dγ ) T
g
correct!
![Page 30: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/30.jpg)
ECMWFNumerics of Physical Parametrizations Slide 30
Coupling interface...
Splitting of some form or another seems currently the only practical way to combine “dynamics” and “physics” (also
conceptual advantage of “job-splitting”)
Example: typical time-step of the IFS model at ECMWF (2TLSLSI)
![Page 31: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/31.jpg)
ECMWFNumerics of Physical Parametrizations Slide 31
two-time-level-scheme
![Page 32: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/32.jpg)
ECMWFNumerics of Physical Parametrizations Slide 32
Sequential vs. parallel splitvdif - dynamics
12 15 18 21 24 27 30 33 36Forecast step (hours)
0
5
10
15
U (
m/s
)
parallel split (ej4k)sequential split (ej4n)bad sequential split (ej4x)sequential split, dt=5 min (ej4m)
(90 W, 60 S) T159 forecasts 2002011512, dt=60 min
parallel split
sequential split
A. Beljaars
![Page 33: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/33.jpg)
ECMWFNumerics of Physical Parametrizations Slide 33
‘dynamics’-’physics’ coupling
gwdragvdifcloudconvradcloudconvrad
t
PPPP
tOgtPgttPgtP
PRGGt
FF
2
1
2
1
))((),(),(2
1),(
2
1
02/1
2022
2/1
2/12/100
!!!box black anot is P
![Page 34: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/34.jpg)
ECMWFNumerics of Physical Parametrizations Slide 34
Noise in the operational forecasteliminated through modified coupling
![Page 35: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/35.jpg)
Splitting within the ‘physics’
‘Steady state’ balance betweendifferent parameterized processes: globally
P. Bechtold
![Page 36: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/36.jpg)
Splitting within the ‘physics’
‘Steady state’ balance betweendifferent parameterized Processes: tropics
![Page 37: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/37.jpg)
ECMWFNumerics of Physical Parametrizations Slide 37
Splitting within the ‘physics’
“With longer time-steps it is more relevant to keep an accurate balance between processes within a single time-step” (“fractional stepping”)
A practical guideline suggests to incorporate slow explicit processes first and fast implicit processes last (However, a rigid classification is not always possible, no scale separation in nature!)
parametrizations should start at a full time level if possible otherwise implicitly time-step dependency introduced
Use of predictor profile for sequentiality
![Page 38: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/38.jpg)
ECMWFNumerics of Physical Parametrizations Slide 38
Splitting within the ‘physics’
ΔtPΔtPP*Δt*PFFgwdragvdifradradcloudconvDpredict
)(5.0
~ 00
Predictor-corrector scheme by iterating each time step and use these predictor values as input to the parameterizations and the dynamics.Problems: the computational cost, code is very complex and maintenance is a problem, tests in IFS did not proof more successful than simple predictors, yet formally 2nd order accuracy may be achieved ! (Cullen et al., 2003, Dubal et al. 2006)
Previous operational:
Currently operational:
, ,
,
*predict conv D cloud guess rad vdif gwdrag
predict cloud D conv rad vdif gwdrag
F F P t P t
F F P t
Choose: =0.5 (tuning) Inspired by results!
![Page 39: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/39.jpg)
ECMWFNumerics of Physical Parametrizations Slide 39
Concluding Remarks
Numerical stability and accuracy in parametrizations are an important issue in high resolution modeling with reasonably large time steps.
Coupling of ‘physics’ and ‘dynamics’ will remain an issue in global NWP. Solving the N.-S. equations with increased resolution will resolve more and more fine scale but not down to viscous scales: averaging required!
Need to increase implicitness (hence remove “arbitrary” sequentiality of individual physical processes, e.g. solve boundary layer clouds and vertical diffusion together)
Other forms of coupling are to be investigated, e.g. embedded cloud resolving models and/or multi-grid solutions (solving the physics on a different grid; a finer physics grid averaged onto a coarser dynamics grid can improve the coupling, a coarser physics grid does not !) (Mariano Hortal + Agathe Untch, work in progress)
![Page 40: ECMWF Numerics of Physical Parametrizations Slide 1 Numerics of physical parametrizations by Nils Wedi (room 007; ext. 2657)](https://reader036.vdocument.in/reader036/viewer/2022081518/55154be755034673228b6284/html5/thumbnails/40.jpg)
ECMWFNumerics of Physical Parametrizations Slide 40
References Dubal M, N. Wood and A. Staniforth, 2006. Some numerical properties of approaches to physics-
dynamics coupling for NWP. Quart. J. Roy. Meteor. Soc., 132, 27-42.
Beljaars A., P. Bechtold, M. Koehler, J.-J. Morcrette, A. Tompkins, P. Viterbo and N. Wedi, 2004. Proceedings of the ECMWF seminar on recent developments in numerical methods for atmosphere and ocean modelling, ECMWF.
Cullen M. and D. Salmond, 2003. On the use of a predictor corrector scheme to couple the dynamics with the physical parameterizations in the ECMWF model. Quart. J. Roy. Meteor. Soc., 129, 1217-1236.
Dubal M, N. Wood and A. Staniforth, 2004. Analysis of parallel versus sequential splittings for time-stepping physical parameterizations. Mon. Wea. Rev., 132, 121-132.
Kalnay,E. and M. Kanamitsu, 1988. Time schemes for strongly nonlinear damping equations. Mon. Wea. Rev., 116, 1945-1958.
McDonald,A.,1998. The Origin of Noise in Semi-Lagrangian Integrations. Proceedings of the ECMWF seminar on recent developments in numerical methods for atmospheric modelling, ECMWF.
Oschlies A.,2000. Equatorial nutrient trapping in biogeochemical ocean models: The role of advection numerics, Global Biogeochemical cycles, 14, 655-667.
Sportisse, B., 2000. An analysis of operator splitting techniques in the stiff case. J. Comp. Phys., 161, 140-168.
Termonia, P. and R. Hamdi, 2007, Stability and accuracy of the physics-dynamics coupling in spectral models, Quart. J. Roy. Meteor. Soc., 133, 1589-1604.
Wedi, N., 1999. The Numerical Coupling of the Physical Parameterizations to the “Dynamical” Equations in a Forecast Model. Tech. Memo. No.274 ECMWF.