introduction the two schemesthetwo schemes · spatial local search method the spatial adaptive...
TRANSCRIPT
Adaptive parameterisation scheme
An adaptive paramterisation consists of two
parts (visualised in the plate below):
Called in fraction of the time steps, grid boxes
More complex and physical
Can be called by the adaptive generalisation
Generalises the results to the full domain and
time steps
Utilises nearby intrinsic calculations
Uses temporal and spatial correlations
Intrinsic calculation of sub scale processes
Simple (statistical) adaptive generalisation
�
�
�
�
�
�
C o n c l u s i o n s & O u t l o o k
�
�
�
�
�
�
�
The poster illustrates two adaptive schemes
Adaptive spatial method
Adaptive temporal algorithm
The schemes can be over a factor two more
accurate or three times more efficient relative to
the common persistence assumption
We expect that the general idea is useful for other
parameterisations
The two schemes may be combined
More info and an article can be found on:
http://www.meteo.uni-bonn.de/
venema/themes/adaptive_parameterisations/
I n t r o d u c t i o n
Almost any geophysical dynamical model will
need parameterisations for of sub-scale pro-
cesses that take place on scales below the model
resolution. We introduce the term
for a scheme, which uses spatial and temporal
correlations in the resolved geophysical fields to
make the parameterisation computationally more
efficient.
This poster presents two adaptive radiative trans-
fer (RT) parameterisation schemes for the
COSMO numerical weather prediction (NWP)
model.
“adaptive parameterisation scheme”
Te m p o r a l a d a p t i v e s c h e m e
�
�
A simple regression scheme is used to cal-
culate the changes
If the changes become too large, a new
intrinsic calculation is performed
)()()(
)()(
tFttFttF
FtFttF
simplesimplesimple
simpleexact
���
���
�������
�����
Perturbation method:
calculate error-Estimator (� ) based on
a very simpleradiation scheme for
each grid point
apply `perturbationmethod‘ for surface-
fluxes
update 2-streamradiation-scheme
�…‘large‘
… ‘moderate‘�
Spatial local search method
�
�
�
�
�
The spatial adaptive scheme computes theradiation at the other 15 columns bysearching for similar column in the vicinity
Search region 5x5 pixels
Search algorithm minimises a similarityindex:
with, w weights, CCL: cloud cover (low
clouds), CCT: cloud cover (all clouds),
LWP: total column cloud water, : surfacealbedo, t: time
The weights were found by optimisation
The result is not sensitive to the weights
i:
�
|||||||||| 43321 twwLWPwCCTwCCLw ���������� ��
15
16
1
2
3
4
5
6
7
8
910
11
12
13
14
15
16
1
2
3
4
5
6
7
8
910
11
12
13
14
15
16
1
2
3
4
5
6
7
8
910
11
12
13
14
15
16
1
2
3
4
5
6
7
8
910
11
12
13
14
15
16
1
2
3
4
5
6
7
8
910
11
12
13
14
15
16
1
2
3
4
5
6
7
8
910
11
12
13
14
15
16
1
2
3
4
5
6
7
8
910
11
12
13
14
15
16
1
2
3
4
5
6
7
8
910
11
12
13
14
15
16
1
2
3
4
5
6
7
8
910
11
12
13
14
S p a t i a l a d a p t i v e s c h e m e
This scheme uses spatial correlations
(mainly in the cloud field)
In every 4x4 intrinsic region, one intrinsic
(2-stream RT calculation) is performed
every 5 minutes
In which column the
intrinsic calculation
is called depends
on the regular pattern
to the right
�
�
�
The two schemesThe two schemes
COSMO ModelCOSMO Model
Efficient and accurate parameterisations for radiative
transfer that utilise spatial and temporal correlations( )Effizientere und genauere Parametrisierungen für den Strahlungstransport durch Verwendung räumlicher und zeitlicher Korrelationen
Victor Venema, Annika Schomburg, Felix Ament, and Clemens [email protected] -- http://www.meteo.uni-bonn.de/venema
Universität
MeteorologischesInstitut
Bonn
Adaptive parameterisation scheme
Past
Future
Medium complexparameterisation
Adaptive parameterisationscheme
Intrinsiccalculation
Adaptivegeneralisation
Request intrinsic calculation
Results intrinsic calculation
Introduction
Rad ia t i ve t r ans fe r scheme
�
�
�
�
COSMO model uses the delta 2-stream radi-
ative transfer approximation
One hour calls lead to
Physical inconsistencies (plate to the right)
Other weather prediction
The standard scheme can be called "too com-
plex", as it should be executed more often
Adaptive scheme allows the utilisation of the
2-stream method at high temporal resolution
�
�
�
�
�
Liquid and ice cloud water, cloud cover profile, gas
absorption, aerosols, ground albedo
Called once an hour
Costs 5-7 % of calculation time
COSMO-LM weather prediction model
�
�
�
�
�
�
Formerly called the Lokal Modell (LM)
Nonhydrostatic dynamical equations for:
Discretisation: horizontal resolution 2.8 km;
50 vertical layers.
Boundary conditions
from coarse resolution
global model (GME)
Case study:
19th September 2001;
12:30 h UTC
the wind vector, pressure perturbation, air temperature,
specific humidity of water vapour, cloud liquid water and
ice, and precipitation in the form of rain, snow and graupel
P h y s i c a l i n c o n s i s t e n c i e s
�
�
�
Comparison of LM runs with 2-stream
parameterisation with T = 2.5 and 60 minutes
T = 60 minutes (blue curve) leads to too much
cases with high insolation and rain at the same
time compared to the 2.5 min run (red curve)
The horizontal lines
indicate the range
from the 25 to the 75
percentile showing
that also the strength
of the relation
between rain rate and
insolation differs
��
Efficiency adaptive schemes
�
�
The RMS error of the schemes for the solar flux
(a) and the infra-red surface net flux (b) as a
function of the number of intrinsic calculations
Even with a strong reduction in the number of
intrinsic calculations, the adaptive schemes
are still quite accurate
ResultsR e c o n s t r u c t e d s o l a r f l u x
Error of the persistence assumption (g) (i.e. the solar
net flux field of 12 h) and the two adaptive schemes (h)
(i) relative to the "truth" (a), a 2-stream calculation on
the full field at 12:30 h. For comparison the liquid water
path (b) and the surface albedo (c) are shown.
Logarithcloud water
Surfacealbedo
Truesolar flux
Errorpersistence
Errortemporal
Errorspatial
Cor re l a t i ons o f t he e r ro r s
�
�
The two adaptive schemes have a shorter cor-
relation length; (a) autocorrelation function for
solar net flux, (b) idem infra-red
Work on stochastic parameterisations shows
that an error field with a weaker correlations
has less influence on the model dynamics
Adaptive temporal:RMSE:Bias: 6
43 W/mW/m
2
2
Adaptive spatial:RMSE:Bias:
31 W/m2 W/m
2
2
Persistence:RMSE: 77 W/mBias: 5 W/m
2
2
Simple RT parameter isat ion
A multiple regression algorithm is used to cal-
culate the changes in:
infra-red flux at the ground (y) and
solar transmittance (y)
�
�
���� �
N
i
ii xbay1
Main predictors solar (x)i
Cloud free Cloudy
Liquid water path
Cloud cover (low clouds)
Cloud cover (total)
Geometric thickness
Surface albedo * Aerosol
Multiple scattering * Aer.
Surface pressure
Surface albedo