Hybrid Variational-Ensemble Data Assimilation - Weather-Chaos
One-Dimensional Variational Assimilation of SSM/I Observations … · 2006-10-27 ·...
Transcript of One-Dimensional Variational Assimilation of SSM/I Observations … · 2006-10-27 ·...
One-Dimensional VariationalAssimilation of SSM/IObservations in RainyAtmospheres atEnvironment Canada (EC)Science & Technology Branch
G. DeblondeJ.-F. MahfoufB. BilodeauD. Anselmo6 October 2006
www.ec.gc.ca
10/6/2006 Page 2
Motivation
• Development at EC of an assimilation system ofsatellite radiances in cloudy and rainy regions (overopen oceans)
• Operational NWP centers– JMA : Mesoscale 4D-Var assimilation of radar
precipitation– NCEP : Global 3D-Var assimilation SSM/I and TMI
derived rainfall rates– ECMWF : Towards a global 4D-Var assimilation of
SSM/I radiances in rainy areas• Two step method (1D-Var+4D-Var) operational since
June 2005 (Bauer et al. 2006 a and b, QJRMS)
10/6/2006 Page 3
Proposed strategy
ObservationsMW Brightness
Temperature
Control variables Profiles of T and Q
1D-Var
Observation operator (1)Precipitation Schemes
ObservationsSurface
Rainfall Rate
Observation operator (2)Radiative Transfer Model
Integrated Water Vapor
Marécal and Mahfouf (2000,2002) -SRRMoreau et al. 2004 -Tb
QC 4D-Var
10/6/2006 Page 4
Experimental set-up• Control variable : T and ln(q) profiles (58 levels)• Observations : SSM/I brightness temperatures (Tb) or derived
surface rain rates (SRR) (Bauer et al. 2002)• Observation operator :
– Moist physical processes : same as in EC globalforecast model (GEM) –jacobians obtained by finitedifference
– Radiative transfer model : RTTOV with scattering effects(Bauer and Moreau, 2002) –jacobians obtained by adjointmethod
• Background error statistics : “NMC” method (lagged forecasts)as in EC operational 4D-Var (incremental)
• Two case studies (40S-40N):– Tropical Cyclone Zoe (F15) & Typhoon Chaba (F14)
10/6/2006 Page 5
Background Term
• Forecast Model:
– Global Environmental Multi-Scale (GEM) MESOGLOBAL -research– Model Resolution: 0.45o longitude x 0.3o latitude grid– 58 vertical levels, model lid at 10 hPa, time-step = 15 minutes
• Moist physical schemes:
– Shallow Convection: KuoTrans -> only cloud liquid water– Non-convective (or stratiform): CONSUN (Sundqvist variant)– Deep Convection: Kain-Fritsch (CAPE) ->highly non-linear
• 12-h precipitation spin-up: 1D-Var background field = 12-h forecast
10/6/2006 Page 6
HclearVclear
HV
TbTb
TbTbP
3737
3737
!
!"
)cos(/22 !"# $%& eP
• P is a measure of visibility of sea-surface relative to expected value inabsence of clouds P=0 => completely opaque rain cloud P=1 => cloud-free ocean sceneIf P > 0.15 ( =Transmittance > 0.4) then Use (19V,H,22V, 37 V,H)Else Use 19V,H, 22V ONLY
!
(Petty 1994)
Number of SSM/I Tb channels assimilated
10/6/2006 Page 7
Brightness Temperature (K)at 19 GHz V 2002 12 27 000 UTC
Tropical Cyclone Zoe
GEM Mesoglobal 12h Forecast SSM/I F15 ObservationsK
10/6/2006 Page 8
Experiments
• 1D-Var Tb – Small ObservationErrors (SOE)– (O+F) = 3 K for V channels &
6 K for H channels (as inMoreau et al. 2004)
• 1D-Var Tb – Large ObservationErrors (LOE)– (O+F) = HBHT
• 1D-Var SRR (Surface RainfallRate) –Observation errorprovided by retrieval algorithm
0
10
20
30
40
50
60
70
19v 19h 22v 37v 37h
SSM/I channel
Ob
serv
ati
on
Err
or
(K)
SOE
LOE
1D-Var Tb Observation Error (K)
10/6/2006 Page 9
Zoe Case: 0000 UTC 27 Dec 2002
N=1414 for 3 channels –more opaqueN=3098 for 5 channels –less opaque
(HBHT)1/2 in Kelvin
19V 19H
22V 37V
37H
10/6/2006 Page 10
TC Zoe: Analyzed Surface Rain Rate (mm/h)
1D-Var LOE
b)
c)
a)
d)
1D-Var SOE
1D-Var SRR PATER observations SSM/I F15
10/6/2006 Page 11
Analyzed fields vs SSM/I retrievals basedon regression equations
O=LWP (Weng & Grody, 1994)
O=IWV (Alishouse et al. 1990)
O=SRR (Bauer et al. 2002 , PATER)
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
2.5
3
(O-P) bias (O-P) SD (O-A) bias (O-A) SDSR
R (
mm
/h)
Tb SOE(N=4512)
Tb LOE (N=4288)
SRR (N=4038)
-10
-8
-6
-4
-2
0
2
4
6
8
(O-P) bias (O-P) SD (O-A) bias (O-A) SD
NO
N-R
AIN
Y IW
V (
kg/m
2)
Tb SOE (N=1242)
Tb LOE (N=1166)
SRR (N=1840)
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
(O-P) bias (O-P) SD (O-A) bias (O-A) SD
LW
P (
kg
/m2
)
Tb SOE (N=4507)
Tb LOE (N=4283)
SRR (N=4034)
10/6/2006 Page 12
Integrated Wapor Vapor (IWV) increments(kgm-2)
TYPHOON CHABA CASE
-3
-2
-1
0
1
2
3
4
5
6
Mean SD
IWV
In
cre
me
nts
(kg
/m2
)
Tb SOE
Tb LOE
Tb LOE2
SRR
TROPICAL CYCLONE ZOE
-3
-2
-1
0
1
2
3
4
5
6
7
Mean SD
IWV
In
cre
me
nts
(kg
/m2
)
Tb SOE
Tb LOE
Tb LOE2
SRR
Increment = (Analysis-Background)
10/6/2006 Page 13
Analyzed IWV error estimateChaba Zoe
A = [ B-1 + HTR-1H ] -1, Rodgers (2000)
Tb SOE TB LOE
SRR
Tb SOE Tb LOE
SRR
10%5% 5%
10%
10%
10% 10%
10%
10/6/2006 Page 14
Size of temperature and humidityincrements (normalized by backgrounderrors) for Zoe
1D-Var Tb SOE1D-Var Tb LOE
T
lnQ
TlnQT
lnQ
TlnQ
(Analysis-Background)
10/6/2006 Page 15
Conclusions
• 1D-Var Tb and SRR developed (RTTOVSCATT-SSM/I or TMI) with GEM moist physicalschemes
• Successful analyses for Tropical Cyclone Zoe and Typhoon Chaba (> 95% convergence).• 1D-Var Tb SOE experiment: weight given to observations is too large and leads to large water
vapor increments. • Largest contribution of observation error comes from the moist physics (in particular deep
convection scheme) and hence affects the 1D-Var behavior– Explicit scheme favored (larger sensitivity to humidity)
• Assimilation of Tb, rather than SRR, should be favored in the variational assimilation context dueto the direct dependence of Tb on cloud liquid water path and integrated water vapor.
• Correlation of observation error between different channels is important for five channel set (lessopaque atmosphere implies more parameters to define).
• Applying moist physical schemes (“linearized”) from other NWP centers (developed specificallyfor DA) not trivial because schemes need to be tuned for model of center. Evaluate the impact ofassimilating 1D-Var IWV in the EC global 4D-Var
– Need to improve the computing efficiency of the 1D-Var– Tb bias correction
• Paper accepted to appear in Monthly Weather Review
10/6/2006 Page 16
www.ec.gc.ca