Development and research of GSI‐based hybrid EnKF‐Var data assimilation for HWRF to improve hurricane prediction
Xuguang Wang, Xu Lu, Yongzuo Li
School of Meteorology
University of Oklahoma, Norman, OK, USA
Acknowledgement
Mingjing Tong , Vijay Tallapragada, NCEP/EMC, College Park, MD
Henry Winterbottom, Jeff Whitaker, NOAA/ESRL, Boulder, CO
WWOSC, Aug. 2014, Montreal, Canada
1
GSI based Var/EnKF/hybrid for global and
regional modeling systems
GSI‐based Var/EnKF/3‐
4DEnVar Hybrid
GFS
WRF‐NMMB
WRF ARW
Hurricane‐
WRF (HWRF)
SCI-PS126.04: Aaron Johnson
talk “A comparison of GSIbased multiscale EnKF and
3DVar for convective scale
weather forecast ”
2
GSI‐based Hybrid EnKF‐Var DA system
Wang, Parrish, Kleist, Whitaker 2013, MWR
EnKF
Whitaker et al. 2008,
MWR
EnKF
analysis 2
member 2
forecast
member k
forecast
control
forecast
Ensemble
covariance
GSI-ACV
Wang 2010, MWR
data assimilation
EnKF
analysis k
control
analysis
Re-center EnKF analysis ensemble
to control analysis
EnKF
analysis 1
member 1
forecast
member 1
analysis
member 1
forecast
member 2
analysis
member 2
forecast
member k
analysis
member k
forecast
control
forecast
First guess
forecast 3
GSI hybrid for GFS: GSI 3DVar vs. 3DEnsVar Hybrid vs. EnKF
 3DEnsVar Hybrid was better than 3DVar due to use of flow‐dependent ensemble covariance
 3DEnsVar was better than EnKF due to the use of tangent linear normal mode balance constraint (TLNMC)
Wang, Parrish, Kleist and Whitaker, MWR, 2013, 141, 4
4098‐4117 GSI hybrid for GFS: 3DEnsVar vs. 4DEnsVar
•
GSI-4DEnsVar: Naturally extended from and unified with GSIbased 3DEnsVar hybrid formula (Wang and Lei, 2014, MWR,
142, 3303-3325).

Add time dimension in 4DEnsVar

J x1' , α  1 J1   2 J e  J o
1 T
1
1 T
1
 1 x1' B static x1'   2 α T C 1α  t 1 ( yto '-Ht x t ) T R -t1 ( yto '-Ht x t )
2
2
2
K

x t  x   α k  (x ek ) t
'
'
1

k 1
B stat 3DVAR static covariance; R observation error covariance; K ensemble size;
C correlation matrix for ensemble covariance localization; x ek kth ensemble perturbation;
x1' 3DVAR increment; x ' total (hybrid) increment; y o ' innovation vector;
H linearized observation operator; 1 weighting coefficient for static covariance;
 2 weighting coefficient for ensemble covariance; α extended control variable.
5
GSI hybrid for GFS: 3DEnsVar vs. 4DEnsVar
Results from Single Reso. Experiments

4DEnsVar improved general global forecasts

4DEnsVar improved the balance of the analysis

Performance of 4DEnsVar degraded if less frequent ensemble perturbations used

4DEnsVar approximates nonlinear propagation better with more frequent ensemble perturbations

TLNMC improved global forecasts
See poster SCI‐POT1040 and
Wang, X. and T. Lei, 2014: GSI‐based four dimensional ensemble‐variational (4DEnsVar) data assimilation: formulation and single resolution experiments with real data for NCEP Global Forecast System. Mon. Wea. Rev., 142, 3303‐3325. 6
GSI hybrid for GFS: 3DEnsVar vs. 4DEnsVar
16 named storms in Atlantic
and Pacific basins during
2010
7
Approximation to nonlinear propagation
–3h increment
propagated by
model integration
4DEnsVar
(hrly pert.)
4DEnsVar
(2hrly pert.)
3DEnsVar
Hurricane Daniel 2010
*
-3h
time
0
3h
8
Verification of hurricane track forecasts
•
•
•
•
3DEnsVar outperforms GSI3DVar.
4DEnsVar is more accurate than 3DEnsVar after the 1-day forecast lead time.
Negative impact if using less number of time levels of ensemble perturbations.
Negative impact of TLNMC on TC track forecasts.
9
Development and research of GSI based
Var/EnKF/hybrid for HWRF
GSI‐based Var/EnKF/3‐4D EnVar Hybrid
GFS
WRF‐NMMB
WRF ARW
Hurricane‐
WRF (HWRF)
10
GSI hybrid for HWRF
Hurricane Sandy, Oct. 2012
 Complicated evolution
 Tremendous size
 147 direct deaths across Atlantic Basin
 US damage $50 billion
New York State before and after
nhc.noaa.gov
11
Experiment Design
• Model: HWRF Sandy 2012 •Observations: radial velocity from Tail Doppler Radar (TDR) onboard NOAA P3 aircraft
• Initial and LBC ensemble: GFS global hybrid DA system
• Ensemble size: 40
12
Experiment Design
• Model: HWRF Oper.
HWRF •Observations: radial velocity from Tail Doppler Radar (TDR) onboard NOAA P3 aircraft
• Initial and LBC ensemble: GFS global hybrid DA system
• Ensemble size: 40
13
TDR data distribution (mission 1) P3 Mission 1
14
Verification against SFMR wind speed
Last Leg
15
Comparison with HRD radar wind analysis
16
Comparison with HRD radar wind analysis
S
N
17
Track forecast (RMSE for 7 missions)
18
Experiments for 2012‐2013 seasons
Correlation between HRD radar wind analysis and analyses from various DA methods
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1
2
3
4
5
Hybrid
6
7
8
9 10 11 12 13 14 15 16 17 18 19 20 21 22
GSI3DVAR
Case#
Hybrid-GFSENS
19
ISSAC 2012 (mission 7)
20
Verification against SFMR and flight level data
Experiments for 2012‐2013 season
Track
MSLP
22
HWRF Hybrid DA with moving nests:
(1) Dual resolution hybrid
9km
•3km movable nest ingests 9km HWRF EnKF ensemble •Two‐way coupling
3km
•Tests with IRENE 2011 assimilating airborne radar data
23
Dual resolution hybrid
IRENE 2011 HWRF Hybrid DA with moving nests:
(2) Self‐consistent system: 3km and 9km ingesting their own EnKF
ensemble
27km
9km
3km
25
HWRF Hybrid DA with moving nests
Near real time experiment during 2014 season
MSLP
Track
hPa
Arthur 2014
 Newly developed self-consistent
HWRF hybrid DA with moving nests
show improvement on MSLP/Vmax
26
than operational HWRF
Summary and ongoing work
a. The GSI‐based hybrid EnKF‐Var data assimilation system was expanded to HWRF. b. Various diagnostics and verifications suggested this unified GSI hybrid DA system provided more skillful TC analysis and forecasts than GSI 3DVar and than HWRF GSI hybrid ingesting GFS ensemble.
c. Airborne radar data improved TC structure analysis and forecast, TC track and intensity forecasts. Impact of the data depends on DA methods.
d. Hybrid DA with movable nests were developed and showed promising results.
e. Ongoing experiments with more cases.
f. Ongoing research to investigate the difference among Var, EnKF, 3DEnsVar and 4DEnsVar hybrid for convective scales.
27
References
Wang, X., 2010: Incorporating ensemble covariance in the Gridpoint
Statistical Interpolation (GSI) variational minimization: a mathematical
framework. Mon. Wea. Rev., 138, 2990-2995.
Wang, X., D. Parrish, D. Kleist and J. S. Whitaker, 2013: GSI 3DVar-based
Ensemble-Variational Hybrid Data Assimilation for NCEP Global Forecast
System: Single Resolution Experiments. Mon. Wea. Rev., 141, 4098-4117.
Wang, X. and T. Lei, 2014: GSI-based four dimensional ensemblevariational (4DEnsVar) data assimilation: formulation and single resolution
experiments with real data for NCEP Global Forecast System. Mon. Wea.
Rev., 142, 3303-3325.
28
GSI-based Hybrid EnKF-Var DA system
•
(4D)EnKF: ensemble square root filter interfaced with GSI
observation operator (Whitaker et al. 2008)
•
GSI-3DEnsVar: Extended control variable (ECV) method
implemented within GSI variational minimization (Wang 2010,
MWR):
J x1' , α   1 J 1   2 J e  J o



1 ' T 1 '
1 T 1
1 o'
' T
 1 x1 B x1   2 α C α  y  Hx R 1 y o '  Hx '
2
2
2
K

x  x   α k  x ek
'
'
1
k 1


Extra term associated with
extended control variable
Extra increment associated
with ensemble
29
DA cycling configuration
Cold Start
GSI3DVar
OBS
Spin-up Forecast
Deterministic Forecast
DA Cycle
OBS
Hybrid
Spin-up Forecast
Deterministic Forecast
Ensemble Perturbation
OBS
HWRF EnKF
Deterministic Forecast
Ensemble
Spin-up Forecast
DA Cycle
30
DA cycling configuration
OBS
Hybrid-GFSENS
Spin-up Forecast
Deterministic Forecast
Ensemble Perturbation
GFS ENS
31

Operational services - Barcelona Dust Forecast Center

Added value to the investigation of Rossby wave-trains by