GSI-based EnKF-Var hybrid data assimilation system: implementation and test for hurricane prediction

Xuguang Wang, Xu Lu, Yongzuo Li, Ting Lei University of Oklahoma, Norman, OK In collaboration with Mingjing Tong , Vijay Tallapragada, Dave Parrish, Daryl Kleist NCEP/EMC, College Park, MD Jeff Whitaker, Henry Winterbottom NOAA/ESRL, Boulder, CO

1

GSI-based Hybrid EnKF-Var DA system Wang, Parrish, Kleist, Whitaker 2013, MWR member 1 forecast member 2 forecast EnKF Whitaker et al. 2008, MWR EnKF analysis 1 EnKF analysis 2 member k forecast control forecast Ensemble covariance GSI-ACV Wang 2010, MWR EnKF analysis k control analysis

data assimilation

member 1 analysis member 2 analysis member 1 forecast member 2 forecast member k analysis member k forecast

First guess forecast

2 control forecast

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 Wang, Parrish, Kleist and Whitaker, MWR, 2013 3

GSI hybrid for GFS: 3DEnsVar vs. 4DEnsVar •

GSI-4DEnsVar

: Naturally extended from and unified with GSI based 3DEnsVar hybrid formula (Wang and Lei, 2014, MWR, in press).

Add time dimension in 4DEnsVar

J

x

' 1   1

J

1   2

J e



J o

x

t

' 

x

1 '    1

k K

  1 1  2

α x

1 '

T k



B

static

 1

x

' 1 (

x

e k

)

t

   2 1 2

α

T

C

 1

α

 1 2 

T t

 1 (

y t o

'-H t

x

t

) T R t 1 (

y t o

'-H t

x

t

)

B

stat 3DVAR static covariance;

R

observation error covariance;

K

ensemble size;

C

correlation matrix for ensemble covariance localization;

x

e k k

th ensemble perturbation;

x

1 ' 3DVAR increment;

x

' total (hybrid) increment;

y

o

' innovation vector;

H

linearized observation operator;  2  1 weighting coefficient for static covariance; weighting coefficient for ensemble covariance;

α

extended control variable. 4

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 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., in press. 5

GSI hybrid for GFS: 3DEnsVar vs. 4DEnsVar 16 named storms in Atlantic and Pacific basins during 2010 6

Approximation to nonlinear propagation –3h increment propagated by model integration 4DEnsVar (hrly pert.) 4DEnsVar (2hrly pert.) 3DEnsVar Hurricane Daniel 2010 * -3h 0 3h time 7

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. 8

Development and research of GSI based Var/EnKF/hybrid for regional modeling system GSI-based Var/EnKF/3D 4DHybrid GFS WRF-NMMB WRF ARW Hurricane WRF (HWRF)

Poster

:

Johnson et al.

“Development and Research of GSI based Var/EnKF/hybrid Data Assimilation for Convective Scale Weather F orecast over CONUS.” 9

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

10

Experiment Design

Sandy 2012 • Model: 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 11

Experiment Design

Oper.

HWRF • Model: 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 12

TDR data distribution (mission 1)

P3 Mission 1

13

Last Leg

Verification against SFMR wind speed

14

Comparison with HRD radar wind analysis

15

S

Comparison with HRD radar wind analysis

N 16

Track forecast (RMSE for 7 missions)

17

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 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 Hybrid GSI3DVAR Hybrid-GFSENS Case# 18

ISSAC 2012 (mission 7)

19

Verification against SFMR and flight level data

Experiments for 2012-2013 season

Track MSLP 21

Two-way Dual Resolution Hybrid for HWRF

9km • 3km movable nest ingests 9km HWRF EnKF ensemble • Two-way coupling • Tests with IRENE 2011 assimilating airborne radar data 3km 22

Two-way Dual resolution hybrid

Summary and ongoing work

GFS a. GSI-based 4DEnsVar for GFS improved global forecast and TC forecast.

b. The analysis generated by 4DEnsVar was more balanced than 3DEnsVar.

c. the performance of 4DEnsVar was in general degraded when less frequent ensemble perturbations were used.

d. The tangent linear normal mode constraint had positive impact for global forecast but negative impact for TC track forecasts.

e. Preliminary tests showed positive impact of the temporal localization on the performance of 4DEnsVar.

HWRF 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. Dual-resolution (3km-9km) two way hybrid for HWRF showed promising results.

e. Developing/enhancing 4DEnsVar hybrid and assimilation of other airborne data and other data from NCEP operational data stream for HWRF.

24

12 km Development and Research of GSI-based Var/EnKF/hybrid DA for Convective Scale Weather Forecasts over CONUS

Poster

: Johnson, Wang, Lei, Carley, Wicker, Yussouf, Karstens • 4 km •

Outer Domain

– assimilate operational conventional surface and mesonet observations, RAOB, wind profiler, ACARS, and satellite derived winds every 3 hours to define synoptic/mesoscale environment

Inner Domain

– assimilate velocity and reflectivity from NEXRAD radar network every 5 min during last 3hr cycle Johnson, Wang et al. 2014 25

Precipitation forecast skill averaged over 10 complex, convectively active cases • • GSI-EnKF forecasts are more skillful than GSI-3DVar forecasts for all thresholds and lead times. Benefits of radar data are more pronounced assimilated by GSI-EnKF than GSI-3DVar.

26

May 8

th

2003 OKC Tornadic Supercell

1hr forecast from 22Z GSI hybrid GSI hybrid Ref and vorticity at 1 km Lei, Wang et al. 2014 W and Vort. at 4 km 27

GSI3DVar Hybrid

DA cycling configuration (mission 1)

Cold Start OBS Spin-up Forecast DA Cycle Deterministic Forecast OBS Spin-up Forecast HWRF EnKF OBS Deterministic Forecast Ensemble Perturbation Deterministic Forecast DA Cycle Ensemble Spin-up Forecast 28

DA cycling configuration (mission 1)

Hybrid-GFSENS Spin-up Forecast OBS Deterministic Forecast Ensemble Perturbation GFS ENS 29

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

x

1 '   1

J

1   2

J e



J o

x

' 

x

1 '   1 1 2

x

1 '

T

B

 1

x

1 ' 

k K

  1 

α

k



x

e k

   2 1 2

α

T

C

 1

α

 1 

y

o

' 

Hx

'

T

 

y

o

' 

Hx

2 Extra term associated with '  extended control variable Extra increment associated with ensemble