1

Improving High Resolution Tropical Cyclone Prediction Using a Unified GSI-based Hybrid Ensemble-Variational Data Assimilation System for HWRF

Xuguang Wang

University of Oklahoma, Norman, OK

In close collaboration with NCEP/EMC, ESRL/PSD, AOML/HRD

Warn On Forecast and High Impact Weather Workshop, Feb. 6, 2013

2

Background History of hybrid DA

Development of theory

Research with simple model and simulated data

System development for real NWP model and test real data

Operational implementation at NWP centers for global model: e.g., GSI-based hybrid Ensemble-Variational DA for NCEP GFS

2

10 y

ears

33

Background GSI is a unified system which provides data assimilation for current operational global and regional forecast system.

Efforts are being conducted to integrate the same GSI-based hybrid DA system with operational regional forecast systems.

GSI-EnKF hybrid

GFS HWRFRapid Refresh (RR)NAM

44

Background

Unifying GSI-based hybrid DA system with operational regional systems facilitates faster transition to operations.

The focus of the project is the extension, application, extensive testing and research of the GSI-based hybrid data assimilation for the HWRF modeling system at high resolutions.

Also motivated by encouraging results of ensemble based data assimilation for tropical cyclones.

55Li et al., 2012, MWR

3DVAR hybrid

700 mb wind 700 mb wind

850 mb temp850 mb temp

•Hurricane IKE 2008

•WRF ARW: Δx=5km

•Observations: radial velocity from two WSR88D radars (KHGX, KLCH)

•WRFVAR hybrid DA system (Wang et al. 2008a)

Background

66

control forecastGSI-ECV

control analysis

data assimilationFirst guess forecast

control forecast

Ensemble covariance

EnKF

EnKF analysis k

member 1 forecast

member 2 forecast

member k forecast

EnKF analysis 2

EnKF analysis 1

member 1 forecast

member 2 forecast

member k forecast

member 1 analysis

member 2 analysis

member k analysis

Re-center E

nKF

analysis ensemble

to control analysis

GSI-based hybrid ensemble-variational DA system

Wang, Parrish, Kleist, Whitaker 2013, MWR

7

''1''1

2'1

1'11

211'1

2

1

2

1

2

1

,

HxyHxyαCαxBx

αx

oToTT

oe JJJJ

R

K

k

ekk

1

'1

' xαxx

7

• GSI-ECV: Extended control variable (ECV) method (Lorenz 2003; Buehner 2005; Wang et al. 2007a, 2008a, Wang 2010):

Extra term associated with extended control variable

Extra increment associated with ensemble

GSI-based hybrid ensemble-variational DA system

• EnKF: square root filter interfaced with GSI observation operator (Whitaker et al. 2008)

88

Why Hybrid? “Best of both worlds” VAR (3D, 4D)

EnKF Hybrid References (e.g.)

Benefit from use of flow dependent ensemble covariance instead of static B

Yes Yes Hamill and Snyder 2000; Lorenc 2003; Etherton and Bishop 2004; Zupanski 2005, Wang et al. 2007ab,2008ab, 2009; Buehner et al. 2010ab; Wang 2011, etc.

Robust for small ensemble or large model error

Yes Wang et al. 2007b, 2009; Buehner et al. 2010b

Better localization for integrated measure, e.g. satellite radiance; radar with attenuation

Yes Campbell et al. 2010

Flexible to add various dynamical/physical constraints

yes Yes Wang et al. 2013

Use of various existing capabilities in VAR (e.g., Outer loops to treat nonlinearity; Variational QC)

yes Yes

Summarized in Wang 2010, MWR

9

Experiment with Airborne radar

• Model: HWRF Δx=9km ( 3km next)

•Observations: radial velocity from Tail Doppler Radar (TDR)

• Case: IRENE 2011

• Initial and LBC ensemble: GFS global hybrid DA system

• Ensemble size: 40

• Initial test focuses on the impact of the use of the ensemble covariance in GSI

IRENE 2011IRENE

2011

Li et al. 2013, AMS presentation

10

DA cycling configuration

GSI 3DVar

EnKF

Hybrid (1 way coupling)

11

TDR data distribution

12

TDR data distribution

m/s

13

700 mb wind increment

m/s m/s

GSI 3DVar 3DEnsVar

14

Verification against independent flight level wind speed

kmkm

GSI 3DVar 3DEnsVarFirst Leg

15

Verification against SFMR wind speed

kmkm

GSI 3DVar 3DEnsVar

First Leg

16

Verification against independent flight level wind speed

kmkm

GSI 3DVar 3DEnsVar

Last Leg

17

Verification against SFMR wind speed

kmkm

3DEnsVarGSI 3DVar

Last Leg

1818

Comparison with radar wind analysiskm

m/s

km km km

m/s m/s

HRD radar wind analysis GSI 3DVAR 3DEnsVar

18

1919

Comparison with radar wind analysis

m/s m/s m/s

GSI 3DVar @ 1km 3DEnsVar @ 1kmHRD radar wind analysis @ 1km

19

20

Comparison with radar wind analysis

m/s m/s m/s

GSI 3DVar @ 3km 3DEnsVar @ 3kmHRD radar wind analysis @ 3km

21

Track forecast

EMC: HWRF official forecastNoDA: no TDR assimilationGSI 3DVar: assimilating TDR using GSI 3DVarEnKF: assimilating TDR using EnKF3DEnsVar: assimilating TDR using 3DEnsVar

3DEnsVar

3DVar

22

MSLP forecast

EMC: HWRF official forecastNoDA: no TDR assimilationGSI 3DVar: assimilating TDR using GSI 3DVarEnKF: assimilating TDR using EnKF3DEnsVar: assimilating TDR using 3DEnsVar

3DEnsVar

3DVar

23

Summary and ongoing work

a. The GSI-based hybrid EnKF-Var data assimilation system was expanded to assimilate TDR data for HWRF.

b. TDR data showed positive impact on TC track and intensity forecasts and verification against independent observations.

c. Various diagnostics and verifications suggested ensemble-based data assimilation (hybrid, EnKF) provided more skillful TC analysis and forecasts than the GSI 3DVar.

d. Testing more missions/cases.e. Testing dual resolution 3km/9km hybrid; 3km’s own hybrid.f. Developing and testing GSI-based hybrid EnKF-Var with

moving nests.g. Add and test more observations.h. Develop and research on various new capabilities for HWRF

hybrid (4DEnsVar, hydrometeor, post balance constraint, etc.).

24

• The current GSI hybrid is 3D (3DEnsVar); temporal evolution of error covariance with the assimilation window not considered

• Observations (e.g., satellite, radar) are spreading through the DA window.

• 4DENSVAR is further developed.

• Different from traditional 4DVAR, 4DENSVAR conveniently avoid the tangent linear and adjoint of the forecast model which is challenging to build and maintain (e.g., Buehner et al. 2010).

• Much cheaper compared to traditional 4DVAR being developed for GSI (Rancic et al. 2012).

24

GSI based 4DENSVAR

2525

4DENSVAR performance:Test with GFS at reduced resolution

2010 hurricane track forecast Anomaly correlation for height

Lei, Wang, Kleist, Whitaker, 2013