1. Assimilate NOAA P3 Tail Doppler Radar Data for Hurricane Initialization at NCEP
Mingjing Tong1
Collaborators: David Perish2, Daryl Kleist3, Russ Treadon2, John Derber2
Xuyang Ge1, Fuqing Zhang4
John Gamache5
NCEP/EMC HWRF team
1UCAR (NCEP/EMC), 2NOAA/NWS/NCEP/EMC, 3IMSG (NCEP/EMC), 4PSU, 5NOAA/AOML/HRD

2. P-3 regular P-3 weak storm P-3 weak storm 200 km 400 km
Jorgensen et al., 1983, J. Climate Appl. Meteor, 22,
400 km
P-3 weak storm
P-3 weak storm

3. HWRF Initialization Data assimilation GSI HWRF vortex initialization
global forecast environment
first guess
126 hr forecast
HWRF 6 hr forecast
HWRF vortex initialization
HWRF modified vortex
Relocation, size correction, intensity correction, adjust mass fields
GDAS forecast
analysis
GSI
Data assimilation
observation
Global hybrid EnKF-3DVAR ensemble forecast member 1
global forecast vortex
HWRF environment
Data thinning, quality control
Iterative minimization
Global hybrid EnKF-3DVAR ensemble forecast member 2
Global hybrid EnKF-3DVAR ensemble forecast member N
HWRF Initialization

4. TC initialization with TDR data assimilation
HWRF model
* Non-hydrostatic coupled atmosphere-ocean model, WRF-NMM dynamic core, rotated E-grid, two way interactive nesting
* Resolution: Parent (Outer) domain 0.18˚ (27 km), inner domain 0.06˚ (9 km), 42 vertical levels
Data assimilated
* Quality controlled tail Doppler radar radial velocity data within [-3 hr, 3 hr] are assimilated
* Rawinsonde, pibal, class sounding, profiler, dropsonde, AIRCFT, AIRCAR, GPSIPW, surface data (marine/land/splash-level/mesonet), satellite wind
* satellite radiance data: HIRS, AMSU-A, AMSU-B, MHS, sounding
GSI
* Analysis variables: streamfunction, unbalanced part of velocity potential, unbalanced part of temperature, unbalanced part of surface pressure, normalized relative humidity, satellite bias correction coefficients
* flow-dependent anisotropic background error covariance [Riishøjgaard (1998), Purser 2003b, 2005].
* Background error statistics are estimated in grid space with the NMC method (3DVAR).
* Hybrid DA option
Four HWRF domains (Fig 4.1 of HWRF USERS’ GUIDE)

5. outer domain ghost domain ghost domain conventional obs
Satellite radiance obs
conventional obs
ps, t, q, uv, TDR vr
UTC

6. Full HWRF vortex initialization
2011 TDR DA Real Time Parallel
experiment
TDRP
TDRA
TDRB
TDRC
NTDR
Assimiate TDR Vr data
yes no
background
Full HWRF vortex initialization
Relocate vortex only
GSI
FGAT
Post balance

7. Real time 3DVAR TDR DA experiments

8. Real time 3DVAR TDR DA experiments

9. Hybrid Ensemble-3DVAR System for HWRF
Current hybrid ensemble-3DAV system is one way coupled system
Ensemble perturbations come from global EnKF-3DVAR system ensemble forecast interpolated to HWRF analysis domains
Ensemble covariance is incorporated as part of the background error covariance though extended control variable method (Lorenc 2003; Buehner 2005; Wang et al. 2007)
May not help much with vortex initialization, hopefully can improve storm environment

10. O-A and O-F (Obs - conventional data)
analysis
6 hr fcst
12 hr fcst
18 hr fcst
24 hr fcst
3DVAR: black, Hybrid: green

11. Pseudo-ensemble hybrid for HWRF
Poterjoy and Zhang (2011): Wavenumber 0 storm structures have the largest influence on forecast uncertainty for TCs with category 1 or higher intensity.
Back ground error covariance is dominated by wavenumber 0 component.
Flow-dependent forecast covariance can be approximated from a sample of near-axisymmetric TC vortices
TC vortices are created by running idealized 3 km resolution HWRF simulations and grouped by intensity (Xuyang Ge)
TC library vortices are mapped to the background vortex region (300 km from background TC center) in ghost domain and perturbations are calculated
Replace global ensemble perturbations with TC vortex library perturbations within 150 km from background TC center
TC library perturbations are gradually blended with global ensemble perturbations from 150 km to 300 km from background TC center

12. Analysis increment from single obs test u obs at 0
Analysis increment from single obs test u obs at 0.5 degree north of storm center at 850 level
PEDA
HYBB
PEDA – hybrid DA with pseudo-ensemble HYBB – hybrid DA with pure global ensemble
Weight given to ensemble B is 0.8

13. Analysis increment from single obs test u obs at 0
Analysis increment from single obs test u obs at 0.5 degree north of storm center at 850 level
PEDA
HYBB
PEDA – hybrid DA with pseudo-ensemble HYBB – hybrid DA with pure global ensemble

14. analysis
PEDA
HYBB
PEDA
HYBB

15. 12 hour forecast
HYBB
PEDA
PEDA
HYBB

16. TDR real time parallel (TDRP) configurations
HWRF
TDRP
Storm environment
GFS analysis
GDAS forecast
Vortex initialization
Full vortex initialization
Relocation only if TDR data are available
Use of GSI
Only for deep storms
Isotropic static B
All storms
Use FGAT
Anisotropic B (some flow
Dependency) in ghost domain
Data assimilated
Data time window [-1.5,1.5] hr
Exclude inner core (150 km from TC center) SYNOP, METAR or sfc marine, ATLAS buoy, Sfc mesonet, synthetic t.c. wind data
Data time window [-3,3] hr
Assimilate all data
Assimilate TDR data
All hybrid experiments follow TDRP configuration, except B is a combination of static B and ensemble B

17. First TDR DA cycle of Hurricane Earl

18. Hybrid TDR test with Hurricane EARL

19. Hybrid TDR test with Hurricane EARL

20. Hybrid TDR test with Hurricane EARL

21. Issues and Discussion Over/under estimate storm intensity
* poor background
* DA configuration
Short term forecast spin up/down
* Mass-wind balance
* storm structure
May need more comprehensive verifications

22. Plans Model: move on to higher resolution (27km-9km-3km) HWRF.
Assimilation method
* Intensive hybrid (1 way coupled) test will be done using ensemble data from the new global EnKF-3DVAR hybrid parallel (prd12q3k).
* Further investigate the performance of PEDA. Improve TC library, e.g. including the size variability.
* Move on to 2 way coupled hybrid system by collaborating with Jeff Whitaker and AOML/HRD
* Analyze cloud variables
DATA
* Investigate the impact of assimilating ground based radar data.
* The impact of satellite radiance data in regional model
* Cloudy radiance data