Initialization Schemes in the Naval Research Laboratory’s Tropical Cyclone Prediction Model
(COAMPS-TC)Eric A. Hendricks1 Melinda S. Peng1
Tim Li2
Xuyang Ge3
1Naval Research Laboratory (NRL), Monterey, CA, USA2University of Hawaii and IRPC, Honolulu, HI
3Pennsylvania State University, State College, PA USA
Acknowledgements: Jim Doyle (NRL), Rich Hodur (SAIC), COAMPS-TC group
CMOS 2012 Congress / AMS 21st NWP and 25th WAF ConferencesMontreal, Canada, 29 May-1 June 2012
Introduction• A crucial part of TC intensity predictions is an accurate and
balanced TC vortex initially• 3DVAR data assimilation systems usually lack proper balance
constraints suitable for multi-scale TC; rapid adjustment often occurs after initialization
• A 4D data assimilation system would alleviate the initial imbalance problem to some degree
• Lack of observational data for TC intensity and structure remains
What do we do in the mean time?
Hybrid 3DVAR/Dynamic Initialization Schemes have the possibility of improving the initial balance and storm intensity/structure, while allowing model physics spin-up, potentially leading to improved intensity and track forecasts
Dynamic Initialization Schemes: TCDI, DI, TCDI/DIApplication to TC Prediction Using COAMPS-TC
NOGAPS/NCEP analysis
3DVAR data assimilation Remove TC vortex
Generate vortex from TCDI
(nudge MSLP)Insert vortex
Run forecast model
Warm Start
Cold Start
12-h forward DI
CNTL
DITCDI
TCDI
TCDI
TCDI
/DI
CNTL: Standard 3DVAR Initialization
DI: 3D Dynamic Initialization to analysis momentum ua (12-h relaxation) after 3DVAR
TCDI: Tropical Cyclone Dynamic Initialization (TC component is dynamic) after 3DVAR
TCDI/DI: Run TCDI, then run DI
)
Synthetic TC obs, Liou and Sashegy (2011)
TCDI: Hendricks et al. (2011) WAF, Zhang et al. (2012) WAF
COAMPS-TC OverviewCurrent and Future Capabilities
• Complex Data Quality Control• Relocation of TC in background• Synthetic Observations: TC vortex• NAVDAS 3DVAR: u, v, T, q, TC option• Initialization: Digital Filter Option• TC Balance Step: (underway)
• Navy Coupled Ocean Data Assimilation (NCODA) System
• 2D OI: SST• 3D MVOI, 3DVAR: T, S, SSH, Ice, Currents• Complex Data Quality Control• Initialization: Stability check
• Numerics: Nonhydrostatic, Scheme C, Moving Nests, Sigma-z, Flexible Lateral BCs
• Physics: PBL, Convection, Explicit Moist Physics, Radiation, Surface Layer
• TC Tools: Moving nests, dissipative heating, spray parameterization, shallow convection
• NRL Coastal Ocean Model (NCOM)• Numerics: Hydrostatic, Scheme C, Nested
Grids, Hybrid Sigma/z• Physics: Mellor-Yamada 2.5• Wave Models (WWIII and SWAN)• Generalized Coupling Layer (ESMF)
Ocean Analysis
Ocean ModelsAtmospheric Model
Atmospheric Analysis
Atmospheric Ensembles• Initial Cond. Perturbation: ET, EnKF• Physics Perturbations: PBL, Convection…• Lateral BCs: Global ensemble (NOGAPS)• Probabilistic Products: Intensity, track…
Ocean Ensembles• Initial Cond. Perturbation: ET• Physics Perturbations: PBL, Fluxes…• Lateral BCs: NCOM• Probabilistic Products: Mixed layer, OHC..
The Coupled Ocean/Atmosphere Mesoscale Prediction System (COAMPS®) is a registered trademark of NRL
COAMPS-TC Control (CNTL) Setup
Dynamics: Non-hydrostatic, compressible, C-grid (Klemp and Wilhemson 1978)
Vertical Discretization: Sigma-z vertical coordinate (40 levels, higher resolution near sfc)
Grids: 3 nests, 45/15/5-km resolution (2-way nesting), 15/5-km meshes move with the TC
PBL: Mellor-Yamada (1.5-order turbulence closure), dissipative heating (Jin et al. 2007)
Cumulus: Kain-Fritsch on 45/15-km, shallow convection, explicit convection on 5-km
Microphysics: NRL scheme, 6 species, based from Rutledge and Hobbs 1984 & Lin et al. 1983
Radiation: Fu-Liou schemeInitialization/DA: 3DVAR scheme (NAVDAS), synthetic observations added
that match observed TC structure and intensity
COAMPS-TC Nest Setup
3 Domains: 45/15/5 km
45 km grid fixed
Inner 2 grids (15/5-km) move
with the TC
DI Case Study: 2011 Hurricane Irene (09L)2011082518, Cold Start (Domain 3)
• During DI, the winds are held quasi-constant• 3DVAR is not able to produce gradient balanced
vortex, rapid adjustment to winds during DI
10-m Winds (kt) Sea Level Pressure (hPa)
CNTL TCDI
TCDI/DIDI
2011 IRENE(09L)
Wind Structure Verification (t=0 h)
H*WIND
Hurricane Irene (09L), 2011082512
COAMPS-TC using CNTL
COAMPS-TC using TCDI/DI
H*WIND courtesy NOAA/AOML/HRD Powell et. al (2010)
COAMPS-TC using DI
10-m Winds (kt)
Case Study: 08W (2011) Ma-On
Significant intensity error reductions for Ma-On by using TCDI/DI
15 cases
10 kt
CNTL TCDI/DIJTWC Best Track in blackCOAMPS-TC in color
Case Study: 07L (2010) Earl
CNTL TCDI/DI
Significant intensity error reductions for Earl by using TCDI/DI
13 cases10 hPa
NHC Best Track in blackCOAMPS-TC in color
Case Study: 12L (2011) KatiaTCDI/DICNTL
TCDI/DI does not over-intensify Katia as much as CNTL earlier, and gets rapid deepening better
NHC Best Track in blackCOAMPS-TC in color
Track Error: Homogenous Large Sample
TCDI/DI (blue curve) has lower track error for ALL cases and < 990 hPa
ALL cases Initial intensity < 990 hPa
Years: 2010-2011Atlantic Storms: Danielle, Earl, Igor, Irene, Katia, Maria, Rina, Julia
Western North Pacific storms: Chaba, Fanapi, Ma-OnCases: 120
Intensity Error: Homogenous Large Sample
ALL cases Initial intensity < 990 hPa
TCDI/DI (blue curve) has lowest intensity error for ALL cases and < 990 hPa cases with more statistical significance, and further reduced errors
Years: 2010-2011Atlantic Storms: Danielle, Earl, Igor, Irene, Katia, Maria, Rina, Julia
Western North Pacific storms: Chaba, Fanapi, Ma-OnCases: 120
Summary
• Three different TC initialization schemes have been developed, tested with COAMPS-TC– TCDI: tropical cyclone vortex spun-up– DI: Full 3D dynamic initialization to analyses winds– TCDI/DI: Run TCDI, then run DI
• TCDI/DI is shown to have superior performance– Average intensity errors reduced by 3-5 hPa and 2-3 kts
over all lead times – Average track errors reduced by 10-30 nm– Better for intense initializations (< 990 hPa)
• The dynamic initialization procedures allow model physics spin-up and “less shock”
• Future work– DI to satellite observed heating profiles