An Overview of Tropical Cyclone An Overview of Tropical Cyclone Track Guidance Models Used by Track Guidance Models Used by

NHCNHC

Michael Brennan, Chris Landsea, James Franklin NCEP/NHC

Mark DeMaria, NOAA/NESDIS/StAR

Andrea Schumacher, CIRA/CSU

Bernard Meisner, NOAA/NWS/Southern Region

General ObjectivesGeneral Objectives

By the end of this module, you should be able to…

• Describe the spectrum of NHC TC track models

• Describe the general strengths and weaknesses of each type of model

• Understand how track models are used in consensus forecasts

• Have an overview of future track forecast model improvements

Why Do Tropical Cyclones Move?Why Do Tropical Cyclones Move?

Primary Factors Affecting TC MotionPrimary Factors Affecting TC Motion

• Steering by deep-layer mean environmental wind

• NW drift due to interaction with N-S gradient of Coriolis parameter – “beta”-effect

• Depends on storm size

• Interaction with other environmental PV-gradients– Drift towards and left of PV gradient

• Interaction with land– Especially with mountainous terrain

• Small scale oscillations due to inner core asymmetries

Hierarchy of TC Track ModelsHierarchy of TC Track Models

• Statistical – Forecasts using relationships with storm-specific information (i.e., location, date)– CLIPER

• Statistical-Dynamical – Statistical models with input from dynamical model output– Best models through 1980s, overtaken by dynamical models in 1990s

• Last statistical-dynamical NHC track model retired in 2006

• Simplified Dynamical – LBAR Barotropic model initialized with vertically averaged (850-200 hPa) winds from

NCEP global model + vortex – BAMD, BAMM, BAMS Forecasts based on simplified dynamic representation of

interaction with vortex and prevailing flow (Modified trajectory in NCEP global model)

• Dynamical Models – Solve the physical equations of motion that govern the atmosphere and ocean– GFDL, GFDN, HWRF, NAM, GFS, NOGAPS, UKMET, ECMWF

• Ensemble and Consensus Forecasts– Model combinations

Statistical: CLIPER Statistical: CLIPER (CLImatology and PERsistence Model)(CLImatology and PERsistence Model)

• Statistical track model developed in 1972– Extended from 72 to 120 h in 2001

• Required Input: – Current and12 h old speed/direction of motion

– Current latitude/longitude

– Julian Day, Storm maximum wind

• Used as a benchmark– Forecasts with errors greater than CLIPER are considered to have no skill

Simplified Dynamical: Simplified Dynamical: BBeta and eta and AAdvection dvection MModel (BAM)odel (BAM)

• Method:

• Follows a trajectory using global model wind fields

• Wind fields smoothed to T25 resolution

• Correction for so-called “Beta Effect” (slow drift to the NW)

• Steering ~ 80-90%, Beta effect ~ 10-20% of motion

• Three different layer averages:

• Shallow (850-700 hPa) - BAMS

• Medium (850-400 hPa) - BAMM

• Deep (850-200 hPa) - BAMD

BAMS

BAMM

BAMD

5,000 ft/850 mb5,000 ft/850 mb

200 mb200 mb

Typical cruising altitude of commercial airplaneTypical cruising altitude of commercial airplane

SurfaceSurface

700 mb700 mb

400 mb400 mb

Which BAM to use? Which BAM to use?

LL

SHALLOW SHALLOW

LL

MEDIUMMEDIUM

LL

DEEPDEEP

Simplified Dynamical: Simplified Dynamical: Limited-area BARotropic (LBAR)Limited-area BARotropic (LBAR)

• Barotropic spectral model

– No temperature gradients

• Initialized with 850-200 hPa average winds/heights from NCEP global model (GFS)

• Idealized vortex and current motion vector added to GFS analysis

• Boundary conditions from GFS

• Still run due to simplicity and minimal computing costs

• Skill limited to ~2 days except in deep tropics

Primary Dynamical Models Used at NHC Primary Dynamical Models Used at NHC ((globalglobal and and regionalregional))

• GFS: U.S. NWS Global Forecast System (relocates first-guess TC vortex)

• UKMET: United Kingdom Met. Office global model (bogus, synthetic data)

• NOGAPS: U.S. Navy Operational Global Atmospheric Prediction System global model (bogus, synthetic data)

• ECMWF: European Center for Medium-range Weather Forecasting global model (no bogus)

• GFDL: U.S. NWS Geophysical Fluid Dynamics Laboratory regional model (bogus, spin-up vortex)

• GFDN: Navy version of GFDL model (bogus, spin-up vortex)

• HWRF: NCEP Hurricane Weather Research and Forecast regional model (vortex relocation and adjustment)

A Note on BogussingA Note on Bogussing• Since the globally analyzed vortex does

not typically represent the structure of a true TC, “Bogussing” is often employed

• Involves an analysis of synthetic data to describe the TC vortex

• Can significantly affect the surrounding environment – Vertical shear

• Creating and inserting a bogus is not straight forward– Forecast can be very sensitive to small

changes in the bogus storm

• Bogus storms tend to be too resilient during Extratropical Transition – Bogus retains warm core too long leading to

poor intensity and structure forecasts

GFS, GFDL, HWRF Initial Vertical ShearGFS, GFDL, HWRF Initial Vertical ShearHurricane Bill Aug 2009Hurricane Bill Aug 2009

Horizontal Resolution in Horizontal Resolution in Global Spectral ModelsGlobal Spectral Models

• Horizontal fields expanded in 2-D wave function series – Associated Legendre functions in latitude

– Fourier series (sines and cosines) in longitude

• T indicates Triangular truncation (e.g., T400) – Latitude and longitude series contain same number

of terms

– Uniform resolution over the sphere

• Rule of thumb for comparison to grid-point models x = 40,000 km/3N, N=truncation number

Global Model PropertiesGlobal Model Properties

GlobalDynamical

Model

Model Physics

Horizontal Grid Spacing

(or equivalent if spectral)

Vertical Levels

Vertical Coordinates

Convective Parameterizati

on

Data Assimilation

CMC GEMHydrostatic Grid Point

0.30° latitude, 0.45° longitude(~33 km at 49°

latitude)

80Hybrid Sigma-

Pressure

Kain-Fritsch(deep)

Kuo-transient (shallow)

4-D Var

ECMWFHydrostatic

Spectral~16 km 91

Hybrid Sigma-Pressure

Tiedtke 4-D Var

GFSHydrostatic

Spectral

~25 km(through FHR 192)

~80 km(FHR 192-384)

64Hybrid Sigma-

Pressure

Simplified Arakawa-Shubert

3-D Var;GSI/GDAS Analysis

NOGAPSHydrostatic

Spectral~40 km 42

Hybrid Sigma-Pressure

Emmanuel4-D Var;NAVDAS Analysis

UKMETNon-

Hydrostatic Grid Point

0.23° latitude, 0.35° longitude(~25 km in mid

latitudes)

70Hybrid Sigma-

PressureGregory/Rowntree

4-D Var

The Geophysical Fluid Dynamics The Geophysical Fluid Dynamics Laboratory (GFDL) Hurricane ModelLaboratory (GFDL) Hurricane Model

• Dynamical model capable of producing skillful intensity forecasts

• Coupled with the Princeton Ocean Model (POM) (1/6° horizontal resolution with 23 vertical sigma levels)

• Replaces the GFS vortex with one derived from an axisymmetric model vortex spun up and combined with asymmetries from a prior forecast

• Sigma vertical coordinate system with 42 vertical levels

• Limited-area domain (not global) with 2 grids nested within the parent grid

• Outer grid spans 75°x75° at 1/2° resolution or approximately 30 km

• Middle grid spans 11°x11° at 1/6° resolution or approximately 15 km

• Inner grid spans 5°x5° at 1/12° resolution or approximately 9 km

GFDL Model Nested GridsGFDL Model Nested Grids

The Hurricane Weather Research & The Hurricane Weather Research & Forecasting (HWRF) Prediction SystemForecasting (HWRF) Prediction System

• Next generation non-hydrostatic weather research and hurricane prediction system

• Movable, 2-way nested grid (9km / 27km; 42 vertical levels; ~75°x75°)• Coupled with Princeton Ocean Model

• POM utilized for Atlantic systems, no ocean coupling in N Pacific systems

• Vortex initialized through use of modified 6-h HWRF first guess• 3-D VAR data assimilation scheme

• But with more advanced data assimilation for hurricane core • Use of airborne and land based Doppler radar data (run in parallel)

• Became operational in 2007• Under development since 2002

• Runs in parallel with the GFDL

2010 HWRF Upgrades2010 HWRF Upgrades

• Fixing known bugs in HWRF model (mostly related to radiation)

– Led to definition of new baseline configuration for HWRF

• Improved initialization (Additional data in GSI near the storm environment)

• Surface physics changes (Cd/Ch coefficients calculated based on observations)

• Gravity wave drag parameterization

2011-2013 HWRF UPGRADE PLAN2011-2013 HWRF UPGRADE PLAN• Data assimilation:

– Advanced initialization for hurricane core – assimilate airborne Doppler radar observations to define storm strength and storm structure (run in parallel in 2010)

– Continuous upgrades to HWRF hurricane core initialization through advanced 4-D data assimilation for winds and reflectivity

• Model resolution upgrades:– Increase resolution: Horizontal resolution 1-6km, Vertical resolution ~100 levels

(dependent on results of current studies). • Hurricane ensembles: High-resolution hurricane model ensembles.

– Development of HWRF ensembles in progress • Model Physics:

– Continuous upgrades to gravity wave drag parameterizations, sea spray parameterization, atmospheric/ocean boundary layer (fluxes), microphysics, deep convection (cloud-resolving scales), radiation

• Ocean coupling– Replacement of POM with the HYCOM ocean model for 2011

• Other upgrades:– Coupling to land surface model with advanced surface physics for improved rainfall

forecasts at landfall. Important input to hydrology and stream flow models which will address inland flooding.

– Advanced Wave Model (WAVEWATCH III) to forecast waves up to the beach, i.e. improve non-linear interactions, surf-zone shallow water physics, wave interactions with currents.

*HWRF *GFDL *HWRF *GFDL Grid configuration 2-nests 3-nests

Nesting Force-feedbackInteraction thru intra-

nest fluxes

Ocean coupling POM (Atlantic only) POM

Convective parameterization

SAS mom.mix. SAS mom.mix.

Explicit condensation Ferrier Ferrier

Boundary layer GFS non-local GFS non-local

Surface layer GFDL (Moon et. al.) GFDL (Moon et. al.)

Land surface model GFDL slab GFDL slab

Dissipative heating Based on D-L Zhang Based on M-Y TKE2.5

Gravity wave drag YES NO

RadiationGFDL (cloud differences)

GFDL

*Configurations for 2010 season

500 hPa Winds in Hurricane Ike Environment 500 hPa Winds in Hurricane Ike Environment 05 Sept 2008 12 UTC05 Sept 2008 12 UTC

Satellite data used in NCEP’sSatellite data used in NCEP’soperational data assimilation systemsoperational data assimilation systems

• MODIS IR and water vapor winds

• GMS, Meteosat, and GOES cloud drift IR and visible winds

• GOES water vapor cloud top winds

• SSM/I wind speeds

• SSM/I precipitation estimates

• TRMM TMI precipitation estimates

• NOAA-17 HIRS 1b radiances

• AQUA AIRS 1b radiances

• NOAA-15, NOAA-16, NOAA-18 and AQUA AMSU-A 1b radiance

• NOAA-15, -16, and -17 AMSU-B 1b radiances

• GOES-12 5x5 cloud cleared radiances

• NOAA-16 and -17 SBUV ozone profiles

Satellite data also used to help estimate initial storm position, motion, intensity and wind structure for “TC Vitals” file used to initialize regional models

Specialized Aircraft DataSpecialized Aircraft Data

• NOAA Gulfstream-IV Jet flies synoptic surveillance missions in tropical cyclones that may impact land

• 20 to 30 GPS dropwindsondes released in storm environment to improve analysis of steering flow

– Wind, T, RH profiles

– ~200 hPa to surface

• Can be supplementedwith U.S. Air Force reserveand NOAA P-3 data

G-IV Dropwindsonde Locations G-IV Dropwindsonde Locations Hurricane Gustav 30 August 2008Hurricane Gustav 30 August 2008

“Late” versus “Early” Models

• “Late Models” = not available at synoptic time

• Dynamical models are not usually available until 4-6 hrs after the initial synoptic time (i.e., the 12Z run is not available until as late as ~18Z in real time)

• Results must be interpolated to latest NHC position (GFDL GFDI, NGPS NGPI, etc)

• Late Models: All global models, GFDL, GFDN, HWRF

• “Early Models” = available shortly after synoptic time

• Early Models: LBAR, BAM, CLIPER, interpolated dynamical models

Ensemble Forecasts (Classic Method)Ensemble Forecasts (Classic Method)

• A number of forecasts from a SINGLE MODEL using perturbed initial conditions that represent the likely initial analysis error distribution

• Each different model forecast is known as a “member model”

• Small spread among the member models may imply high confidence

• Large spread among the member models may imply low confidence

GFS Ensembles –solid linesCMC Ensembles – dashed

192 hr forecasts from 10 June 2009

Ensemble ForecastsEnsemble ForecastsHurricane Rita 9 Sep 2005 12zHurricane Rita 9 Sep 2005 12z

Ensemble Forecasts (Multi-Model Method)Ensemble Forecasts (Multi-Model Method)

• A group of forecast tracks from DIFFERENT PREDICTION MODELS at the SAME INITIAL TIME

• A multi-model ensemble is usually superior to an ensemble from a single model

• Different models typically have different biases, or random errors that will cancel or offset each other when combined.

• The multi-model ensembles are used to create CONSENSUS forecasts.

• Consensus Model Types

• Fixed: All members must be present, linear average

• Variable: Some members can be missing, linear average

• Corrected: Unequally weighted based on expected performance

Consensus Track Models for 2010Consensus Track Models for 2010

• Fixed

– TCON: GFS, UKMET, NOGAPS, GFDL, HWRF

– GUNA: GFS, UKMET, NOGAPS, GFDL

• Variable

– TVCN: GFS, UKMET, NOGAPS, GFDL, HWRF, GFDN, ECMWF

• Corrected

– TCCN: Corrected consensus version of TCON

– TVCC: Corrected consensus version of TVCN

– CGUN: Corrected consensus version of GUNA

– FSSE: Florida State Super Ensemble

Excellent example of TVCN consensus: HURRICANE IKE, 1800 UTC 6 SEP 2008Excellent example of TVCN consensus:

HURRICANE IKE, 1800 UTC 6 SEP 2008

120 Hr Verifying Point120 Hr Verifying Point

Track Model VerificationTrack Model Verification

• Calculate distance from forecast to best track position

• Homogeneous sample– All models must be available for inclusion of a

case

• NHC verifies tropical and subtropical stages– Extra-tropical stage excluded

• CLIPER model error is baseline for track forecast skill– Skill = 100*(ECLIPER-Emodel)/ECLIPER

Atlantic Simplified Track Model Verification

3-year Sample (2007-2009) OFCL shown for comparison

Mean Absolute Error Forecast Skill

Dynamical Model Skill (Early) Atlantic 2007-2009

Best Performing AtlanticBest Performing Atlantic Track Model Variability Track Model Variability

Year 48-hr 96-hr      

1988 NHC83        

1989 NHC83        

1990 BAMD    Statistical Dynamical

1991 NHC90    1992 BAMM        

1993 BAMM        

1994 BAMD    Simplified Dynamical

1995 GFDI    1996 GFDI        

1997 GFDI        

1998 UKMI    Global Dynamical

1999 UKMI    2000 GFSI        

2001 GFSI GFSI      

2002 NGPI BAMS  Regional Dynamical

2003 GFDI GFSI  2004 GFDI GFSI      

2005 GFDI UKMI      

2006 GFSI NGPI      

2007 UKMI UKMI      

2008 EMXI EMXI      

2009 GFSI CMCI      

Official forecast performance was very close to the consensus models. Good year for FSSE.

First year of availability for CMCI. Competitive, and potentially better than that (small sample).

Best dynamical models were ECMWF and GFS. UKMET and NOGAPS have been consistently weaker performers over the past few years.

BAMD performed poorly (strong shear). GFNI had insufficient availability.

Official forecast performance was very close to the TVCN consensus model. OFCL beat TVCN at 12 and 72 h.

EMXI best individual model. BAMD did about as well as any of the other 3-D models.

HWFI competitive with GFDL (neither was outstanding).

FSSE was the best consensus model in 2009.

TVCN (with GFNI and EMXI) did better than TCON.

Corrected consensus models TCCN, TVCN, CGUN did not do as well as their uncorrected counterparts. This was also true in 2008

TVCN slightly better than FSSE.

Single-model ensemble not nearly as effective as multi-model ensemble.

Corrected consensus model TVCN did not do as well as its uncorrected counterpart.

Serial Correlation of Track Model Serial Correlation of Track Model ErrorsErrors

Hurricane Dean Track Models18 UTC 17 Aug 2007

06 UTC 18 Aug 2007

GFDI

GFDI

HWFI

HWFIOFCL

Best Track

GFSI

OFCL

GFSI

Best Track

Errors have been cut in half over the past 15 years. 2009 was best year ever at 24-72h. Smaller samples give more erratic trends at days 4-5.

Atlantic Track Error Trends

E. Pacific Track Error TrendsE. Pacific Track Error Trends

Other Forecast Parameters Other Forecast Parameters in NHC Productsin NHC Products

1. Intensity

2. Genesis (formation within 48 hr)

3. Size (radius of 34, 50, and 64 kt)

4. Storm Surge (within about 24 hr of landfall)

5. Rainfall (provided by the Hydrometeorological Prediction Center)

6. Tornadoes (provided by the Storm Prediction Center)

7. Radius of 12-ft seas (provided by NHC’s Tropical Analysis and Forecast Branch and the Ocean Prediction Center)

Concluding Remarks on Concluding Remarks on TC Track Model GuidanceTC Track Model Guidance

• Dynamical models generally outperform simpler techniques

• No single model is always best

• Consensus forecasts generally better than individual models

• Year to year, storm to storm, forecast to forecast variability

• Serial correlation of same model for same storm

• HWRF and global models will continue to improve

Additional InformationAdditional Information• Overview of NHC Models: http://www.nhc.noaa.gov/aboutmodels.shtml• Operational Model Matrix (Comet Password Required):

http://www.meted.ucar.edu/nwp/pcu2/• Global Forecast System Home Page: http://www.emc.ncep.noaa.gov/• HPC’s Subjective Model Biases Page:

http://www.hpc.ncep.noaa.gov/mdlbias/biastext.shtml• HWRF’s Main Page: http://www.emc.ncep.noaa.gov/HWRF/index.html• GFDL Model Description (AMS Publication):

http://ams.allenpress.com/archive/1520-0493/135/12/pdf/i1520-0493-135-12-3965.pdf

• UKMET Model Technical Specifications: http://www.metoffice.gov.uk/research/nwp/numerical/unified_model/new_dynamics.html

• User’s Guide to the ECMWF: http://www.ecmwf.int/products/forecasts/guide/Preface.html

• NOGAPS Technical Specifications: http://www.nrlmry.navy.mil/metoc/nogaps/nogaps_char.html