Improved Statistical Intensity Forecast Models:A Joint Hurricane Testbed Project Update

Mark DeMaria, NOAA/NESDIS, Fort Collins, COJohn A. Knaff, CIRA/CSU, Fort Collins, COJohn Kaplan, NOAA/AOML/HRD, Miami, FL

Presented at the Interdepartmental Hurricane Conference

March 22, 2006

Outline

• Current Status of SHIPS and the Rapid Intensity Index (RII)

• Project Goals– New decay model in SHIPS– Modified vertical shear calculation– Discriminant analysis version of the RII

• Progress and Plans for the 2006 Season– Additional SHIPS modifications from 2005

results

2005 Operational Version of SHIPS

• 16 basic predictors– atmospheric from GFS forecast fields– oceanic from Reynold’s weekly SST– climatology and persistence from ATCF input

• Correction for ocean heat content (Atlantic only) and GOES predictors – previous JHT project

• Adjusted SST from Joe Cione cooling algorithm – previous JHT project, Atlantic only

• New decay model formulation– Less decay over islands and narrow landmasses– current JHT project

SHIPS Skill in 2005

0

5

10

15

20

25

30

12 24 36 48 60 72 84 96 108 120

Forecast Interval (hr)

Imp

rov

em

en

t O

ve

r S

HF

5 (

%)

Atlantic

East Pacific

Impact of New Decay Model and Cione Ocean Cooling

(based on re-runs of 2005 Atlantic forecasts)

0.0

2.0

4.0

6.0

8.0

10.0

12.0

12 24 36 48 60 72 84 96 108 120

Forecast Interval (hr)

Pe

rce

nt

Imp

rov

em

en

t

New Decay Model

SST Cooling

New Decay and SSTCooling

Vertical Shear Evaluation

• Shear evaluated from GFS forecast along OFCI track

• Large annulus utilized – 200-800 km radius

• Proposed new method– Remove vortex from GFS fields– Test other averaging areas

• Smaller radii• Elliptical region aligned with storm motion (Stacy Stewart

method)

SHIPS Shear Calculation Region200 hPa Wind Wilma 24 Oct 2005 06 UTC

Impact of OFCI and GFS Track Mismatch96 h Forecast for Frances from 27 Aug 2004 12 UTC

850 hPa 200 hPa

G G

O O

G = GFS position O = OFCI Position

GFS Fields After Vortex Removal Procedure 96 h Forecast for Frances from 27 Aug 2004 12 UTC

G G

OO

850 hPa 200 hPa

G = GFS position O = OFCI Position

Additional Changes based on 2005 Season

• Add storm translation to MPI calculation– Wilma (2005), Florence (1994)

• Add 250 hPa temperature predictor for cold upper atmosphere, but low tropopause– Epsilon and Zeta (2005)

• Print recon variables for qualitative use– GRIP model sample too small

• Test forecast with more general predictive equation – Will be run after SHIPS run, but saved for evaluation

Storm Translational Speed and MPI

– Current version adds constant 12 kt to DK (1994) and WH (1997) formulas

– Acceleration not accounted for– Add fraction of translational speed from

Schwerdt formula (a = 1.5c0.63) to MPI– Improves fit to 1982-2005 sample

Hurricane Epsilon and Tropical Storm Zeta(motivation for 250 hPa T predictor)

Zeta 03 Jan 2006 1745 UTCMax Wind = 55 kt

SST = 24.3 oCT’(200 hPa) = -4.1oCT’(250 hPa) = -6.6oC

Epsilon 04 Dec 2005 1445 UTCMax Wind = 70 kt

SST = 21.7 oCT’(200hPa) = -1.8oCT’(250hPa) = -6.4oC

Recon and Additional GOES Predictors

• GOES and Recon Intensity Prediction (GRIP) Model– Part of previous JHT project

• 808 cases 1995-2004– 6554 for total SHIPS sample

• Significant new predictive information in recon data

• Independent tested on 2005 Atlantic forecasts• Improvement only at 12-24 hours• Developmental sample still too small

Error Reduction of SHIPS Model Fit

From Inclusion of Recon and Satellite Data

0

1

2

3

4

5

6

7

8

9

10

6 18 30 42 54 66 78 90 102

114

Forecast Interval (hr)

Var

ian

ce In

crea

se (

%)

ALL 5

OHC

GOES1

GOES2

RECON1

RECON2

-15

-10

-5

0

5

10

15

20

12 24 36 48 60 72 84 96 108 120

Forecast Interval (hr)

Per

cen

t Im

pro

vem

ent

Independent Coefficients

Dependent Coefficients

Intensification is favored with large symmetric tangential wind near the RMW, but small area-integrated KE

2005 EvaluationVariance Increase for Developmental Sample

0

2E+16

4E+16

6E+16

8E+16

1E+17

1.2E+17

20 40 60 80 100 120 140 160

Maximum Wind (kt)

Kin

etic

En

erg

y (J

)Katrina2 05

Ivan 04

Rita 05

Fran 96 Wilma 05

Isabel 03

Jeanne 04Floyd 99

Frances 04

Georges2 98 Georges1 98

Bonnie 98Lili 02

Opal 95Dennis 05

Bret 99

Bertha 96

Charley1 04

Erin1 95Charley2 04

Earl 98 Claudette 01

Erin2 95Katrina1 05

Danny1 97

Danny2Gaston 04

KE-5

KE-4

KE-3

KE-2

KE-1

KE-0

SS-0 SS-1 SS-2 SS-3 SS-4 SS-5

Max Wind vs. KE categories of 1995-2005 U.S. Landfalling Hurricanes

Generalized Prediction Equation

dV/dt = (t) Continuous form of SHIPS model

assumed constant over entire forecast intervale.g., V(96) = V(0) + t

Replace with generalized prediction equation

dV/dt = V - (V/Vmpi)nV

n=3, -1 = 26 hr, Vmpi from MPI formulaestimate (t) statistically.

Analytic Solution for Constant Vmpi and Vs = Steady State V = Vmpi(/)1/3

V/Vs = (Vo/Vs)et/[1 + (e3t-1)(Vo/Vs)3]1/3

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

0 0.5 1 1.5 2 2.5 3 3.5 4

t/k

V/V

s

Vo/Vs=1.6Vo/Vs=1.3Vo/Vs=1.0Vo/Vs=0.7Vo/Vs=0.4Vo/Vs=0.1

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

0 0.5 1 1.5 2 2.5 3 3.5 4

t/k

V/V

s

Vo/Vs=1.6Vo/Vs=1.3Vo/Vs=1.0Vo/Vs=0.7Vo/Vs=0.4Vo/Vs=0.1

0 0

Plans for 2006 SHIPS Model

• Operational Version– 16 basic predictors + OHC/GOES correction– New speed adjusted MPI– New 250 hPa T predictor– Cione SST cooling algorithm (pending JHT approval)– New decay formulation (pending JHT approval)– Recon info included on SHIPS output, but not included

in the prediction– Can also show forecast with generalized prediction

equation• Parallel version

– Modified shear calculation

Rapid Intensity Index

• Uses subset of SHIPS input most correlated with rapid intensity change

• Estimates probability of 25 kt increase in next 24 hours– Original version used 30 kt threshold

• Atlantic and east Pacific versions• Results included on SHIPS text output

RII Brier Skill Score 2004-2005

RII Improvements

• Original version used binary method– 0 or 1 depending on if predictor exceeded threshold

• Updated version scales predictors between 0 and 1

• Current project– Use discriminant analysis to determine optimal

weights for combining predictors– To be developed during J. Kaplan visit to CIRA April

10-14th – Will be evaluated on independent 2006 cases

Acknowledgments

• Thanks to the TPC and JHT support staff– Chris Sisko– Alison Krautkramer– Chris Lauer– Jim Gross– Michelle Mainelli