+ All Categories
Home > Documents > Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint...

Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint...

Date post: 12-Jan-2016
Category:
Upload: hillary-mosley
View: 219 times
Download: 0 times
Share this document with a friend
Popular Tags:
25
Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA, Fort Collins, CO ohn A. Knaff, Jack Dostalek and Kimberly J. Mueller CSU/CIRA, Fort Collins, CO Collaborators : Jim Gross (TPC), Charles Anderson (CSU), Buck Sampson (NRL),Miles Lawrence(TPC), Chris Sisko (TPC) Presented at The Inter-Departmental Hurricane Conference March 3, 2004 Charleston, SC
Transcript
Page 1: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Improvements in Deterministic and Probabilistic

Tropical Cyclone Surface Wind Predictions  

Joint Hurricane Testbed Project Status Report

Mark DeMariaNOAA/NESDIS/ORA, Fort Collins, CO

 John A. Knaff, Jack Dostalek and Kimberly J. Mueller

CSU/CIRA, Fort Collins, CO

Collaborators: Jim Gross (TPC), Charles Anderson (CSU),Buck Sampson (NRL),Miles Lawrence(TPC), Chris Sisko (TPC) 

Presented at The Inter-Departmental Hurricane ConferenceMarch 3, 2004 Charleston, SC

Page 2: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

OUTLINE• Deterministic Intensity Forecast Improvements

– Can inner core data from aircraft and satellite improve SHIPS forecasts?

• Automated objective analysis and EOF analysis

– Compare neural network and linear regression models

• Probabilistic Surface Wind Forecast Improvements– Calculate operational track/intensity and wind radii-

CLIPER error distributions– Randomly sample errors using Monte Carlo method

• Generate probabilities of 34, 50, 64 and 100 kt winds

Page 3: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Decay-SHIPS and NHC Intensity Forecast Skill 2001-2003

-30

-25

-20

-15

-10

-5

0

5

10

15

20

12 24 36 48 60 72 84 96 108 120

Forecast Interval (hr)

Err

or

Rel

ativ

e to

SH

IFO

R (

%)

Decay SHIPS

NHC Official

Page 4: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

5 Basic Radial Profiles (Samsury and Rappaport 1991)

1

2

3

4

5

• Develop objective method for extracting similar information• Supplement with inner-core GOES data

Page 5: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Variational Wind Analysis for Aircraft Data

• Combine 12 hours of recon data in storm-relative coordinates

• Perform automated quality control– Analyze data to determine if coverage is sufficient

• Designed to measure at least azimuthal wavenumber 0 and 1 – Compare data to “pre-analysis” to eliminate bad points

• Perform “variational” analysis to provide u,v on radial, azimuthal grid – azimuthal smoothing >> radial smoothing– Based on Thacker and Long (1988)

• Preliminary prediction based upon azimuthal average tangential wind

Page 6: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

AF Recon Flight Level Winds for Hurricane LiliEarth-Relative 10/02/02 0000-1200 UTC

Page 7: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

AF Recon Flight Level Winds for Hurricane LiliStorm-Relative 10/02/02 0000-1200 UTC

Page 8: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Variational Wind Analysis for Lili10/02/02 0000-1200 UTC

Page 9: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Isotachs (kt) from Variational Wind Analysis for Lili10/02/02 0000-1200 UTC

Page 10: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Isotach Analyses for Hurricane Lili10/01 0000 UTC – 10/03 1200 UTC

Page 11: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

0

10

20

30

40

50

60

0 20 40 60 80 100

120

140

160

180

200

Radius (km)

Tan

gen

tial

Win

d (

m/s

)

10/01 00

10/01 12

10/02 00

10/02 12

10/03 00

10/03 12

Azimuthally Averaged Tangential Wind (r=0 to 200 km)Hurricane Lili 10/01 00 UTC to 10/03 12 UTC

Page 12: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Comparison of Best Track and Variational Analysis Maximum Wind

(1995-2002 Cases)

y = 0.9743x - 0.1111

R2 = 0.903

0

20

40

60

80

100

120

140

160

180

0 20 40 60 80 100 120 140 160 180

Analysis Max Wind (kt)

Bes

t T

rack

Max

Win

d (

kt)

Page 13: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

EOF Analysis

• ~400 cases with recon and IR data (95-03)• 51 radial grid points, r = 4 km• How to relate 102 IR and wind values to intensity

change?– Empirical Orthogonal Function (EOF) Analysis

– Mathematical technique for extracting common patterns from datasets

– Apply to tangential wind and IR radial profiles

– Work with small set of patterns instead of the entire profiles

Page 14: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Variance Explained by each EOF

0

10

20

30

40

50

60

70

80

90

1 6 11 16 21 26 31 36 41 46 51

Eigenvalue Number

Var

ian

ce E

xpla

ined

(%

)

Tangential Wind

IR Brightness TTang. Wind:99% w\ 6 EOF

IR Brightness T:99% w\ 4 EOF

Page 15: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Tangential Wind and IR EOFs

-0.2

0

0.2

0.4

0.6

0.8

2 22 42 62 82 102 122 142 162 182 202

Radius (km)

No

rmal

ized

Val

ue

EOF 1

EOF 2

EOF 3

-0.4

-0.2

0

0.2

0.4

0.6

2 22 42 62 82 102 122 142 162 182 202

Radius (km)

No

rmal

ized

Val

ue

EOF 4

EOF 5

EOF 6

-0.4

-0.2

0

0.2

0.4

0.6

2 22 42 62 82 102 122 142 162 182 202

Radius (km)

No

rmal

ized

Val

ue

EOF 1

EOF 2

EOF 3

EOF 4

Tang. Wind 1-3

IR 1-4

Tang. Wind 4-6

Page 16: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Part 1 Project Schedule

• Spring 2004: Develop statistical intensity model using EOF amplitudes – Provide adjusted SHIPS forecast based upon inner core

information

• Spring 2004: Compare neural network and regression techniques– Collaboration with Dr. Charles Anderson, CSU Computer Science

Department (Expert in Machine Learning Techniques)

• Summer 2004: Implement variational aircraft analysis at NHC/JHT

• Summer/Fall 2004: Test results on real-time forecasts

Page 17: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Preliminary Neural Network ResultsDependent data test with 1989-2002 Sample

Page 18: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Monte Carlo Model for Tropical Cyclone Surface Wind Probabilities

(Initial support from Insurance Friends of the National Hurricane Center)

• Calculate NHC track and intensity errors (along track and cross track) from multi-year sample

• Determine large set of tracks and intensities (realizations) centered around official forecast by randomly sampling from error distributions

• Estimate wind radii distributions from errors of radii-CLIPER model

• Calculate probabilities by number of times specified point comes with radii of specified wind speed relative to total number of realizations

• Run in real-time in 2003 season (starting August)

Page 19: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Monte CarloWind ProbabilityModel

Example:

Hurricane Fabian

Aug 31 2003 18Z

Vmax=115 kt

R34 100 75 75 100R50 30 30 30 30R64 20 20 20 20

Page 20: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Modifications based on 2003 Results

• Model Changes– Improved portable random number generator– Complete error field sampling (instead of 1-99th percentiles)– Modified for use in the Atlantic, East/Central Pacific, and western North

Pacific basins (i.e., Longitude … 0-360)– Option for 100 kt radii added for JTWC

• Error Components– Improved radii-CLIPER model

• Inclusion of initial wind radii asymmetries – radii match observed at t=0 hr

• R34 bias correction

– Intensity errors account forecast intensity and distance to land– Distributions being updated with 2003 cases

Page 21: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Old New

Impact of Model Changes(Fabian 2003 Example)

Page 22: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

3.34

1.62

0.600.33

0.18

0.06

0.02

15.759.53

4.452.96 2.45

1.09

0.48

0.01

0.1

1

10

100

10 100 1000 10000 100000

Number of Realizations

Pro

ba

bil

ity

Err

or

(%) Average Maximum

Effect of Number of Realizations on Probability Estimate

Page 23: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Sensitivity to the number of realizations

N=500 N=1000

N=2000 N=500000

Page 24: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

R34 R50

R64 R100

TyphoonMaemi

9/9/04 06 Z

Vmax=115 kt

R34 130 130 130 130R50 50 50 50 50R100 20 20 20 20

N=2000

Page 25: Improvements in Deterministic and Probabilistic Tropical Cyclone Surface Wind Predictions Joint Hurricane Testbed Project Status Report Mark DeMaria NOAA/NESDIS/ORA,

Part 2 Project Schedule

• Spring 2004: Investigate variable grid options – Improve efficiency and for NDFD applications

• Spring 2004: Finalize probability model for 2004 season

• Summer/Fall 2004: Run at NHC in real-time for Atlantic and East Pacific cases

• Summer/Fall 2004: Coordinate with JTWC for real-time tests (directly on their ATCF)

• Winter 2004: Evaluate results from 2004 runs

• 2004 “Freebie”: Provide NHC and JTWC with updated Radii-CLIPER models


Recommended