Improving the Validation and Prediction of Tropical Cyclone Rainfall Timothy P. Marchok

Download Report

Transcript Improving the Validation and Prediction of Tropical Cyclone Rainfall Timothy P. Marchok

Improving the Validation and
Prediction of Tropical Cyclone Rainfall
Timothy P. Marchok
NOAA/GFDL Princeton, NJ
Robert F. Rogers
NOAA/AOML/HRD Miami, FL
Robert E. Tuleya
NOAA/NCEP/EMC/SAIC Camp Springs, MD
58th Interdepartmental Hurricane Conference, Charleston, SC
March 1-5, 2004
* This project is being funded by the Joint Hurricane Testbed (JHT)
Goals
• Goal: Develop a set of rainfall validation schemes
more suited specifically for TCs
 Motivation: Conventional precip skill scores are often difficult
to interpret in the context of tropical cyclones
• Goal: Produce model QPF error statistics for historic
U.S. landfalling storms.
 Motivation: No systematic QPF verification has been done
on NWS operational models for the specific subset of
landfalling U.S. tropical cyclones
• Goal: Develop a modified R-CLIPER that incorporates
vertical shear & storm track data.
 Motivation: Recent research has shown that including
vertical shear information can add structure to R-CLIPER's
currently symmetric rainfall distribution
Outline
• Standard QPF verifications
 GFDL model: 1995-2002 storms; rain gauge data
 NWS operational suite: 1998-2003 storms; Stage
IV gridded rainfall analyses
• Application of a new verification technique to
Hurricane Isabel’s landfall
• Future work
GFDL Model QPF Verification, 1995-2002
• Verification data: Hourly & daily rain gauges
• Gauge data within 800 km of storm track was
summed over model forecast interval
• Interpolated model storm total rainfall to
gauge location
• Average of 211 hourly gauges per storm
• Nearly 1200 daily gauges per storm
• Only used forecasts initialized at 12 UTC, to
coincide with daily rain gauge network
• All storms within ~24h of landfall
U.S. Landfalling Cases for Model Evaluation
NWS Operational Suite
GFDL, R-CLIPER
1995
1996
1997
1998
1999
2000
2001
2002
2003
Allison
60
Bertha
90
Danny
65
Bonnie
95
Bret
100
Gordon
55
Allison
45
Bertha
35
Bill
50
Dean
40
Fran
100
Charley
40
Dennis
60
Helene
65
Barry
60
Edouard
35
Claudette
75
Erin
75
Josephine
60
Earl
70
Floyd
90
Gabrielle
60
Fay
50
Grace
35
Frances
45
Harvey
50
Hanna
45
Henri
30
Georges
90
Irene
70
Isidore
55
Isabel
90
Opal
100
Hermine
35
Kyle
35
Lili
85
QPF Bias Scores: GFDL vs. R-CLIPER
Equitable Threat Scores
Model vs. obs correlation: A first
guess at pattern matching
Track error <-> QPF error
Storm Intensity <-> QPF Error
QPF verification for NWS model
suite: 28 storms from 1998-2003
• GFDL, NCEP/GFS, NCEP/Eta, R-CLIPER, 2xR-CLIPER
• GFDL 2003 version used for all 28 cases
• Obs data: Hourly multi-sensor (radar, gauge)
gridded data available online from NCAR
 2002-2003: NCEP/EMC Stage IV data
 1998-2001: NCEP/EMC Stage II data
QPF Bias Scores
QPF Equitable Threat Scores
GFS: 0.65
Eta:
0.56
GFDL: 0.52
RCLIP: 0.47
GFS: 0.65
Eta:
0.56
GFDL: 0.52
RCLIP: 0.47
Stage
IV
GFS
GFDL
Eta
Stage
IV
GFS
GFDL
Eta
Rainfall statistics for observations and forecasts of 24-h rain
from 12 UTC 18 to 12 UTC 19 September for Isabel (2003)
Total areal
coverage of rain (x
106 km2)
Total rain flux (x
106 in-km2)
Mean R (in)
Stage IV
1.201
1.536
1.28
R-CLIPER
1.936
2.191
1.13
GFDL
1.933
2.453
1.27
GFS
1.877
1.485
0.79
Eta
1.404
2.528
1.80
Plot of 24-h rain (in) from 12 UTC 18 to 12 UTC 19 September
for Isabel (2003) for NPVU and R-CLIPER
25
100
flux PDF (RCLIPER)
90
flux PDF (NPVU)
flux CDF (RCLIPER)
Stage IV
20
80
flux CDF (NPVU)
Frequency
(%) (%)
Frequency
15
60
50
10
40
5
20
10
R-CLIPER
0
0
0.1
1
10
100
Rain (in)
PDF and CDF comparisons of
rain flux binned by rain amount
(%)
30
(%)
frequency
Cumulative
Cumulative
frequency
70
Plot of 24-h rain (in) from 12 UTC 18 to 12 UTC 19
September for Isabel (2003) for NPVU and GFDL
25
100
90
flux PDF (GFDL)
flux PDF (NPVU)
Stage IV
flux CDF (GFDL)
20
80
flux CDF (NPVU)
Frequency
(%) (%)
Frequency
15
60
50
10
40
5
20
10
GFDL
0
0
0.1
1
10
100
Rain (in)
PDF and CDF comparisons of
rain flux binned by rain amount
(%)
30
(%)
frequency
Cumulative
Cumulative
frequency
70
Plot of 24-h rain (in) from 12 UTC 18 to 12 UTC 19
September for Isabel (2003) for NPVU and GFS
25
100
flux PDF (GFS)
90
flux PDF (NPVU)
flux CDF (GFS)
Stage IV
20
80
flux CDF (NPVU)
(%) (%)
Frequency
Frequency
15
60
50
10
40
5
20
10
GFS
0
0
0.1
1
10
100
Rain (in)
PDF and CDF comparisons of
rain flux binned by rain amount
(%)
30
(%)
frequency
Cumulative
Cumulative
frequency
70
Plot of 24-h rain (in) from 12 UTC 18 to 12 UTC 19
September for Isabel (2003) for NPVU and Eta
25
100
flux PDF (ETA)
90
flux PDF (NPVU)
flux CDF (ETA)
Stage IV
20
80
flux CDF (NPVU)
(%) (%)
Frequency
Frequency
15
60
50
10
40
5
20
10
Eta
0
0
0.1
1
10
100
Rain (in)
PDF and CDF comparisons of
rain flux binned by rain amount
(%)
30
Cumulative
frequency
(%)
frequency
Cumulative
70
Probability-matched 24-h rain estimates from Stage IV data vs.
R-CLIPER, GFDL, GFS, and Eta models for Hurricane Isabel
10
RCLIPER
GFDL
GFS
ETA
90%
90%
forecast R (in)
90%
1
10%
10%
10%
10%
0.1
0.1
1
observed R (in)
10
Summary & Future Plans
• QPF error stats for landfalling TCs:
 GFDL only: 1995-2002 vs. rain gauge data
 NWS suite: 1998-2003 vs. gridded Stage IV data
 Compare operational GFDL vs. 2003 GFDL for all storms,
1995-2003
• Development of a set of QPF verification schemes
more suited to TCs
 Apply a PDF/CDF method to TC rain flux
 Use PDF/CDF method on NWS suite, 1998-2003
 Investigate other spatial-based methods?
• Develop a modified R-CLIPER that incorporates
vertical shear & storm track data (Year 2)