Assessment of Tropical Rainfall Potential TRaP forecasts during the 200304 Australian tropical cyclo

1
Assessment of Tropical Rainfall Potential (TRaP)
forecasts during the 2003-04 Australian tropical
cyclone season

• Beth Ebert
• BMRC, Melbourne, Australia
• with thanks to Sheldon Kusselson, Mike Turk,
  Ralph Ferraro and Bob Kuligowski

2nd IPWG Meeting, Monterey, 25-28 October 2004
2
TRaP - Tropical Rainfall PotentialNESDIS
nowcasts of rain in tropical cyclones

• Generation of TRaP
• Compute areal rain rates from passive microwave
  sensor (SSM/I, AMSU, or TRMM)
• Using operational forecast cyclone track, advect
  rainfall for 24 h, assuming steady state storm
  structure
• Analyst vets TRaP prior to public release

SSM/I "snapshot"
DARWIN
TRaP
DARWIN
TC Craig, 10 March 2003
3
Validation of TRaP over U.S. for 2002 Atlantic
hurricane season

• (Ferraro et al., 2004, Wea. Forecasting,
  submitted)

42 TRaPs verified against Stage IV radar/gauge
analyses at 4 km resolution. TRaP
under-estimated rain rate, volume, max. TRaPs
from TRMM performed best, closely followed by
AMSU. TRaP outperformed Eta NWP model forecasts
at 50 km resolution.
SSM/I AMSU TRMM
4
Atlantic vs. South Pacific hurricane rainfall

• Mean rain rate from TRMM TMI as a function of
  radial distance from storm center, 1998-2000

Atlantic
South Pacific
Lonfat et al., 2004, Mon. Wea. Rev.
5
2003-2004 Australian tropical cyclones

• Validation strategies
• maximum 24 h rain at landfall vs. rain gauge
  observations 3h (12 h)
• spatial rainfall distribution in 10 box vs.
  operational 0.25 gauge analysis 3h
• contiguous rain area (CRA) bounded by 20 mm d-1
  threshold vs. operational 0.25 gauge
  analysis 3h

6
Tropical Cyclone Fay (17-28 March 2004)
landfall

• TRaP too great on most days, especially near
  landfall
• Some extreme values for SSM/I and TRMM
• Areal TRaP vs gauge observations not ideal but
  no radar data available

7
Tropical Cyclone Fay (28 March 2004)
AMSU
OBS

• Maximum 24 h rain (mm)
• Observed 159.4
• 0100 UTC 28 March 2004
• AMSU 111.6
• 0233 UTC 28 March 2004
• SSM/I 478.1
• 1304 UTC 27 March 2004
• TRMM 251.0
• 0156 UTC 28 March 2004

SSM/I
TRMM
8
Maximum rain at landfall
Mean

• TRaP estimated maximum rain well for some TCs,
  overestimated for others
• AMSU less likely to overestimate

9
Spatial validation - TC Fay (28 March 2004)

• statistics for land grid boxes only

10
Aggregated results all vs. vetted (checked by
analyst) TRaPs

• Rain area and volume too small by 50
• POD for heavy rain is 0.2-0.6, FAR is 0.2-0.6
• Vetted TRaPs perform better than all (unvetted
  vetted) TRaPs

11
Aggregated results sensor intercomparison

• SSM/I TRaPs had some large errors, AMSU had
  smallest errors
• AMSU TRaPs gave largest rain area
• AMSU TRaPs showed best performance, then TRMM,
  then SSM/I

12
CRA verification method (Ebert and McBride, 2000)

• Define entities using threshold (Contiguous Rain
  Areas)
• Location error determined by
• pattern matching (minimum total squared error,
  maximum correlation, or maximum overlap)
• external specification using best track data
• Verify properties of CRA (size, mean and maximum
  intensity, etc.)
• Error decomposition
• MSEtotal MSEdisplacement MSEvolume
  MSEpattern
• Version for pattern matching using correlation
• (rcorrelation, sstd.dev.)

13
CRA validationTC Fay (0303 UTC 25 March 2004)
14
CRA validationTC Monty (2216 UTC 1 March 2004)
15
CRA validation results for vetted TRaPs

• Pattern error most important, followed by volume
  error, then displacement error

16
Comparison to operational NWP

• Mesoscale model (mesoLAPS, 12 km resolution)
• TC-centered mesoscale model (TC-LAPS, 15 km
  resolution)
• 24 h rain forecasts for TC Monty, 00 UTC 2
  March 2004
• Verification on 0.25 grid consistent with TRaP
  verification

17
Comparison to operational NWP

• NWP models overestimated rain area and volume
• Correlations comparable between TRaP and models
• Threat score best for TC-LAPS
• Fairer comparison might use vetted TRaPs but not
  enough days in common

18
Comparison of Australian and US results (median
values for vetted TRaPs)
19
Reasons for differences
20
CRA validation suggests...
Location Error 18
Pattern error 48
Volume error 34
21
Living with uncertainty Ensemble TRaP

• Perturb or vary
• Cyclone track
• Parameters of microwave rain rate retrieval
• Satellite sensors included in the ensemble,
  including VIS/IR
• Sources of TC rain forecasts R-CLIPER, NWP, ...

TC Monty, 00 UTC 2 March 2004 Ensemble of 27 TRaP
forecasts (15 AMSU, 8 SSM/I, 4 TRMM) valid
within 12 h Mean includes histogram
transformation
22
(No Transcript)