Predicting Tropical Cyclone Track Forecast Error Joint Hurricane Testbed Project James S. Goerss

Predicting Tropical CycloneTrack Forecast Error

Joint Hurricane Testbed Project

James S. GoerssNRL Monterey

Consensus Forecast Properties

Consensus TC track forecast error is dependent upon two things (Goerss 2000):

1) The TC track forecast error of the individual NWP models that make up the consensus

2) The degree of independence (or effective degrees of freedom) of the TC track forecast errors of the individual NWP models that make up the consensus

Consensus Forecast Models

• The interpolated versions of five high-quality TC track forecast models available to NHC were used in this study. The five models are: NOGAPS and

GFDN (NGPI and GFNI; FNMOC), GFS and GFDL (AVNI and GFDI; NCEP), and the UKMO global model (UKMI).

• GUNA is a consensus model computed when the track forecasts from all four models (GFDI, AVNI, NGPI, and UKMI) are available

• CONU is a consensus model computed when track forecasts from at least two of the five models (GFDI, AVNI, NGPI, UKMI, and GFNI) are available

0

50

100

150

200

250

300

350

400

24 48 72 96 120

GFDIAVNIUKMINGPIGUNA

2001-2003 Atlantic GUNA EnsembleTC Forecast Error (nm)

The only models available out to 120 h are NOGAPS, the NCEP global forecast system, and the UK Met Office global model. Thus, they are the only models available to CONU at 96h and 120h. Still, the CONU 120h forecast error is comparable to typical 72h forecast error in the western North Pacific in the 1990’s.

The only models available out to 120 h are NOGAPS, the NCEP global forecast system, and the UK Met Office global model. Thus, they are the only models available to CONU at 96h and 120h. Still, the CONU 120h forecast error is comparable to typical 72h forecast error in the western North Pacific in the 1990’s.

619 358 176 Number of Forecasts467 229

0

50

100

150

200

250

300

24 48 72 96 120

OFCLGUNACONU

2001-2003 Atlantic TC Forecast Error (nm) Homogeneous Comparison

The only models available out to 120 h are NOGAPS, the NCEP global forecast system, and the UK Met Office global model. Thus, they are the only models available to CONU at 96h and 120h. Still, the CONU 120h forecast error is comparable to typical 72h forecast error in the western North Pacific in the 1990’s.

The only models available out to 120 h are NOGAPS, the NCEP global forecast system, and the UK Met Office global model. Thus, they are the only models available to CONU at 96h and 120h. Still, the CONU 120h forecast error is comparable to typical 72h forecast error in the western North Pacific in the 1990’s.

621 469 176 Number of Forecasts360 231

0102030405060708090100

24 48 72 96 120

GUNACONU

2001-2003 Atlantic Forecast Availability Percentage

The only models available out to 120 h are NOGAPS, the NCEP global forecast system, and the UK Met Office global model. Thus, they are the only models available to CONU at 96h and 120h. Still, the CONU 120h forecast error is comparable to typical 72h forecast error in the western North Pacific in the 1990’s.

The only models available out to 120 h are NOGAPS, the NCEP global forecast system, and the UK Met Office global model. Thus, they are the only models available to CONU at 96h and 120h. Still, the CONU 120h forecast error is comparable to typical 72h forecast error in the western North Pacific in the 1990’s.

Predicting TC Track Forecast Error

• The purpose of this study is to determine to what extent the TC track forecast error of the consensus models

GUNA and CONU can be predicted prior to the time when official forecasts must be issued.

• Predictors of consensus forecast error must be quantities that are available prior to the time when official forecasts must be issued.

• Consensus model spread is defined to be the average distance of the member forecasts from the consensus forecast.

• The possible predictors examined in this study are consensus model spread; initial and forecast TC intensity; initial and forecast displacement of TC location (latitude and longitude); TC speed of

motion; and number of members available (for CONU).

76%

Spread, Lati, Inti, Lond, Loni, Latd, Speed (R=.717)

Hurricane Isabel – 00Z September 13, 2003Predicted CONU 76% Confidence Radius

Hurricane Kate – 00Z September 30, 2003Predicted CONU 76% Confidence Radius

Fit to Independent Data

76%

Spread, Lati, Inti, Lond, Loni, Latd, Speed (R=.717)

71%

Summary

• While the TC track forecast errors for the consensus model, CONU, are comparable to those for GUNA, the forecast

availability for CONU is superior to that for GUNA.

• At all forecast lengths, the consensus model spread was found to be related to consensus model TC track forecast error. The strongest correlation was found for 96-h and 120-h forecasts. TC intensity was found to be consistently but less strongly related to track error.

• Using stepwise linear regression and a pool of predictors consisting of spread, initial and forecast intensity, initial and forecast displacement of position (latitude and longitude), speed of motion, and number of members, the percent variance of CONU TC track forecast error explained for the 2001-2003 Atlantic seasons ranged from just under 20% at 24h and 48h to roughly 50% at 96h and 120h. Similar results were found for GUNA.

Summary

• The range of predicted uncertainty for CONU forecasts containing the verifying TC position 73-76% of the time varied from 15-140 nm at 24 h, 30-250 nm at 48 h, 50-580 nm at 72 h, 60-1060 nm at 96 h, and 100-1200 nm at 120

h. (NHC Potential Day 1-5 Track Area graphic uses 81, 150, 225, 282, and 374 nm, respectively)

• Independent data testing indicated that one can expect only a small degradation in this predictive capability in practice.

• The relationship between ensemble spread and the TC forecast track error was found to be weak for both the NCEP GFS and ECMWF ensemble prediction systems.

Questions?

Why use a “poor man’s ensemble” when you have the NCEP GFS or ECMWF ensemble prediction

systems available?

0

50

100

150

200

250

300

24 48 72

AP01AP02AP03AP04AP05AEMNAVNO

2001-2003 Atlantic GFS EnsembleTC Forecast Error (nm)

The only models available out to 120 h are NOGAPS, the NCEP global forecast system, and the UK Met Office global model. Thus, they are the only models available to CONU at 96h and 120h. Still, the CONU 120h forecast error is comparable to typical 72h forecast error in the western North Pacific in the 1990’s.

The only models available out to 120 h are NOGAPS, the NCEP global forecast system, and the UK Met Office global model. Thus, they are the only models available to CONU at 96h and 120h. Still, the CONU 120h forecast error is comparable to typical 72h forecast error in the western North Pacific in the 1990’s.

312 223 156 Number of Forecasts

0

50

100

150

200

250

300

24 48 72

AN01AN02AN03AN04AN05AEMNAVNO

2001-2003 Atlantic GFS EnsembleTC Forecast Error (nm)

The only models available out to 120 h are NOGAPS, the NCEP global forecast system, and the UK Met Office global model. Thus, they are the only models available to CONU at 96h and 120h. Still, the CONU 120h forecast error is comparable to typical 72h forecast error in the western North Pacific in the 1990’s.

The only models available out to 120 h are NOGAPS, the NCEP global forecast system, and the UK Met Office global model. Thus, they are the only models available to CONU at 96h and 120h. Still, the CONU 120h forecast error is comparable to typical 72h forecast error in the western North Pacific in the 1990’s.

312 223 156 Number of Forecasts

05101520253035404550

24 48 72 96 120

GUNAAEMN

2001-2003 AtlanticGUNA and AEMN Ensembles

Percent Variance of Forecast Error Explained by Spread

The only models available out to 120 h are NOGAPS, the NCEP global forecast system, and the UK Met Office global model. Thus, they are the only models available to CONU at 96h and 120h. Still, the CONU 120h forecast error is comparable to typical 72h forecast error in the western North Pacific in the 1990’s.

The only models available out to 120 h are NOGAPS, the NCEP global forecast system, and the UK Met Office global model. Thus, they are the only models available to CONU at 96h and 120h. Still, the CONU 120h forecast error is comparable to typical 72h forecast error in the western North Pacific in the 1990’s.

73%

Inti, Spread, Speed, Lond, Lati (R=.438)

Intf, Spread, Lati, Num, Lond, Latd (R=.433)

74%

Intf, Spread, Latd, Num, Lati (R=.544)

73%

Spread, Latd, Num, Lond, Inti (R=.696)

73%