Ensemble Corrected Consensus and Weighted Consensus TC Track

Ensemble, Corrected Consensus, and Weighted Consensus TC Track Forecasts James S. Goerss NRL Monterey March 4, 2009

or How to Beat Consensus* *at least for the Atlantic basin in 2008

Track Forecast Guidance Interpolated Model Guidance AVNI – Global Forecast System (GFS) run at NCEP GFDI – GFDL model run at NCEP HWFI – Hurricane WRF run at NCEP NGPI – Navy global model (NOGAPS) run at FNMOC GFNI – GFDL model run at FNMOC EGRI – UK Met Office global model EMXI – ECMWF global model TVCN – Consensus of above models (at least two) AEMI – NCEP GFS ensemble mean

Corrected Consensus (TVCC) • Possible predictors are consensus model spread; initial and forecast TC intensity; initial TC location and forecast displacement of TC location (latitude and longitude); TC speed of motion; and number of members available. • Regression models to predict TVCN east-west and northsouth forecast error for all forecast lengths were derived using the aforementioned pool of predictors for the 2002 -2007 seasons • The means of the TVCN east-west and north-south forecast errors for all forecast lengths were also found for the 20022007 seasons to be used as bias correctors. • Optimal combinations of the statistical and bias correctors are used to produce corrected consensus (TVCC) forecasts for the Atlantic and eastern North Pacific basins.

2008 Atlantic Non-Homogeneous TC Forecast Error (nm) * Best Non-Consensus Guidance * * 314 251 204 160 * 131 Number of Forecasts

2008 Atlantic Forecast Availability Percentage

2008 Atlantic TC Forecast Error Percent Improvement over GFDI

Weighted Consensus Guidance Based on the forecast performance for EMXI and AEMI in the Atlantic for 2008, the following weighted consensus forecast aids were computed: CNW 6 – Determined just like TVCN except, that instead of equal weight for all models, 6 times the weight was applied to EMXI (roughly the same as taking the average of TVCN and EMXI) CON 3 – Determined by taking the average of the TVCN, EMXI, and AEMI forecasts (at least TVCN).

2008 Atlantic Homogeneous TC Forecast Error (nm) 306 249 202 158 131 Number of Forecasts

2008 Atlantic TC Forecast Error Percent Improvement over TVCN

2008 Eastern North Pacific Non-Homogeneous TC Forecast Error (nm) * Best Non-Consensus Guidance * * * 238 178 130 * * 90 57 Number of Forecasts

2008 Eastern North Pacific Forecast Availability Percentage

2008 Eastern North Pacific TC Forecast Error Percent Improvement over Best Model

2008 Eastern North Pacific Homogeneous TC Forecast Error (nm) 236 178 129 89 56 Number of Forecasts

2008 Eastern North Pacific TC Forecast Error Percent Improvement over TVCN

Summary • For both the Atlantic and eastern North Pacific basins in 2008, the corrected consensus (TVCC) forecast errors were not much better or worse than those for TVCN. • Due to the superior performance of EMXI and AEMI (at 96 h and 120 h) in the Atlantic, the forecast errors for CNW 6 were 6 -9 percent better than those for TVCN at all forecast lengths while those for CON 3 were 14 -16 percent better at 96 h and 120 h. • Due to the average performance of EMXI and inferior performance of AEMI in the eastern North Pacific, the forecast errors for CNW 6 were 5 -12 percent worse than those for TVCN forecast lengths greater than 72 h while those for CON 3 were 14 -26 percent worse forecast lengths greater than 24 h. • The Atlantic results indicate that weighted consensus TC track forecast aids are capable of producing improved track forecasts if one or more of the models perform considerably better than the others. However, whether such improved model performance is consistent from season to season has yet to be determined. The eastern North Pacific results suggest that such performance and the utility of weighted consensus track forecasts are likely to be basin dependent.

Questions?