Historic Look at the Tropical Cyclone Track Forecasting Skill
of Operational NWP Models and their Consensus

James S. Goerss and Charles R. Sampson

Marine Meteorology Division
Naval Research Laboratory, Monterey, CA

James M. Gross

Tropical Prediction Center, National Hurricane Center
NOAA/NWS, Miami, FL

Abstract:

The tropical cyclone (TC) track forecasting skill of operational NWP models and their consensus is examined for the Atlantic from 1998 to 2002 and for the western North Pacific from 1992 to 2002. The operational models available for use by the NHC forecasters for the Atlantic today include NOGAPS, UKMO, GFDL, AVN, GFDN, and COAMPS, while those currently available for use by the JTWC forecasters for the western North Pacific include NOGAPS, UKMO, AVN, GFDN, COAMPS, the Japanese global spectral model (JGSM), the Japanese typhoon model (JTYM), and AFWA's MM5. All of the model forecast tracks used in this study come from the interpolated versions of the operational models and are representative of the guidance that would normally be available in real time to the operational forecasters.

The TC track forecasting skill of the operational NWP models is steadily improving. For the western North Pacific, the typical 72-h model forecast error has decreased from roughly 320 nm to roughly 220 nm over the past ten years and is now comparable to the typical 48-h model forecast error ten years ago. For the Atlantic, the ranges of model forecast errors in 1998 were 96-107 nm, 169-221 nm and 273-337 nm at 24 h, 48 h, and 72 h, respectively. The respective ranges for 2002 were 73-86 nm, 144-162 nm, and 209-248 nm. Furthermore, over the years the number of operational NWP models capable of making quality TC track forecasts has increased. For example, with the marked improvement in the AVN and the availability of GFDN and COAMPS, the number of models available to the NHC forecasters for the Atlantic rose from 3 to 6 in 2000. Prior to 1996, TC track forecasts from only 3 operational models (NOGAPS, UKMO, and JTYM) were available to the forecasters at JTWC. By 2001, the JTWC forecasters had 8 models available with the addition of JGSM and GFDN in 1996, the addition of COAMPS and the MM5 in 2000, and the addition of AVN in 2001.

Consensus forecast aids are routinely utilized by the forecasters at both NHC and JTWC. In this study we examine the performance of consensus aids that require that TC track forecasts from at least two models from a specified pool of models be available. For the 1998-1999 Atlantic hurricane seasons, the TC track forecast errors for a consensus utilizing NOGAPS, UKMO, and GFDL were 76 nm, 146 nm, and 209 nm at 24 h, 48 h, and 72 h, respectively. For the 2000-2002 Atlantic hurricane seasons, the respective errors for this 3-model consensus were 72 nm, 134 nm, and 208 nm. With the addition of the AVN, the forecast errors for 2000-2002 for the 4-model consensus improved to 69 nm, 127 nm, and 198 nm, respectively. The respective consensus forecast errors again improved slightly to 68 nm, 126 nm, and 194 nm with the addition of GFDN and COAMPS. While the addition of models to the consensus has a modest impact on the forecast skill, it has a more marked impact on consensus forecast availability. In 1998, the forecast availability of the 3-model consensus was 90%, 88%, and 83% at 24 h, 48 h, and 72 h, respectively. The forecast availability of the 6-model consensus in 2002 was 95% for all three forecast lengths. Similarly, for the western North Pacific in 2002, the 72-h forecast errors for a 2-model, 5-model, 7-model, and 8-model consensus were 188 nm, 177 nm, 170 nm, and 159 nm, respectively. The forecast availability for these consensus forecasts was 84%, 89%, 92%, and 97%, respectively.