A hurricane model developed at GFDL, NOAA, was combined with each of AVN and NOGAPS global analyses to construct typhoon prediction systems GFDS and GFDN, respectively. The GFDS system performed 125 (178) forecast experiments for 16 (24) storms in the western North Pacific basin during 1995 (1996). It exhibited considerable skill in the forecast of tropical cyclone tracks. The average forecast position errors at 12, 24, 36, 48 and 72h in 1995 (1996) were 95 (108), 146 (178), 193 (227), 249 (280), and 465 (480) km. The improvement with GFDS in the typhoon position forecast over CLIPER was roughly 30%. The reduction of position errors in both average and standard deviations indicates superior forecast accuracy and consistency of GFDS, although there existed systematic northward bias in the forecast motion at low latitudes. On the other hand, intensity forecast was not satisfactory, showing a tendency to overpredict weak storms and underpredict strong storms, similar to the tendency in the Atlantic.
Two sets of forecasts performed in the 1996 season, the one by GFDS and the other by GFDN, were compared with each other. Forecast skills of the storm position with the two systems were comparable. However, the two forecast positions tended to be systematically biased toward different directions. As a result, when the two forecasts were averaged, the mean error was 10% smaller than that of each forecast. Also, overall improvement in track forecast was obtained in supplemental experiments in which individual forecasts were corrected for systematic biases. Though systematic bias is not steady, there may be ways to utilize it for improvement of tropical cyclone forecasts.
Meteorological Society of Japan