GFDL颱風預報系統對於1995年及1996年西北太平洋地區颱風測試結果評估

Evaluation of the GFDL Hurricane Prediction System in the Western North Pacific in 1995 and 1996

吳俊傑(Chun-Chieh Wu) ； 卓瑜甄(Yu-Tseng Cho)

《大氣科學》 25卷4期 (1997/12) Pp. 449-481

摘要

本研究測試GFDL（GeophysicaI Fluid Dynamics Laboratory）颱風預報系統對於西太平洋地區颱風的預報情形，以瞭解此系統在現階段對於西太平洋颱風之預報能力及被參考價值。在1995年颱風季節中，我們對西北太平洋16個颱風做125個案的真時GFDL颱風模式預報。在路徑預報方面GFDL颱風模式的預報結果相較於CLIPER、中央氣象局的TFS和EBM、及美國的AVN與NOGAPS等颱風模式的平均路徑誤差約減少35%。另外GFDL颱風路徑預報誤差亦低於關島和台灣之官方預報，其改進超過30%。1996年之颱風季節中，我們復對西北太平洋23個颱風做171個案的真時GFDL颱風模式預報。在路徑預報方面，GFDL颱風預報系統對1996年西北太平洋颱風之路徑預報在所有預報個案中，對12，24，36，48，60及72小時的各預報時間，其平均預報路徑誤差分別為108，174，220，274，352及436公里。路徑預報偏差及強度預報偏差之結果則顯示雖然GFDL颱風預報系統對於西北太平洋颱風預報之系統偏差會因年份不同而有所變化，但基本上仍略具一定的系統偏差：即在路徑預報方面GFDL系統具有偏北與相對於行進方向偏右的趨勢（特別是在1995年），且會低估高緯度颱風之移速，在強度預報偏差上，GFDL系統明顯具有高報強度較弱之颱風而低報強度較強之颱風的趨勢，這顯示模式本身及初始化過程尚有許多需要改進之空間。由賴恩颱風（1995）個案結果凸顯GFDL颱風預報對於初始資料之敏感度問題，及台灣地區對於颱風預報誤差之較嚴格需求。此外，賀伯颱風（1996）個案分析結果則顯示GFDL預報系統對於風雨及中小尺度現象已稍具預報能力。目前我們正在規劃更多之測試以探討GFDL預報系統表現與其他現有模式差異之主因，以及瞭解颱風預報模式對初始資料場之敏感度。我們期望藉由這些測試以釐清現階段颱風預報所面臨之關鍵問題，並協助改善國內之颱風預報。

關鍵字

颱風 ； GFDL颱風預報系統

並列摘要

The Geophysical Fluid Dynamics Laboratory (GFDL) hurricane prediction system was tested at the National Centers for Environmental Prediction in experimental mode and in near real time for western North Pacific storms during 1995 and 1996.Altogether, 125 (171) forecasts were run for 16 (23) storms in 1995 (1996). The forecast tracks in 1995 were compared with the official JTWC (Joint Typhoon Warning Center) and CWB (Central Weather Bureau) forecast available just after the initial synoptic time, as well as forecasts from other operational models, such as CLIPER (CLImatology and PERsistence), TFS (Typhoon Forecasting System) of CWB, NOGAPS (Navy Operational Global Prediction System) and AVN (AViatioN model). Results in 1995 indicate that the GFDL system exhibited considerable skill in the forecast of the storm track for the cases run. The average improvement of the GFDL track forecasts over other models is roughly 35%. In addition, the improvement of the GFDL forecasts over the official forecasts from JTWC and CWB is about 30%. The forecasts from 1996 also show comparable performance, with mean track errors of 108, 174, 220, 274, 352 and 436 km for 12, 24, 36, 48, 60 and 72h, individually.Despite the demonstrated skill of the GFDL forecast system, some biases were revealed in the storm track and intensity. In general, a northward and a rightward (relative to the storm's heading direction) bias is found (especially in 1995). As to the intensity forecasts, an over-prediction of the intensity of weak storms and an under-prediction of the intensity of strong storms by the GFDL system is found, thus suggesting a need for some improvement both in the prediction model and its initialization.The case studies of Typhoon Ryan (1995) and Typhoon Herb (1996) are also discussed. The sensibility of the GFDL system to its initial condition is identified. Certain capability, in forecasting the distribution of precipitation and wind is also demonstrated. Overall, results from this study clearly indicate that the GFDL hurricane prediction system can serve as a useful dynamical model guidance for tropical cyclone forecasters in the western North Pacific region. More tests and analyses are still needed to improve the current GFDL system and to identify the key factors that affect the typhoon forecasting system. It is also hoped that results from these studies can help improving typhoon forecasts in Taiwan.

並列關鍵字

Typhoon ； GFDL hurricane prediction system