- 學位論文
西北太平洋熱帶氣旋之展期預報評估
Evaluations of the Extended-Range Tropical Cyclone Forecasts in the Western North Pacific
摘要
本論文利用2006~2015年之NCEP GEFS模式(Global Ensemble Forecast System)之歷史再預報資料(reforecast)，配合熱帶氣旋偵測程式(tropical cyclone tracker)，與相同年份之颱風最佳路徑資料(best tracks)進行模式預報校驗。藉由颱風個數、二元式及機率式等不同的校驗方法，評估西北太平洋地區之熱帶氣旋預報技術，並分析模式預報技術與大尺度環境之關聯性，例如：聖嬰現象(El Niño-Southern Oscillation)及MJO(Madden-Julian Oscillation)。 二元式校驗結果顯示，在使用完整偵測條件的情形下，Very Strong El Nino(強度聖嬰年)有較好的預報技術，預兆得分(Threat Score)於第一週(6~168 小時)及第二週(174~336小時)分別為0.5及0.3。MJO的影響評估顯示，模式預報之初始時間落於MJO相位6具有較佳之熱帶氣旋活動的預測技術，第一週之預兆得分可接近0.5，第二週為0.32。機率式校驗結果顯示，ROC曲線(Receiver Operating Characteristic Curve)之AUC(Area Under Curve)於第一週及第二週約為0.78及0.69，顯示模式之颱風機率預報具有可接受的判別能力。颱風個數校驗結果顯示，NCEP GEFS預報可產生與觀測資料大致相符的颱風數量。 本論文亦分析NCEP GEFS於台灣地區之熱帶氣旋預報技術。二元式校驗結果顯示，強度聖嬰現象於第一週的有較好的預報技術，預兆得分可達0.61。當模式初始時間落於MJO相位6時，模式具有較佳之熱帶氣旋預報技術，第一週及第二週之預兆得分為0.6及0.26。機率式校驗之ROC曲線分析顯示，第一週預報之AUC約0.89，第二週預報約為0.75，顯示模式之颱風機率預報具有良好的判別能力。台灣區域之颱風個數預報校驗結果顯示，模式於第一週具有較佳之預報技術。因此NCEP GEFS適合提供台灣區域「是否可能有颱風侵台」或「颱風侵台機率」之預報參考資訊。
並列摘要
This study uses the NCEP Global Ensemble Forecast System (GEFS) reforecasts from 2006 to 2015 to evaluate the predictability of extended-range tropical cyclone (TC) forecasts in the western North Pacific. A TC trcker is used to detect TCs in the reforecasts, and the best tracks are used for forecast verifications. Various verification methods are used to analyze the TC activity forecast skills in week-1 and week-2. This study also investigates the relationships between forecast skill and the large scale environments, such as the El Niño-Southern Oscillation (ENSO) and Madden-Julian Oscillation (MJO). The verification results show that the GEFS has better foreast skills during the Very Strong El Nino years. The threat score (TS) in weeks 1 and 2 are 0.5 and 0.3, respectively. The NCEP GEFS have better skills if the reforecasts initialized in phase 6, and the TS in weeks 1 and 2 are 0.5 and 0.32, respectively. The probabilitic verifications show that the GEFS reforecasts have acceptable capability in providing TC activity forecasts. The Area Under Curve (AUC) of the Receiver Operating Characteristic Curve is 0.78 in week-1 and 0.69 in week-2. The verification of TC number forecast shows that the TC numbers in the GEFS reforecasts are roughly consistent with the observations. TC forecasts in Taiwan area are also investigated in this study. It is shown that the GEFS has better forecast skill during Very Strong El Nino years in which the TS can reach 0.61 in week-1. Also, the TSs in weeks 1 and 2 are 0.6 and 0.26 if the reforecasts initialized in phase 6. The probabilistic forecast verifications show that the AUC is 0.89 in week-1 and 0.75 in week-2 for the forecasts in the Taiwan area. The verification of TCs number forecast in Taiwan area indicates that the GEFS has better skill in week-1. Thus the GEFS reforecast is capable of providing forecast information on whether TCs will affect Taiwan or not, and also the strike probability in Taiwan area.