文章詳目資料

大氣科學

  • 加入收藏
  • 下載文章
篇名 蒙地卡羅法在颱風侵襲機率估計的應用
卷期 39:3
並列篇名 An Application of the Monte Carlo Method:Tropical Cyclone Strike Probabilities
作者 蔡孝忠呂國臣許乃寧賈愛玫
頁次 269-288
關鍵字 颱風蒙地卡羅法颱風侵襲機率TyphoonMonte Carlo methodTropical Cyclone Strike Probabilities
出刊日期 201109

中文摘要

中央氣象局自2007 年起提供颱風路徑預報輔助產品:颱風路徑侵襲機率。此產品是以量化的方式估計未來各地可能受到颱風中心120 公里範圍通過的機率,以提供預報之不確定性資訊。本研究參考DeMaria et al.(2009)以蒙地卡羅模擬法進行颱風侵襲機率的估計,資料採用中央氣象局2006至2010 年之官方颱風預報路徑。相較於過去使用二維常態分布函數對各個預報時段的颱風中心預報誤差進行函數擬合的方式,蒙地卡羅法具有不需對於資料分布型態進行假設,且可同時考量多種影響因子的優點。歷史路徑預報資料分析結果顯示,相鄰預報時段之預報誤差具有明顯相關性,相關性最高可達到0.91,最小值亦有0.72。在考慮誤差相關性之後,可使得模擬路徑及颱風侵襲機率的估計更加合理。此外,採用Sobol Sequence產生隨機亂數進行颱風侵襲機率估計,可得到較小之誤差平均值和誤差最大值,且當路徑預報的時間越短,誤差值與收斂速度的改進將更加明顯。研究結果顯示,採用蒙地卡羅法模擬颱風預報路徑,可有效且客觀的估計颱風路徑預報誤差機率,並可保留官方路徑預報的原始特性。未來待中央氣象局發布颱風強度及結構之官方預報資訊之後,本研究方法將可擴充進行颱風風速機率的估計。

英文摘要

Central Weather Bureau (CWB) has issued the probabilities of strike (POS) since 2007. This product estimates the probability that the center of a tropical storm (or stronger) would pass within 120 km of the specified locations to represent the track forecast uncertainty. In this study, the official track forecasts during 2006-2010 issued by CWB are used, and the Monte Carlo method (DeMaria et al. 2009) is applied to estimate the POS. Compared with the bivariate normal distribution fitting which was used in the previous research, it is not necessary to assume the type of data distributions in advance, and multiple factors could also be considered by using the Monte Carlo method. The track error analysis shows that the serial correlations are significant between two consecutive forecast time steps. The correlation values are within the range of 0.72 to 0.91. Thus the serial correlation of the error is accounted in the Monte Carlo track realization to generate reasonable track realizations. The Sobol sequence is used to generate random numbers. Comparing with the standard random number generator, the estimated POS errors are smaller and the error convergent rates are higher by using the Sobol sequence, especially when the track forecast duration is shorter. It is shown that the POS could be objectively estimated by utilizing the Monte Carlo method. The error
statistics and serial correlations of the CWB official track forecasts can also be preserved. In the future, this scheme can be extended to estimate the wind speed probabilities when the intensity and structure forecasts are issued by CWB.

相關文獻