應用線性判別分析法改善衛星資料估算颱風生成潛勢指標之研究

NCUIR > College of Earth Sciences > Department of Atmospheric Sciences and Graduate Institute of Atmospheric Physics > Department of Earth Sciences > Item 987654321/64829

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/64829

Title:	應用線性判別分析法改善衛星資料估算颱風生成潛勢指標之研究
Authors:	黃建齊;Hwang,Jiann-chyi
Contributors:	大氣物理研究所
Keywords:	颱風生成;熱帶雲簇;海氣參數;線性判別分析;Tropical cyclogenesis;Tropical cloud cluster;Air-sea interactive parameter;Linear Discriminant Analysis
Date:	2014-07-22
Issue Date:	2014-10-15 14:28:45 (UTC+8)
Publisher:	國立中央大學
Abstract:	颱風的生成發展機制目前尚未能完全被瞭解，因此本研究嘗試分析熱帶雲簇能否發展成為颱風的關鍵海氣環境條件。首先收集洋面上發展雲簇與非發展雲簇個案，應用統計學上的線性判別分析（Linear Discriminant Analysis，LDA）方法，以歸納此兩類雲簇間海氣環境條件的差異性，進而做為颱風生成潛勢之預報參考。在分類工作上，LDA常用於決定未分類樣本的類別歸屬，實務上則需先建立判別函數（discriminant function）做為分類依據。因此，本研究收集西北太平洋2000年至2009年間的聯合颱風警報中心（JTWC）颱風最佳路徑、全球熱帶雲簇資料庫（Global Tropical Cloud Cluster Dataset），及NCEP Final Analysis（FNL）再分析資料，挑選七項海氣環境參數以建立完整的判別函數。此七項參數分別為：1）850hPa 相對渦度、2）850hPa－200hPa 垂直風切、3）925hPa 輻散值、4）緯度、5）850hPa 相對溼度、6）SSM/I 反演之合成熱能、7）海表面溫度。而後，即可應用LDA方法及所得之判別函數量化歸納出雲簇能發展為颱風的海氣條件閥值。接下來，再以發生於2010年至2013年間83個獨立颱風個案進行驗證，並將判別函數換算為颱風生成機率分佈圖。本研究結果顯示：若設定60% 發展機率為門檻值，本方法預報發展個案命中率達96.4%，且平均可早於JTWC發佈TS警報前53.1小時發佈颱風生成預警。 ;Up until now, the mechanisms behind tropical cyclogenesis is still not fully understood. Thus, this study seeks to find the primary air-sea environmental parameters of tropical cloud clusters (TCC) that may eventually lead to the development of a typhoon. First, this study collected data related to developing and nondeveloping TCCs, and applied the linear discriminant analysis (LDA) to analyze the air-sea parameters differences between developing and nondeveloping TCC sets. The thresholds from the differences can thus be used to establish the tropical cyclogenesis potential index (TCPI) for early typhoon formation detection. LDA is usually used to classify unknown cases to different groups according to discriminant functions. In this study, the Western Pacific JTWC best tracks, global tropical cloud cluster dataset, and NCEP FNL Analyses data are collected and used to establish a comprehensive environmental LDA discriminant function with seven air-sea parameters: 1) 850hPa relative vorticity, 2) 850hPa-200hPa vertical wind shear, 3) 925hPa divergence, 4) latitude, 5) 850hPa relative humidity, 6) SSM/I composite heat energy, and 7) sea surface temperature. Eventually, we can apply the discriminant function established by the historical cases to classify future unknown cases. Meanwhile, the air-sea environmental thresholds for cyclonegensis from the LDA method are verified by 83 independent typhoon cases that occurred during 2010 to 2013. The discriminant functions are then converted into tropical cyclogenesis probability (TCP) maps. Results show that if the threshold for tropical cyclogenesis probability is set to 60%, the average hit rate of the method is 96.4%, and the average forecast time is roughly 53.1 hours ahead of JTWC’s official tropical cyclone warning.
Appears in Collections:	[Department of Atmospheric Sciences and Graduate Institute of Atmospheric Physics ] Department of Earth Sciences

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	771	View/Open

社群 sharing

Loading...