台灣地區之颱風路徑軌跡分群與災情統計檢定

西北太平洋地區乃全球颱風發生頻率最高的區塊，颱風挾著狂風驟雨侵襲，每每對台灣造成重創，因此，提升對颱風動向的瞭解與掌握對於台灣而言乃是一大要務。本研究將針對西元1964至2015年間台灣警報範圍內之颱風軌跡進行長時段的分析與探索，而對颱風之分析時間精度則控制在5天內的每6小時之經緯度位置。本研究除了從一些知名的颱風資料庫下載並整理歷年的颱風訊息外，亦從不同網站或報紙等的資料源補充各颱風災損訊息，這個特點使得本研究的成果將有別於過往大都僅分析颱風的典型路徑以及基本天候特徵之研究。透過k-means演算法萃取出典型的颱風路徑之後，本研究藉由Kruskal-Wallis檢定解析典型颱風路徑間造成的財損、農損、傷亡等特徵之間是否存在顯著差異性，藉此辨識出容易造成重大災害的颱風路徑。未來台灣在面對颱風侵襲時，中央氣象局可根據估計出的颱風路徑，推算出可能造成的災損數據，進而對其加強防護措施，一定程度地強化防災的意識與警戒心。

關鍵字

颱風；路徑軌跡分群；災情統計檢定

並列摘要

The Northwest Pacific region is one of the areas with the highest frequency of typhoons in the world. Typhoons are invaded by squally showers and often cause severe damage to Taiwan. Therefore, improving the understanding of typhoon trends is a major task for Taiwan. This research conducted a long-term analysis to explore the typhoon trajectories within the warning range of Taiwan from 1964 to 2015. The resolution of the analysis time of the typhoon will be controlled at the latitude and longitude position every 6 hours in 5 days. In addition to downloading and collating typhoon information from some well-known typhoon databases, this research also supplements information on typhoon damages from different websites or newspapers. This feature makes the results of this research different from those in the past that only analyze the typical trajectories and basic weather characteristics of typhoons. After extracting the typical typhoon trajectories through the k-means algorithm, this study uses the Kruskal-Wallis test to analyze whether there are significant differences in the characteristics of damages such as money loss, agricultural loss, and casualties caused by those typhoon trajectories. In this way, we can identify the typhoon trajectories that prone to cause major disasters. When Taiwan faces a typhoon in the future, the Central Weather Bureau can evaluate the possible loss based on the estimated typhoon path, and then strengthen the vigilance of disaster prevention to a certain extent.