| 摘要: | 颱風帶來最直接的破壞就是狂風及豪雨，而每場颱風事件的特性差異性甚大，例如：影響時間、侵襲路徑、生成位置以及所帶來不同強度的豪雨及強風，不同強風豪雨致災因子的特性使得災損情形也不相同，臺灣無法避免颱風的侵襲，因此必須探討颱風特性及所造成的災害現象，才能於颱風來臨時，做好相關防災準備，降低颱風所造成的損害。
傳統上，描述颱風強度分類的依據為風速，然而在探討颱風災害時卻多以降雨量作為致災因子之參數，使得在災害評估中可能忽略了強風的影響性，因此本研究從颱風的風、雨兩特性之角度將颱風分為「風型颱風」以及「雨型颱風」，發展了創新之颱風分類指標(Typhoon Type Index)以將颱風於不同地區之風雨特徵偏向量化，並探討颱風的風雨特性於農作物損失與坡地災害觸發之影響機制。例如：2015年蘇迪勒颱風為典型的風型颱風(全台平均TTI =2.66 )，其所造成之作物損失主要以香蕉、柚子、芭樂等果樹類作物，而造成之崩塌點位之氣象資料中風速特徵整體來說高於降雨(崩塌點位TTI =2.74 )，說明該崩塌事件可能由強風所觸發。而2008年卡玫基颱風為一雨型颱風(全台平均TTI =-0.62)，所造成之農損以香瓜、蔬菜、西瓜等作物浸泡為主，而卡玫基引致之崩塌附近亦為雨型颱風分布(崩塌點位TTI = -0.76)，說明該崩塌事件由降雨誘發之貢獻更大。此外，本研究依照不同颱風風雨特徵的空間分布特性將颱風歷史事件分類，得到TTI空間分布相關性高之颱風通常具有相似颱風路徑。
在本研究最後一部分提出了以隨機森林機器學習模型建立多光譜遙測影像之颱風前後崩塌地自動辨識系統建立方法，結果透過圖像差分法結合雲指標之濾雲程序，可有效地偵測出風災後大規模崩塌以及崩塌聚集區之影像像素，崩塌分類準確度為0.77，已可辨識出崩塌地之主要輪廓。
;The most direct hazard factor caused by typhoons is strong winds and intense rainfall. Traditionally, the basis for describing the classification of typhoon intensity is wind speed. However, when discussing typhoon disasters, rainfall is often used as the hazard factor parameter, so that the influence of strong winds may be ignored in disaster assessment. Therefore, in this study, from the perspective of the wind and rain characteristics, the typhoon is divided into "wind-type typhoon" and "rain-type typhoon." An innovative Typhoon Type Index has been developed to quantitatively describe the typhoon′s rain-wind characteristics in different regions. We further analyze the impact mechanism of crop loss and landslides disaster triggering caused by typhoons. For example, Typhoon SOUDELOR(2015) was a typical wind-type typhoon (average TTI of the whole of Taiwan = 2.66). The crop losses caused by it were mainly fruit crops such as bananas, grapefruits, and guava. Moreover, for the landslides triggered by SOUDELOR, the overall wind speed characteristics are higher than the rainfall around the site of landslides (landslide site TTI = 2.74), indicating that strong winds may mainly trigger the landslide events. Typhoon KALMEAGI(2008) was a rain-type typhoon (average TTI of the whole Taiwan = -0.62). The agricultural damage caused was mainly soaking melons, vegetables, watermelons. Furthermore, the trigger of the landslides caused by KALMEAGI was mainly influent by the rainfall dominate (landslide site TTI = -0.76). Also, this study classifies historical typhoon events according to the spatial distribution characteristics of typhoon rain-wind characteristics. The result shows that the typhoons with high TTI spatial distribution correlation usually have similar typhoon track.
In the last part of this study, the Random Forest machine learning model is built to recognize the landslide pixels in the multispectral satellite image. The classification accuracy of the landslide is 0.77, and the main outline of the landslide area can be identified. |
