熱帶氣旋快速增強之大尺度環境特徵分析與預報

Large-Scale Environment Analysis and Prediction of Tropical Cyclone Rapid Intensification Events

張有元(Yu-Yuan Chang)

指導教授 ： 蔡孝忠

淡江大學/工學院/水資源及環境工程學系碩士班/碩士(2018年)

https://doi.org/10.6846/TKU.2018.00892

摘要

在熱帶氣旋(Tropical Cyclone)的預報中，強度快速增強(Rapid intensification，RI)是最具挑戰性的預報項目之一。美國國家颶風中心(National Hurricane Center)將 RI 事件定義為 24 小時內之近中心最大風速增加至少 30 kt 以上。在快速增強階段，熱帶氣旋之強度預報誤差將顯著增加。

本研究利用 SHIPS (Statistical Hurricane Intensity Prediction Scheme)開發資料探討 RI 和非RI 事件之環境特徵。近期研究顯示，RI 事件除了與上層海洋熱力結構相關之外，亦可能與海水密度、鹽度結構有密切關聯，因此本研究亦採用全球混合坐標海洋模式(HYbrid Coordinate Ocean Model，HYCOM)之上層海洋垂直分層資料，討論海洋環境條件對於熱帶氣旋發展及強度增強的影響。最後，使用分量迴歸(Quantile Regression，QR)和羅吉斯迴歸(Logistic Regression，LR)兩種統計迴歸方法開發 RI 事件機率之預報模式，並藉由預報技術校驗之結果，進行兩種不同方法之差異與比較分析。

研究結果顯示，包含 SHIPS 大氣變數和 ETCHP(Effective Tropical Cyclone Heat Potential)的 QR 預報模式具有最佳表現，其 ROC 曲線下面積(area under the Receiver Operating
Characteristic curve)為 0.88，偵測率(Probability of Detection)為 94％，可較最佳 LR 預報模式改進 10%，且誤報率(False Alarm Ratio)降低 10％。因此，本研究認為 QR 模式可提供具有預報技術之熱帶氣旋 RI 事件預報。

關鍵字

熱帶氣旋 ； 強度快速增強 ； 分量迴歸 ； 羅吉斯迴歸

並列摘要

Rapid intensification (RI) is one of the most challenging issues for operational tropical cyclone (TC) forecasting. According to the National Hurricane Center, a RI event is defined as an increase in the maximum sustained wind speed by at least 30 kt within a 24-h period. The 24-h TC intensity forecast errors are significantly larger during the rapid intensification stage.

In this study, various datasets are utilized to investigate the RI events of the western North Pacific TCs. The SHIPS (Statistical Hurricane Intensity Prediction Scheme) Developmental Dataset is used to explore the characteristics of the RI and non-RI events. Recent studies show that the RI events are related to not only the upper ocean thermal structure but also the density and salinity structure. Thus the HYCOM (Hybrid Coordinate Ocean Model) ocean analysis is also used to study the impact of the pre-existing ocean conditions on TC development and intensification. Finally,probabilistic forecast models for the prediction of RI events are developed by using the Quantile Regression(QR) and Logistic Regression(LR) methods.

Results show that the QR model that includes the SHIPS predictors and the ETCHP (Effective TC Heat Potential) has the best performance. The area under ROC curve (AUC) of the QR model is 0.88. The probability of detection (POD) is 94%, which is 10% better than for the best LR model. In addition, the false alarm ratio (FAR) is 10% lower. Thus, the QR model can provide more skillful probability forecasts of TC RI events.

並列關鍵字

Tropical Cyclone ； Rapid Intensification ； Quantile Regression ； Logistic Regression

參考文獻

參考文獻

1. 周仲島. (2009). 氣候變遷對災害防治衝擊調適與因應策略整合研究－子計畫一:台灣地區劇烈降雨與侵台颱風變異趨勢與辨識研究(II)：行政院國家科學委員會.

2. Akaike, H., 1973. Information theory and an extension of the maximum likelihood principle. In

B. N. Petrov & F. Caski (Eds.), Proceedings of the Second International Symposium on

Information Theory (pp. 267–281). Budapest: Akademiai Kiado.