臺灣颱風暴潮之季度預測與評估

Abstract

臺灣位處西北太平洋颱風侵襲之敏感區域，每逢夏、秋兩季颱風侵襲時常造成沿海區域暴潮侵襲機率增加，此時易導致下游低窪地區排水不易等，若不幸再碰上大潮或強降雨事件可能發生堤防溢淹情況，其颱風暴潮形成之災害勢必更加嚴重。有鑑於颱風暴潮事件對於沿海區域威脅之嚴重性，且傳統暴潮模式推估之時間期距多以天、週等短期距作為單位，使得預警發出後準備時間緊迫，難以提早進行相關防範措施，因此本研究擬以中長期觀點出發，建立一套能提供海岸區域評估颱風暴潮風險的模式。 本研究根據相關文獻與理論挑選出三項因子作為暴潮風險評估之依據，分別為聖嬰與反聖嬰現象對臺灣颱風季節分佈的影響、颱風路徑季度預報、颱風累積動能。而本研究根據此三項因子在臺灣所顯現的統計特性定義出各因子之風險評估，並依照不同季節海岸區域對颱風敏感程度不一的特性，將此三個因子的風險評估結果分成低、中、高三種不同風險層級。並使用水文頻率分析方法計算暴潮觀測資料之風險層級進行模式驗證，最後利用列聯表計算各項校驗指標評析本模式預報結果，其預報之命中率皆有不錯之預報成效。 本研究亦將此模式實際應用於北區淡水河流域海岸與南區高屏溪流域外海兩區域，其結果中發現，若測站驗證年數夠多、潮位資料夠完整，兩區域之準確度可達0.6以上。將兩測站分區討論，北區因颱風路徑風險定義符合的個數較少，模式容易發生低估之情形；而南部則因地形因素影響颱風強度發展，導致模式容易有高估之過度預警狀況出現。總體而言，本模式在進行季度預報已有初步之成效。

Located in Northwestern Pacific ocean area, Taiwan is often stricken by typhoon. Whenever summer or fall typhoon strikes, typhoon surge in coastal area rises, in this time, the embankment is possibly suffering from overflooding. If the intensive precipitation also takes a part in at the same time, the loss of disaster is countless. Moreover, traditional typhoon surge prediction model often provides us with daily or weekly forecast, this may result in no time and precaution measures falling short. In order to prevent such condition from happening, this research is trying to create a surge prediction model in a long-term perspective and combine Central Weather Bureau numerical products, hoping to connect both sides of meteorology and hydrology. In the surge prediction model of this research selects 3 factors to assess the risk of coastal typhoon surge based on related papers, which are :the effects of El Nino and La Nina incidents, seasonal typoon path forecast, and accumulated cyclone energy. According to each factor’s stastistical properties, we define its risk score respectively and turn it into 3 different risk levels, which are Low, Medium, and High. When proceeding this model’s varifications, we use frequency analysis method to calculate the historical surge events’ risk levels, and we also use contingency table’s several score indices to compare them with the forecast results. Applying this model to northern and southern areas, we found it useful and the accuracy of both areas can reach to 0.6 if the data records permits. Discussing the results of each area, model is often underestimated in the northern area due to the less number of typhoon path risk definition, and often overestimated in the southern area because the inensity of typhoon is usually restricted to its landforms. Generally speaking, this model has a fairly outcome in seasonal forecast.