| 摘要: | 致災性強對流與冰雹事件好發於台灣夏季，如何針對此類劇烈天氣事件進行更快速、準確的預警，為當前氣象局亟欲提升的能力。本研究利用中央氣象局預報中心所使用的對流監測平台(System for Convection Analysis and Nowcasting, SCAN)歷史事件，進行2011~2018年台灣夏季(5~8月)於不同天氣系統(綜觀與弱綜觀)及地理環境(北部及南部)下的對流胞統計特徵分析，來建構對流胞即時預報產品，使得災害潛勢區域能有定量上的依據。此外，收集2011~2020年間的夏季冰雹事件的目擊資訊，以檢驗SCAN系統在台灣地區對冰雹的預警能力。
特徵統計結果顯示，台灣夏季對流胞的移動速度分布以2~10 m/s居多、最大反射率集中在45~55dBZ、生命期則以1小時以內為主。此外，分布在海上的對流胞移動速度相較陸地上的對流胞移動速度為快，因此由綜觀天氣產生之對流胞移動速度會相較於內陸弱綜觀天氣生成之對流胞快。比較一小時預報的路徑誤差，南部地區的誤差比北部大、綜觀天氣系統下的誤差比弱綜觀大。而經緯向誤差大致相同，統計結果主要受到綜觀天氣系統對流所主導。本研究進一步結合過去開發的對流胞路徑潛勢預報(Potential Track Area for Storm, PTAS)與颱風侵襲機率預報(Wind Speed Probability, WSP)技術，利用統計誤差特性建構對流胞侵襲機率預報(Storm Threat Optimal Probability, STOP)產品。校驗結果顯示，該產品於易受強對流侵襲地區的整體預報誤差在10%以內。此外，區分天氣系統所建構出之即時預報在一般對流胞速度下(10 m/s 以下)並無太大差異。而針對冰雹預警能力的方面，當前SCAN系統對於冰雹事件的預警時間(lead time)以15分鐘以內居多，且有二大預警問題：(1)部分目擊事件未被預警以及(2)假警報過多。本研究分別提出(1)以冰雹目擊個案特徵統計結果下修預警門檻以及(2)以台灣探空資料修正SCAN 系統背景溫度高度場等改善建議。最終，以個案證實對流胞即時預報於冰雹預警上的適用性。;System for Convection Analysis and Nowcasting (SCAN) is one of the operational forecast systems used by the Central Weather Bureau (CWB), which can provide storm cells and hail information. This study presents an 8-year analysis of summer storm cells in Taiwan (from 2011 to 2018, May-Aug) based on the SCAN, discusses the characteristics of cells under different weather systems (synoptic or weak synoptic), and geographical environments (north or south of Taiwan). The statistical results of storm track errors are used to establish the storm cell nowcasting, in order to define 0-1-hour severe storm warning area and quantitatively improve the warning capability of severe weather. In addition, this study also collected summer hail events from 2011 to 2020 in Taiwan to check the hail warning capability of the SCAN.
The results show that the cell speed in Taiwan is mostly 2~10 m/s, the maximum reflectivity is concentrated at 45~55dBZ, and the lifetime is mainly within 1 hour. The cells that come from the ocean (synoptic) will have a higher moving speed than those generated in the inland (weak synoptic). Comparing the one-hour forecast error, the error in the southern region is higher than the error in the northern region, the error in the synoptic day is higher than which in the weak synoptic day, and the latitude and longitude errors are roughly the same. Furthermore, the statistical results are dominated by the synoptic weather. Construct the Storm Threat Optimal Probability (STOP) forecast using the statistical data. The verification shows that the forecast error of the product is within 10%. Besides, the forecast results in the different weather systems are nonsignificant differences in most cases (below 10 m/s). In terms of hail warning, the lead time of hail that SCAN can provide is around 15 minutes, and there have two major warning problems: (1) some hail cases were not warning in the SCAN system, and (2) there are too many false alarms. This research proposes (1) lowering the warning threshold based on the statistical results of the hail report and (2) correcting the default value of the SCAN system by using the sounding data in Taiwan to solve the problems. Finally, confirm that the probability forecast can be used for hail warning. |
