本研究使用加拿大麥基爾大學(McGill University)所發展之雷達回波外延預報系統 (McGill Algorithm for Precipitation nowcasting using Lagrangian Extrapolation, MAPLE)，利用 中央氣象局(Central Weather Bureau, CWB)雷達回波整合資料，檢視該系統在臺灣地區即 時天氣預報(0~6小時)的表現。MAPLE系統包含兩個程序，依序為：1.雷達回波變分追 蹤法(Variational Echo Tracking, VET)決定降水系統的移動場；2.由半拉格朗日後推平流法 (semi-Lagrangian backward advection)決定降水系統的預報場。本研究選取颱風與梅雨鋒面 兩種天氣型態共三個個案進行分析，並評估MAPLE的可行性。 研究首先針對VET過程所使用的兩種參數一移動向量密度及回波資料時間間隔 進行敏感度測試。在獲得臺灣地區最佳化設定後，定性討論雷達回波外延在預報平移與 旋轉天氣系統之特性。本篇分別選取兩個颱風及梅雨鋒面共三個個案進行測試，由測試 結果顯示，MAPLE回波預報結果與觀測回波之間的相似程度在分數表現上相當理想。 進一步檢視移動向量，但當天氣系統接近臺灣陸地時，移動向量速度減緩，顯示外延法 能夠反應出地形阻擋之影響。而MAPLE系統定量降水表現方面，綜合不同預報得分指 標，此系統對於颱風和梅雨鋒面的降水預報能力可維持在2~3小時。與持續法相比較， 其降水預報誤差隨時間明顯改進。然而，當預報時間(lead time)超過3小時後，外延預報 能力迅速下降，主要與外延回波缺乏掌握天氣系統生成與消散的機制有關。整體而言， 透過高時空間解析度之雷達觀測網的降水觀測，進行外延預報，能迅速提供即時定量降 水預報資訊。此技術可在極短時間內提供0-3小時之預報，彌補複雜天氣數值模式初期 起轉過程在降水資訊之不足。

The nowcasting system, McGill Algorithm for Precipitation nowcasting using Lagrangian Extrapolation (MAPLE), is used to examine very short-term forecast of precipitation over Taiwan area in this study. There are two components in the MAPLE system: First, the motion field of precipitation is determined by Variational Echo Tracking (VET) technique. Second, a radar map is advected by means of a semi-Lagrangian backward advection scheme to be a precipitation forecast. By using composite radar data from the Central Weather Bureau (CWB), the cases from typhoon and Mei-Yu front are investigated in this study. At first, sensitivity tests determine the best setting of VET parameters for nowcast over Taiwan area. Then the features of Quantitative Precipitation Nowcasting after running the MAPLE are investigated. The similarity of observed and predicted reflectivity can last for couple hours. While examining the motion vector of VET, it shows that the speed decreases if the weather system approaches to Taiwan. This indicates that the extrapolation skill can reflect the terrain effect. By examining of different forecast skill, the MAPLE system is able to perform well around 2 - 3 hour forecast lead time in general. Compared with the persistence method, the error of precipitation forecasting has obviously improved. However, after 3 hour, the scores of forecast skill decrease significantly due to lack of growth and decay mechanism from movement of radar echoes. Over all, the nowcasting system can provide 0-3 hours precipitation information when the complex numerical model is under ‘spin-up’ process.