雷達回波外延法使用過去時間的觀測回波資料，計算出可提供天氣系統移動及旋轉特性資訊的回波移動場，並根據此移動場結果來進行外延預報。根據前人研究，移動場計算上的不確定性是外延預報當中最主要的誤差來源之一，且天氣系統在不同高度上亦有不同的移動方向，這些因素都將影響外延預報的能力。 本研究針對此一不確定性來源，將三維的回波移動場應用至MAPLE (McGill Algorithm for Precipitation nowcasting using Lagrangian Extrapolation) 即時預報系統中，並挑選臺灣秋季及梅雨鋒面降水事件來進行分析討論。根據三維移動場的時間與空間分析結果，在秋季降水事件中，東西方向(u)分量和南北方向(v)分量移速大小相近；在梅雨鋒面事件中，東西方向之移速則大於南北方向之移速，這樣的結果顯示出不同季節的降水系統移動特性。三維回波移動場資訊也被應用至即時預報方法中，將其預報能力以連續校驗法以及絕對校驗法進行分析，結果顯示，三維雷達外延方法之預報能力有所提升，可改善至三小時的預報結果。 此外，本研究亦根據三維移動場的時空分析結果，建立一系集即時預報系統，考慮在移動場計算過程所包含之不確定性。校驗結果顯示，應用此系集即時預報系統至秋季降水及梅雨鋒面個案時，皆具有正確判斷降水事件發生之區辨能力，且可以維持約三小時的預報可信度表現。並且於累積雨量預報上也能正確的掌握和觀測結果相似的降雨分布。整體而言，三維雷達外延即時預報方法以及系集即時預報方法皆可以針對移動場上的不確定性進行改善，也較原先的外延方法有更好的預報能力。 ;Radar echo extrapolation utilizes the observed composite reflectivity to estimate the motion fields of radar echoes and provide advection and rotation information for extrapolation. The uncertainty of motion fields is one of the major error in radar extrapolation. In addition, the weather system may have different moving directions at different heights. In this study, 3-Dimension motion fields are estimated by the entire volume scanned data and applied to MAPLE (McGill Algorithm for Precipitation nowcasting using Lagrangian Extrapolation) nowcasting system. With autumn precipitation events in the Yilan area and the Meiyu front events in Taiwan, the characteristic of 3D motion fields in space and time are analyzed. The results show that u-component varies more than v-component both in time and space for the Mei-yu front event. As for autumn precipitation, the diversity of u- and v- components are similar. Then, the added value of 3D motion fields for the nowcasting is evaluated by continuous and categorical verification. It is found that the improvement of the nowcasting with 3D motion fields can be up to 3-h. Furthermore, based on the analysis of 3D motion field, an ensemble nowcasting scheme was developed by considering the uncertainty of motion field. The verification of ensemble nowcasting shows a good ability to correctly predict the occurrence of the precipitation and the reliability can up to nearly 3-h in autumn precipitation and Meiyu front events. The forecast of accumulated rainfall also accurately captures rainfall distribution similar to the observation. Overall, both the utilization of 3D motion fields in nowcasting and the ensemble nowcasting scheme can improve the uncertainty caused by motion field estimation and demonstrate better forecasting ability compared to the original extrapolation method.
