本研究藉由分析2021年宜蘭平原冬季降水事件中系集模擬成員之表現，進一步了解導致宜蘭平原地區強降水的關鍵因素。此外，運用多都卜勒雷達風場合成（WISSDOM）及多種觀測資料（包括無人機、剖風儀、微型探空等），並採用K-means分群方法將各個系集成員依照宜蘭地區降水極值位置進行空間上的分類，以分析不同群集之間的風場結構、降水分布及其不確定性。 研究結果顯示，降水預報表現最佳的群集並不一定能準確地捕捉降雨事件期間的動力結構，而最能重現觀測中降水演變過程的群集，並不一定能夠產生最佳的降水量。此外，從分群結果中，第二群能夠較準確地捕捉到降水先增後減的變化趨勢，尤其是在降水事件的後期，當東北風進入平原後產生地形下沉作用，進而引發西風出現並抑制降水。相較之下，第三群與第四群雖然產生的總雨量較接近觀測，但其空間分布有所差異，與底層風場強度與低層環流結構不同有關；而第一群則因為環境偏冷且水氣量少，因此降水最弱。本研究指出，即便降水總量相似，其背後的動力結構可能截然不同。因此，在探討像宜蘭這樣複雜地形區的降水機制時，除了累積降水量外，亦應評估低層水氣與風場、環流結構的交互作用，才能更全面地掌握降水的發展過程。 ;This study analyzes performance of ensemble members in the 2021 winter severe precipitation event over Yilan using the Wind Synthesis System using Doppler Measurements (WISSDOM) and various observational data (Unmanned Aerial Vehicle, Wind Profiler, Storm Tracker, etc.). K-means clustering spatially classifies ensemble members based on the location of precipitation extreme value to analyze wind field structure, precipitation distribution, and uncertainties among clusters. Results show that clusters with the best precipitation performance do not necessarily capture dynamic structures accurately, while those reproducing observed precipitation evolution may not produce optimal amounts. Cluster 2 best captured the precipitation increase-decrease pattern, particularly when northeasterly winds generated orographic subsidence and westerly return flow that suppressed precipitation. Clusters 3 and 4 produced rainfall amounts closer to observations but with different spatial distributions due to varying low-level wind intensity and circulation structures. Cluster 1 showed weakest precipitation under colder, drier conditions. The study demonstrates that similar precipitation totals can result from fundamentally different dynamic structures. Therefore, investigating precipitation mechanisms in complex terrain requires analyzing low-level wind fields and local circulation structures beyond accumulated precipitation amounts to comprehensively understand precipitation development processes.
