颱風凱米生成於西太平洋，並由東側接近臺灣，在登陸前於臺灣東北部近海呈現迴轉路徑。本研究使用具20–2公里可變水平解析度之區域性跨尺度預報模式（Model for Prediction Across Scales， MPAS），探討凱米颱風接近臺灣期間的路徑演變及其相關降雨響應。控制組（CTL）實驗顯示，中央山脈（Central Mountain Range， CMR）會在臺灣東側誘發明顯的氣流通道效應，增強北風，進而促使凱米初期向南偏折，之後快速轉向北移。為找出較適合重現凱米迴轉路徑的模式設定，本研究針對不同微物理參數化方案進行敏感度實驗，並進一步利用系集模擬進行波數一（wavenumber-one， WN-1）位渦（potential vorticity， PV）收支分析，以探討迴轉期間路徑改變的動力機制。PV收支分析結果顯示，水平平流項在迴轉期間對颱風路徑改變具有主要貢獻，而垂直平流項則對後續向北轉向具有調節作用。模擬與觀測結果皆顯示，凱米迴轉期間臺灣西南部出現顯著強降雨。進一步以兩組採用不同微物理方案、各包含30個成員的系集進行系集敏感度分析（Ensemble Sensitivity Analysis， ESA），分別針對臺灣累積降雨與颱風中心駛流指標，辨識與降雨及路徑變異相關的前期環境特徵。降雨ESA結果顯示，臺灣西北側的比濕與風場呈現相互對應的敏感度分布，顯示颱風環流可能將暖濕空氣輸送至所選定的迎風面降雨區域。駛流ESA則顯示，最明顯的有利敏感度出現在臺灣西北側500 hPa緯向風分量，並向中國東南沿海延伸，顯示上游環境氣流對颱風移動具有重要調節作用。局地化多變量ESA在降雨與駛流響應中皆呈現空間上較一致的敏感度分布，其敏感度量值約為單變量ESA的三分之一，顯示局地化處理能有效抑制遠端雜訊的影響。;Typhoon Gaemi formed over the western Pacific and approached Taiwan from the east， exhibiting a looping track offshore of northeastern Taiwan before landfall. This study uses the regional Model for Prediction Across Scales (MPAS) with a variable horizontal resolution of 20–2 km to investigate the track evolution of Gaemi near Taiwan and the associated rainfall response. The CTL experiment shows that the Central Mountain Range (CMR) induces strong flow channeling east of Taiwan， enhancing northerly winds that contribute to the initial southward deflection of Gaemi before its rapid northward turn. Sensitivity experiments with different microphysics parameterization schemes were conducted to identify suitable configurations for reproducing the looping track， and ensemble simulations were further analyzed using wavenumber-one (WN-1) potential vorticity (PV) budgets. The PV budget analysis indicates that horizontal advection provides the dominant contribution to the track change during the looping period， while vertical advection modulates the subsequent northward turn. Both simulations and observations show heavy rainfall over southwestern Taiwan during this period. Ensemble Sensitivity Analysis (ESA)， based on two 30-member ensembles using different microphysics schemes， was applied to accumulated rainfall and storm-centered steering-flow metrics to identify precursor environmental features associated with rainfall and track variability. The rainfall ESA reveals corresponding sensitivity patterns in specific humidity and winds northwest of Taiwan， suggesting that the typhoon circulation transports warm and moist air toward the selected windward rainfall region. The steering-flow ESA identifies the strongest favorable sensitivity to the 500-hPa zonal wind component northwest of Taiwan， extending toward the southeastern coast of China， indicating the importance of the upstream environmental flow in modulating storm motion. A localized multivariate ESA produces spatially more coherent sensitivity patterns for both rainfall and steering-flow responses， with sensitivity magnitudes approximately one-third of those obtained from the univariate ESA， consistent with the suppression of remote noise through localization.
