2025 年 7 月，中度颱風丹娜絲以特殊路徑影響臺灣西南沿岸，並於臺南七股等低窪沿海地區造成海水倒灌。潮位觀測顯示，本次事件具有明顯之空間局部性與短延時特徵，其中東石潮位站出現暴潮偏差由負值快速轉為正值之「先降後昇」現象，將軍潮位站則出現短時間、高振幅之正暴潮尖峰。現行氣象署作業化 COMCOT-SS 暴潮預報設定雖可描述部分背景潮汐變化，但未能有效重現上述暴潮偏差之相位與振幅，顯示模式於近岸地形、颱風風場結構及海氣動量交換之描述仍有改善空間。 本研究以 COMCOT-SS 暴潮模式為基礎，採分階段敏感度與交叉敏感度分析架構，依序檢驗地形解析度、參數化理想風場、最大風速及海表面拖曳係數對模擬結果之影響。首先，將臺灣西南沿岸巢狀網格解析度由公里級提高至約 200 m，以改善潮間帶、淺灘與局部岸線之地形表示；其次，建立 Two-Segment 參數化風場，分別描述內圈風速增長與外圈風速衰減，並進一步進行最大風速與拖曳係數敏感度試驗，其中拖曳係數係由 Charnock 關係、對數風速律及粗糙長度上限動態計算。 結果顯示，提高地形解析度可改善近岸水深與測站格點之描述，但不足以單獨重現東石站之強烈負暴潮偏差與將軍站之正暴潮尖峰。Two-Segment 風場可使暴潮偏差歷線較為集中，並改善東石站負轉正之相位及將軍站整體模擬誤差；提高最大風速則可增強暴潮振幅，但仍無法單獨完整重現觀測極值。進一步調整拖曳係數後，海氣動量交換增強，使東石與將軍站之水位歷線及暴潮偏差獲得較一致之改善。研究結果顯示，丹娜絲颱風暴潮模擬誤差並非由單一因素造成，而是受到近岸地形、風場結構、颱風強度及海氣動量交換共同影響。本研究建立之控制變因分析流程與近岸拖曳係數敏感度架構，可作為臺灣西南沿岸暴潮預報系統在地化校正之事件型科學參考；惟相關參數仍需透過更多颱風案例及獨立觀測資料進一步驗證。;In July 2025， Typhoon Danas followed an atypical track toward southwestern Taiwan and caused coastal inundation in low-lying areas， including Qigu， Tainan. Tide-gauge observations revealed that the event exhibited highly localized and short-duration storm-surge characteristics. At the Dongshi tide gauge， the surge residual recorded a rapid transition from negative to positive， whereas the Jiangjun tide gauge recorded a brief， high-amplitude positive surge peak. Although the operational COMCOT-SS storm-surge forecasting configuration used by the Central Weather Administration could reproduce part of the background tidal variation， it failed to adequately capture the phase and amplitude of these surge residuals. This indicates that improvements are still needed in the representation of nearshore topography， typhoon wind-field structure， and air–sea momentum exchange. Based on the COMCOT-SS storm-surge model， this study employed a controlled sensitivity-analysis framework to sequentially examine the effects of topographic resolution， parametric wind-field structure， maximum wind speed， and sea-surface drag coefficient on the simulated storm surge. First， the resolution of the nested grids along the southwestern coast of Taiwan was increased from the kilometer scale to approximately 200 m to improve the representation of intertidal zones， shallow banks， and local shorelines. Second， a Two-Segment parametric wind-field model was developed to separately describe the increase in inner-core wind speed within the radius of maximum wind and the decay of the outer wind field beyond it. Sensitivity experiments were then conducted for the maximum wind speed and drag coefficient. The drag coefficient was dynamically calculated using the Charnock relation， the logarithmic wind profile， and an upper limit on aerodynamic roughness length. The results show that increasing topographic resolution improved the representation of nearshore bathymetry and tide-gauge grid locations， but was insufficient to independently reproduce the strong negative surge residual at Dongshi and the positive surge peak at Jiangjun. The Two-Segment wind field produced a more temporally concentrated surge response and improved the transition from negative to positive residuals at Dongshi， as well as the overall simulation error at Jiangjun. Increasing the maximum wind speed enhanced the surge amplitude， but could not independently reproduce the observed extreme values. Further adjustment of the drag coefficient strengthened air–sea momentum transfer and resulted in more consistent improvements in the simulated water-level and storm-surge residual time series at both stations. These findings indicate that the simulation errors during Typhoon Danas were not caused by a single factor， but were jointly influenced by nearshore topography， wind-field structure， typhoon intensity， and air–sea momentum exchange. The controlled sensitivity-analysis framework and nearshore drag-coefficient sensitivity approach developed in this study provide an event-based scientific reference for the localized calibration of storm-surge forecasting systems along the southwestern coast of Taiwan. However， the applicability of the proposed parameters requires further verification using additional typhoon cases and independent observational datasets.
