應用WRF模式於颱風期間離岸風場風速模擬與參數敏感性分析

Abstract

本研究針對臺灣離岸風電在颱風情境下所面臨之風場模擬挑戰，採用 WRF（Weather Research and Forecasting）模式，評估不同海氣象初始場來源、起始模擬時間與物理參數化方案對模擬結果之敏感性與準確度。研究選取 2017 年尼莎及 2023 年杜蘇芮與海葵三個颱風為個案，並將模擬結果與竹南、彰化觀測塔及龍鳳氣象站資料比對，從颱風路徑與風場特性進行分析與驗證。結果顯示，WRF(ERA5) 資料因具較高解析度，能較準確重現颱風結構與風場強度，惟常高估風速，且峰值時間多偏提前； WRF(FNL) 模擬風速趨勢較保守，更容易出現較高的RMSE與MAE，如海葵颱風案例中風速 RMSE 高達 12.27 m/s， KGE 值為 -0.1，顯示其在特定情境下仍存有顯著誤差。起始時間敏感性分析指出，模擬啟動時間接近颱風登陸前 6 至 12 小時，有助於精準掌握風場結構與演變；若啟動時間過短，可能因邊界場資訊尚未完整建立，導致模擬穩定性與準確性下降。參數組合方面，分成PC1以及PC2，其中PC1選擇WSM5(WRF Single-Moment 5-Class)微物理方案搭配GD(Grell–Dévényi)積雲參數方案；PC2使用Thompson微物理方案與KF(Kain–Fritsch)積雲參數方案。PC1 組合整體較保守；PC2 在颱風結構較為準確，惟易出現風速高估，杜蘇芮案例峰值高估15 %。此外，本研究也使用龍鳳氣象站所觀測之風速資料進行模擬結果比對。由於WRF較無法還原都市複雜地形，導致再使用陸上型觀測站時，風速有高估的現象。此類型測站具輔助驗證價值，但其準確度不如海上型觀測塔。整體而言，WRF 模式具備合理再現颱風風場之能力，但模擬精度高度依賴海氣象資料、參數設定與起始時間選擇。本研究成果可作為後續進行離岸風場極端載重評估與風險管理之參考依據。

This study uses the Weather Research and Forecasting (WRF) model to address the challenges of simulating typhoon wind fields for offshore wind power development in Taiwan. The study evaluates the sensitivity and accuracy of the simulated wind fields under various meteorological input datasets, simulation start times, and physical parameterization schemes. The study selects three typhoons—Nesat (2017), Doksuri (2023), and Haikui (2023)—as case studies. The simulated results were compared against observational data from the Zhunan and Changhua offshore meteorological towers, as well as the Longfeng weather station, focusing on typhoon tracks and wind field characteristics for analysis and validation. The results indicate that WRF initialized with ERA5 data provides a more accurate representation of typhoon structures and wind field intensity due to its higher spatial resolution. However, it tends to overestimate wind speeds and predict peak wind events earlier than observed. Conversely, simulations using FNL data show more conservative wind speed trends but frequently yield higher RMSE and MAE values. For instance, the wind speed RMSE for Typhoon Haikui reached 12.27 m/s with a KGE value of -0.1, thereby demonstrating significant errors under specific scenarios. Sensitivity tests of the simulation's initialization time demonstrated that initiating runs 6–12 hours prior to landfall enhanced accuracy, while overly brief lead times resulted in incomplete boundary conditions and diminished model stability. Two parameterization combinations were assessed: PC1 utilized the WRF Single-Moment 5-Class (WSM5) microphysics scheme with the Grell–Dévényi (GD) cumulus parameterization, while PC2 employed the Thompson microphysics scheme and Kain–Fritsch (KF) cumulus scheme. The PC1 ensemble demonstrates a more conservative overall tendency, while the PC2 ensemble provides a more accurate representation of the typhoon's structure but tends to overestimate wind speeds. In the case of Typhoon Doksuri, the peak was estimated to be 15% higher than the actual measurement. Additionally, wind speed data from the Longfeng weather station were used for further validation. Due to WRF's limitations in resolving complex urban terrain, simulations tend to overestimate wind speeds when validated against land-based stations. While such stations offer supplementary verification, their accuracy is generally inferior to that of offshore towers. Overall, the WRF model demonstrates the capability to reasonably reproduce typhoon wind field characteristics; however, its accuracy is highly dependent on initial conditions, parameter settings, and simulation timing. The findings of this research can serve as a reference for future offshore wind load estimation and risk assessments under extreme weather conditions.