本研究針對在台灣冬季實行的觀測實驗(Taiwan atmospheric PBL Observation, Modeling, and Data Assimilation)或稱T-POMDA，獲得的高垂直解析度行星邊界層觀測資料進行資料同化實驗並分析其對於邊界層分析場、預報及空氣品質模擬的影響作系列的驗證。同化觀測包含微型探空(Aerosond)、無人機(UAV)和剖風儀。本研究提出在WRF局地化系集卡爾曼濾波系統(WRF-LETKF)下，邊界層高解析度觀測系集資料同化的策略。包含變數相關性的局地化(variable localization)和隨流場變化的觀測誤差斜方差矩陣估計法(Adaptive Observation Error Inflation, AOEI)。 由於物理參數化的限制，近地面溫度與風場有不同於真實世界的相關性。跨變數更新會降低同化變數預報的能力，因為錯誤的跨變數增量(unobserved variable increment)造成相反於同化變數增量(observed variable increment)的物理作用，並且在後續的預報中限制同化變數增量的效益。結果顯示進行跨變數相關性的局地化有助於維持同化變數的效果，並且增進同化變數的預報能力。此外，無人機的溫度觀測採用AOEI的方法估計其觀測誤差斜方差矩陣，以克服在複雜地形的邊界層下，溫度增量(innovation)容易較大的問題。 AOEI 法藉由ensemble spread擴張觀測誤差，結果顯示壓制分析場中過大的增量(increment)有助於減緩同化所造成的非物理性流場，並且增進熱力主導的海陸風模擬。 藉WRF-LETKF系統所估計出的氣象分析場將提供予WRF-CMAQ空氣品質模擬 iii 的氣象參考場使用。同化邊界層中變化快速的風場調整山坡上的垂直環流結構，使得污染物沿地形的垂直傳輸獲得改善。日間混和層內上升氣流的減緩使臺中市及山坡上有較多的污染物累積，中央山脈山壁上的上升氣流則受到加強，沿地形傳遞山腳混和層內的污染物自山頂附近，形成雙層分布的結構。夜間山風的加強使得山腳與平地有較高的穩定度，易於累積污染物。此外，日間傳輸至高層的污染物藉由此下沉氣流再次回到地表加劇夜間高污染事件。同化邊界層溫度則調整熱力主導的海陸風與山谷風結構，調整地面夜間過冷的情況，減弱過強的陸風使污染物較容易滯留於陸地上。並且臺中盆地與大肚臺地間的溫度梯度也受到加強，增加污染物向盆地內傳入的通量。;This study investigates the impact of assimilating the high vertical-resolution observations on the Planetary Boundary Layer (PBL) analysis, forecast, and air quality prediction during the Taiwan Air Pollution Modeling and Data assimilation (T-POMDA) experiment. The observations include wind profilers, and unmanned aerial vehicles (UAV), and Aerosond collected from 16 to 17 March 2021. The assimilation is conducted with the Weather Research and Forecasting model-Local Ensemble Transform Kalman Filter (WRF-LETKF) with rapid update cycles. Assimilation strategies with variable localization and adaptive observation error variance inflation (AOEI) are adopted to highlight the impact of assimilating the PBL observations with a high vertical resolution. Assimilating the wind observations from wind profilers and Aerosond can improve the PBL wind structure over central Taiwan. However, the near-surface temperature analysis degraded. The cross-variable error correlation is opposed to the nature which will prevent the observed variable to correct the model field. Applying variable error covariance localization becomes crucial that maintains the impact of wind assimilation and improves the forecast. In addition, the impact of the UAV temperature observation is optimized by applying AOEI to improve the estimation of the representative error because the large innovation occurred easily in PBL due to the high variation of temperature and cold bias in the simulation. The result shows that the suppression of large increment can prevent the unrealistic flow in the analysis. Then it improves the thermal-direct land-sea breeze v simulation. The WRF-LETKF analysis is used to drive the Community Multiscale Air Quality Model (CMAQ) to obtain the air quality simulation. The assimilation of wind observation with high vertical resolution corrects the structure of vertical circulation near the terrain which better represents the vertical transportation of the pollutants. During the daytime, the suppression of upslope wind in mixing layer causes more pollutants accumulated in Taichung and hillside. At the mountain wall of Central Mountain Range (CMR), upslope wind is intensified and transports the pollutants in mixing layer to the top of mountain. Finally, the polluted parcels separate to two parts. During the nighttime, the intensification of mountain breeze leads to more stable environment at hillside which accumulates pollutants easier. Furthermore, the pollutants at upper level transport downward and intensify the pollution event in the plain. The assimilation of temperature observation corrects the thermal-direct land-sea breeze and mountain-valley breeze. The correction of overcooling at night suppresses the land breeze and prevents the pollutants to diffuse offshore. Stronger temperature gradient between Taichung Basin and Dadu Tableland increases the transportation of pollutants from the coastline which intensifies the PM2.5 concentration in the basin.
