輻射傳輸模式能夠將大氣變數（例如大氣熱力狀態等）轉變為輻射參數。本研究使用CRTM (Community Radiative Transfer Model)來連結由數值模式所輸出的變數與衛星觀測做驗證。研究內容著重於利用多頻道的地球同步衛星影像-日本向日葵八號衛星來客觀檢驗區域模式-WRF (Weather Research and Forecasting)在使用不同微物理參數化法的模擬表現。四種微物理參數化法被選擇為本研究的參考方案，包括Goddard (GCE)、WRF single-moment 6 class (WSM6)、WRF double-moment 6 class (WDM6)及Morrison (MOR)。我們選擇了2017年6月1日造成臺灣強降雨事件的梅雨天氣型態為本研究的模擬個案來校驗數值模式的表現。 研究結果顯示數值模式在9公里水平解析度的東亞地區大範圍綜觀天氣模擬中，於四種微物理方案皆產生較衛星觀測更多的雲網格覆蓋量，其中又以使用MOR微物理方案的模擬結果產生了較為可觀的高雲量。而使用GCE方案在晴空網格點的水氣頻道模擬結果顯示其具有較佳的水氣收支設定。在敏感性測試上顯示MOR的模擬結果在高度14公里以上的較大量冰雲，可能與模擬產生過多高雲有關。評估3公里水平解析度之高解析數值模擬在臺灣附近的表現則發現，各模擬方案皆能有效掌握有雲格點的分布情形。在點對點的分析中，高雲呈現出三種雲種中最好的分布表現。同時，模擬結果也顯示模式需要較長的模擬時間來掌握低雲的發展情形。 ;Radiative transfer model could be used to convert the geophysical variables (e.g., atmospheric thermodynamic state) to the radiation field. In this study, the Community Radiative Transfer Model (CRTM) is used to connect numerical weather prediction model outputs and satellite observations. The performance of the regional Weather Research and Forecasting (WRF) model with different microphysics schemes was investigated objectively using multichannel observed satellite radiances at different wavelengths from a Japanese geostationary satellite Himawari-8. Among the schemes, 4 microphysics schemes in WRF (i.e., Goddard [GCE], WRF single-moment 6 class [WSM], WRF double-moment 6 class [WDM], and Morrison schemes [MOR]) were configured in the WRF model to simulate a heavy rainfall event caused by the Mei-Yu front on the June 1, 2017, in the vicinity of Taiwan. The results over the East Asia domain (9 km) illustrate that all four microphysics schemes produced more cloud cover than observed in the satellite data. A considerable amount of high cloud was displayed in the MOR scheme simulation. However, the simulation with the GCE scheme displayed an improved water vapor budget in clear scenes. Sensitivity tests reveal that the excess condensation of ice at ≥14 km might associate with the high cloud cover and higher cloud-top altitudes. When focusing on Taiwan using a higher (3-km) model resolution, each scheme displayed a decent performance on cloudy pixels. In the grid-by-grid analysis, the distribution of high clouds was the most accurate among the three cloud types. The results also suggested that all schemes required a longer simulation time to describe the low cloud property.
