| 摘要: | 本研究旨在探討GNSS-RO折射率觀測中，因槽化所造成的低層資料負偏差。由於無法於觀測資料中直接得知槽化的發生與否，本研究採用全球模式預報資料作為判定的依據。經過比對，當預報中出現槽化時，對應的觀測資料在槽化層以下也出現顯著的負偏差。
檢視模式預報出現槽化的熱力剖面，槽化的發生主因有三：邊界層的混合逆溫、降雨系統冷池造成近地表劇烈降溫以及其他系統造成的沉降逆溫，垂直大幅度的溫、濕度梯度為形成槽化的主要因素。
為了檢視同化這些含負偏差的GNSS-RO資料對於區域模式預報的影響，吾人選取2020/05/22的鋒面大雨個案作為研究標的，進行四組實驗：1. 未同化GNSS-RO資料的CTRL、2. 同化所有GNSS-RO資料的REF_ALL、3. 使用LSW>35%作為資料檢定（QC）策略的REF_LSW、4. 移除模式槽化層以下資料的REF_NODUCT。經過三天半的資料同化循環後，比較各實驗分析場、預報場。並以其探討降雨預報差異的原因。
在分析中，可以歸納出同化GNSS-RO資料後可以顯著調整模式場中南海至中國華南區域的動、熱力環境為適合鋒面強度增加之環境。使用QC策略的兩組實驗結果顯示可進一步增加水氣量，但動力場的表現依QC策略而有所不同：剔除較多資料的REF_LSW低層輻合較CTRL減弱，導致鋒面強度下降；保留較多資料的REF_NODUCT則營造出最強的低層輻合與鋒面強度，顯示移除低層負偏差及盡可能保留低層資料的價值。
預報上，REF_NODUCT出現最強降雨，REF_ALL與REF_LSW接近，而CTRL最弱。但是技術得分則顯示REF_ALL表現最佳、CTRL次之，REF_NODUCT及REF_LSW的表現不彰，主要是由於採用QC策略的兩組實驗整體鋒面的雨帶南偏，這可能是因為移除低層資料所造成中國華南區域西南渦東移，使鋒面移速較原先更快。
增量分析實驗指出同化GNSS-RO低層資料會導致局地的水氣明顯減少，且會因為模式平流而移動，但對於水氣傳輸的影響仍要視其與動力場的交互作用而定。
本研究在槽化判定使用模式，可能因模式誤差而錯誤地刪除資料；且移除資料可能導致西南渦東移而使降雨出現位移，未來結合由GNSS-RO觀測資料直接判定槽化以及重建槽化層以下資料的技術，可望改善上述問題，並達成最大化使用GNSS-RO低層資料的目的。
;This study aims to explore the presence of negative bias in lower-level data in GNSS-RO refractivity data caused by ducting. Since ducting occurrences cannot be directly detected from observational data, this study relies on model data for assessment. Comparisons show that when ducting occur in the model, the significant negative bias exists in corresponding observational data below the ducting layer.
An examination of thermal profiles during ducting occurrences in the model identifies three primary factors contributing to ducting: mixing inversion in boundary layer, cold pools from rainfall systems causing severe near-surface cooling, and subsidence inversions induced by other systems like anticyclone, with sharp vertical temperature and humidity gradients acting as primary drivers.
To evaluate the impact of these negatively biased GNSS-RO data on the model, a heavy rainfall event on May 22, 2020, is selected as the focal case. The study divides experiments into four groups: 1. CTRL, without assimilating GNSS-RO data; 2. REF_ALL, assimilating all GNSS-RO data; 3. REF_LSW, using LSW>35% as a QC strategy; and 4. REF_NODUCT, removing data below the ducting layer. After three and a half days of data assimilation cycles, differences in analysis fields, forecast fields, and forecast rainfall are compared among the experiments.
Analysis reveals that assimilating GNSS-RO data significantly adjusts the dynamical and thermal environments in the region from the South China Sea to South China, making them more conducive to increased frontal intensity. Additionally, the utilization of QC strategies further enhances water vapor content. However, the performance of the dynamic field varies based on the QC strategy employed: removing more data in REF_LSW leads to weakened low-level convergence and decreased frontal strength, while retaining more data in REF_NODUCT yields the strongest low-level convergence and frontal strength, showing the importance of mitigating negative bias and preserving lower-level data.
As for forecast outcomes, REF_NODUCT exhibits the highest rainfall, REF_ALL is comparable to REF_LSW, and CTRL produces the weakest results. However, skill scores suggest that REF_ALL performs the best, followed by CTRL, while the performances of REF_NODUCT and REF_LSW are less satisfactory. This could be attributed to the eastward shift of the southwest vortex in South China caused by the removal of lower-level data, resulting in a faster movement of the front compared to its original speed.
Incremental analysis experiments indicate that assimilating GNSS-RO low-level data leads to a significant reduction in local water vapor and can cause its movement due to the model advection. However, the impact on water vapor transport depends on its interaction with the dynamic field.
This study′s reliance on model data for ducting determination may lead to erroneous data removal due to model errors. Furthermore, removing data could lead to the eastward shift of the southwest vortex, causing rainfall displacement. Future efforts should focus on integrating techniques for directly determining ducting from GNSS-RO observational data and reconstructing data below the ducting layer to address these issues and maximize the utility of GNSS-RO lower-level data. |
