本研究分為兩階段，第一部分利用觀測增量樣本估計徑向風水平方向上的觀測誤差相關性。結果顯示即便使用不同解析度之超級觀測，徑向風的誤差尺度皆約為25公里。接著，第二部分實驗將徑向風誤差相關性的估計結果加入至雷達資料同化系統中。由同化實驗結果可知，相較於不考慮觀測誤差相關性的實驗，同化徑向風觀測時加入觀測誤差相關性可產生空間上較小尺度的風場修正，而此風場修正會使得分析場在梅雨鋒面區域擁有較強的輻合、激發較多的局部對流，並伴隨較多的水氣量以及較強的西南風，進而提供對流良好的發展條件。風場與水氣場的修正隨後改善了回波及降水的預報。除了更好的單一預報表現之外，在機率定量降水預報(PQPF)當中，可以得知在同化系統中加入徑向風觀測誤差相關性後，系集能有更高機率預報出較為接近觀測的降水強度以及空間分布。另外，由每小時累積雨量預報FSS (Fractions Skill Score)得分之校驗結果得知，加入徑向風觀測誤差相關性之改善的效果可以維持約六小時。;An assumption of uncorrelated observation errors is commonly adopted in conventional data assimilation. For this reason, high-resolution data is re-sampled with strategies like superobbing or data thinning. This also sacrifices the advantage of high temporal and spatial resolution observations that can provide essential detailed structures. However, assimilating the high-resolution data, such as radar radial wind, without considering the observation error correlation can lead to overfitting and thus degrade the performance of data assimilation and forecast. This study uses the radar ensemble data assimilation system, which couples the Weather Research and Forecasting model and Local Ensemble Transform Kalman Filter (WRF-LETKF), to assimilate the radar radial velocity and reflectivity data. We present a strategy to include the error correlation of the Doppler radar radial velocity in the WRF-LETKF radar assimilation system and examine their impact on short-term precipitation prediction based on the heavy rainfall case on 2nd June 2017 in Taiwan.
We first estimate the horizontal error correlation scale for radial velocity based on the innovation statistics. The error correlation scale is approximately 25 km. The introduction of correlated observation error for Doppler radar radial winds exhibits more small-scale features in the wind analysis corrections compared to the experiment using the independent observation assumption. Consequently, the modification on wind corrections leads to stronger convergence accompanied by higher water vapor content, and induces subsequent local convections, resulting in more accurate simulated reflectivity. For probability quantitative precipitation forecast (PQPF), our results show that the experiment using the correlated observation error has higher probability to generate heavy rainfall and agrees better with the observation.
