贝叶斯膨胀算法对EnSRF雷达资料同化的影响研究

本文针对2009年6月5日发生在我国华东地区的一次中尺度对流过程（Mesoscale Convective System，简称MCS），基于集合均方根滤波（Ensemble Square Root Filter，简称EnSRF）方法同化多部多普勒天气雷达资料，引入具有时空自适应理论优势的贝叶斯膨胀算法，通过与常数膨胀算法的对比，分析了两种膨胀算法对EnSRF同化效果的影响。结果表明:贝叶斯膨胀算法同化的雷达组合反射率因子在强对流中心处有所增强，改善了基于常数膨胀算法的EnSRF同化强对流系统偏弱的问题。相比常数膨胀算法，贝叶斯膨胀算法同化的冷池结构更合理，径向风和反射率因子的均方根误差均减少。进一步探讨贝叶斯膨胀算法对同化效果改善的原因，结果发现:贝叶斯膨胀参数的分布与反射率因子的均方根误差分布十分吻合，这表明贝叶斯膨胀算法可以在背景场均方根误差较大，即背景场与观测差距较大时，给出较大的膨胀参数，进而增加集合的背景场误差，使得观测权重增大，从而给出了较大的分析增量。对集合平均分析场进行了1小时的确定性预报发现，贝叶斯膨胀算法提高了预报模式对安徽与江苏交界处的强对流系统的模拟效果，回波强度更强，冷池强度和范围更大，且对于不同组合反射率因子的阀值，贝叶斯膨胀算法的评分（Equitable Threat Score，简称ETS）均高于常数膨胀算法。这表明贝叶斯膨胀算法有效地改进了基于常数膨胀算法的EnSRF同化雷达资料的效果。

Impact of Bayesian Inflation Method on Assimilation of Doppler Radar Data with EnSRF Method

The mesoscale convective system (MCS) occurred on 5 June 2009 in eastern China is simulated using the Advanced Regional Prediction System (ARPS) model and Doppler Radar data is assimilated with EnSRF. Bayesian inflation method is introduced in this study, which allows the inflation parameter to vary in space and time. The impact of Bayesian inflation method on assimilation of radar data with the ensemble square root filter (EnSRF) is investigated by comparing with the simulation using the multiplicative inflation method. Experimental results show that: the simulated composite reflectivity and cold pool from the Bayesian inflation experiment are stronger than that from the multiplicative inflation experiment; Bayes inflation method improves the performance of EnSRF, which always underestimates convection at the storm center. In the convective region, root mean square innovation of radial velocity and reflectivity in the Bayes inflation experiment are lower than that in the multiplicative inflation experiment. Further analysis indicates that the structure of Bayes inflation parameter corresponds very well to the root mean square innovation of reflectivity, which explains why the performance of EnSRF based on Bayes inflation method is improved. It is found that Bayes inflation method can give more weight to radar observations by increasing background error and provides bigger analysis increment when the root mean square innovation (RMSI) of background is bigger. Simulations of the two analysis fields show that the reflectivity near Hefei is stronger and the convective area of MCS is larger in Bayes inflation experiment than in the multiplicative inflation experiment. The simulated cold pool is colder and the area is bigger from Bayes inflation experiment than from the multiplicative inflation experiment, and corresponds well with observed reflectivity. ETS (Equitable Threat Score) of composite reflectivity from Bayes inflation experiment is higher than that from the multiplicative inflation experiment for various thresold. These resulsts suggest that Bayes inflation method improves the performance of EnSRF in radar data assimilation compared to that based on multiplicative inflation method.