利用四维集合卡尔曼滤波迭代同化多普勒雷达资料的研究

A iterative four dimensional ensemble kalman filter (4DEnKF) method will be researched for assimilating successive multi-observations from Doppler radar. The radar observations, velocity and reflectivity, have two characteristics. One is they are not model forecast variable. Second is observation temporal density is very high. Unlike traditional sequential assimilation method, 4DEnKF could assimilate multi-time observations in single cycle. This method could reduce the frequency of assimilation contrasting high temporal frequency of radar observations. The temporal developing information from observations also could be included during assimilation. More informations form multi-time observations have the potential to improve the analysis of unobserved model variables in one assimilation cycle. A "Running in Place (RIP)" idea will be applied to 4DEnKF to assimilating observation iteratively. This iterative 4DEnKF method could decrease negative affecting from the quality of initial forecasting ensemble. The background error correlation in forecast ensemble will be improved step by step. The "spin-up" will be accelerated by more information extracted from beginning assimilation cycles. It could lead to a faster responding to severe weather. The combination of these two methods could improve assimilaton analysis in both single and multi cycle to give better forecast of severe weather in meso and small scale.

将开展四维集合卡尔曼滤波（4DEnKF）迭代同化连续多时次雷达观测的研究。针对雷达径向风和反射率不是直接的模式变量的问题，利用观测时间密度大的特点，使用4DEnKF在一次分析中同化连续多个时次的观测。减少同化的频繁度，并通过连续观测引入时间演变信息，改进单次同化的效果。更丰富的观测信息有望进一步改善非观测变量的分析结果，使分析更协调。在4DEnKF基础上引入"原地运行"（Running in Place，RIP）的迭代同化思路，研究迭代同化多时次观测资料的效果。通过该方法减小同化对初始集合样本质量的依赖程度。在先验信息不充足的条件下，通过对一系列雷达观测的反复同化，逐步改进样本构造的相关结构，从而帮助同化最大程度提取雷达观测信息，加快同化收敛速度。通过这两种方法的结合，改进雷达资料单次和循环同化的效果，使得预报能更快更好地反映出中小尺度高影响天气系统。

开展四维集合卡尔曼滤波（4DEnKF）迭代同化连续多时次雷达观测的研究。期望通过该方案可以利用雷达资料高时间密度的特点，在单次同化中分析连续多个时次的观测信息，在先验信息不足情况下，最大程度从观测中获取有用信息，减小初始样本质量问题的影响，加速同化的收敛。并建立一套基于4DEnKF-RIP同化雷达资料的方案，实现了对多个时次雷达资料的迭代同化能力。利用浅水方程和模拟雷达资料进行了循环同化的理想试验，试验结果验证了方法的有效性。在多同化了非分析时刻的观测后，分析场有了进一步的改善。而迭代同化主要在同化循环的前1-2次中进行迭代有改进。在实际个例的同化试验中，4DEnKF-RIP方案并没有明显的优于EnKF的分析结果，但是1小时预报结果从系统的落区、形态和虚假回波等几个方面都与实况较为接近。通过研究，期望能够结合两种方法的优点，改进雷达资料单次和循环同化的效果，使得预报能更快更好地反映出中小尺度高影响天气系统。