【作者】 楚志刚；

【导师】 银燕；

【作者基本信息】 南京信息工程大学 ， 大气物理学与大气环境， 2013， 博士

【摘要】 新一代天气雷达径向速度场存在速度模糊的问题。模糊一直困扰着雷达速度场的应用,使主观分析变得困难,使资料同化和风场反演结果变得不可靠。基于软件的退模糊算法是解决这个问题的途径之一。本文以CINRAD-SA业务雷达观测数据为研究对象,围绕着速度退模糊研究中存在的问题(速度模糊特征、退模糊算法的污染作用、连续噪声抑制、算法验证),开展了四个方面的研究。首先,用业务雷达观测数据,初步分析了速度模糊的特征,包括模糊的发生频率及其在站点、类型、时间和空间上的分布。分析结果认为：CINRAD-SA的速度模糊现象是普遍存在的,沿海雷达的模糊略多于内陆雷达,弱对流回波的模糊发生频率最高,冬半年的模糊远多于夏半年,而且模糊多出现在仰角为6.0。,方位角为70。(或250。),径向距离为50-60km,高度为5-6km的地方。这些特征为之后速度退模糊新算法的设计和验证提供依据。接着,从速度退模糊的基本原理出发,分析了退模糊错误的类型以及错误产生的原因。本文把错误分为两类,一类是一般错误,另一类是污染错误。通过分析发现,这两类错误的产生原因和危害不同,一般错误是由缺测引起的,而污染错误是由连续噪声所致。污染错误使退模糊后数据质量的改善情况变得不确定,这也是退模糊算法在应用中遇到的主要问题。因此,减小连续噪声的影响是提升退模糊算法性能的重要途径。基于上述的研究结果,本文提出了一种抗噪声的速度退模糊新算法(AND算法)。AND以减少污染错误为首要目标,通过消除连续噪声的影响提升算法性能。算法有三个步骤,噪声分离、曲线退模糊、噪声恢复,其中包含了一个无损的“分离-恢复”噪声抑制方案。这个方案解决了噪声抑制过程中误删除非噪声数据的问题,可以在有效抑制连续噪声的同时又不损失任何风场信息。退模糊时,以三条拟合曲线加权的方式来进一步抑制残留噪声,同时可以动态匹配不同尺度的风场。最后,为了验证AND算法的实际性能,从四站三年的业务雷达数据中选出所有模糊文件作为数据集,以文件正确率作为评分方法进行了全面的验证,与现有退模糊算法进行了直接和间接的比对,并给出了退模糊个例说明。验证结果表明：总的退模糊正确率接近90%,台风、强对流的退模糊正确率>94%,不连续层状云的退模糊正确率也达到了70%。AND算法明显优于WSR-88D退模糊算法,退模糊正确率比其高30个百分点,且污染错误的发生概率有较大幅度的降低。更多还原

【Abstract】 Radial velocity fields from CINRAD weather radars are faced by the problem of velocity ambiguity. Ambiguity which inhibits the application of radar velocity fields will make subjective analysis difficult and make data assimilation and wind field retrieval unreliable. One effective solution is software-based dealiasing algorithm. In this paper, observations from CINRAD-SA operational radars were studied, and in response to the problems in velocity dealiasing (velocity ambiguity characteristics, contaminated errors, continuous noise inhibition ability, and algorithm validation), studies were conducted from four aspects.First, operational radar observations were used to pre-characterize velocity ambiguity, including the occurrence rate of ambiguity, and the inter-station, inter-type, temporal, and spatial distributions of ambiguity. The results showed that:velocity ambiguity was ubiquitous in CINRAD-SA; seashore radars were slightly more ambiguous than inland radars; echoes of weak convections had the highest occurrence rate of ambiguity; ambiguity occurred more frequently in winter half year than in summer half year; ambiguity occurred mainly at an elevation angle of6.0°, azimuth of70°or250°, radial distance of50-60km, or height of5-6km. Knowledge on these characteristics provides a foundation for design and validation of dealiasing algorithms.Second, the types and causes of dealiasing errors were analyzed by using the basic principles of velocity dealiasing. Errors were divided into two types:general error and contamination error. The two error types had different causes and harms. General error and contamination error were caused by missing data and continuous noise respectively. Contamination error showed negative effect on dealiasing algorithms and made data quality improvement unclear after dealiasing, so contamination error is a major problem for the application of dealiasing algorithms. Therefore, an important pathway to improve dealiasing performance is to reduce the impact from continuous noise.Third, based on the above results, a new anti-noise velocity dealiasing algorithm (AND algorithm) was proposed. The primary objectives of AND were to reduce contamination error and to improve algorithm performance by eliminating the impact of "continuous noise". AND was finished in three steps:noise separation, curve dealiasing, and noise restoration; it involved a lossless "separation-restoration" repressing scheme. This scheme could avoid the mistaken deletion of non-noisy data during noise suppression, effectively repress "continuous noise", and avoid the loss of any wind field information. During dealiasing, three fitted curves were weighted to further repress residual noise and dynamically match wind fields at varying scales.Fourth, all ambiguous files from4stations×3years of operational radar data were used to validate the actual performance of AND based on file precision; AND was also directly and indirectly compared with an existing dealiasing algorithm, with illustration by special cases. The results showed that:total dealiasing precision of AND was close to90%; precisions of typhoons and strong convections were>94%; precision of discontinuity stratus was70%. AND (with a dealiasing precision30%higher) was obviously superior over WSR-88D algorithm and greatly reduced the rate of contamination error.更多还原