基于数学形态学的三维风暴体自动识别方法研究

基于雷达数据的风暴体识别、追踪及预警方法是重要的临近预报技术之一,其中准确的风暴体自动识别是进行风暴体自动追踪和预警的前提。在风暴体识别中常会碰到的两个问题是:虚假合并和从风暴簇中分离出相距较近的风暴单体。美国国家大气科学研究中心提出的TITAN(Thunderstorm Identification,Tracking,Analysis,and Nowcasting)算法使用单阈值进行识别,容易将相邻的多单体回波识别为一个风暴体。美国国家强风暴实验室提出的SCIT(Storm Cell Identification and Tracking)算法使用7个反射率因子阈值进行识别,可以较好地分离出风暴簇中的风暴单体,但它直接抛弃了低阈值的识别结果,导致风暴体内部结构信息的丢失。SCIT的这种识别策略可能会使处于初生阶段、强度较低的风暴体被错误地抛弃掉。TITAN和SCIT都无法完全识别出相邻风暴的虚假合并。为了解决这两个问题,文章提出了基于数学形态学的识别方法。该方法首先使用第1级阈值进行单阈值识别;其次,对识别得到的风暴体执行基于动态卷积模板的腐蚀操作,以消除虚假合并;然后,使用高一级阈值进行识别,并对识别得到的风暴体进行膨胀操作,当风暴体的边界在膨胀的过程中相互之间接触,或接触到了原来较低阈值识别的风暴体的边界时,则停止膨胀过程;最后,逐次使用更高级别的阈值进行识别,并在每一级阈值的识别过程中执行腐蚀和膨胀操作。试验结果表明,通过在多阈值识别的过程中综合使用膨胀和腐蚀操作,基于数学形态学的三维风暴体识别方法不仅能够成功地识别出风暴体的虚假合并,同时还能在从风暴簇中分离出相距较近的风暴单体时,尽可能多地保留风暴单体的内部结构信息。

3D STORM AUTOMATIC IDENTIFICATION BASED ONMATHEMATICAL MORPHOLOGY

The storm identification,tracking and forecasting method is one of the important nowcasting techniques and accurate storm identification is the prerequisite of successful storm tracking and forecasting.Storm identification faces two difficulties:one is false merger and the other is to isolate adjacent storms in a cluster of storms.The TITAN(Thunderstorm Identification,Tracking,Analysis,and Nowcasting) algorithm is apt to identify adjacent storm cells as one storm because it uses a single reflectivity threshold.The SCIT(Storm Cell Identification and Tracking) algorithm uses 7 reflectivity thresholds and therefore is capable of isolating adjacent storm cells,but it discards the results identified by the lower threshold,leading to the loss of the internal structure information of storms.This strategy of SCIT may erroneously miss initiating storms with low reflectivity.Both the TITAN and SCIT have the problem of failing to satisfactorily identify false merger.To overcome these shortcomings,this paper proposes a novel approach based on mathematical morphology.The approach first applies the single threshold identification followed by implementing a special erosion process using dynamic convolution mask to resolve false merger problem.During multi-threshold identification stages,dilation operation is performed against the storm cells which are just obtained by the higher threshold identification,until the storm edges touch each other or touch the edges of the previous storms identified by the lower threshold.The results of experiment show that by combining the strengths of the dilation and erosion operation,this approach is able to successfully recognize false merger as well as to keep the internal structure of sub-storms when isolating storms from a cluster of storms.