The Development On A Multiscale Identifying Algorithm For Heavy Rainfall And Methods Of Extracting Evolvement Information

A multi-scale rainstorm echoes identification method is developed by using hierarchical clustering algorithm, and echoes evolution feature with different scales are analysed. The quality of 1h accumulative rainfall amount both using the multi-scale and tracking radar echoes by correlation (TREC) have been tested using real radar retrieval.1 Hierarchical k-means clustering method has been applied to segment radar image ,which has been investigated by squall line rainstorm observed by the GuangZhou radar and typhoon KANU rainstorm observed by the WenZhou radar. It is found that hierarchical k-means clustering method can realize multi-scale storm identification, and the result of identification is steady and reasonable by analyzing the result of segmenting consecutive CAPPI.2 The multi-scale identification method is insensitive to varying parameter, which is favorable for applying the technique to different weather system. The different scale convective systems can be identified, tracked and extrapolated with different lead time of forecasting . 1h accumulative rainfall amount extrapolating with clustering identification and different trends are slightly closer to that retrieving by radar observation than that extrapolating with TREC technique and linear trend.3 The evolution character of different scale rainfall echoes has been studying based on multi-scale identification of squall line and typhoon rainstorm. It has been revealed that small scale convective cells with mean lifetimes of tens of minutes, large spatio-temporal variation, feeblish regularity is difficult to nowcasting, while larger scale rainfall echoes with mean lifetimes of many hours, better regularity and obvious evolution stages has a certain extent predictability.4 The TREC have method the capability of tracking rain-band with appropriate parameter. The "noisiness" of TREC vector can be reduced by checking the consistency of each motion vector. The forecast radar reflectivity pattern is obtained by extrapolation based on smoothed TREC vectors. 5 The TREC technique is applied to typhoon KANU observed by the WenZhou single radar and tropic storm BILIS observed by FuJian radar netting . The studies have clearly demonstrated the ability of the TREC technique tracking rotating rain-band on cyclic character of the TREC vectors and position of typhoon eye. The motion vectors of typhoon rain-band are models by a Rankine combined vortex containing a core region where velocity increase out to a ring of maximum velocity. The forecast results extrapolating by temporal integration of the two-time-level semi-Lagrangian scheme excel that extrapolating linearly.6 Rainfall echoes are extrapolated by combining cluster identification method with TREC tracking technology. The position error and intensity error are discussed by analyzing forecast error using both clustering and TREC technique.