Research On Classification And Nowcasting Of Convective Weathers Based On Pattern Recognition Theory

Convective weathers(hail,heavy rain,convective wind and tornado)are great disasters for the public;therefore,nowcasting(0-2 hours forecasting)severe convective weathers is an essential part of weather forecast.Convective storms that cause convective weathers often have small spatial scales and rapid structural evolution.So it is difficult to forecast the occurrence of convective weathers.This paper describes a system used for nowcasting of severe convective weathers using Doppler weather radar.The system is capable of(1)automatic identification,tracking and extrapolation of various convective objects,such as convective storms and convective cells and(2)classification of severe convective storms.The main work and results are as follows:(1)Identification,tracking and extrapolation methods for convective objectsTo solve the problems of identification,tracking and extrapolation of convective objects,this paper proposes a new convective nowcasting method that bases on region tree structure representation of radar reflectivity images.In this method,various convective objects are identified by clustering regions in the region tree;tracking of convective systems is solved by matching region trees;position and structure nowcasting of convective systems is achieved by extrapolating the whole region tree structure.The experimental results show that the tree-based storm tracking and extrapolation algorithms achieve better performance.(2)Automatic detection of mesocyclonesMesocyclones can be used to distinguish common storms and supercells.This paper proposes an automatic mesocyclone detection algorithm,which first identities and clusters two dimensional velocity couplets to form three dimensional vortices,and then tracks these vortices to obtain their durations.Mesocyclones are diagnosed from vortex detections based on a group of criteria that are obtained using a training dataset.The experimental results show that the proposed algorithm produces small false alarm rate.(3)Automatic detection of bow echoesThe bow echo is a kind of bond echo that is often associated with damaging winds.This paper proposes an automatic bow echo detection algorithm that is used for identification of large scale bow echoes.The structural characteristics of the bow echo that are indicators of damaging winds are identified,tracked and extrapolated to predict convective winds.The main steps of the method include: the detection of quasi-linear convective system,the extraction and analysis of the boundary of the quasi-linear convective system,and the detection and tracking of the contour notches.Finally,this paper presents the results of this method running on typical bow echo cases.