Research On Automatic Identification And Location Of Weather System Based On High Resolution Wind Field Pattern

The analysis and identification of mesoscale weather systems is the main basic work in analyzing and forecasting mesoscale convective weather.The data includes weather elements such as wind,temperature,pressure and humidity at the relevant pressure level.The item includes vortex,shear line,front,jet,and significant streamline.Vortex and shear line system of wind field contribute to form and accompany strong convective weather.At present,systems like vortex or shear line are still analyzed by manual and computer-aided production.This paper proposes a series of methods for objective identification and localization of vortex and shear line systems in high-precision wind field.The intelligent analysis method of weather system is used for the Weather forecast production system,ie MICAPS,which contribute to intelligent weather forecasting.The main contents and contributions of this paper are introduced into the following aspects:1)Flow pattern classification and feature analysis of 2-D wind direction fieldThe wind data on the isostatic surface processed by the MICAPS system is stored and displayed in the form of a 2-D matrix.Each wind system has its own flow pattern and structure.This paper takes the northern hemisphere as an example to divide the wind flow into ten patterns,which are clockwise circular,counter-clockwise circular,counterclockwise convergence,clockwise divergence,convergence,divergence,saddle,parallel,shear and random.The characteristics of standard deviation,kurtosis,vorticity and divergence are analyzed respectively.It is proposed a feature vector extraction method based on the directional angle deviation for the classification of the standard center symmetrical flow field.2)Methods of automatically identifying and positioning vortexes in wind fieldFor the data preprocessing,a local statistical feature used to filter uniform and invalid random direction areas is proposed to obtain the optimal vortex candidate regions in the complex high-precision Numerical Weather Prediction data.In the vortex flow recognition part,a discriminant rule based on the directional angle deviation and the vortex deformation mode is proposed to retrieve the candidate point set that covers the center area of the vortex.Then,the point set is clustered to obtain the vortex system.Finally,an algorithm of precise center location based on points expansion and extream vorticity is provided.The contrast experiments verify the proposed method can solve the multi-distribution,multi-scale and polymorphic problems of vortex,and has more advantages in locating and tracking tropical cyclone centers.It is also proposed a method based on decision tree for classifying the complicated central symmetric wind field.The directional angle difference,vorticity,divergence,and the template similarity are taken as the classification features.The experimental part verifies that those features have lower correlation and better classification effect.3)Methods of automatically identifying and positioning shear lines in wind fieldFirst,a wind direction matrix is transformed into grayscale map for filtering and edge point extraction.Then,the wind flow and the quantitative models of the single-type shear line are constructed,which used for the extraction of the shear region as to extract the shear candidate point set.Finally,each shear line system is determined by clustering the point set and each line is refined and displayed.In addition,this paper also proposes a rapid unified identification method of multi-class shear line.Kurtosis,the statistical feature of the direction data,is used to estimate the wind direction shear area.The evaluation experiment proves the advantage of single-class recognition method in recognition accuracy,and the advantage of multi-class recognition method in processing efficiency.