Research On Intelligent Nowcasting Method Of Severe Convective Wind Based On Weather Radar Data

Severe convective wind(SCW)is one of the most common severe convective weather in China,which causes huge economic losses every year.The SCW shows the characteristics of locality,suddenness and great destructiveness,which makes it difficult to nowcast SCW.The Doppler weather radar can generate data with high spatial and temporal resolution and is an important tool for observing and nowcasting the SCW.However,the existing intelligent algorithms for nowcasting the SCW cannot fully utilize the information provided by radar data,and the performance of nowcasting needs to be improved.In order to improve the performance of intelligent nowcasting algorithms for the SCW,this paper studies the use of image processing and machine learning method to control the radar data quality,automatically detect some radar phenomena related to the SCW,design the features about the SCW,construct various nowcasting models of the SCW,and so on.The main contributions and results are as follows.(1)In order to improve the prediction of the SCW based on radial velocity data,a new velocity dealiasing scheme is proposed,and the nowcasting model of the SCW based on large radial velocities is improved.The strong wind shear,noise data and isolated echo in radial velocity data severely interfere with the velocity dealiasing algorithm.This paper proposes a two-step velocity dealiasing algorithm,including intra-region and inter-region dealiasing.According to the abnormal velocity difference between regions,the operation of velocity dealiasing is carried out.On the basis of velocity dealiasing,the automatic identification algorithm of large radial velocities is improved by considering their correlation with SCW and filtering out the noise.Besides,an SCW nowcasting model based on large radial velocities is designed based on the relationship between the parameters of large radial velocities and SCWs.(2)An automatic gust front identification algorithm and an SCW detection model based on the gust front are designed.As the gust front is shown as a weak narrow-band echo in radar reflectivity image,an automatic gust front identification algorithm is designed.First,a local binary operator is designed to extract the potential weak narrow-band echoes.Then,image processing methods such as skeletalization,principal component analysis(PCA),and local binary pattern(LBP)are used to filter out various types of interference and extract the complete weak narrow-band echo.Finally,according to the location relationship between the gust front and its generating storm,the real gust front is extracted from the weak narrow-band echoes to reduce false positives and improve the performance of automatic recognition of the gust front.Based on the automatic identification of the gust front and the correlation between the parameters of the gust front and the SCW,an SCW nowcasting model is designed.(3)An automatic identification algorithm for the linear mesoscale convective system(MCS)and an SCW nowcasting model induced by linear MCS are designed.The algorithm for automatically identifying linear MCS consists of three steps.First,the multiscale idea is used to segment the convective region of MCS.Then,a standardized window(8 × 8 region)is used as the description of MCS.Finally,a 64 D feature from the 8 × 8 region is used to train the support vector machine model for identifying linear MCS.Based on the automatic identification of linear MCS,two typical structures of inflow gaps and high gradient leading edge associated with the SCW in linear MCS are extracted.With the help of the related research,an SCW nowcasting model based on the linear MCS is designed.(4)Two recognition models of SCW based on machine learning are proposed,including a random forest model based on features of the storm and an ensemble model based on the Stacking learning method.The random forest model is composed of the following steps.First,according to the local similarity properties of various radar phenomena related to convective winds,seven groups of features were designed.Second,the parameters of these features are determined by mutual information,and the features are selected by L1-norm.Last,the selected features are used to create the random forest model for predicting SCW.The ensemble model is constructed by the Stacking learning method.A variety of SCW detection models are considered as the base learners,and their outputs are used as the input to train the ensemble model.In this way,the performance of the SCW nowcasting model can be improved.