Identification And Intensity Estimation Of Convective Wind Based On Doppler Radar Data

Convective wind is one of the strong convective disaster weather,which can cause huge economic losses and has the characteristics of local,sudden and destructive,causing difficulties in accurate proximity forecasting.Traditional convective wind prediction relies on the identification of typical echo phenomena of convective winds,which requires accurate identification algorithms and has certain false alarms.Therefore,the proximity forecasting of convective winds is challenging.In this thesis,based on Doppler radar data,we use computer vision and machine learning methods to study the problem of identifying convective monoliths associated with convective winds and the problem of estimating the intensity of convective winds on the basis of segmenting out convective storms,and our main work is as follows:(1)In this thesis,based on the traditional radar phenomena related to convective gale(such as band echo,mid-level radial convergence,mesocyclone,etc.),five kinds of convective wind storm identification features are designed using computer vision methods based on Doppler radar reflectivity data and radial velocity data of convective storm,including storm moving velocity feature and core sink velocity feature,storm reflectivity high value feature,storm radial velocity high value feature,storm radial velocity shear feature,and storm radial velocity texture feature.(2)Based on the design of convective wind storm features,this thesis uses mutual information to select features,and a total of 16-dimensional features are selected,and based on the random forest model,a model that can effectively distinguish strong/nonstrong convective wind storm is obtained.Using convective storm samples from June,July and August 2016 in 13 cities of China,the probability of detection(POD)of SCW is 78.9%,the false alarm ratio(FAR)is 26.4%,and the critical success index(CSI)is61.5%.For the convective wind samples that carry typical echo phenomena,the POD,FAR,and CSI range from 89.4% to 99.3%,4.2% to 16.0%,and 76.4% to 95.1%,respectively.Meanwhile,the POD of samples without typical echo phenomena is66.8%.(3)Based on the identification of convective gale monomers,the intensity of convective gale is estimated in this thesis.A convective gale monomer time series dataset is constructed using radar data from 13 cities in China in June,July,and August2016,and the convective wind storm time series are analyzed to find the abrupt variability of convective wind sequence wind speeds.Later,for the sequence maximum wind speed,the convective storm characteristics are extracted and a regression model is constructed for intensity estimation.The prediction score indexes RMSE and MAE for convective wind speed above 17.2m/s reach 3.124 and 2.318 respectively.