Characteristic Analysis Of Thunderstorm In South China And Potential Prediction Of XDBoost Algorithm Under Multiple Factors

For a long time,thunderstorms have brought certain hazards to people's lives and property and public facilities.Correctly understanding the characteristics of thunderstorms and accurate forecasting are important ways to reduce lightning damage and ensure personal safety,it is also a question that scientists have been discussing.In recent years,some researchers have used changes in atmospheric nitrogen oxides(NO_x)and ozone concentrations(O₃)to analyze the characteristics of thunderstorms.Similarly,because thunderstorms occur in the troposphere,whether the impact has a disturbance error of the tropospheric delay ZTD on the navigation and positioning system,in this aspect,few scholars have discussed whether the above two can be used as factors for thunderstorm potential prediction like convective parameters.Based on the existing classifier algorithm to predict the potential of thunderstorms,when the recall rate,AUC value and accuracy rate of the forecast model cannot be balanced,this paper introduces the XGBoost algorithm based on grid search combined with cross validation.The XGBoost GS-LOO-CV model has better performance in recall rate,AUC value and accuracy rate under experimental comparison,which can be better applied to the forecast of thunderstorm potential in South China.The specific research process of this paper includes:1.This paper converts the map coordinate system used in domestic lightning positioning,and converts the Mars coordinate system to the coordinate system WGS84,the data source corresponds to the data on ZTD,convection parameters and atmospheric NO₂ and O₃concentrations,and standardizes the frame of the coordinate system where the data is located.In addition,the NO₂ and O₃ concentration data are interpolated and compared in 8 cases through3 interpolation methods,the KNN5 interpolation method whose average deviation is smaller is selected as the selection.In the same way,three interpolation methods are used to extract the grid data CAPE,CIN,KI,and TT,and the above indicators are used to select the data obtained by the inverse distance interpolation method.The source and extraction method of LI and ZTD data are also explained.2.By analyzing the relationship between the time of occurrence,rainfall,and latitude area under thunderstorm weather,it is concluded that the thunderstorm period is most concentrated at 16:00 pm,accounting for 34%of the statistical thunderstorm period.The relationship between rainfall and thunderstorms is not absolute,rainfall in thunderstorm weather depends on whether the dry₂ and cold clouds of convective clouds change the hot and humid clouds,causing the precipitation of water vapor.Analyzing the topography of South China and explain why the frequency of thunderstorms in higher latitudes is higher than that in mid-latitude and lower latitudes.At the same time,based on the chemical point of view,the reasons why NO₂ and O₃concentration changes during thunderstorms will increase and deviations are analyzed.At the same time,the reasons why the changes of NO₂,O₃ and ZTD during thunderstorms will become higher and deviations are analyzed.Finally,through ZTD,convection parameters and NO₂ and O₃ concentrations,a bivariate correlation analysis of thunderstorm potential is done,and the absolute value is arranged according to the correlation size:CAPE=CIN>LI>TT>NO₂=KI>O₃>ZTD.3.Compare the classifier algorithm with the XGBoost algorithm by introducing naive Bayes classification and binary logistic regression,by comparing the performance of the three in AUC value,recall rate and accuracy rate.T he experiment shows that the XGBoost GS-LOO-CV model constructed by the XGBoost algorithm has the best overall performance.In all indicators are better than the first two,which can be well applied to the forecast of thunderstorm potential in southern China.