A Forecast Method Of Downburst Base On Doppler Radar

Downburst is a kind of meteorological disaster that can cause local wind share which has the characteristics of short life cycle and fast generation.These characteristics bring big obstacle to the identification and forecast of downburst.At present,weather forecast mainly relies on Doppler weather radar products.The main work of this paper is as follows:Get severe convection weather disaster event report from common database and select disaster event report related with downburst and hail and rain.According to information find the radar catching this disaster and batch download Doppler weather radar data.Taking full advantage of SCIT algorithm,calculate the feature of reflectivity and feature of radial velocity and get monomer array and calculate time difference sequence of monomer array.According to the information of severe convection weather disaster event report,find the monomer array leading to the disaster.Split monomer array and get monomer array segment set.Organize the negative sample of the downburst monomer array segment set and taking advantage of statistical significance analysis,get effective feature.Use principal component analysis(PCA)algorithm and select two-dimensional principal component and train support vector machine(SVM)model classifier and find the best value.Calculate the maximum mutual information coefficient(MIC)of every positive and negative sample and selecting the subset of effective feature with less-than 0.4 as the effective feature of Logistic regression model and find the best value.Build the balloting system of every classifier and according to joint decision,improve forecast accuracy.Compare the effect of dual model classifier and analyse the effect.In summary,this paper proposesa downburst recognition and forecast method basing on the array segment(time window).This method is able to forecast downburst before 4 to 8 minute based on increasing the hit ratio and decreasing the rate of false.This paper provides a new way for severe convective weather recognition basedon radar.