Research On The Algorithm Of Three-Dimensional Precipitation Structure Inversion With GPM/DPR And CINRAD Data

Using of GPM/DPR data and the Tianjin CINRAD ground-base radar data,a three-dimensional precipitation inversion model is obtained through Support Vector Machine(SVM)algorithm.In order to evaluate the effectiveness of the precipitation inversion model,the estimation of precipitation by SVM method is compared with Z-I relationship and the subsection optimized Z-I relationship.This article mainly completed the following work:Firstly,screen out all the precipitation.Cases jointly detected by GPM/DPR and Tianjin CINRAD from May to September 2017 to 2019.Secondly,the geometric matching method is used to match the GPM/DPR data with CINRAD data in time and space.The accuracy of the matching is checked.The results show that the reflectivity of DPR and CINRAD have a good correlation,indicating that the matched samples have good temporal and spatial consistency.Therefore,the matched samples can be used to establish a three-dimensional precipitation inversion model.Finally,the SVM algorithm is used to obtain the three-dimensional precipitation inversion model,combined with particle swarm optimization(PSO)algorithm to optimize the parameters of the SVM.Compared with the estimation of the Z-I relationship,optimized Z-I relationship and the SVM algorithm,the conclusions are as follows:(1)The estimation precision for precipitation of SVM algorithm is better than other methods.(2)The relative errors of the three algorithms are analyzed on different reflectivity intervals.For the Z-I relationship,the precipitation is overestimated in the interval greater than 50 dBZ.The optimized Z-I relationship performances better than Z-I relationship,but there is still an overestimation.The estimation error for precipitation of SVM algorithm is stable,which won't be affected by the radar reflectivity.(3)At different distances from the radar,the estimation error for precipitation is relatively small within the range of 50?150 km,but the error increases outside the range of 150 km.However,the error of the SVM algorithm estimation value increasing with the change of the distance is relatively minimal,which shows that the SVM algorithm has a certain optimization for the precipitation estimation of the samples at a long distance.(4)From two special cases,the results show that the precipitation distribution and the vertical precipitation profile obtained by the SVM algorithm are closer to the GPM/DPR measured,which indicated that the SVM algorithm is the best of these three methods.