Weather Radar Identification And Nowcasting Of Thunderstorms

Weather radar is one of the effective tools for detecting convective weather and nowcasting due to its high temporal and spatial resolution.In this paper,thunderstorm and lightning,are detected by weather radar.In order to solve the problem of insufficient utilization of historical radar data and short extrapolation timeliness of traditional methods,a model based on deep learning is used for nowcasting and the results are compared with the optical flow method.The main contents and conclusions are as follows:(1)The DBSCAN clustering algorithm is used to identify the two-dimensional components of thunderstorms,the corrosion expansion in the morphology is introduced to eliminate the false combination of thunderstorm components,and then the three-dimensional thunderstorm cells are obtained by vertical correlation through the overlapping area.Finally,the characteristic quantity of each thunderstorm monomer is calculated.This method can effectively identify thunderstorm.After identifying the thunderstorm cloud,the weather radar was used to identify and explore the lightning.The result showed that the index of the 40 dBZ radar echo top higher than the temperature layer of-10 ? can better identify the lightning occurrence area.(2)The accuracy of the optical flow method for the prediction of thunderstorm cloud precipitation and lightning occurrence area decreases with time,but its prediction effect on the evolution of echo shape and the change of echo intensity is poor.The predictive neural network is better than the optical flow method in the prediction of the evolution of the radar echo shape,and it is suitable for the prediction of precipitation.However,neural network has a fuzzy effect,the optical flow method is more accurate in predicting details.The optical flow is more suitable for the forecast of Small and medium-scale weather systems than the predictive neural network.(3)An algorithm based on deep learning is used to extrapolate the radar echo sequence.Considering the frequencies of different rainfall levels are highly imbalanced and to improve the prediction accuracy of strong echoes,the network is trained by weighted loss function.The test set and individual case evaluation show that the CSI?POD of predictive neural network is higher than that of optical flow method and FAR lower than that of optical flow method under the same extrapolation aging and reflectivity threshold.Therefore,the predictive neural network has a higher ability to predict echoes than optical flow.(4)Through the forecast of precipitation in stratiform clouds and convective clouds,it is found that the accuracy of the optical flow method and predictive neural network for the precipitation process of stratiform clouds is higher than that of convective cloud precipitation,mainly due to the slow change of echo of stratified cloud precipitation and the rapid change of strong echo of convective cloud precipitation.Although the accuracy of the two nowcasting schemes decreases with the extension of forecasting time,the predictive neural network is still superior to the optical flow method.