Region-adaptive Modeling For Severe Convection Weather Classification And Recognition

Severe convection weather(hail,short-term heavy rainfall,etc.)has the characteristics of short life cycle,strong locality and great destructive power,which is one of the main problems of meteorological disaster prevention and mitigation.Although many achievements have been made in the study of severe convective weather,due to the complex topographic environment in China,there are obvious differences in severe convection weather between different regions.Aiming at this problem that the applicability of existing algorithms to regions is poor,this paper proposes a region-adaptive method for severe convection weather on the basis of summarizing the previous work,which constructs a classification and recognition model of severe convection weather from the traditional machine learning method and deep learning method.The main work of this paper is as follows:(1)Doppler weather radar is one of the main tools for monitoring severe convective weather.Extracting a large number of hail and short-term heavy rainfalls samples from radar data is the basis of this paper.Improving sample extraction method and realizing semi-automatic extraction of samples can improve the work efficiency by nearly 10 times.Firstly,the record of hail reports is analyzed based on semantic analysis,from which key information such as time,place and hail size is extracted.Secondly,the samples is marked by using the "human-computer interactive convective monomer sequence marking subsystem" designed in this paper.Finally,the modeling database is formed.(2)The modeling of C band radar classification for severe convection weather is studied.Considering the influence of terrain environment,the melting distance and the terrain slope are extracted as features.A weighted support vector machine was established to distinguish between hail and short-term heavy rainfall,it has a good performance on hail recognition.The POD is 84.3% and the CSI is 79.8%.(3)The classification and recognition model of severe convection weather based on convolution neural network is designed,which avoids the artificial design features and is convenient to use adapt to the regional characteristics.By analyzing the characteristics of hail cloud on radar reflectivity factor map and combining with meteorological background knowledge,the radar data are pre-processed from three aspects: constructing three-dimensional monomer structure,image direction correction and image size processing.The overall architecture of convolution neural network is designed,and some strategies are adopted to avoid network overfitting.Experiments show that the model based on convolution neural network has a good hit ratio of 81.0% and the critical success index of 79.8%.