Analysis Of The Relationship Between Radar Echo And Rainfall Grade Based On Machine Learning

In the field of meteorology,using radar data to analyze and forecast rainfall has always been a hot topic,and it is of great significance.This is because rainfall has varying degrees of impact on commercial activities,agricultural production,transportation and all aspects of daily life.If we can accurately analyze the level of rainfall,we can take different measures for different levels of rainfall,so as to "prepare for a rainy day" and reduce or avoid economic losses.The traditional methods of rainfall grade analysis have some problems,such as the complexity of the model,the excessive amount of calculation and the insufficient use of radar data,which lead to the low accuracy of the analysis.This topic is based on the actual business demand of Liaoning Meteorological Disaster Warning Center.Using the S-band Doppler weather radar observation data and historical rainfall data throughout Liaoning Province,XGBoost method in machine learning is used to analyze the precipitation intensity level.Firstly,in terms of data processing,data cleaning and processing are carried out respectively for multilayer radar echo data and rainfall data to eliminate data noise.The time stamp was used as the matching basis,and the rainfall data and radar data were integrated to form the final data set of "eight-neighborhood multi-layer radar echo-rainfall data".Secondly,XGBoost ensemble learning method was used to analyze multilayer radar echo and rainfall intensity grade.In order to improve the accuracy of analysis results,the precipitation grade analysis model was optimized by combining grid search and crossover test.After the full study of the characteristics of different types of rainfall,this paper studied a drop water model method,this method in convective clouds and stratiform cloud precipitation rainfall on the basis of the differences between characteristics of radar observations,according to the characteristics of radar rainfall classification work in the first place,and then further,according to different types of rainfall adopt different analysis method.In order to verify whether XGBoost machine learning method can effectively analyze rainfall levels based on multi-layer radar echo data,this paper compares the results obtained from XGBoost precipitation level analysis model with the currently widely used GBDT machine learning methods.The results show that the analysis accuracy of XGBoost precipitation grade analysis model is higher than that of the above mainstream machine learning methods,and the analysis results are basically consistent with the subjective analysis experience.Finally,in order to facilitate users to operate a large number of multilayer radar data and historical rainfall,and ensure the transparency of the model to users,a simple and easy to use precipitation grade analysis system is constructed in this paper.The system construction process includes requirements analysis,system design,coding,testing and so on.At the end of the paper,the progress of all the work is analyzed and summarized,the progress and shortcomings are objectively evaluated,and the prospect of the follow-up work is proposed.