Prediction Of Hail And Short-term Heavy Rainfall Based On Physics Data

Hail and short-term heavy precipitation are severe weather phenomenon caused by a strong convective weather system,which could cause huge losses to agriculture,construction,communications,electricity,transportation,people's lives and property.Many scholars have conducted related research on the identification technology of strong convective weather radar echoes.Prediction by radar information is only a reflection of the situation,lacking long time forecasting advancement.Therefore,based on the weather background of hailstorm short-term heavy precipitation,we can use machine learning to obtain a more reliable method.This paper proposes to use machine learning models to predict hail and short-term heavy precipitation based on physics field data of Tianjin meteorological stations and surrounding sites,and finally achieves the accurate forecast within one hour of forecasting time.The main tasks of this paper are given as follows:1 The model is established based on the data observed by the ground observation site.It integrates the physical field data of the first three hours of the current time observed at the Tianjin ground station.2 Under sampling the data on the sunny day.We use SMOTE oversampling method to increase the number of hail samples to solve the problem of category imbalance.3 Using PCA to reduce data dimensions and solve overfitting problems.4 Establishing the GBDT+LR model to predict hail and short-term heavy precipitation.After training GBDT+LR model,the probability of detection about hail prediction obtained by the cross validation verification set is 90.2% and the critical success index is 85.9%.The probability of detection about short-term heavy precipitation is 94.5% and the critical success index of short-term heavy precipitation is 85.5%.Meanwhile,we forecast the meteorological data of Tianjin in 2019 and accurately forecast the process of most hail and short-term heavy precipitation.