Hail Identification Based on Multiple Machine Learning Algorithms
Jin Yuchen
Zheng Xucheng
Su Lijuan
Li Hui
Xin Yue
Yi Na'na
Abstract：Based on the hail data of 119 observation stations in Inner Mongolia Autonomous Region from 1959 to 2020 and the ERA5 reanalysis data,the objective hail sample labeling procedure is pro-posed.Hail recognition model is constructed by using Support Vector Machine(SVM),K Nearest Neigh-bor(KNN)algorithm,Logistic Regression Model and ensemble model for training,testing and optimiza-tion,and the guidance ability of the models is evaluated.The research results show that the models con-structed by SVM and KNN are more accurate in hail prediction.The probability of detection(POD)of hail based on independent test samples is more than 82%,and the TS score surpasses 0.70.The ensem-ble model constructed based on the fusion of the three models has a score of 0.74,and is capable of provi-ding better hail prediction.The comparison of the performance of the three optimized models and the en-semble model reveals that the SVM and KNN have higher recognition accuracy in identifying hail and non-hail,while the Logistic Regression Model has the lowest recognition accuracy.The ensemble model has a higher POD and a higher false alarm rate(FAR)than KNN algorithm,so it is more capable of giv-ing correct recognition results for hail samples.For the prediction of 136 hail processes with specific time records in 2021 and 2023,KNN has the highest POD and the Logistic Regression Model has the poorest prediction effect.
Keywords：hail identificationartificial intelligencemachine learning algorithmsERA5 reanalysis data
Publication Date：2025-11-30
Online Publishing Date：2026-09-12(First online date of this platform, not the publication date of the document)
Pages：9( 90-98 )