基于机器学习算法的雷达回波特征与闪电数量验证  

Verification of Radar Echo Features and Lightning Quantities Based on Machine Learning Algorithms

作　　者：计博严 吴刚 余博嵩 Ji Boyan;Wu Gang;Yu Bosong(Jiangxi Meteorological Detection Center,Nanchang 330095,China)

机构地区：[1]江西省气象探测中心,江西南昌330095

出　　处：《气象与减灾研究》2024年第4期271-278,共8页Meteorology and Disaster Reduction Research

基　　金：江西省气象局青年人才培养项目(编号:JX2020Q01).

摘　　要：全球范围内极端天气事件的频发,尤其是强对流天气的增多,对民众的日常生活及社会经济活动产生了显著影响。雷电作为这类灾害天气的标志性现象,对其开展有效的监测与深入研究,对于增强防灾减灾能力至关重要。文中采用江西2019—2021年汛期11次雷暴案例数据和雷达回波数据,结合4种主流机器学习算法验证闪电数量。实验结果表明,随机森林与XGBoost算法在验证闪电数量方面显著优于传统的多元线性回归和决策树算法,展现了更高的验证效能。由于XGBoost模型的参数复杂度高且训练效率相对较低,其在实际应用中可能存在一定的局限性。基于对数据规模与模型运行效率的全面考量,采用随机森林算法来解析雷达回波信息,进而验证闪电活动频次,是一种高效且实用的方法路径。The frequent occurrence of extreme weather events worldwide,particularly the increase in severe convective weather,presented a significant impact on people’s daily lives and socioeconomic activities.Lightning,as a hallmark phenomenon of such disastrous weather,effective monitoring and in-depth research on which is crucial for enhancing our capabilities in disaster prevention and mitigation.Based on the data from 11 thunderstorm cases and radar echo data during the flood seasons from 2019 to 2021 in Jiangxi Province,the lightning occurrences were validated combined with four mainstream machine learning algorithms.Experimental results indicated that the Random Forest and XGBoost algorithms were significantly better than traditional multiple linear regression and decision tree algorithms in validating lightning occurrences,exhibiting superior validation efficiency.However,due to the complexity of model parameters and relatively lower training efficiency,XGBoost might have limitations in practical applications.Considering the data scale and the operational efficiency of the model,the application of Random Forest algorithm to analyze radar echo information and subsequently validate lightning activity frequency proved to be an efficient and practical approach.

关 键 词：闪电数量验证 机器学习 雷达回波 

分 类 号：P427.32[天文地球—大气科学及气象学]