深度学习方法预测阿克苏地区冰雹云雷达回波个例分析  

Case Analysis of Deep Learning Methods for Predicting Radar Echoes of Hail Clouds in Aksu Region

作　　者：黄静 佘勇 樊予江 HUANG Jing;SHE Yong;FAN Yujiang(College of Electronic Engineering(College of Meteorological Observation),Chengdu University of Information Technology,Chengdu 610225,China;Xinjiang Municipality Weather Modification Office,Urumqi 830002,China)

机构地区：[1]成都信息工程大学电子工程学院(大气探测学院),四川成都610225 [2]新疆人工影响天气办公室,新疆乌鲁木齐830002

出　　处：《沙漠与绿洲气象》2024年第2期107-113,共7页Desert and Oasis Meteorology

基　　金：新疆气象局科技创新基金项目(MS202229)。

摘　　要：利用2021—2022年4—9月阿克苏地区冰雹云的雷达回波资料,基于轨迹GRU模型和GAN模型共同构建一个深度学习的回波外推模型,应用于强对流(冰雹)天气监测预警。采用分阈值和预报时效的评估方法,对深度学习的回波外推模型预测回波的效果进行分析,结果表明:(1)在30 min预测时间内,随反射率阈值增加,临界成功指数(CSI)和命中率(POD)逐渐降低,虚警率(FAR)先降低后升高,FAR在反射率阈值为35 dBZ时最低。(2)在反射率阈值为35 dBZ和相同外推时效的情况下,基于深度学习的回波外推模型和光流法相比,CSI提高0.05~0.15,POD提高0.05~0.15,FAR降低0.05~0.12。(3)在预测反射率阈值为35 dBZ的强对流单体移动路径方面,基于深度学习的回波外推模型与TITAN法相比,预测的单体移动路径更接近实况单体移动路径。Utilizing the radar echo data from the hail cloud in the Aksu region from April to September in 2021-2022,a deep learning echo extrapolation model was jointly constructed based on the trajectory GRU model and generative adversarial networks(GAN)model.This model was applied for severe convective(hail)weather monitoring and early warning.An evaluation method based on different thresholds and forecast duration was employed to analyze the echo prediction performance of the deep learning model.Results indicate that:(1)Within 30 min forecast period,as the reflectivity threshold increases,both the Critical Success Index(CSI)and the Probability of Detection(POD)gradually decrease,while the False Alarm Rate(FAR)first decreases and then increases,reaching its lowest at the reflectivity threshold of 35 dBZ.(2)At the reflectivity threshold of 35 dBZ and the same extrapolation duration,compared to the optical flow method,the deep learning echo extrapolation model increases the CSI and POD by 0.05-0.15,and reduces FAR by 0.05-0.12.(3)In terms of predicting the movement path of severe convection cells at a reflectivity threshold of 35 dBZ,the deep learning echo extrapolation model,compared to the TITAN method,provides a prediction path that is closer to the actual movement path of the cells.

关 键 词：强对流 深度学习 轨迹GRU GAN 

分 类 号：P456[天文地球—大气科学及气象学]