期刊文献+
共找到2篇文章
< 1 >
每页显示 20 50 100
基于深度学习的天气雷达回波序列外推及效果分析 被引量:14
1
作者 黄兴友 马玉蓉 胡苏蔓 《气象学报》 CAS CSCD 北大核心 2021年第5期817-827,共11页
天气雷达探测资料是进行强对流天气临近预报的主要参考数据。针对传统雷达回波外推方法中存在资料信息利用率不足和外推时效有限的问题,文中利用神经网络进行雷达回波的外推、利用预测神经网络模型进行2 h以内的回波变化预报。回波外推... 天气雷达探测资料是进行强对流天气临近预报的主要参考数据。针对传统雷达回波外推方法中存在资料信息利用率不足和外推时效有限的问题,文中利用神经网络进行雷达回波的外推、利用预测神经网络模型进行2 h以内的回波变化预报。回波外推问题的关键是回波时、空序列预测问题,该网络具有解决时间记忆问题的长、短时记忆单元(Long Short-Term Memory,LSTM)和提取空间特征的卷积模块。应用福建、江苏和河南多年的雷达探测资料构造训练和测试数据集。为消除降水的不平衡和提高对强回波的预报准确率,网络采用带权重的损失函数进行训练。对光流法和预测神经网络进行测试集检验以及个例分析,结果表明,在相同外推时效和检验反射率阈值的情况下,预测神经网络的临界成功指数、命中率均高于光流法,虚警率低于光流法。不同类型降水预测神经网络的SSIM值(structural similarity)均高于光流法,且层状云降水的SSIM值比对流云降水的大。因此,预测神经网络对强回波的预报能力高于光流法;在预报时效性上,预测神经网络模型具有一定的优越性;预测神经网络对层状云降水预报的准确率比对流云降水的高。 展开更多
关键词 雷达临近预报 循环神经网络 卷积神经网络 损失函数 深度学习
下载PDF
Turbulence and Rainfall Microphysical Parameters Retrieval and Their Relationship Analysis Based on Wind Profiler Radar Data 被引量:1
2
作者 胡苏蔓 黄兴友 马玉蓉 《Journal of Tropical Meteorology》 SCIE 2021年第3期291-302,共12页
Rainfall is triggered and mainly dominated by atmospheric thermo-dynamics and rich water vapor.Nonetheless, turbulence is also considered as an important factor influencing the evolution of rainfall microphysical para... Rainfall is triggered and mainly dominated by atmospheric thermo-dynamics and rich water vapor.Nonetheless, turbulence is also considered as an important factor influencing the evolution of rainfall microphysical parameters. To study such an influence, the present study utilized boundary layer wind profiler radar measurements. The separation point of the radar power spectral density data was carefully selected to classify rainfall and turbulence signals;the turbulent dissipation rate ε and rainfall microphysical parameters can be retrieved to analyze the relationship betweenε and microphysical parameters. According to the retrievals of two rainfall periods in Beijing 2016, it was observed that(1) ε in the precipitation area ranged from 10^(-3.5) to 10^(-1) m^(2) s^(-3) and was positively correlated with the falling velocity spectrum width;(2) interactions between turbulence and raindrops showed that small raindrops got enlarge through collision and coalescence in weak turbulence, but large raindrops broke up into small drops under strong turbulence, and the separation value of ε being weak or strong varied with rainfall attributes;(3) the variation of rainfall microphysical parameters(characteristic diameters, number concentration, rainfall intensity, and water content) in the middle stage were stronger than those in the early and the later stages of rainfall event;(4) unlike the obvious impacts on raindrop size and number concentration, turbulence impacts on rain rate and LWC were not significant because turbulence did not cause too much water vapor and heat exchange. 展开更多
关键词 turbulent dissipation rate rainfall microphysical parameters wind profiler radar spectrum width collision-coalescence BREAKUP RETRIEVAL
下载PDF
上一页 1 下一页 到第
使用帮助 返回顶部