周康辉,郑永光. 2021. 基于多源数据和深度学习的闪电短时预报方法[J]. 气象学报, (0):-, doi:10.11676/qxxb2021.002
基于多源数据和深度学习的闪电短时预报方法
The very short-range lightning forecasting with multisource data by deep learning
投稿时间:2020-03-12  修订日期:2020-06-04
DOI:10.11676/qxxb2021.002
中文关键词:  强对流 短时预报 深度学习 观测数据 数值模式预报
英文关键词:Convective weather, Very short-range forecast, Deep learning, Observation data
基金项目:国家重点研发计划项目(2018YFC1507504和2017YFC1502003),国家重点基础研究规划项目
作者单位E-mail
周康辉 国家气象中心 zhoukh@cma.gov.cn 
郑永光 国家气象中心 zhengyg@cma.gov.cn 
摘要点击次数: 85
全文下载次数: 76
中文摘要:
      强对流短时预报时段(2~6 h),具有较大难度。一方面,基于观测数据的外推已基本不可用;另一方面,高分辨率数值模式(High-resolution Numerical Weather Prediction, HNWP)的预报性能有待提升。本文利用深度学习方法,将卫星、雷达、闪电等观测数据和HNWP进行融合,得到了更有效的闪电短时预报结果。基于多源观测数据和高分辨率数值天气预报数据的特性,我们构建了一个双输入单输出的深度学习语义分割模型(LightningNet-NWP),使用了包括闪电密度、雷达组合反射率拼图、卫星成像仪6个红外通道,以及GRAPES-3km模式预报的雷达组合反射率等共9个预报因子。深度学习模型使用了编码-解码的经典全卷卷积结构,并使用池化索引共享的方式,尽可能保留不同尺度特征图上的细节特征信息;利用三维卷积层提取观测数据时间和空间上的变化特征。结果表明,LightningNet-NWP能够较好地实现0~6 h的闪电预报,具备比单纯使用多源观测数据、HNWP数据更好的预报结果。深度学习能够有效实现多源观测数据和NWP数据的融合,在2~6 h时效预报效果优于单独使用观测数据或NWP数据;预报时效越长,融合的优势体现的越明显。
英文摘要:
      It is a great challenge for the very short-range (VSR, 2-6 h) convective weather(CW) forecasting. On the hand, the extrapolation of observation data is no longer usable. On the other hand, the High-resolution Numerical Weather Prediction (HNWP)’s performance needs to be improved. For the above confusions, a semantic segmentation deep learning network, named LightningNet-NWP, was used to merge the multi-source observation data and HNWP data to get better VSR lightning forecasts. The predictors of LightningNet-NWP including lightning density, radar reflectivity, 6 infrared bands of Himawari-8, as well as the variable of radar composite reflectivity from GRAPES3km. Because the observation and HNWP data differs a lot, we designed two encode-decode symmetry sub-networks to extract the future in above two data sources. The pooling index is shared in upsampling process, so the details of shallow feature maps is transmitted and fully used. Three dimensional convolutional layers were utilized to extract the spatial and temporal features. The experimental results show that LightningNet-NWP could combine the observation and HNWP data effectively and make a good lightning prediction for next 0-6 hours. The performance of LightningNet-NWP with observation and HNWP data is much better than observation or HNWP data was used alone. The longer is the prediction, the advantage of combinational use of observation and HNWP data is larger.
查看全文   查看/发表评论  下载PDF阅读器
分享按钮