Radar Echo Extrapolation Model Based on Deep Spatio-Temporal Fusion Neural Network

FANG Wei; PANG Lin; YI Weinan

doi:10.12263/DZXB.20211551

您当前的位置：

首页 >

文章列表页 >

Radar Echo Extrapolation Model Based on Deep Spatio-Temporal Fusion Neural Network

PAPERS | 更新时间：2025-12-08

- Radar Echo Extrapolation Model Based on Deep Spatio-Temporal Fusion Neural Network
- ACTA ELECTRONICA SINICA Vol. 51, Issue 9, Pages: 2526-2538(2023)
- 作者机构：
  
  1.南京信息工程大学计算机学院数字取证教育部工程研究中心，江苏南京 210044
  2.气象灾害国家重点实验室，北京 100081
  3.南京信息工程大学江苏省大气环境与装备技术协同创新中心，江苏南京 210044
  4.中国气象局气象干部培训学院，北京 100081
- 作者简介：
- 基金信息：
  
  National Natural Science Foundation of China(42075007);Open Grants of the State Key Laboratory of Severe Weather(2021LASWB19)
- DOI：10.12263/DZXB.20211551
  CLC： TP399
- Received：19 November 2021，
  
  Revised：2022-07-10，
  
  Published：25 September 2023
- 稿件说明：
移动端阅览
FANG Wei,PANG Lin.Radar Echo Extrapolation Model Based on Deep Spatio-Temporal Fusion Neural Network[J].ACTA ELECTRONICA SINICA,2023,51(09):2526-2538.
DOI：

FANG Wei,PANG Lin.Radar Echo Extrapolation Model Based on Deep Spatio-Temporal Fusion Neural Network[J].ACTA ELECTRONICA SINICA,2023,51(09):2526-2538. DOI： 10.12263/DZXB.20211551.

摘要

基于深度学习的雷达回波外推是实现短临降水预报的重要方法，由于雷达回波数据具有显著的非刚性的运动特征，使得数据的统计特性随时间在不断变化，意味着其具有高阶非平稳性，而现有深度学习方法难以捕捉回波序列的非刚性运动特征，且难以建模雷达数据的高阶非平稳性.为此，本文针对雷达数据特征提出了一种新的时空融合网络STUNNER（Spatio-Temporal Fusion Neural Network）.STUNNER设计了一种两路时空融合架构，通过交叉连接时间差分网络和时空轨迹网络实现高效的雷达回波外推.时间差分网络通过引入差分的思想提取高阶非平稳数据中平稳性特征来学习雷达回波的长期趋势，时空轨迹网络利用动态卷积将卷积循环神经网络中普通卷积固定的参数采样位置优化为随时间变化的动态位置来提取雷达回波的瞬时变化，同时采用两路融合策略将长期趋势与瞬时变化融合，实现长短时关联记忆.所提模型与其他四个模型在两个公开数据集上进行了实验对比.在雷达回波外推任务中当雷达反射率阈值为45 dBZ时，STUNNER在POD （Probability Of Detection）、CSI（Critical Success Index）、FAR（False Alarm Rate）上相比MIM（Memory In Memory）分别优化了0.020，0.023，0.043.实验结果表明新模型在处理雷达回波外推任务上具有更高的准确率.

Abstract

Radar echo extrapolation based on deep learning is an important method for precipitation nowcasting. Since radar echo data has significant non-rigid motion characteristics

the statistical characteristics of the data are constantly changing with time

which means that it has high-order non-stationarity. However

it is difficult for existing deep learning methods to capture the non-rigid motion characteristics of echo sequences and model the high-order non-stationarity of radar data. In this paper

we propose a new spatio-temporal fusion neural network (STUNNER) for radar data features

which designs a two-stream spatio-temporal fusion architecture to achieve efficient radar echo extrapolation by cross-connecting the temporal differencing network and the spatio-temporal trajectory network. The temporal differencing network learns the long-term trend of the radar echo by introducing the idea of difference to extract the stationary features in the high-order non-stationary data. The spatio-temporal trajectory network uses dynamic convolution to optimize the fixed parameter sampling position of conventional convolution in the convolutional recurrent neural network to a dynamic position that changes with time to extract the transient variation of radar echo. And the long-term trend and the transient variation are fused by a two-stream fusion strategy to realize long- and short-term association memory. The proposed model is experimentally compared with four other models on two public datasets. In the radar echo extrapolation task

when the radar reflectivity threshold is 45 dBZ

compared with MIM (Memory in Memory)

STUNNER has a 0.020 higher POD (Probability of Detection)

0.023 higher CSI (Critical Success Index) and 0.043 lower FAR (False Alarm Rate). The experimental results show that the innovative model has higher accuracy in dealing with the extrapolation of radar echoes.

关键词

Keywords

references

JI S W , XU W , YANG M , et al . 3D convolutional neural networks for human action recognition [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence , 2013 , 35 ( 1 ): 221 - 231 .

ZHOU Y Z , SUN X Y , LUO C , et al . Spatiotemporal fusion in 3D CNNs: A probabilistic view [C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway : IEEE , 2020 : 9826 - 9835 .

WANG L M , LI W , LI W , et al . Appearance-and-relation networks for video classification [C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2018 : 1430 - 1439 .

ZHOU Y Z , SUN X Y , ZHA Z J , et al . MiCT: Mixed 3D/2D convolutional tube for human action recognition [C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2018 : 449 - 458 .

SHI X J , CHEN Z R , WANG H , et al . Convolutional LSTM network: A machine learning approach for precipitation nowcasting [C]// Proceedings of the 28th International Conference on Neural Information Processing Systems - Volume 1 . Montreal : MIT Press , 2015 : 802 - 810 .

DÜZÇEKER A , GALLIANI S , VOGEL C , et al . DeepVideoMVS: Multi-view stereo on video with recurrent spatio-temporal fusion [C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway : IEEE , 2021 : 15319 - 15328 .

TULYAKOV S , LIU M Y , YANG X D , et al . MoCoGAN: Decomposing motion and content for video generation [C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2018 : 1526 - 1535 .

WANG L M , XIONG Y J , WANG Z , et al . Temporal segment networks: Towards good practices for deep action recognition [C]// European Conference on Computer Vision . Cham : Springer , 2016 : 20 - 36 .

SIMONYAN K , ZISSERMAN A . Two-stream convolutional networks for action recognition in videos [C]// Proceedings of the 27th International Conference on Neural Information Processing Systems - Volume 1 . Montreal : MIT Press , 2014 : 568 - 576 .

WANG Y B , ZHANG J J , ZHU H Y , et al . Memory in memory: A predictive neural network for learning higher-order non-stationarity from spatiotemporal dynamics [C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway : IEEE , 2020 : 9146 - 9154 .

WANG Y B , LONG M S , WANG J M , et al . PredRNN: Recurrent neural networks for predictive learning using spatiotemporal LSTMs [C]// Proceedings of the 31st International Conference on Neural Information Processing Systems . Long Beach : Curran Associates Inc , 2017 : 879 - 888 .

LIN Z H , LI M M , ZHENG Z B , et al . Self-attention ConvLSTM for spatiotemporal prediction [J]. Proceedings of the AAAI Conference on Artificial Intelligence , 2020 , 34 ( 7 ): 11531 - 11538 .

SHI X J , GAO Z H , LAUSEN L , et al . Deep learning for precipitation nowcasting: A benchmark and a new model [C]// Proceedings of the 31st International Conference on Neural Information Processing Systems . Long Beach : Curran Associates Inc , 2017 : 5617 - 5627 .

SONG K , YANG G W , WANG Q X , et al . Deep learning prediction of incoming rainfalls: An operational service for the city of Beijing China [C]// 2019 International Conference on Data Mining Workshops (ICDMW) . Piscataway : IEEE , 2020 : 180 - 185 .

TIAN L , LI X T , YE Y M , et al . A generative adversarial gated recurrent unit model for precipitation nowcasting [J]. IEEE Geoscience and Remote Sensing Letters , 2020 , 17 ( 4 ): 601 - 605 .

SØNDERBY C K , ESPEHOLT L , HEEK J , et al . MetNet: A neural weather model for precipitation forecasting [EB/OL]. ( 2020-03-24 )[ 2021-09-10 ]. https://arxiv.org/abs/2003.12140 https://arxiv.org/abs/2003.12140 .

JING J R , LI Q , PENG X , et al . HPRNN: A hierarchical sequence prediction model for long-term weather radar echo extrapolation [C]// ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . Piscataway : IEEE , 2020 : 4142 - 4146 .

RAVURI S , LENC K , WILLSON M , et al . Skilful precipitation nowcasting using deep generative models of radar [J]. Nature , 2021 , 597 ( 7878 ): 672 - 677 .

JADERBERG M , SIMONYAN K , ZISSERMAN A , et al . Spatial transformer networks [C]// Proceedings of the 28th International Conference on Neural Information Processing Systems - Volume 2 . Montreal : MIT Press , 2015 : 2017 - 2025 .

PATRAUCEAN V , HANDA A , CIPOLLA R . Spatio-temporal video autoencoder with differentiable memory [EB/OL]. ( 2015-11-25 )[ 2021-09-10 ]. https://arxiv.org/abs/1511.06309 https://arxiv.org/abs/1511.06309 .

KLEIN B , WOLF L , AFEK Y . A dynamic convolutional layer for short rangeweather prediction [C]// 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway : IEEE , 2015 : 4840 - 4848 .

BRABANDERE D B , JIA X , TUYTELAARS T , et al . Dynamic filter networks [M]// Advances in Neural Information Processing Systems . Montreal : MIT Press , 2016 : 667 - 675 .

DAI J F , QI H Z , XIONG Y W , et al . Deformable convolutional networks [C]// 2017 IEEE International Conference on Computer Vision (ICCV) . Piscataway : IEEE , 2017 : 764 - 773 .

SRIVASTAVA N , MANSIMOV E , SALAKHUTDINOV R . Unsupervised learning of video representations using LSTMs [C]// Proceedings of the 32nd International Conference on International Conference on Machine Learning-Volume 37 . New York : ACM , 2015 : 843 - 852 . .

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

No data

Related Author

YI Wei-nan

Related Institution

No data

⁰