张永轩,黄兴友,王雪婧,于华英,楚志刚

• 1:南京信息工程大学大气物理学院

摘要(Abstract)：

【目的】中尺度对流降水预报是天气预报的重点和难点之一，天气雷达探测的高时空分辨率降水数据是开展0～2 h临近预报的重要依据，对高分辨率中小流域雨洪预报具有重要意义。利用雷达数据开展对流降水临近预报为人们出行、农业生产指导、防灾减灾提供便利，在气象和水文领域具有实际应用价值。【方法】以广州新一代S波段天气雷达体扫资料为基础，采用基于Causal-LSTM记忆模块的条件可逆网络CrevNet开展对流降水临近预报的能力研究，并将模型的预报效果与基于普通ST-LSTM的模型进行对比分析，验证其优越性。为提升预报模型对强回波的记忆能力使用了带权重Huber损失。采用临界成功指数(CSI)、命中率(POD)、虚警率(FAR)评估了测试集在不同预报时效和检验阈值下的预测效果。此外，还采用峰值信噪比(PSNR)、图像结构相似性(SSIM)以及偏差评分(BIAS)检验新生对流回波的预测能力。【结果】结果显示：基于Causal-LSTM记忆单元的CrevNet模型在预报时段内CSI、POD较高，FAR较低。在两次对流个例的预测中，该模型在多个预测时效下具有较高的PSNR、SSIM以及更接近1的BIAS。【结论】研究表明：条件可逆网络CrevNet深度学习模型对于时空序列有较强的时空特征提取能力，搭配不同的卷积循环神经单元，预报效果会有不同；结合Causal-LSTM记忆单元的CrevNet模型能更好地保留对流回波形态，适用于对流降水临近预报。

关键词(KeyWords)： 对流临近预报;深度学习;CrevNet;Huber损失;雷达反射率;降水;气候变化;防灾减灾

Abstract:

Keywords:

基金项目(Foundation): 国家重点研发计划项目“重大自然灾害监测预警”课题(2018YFC1506102);; 山东省自然科学基金双偏振雷达资料在强对流天气中的应用研究(ZR2020MD053)

作者(Author): 张永轩,黄兴友,王雪婧,于华英,楚志刚

DOI: 10.13928/j.cnki.wrahe.2024.11.001

参考文献(References)：

• [1] 俞小鼎，郑永光.中国当代强对流天气研究与业务进展 [J].气象学报，2020,78 (3):391-418.YU Xiaoding,ZHENG Yongguang.Research and business progress of severe convective weather in contemporary China [J].Acta Meteor Sinica,2020,78 (3):391-418.
• [2] 陈晓娟，王咏青，毛璐，等.贺兰山区两次极端暴雨动力作用数值模拟分析[J].干旱区研究，2020,37(3):680-688.CHEN Xiaojuan,WANG Yongqing,MAO Lu,et al.Numerical simulation analysis of the dynamic action of two extreme rainstorms in Helan Mountain[J].Arid Zone Research,2020,37(3):680-688.
• [3] 崔春光，杜牧云，肖艳姣，等.强对流天气资料同化和临近预报技术研究 [J].气象，2021,47 (8):901-918.CUI Chunguang,DU Muyun,XIAO Yanjiao,et al.Research on data assimilation and nowcasting technology of severe convective weather [J].Meteorological Calendar,2021,47 (8):901-918.
• [4] 程丛兰，陈敏，陈明轩，等.临近预报的两种高时空分辨率定量降水预报融合算法的对比试验 [J].气象学报，2019,77 (4):701-714.CHENG Conglan,CHEN Min,CHEN Mingxuan,et al.Comparative experiment of two fusion algorithms for quantitative precipitation forecasting with high spatiotemporal resolution for nowcasting [J].Acta Meteor Sinica,2019,77 (4):701-714.
• [5] DOVIAK R J.Doppler radar and weather observations[M].Chelmsford,Massachusetts:Courier Corporation,2006.
• [6] 邸苏闯，李卓蔓，刘玉，等.基于气象雷达反演和云图外推法的临近期降雨预报方法研究[J].水利水电技术(中英文),2022,53(5):13-21.DI Suchuang,LI Zhuoman,LIU Yu,et al.Research on near-term rainfall forecasting method based on meteorological radar inversion and cloud map extrapolation method[J].Water Resources and Hydropower Engineering,2022,53(5):13-21.
• [7] 庄潇然，郑玉，王亚强，等.基于深度学习的融合降水临近预报方法及其在中国东部地区的应用研究 [J].气象学报，2023,81 (2):286-303.ZHUANG Xiaoran,ZHENG Yu,WANG Yaqiang,et al.Deep learning-based converged precipitation nowcasting method and its application in Eastern China [J].Acta Meteor Sinica,2023,81 (2):286-303.
• [8] 王永灿，魏加华，李琼，等.基于雷达回波的临近降水预报卷积循环神经网络模型研究 [J].水利水电技术(中英文),2023,54 (1):24-41.WANG Yongcan,WEI Jiahua,LI Qiong,et al.Research on convolutional recurrent neural network model for nowcasting precipitation prediction based on radar echo [J].Water Resources and Hydropower Engineering,2023,54 (1):24-41.
• [9] 曹雪健，许金玉，戚友存.“三道防线”建设赋能城市洪水预报：内在机理和技术路径[J].水利发展研究，2024,24(8):27-30.CAO Xuejian,XU Jinyu,QI Youcun empowering urban flood forecasting with the construction of “three lines of defense”:internal mechanism and technical path[J].Water Resources Development Research,2024,24 (8):27-30.
• [10] 袁凯，李武阶，李明，等.四种机器深度学习算法对武汉地区雷达回波临近预报的检验和评估 [J].气象，2022,48(4):428-441.YUAN Kai,LI Wujie,LI Ming,et al.Test and evaluation of four machine deep learning algorithms for radar echo nowcasting in Wuhan [J].Meteorological Journal,2022,48(4):428-441.
• [11] HOCHREITER S,SCHMIDHUBER J.Long short-term memory[J].Neural Computation,1997,9(8):1735-1780.
• [12] SHI X,CHEN Z,WANG H,et al.Convolutional LSTM network:A machine learning approach for precipitation nowcasting[C]// CORTES C,SUGIYAMA M,LEE D D,et al.NIPS′15:Proceedings of the 28th International Conference on Neural Information Processing Systems-Volume 1.Cambridge,MA,USA:MIT Press,2015:802-810.
• [13] SHI X,GAO Z,LAUSEN L,et al.Deep Learning for Precipitation Nowcasting:A Benchmark and A New Model[C]//ULRIKEL,ISABELLE G,SAMY B,et al.NIPS′17:Proceedings of the 31st International Conference on Neural Information Processing Systems.Long Beach,California,USA:Curran Associates Inc,2017:5622-5632.
• [14] WANG Y,LONG M,WANG J,et al.Predrnn:Recurrent neural networks for predictive learning using spatiotemporal lstms[C]//ULRIKE L,ISABELLE G,SAMY B et al.NIPS′17:Proceedings of the 31st International Conference on Neural Information Processing Systems.Long Beach California USA:Curran Associates Inc,2017:879-888.
• [15] WANG Y,GAO Z,LONG M,et al.Predrnn++:Towards a resolution of the deep-in-time dilemma in spatiotemporal predictive learning[C]//LUXBURG U.Advances in Neural Information Processing Systems 30 (NIPS 2017).Stockholm,Sweden:Neural Information Processing Systems Foundation Inc,2018:5110-5119.
• [16] WANG Y,WU H,ZHANG J,et al.Predrnn:A recurrent neural network for spatiotemporal predictive learning[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2022,45(2):2208-2225.
• [17] RAVURI S,LENC K,WILLSON M,et al.Skillful precipitation nowcasting using deep generative models of radar[J].Nature,2021,597(7878):672-677.
• [18] 黄兴友，马玉蓉，胡苏蔓.基于深度学习的天气雷达回波序列外推及效果分析 [J].气象学报，2021,79 (5):817-827.HUANG Xingyou,MA Yurong,HU Suman.Extrapolation and effect analysis of weather radar echo sequence based on deep learning [J].Acta Meteor Sinica,2021,79 (5):817-827.
• [19] 张秀再，李景轩，杨昌军，等.基于时空注意力门控循环单元的特征增强云图预测算法 [J].激光与光电子学进展，2023,60 (24):77-86.ZHANG Xiuzai,LI Jingxuan,YANG Changjun,et al.Feature-enhanced cloud map prediction algorithm based on spatio-temporal attention-gated recurrent unit [J].Advances in Laser and Optoelectronics,2023,60 (24):77-86.
• [20] 罗健文，邹茂扬，杨昊，等.面向降雨预报的雷达回波预测序列外推方法 [J].计算机应用研究，2024,41(4):1138-1142.LUO Jianwen,ZOU Maoyang,YANG Hao,et al.Extrapolation Method of Radar Echo Prediction Sequence for Rainfall Forecast[J].Computer Application Research,2024,41(4):1138-1142.
• [21] 方巍，沈亮，邹立尧，等.基于GCA-ConvLSTM预测网格的短临降水雷达回波外推方法[J].暴雨灾害，2023,42(4):427-436.FANG Wei,SHEN Liang,ZOU Liyao,et al.Short-term precipitation radar echo extrapolation method based on GCA-ConvLSTM prediction grid[J].Heavy rain disaster,2023,42(4):427-436.
• [22] LIN Z,LI M,ZHENG Z,et al.Self-attention convlstm for spatiotemporal prediction[C]//AAAI.Proceedings of the AAAI Conference on Artificial Intelligence.Washington,DC,USA:AAAI Press,2020,34(7):11531-11538.
• [23] GUEN V L,THOME N.Disentangling physical dynamics from unknown factors for unsupervised video prediction[C]// IEEE.Proceedings of the IEEE/CVF conference on computer vision and pattern recognition.Seattle,WA,USA:IEEE,2020:11474-11484.
• [24] FAN H,ZHU L,YANG Y.Cubic LSTMs for video prediction[C]//AAAI.Proceedings of the AAAI conference on artificial intelligence.Washington,DC,USA:AAAI Press,2019,33(1):8263-8270.
• [25] 叶玉玲.模糊粗糙神经网络的结构与参数优化[J].系统工程与电子技术，2009,31(12):2988-2993.YE Yuling.Structure and parameter optimization of fuzzy rough neural network[J].Systems Engineering and Electronic Technology,2009,31(12):2988-2993.
• [26] 马志峰，张浩，刘劼.基于深度学习的短临降水预报综述[J].计算机工程与科学，2023,45(10):1731-1753.MA Zhifeng,ZHANG Hao,LIU Jie.A review of short-term precipitation forecasting based on deep learning[J].Computer Engineering and Science,2023,45(10):1731-1753.
• [27] LIN X,WANG H,ZHANG Q,et al.A spatiotemporal network model for global ionospheric TEC forecasting[J].Remote Sensing,2022,14(7):1717.
• [28] 张佳洛.基于深度学习的强对流天气降水临近预报的方法研究[D].南京：南京理工大学，2020.ZHANG Jialuo.Research on precipitation nowcasting method of severe convective weather based on deep learning[D].Nanjing:Nanjing University of Science and Technology,2020.
• [29] DAVIS E W.The ecological approach to visual perception [J].Adapted Physical Activity Quarterly,2012,29 (1):98-99.
• [30] 滕志伟.基于深度学习的多普勒气象雷达回波外推算法研究[D].长沙：湖南师范大学，2017.TENG Zhiwei.Research on Doppler Weather Radar Echo Extrapolation Algorithm Based on Deep Learning[D].Changsha:Hunan Normal University,2017.
• [31] 朱丽娟，赵青兰.基于结构相似性的全参考图像质量评价 [J].信息通信，2013(1):1-2.ZHU Lijuan,ZHAO Qinglan.Quality Evaluation of Full Reference Image Based on Structural Similarity [J].Information and Communications Technology,2013(1):1-2.
• [32] ZHAO H,GALLO O,FROSIO I,et al.Loss functions for image restoration with neural networks [J].IEEE Transactions on Computational Imaging,2016,3(1):47-57.