设为首页
加入收藏
Squall Line Identification Method Based on Convolution Neural Network
-
摘要: 为了探究深度学习用于飑线识别的可行性，基于2008—2020年河南省郑州和驻马店雷达数据，采用卷积神经网络（convolutional neural network，CNN）算法构建飑线识别模型，引用临界成功指数、公平风险评分、命中率和误判率定量评价模型的识别效果，对比不同样本组成比例和网络结构对飑线识别效果的影响。结果表明：建模所用的样本组成比例对飑线识别有一定影响，通过改变采样方式和优化网络结构均能够改善样本比例不平衡的问题，提高飑线识别效果，且后者提升的幅度更大，而两种方法的结合无明显提升。测试结果表明：该模型临界成功指数为0.66，公平风险评分为0.58，命中率为0.86，误判率为0.24。研究揭示了卷积神经网络能够提取并学习飑线和非飑线回波的图像特征，对飑线有一定识别能力。Abstract: Squall line often leads to heavy rain, gale and hail, which is a difficult key problem in nowcasting. In order to explore the feasibility of deep learning for squall line identification, the training, validation and test set sample sets are established based on the radar data of Zhengzhou and Zhumadian in Henan Province during 2008-2020. The convolutional neural network (CNN) algorithm is used to construct a squall line identification model. The critical success index (CSI), equitable threat score (ETS), hit rate (POD) and false positive rate (FAR) are used to quantitatively evaluate the identification effect of the model. The influence of different sample composition and network structure on squall line identification effect are compared. The results show that the composition ratio of sample is imbalanced, because squall line accounts for very small proportion in all kinds of weather processes. This imbalance will degrade the classification performance of the identification model to squall line samples. The imbalance of sample composition can be improved by changing sampling mode and optimizing network structure, both can improve the identification efficiency, especially the latter. However, the combination of the two methods does not bring further improvement. The over fitting problem in network training can be alleviated by increasing the sparsity and randomness of the network structure. The validation set shows that CSI is 0.87, ETS is 0.82, POD is 0.96, and FAR is 0.10. Based on the test set, the echo can be correctly identified by network as non-squall line in the weak stage of convection development, and as squall line in the strong stage of squall line development. The echo intensity and spatial distribution of squall line cases differ greatly, and the samples in the test set have the image features which are not included in the training set, and therefore the identification effect reduces. The test set show that CSI is 0.66, ETS is 0.58, POD is 0.86, and FAR is 0.24. The research reveals that CNN can extract and learn the image features of squall line echo, and it has a certain ability to identify squall line.
-
图 1 2018年6月26日郑州雷达探测飑线过程的识别结果
Fig. 1 Identification results of squall line detected by Zhengzhou radar on 26 Jun 2018
图 2 2019年6月3日驻马店雷达探测飑线过程的识别结果
Fig. 2 Identification results of squall line detected by Zhumadian radar on 3 Jun 2019
图 3 2020年6月24日郑州雷达探测飑线过程的识别结果
Fig. 3 Identification results of squall line detected by Zhengzhou radar on 24 Jun 2020
表 1 飑线过程发生时间
Table 1 Occurrence time of squall line processes
序号 时间 序号 时间 1 2008-05-09T06:00—11:00 12 2014-07-29T05:00—22:00 2 2008-06-03T07:00—11:00 13 2015-07-14T12:00—15:00 3 2009-05-16T12:00—15:00 14 2016-06-04T16:00—22:00 4 2009-06-03T11:00—16:00 15 2016-06-05T09:00—16:00 5 2009-06-14T11:00—16:00 16 2016-06-13T14:00—22:00 6 2011-07-26T08:00—16:00 17 2017-05-22T10:00—13:00 7 2013-06-02T10:00—14:00 18 2018-06-10T10:00—14:00 8 2013-07-04T06:00—13:00 19 2018-06-13T05:00—09:00 9 2013-07-31T20:00—24:00 20 2018-06-26T06:00—13:00 10 2013-08-01T08:00—16:00 21 2019-06-03T12:00—15:00 11 2013-08-11T09:00—18:00 22 2020-06-24T11:00—16:00 
下载: 导出CSV
表 2 飑线判别依据
Table 2 Criteria of squall line discrimination
相关文献 判别依据 文献[22] ①大于12 dBZ的回波带不小于150 km，持续时间超过5 h；②大于36 dBZ的回波带长宽比不小于3:1 文献[23] ①大于20 dBZ的回波带不小于100 km，持续时间超过4 h；②大于40 dBZ的回波带长宽比不小于5:1，持续时间超过2 h 文献[24] ①大于40 dBZ的回波带呈连续或准连续且不小于100 km，持续时间超过3 h；②线性或准线性对流共有一个回波前缘 文献[25] 大于35 dBZ的回波带长宽比大于5:1，且长度大于50 km 文献[9] ①由层云(15 dBZ)连接的对流云(40 dBZ)组成的系统长轴大于100 km，持续时间超过4 h；②系统长宽比大于5:1，持续时间超过2 h
下载: 导出CSV
表 3 数据集信息
Table 3 Information of datasets
数据集 飑线样本量 非飑线样本量 总样本量 训练集 1319 3035 4354 验证集 439 1011 1450 测试集 309 1075 1384
下载: 导出CSV
表 4 改进方案识别结果
Table 4 Identification result of different schemes
方案 CSI ETS POD FAR 原方案 0.59 0.47 0.77 0.27 过采样 0.65 0.52 0.80 0.26 降采样 0.74 0.63 0.93 0.23 网络结构优化 0.87 0.82 0.96 0.10 降采样及网络结构优化 0.88 0.83 0.97 0.10
下载: 导出CSV
表 5 训练集对比
Table 5 Comparison of training sets
训练集 飑线样本量 非飑线样本量 总样本量 原训练集 1319 3035 4354 过采样 2638 3035 5673 降采样 1319 2035 3354
下载: 导出CSV
表 6 验证集与测试集识别结果
Table 6 Identification results of validation set and test set
数据集 CSI ETS POD FAR 验证集 0.87 0.82 0.96 0.10 测试集 0.66 0.58 0.86 0.24
下载: 导出CSV
-
[1] 姚建群, 戴建华, 姚祖庆.一次强飑线的成因及维持和加强机制分析.应用气象学报, 2005, 16(6):746-752. doi: 10.3969/j.issn.1001-7313.2005.06.005Yao J Q, Dai J H, Yao Z Q. Case analysis of the formation and evolution of 12 July 2004 severe squall line. Journal of Applied Meteorological Science, 2005, 16(6): 746-752. doi: 10.3969/j.issn.1001-7313.2005.06.005 [2] 梁建宇, 孙建华. 2009年6月一次飑线过程灾害性大风的形成机制. 大气科学, 2012, 36(2): 316-336. https://www.cnki.com.cn/Article/CJFDTOTAL-DQXK201202010.htmLiang J Y, Sun J H. Formation mechanism of disastrous gale in a squall line process in June 2009. Chinese Journal of Atmospheric Sciences, 2012, 36(2): 316-336. https://www.cnki.com.cn/Article/CJFDTOTAL-DQXK201202010.htm [3] 罗琪, 郑永光, 陈敏. 2017年北京北部一次罕见强弓状飑线过程演变和机理. 气象学报, 2019, 77(3): 371-386. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXB201903001.htmLuo Q, Zheng Y G, Chen M. Evolution and mechanism of a rare strong bow squall line in northern Beijing in 2017. Acta Meteorologica Sinica, 2019, 77(3): 371-386. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXB201903001.htm [4] Rinehart R E, Garvey E T. Three-dimensional storm motion detection by conventional weather radar. Nature, 1978, 273(5660): 287-289. doi: 10.1038/273287a0 [5] Dixon M, Wiener G. TITAN: Thunderstorm identification, tracking, analysis, and nowcasting-A radar-based methodology. Journal of Atmospheric and Oceanic Technology, 1993, 10(6): 785-797. doi: 10.1175/1520-0426(1993)010<0785:TTITAA>2.0.CO;2 [6] Johnson J T, MacKeen P L, Witt A, et al. The storm cell identification and tracking algorithm: An enhanced WSR-88D algorithm. Weather and Forecasting, 1998, 13(2): 263-276. doi: 10.1175/1520-0434(1998)013<0263:TSCIAT>2.0.CO;2 [7] 韩丰, 沃伟峰. SWAN2.0系统的设计与实现. 应用气象学报, 2018, 29(1): 25-34. doi: 10.11898/1001-7313.20180103Han F, Wo W F. Design and implementation of SWAN2.0 platform. Journal of Applied Meteorological Science, 2018, 29(1): 25-34. doi: 10.11898/1001-7313.20180103 [8] 郑永光, 周康辉, 盛杰, 等. 强对流天气监测预报预警技术进展. 应用气象学报, 2015, 26(6): 641-657. doi: 10.11898/1001-7313.20150601Zheng Y G, Zhou K H, Sheng J, et al. Advances in techniques of monitoring, forecasting and warning of severe convective weather. Journal of Applied Meteorological Science, 2015, 26(6): 641-657. doi: 10.11898/1001-7313.20150601 [9] 杨吉, 刘黎平, 夏文梅, 等. 基于动态模板函数的线状中尺度对流系统自动识别. 气象, 2014, 40(11): 1389-1397. doi: 10.7519/j.issn.1000-0526.2014.11.012Yang J, Liu L P, Xia W M, et al. Automatic identification of linear mesoscale convective system based on dynamic template function. Meteorological Monthly, 2014, 40(11): 1389-1397. doi: 10.7519/j.issn.1000-0526.2014.11.012 [10] 程凌舟, 何建新, 曾宪军. 基于小波和Hu矩的飑线雷达回波识别. 成都信息工程大学学报, 2017, 32(4): 369-374. https://www.cnki.com.cn/Article/CJFDTOTAL-CDQX201704005.htmCheng L Z, He J X, Zeng X J. Squall line radar echo recognition based on Wavelet and Hu moment. Journal of Chengdu University of Information Technology, 2017, 32(4): 369-374. https://www.cnki.com.cn/Article/CJFDTOTAL-CDQX201704005.htm [11] 李哲, 李国翠, 刘黎平, 等. 飑线优化识别及雷暴大风分析. 高原气象, 2017, 36(3): 801-810. https://www.cnki.com.cn/Article/CJFDTOTAL-GYQX201703019.htmLi Z, Li G C, Liu L P, et al. Optimal identification of squall line and analysis of thunderstorm and gale. Plateau Meteorology, 2017, 36(3): 801-810. https://www.cnki.com.cn/Article/CJFDTOTAL-GYQX201703019.htm [12] 唐伟, 周勇, 董昊, 等. 我国气象领域应用人工智能技术的现状和国际对比. 气象科技进展, 2019, 9(5): 55-56;62. doi: 10.3969/j.issn.2095-1973.2019.05.009Tang W, Zhou Y, Dong H, et al. Current situation and international comparison of artificial intelligence technology in meteorological field in China. Advances in Meteorological Science and Technology, 2019, 9(5): 55-56;62. doi: 10.3969/j.issn.2095-1973.2019.05.009 [13] 黄小刚, 费建芳, 陈佩燕. 利用神经网络方法建立热带气旋强度预报模型. 应用气象学报, 2009, 20(6): 699-705. doi: 10.3969/j.issn.1001-7313.2009.06.007Huang X G, Fei J F, Chen P Y. A neural network approach to predict tropical cyclone intensity. Journal of Applied Meteorological Science, 2009, 20(6): 699-705. doi: 10.3969/j.issn.1001-7313.2009.06.007 [14] 韩丰, 杨璐, 周楚炫, 等. 基于探空数据集成学习的短时强降水预报试验. 应用气象学报, 2021, 32(2): 188-199. doi: 10.11898/1001-7313.20210205Han F, Yang L, Zhou C X, et al. An Experimental study of the short-time heavy rainfall event forecast based on ensemble learning and sounding data. Journal of Applied Meteorological Science, 2021, 32(2): 188-199. doi: 10.11898/1001-7313.20210205 [15] 张振华, 苗春生, 曾智华, 等. 一种人工神经网络云分类方法的改进与应用. 应用气象学报, 2012, 23(3): 355-363. doi: 10.3969/j.issn.1001-7313.2012.03.012Zhang Z H, Miao C S, Zeng Z H, et al. Improvement and application of artificial neural networks to cloud classification. Journal of Applied Meteorological Science, 2012, 23(3): 355-363. doi: 10.3969/j.issn.1001-7313.2012.03.012 [16] 张雪芬, 王志诚, 茆佳佳, 等. 微波辐射计温湿廓线反演方法改进试验. 应用气象学报, 2020, 31(4): 385-396. doi: 10.11898/1001-7313.20200401Zhang X F, Wang Z C, Mao J J, et al. Experiments on improving temperature and humidity profile retrieval for ground-based microwave radiometer. Journal of Applied Meteorological Science, 2020, 31(4): 385-396. doi: 10.11898/1001-7313.20200401 [17] 郑远攀, 李广阳, 李晔. 深度学习在图像识别中的应用研究综述. 计算机工程与应用, 2019, 55(12): 20-36. doi: 10.3778/j.issn.1002-8331.1903-0031Zheng Y P, Li G Y, Li Y. Survey of application of deep learning in image recognition. Computer Engineering and Applications, 2019, 55(12): 20-36. doi: 10.3778/j.issn.1002-8331.1903-0031 [18] 郭瀚阳, 陈明轩, 韩雷, 等. 基于深度学习的强对流高分辨率临近预报试验. 气象学报, 2019, 77(4): 715-727. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXB201904009.htmGuo H Y, Chen M X, Han L, et al. Experiment on high resolution near prediction of severe convection based on deep learning. Acta Meteorologica Sinica, 2019, 77(4): 715-727. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXB201904009.htm [19] 韩丰, 龙明盛, 李月安, 等. 循环神经网络在雷达临近预报中的应用. 应用气象学报, 2019, 30(1): 61-69. doi: 10.11898/1001-7313.20190106Han F, long M S, Li Y A, et al. Application of cyclic neural network in radar nowcasting. Journal of Applied Meteorological Science, 2019, 30(1): 61-69. doi: 10.11898/1001-7313.20190106 [20] Gu J, Wang Z, Kuen J, et al. Recent advances in convolutional neural networks. Pattern Recognition, 2018, 77: 354-377. doi: 10.1016/j.patcog.2017.10.013 [21] Zhang J, Howard K, Gourley J. Constructing three-dimensional multiple-radar reflectivity mosaics: Examples of convective storms and stratiform rain echoes. Journal of Atmospheric and Oceanic Technology, 2005, 22(1): 30-42. doi: 10.1175/JTECH-1689.1 [22] Chen G T J, Chou H C. General characteristics of squall lines observed in TAMEX. Monthly Weather Review, 1993, 121(3): 726-733. doi: 10.1175/1520-0493(1993)121<0726:GCOSLO>2.0.CO;2 [23] Geerts B. Mesoscale convective systems in the southeast United States during 1994-95: A survey. Weather and Forecasting, 1998, 13(3): 860-869. doi: 10.1175/1520-0434(1998)013<0860:MCSITS>2.0.CO;2 [24] Parker M D, Johnson R H. Organizational modes of midlatitude mesoscale convective systems. Monthly Weather Review, 2000, 128(10): 3413-3436. doi: 10.1175/1520-0493(2001)129<3413:OMOMMC>2.0.CO;2 [25] 俞小鼎. 多普勒天气雷达原理与业务应用. 北京: 气象出版社, 2006.Yu X D. Doppler Weather Radar Principle and Application. Beijing: China Meteorological Press, 2006. [26] 胡涛. 基于深度学习的鱼类识别研究. 杭州: 浙江工业大学, 2019.Hu T. Research on Fish Recognition Based on Deep Learning. Hangzhou: Zhejiang University of Technology, 2019. [27] Simonyan K, Zisserman A. Very Deep Convolutional Networks for Large-scale Image Recognition. arXiv preprint arXiv: 1409.1556, 2014. [28] Kingma D P, Ba J L. Adam: A Method for Stochastic Optimization. arXiv: 1412.6980, 2014. [29] 蔡金圻, 谭桂容, 牛若芸. 基于迁移CNN的江淮持续性强降水环流分型. 应用气象学报, 2021, 32(2): 233-244. doi: 10.11898/1001-7313.20210208Cai J Q, Tan G R, Niu R Y. Circulation pattern classification of persistent heavy rainfall in jianghuai region based on the transfer learning cnn model. Journal of Applied Meteorological Science, 2021, 32(2): 233-244. doi: 10.11898/1001-7313.20210208 [30] 陶新民, 郝思媛, 张冬雪, 等. 基于样本特性欠取样的不均衡支持向量机. 控制与决策, 2013, 28(7): 978-984. https://www.cnki.com.cn/Article/CJFDTOTAL-KZYC201307004.htmTao X M, Hao S Y, Zhang D X, et al. Unbalanced support vector machine based on sample characteristics under sampling. Control and Decision, 2013, 28(7): 978-984. https://www.cnki.com.cn/Article/CJFDTOTAL-KZYC201307004.htm [31] 李彦冬, 郝宗波, 雷航. 卷积神经网络研究综述. 计算机应用, 2016, 36(9): 2508-2515;2565. https://www.cnki.com.cn/Article/CJFDTOTAL-JSJY201609029.htmLi Y D, Hao Z B, Lei H. A review of convolutional neural networks. Journal of Computer Applications, 2016, 36(9): 2508-2515;2565. https://www.cnki.com.cn/Article/CJFDTOTAL-JSJY201609029.htm [32] Srivastava N, Hinton G, Krizhevsky A, et al. Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 2014, 15(1): 1929-1958. http://dl.acm.org/citation.cfm?id=2670313 [33] Goodfellow I J, Warde-Farley D, Mirza M, et al. Maxout Networks. arXiv preprint arXiv: 1302.4389, 2013. [34] Zeiler M D, Fergus R. Stochastic Pooling for Regularization of Deep Convolutional Neural Networks. arXiv preprint arXiv: 1301.3557, 2013. -
图(3) / 表(6)
计量
- 摘要浏览量: 1935
- HTML全文浏览量: 487
- PDF下载量: 249
- 被引次数: 0
