留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于雷达组合反射率拼图和深度学习的中尺度对流系统识别、追踪与分类方法

南刚强 陈明轩 秦睿 韩雷 曹伟华

南刚强,陈明轩,秦睿,韩雷,曹伟华. 2021. 基于雷达组合反射率拼图和深度学习的中尺度对流系统识别、追踪与分类方法. 气象学报,79(6):1002-1021 doi: 10.11676/qxxb2021.062
引用本文: 南刚强,陈明轩,秦睿,韩雷,曹伟华. 2021. 基于雷达组合反射率拼图和深度学习的中尺度对流系统识别、追踪与分类方法. 气象学报,79(6):1002-1021 doi: 10.11676/qxxb2021.062
Nan Gangqiang, Chen Mingxuan, Qin Rui, Han Lei, Cao Weihua. 2021. Identification, tracking and classification method of mesoscale convective system based on radar composite reflectivity mosaic and deep learning. Acta Meteorologica Sinica, 79(6):1002-1021 doi: 10.11676/qxxb2021.062
Citation: Nan Gangqiang, Chen Mingxuan, Qin Rui, Han Lei, Cao Weihua. 2021. Identification, tracking and classification method of mesoscale convective system based on radar composite reflectivity mosaic and deep learning. Acta Meteorologica Sinica, 79(6):1002-1021 doi: 10.11676/qxxb2021.062

基于雷达组合反射率拼图和深度学习的中尺度对流系统识别、追踪与分类方法

doi: 10.11676/qxxb2021.062
详细信息
    作者简介:

    南刚强,主要从事机器学习算法应用研究。E-mail:gqnan_ouc@163.com

    通讯作者:

    陈明轩,主要从事短时临近天气预报研究。E-mail:mxchen@ium.cn

  • 中图分类号: P458.2

Identification, tracking and classification method of mesoscale convective system based on radar composite reflectivity mosaic and deep learning

  • 摘要: 中尺度对流系统(Mesoscale Convective System,MCS)是很多对流性天气的主要致灾体,可导致严重的气象和水文灾害,如雷暴大风、冰雹、龙卷风和山洪。对MCS进行准确的识别和追踪,并根据追踪轨迹及获得的MCS特征实现MCS的分类,对灾害天气的分析和预报有重要意义。基于京津冀地区2010—2019年的雷达组合反射率拼图资料,分别使用支持向量机(SVM)、随机森林(RF)、极度梯度提升决策树(XGBoost)和深度神经网络(DNN)4种机器学习方法,研发了京津冀地区MCS的自动识别算法。使用时、空重叠追踪法对识别的MCS进行追踪匹配,得到包含强度、时间和空间信息的MCS追踪数据资料。在区分线状对流系统和非线状对流系统的基础上,进一步从经典的尾随层云(Trailing Stratiform,TS)、前导层云(Leading Stratiform,LS)和平行层云(Parallel Stratiform,PS)三类准线性MCS的概念模型和结构特征出发,根据追踪轨迹计算MCS的运动方向和MCS近似长轴两侧层状云和强对流云的面积占比,建立准线性MCS的分类算法。MCS的识别属于二分分类问题,以命中率(POD)、虚警率(FAR)、临界成功指数(CSI)和准确率(ACC)为评价指标,综合对比各项指标发现DNN模型较SVM、RF和XGBoost模型对MCS的识别效果更好。使用时、空重叠追踪法对DNN模型识别的MCS进行追踪,结合对两个追踪实例的分析,发现本研究所用的算法取得了很好的追踪结果,也进一步说明了深度学习方法识别MCS的准确性和优势。根据追踪轨迹计算某时刻MCS的运动方向,结合识别的层状云和强对流云的分布位置,准确实现了TS、LS和PS型准线性MCS的分类,为准线性MCS的生命史预测及其致灾天气特别是短时强降水的强度、位置和持续时间的客观预报提供了一种技术思路。

     

  • 图 1  原始雷达拼图数据 (2014年6月17日11时59分36秒(世界时,下同))

    Figure 1.  Original radar mosaic data (11:59:36 UTC 17 June 2014)

    图 2  使用雷达拼图数据 (2014年6月17日 11时59分36 秒) 演示候选MCS切片的分割过程 (a. 包含强对流单元且面积大于40 km2的对流区域;b. 连接指定半径24 km内的对流区域,将主轴长度超过100 km的连接区域确认为MCS核;c.关联MCS核指定半径96 km内的层云区域得到候选MCS切片)

    Figure 2.  Demonstration of segmentation steps for candidate MCS slices using radar mosaic data (11:59:36 UTC 17 June 2014)(a. convection areas greater than 40 km2 with intense convection;b. connected convection area within a specified radius (24 km),and the connected area is considered to be the MCS core if its major axis length is at least 100 km;c. candidate MCS slice is identified by connecting the strtatiform pixels that are within the specified radius (96 km) of MCS core)

    图 3  MCS切片的凸包 (a) 和拟合椭圆 (b) 示意

    Figure 3.  Convex hull (a) and fitting ellipse (b) of MCS slice

    图 4  追踪过程示意 (N为当前时刻的MCS切片,S1S2为下一时刻的2个MCS切片)

    Figure 4.  Tracking process (N is a MCS slice at the current moment,S1 and S2 are the two MCS slices at the next moment)

    图 5  MCS正方向的定义 (a. k≥0,b. k<0;红色箭头为短轴的正方向)

    Figure 5.  Definition of the positive direction of MCS (a. k≥0,b. k<0;red arrow is the positive direction of the minor axis)

    图 6  2019年5月17日09时18分—15时MCS追踪轨迹

    Figure 6.  Tracking path of MCS during 09:18—15:00 UTC 17 May 2019

    图 7  2019年5月17日13时56分—14时30分原始雷达拼图数据 (a—f,时间间隔:6 min)

    Figure 7.  Original radar mosaic data at 13:56—14:30 UTC 17 May 2019 (a—f,interval:6 min)

    图 8  2019年5月17日13时56分—14时30分的MCS切片 (a—f,间隔: 6 min)

    Figure 8.  Display of MCS slices during 13:56—14:30 UTC 17 May 2019 (a—f,interval:6 min)

    图 9  2019年5月17日09时18分—15时MCS追踪路径 (已匹配)

    Figure 9.  Tracking path of MCS during 09:18—15:00 UTC 17 May 2019 (rematched)

    图 10  2019年7月13日13时42分—22时54分MCS追踪路径

    Figure 10.  Tracking path of MCS during 13:42—22:54 UTC 13 July 2019

    图 11  2019年7月13日13时42分—22时54分MCS追踪路径 (已匹配)

    Figure 11.  Tracking path of MCS during 13:42—22:54 UTC 13 July 2019 (rematched)

    图 12  2019年7月13日18时41分—19时11分 (a—f,间隔:6 min) 的原始雷达拼图数据

    Figure 12.  Original radar mosaic data during 18:41—19:11 UTC 13 July 2019 (a—f,interval:6 min)

    图 13  2019年7月13日18时41分—19时11分 (a—f,间隔:6 min) 的MCS切片展示 (b—e子图中有2个MCS切片)

    Figure 13.  Display of MCS slices during 18:41—19:11 UTC 13 July 2019 (a—f,interval:6 min)(there are two MCS slices in the b—e panels)

    图 14  2019年5月17日的LS型MCS雷达回波 (a. 12时41分,b. 12时47分,c. 12时53分,d. 13时05分)

    Figure 14.  Classified LS MCS radar reflectivity on 17 May 2019 (a. 12:41 UTC,b. 12:47 UTC,c. 12:53 UTC,d. 13:05 UTC)

    Continued

    图 15  2019年7月13日TS型MCS雷达回波 (a. 14时17分,b. 14时47分,c. 15时17分,d. 15时47分)

    Figure 15.  Classified TS MCS radar reflectivity on 13 July 2019 (a. 14:17 UTC,b. 14:47 UTC,c. 15:17 UTC,d. 15:47 UTC)

    图 16  2019年7月25日PS型MCS雷达回波 (a. 05时47分,b. 06时11分,c. 06时41分,d .07时05分)

    Figure 16.  Classified PS MCS radar reflectivity on 25 July 2019 (a. 05:47 UTC,b. 06:11 UTC,c. 06:41 UTC,d. 07:05 UTC)

    表  1  DNN模型主要参数

    Table  1.   Main parameters of the DNN model

    DNN
    输入层节点14
    输出层节点2
    激活函数Relu
    优化器GradientDescent
    下载: 导出CSV

    表  2  用于分割雷达拼图中MCS的指标阈值

    Table  2.   Various thresholds used to segment MCS in radar mosaic data

    指标名称阈值
    层状云(dBz) 20
    对流(dBz) 40
    强对流(dBz) 50
    对流区域面积(km2 40
    MCS核长度(km)100
    对流区域搜索半径(km) 24
    层状云区域搜索半径(km) 96
    下载: 导出CSV

    表  3  MCS样本特征列表

    Table  3.   Sample features of MCS

    特征定义
       面积特征面积切片总面积(km2
    强对流面积超过强对流阈值的像素所覆盖面积(km2
    对流面积超过对流阈值的像素所覆盖面积(km2
       比值特征面积-凸包面积比值总面积与凸包面积之比
    强对流-层状云比值强对流面积与总面积之比
    强对流-对流比值强对流面积与对流面积之比
    对流-层状云比值对流面积与总面积之比
       几何特征椭圆长轴长度切片最佳拟合椭圆的长轴长度(km)
    椭圆短轴长度切片最佳拟合椭圆的短轴长度(km)
    椭圆离心率切片最佳拟合椭圆的离心率
    凸包面积切片的外接多边形所覆盖的面积(km2
       统计特征方差切片区域像素值的方差
    平均值切片区域像素值的均值(dBz)
    最大值切片区域像素值的最大值(dBz)
    下载: 导出CSV

    表  4  不同类别和年份的训练集和测试集样本数

    Table  4.   Training and testing counts by classification and year

    年份MCS样本数non-MCS样本数
    训练集2010741635
    20111606903
    201213671018
    201322781070
    20149651042
    2015516323
    201621451848
    201714451644
    总计110638483
    测试集20189671737
    201917651211
    总计27322948
    下载: 导出CSV

    表  5  TS、LS和PS型MCS的分类规则

    Table  5.   MCS classification rules for TS, LS and PS

    RI≥threRI≤1/threRS≈1 且
    1/thre<RI<thre
    θ(0,90°)TSLSPS
    θ(−90°,0)LSTS
    下载: 导出CSV

    表  6  预测和实际标签的混淆矩阵

    Table  6.   Confusion matrix for predictions and actual labels

    实际
    MCSnon-MCS
    预测MCSTPFP
    non-MCSFNTN
    下载: 导出CSV

    表  7  SVM、RF、XGBoost和DNN模型在测试集上的混淆矩阵

    Table  7.   Confusion matrix of the SVM,RF,XGBoost and DNN models on testing set

    TPFNFPTN
    SVM24073253032645
    RF24702624082540
    XGBoost24922404092539
    DNN24283042902658
    下载: 导出CSV

    表  8  SVM、RF、XGBoost和DNN模型在测试集上的评分

    Table  8.   Scores of the SVM,RF,XGBoost and DNN models on testing set

    CSIPODFARACC
    SVM0.79310.88100.11180.8894
    RF0.78660.90410.14180.8820
    XGBoost0.79340.91220.14100.8857
    DNN0.80340.88870.10670.8953
    下载: 导出CSV

    表  9  2018和2019年MCS切片中TS、LS和PS型的个数统计

    Table  9.   Numbers of TS,LS and PS in MCS slices in 2018 and 2019

    年份20182019总计
    PS415798
    LS266389
    TS112356468
    下载: 导出CSV

    表  10  分类出的LS、TS和PS型准线性MCS所对应的RSRIθ的计算值 (比值的分母为0时用−9999.000表示计算值;此处只选择了3个时间段)

    Table  10.   Calculated values of RSRI and θ,which correspond to the classified LS,TS and PS of Quasi-linear MCSs (−9999.000 is used to represent their values when the denominator of RS and RI is 0,only three time periods are selected here)

    日期 时间(UTC)RSRIθMCS类型日期 时间(UTC)RSRIθMCS类型
    20190517 12:411.161 0.00266.626LS20190713 15:230.711880.000 32.700TS
    20190517 12:471.217 0.01618.913LS20190713 15:470.661 53.773 3.644TS
    20190517 12:531.223 0.00213.530LS20190713 15:590.703−9999.000 7.494TS
    20190517 12:591.195 0.01727.797LS20190725 05:470.979 1.614 64.397PS
    20190517 13:051.195 0.02352.510LS20190725 05:590.963 1.749−73.618PS
    20190713 14:170.669 81.25084.425TS20190725 06:110.853 2.461 39.279PS
    20190713 14:290.625322.50050.068TS20190725 06:290.860 5.632 33.337PS
    20190713 14:470.659344.50015.620TS20190725 06:410.834 3.686 12.589PS
    20190713 14:590.701 43.824 3.385TS20190725 07:050.876 1.827−49.447PS
    20190713 15:110.679−9999.000 9.676TS
    下载: 导出CSV
  • [1] 曹伟华,陈明轩,高峰等. 2019. 雷暴区域追踪矢量与雷暴单体追踪矢量融合临近预报研究. 气象学报,77(6):1015-1027

    Cao W H,Chen M X,Gao F,et al. 2019. A vector blending study based on object-based tracking vectors and cross correlation tracking vectors. Acta Meteor Sinica,77(6):1015-1027 (in Chinese)
    [2] 陈明轩,俞小鼎,谭晓光等. 2006. 北京2004年“7.10”突发性对流强降水的雷达回波特征分析. 应用气象学报,17(3):333-345 doi: 10.3969/j.issn.1001-7313.2006.03.011

    Chen M X,Yu X D,Tan X G,et al. 2006. Radar echoes characteristics of the sudden convective rainstorm over Beijing area on July 10,2004. J Appl Meteor Sci,17(3):333-345 (in Chinese) doi: 10.3969/j.issn.1001-7313.2006.03.011
    [3] 陈明轩,高峰,孔荣等. 2010. 自动临近预报系统及其在北京奥运期间的应用. 应用气象学报,21(4):395-404 doi: 10.3969/j.issn.1001-7313.2010.04.002

    Chen M X,Gao F,Kong R,et al. 2010. Introduction of auto-nowcasting system for convective storm and its performance in Beijing olympics meteorological service. J Appl Meteor Sci,21(4):395-404 (in Chinese) doi: 10.3969/j.issn.1001-7313.2010.04.002
    [4] 韩雷,郑永光,王洪庆等. 2007. 基于数学形态学的三维风暴体自动识别方法研究. 气象学报,65(5):805-814

    Han L,Zheng Y G,Wang H Q,et al. 2007. 3D storm automatic identification based on mathematical morphology. Acta Meteor Sinica,65(5):805-814 (in Chinese)
    [5] 雷蕾,邢楠,周璇等. 2020. 2018年北京“7.16”暖区特大暴雨特征及形成机制研究. 气象学报,78(1):1-17

    Lei L,Xing N,Zhou X,et al. 2020. A study on the warm-sector torrential rainfall during 15—16 July 2018 in Beijing area. Acta Meteor Sinica,78(1):1-17 (in Chinese)
    [6] 王晓芳,崔春光. 2011. 暴雨中尺度对流系统研究的若干进展. 暴雨灾害,30(2):97-106

    Wang X F,Cui C G. 2011. A number of advances of the research on heavy rain mesoscale convective systems. Torrential Rain Disaster,30(2):97-106 (in Chinese)
    [7] 杨吉,郑媛媛,夏文梅等. 2015. 雷达拼图资料上中尺度对流系统的跟踪与预报. 气象,41(6):738-744 doi: 10.7519/j.issn.1000-0526.2015.06.008

    Yang J,Zheng Y Y,Xia W M,et al. 2015. Mesoscale convective systems (MCSs) tracking and nowcasting based on radar mosaic data. Meteor Mon,41(6):738-744 (in Chinese) doi: 10.7519/j.issn.1000-0526.2015.06.008
    [8] Ashley W S,Haberlie A M,Strohm J. 2019. A climatology of quasi-linear convective systems and their hazards in the United States. Wea Forecast,34(6):1605-1631 doi: 10.1175/WAF-D-19-0014.1
    [9] Bengio Y. 2009. Learning deep architectures for AI. Found Trends Mach Learn,2(1):1-127 doi: 10.1561/2200000006
    [10] Davis C,Brown B,Bullock R. 2006a. Object-based verification of precipitation forecasts. Part Ⅰ:Methodology and application to mesoscale rain areas. Mon Wea Rev,134(7):1772-1784 doi: 10.1175/MWR3145.1
    [11] Davis C,Brown B,Bullock R. 2006b. Object-based verification of precipitation forecasts. Part Ⅱ:Methodology and application to convective rain systems. Mon Wea Rev,134(7):1785-1795 doi: 10.1175/MWR3146.1
    [12] Dixon M,Wiener G. 1993. TITAN:Thunderstorm identification, tracking,analysis,and nowcasting:A radar-based methodology. J Atmos Ocean Technol,10(6):785-797 doi: 10.1175/1520-0426(1993)010<0785:TTITAA>2.0.CO;2
    [13] Haberlie A M,Ashley W S. 2018a. A method for identifying midlatitude mesoscale convective systems in radar mosaics. Part Ⅰ:Segmentation and classification. J Appl Meteor Climatol,57(7):1575-1598 doi: 10.1175/JAMC-D-17-0293.1
    [14] Haberlie A M,Ashley W S. 2018b. A method for identifying midlatitude mesoscale convective systems in radar mosaics. Part Ⅱ:Tracking. J Appl Meteor Climatol,57(7):1599-1621 doi: 10.1175/JAMC-D-17-0294.1
    [15] Han L,Fu S X,Zhao L F,et al. 2009. 3D convective storm identification,tracking,and forecasting:An enhanced TITAN algorithm. J Atmos Ocean Technol,26(4):719-732 doi: 10.1175/2008JTECHA1084.1
    [16] Houze R A Jr. 2018. 100 years of research on mesoscale convective systems. Meteor Monogr,59(1):17.1-17.54
    [17] Jergensen G E,McGovern A,Lagerquist R,et al. 2020. Classifying convective storms using machine learning. Wea Forecast,35(2):537-559
    [18] Johnson J T,MacKeen P L,Witt A,et al. 1998. The storm cell identification and tracking algorithm:An enhanced WSR-88D algorithm. Wea Forecast,13(2):263-276 doi: 10.1175/1520-0434(1998)013<0263:TSCIAT>2.0.CO;2
    [19] Lakshmanan V,Hondl K,Rabin R. 2009. An efficient,general-purpose technique for identifying storm cells in geospatial images. J Atmos Ocean Technol,26(3):523-537 doi: 10.1175/2008JTECHA1153.1
    [20] Mueller C,Saxen T,Roberts R,et al. 2003. NCAR auto-nowcast system. Wea Forecast,18(4):545-561 doi: 10.1175/1520-0434(2003)018<0545:NAS>2.0.CO;2
    [21] Parker M D,Johnson R H. 2000. Organizational modes of midlatitude mesoscale convective systems. Mon Wea Rev,128(10):3413-3436 doi: 10.1175/1520-0493(2001)129<3413:OMOMMC>2.0.CO;2
    [22] Pedregosa F,Varoquaux G,Gramfort A,et al. 2011. Scikit-learn:Machine learning in python. J Mach Learn Res,12:2825-2830
    [23] Rinehart R E,Garvey E T. 1978. Three-dimensional storm motion detection by conventional weather radar. Nature,273(5660):287-289 doi: 10.1038/273287a0
    [24] Schumacher R S,Johnson R H. 2005. Organization and environmental properties of extreme-rain-producing mesoscale convective systems. Mon Wea Rev,133(4):961-976 doi: 10.1175/MWR2899.1
    [25] Schumacher R S,Johnson R H. 2006. Characteristics of U. S. extreme rain events during 1999–2003. Wea Forecast,21(1):69-85 doi: 10.1175/WAF900.1
    [26] Schumacher R S,Rasmussen K L. 2020. The formation,character and changing nature of mesoscale convective systems. Nat Rev Earth Environ,1(6):300-314 doi: 10.1038/s43017-020-0057-7
    [27] Skok G,Tribbia J,Rakovec J,et al. 2009. Object-based analysis of satellite-derived precipitation systems over the low-and midlatitude Pacific Ocean. Mon Wea Rev,137(10):3196-3218 doi: 10.1175/2009MWR2900.1
    [28] van der Walt S,Schönberger J L,Nunez-Iglesias J,et al. 2014. Scikit-image:Image processing in python. Peer J,2:e453 doi: 10.7717/peerj.453
    [29] Wang X F,Cui C G,Cui W J,et al. 2014. Modes of mesoscale convective system organization during Meiyu season over the Yangtze River basin. Acta Meteor Res,28(1):111-126 doi: 10.1007/s13351-014-0108-4
    [30] Zheng A. 2015. Evaluating Machine Learning Models:A Beginner's Guide to Key Concepts and Pitfalls. O'Reilly Media,Inc:8-9
  • 加载中
图(17) / 表(10)
计量
  • 文章访问数:  207
  • HTML全文浏览量:  51
  • PDF下载量:  138
  • 被引次数: 0
出版历程
  • 收稿日期:  2021-03-29
  • 修回日期:  2021-08-23
  • 网络出版日期:  2021-11-08
  • 刊出日期:  2021-12-27

目录

    /

    返回文章
    返回