南刚强,陈明轩. 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、追踪和分类方法[J]. 气象学报, (0):-, 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、追踪和分类方法
Identification, Tracking and Classification Method of Mesoscale Convective System Based on Radar composite reflectivity Mosaic and Deep Learning
投稿时间:2021-03-25  修订日期:2021-08-02
DOI:10.11676/qxxb2021.062
中文关键词:  深度学习,中尺度对流系统,识别,追踪,分类
英文关键词:Deep learning, Mesoscale convective system, Identifying, Tracking, Classification
基金项目:
作者单位邮编
南刚强 中国海洋大学 266100
陈明轩 北京城市气象研究院 100089
摘要点击次数: 15
全文下载次数: 18
中文摘要:
      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。[目的]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,这对灾害天气的分析和预报有着重要意义。[资料和方法]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)四种机器学习方法,研发了京津冀地区MCS的自动识别算法。其次,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。最后,在区分线状对流和非线状对流的基础上,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。[结果和结论] MCS的识别属于二分分类问题,以命中率(POD)、误警率(FAR)、关键成功指数(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。根据追踪轨迹计算单时刻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。
英文摘要:
      Mesoscale convective system is the main cause of a lot of convective weather, which can cause severe meteorological and hydrological disasters, such as thunderstorms, tornadoes, and flash flooding. Making accurate identification and tracking on MCS and realizing MCS’s classification based on the tracking trajectory and obtained MCS features attach great importance to analyze and forecast catastrophic weather. Based on the radar composite reflectivity mosaic data in the Beijing-Tianjin-Hebei region from 2010 to 2019, support vector machines (SVM), random forest (RF), Extreme Gradient Boosting (XGBoost) and deep neural network (DNN) are used to develop an automatic recognition algorithm of MCS in the Beijing-Tianjin-Hebei region. Secondly, the tracking and matching of identified MCS slices are completed according to spatiotemporal overlap tracking, and a tracking database of MCS is obtained, including intensity, spatial and temporal information. Finally, on the basis of distinguishing linear convection from non-linear convection, and starting from three conceptual models and structural characteristics of quasi-linear MCS that are classical trailing, leading, and parallel stratiform precipition, the classification algorithm of quasi-linear MCS is established by the area ratio of stratiform and intense convection on both sides of approximate major axis of MCS and the movement direction of MCS which is calculated according to tracking trajectory. The recognition of MCS is subsumed under binary classification. Taking POD, FAR, CSI and ACC as evaluation indexes, the DNN model is better than SVM, RF and XGBoost models in MCS recognition after comprehensive comparison. Spatiotemporal overlap tracking is used to track MCS slices identified by DNN model. The analyses about two tracking examples suggest that the algorithm used in this research has achieved good tracking results and also has further explained the accuracy and advantage of deep learning to identify MCS. The accurate realization of MCS’s classification including TS, LS and PS provides a technical idea for the life cycle prediction of quasi-linear MCS and the objective prediction of disaster-causing weather, especially the intensity, location and duration of short-term heavy precipitation by combining the movement direction of MCS at a single radar snashot with the distribution positions of stratiform and intense convection in MCS slices.
查看全文   查看/发表评论  下载PDF阅读器
分享按钮