Cloud Classification and Distribution of Cloud Types in Beijing Using Ka-Band Radar Data

Abstract

A cloud clustering and classification algorithm is developed for a ground-based Ka-band radar system in the vertically pointing mode. Cloud profiles are grouped based on the combination of a time-height clustering method and the k-means clustering method. The cloud classification algorithm, developed using a fuzzy logic method, uses nine physical parameters to classify clouds into nine types: cirrostratus, cirrocumulus, altocumulus, altostratus, stratus, stratocumulus, nimbostratus, cumulus or cumulonimbus. The performance of the clustering and classification algorithm is presented by comparison with all-sky images taken from January to June 2014. Overall, 92% of the cloud profiles are clustered successfully and the agreement in classification between the radar system and the all-sky imager is 87%. The distribution of cloud types in Beijing from January 2014 to December 2017 is studied based on the clustering and classification algorithm. The statistics show that cirrostratus clouds have the highest occurrence frequency (24%) among the nine cloud types. High-level clouds have the maximum occurrence frequency and low-level clouds the minimum occurrence frequency.

摘 要

云类是开展雷达数据云微物理参数精确反演的前提, 也是研究地区天气或气候分布特征的典型参数之一. 本文基于雷达探测特点, 提出了一种适用于地面Ka波段雷达垂直探测数据的云聚类及云分类算法. 所提的云聚类方法由时间-高度聚类法与k均值聚类法相结合, 实现对雷达廓线的云团检测及聚类; 云分类算法是基于模糊逻辑法, 利用9个云物理参数将云团分为九种类型: 卷层云, 卷积云, 高层云, 高积云, 层积云, 层云, 积云, 雨层云和积雨云. 文章利用2014年1-6月同期观测的全天空图像开展了对比分析, 研究结果表明92%的云廓线被成功聚类, 云分类平均一致性为87%. 基于所提云分类算法和雷达探测数据, 本文研究了北京市2014年1月至2017年12月的云类分布特征, 统计结果表明, 九种云类型中, 卷层云的出现频率最高达24%, 而低层云的出现频率最小.