引用本文: 段赛男, 焦瑞莉, 吴成来. 2024. 一种基于K-means聚类算法的沙尘天气客观识别方法[J]. 气候与环境研究, 29(2): 178−192.
DUAN Sainan, JIAO Ruili, WU Chenglai. 2024. Objective Identification Method for Dust Weather Based on the K-means Clustering Algorithm [J]. Climatic and Environmental Research (in Chinese), 29 (2): 178−192.
 Citation: DUAN Sainan, JIAO Ruili, WU Chenglai. 2024. Objective Identification Method for Dust Weather Based on the K-means Clustering Algorithm [J]. Climatic and Environmental Research (in Chinese), 29 (2): 178−192.

## Objective Identification Method for Dust Weather Based on the K-means Clustering Algorithm

• 摘要: 鉴于以往基于污染物浓度时间序列进行分析的沙尘天气识别方法在判断标准上存在一定的主观性，本文提出一种基于K-means聚类算法的沙尘天气客观识别方法。本方法利用环境监测总站的PM2.5和PM10小时浓度资料进行聚类，首先选取最优的分类数目K进行聚类，其次对聚类结果中离散程度较高的类别进行再次聚类，直到无需分类。将本方法应用于西安市2018年2～4月沙尘天气的识别中，结果表明，本方法可有效识别主要沙尘天气。此外，利用本方法可得到沙尘天气典型特征：PM2.5占PM10浓度的比例小于43.5%、PM10浓度高于228 μg/m3，符合沙尘天气期间PM10浓度较高且以粗颗粒物为主的物理特征。总体上看，本方法物理基础清晰，可操行性强，适用于大规模数据处理，具有较好的实用价值和应用前景。

Abstract: Time-series analysis methods have been developed previously to identify dust weather based on pollutant concentrations; however, the criteria used are subject to considerable uncertainties. Therefore, herein, we propose an objective identification method for dust weather. This method is based on the K-means clustering algorithm and involves using hourly concentrations of PM2.5 and PM10 obtained from environmental monitoring stations. The flow path of this method involves first selecting the optimal number of classifications K for cluster analysis, followed by classifying the cluster groups that show substantial scattering in the distribution of PM2.5 and PM10 concentrations until no further classification is needed. Further, this method is applied to identify the dust weather in Xi'an from February to April 2018. The results show that this method can effectively identify the main dust weather events. Based on this method, typical characteristics of dust weather can be obtained: The ratio of PM2.5 to PM10 concentration being less than 43.5%, and the PM10 concentration being greater than 228 μg/m3, which is consistent with physical characteristics that the PM10 concentration is high and mainly consists of coarse particles during the dust event. Overall, this method has a clear physical basis; it is also easy to operate, suitable for massive data processing, and promising for applications in relevant areas.

/

• 分享
• 用微信扫码二维码

分享至好友和朋友圈