地球信息科学学报 ›› 2017, Vol. 19 ›› Issue (4): 475-485.doi: 10.3724/SP.J.1047.2017.00475

• 地理空间分析综合应用 • 上一篇    下一篇

耦合卡尔曼滤波和多层次聚类的中国PM2.5时空分布分析

刘鹏华1(), 姚尧1,2,*(), 梁昊3, 梁兆堂1, 张亚涛1, 王昊松1   

  1. 1. 中山大学地理科学与规划学院,广州 510275
    2. 中山大学 广东省城市化与地理环境空间模拟重点实验室,广州 510275
    3. 南京大学 江苏省地理信息技术重点实验室,南京 210023
  • 收稿日期:2016-07-01 修回日期:2016-11-01 出版日期:2017-04-20 发布日期:2017-04-20
  • 作者简介:

    作者简介:刘鹏华(1995-),男,本科生,研究方向为遥感与地理信息系统。E-mail:liuphhhh@foxmail.com

  • 基金资助:
    国家自然科学重点基金项目(41531176);国家自然科学基金项目(41671398、41601420)

Analyzing Spatiotemporal Distribution of PM2.5 in China by Integrating Kalman Filter and Multi-level Clustering

LIU Penghua1(), YAO Yao1,2,*(), LIANG Hao3, LIANG Zhaotang1, ZHANG Yatao1, WANG Haosong1   

  1. 1. School of Geography and Planning, Sun Yat-sen University, Guangzhou 510275, China
    2. Key Laboratory for Urbanization and Geo-simulation of Guangdong Province, Sun Yat-sen University, Guangzhou 510275, China
    3. Key Laboratory for Geographical Information Science and Technology of Jiangsu Province, Nanjing University, Nanjing 210023, China
  • Received:2016-07-01 Revised:2016-11-01 Online:2017-04-20 Published:2017-04-20
  • Contact: YAO Yao

摘要:

近年来,细颗粒物污染尤其是PM2.5受到人们越来越多的关注,研究PM2.5的时空分布规律也具有越来越重大的意义。传统的遥感反演方法模型复杂,且不能揭示近地表面的PM2.5分布规律。地面监测站的建设为PM2.5的研究提供了更实时的观测数据,但由于测量噪声的影响,观测数据存在不准确的极端异常值。为了揭示中国PM2.5的时空分布特征,本研究采用Kalman滤波对2015年中国338个城市的空气质量监测网络大数据进行最佳估计,并分析其时空特征。同时,根据中国各城市的PM2.5浓度的时序分布,采用基于DTW的K-Medoids聚类方法将其分为4个等级,并采用q统计量来评估PM2.5浓度分布的空间分层异质性。结果表明,采用Kalman滤波能有效去除数据噪声,峰值信噪比(PSNR)明显增大。在时空分布上,中国PM2.5时间分布曲线呈现“U”形,冬季PM2.5浓度明显高于夏季,且日变化曲线呈现“W”形;秋冬季PM2.5浓度的空间分层异质性非常显著,且空间分布呈现“双核分布”,重污染区主要分布在华北平原、新疆等地,西藏、广东、云南等地是稳定的空气质量优良区。

关键词: PM2.5, 大数据, 卡尔曼滤波, 时空分析, K-Medoids

Abstract:

Serious air pollution has recently aroused wide public concerns in China. The traditional method of quantitative remote sensing model is not only sophisticated but also inaccurate to fetch the exact PM2.5 data near the ground. Though the built-up ground monitoring stations can now provide sufficient PM2.5 observation data with high sampling frequency, there still exist many extreme outliers due to inevitable observation noise. Therefore, in this study, we adopted Kalman filter for optimal estimation of time-series of air quality data in 338 cities of China and comprehensively analyzed the spatiotemporal distribution pattern during the period of 2015. In our detailed analysis, we used DTW based K-Medoids clustering to classify cities into 4 levels according to their contamination degree, and utilized q statistic technique to evaluate the spatial stratified heterogeneity of PM2.5. The results show that by using Kalman filter, noise can be effectively reduced and value of PSNR can be significantly improved. In the study of temporal distribution, we found that PM2.5 followed a ‘U’ curve in yearly temporal distributions while daily temporal distributions obeyed a ‘W’ curve. PM2.5 density is much higher in winter than in summer in China, and spatial stratified heterogeneity is even more pronounced during the fall-winter stage. In the study of spatial distribution, it can be clearly seen that PM2.5 appears a ‘Dual-core’ pattern across China where concentration of PM2.5 spiked at Xinjiang and North China plain. In contrast, Xizang, Guangdong and Yunnan are more stable areas with excellent air quality, ranking first-tier nationwide.

Key words: PM2.5, big Data, kalman filter, spatiotemporal analysis, K-Medoids