耦合卡尔曼滤波和多层次聚类的中国PM2.5时空分布分析
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
收稿日期: 2016-07-01
要求修回日期: 2016-11-01
网络出版日期: 2017-04-20
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
Received date: 2016-07-01
Request revised date: 2016-11-01
Online published: 2017-04-20
Copyright
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。
刘鹏华 , 姚尧 , 梁昊 , 梁兆堂 , 张亚涛 , 王昊松 . 耦合卡尔曼滤波和多层次聚类的中国PM2.5时空分布分析[J]. 地球信息科学学报, 2017 , 19(4) : 475 -485 . DOI: 10.3724/SP.J.1047.2017.00475
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
Fig.1 Air quality monitoring stations of China图1 中国空气质量监测站点分布图 |
Fig.2 Research workflow图2 研究流程图 |
Fig.3 Pseudo code of Kalman filter图3 Kalman滤波伪代码 |
Tab.1 PSNR value before and after Kalman filtering表1 Kalman滤波前后PSNR值对比表 |
地区 | 滤波前 | 滤波后 |
---|---|---|
北京 | 12.8132 | 21.0545 |
上海 | 9.7523 | 14.9660 |
广州 | 13.2474 | 19.1636 |
南京 | 17.9053 | 27.0554 |
Fig.4 Quarterly and monthly average concentrations of PM2.5 in China图4 中国PM2.5季度和月度平均浓度 |
Fig.5 Hourly concentration of PM2.5 in China图5 中国PM2.5逐小时浓度 |
Fig.6 Spatial distribution of average PM2.5 concentration in China in 2015图6 2015年中国PM2.5平均浓度空间分布图 |
Fig.7 Spatial distribution of PM2.5 concentration in China in 2015图7 中国2015年1-12月PM2.5浓度空间分布图 |
Fig.8 The variation of Silhouette values with the changes in number of categories图8 Silhouette值随类别数目变化图 |
Fig.9 Features distribution of clustering centers图9 聚类中心特征分布图 |
Fig.10 q statistic distribution图10 q 统计量分布图 |
Fig.11 Clustering results of PM2.5 concentrations of cities based on K-Medoids图11 基于K-Medoids的城市PM2.5浓度聚类结果图 |
The authors have declared that no competing interests exist.
[1] |
[
|
[2] |
|
[3] |
|
[4] |
|
[5] |
|
[6] |
|
[7] |
|
[8] |
|
[9] |
|
[10] |
[
|
[11] |
[
|
[12] |
[
|
[13] |
[
|
[14] |
[
|
[15] |
|
[16] |
|
[17] |
[
|
[18] |
[
|
[19] |
|
[20] |
|
[21] |
|
[22] |
|
[23] |
|
[24] |
|
[25] |
|
[26] |
|
[27] |
|
[28] |
|
[29] |
[
|
[30] |
|
[31] |
[
|
[32] |
|
[33] |
|
[34] |
[
|
[35] |
|
[36] |
|
[37] |
|
[38] |
[
|
[39] |
[
|
[40] |
[
|
[41] |
|
[42] |
[
|
[43] |
[
|
[44] |
[
|
[45] |
|
[46] |
|
[47] |
[
|
[48] |
[
|
[49] |
[
|
/
〈 |
|
〉 |