Routes and clustering features of PM2.5 spillover within the Jing-Jin-Ji region at multiple timescales identified using complex network-based methods

Highlights

• A novel perspective for the regional joint control of air pollution.

• PM2.5 spillover networks at multiple timescales were constructed.

• An integrated method of discrete wavelet transform, GARCH-BEKK and complex networks.

• Jing-Jin-Ji cities are divided into two to three clusters under different timescales.

• Interaction of the PM2.5 spillover behavior within Jing-Jin-Ji is active for one week.

Abstract

As a resource-dependent local economy and heavy industry cluster, the Jing-Jin-Ji region has become China's largest source of air contaminant emissions and experiences frequent pollutant spillovers. integration process. One of the most urgent steps in the integration development process in Jing-Jin-Ji is achieving joint regional air pollution control in the cities comprising the region. Thus, this paper aims to provide a decision-making foundation for the collaborative governance of Jing-Jin-Ji cities with respect to PM2.5. To achieve this goal, spillover networks at multiple timescales were constructed using discrete wavelet transform, the GARCH-BEKK model and complex networks. The results showed that the interaction of the PM2.5 spillover behavior within Jing-Jin-Ji is notably strong and that the spillover relationships are normally active for up to approximately one week. The Shijiazhuang and Baoding cities require more control due to their wide PM2.5 spillover range to other cities, such as Beijing and Chengde. Moreover, the PM2.5 spillover routes differ at different timescales. In the short term, the initial cities of spillover routes, e.g., the “Cangzhou–Tianjin-Handan–Beijing” route, are the critical control cities for the government, and the cities along the routes should take advanced measures to prevent bad air conditions when the former cities have heavy haze days. In addition, the Jing-Jin-Ji cities are divided into two to three clusters, and cities in the same cluster are mostly adjacent to each other. Therefore, the most cost-effective method for achieving joint regional air pollution control in Jing-Jin-Ji cities is to treat the cities in the same clusters as a whole. This study proposes a novel perspective for the regional joint control of air pollution based on complex networks, an approach that can be used to holistically and cost-effectively select areas of collaborative governance.

Introduction

In 2014, the Chinese government promulgated a policy of integrated development of Jing-Jin-Ji cities, which include Beijing, Tianjin and Hebei. Due to the severe air pollution in these cities, particularly after the frequent observation of smog in Beijing and its surrounding areas in 2013, the most urgent step in this integrated development process is achieving joint regional air pollution control in Jing-Jin-Ji cities (Zhang et al., 2016). The air pollution in Jing-Jin-Ji cities is caused by several factors, such as the coal consumption (Ji et al., 2017; Sun et al., 2018) and biomass burning (Wang et al., 2015). Moreover, air pollution is transmitted among Jing-Jin-Ji cities by wind (Jiang et al., 2015), aerosol and meteorological variables (Gao et al., 2015). Although joint regional air pollution control could be more cost-effective (Wu et al., 2015), this does not necessarily mean that we should jointly control all 13 Jing-Jin-Ji cities at the same time. Is these a more effective way to achieve joint control of regional air pollution in this region? Are there any specific routes and clustering features of air pollution spillover among Jing-Jin-Ji cities, particularly at multiple timescales? Answers to these questions will provide a cost-effective solution.

Due to the adverse influences of haze (Zhang et al., 2018) and the easy movement of contaminants, particularly the PM2.5 concentration, the spillover of PM2.5 has become a hot topic. Yan et al. (2018), Cheng et al. (2017) and Zhang and Gong (2018) investigated the temporal and spatial characteristics of PM2.5. Wu et al. (2017) and Liu et al. (2017c) identified the natural and anthropogenic factors of PM2.5, and Du et al. (2018), Ma et al. (2016) and Jiang et al. (2018) emphasized a single factor associated with the spillover of PM2.5. However, few studies have focused on the whole spillover effects, which include all the factors that might influence PM2.5 spillover among cities or regions. Studying the spillover effect of PM2.5 can not only elucidate the routes and clustering features of PM2.5 spillover among cities from a holistic perspective but also identify early warning measures for cities when related cities have abnormal air conditions and provide a decision-making foundation for the collaborative governance of Jing-Jin-Ji cities with respect to PM2.5. Using North China as a model, Yang et al. (2018) revealed that North China exported a large amount of embodied PM2.5 to other domestic provinces, mainly China's central coastal area and the Beijing-Tianjin region, but neglected the time attribute of PM2.5 spillover because the transmission of PM2.5 from city A to city B might experience a time lag (Feng et al., 2015; Wang and Dai, 2016). In addition, hidden time-frequency information in a time series can be discovered through multi-timescales analysis. Therefore, the multi-timescale spillover of PM2.5 among cities should be studied. The use of multiple timescales is a relatively new but actively used approach in time-series research. Scholars usually analyze the relationships in multiple time series based on the original information, but as with the wavelets theory used in time-series research, scholars have found that the spillover relationship can also show different characteristics in a multi-timescale period (Liu et al., 2017a). Wavelets are widely used in different fields, such as weapons detection and military intelligence (Hussein and Hu, 2016), medical diagnoses (Koh et al., 2017), and seismic data processing (Liu et al., 2016). In recent years, wavelet transform, particularly discrete wavelet transform (DWT), has been widely used in time-series research. Scholars have used the DWT method to capture multi-timescale information from original time series and their effect on other time series; for example, Huang et al. (2017) used the DWT method to transform oil prices into multiple time horizons and later analyzed their effects on the stock market from a multi-scales perspective. Moreover, increasing numbers of scholars are examining the relationships in multiple time series at multiple timescales; for example, Sui et al. (2018) used DWT methods to transform all of fluctuations of the prices of energy stocks in China into multi-scales time series and analyzed their correlation structure under multiple threshold scenarios, and Liu et al. (2017b) used DWT methods to transform all the stock indexes of G20 countries into multi-scale time series and then analyzed the spillover relations of the G20 countries at different timescales. These studies provide a good basis for analyzing the features of air pollution spillover among Jing-Jin-Ji cities at multiple timescales.

In this study, we used a complex network-based method and the GARCH-BEKK model to more effectively identify the structures, routes and clustering features of air pollution spillover among Jing-Jin-Ji cities. Complex network analysis is an effective modeling method for elucidating the relationships among multiple agents. Briefly, a complex network is composed of nodes (agents) and links (relations of agents), and this approach has been widely used for modeling the relationships among agents, which could be countries (Li et al., 2017), sectors (Shi et al., 2017), companies (An et al., 2017), human beings, animals (Ivens et al., 2016), papers (Li et al., 2016), or words (Li et al., 2015). Various relationships are used in complex network modeling, including direct relationships, such as holding relationships (Li et al., 2014), affiliation relationships (Li et al., 2017), and trading relationships (Guan et al., 2016), as well as indirect and calculated relationships, such as co-holding and competition relationships (An et al., 2017), causality relationships (Jiang et al., 2017), correlations (Sui et al., 2018), and spillover relationships (Liu et al., 2017a). The GARCH-BEKK model is used to find the spillover relationships among multiple agents based on time series, and based on the spillover coefficients, we can obtain bidirectional results for the whole spillover effect regarding both the spillover effect of agent A on B and that of B on A. Based on the spillover relationships, we can construct a directed spillover network of all the agents in a holistic manner, and we can also use different algorithms and formulas of complex network theory to identify the routes and clustering features of air pollution spillover among Jing-Jin-Ji cities.

In this paper, we mainly focus on identifying the routes and clustering features of air pollution spillover among Jing-Jin-Ji cities to provide a decision-making basis for the joint regional control of air pollution in Jing-Jin-Ji cities. First, we used grasping technology to achieve the daily air pollution value (PM 2.5) of 13 cities in the Jing-Jin-Ji area. Second, using the DWT method, we transformed the air pollution time series of each city into multiple timescales. Third, wed use the GARCH-BEKK model to calculate the air pollution spillover relationships between any two cities at multiple timescales. Fourth, using a complex network-based method, we constructed directed air pollution spillover networks at multiple timescales by taking the cities as nodes, the significant spillover relationships as edges, and the spillover coefficients as weights. Using methods for route and community detection, we then obtained the routes and clustering features of air pollution spillover among Jing-Jin-Ji cities at multiple timescales.