Introduction

Air pollution is a key environmental threat to human health worldwide (Cohen et al. 2017; Burnett et al. 2018). In China, long-term exposure to fine particulate matter (PM2.5) may have caused 1.5 to 2 million premature deaths annually (Liang et al. 2020), prompting the government to place air pollution control as a top priority in its policy agenda. Since 2013, the central government has promulgated two large-scale national action plans to tackle the PM2.5 air pollution problem: first the 2013–2017 Air Pollution Prevention and Control Action Plan (the 2013–2017 Action Plan hereafter) with ambitious reduction targets set to be achieved by the end of 2017, and then the Three-year Action Plan to Win the Blue-Sky Defense War (2018–2020) as a spinoff (Feng et al. 2019; Li et al. 2020b; Jiang et al. 2021). These two national movements together with stringent enforcement have reduced PM2.5 concentrations and improved the air quality greatly (Zheng et al. 2018; Ma et al. 2019; Editorial 2019), and the annual average PM2.5 concentrations declined by 34–49% across China from 2013 to 2018 (Zhang et al. 2019b).

However, few studies have examined the post-2018 changes in PM2.5 concentrations in China in detail (Zhai et al. 2019; Chen et al. 2020; Bae et al. 2021) (see Table 3 for more detail), or questioned if PM2.5 concentration levels had continued to decline after the successful experience by the Action Plan (Yin and Zhang 2020; Zhong et al. 2021). One recent study on the PM2.5 pollution in Beijing-Tianjin-Hebei and the surrounding areas (the so-called “2 + 26” cities) estimated that PM2.5 concentration levels increased by 6.8 µg/m3 (9.46%) in the winter of 2018 (Dec–Feb). More recently, a few studies discovered that PM2.5 air pollution also deteriorated during the coronavirus disease 2019 (the COVID-19 hereafter) epidemic in early 2020 (Wang et al. 2020a; Le et al. 2020). However, one big issue with these studies was that such a rebound in PM2.5 air pollution was more than 100% attributed to meteorological conditions, and human efforts continued to reduce PM2.5 concentrations (Yin and Zhang 2020). These results and conclusions run contrary to the observation that government policies and actions may have been relaxed after the great success in 2017 (Ministry of Ecology and Environment of China 2017, 2018, 2019). According to the official statistics (Ministry of Ecology and Environment of China 2020), the national average PM2.5 concentration levels in 2019 stayed the same as that in 2018, both at 36 µg/m3, pointing to a strong possibility of a reversed trend for at least some regions.

Our study aims to reassess the changes in China’s PM2.5 pollution from 2018 to early 2020 before the outbreak of COVID-19 based on a large sample of cities with adjusted PM2.5 data and decompose the relative contributions of the meteorological and anthropogenic factors to the recent changes. The contributions of our study are threefold: First, by focusing on the most polluting areas in China with district heating in the north of the Huai River (Ebenstein et al. 2017), we expand the study area to 136 cities in 15 heating provinces in northern China from 2015 to early 2020 right before the COVID-19 crisis. Second, we adjust the PM2.5 concentration data following the official monitoring method change for comparison consistency (see “Materials and methods” for more detail). Third, we apply a stepwise multiple linear regression (MLR) analysis (Li et al. 2019a, 2020a; Zhai et al. 2019; Chen et al. 2020; Bae et al. 2021) to decompose PM2.5 changes into contributions from meteorological and anthropogenic factors. We further relate the human factor contributions to actual policy adjustment and the rising marginal cost of these policy measures.

Materials and methods

Ground-based PM2.5 data with qualitative adjustment

We used data of 136 northern cities with central heating in China. These cities spread over 15 northern provinces: Beijing, Tianjin, Hebei, Shanxi, Heilongjiang, Jilin, Liaoning, Inner Mongolia, Shandong, Henan, Shaanxi, Ningxia, Gansu, Qinghai, and Xinjiang (Fig. 1). Among those 136 cities, “2 + 26” cities were the most polluted area in China, and the targeted region by the Ministry of Ecology and Environment to mitigate PM2.5 pollution in February 2017 (Ministry of Ecology and Environment of China 2017); other northern cities were not targeted at this time. Given the heterogeneity in these two regions, below we analyze the changes in PM2.5 concentrations of “2 + 26” cities and other northern cities separately.

Fig. 1
figure 1

Changes in mean PM2.5 concentrations in northern China from 2016 to 2019. a–c Inter-annual changes in northern China between two consecutive years. d Time trend in PM2.5 concentrations for all northern cities and “2 + 26” cities. Ground-based PM2.5 data from 136 northern cities were used here (for the other 19 cities in northern China, data were imputed from their direct neighbors and then used in the maps). The 2019 heating season excludes post-Feb 2020 data since the lockdown measures led to considerable PM2.5 reductions. Rectangles in a–c indicate the Beijing-Tianjin-Hebei region and the surrounding areas (i.e., the “2 + 26” cities), and shaded areas in d denote the heating seasons

Full size image

The data source for ground-based PM2.5 concentrations is an online platform (https://www.aqistudy.cn/historydata/), which has been used by several studies before (Zhang et al. 2019c; Zhu et al. 2020). This data source collects real-time data of national monitoring sites from the Ministry of Ecology and Environment of China (https://www.mee.gov.cn/hjzl/) and then calculates monthly mean PM2.5 concentrations for each city in China. For the purpose of this research, we collected monthly PM2.5 concentrations data for each northern city from 2015 to early 2020 before the COVID-19 crisis.

The biggest challenge to use ground-based PM2.5 concentrations data is that the Ministry of Ecology and Environment changed its monitoring method of PM2.5 concentrations in September 2018 (Li et al. 2020a), so that PM2.5 concentration data are not directly comparable before and after September 2018. According to the Amendment for Ambient Air Quality Standards (hereafter referred to as “the Amendment”), the local ambient state instead of the standard state (273 K, 1013 hPa) should be used during the monitoring process of PM2.5 concentrations before this Amendment. The local ambient state refers to the real-time condition of monitoring stations, i.e., real-time temperature and surface pressure. Thus, to achieve a consistently measured time series of PM2.5 concentrations, we followed the literature (Laugier and Garai 2007; Li et al. 2020a) and applied the Ideal Gas Equation of State to adjust pre-September 2018 PM2.5 concentration data:

$$\begin{array}{c}{PM}_{adj}={PM}_{obs}\times \frac{273+0}{273+T}\times \frac{Pressure}{1013}\end{array}$$
(1)

where \({PM}_{adj}\) and \({PM}_{obs}\) are PM2.5 concentrations with and without adjustment, respectively; \(T\) is the real-time temperature, and \(Pressure\) is the real-time surface pressure. Since the temperature and surface pressure should be roughly the same within a city, we directly applied the above adjustment to our city-level PM2.5 concentration data.

Meteorological data

We integrated meteorological data from several sources. Meteorological variables were mainly from Airwise, an online data platform (http://hz.zc12369.com/home) that has been used by previous research (Xu et al. 2020; Li et al. 2020b). The original data source of Airwise is the ground-based data from the National Meteorology Center (NMC) of China (http://www.nmc.cn/). This platform collects the real-time data of national sites from NMC and provides monthly meteorological data for 137 northern cities from 2015 to 2020. We used six meteorological variables including mean temperature, humidity, wind speed, surface pressure, total precipitation, and total cloud cover. We chose these variables since they are strongly correlated with PM2.5 concentrations (Xu et al. 2018; Liu et al. 2020). To fill in possible missing values, we further collected surface pressure data from the China Meteorological Data Service Center (http://data.cma.cn; last accessed: 2020–10-17) and the National Climate Data Center (https://quotsoft.net/air/#archive; last accessed: 2020–11-25). We also collected three more meteorological variables used in the sensitivity analyses (described below): 850 hPa meridional wind velocity (V850), 10 m meridional wind (V10), and zonal wind (U10), all from MERRA-2 reanalysis by NASA Global Modeling and Assimilation Office (https://gmao.gsfc.nasa.gov/reanalysis/MERRA-2, last accessed: 2021–01-10). The MERRA-2 data have a spatial resolution at 0.5° \(\times\) 0.625° and we further averaged these data to the city level.

Multiple linear regression model

Several studies have applied stepwise MLR models to quantify the relative impacts of meteorological and anthropogenic factors on air pollutant trends (Li et al. 2019a, 2020a; Zhai et al. 2019). We employed the same method to estimate the contributions of human and nature factors to the PM2.5 rebound. First, we regressed monthly PM2.5 concentrations on six meteorological variables from February 2017 to January 2020, which are highly relevant to PM2.5 concentrations. We specified the following MLR model for each city:

$$\begin{array}{c}{PM}_{adj}={\beta }_{0} + \sum\limits_{k = 1}^{6}{\beta }_{k}{X}_{k}+\varepsilon \end{array}$$
(2)

where \({PM}_{adj}\) is the monthly mean PM2.5 concentrations after adjusting, \({X}_{1}\)-\({X}_{6}\) are six meteorological variables as described above, and \({\beta }_{k}\) are the corresponding regression coefficients.

We further conducted a series of sensitivity analyses by adding more meteorological variables as well as the second order effect of all variables into the regression. Specifically, we added V850, V10, and U10 into the baseline model. Since V850 has a strong correlation with PM2.5 concentrations in Northern China, we also ran a MLR model with V850 and our first six meteorological variables (Cai et al. 2017). For all model runs at the city level, only the first three key variables based on R-square were kept in the final regression to avoid overfitting (Li et al. 2020a). These three key variables varied from one city to another (Supplementary Figs. S1–3). The fitted values of the MLR model reflect the contributions from meteorological factors.

To estimate the contributions from anthropogenic factors, we further distilled the nonlinear trend in the residuals from the MLR model. Specifically, we included seven dummy variables in the regression of MLR residuals from the last step:

$$\begin{array}{c}Res={\alpha }_{0}+ \sum\limits_{k=1}^{7}{\alpha }_{k}{T}_{k}+\varepsilon \end{array}$$
(3)

where \(Res\) is the residual from Eq. (2) and \({T}_{1}\)-\({T}_{7}\) are time-dummy variables covering the whole study period and indicating whether each month belongs to the heating season or non-heating season in a specific year (e.g., 2017 heating season and 2017 non-heating season, and so on). The regression coefficients \({\alpha }_{k}\) are the anthropogenic contributions for each season.

Results

PM2.5 pollution rebounds in northern China

Annual mean PM2.5 concentrations in northern cities declined from 2015 to 2018 but rebounded in 2019. Table 1 shows that the annual mean PM2.5 concentrations in northern China dropped continuously from 2015 to 2018, with a total effect of 16.56 µg/m3. However, this declining trend was reversed in 2019, when PM2.5 concentrations increased by 5.16 µg/m3 (14.60%), or 80% of the reduction (6.4 µg/m3) achieved in 2018. Figure 1c further shows that such a PM2.5 rebound in 2019 occurred to most cities in northern China, though some of them continued to reduce their PM2.5 concentrations in 2019, such as Beijing (Li et al. 2020b). The rebound was even more significant during the heating seasons (HS; Nov to next Mar) when the PM2.5 pollution was the most severe (Fig. 1d). In the 2018 HS, northern cities saw the PM2.5 concentrations rebounded by 5.71 µg/m3 (10.80%) after a drastic drop of 16.17 µg/m3 in the 2017 HS (Table 1). However, PM2.5 concentrations declined remarkably in the 2019 HS due to the impact of COVID-19 lockdown (Supplementary Table S1) (Huang et al. 2021), resulting in a smaller rebound in northern cities during 2017–2019 HS (Supplementary Table S2). Given that lockdown only had a short-term impact on PM2.5 trends (Lu et al. 2021), we excluded its impacts by removing data in February and March 2020 (post-Feb 2020) from the baseline results. As such, PM2.5 rebound continued into the 2019 HS, with the PM2.5 concentrations further increased by 4.79 µg/m3 (8.1%). These two rebounds increased PM2.5 concentrations by 10.49 µg/m3 and offset 65% of the largest reduction in PM2.5 concentrations during the 2017 HS.

Table 1 Year-to-year changes in mean and max PM2.5 concentrations for different time ranges from 2015 to early 2020 in northern China
Full size table

The so-called “2 + 26” cities, consisting of Beijing, Tianjin, and 26 surrounding cities in Northern China, are the key area in China’s air pollution control program since 2013, and also the targeted region in the annual action plan during heating seasons since 2017. Compared with other northern cities, “2 + 26” cities had a greater drop of 26.67 µg/m3 (33.43%) in their annual mean PM2.5 concentrations from 2015 to 2018, but with a smaller rebound of 3.02 µg/m3 (5.68%) in 2019 (Table 1). During heating seasons, these 28 cities also reduced their PM2.5 concentrations much more than other cities in the 2017 HS and then experienced a greater rebound in PM2.5 pollution in the 2018 HS. Mean PM2.5 concentrations in the 2018 HS increased by 7.30 µg/m3 (9.62%) compared to the previous year (Supplementary Fig. S4). Such a rebound effect continued into the 2019 HS (2.14 µg/m3, 2.57%). These two rebounds together offset 29.44% of the biggest reduction of 32.06 µg/m3 in the 2017 HS. Other northern cities had an even larger and more evenly distributed rebounding effect over the two heating seasons: 5.35 µg/m3 in the 2018 HS and 5.38 µg/m3 in the 2019 HS.

It is worth noting that the maximum of PM2.5 concentrations consistently show greater rebound than the means, implying that the PM2.5 rebound made the worst pollution month even worse (Table 1).

Decomposing PM2.5 rebounds into anthropogenic and meteorological factors

We decomposed the contribution of meteorological and anthropogenic factors to the rebound of PM2.5 concentrations using a stepwise MLR model (Li et al. 2019a), one of the two common methods used in this task (see Table 3 for our summary). The results showed that anthropogenic factors contributed to 55% (2.82 µg/m3) of the PM2.5 rebound in northern cities from 2018 to 2019, while meteorological factors only contributed to 45% (2.34 µg/m3) of the rebound (Fig. 2e). A similar pattern appeared over the 3-year heating seasons from 2017 to 2019, though with a much larger absolute rebound (Fig. 2f). For the “2 + 26” cities, around 40–50% of the PM2.5 rebound was attributed to anthropogenic factors, and the rebound during the heating seasons was much larger, too. Results from sensitivity analyses with respect to variable choices and functional forms further show a range of 50–60% of human contributions for northern cities, and 20–80% for “2 + 26” cities from 2018 to 2019 (Supplementary Table S3). Our decomposition results are also robust even if post-Feb 2020 data are included in the analysis (Supplementary Table S3).

Fig. 2
figure 2

Decomposition of the relative impact of anthropogenic and meteorological factors on changes in PM2.5 concentrations. a-d Monthly anomalies (de-meaned) of observed and fitted PM2.5 concentrations in all northern cities and “2 + 26” cities either annually or for the heating seasons (HS: Nov, Dec, Jan, Feb, and Mar). e–h Decomposition results in all northern cities and “2 + 26” cities either annually or for the heating seasons only. The 2019 heating season excludes post-Feb 2020 data since the lockdown measures led to considerable PM2.5 reductions

Full size image

The contributions of anthropogenic drivers in different cities varied greatly. For the 136 northern cities used in our MLR model, the contributions of anthropogenic drivers to PM2.5 rebound during the 2017–2019 HS ranged from − 10.82 to 32.73 µg/m3, and 112 of them had seen positive contributions of anthropogenic factors to the PM2.5 rebound (Supplementary Fig. S6). Considering the important role of anthropogenic factors in PM2.5 rebound, we focus on the adjustment of policy targets and mitigation measures below, which would directly affect emissions from human activity. That being said, the underlying physical and chemical mechanisms of PM2.5 rebounds also merit further investigation.

Correlation of PM2.5 rebounds with previous reduction

The significant PM2.5 rebound driven by anthropogenic factors in the heating seasons was tightly linked to the significant reduction of PM2.5 before. Anthropogenic contributions to PM2.5 concentrations were correlated over space and time. As Fig. 3 shows, for every 100 km reduction in the distance to Beijing (within a 400 km range), the rebounding effect of PM2.5 concentrations driven by human factors increased by about 3 µg/m3 during the 2018 HS for the “2 + 26” cities. This seems to suggest a counterintuitive and declining pressure from Beijing to surrounding cities for the latter to reduce PM2.5 emissions. However, it might be well explained by the more significant reduction in PM2.5 concentrations due to human factors by cities closer to Beijing in the 2017 HS. For every 100 km reduction in the distance to Beijing, the PM2.5 concentrations caused by human factors dropped by 11 μg/m3 in the 2017 HS, corresponding to a 27% rebounding effect in the 2018 HS. Interestingly, the total rebound effect during the 2017–2019 HS had little to do with the city distance to Beijing (Supplementary Fig. S7).

Fig. 3
figure 3

Correlation between city distance to Beijing and inter-annual changes in anthropogenic impacts on PM2.5 concentrations for the “2 + 26” cities during the heating seasons from 2016 to 2018. The geographical distance is measured by the transport distance extracted from the Chinese largest search engine (https://map.baidu.com/). Three cities in Henan province, Puyang, Hebi, and Jiaozuo, are excluded in the regression analysis due to their missing data problem

Full size image

As for other northern cities, the magnitude of the rebounding effect due to human factors in the 2018 HS was also negatively related to their contribution to the reduction in PM2.5 concentrations in the previous year. Specifically, for 1 µg/m3 reduction in PM2.5 concentrations resulting from human factors in the 2017 HS, PM2.5 concentrations increased by 0.18 µg/m3 in the 2018 HS (Fig. 4), pointing to an 18% rebounding effect in the 2018 HS. This rebounding effect was less than the 26% rebound effect for “2 + 26” cities in the 2018 HS (Fig. 4), probably because other northern cities did not reduce their PM2.5 concentrations in the 2017 HS as much as “2 + 26” cities did. However, the total human-related rebound effect over the 2-year heating seasons from 2018 to 2019 amounted to 40% of the reduction in the 2017 HS for other northern cities, which became even larger than the 28% for “2 + 26” cities (Supplementary Fig. S8). The differences in the rebounding effect driven by human factors between the two regions point to different reasons behind their PM2.5 rebounds.

Fig. 4
figure 4

Correlation of two inter-annual changes in anthropogenic impacts on PM2.5 concentrations for northern cities during the heating seasons from 2016 to 2018

Full size image

Policy and cost reasons behind PM2.5 rebounds

PM2.5 rebound was closely related to the policy re-adjustment and rising marginal cost of mitigation measures, at least for the “2 + 26” cities. By comparing policy targets and mitigation measures from 2017 to 2019 during the heating seasons for the “2 + 26” cities (Fig. 5), we find that the targets for reducing PM2.5 concentrations and severe haze-polluted days decreased from 15% in the 2017 HS to 3% in the 2018 HS and 4–6% in the 2019 HS. Furthermore, specific regulations on mitigating PM2.5 pollution also loosened in their intensity and stringency, though with more flexibility being added. For instance, the strong one-size-fits-all requirements were removed with respect to the coal-to-gas and coal-to-power programs; as a replacement, it was then encouraged to use power, gas, coal, and centralized heating in their best suitable way. Furthermore, industrial production was staggered, coal-fired boilers were phased out, and dust control was managed with a new credit system. The shrunken policy goal and increased flexibility and freedom were consistent with the observed rebounding effect of PM2.5 concentrations and the significant role of human factors in the “2 + 26” cities during the heating seasons, irrespective of the influence of natural factors. One possible reason for the systemic policy re-adjustment after 2017 was due to the economic losses born by local governments to achieve the milestone targeted by the Action Plan (Li et al. 2019c). Moreover, the policy re-adjustment for “2 + 26” cities in the 2018 HS might contribute to the greater PM2.5 rebound than that of other northern cities.

Fig. 5
figure 5

A comparison of official policy target and mitigation measures for the PM2.5 air pollution control in the “2 + 26” cities during the heating seasons from 2017 to 2019. The data are obtained from Ministry of Ecology and Environment of China (2017, 2018, 2019)

Full size image

The rising marginal cost of PM2.5 mitigation measures was also astonishing. As illustrated in Fig. 6, only four out of 13 measures cost less than 100 yuan (~ $15.64) per kg of PM2.5 emission reduction, including removing small coal-fired boilers, lowering pollution in key industries, controlling road dust, and phasing out outdated industrial capacities. These measures together account for roughly 70% of the existing PM2.5 emission reduction capacity. By contrast, the heavily used coal-to-gas and coal-to-power programs were very costly with the highest average abatement cost at 583 yuan/kg (~ $91.20/kg), though these programs brought about significant PM2.5 emission reduction in the 2017 HS (Li et al. 2020c; Wang et al. 2020b). After a dramatic cut in PM2.5 emissions in 2017 to meet the policy goal stipulated by the Action Plan, there was little room for these measures to further reduce emissions cost-effectively—an inconvenient truth consistent with the PM2.5 rebound.

Fig. 6
figure 6

An average cost curve for PM2.5 abatement in China. Data on the abatement effect and average cost of each measure are obtained from Zhang et al. (2019a)

Full size image

Discussion

Thanks to rigid mitigation measures and favorable meteorological conditions, China has achieved significant reduction in PM2.5 concentrations during 2013–2018, especially in northern cities; nevertheless, the post-2018 rebound in PM2.5 pollution cannot be overlooked. This study found a PM2.5 rebound in northern China in 2019 and early 2020, even though the nationwide PM2.5 concentrations remained unchanged from 2018 to 2019. To mitigate the health risk linked to PM2.5 pollution, how to maintain a sustainable PM2.5 reduction merits more attention. So, it is necessary to quantify the relative contribution of both anthropogenic and meteorological factors to the PM2.5 rebound and link decomposition results to underlying human factors, such as the actual policy readjustment and the increasing marginal cost of mitigation measures.

Our decomposition results differ considerably from the existing study (Yin and Zhang 2020) claiming that meteorological conditions explained 122% of the PM2.5 rebound in the 2018 winter and that human factors offset rather than contributed to the PM2.5 rebound in the “2 + 26” cities. There may be two explanations for such a qualitative difference. First, we used adjusted PM2.5 concentration data to start with, while the previous study did not (see Table 2 for a direct comparison with the literature and the impact of such our data adjustment). The second reason is that different methods were used to fit the observed PM2.5 trend: While we applied a stepwise MLR model, the previous study used a chemical transport model (CTM). Though there is no consensus concerning which method performs better, our model had a better goodness-of-fit. The mean correlation of model fitted values with actual PM2.5 concentrations was 0.86 (Fig. 2a; Supplementary Fig. S5), compared to a range of 0.67–0.72 in the previous study based on CTM (Yin and Zhang 2020).

Table 2 Comparison of mean PM2.5 concentrations with and without data adjustment in the winters of 2017 and 2018 for the “2 + 26” cities
Full size table

Moreover, our results were generally consistent with those from most previous studies (Table 3). Although few studies investigated the drivers of PM2.5 concentrations in post-2018 China, many studies on the previous declining trend of PM2.5 suggested that meteorology only played a minor role in the process, regardless of the method being used. For example, a study using MLR indicated that 12% of the PM2.5 reduction from 2013 to 2018 was attributable to meteorology (Zhai et al. 2019), while another study used the CTM model and found that meteorology contributed 16% to the reduction from 2013 to 2017 (Zhang et al. 2019b). Thus, it is unreasonable to fully attribute the PM2.5 rebound to meteorology while neglecting the anthropogenic factors (Yin and Zhang 2020).

Table 3 A summary of meteorological impacts on PM2.5 level changes by previous studies
Full size table

On the methodology front, both MLR and CTM have been widely used to decompose the meteorological and anthropogenic factors behind pollution level changes; however, it is still too early to determine which one is better. While CTM can simulate the chemical process and predict chemical species with more detail (Bey et al. 2001), one limitation of this method is its inherent uncertainty regarding its emission inventory (Chen et al. 2019; Zhong et al. 2021): Even if two studies used the same CTM and focused on the same domain, their decomposition results can vary much (Ding et al. 2019; Dong et al. 2020). By contrast, the uncertainty of the MLR mainly comes from model specifications and its sensitivity to data outliers. Moreover, MLR is unable to provide detailed information on physical and chemical mechanisms of PM2.5 trends as CTM did in previous studies (Wang et al. 2020a; Le et al. 2020; Huang et al. 2021). This, admittedly, is one limitation of our research that merits more future research. Nevertheless, MLR is sufficient to fulfill the main research objective of the paper, i.e., to analyze the long-term trend of PM2.5 and decompose the PM2.5 rebounds into human and natural factors. Such a statistical approach has been used in other similar research such as Li et al. (2019a) and Zhai et al. (2019). The obtained MLR results can provide useful hints on high-level factors behind the PM2.5 rebound and thus help achieve sustainable PM2.5 emission reductions.

Broadly speaking, the PM2.5 rebound during the heating season characterizes the typical campaign-style governance model of China (Liu et al. 2015; Wang et al. 2021). First is the top-down campaign-style governance with quick and highly visible effects, which may create long-term issues and side effects (Zhao et al. 2020b). Specifically, this model focuses on solving the most-pressing issue with top priorities, at a very quick pace. To solve this problem, the governments set an ambitious goal that needs to be achieved within a limited amount of time and then mobilize and coordinate all types of resources to fulfill that goal. Such a campaign-style model has the merit of taking effect very quickly but cannot sustain the effort due to high costs of implementation. Once the initial goal is reached, the governments will shift their attention and focus to other important issues. This is the case for PM2.5 rebound in northern China. The policy goal of a 10% reduction in annual PM2.5 from 2012 to 2017 set up by the Action Plan was first over-met by governments leveraging heavy one-size-fits-all measures regardless of their relative cost-effectiveness (Zhang et al. 2019b). However, this was later followed by a rebounding effect of as high as 65% in the following heating seasons for northern cities in China.

Given the increasing marginal costs of emission reductions (Fig. 6), we do not believe that the Chinese government intended to adjust the goals and measures to result in the PM2.5 rebounds, but only the result of a dynamic process characterized by the campaign-style governance model. In other words, the PM2.5 rebound we found for northern cities only followed after deep cuts in PM2.5 concentrations previously. It is interesting to note that the magnitude of and the reasons for the PM2.5 rebounds in “2 + 26” cities and other northern cities was somewhat different. Specifically, the rebounding effect in “2 + 26” cities was greater than that of other northern cities in the 2018 HS. This is because “2 + 26” cities were chosen as the targeted region in the annual action plan since 2017 and thus had to reduce PM2.5 pollution more seriously in the 2017 HS. Owing to the policy re-adjustment and increasing marginal costs later on, PM2.5 concentrations in “2 + 26” cities rebounded more in the 2018 HS than in other northern cities. If without the previous deeper cuts in PM2.5 concentrations, we should not expect “2 + 26” cities to perform worse than other cities in terms of either PM2.5 mitigation or PM2.5 rebounds. Such a comparison between “2 + 26” and other northern cities suggests that China’s campaign-style governance model had its spatial range limitations. Furthermore, the campaign on PM2.5 also produced other repercussions. Studies found that southern cities experienced natural gas shortages owing to the hyped coal-to-gas program in the north and these cities had to revert to coal for production and heating (Wang et al. 2020b). The decrease in PM2.5 concentrations in northern China even contributed to the increase in O3 concentrations in the summer owing to the excessive attention on PM2.5 controls (Li et al. 2019b; Wang et al. 2020c; Zhao et al. 2020a).

Second is the system’s ability to adapt to emerging problems and new situations in the process. In the case of controlling China’s PM2.5 pollution, the re-adjustment of policy targets and the insertion of flexibilities into policy measures after 2017 may reflect certain degrees of self-learning and system re-calibration, rather than pure rebounding. After all, the rising abatement cost and potential GDP loss (Li et al. 2019c) cannot simply be ignored and must be carefully balanced with PM2.5 emission reduction. The end result is that even considering the relaxation of policy stringency and the rebound effect, China’s overall speed in reducing PM2.5 concentrations is still faster than that of its counterparts from the developed world (Greenstone et al. 2021). Similar findings also exist in relation to China’s reduction of its coal overcapacity (Dong et al. 2021). Whether a smooth or curly path can better address these complex collective action problems deserves more attention and fundamental research.

In sum, the main contribution of this paper is to confirm the PM2.5 rebound in northern China from Nov 2018 to Jan 2020. After decomposing the changes in PM2.5 concentrations via a statistical approach, this study highlights the important role played by human factors, in particular the policy re-adjustment and the increasing marginal costs of PM2.5 abatement. However, there are still some data and methodological limitations that need to be addressed in future research. First, though the statistical approach was straightforward, it was unable to offer detailed information about the contribution of different mitigation measures. Future studies could collect more data and quantify the relative contributions of those different measures to the PM2.5 rebound. Second, due to the COVID-19 lockdown, the statistical approach could not separate its confounding effect on PM2.5 from the other non-COVID-related human factors that we care about. More future work should be done to control for confounding factors such as COVID-19 in order to examine more recent trends of PM2.5. Lastly, it would also be interesting to compare the relative performance of our statistical approach and chemical transport modeling approach with more data inputs and more research efforts in the future.

Conclusion and policy implications

This study found that mean PM2.5 concentrations in northern China continued to decline from 2015 to 2018. However, the PM2.5 pollution in this heavily polluted region has rebounded in 2019 and early 2020 before COVID-19, mostly due to changes during the heating season. Furthermore, anthropogenic factors have contributed roughly 50% of the rebound on average to this rebound after contributing to a severe cut of PM2.5 emissions in 2017. The underlying reasons for this PM2.5 rebound and the contribution from human factors have a lot to do with the systemic re-adjustment of policy targets and mitigation measures by the central government, as well as the rising abatement cost of many mitigation measures, rather than mere meteorological factors. In other words, such a human-induced PM2.5 rebound is the result of a dynamic process under the campaign-style governance model.

Regardless of the merits or demerits of China’s governance model, successfully addressing PM2.5 pollution brings about substantial environmental and human health benefits, as well as co-benefits for climate change. Challenges, however, still exist concerning how to maintain the previous efforts and momentums, potentially in a smarter and more cost-effective way. Based on our results and discussion, we make the following policy recommendations. First, chart a more sustainable path for future PM2.5 emission reductions to meet the 2035 policy target of 35 µg/m3 while avoiding possible rebounding effects. This includes paying more attention to key cities and key polluting seasons, both in the north and south, while not only focusing on the national annual mean concentration levels. Second, dynamically adjust the portfolio of mitigation measures based on their relative cost-effectiveness and social acceptance, strike a good balance between policy stringency and flexibility, and differentiate the contributions from natural and human factors. Third, address PM2.5 pollution together with other energy and environmental issues, leverage system synergies and co-benefits, and employ a multi-goal governance system.