Evaluating air pollution and economic impacts in Chinese megacity clusters under clean air policies

Summary

To address worsening air pollution issues, the Chinese government has incurred substantial economic costs by implementing emission reductions. This study employed the Environmental Kuznets Curve (EKC) framework to investigate the relationship between emission reductions and economic growth. The EKC exhibited similar characteristics in the Yangtze River Delta (YRD), Pearl River Delta (PRD), and Chengyu (CC), with a monotonically decreasing trend in NO_x, PM₁₀, PM_2.5, and SO₂ emissions. In contrast, completely inverted-U-shaped EKC curves were observed in the Beijing-Tianjin-Hebei (BTH) and Fenwei plain (FWP). Notably, all megacities exhibit an inverted U-shaped curve for VOC emissions in relation to GDP. The constructed emission-GDP functional model reveals distinct regional patterns, with nonlinear structures characterizing the PRD and YRD, a pronounced linear relationship dominating in the FWP, and hybrid patterns emerging in the CC and BTH. We further quantify heterogeneous economic losses from PM_2.5 and O₃ pollution, revealing distinct regional patterns tied to local development pathways. This study provides key scientific insights for exploring air pollution control strategies.

Graphical abstract

Highlights

• •Environmental Kuznets Curve used to develop pollutant emissions-GDP functional model
• •Emission reduction and economic growth align with the EKC theory
• •Spatial heterogeneity exists in the turning points of the EKC curve
• •Inverted EKC curves observed in the Beijing-Tianjin-Hebei and Fenwei plain

Environmental management; Environmental science; Pollution

Introduction

Air pollution has emerged as a critical environmental challenge in China, driven by rapid industrialization, urbanization, and energy-intensive economic development. Despite stringent regulatory measures implemented in recent decades, many regions continue to experience severe particulate matter (PM_2.5/PM₁₀), ozone (O₃), and nitrogen dioxide (NO₂) pollution, posing significant threats to public health and ecosystems.¹ Concurrently, the economic ramifications of air pollution are increasingly evident, including productivity losses, rising healthcare expenditure, and disruptions to agriculture and transportation.^2,3,4 Identifying the intrinsic linkage between anthropogenic emissions and economic development facilitates the formulation of scientific and dynamic emission reduction strategies. This is crucial for steering China’s megacities toward a win-win phase of sustained economic growth alongside reduced anthropogenic emissions.

After years of concerted efforts, primary PM_2.5 emissions in China have been significantly reduced. In some regions, PM_2.5 has transitioned to being predominantly secondary in origin.⁵ Atmospheric NOx and VOCs undergo photochemical reactions with OH radicals to form O₃, while simultaneously oxidizing NOx, SO₂, and VOCs to generate secondary particulate matter.^6,7,8 This process, in the simultaneous presence of elevated O₃ and PM_2.5 concentrations, results in a complex composite air pollution.⁹ As industrial structures evolve and emission reduction policies take effect, atmospheric chemical processes give rise to new pollution challenges, such as elevated O₃ and secondary particulate matter, which may offset previous environmental gains.¹⁰ Understanding these mechanisms is crucial for designing cost-effective pollution control strategies that balance economic development with air quality improvements.¹¹ By integrating atmospheric chemistry with economic modeling, policymakers can optimize emission reduction pathways, reduce economic burdens, and maximize health and environmental benefits, thereby supporting sustainable development in rapidly industrializing regions.

Numerous studies have confirmed that PM_2.5 and O₃ cause substantial economic losses to public health and cash crops in China.^12,13 Feng et al.¹⁴ estimated that in 2015, O₃ pollution resulted in combined health and agricultural economic losses equivalent to 7% of China’s GDP, including $7.5 billion and $11.1 billion in yield losses for rice and wheat, respectively. Yin et al.¹⁵ further assessed the global health costs of combustion-related PM_2.5, revealing that China incurred $1.1 trillion in health damages from fossil fuel and biomass combustion in 2019. Xie et al.¹⁶ focused on the Guangdong-Hong Kong-Macao Greater Bay Area, demonstrating that end-of-pipe control technologies could reduce PM_2.5 concentrations by 56.4% by 2050, thereby avoiding $31 billion in health-related economic losses, while highlighting that O₃ control requires more advanced technological investments, underscoring the necessity for coordinated multi-pollutant control strategies. However, significant research gaps remain regarding the scientific evaluation of emission reduction strategies and their economic benefits in China’s megacity clusters over the past decade. This knowledge gap hinders evidence-based optimization of regional air quality management strategies and the formulation of differentiated policies tailored to the varying developmental stages of megacity clusters.

This study focuses on the intrinsic relationship between air pollution and economic benefits across five megacity clusters in China. The objectives of this study were: (1) to investigate the general characteristics of PM_2.5 and O₃ pollution, (2) to analyze the interaction between air pollution and economic development in China, and (3) to construct a pollutant emissions-GDP functional relationship model. Our findings provide scientific support for formulating regionally differentiated emission reduction policies, enabling improvements in environmental quality while ensuring sustained economic growth.

Results

General characteristics of PM_2.5 and ozone pollution in the five megacity clusters

With the implementation of the Air Pollution Prevention and Control Action Plan (APPCP), China has made remarkable progress in air quality management, drawing global attention. Between 2013 and 2019, the annual average PM_2.5 concentration in 74 key cities across China declined by 47.2%. As shown in Figure 1, a comparative analysis was conducted to examine the monthly variation in PM₂.₅ concentrations across five megacity clusters—BTH, Yangtze River Delta (YRD), Pearl River Delta (PRD), Fenwei plain (FWP), and Chengyu (CC)—between 2015 and 2024. Elevated PM₂.₅ concentrations in these regions were predominantly observed during winter (December to February). In 2015, PM₂.₅ concentrations peaked at over 200 μg/m³ in BTH, followed by FWP and CC, where levels exceeded 180 μg/m³ and 140 μg/m³, respectively. In contrast, the YRD region exhibited comparatively lower PM₂.₅ concentrations, approximately 100 μg/m³. Unlike the other regions, PRD exhibited a delayed onset of PM₂.₅ pollution, beginning in January, with lower concentrations of approximately 65 μg/m³. Additionally, a notable increase in PM₂.₅ concentrations was observed in PRD during October, primarily attributed to the autumnal peak in O₃ pollution. Photochemical reactions associated with O₃ formation provide oxidants and precursors that facilitate the secondary formation of PM₂.₅, thereby enhancing its atmospheric production.

Figure 1.

Heatmaps of PM_2.5 concentrations (unit: μg/m³) across five megacity clusters in 2015 (left panel) and 2024 (right panel)

Compared with 2015, the implementation of PM_2.5-focused air pollution control strategies led to significant wintertime reductions in PM_2.5 concentrations by 2024 across the megacity clusters, with decreases of 58.13%, 35.81%, 34.07%, 36.31%, and 31.72% in BTH, YRD, PRD, FWP, and CC, respectively. As illustrated in Figure 1, the emission source intensities in the five megacity clusters declined by 53.73%, 56.61%, 38.08%, 56.47%, and 47.3% between 2010 and 2020, contributing to the varied downward trends in PM₂.₅ concentrations.

The substantial decline in PM_2.5 concentrations across China’s major megacity clusters is primarily attributed to stringent emission controls, industrial restructuring, and strengthened regional collaboration. The implementation of the APPCP (2013) and subsequent policies targeted coal combustion, industrial emissions, and vehicular pollution, substantially reducing primary PM_2.5 and its precursors (SO₂, NO_x, and VOCs). Stricter regulations on industrial and energy sectors, coupled with the promotion of clean energy technologies, proved particularly effective in high-emission regions such as BTH and FWP. Additionally, enhanced regional air quality governance and interprovincial enforcement helped mitigate transboundary air pollution. The YRD and PRD benefited from modernization and favorable coastal meteorological conditions, while the suppression of secondary aerosol formation through precursor control further contributed to PM_2.5 reduction.

In contrast to the effective control of PM_2.5 pollution, secondary pollution characterized by elevated O₃ and secondary particulate matter has become increasingly prominent, with intensified O₃ pollution observed both nationally and in key regions. As illustrated in Figure 2, distinct O₃ pollution patterns are evident across the five major urban agglomerations. In BTH and FWP, peak O₃ concentrations are primarily observed from May to August, whereas the YRD exhibits elevated levels beginning in April and continuing through October. The PRD exhibits high O₃ concentrations primarily from August to October, while the CC region shows increased O₃ levels between April and August, with peak monthly maximum values frequently occurring in August in recent years. Collectively, the periods of elevated O₃ concentrations in these megacity clusters have expanded from spring-summer to include both early spring and late autumn. Furthermore, O₃ pollution increased significantly between 2015 and 2024, with rises of 31.07% in BTH, 3.95% in YRD, 3.49% in PRD, 37.58% in FWP, and 22.41% in CC.

Figure 2.

Heatmaps of O₃ concentrations (unit: μg/m³) in five megacity clusters in 2015 (left panel) and 2024 (right panel)

The rising O₃ levels across China’s megacities result from complex interactions between emission controls and meteorological conditions. Stringent PM₂.₅ mitigation has paradoxically enhanced O₃ formation by weakening aerosol-mediated suppression of photochemical reactions, while disproportionate reductions in NO_x and VOC emissions have shifted O₃ formation regimes.

Climate change has extended the high-O₃ season into spring and autumn by elevating temperatures and altering atmospheric circulation patterns. These findings highlight a critical transition period in which conventional pollution control strategies must be optimized to effectively address emerging secondary pollution challenges.

We further calculated the population-weighted annual mean PM_2.5 and MDA8 O₃ concentrations, as presented in Figures S2 and S3. These population-weighted metrics exhibit spatiotemporal distribution patterns similar to those of the unweighted concentrations. Notably, under the population-weighted measure, the FWP ranks first for PM_2.5, with a concentration of 33.00 μg/m³, followed by the BTH region at 19.00 μg/m³, the CC region at 16.25 μg/m³, the PRD at 7.92 μg/m³, and the YRD at 4.80 μg/m³. For MDA8 O₃, the FWP also ranks highest with 26.70 μg/m³, followed by BTH at 13.15 μg/m³, CC at 12.20 μg/m³, PRD at 11.18 μg/m³, and YRD at 5.76 μg/m³. Despite having the largest population, the YRD has the lowest per capita pollution level among the clusters, reflecting the significant influence of population distribution on exposure assessment.

Relationship between anthropogenic emissions and associated economic benefits

As illustrated in Figure 3A, NO_x emissions declined by 17.6% from 170,000 tons/year in 2010 to approximately 140,000 tons/year in 2020. The most substantial reduction was observed in SO₂ emissions, which dropped by 83.3%, from 120,000 tons/year to just 20,000 tons/year over the same period. Particulate emissions showed consistent downward trends, with PM₁₀ and PM_2.5 decreasing by 60% and 62.5%, respectively, between 2010 and 2020. VOC emissions initially increased, peaking at 180,000 tons/year, then entered a gradual decline phase after 2016 following the implementation of control measures, although with limited effectiveness. Concurrently, the BTH region, as China’s political and economic center, exhibited steady GDP growth from 1,400 billion to 2,200 billion yuan during 2010 and 2020, corresponding to an average annual growth rate of approximately 4.6%. These findings underscore the importance of developing differentiated pollution control strategies—preserving effective measures for conventional pollutants while innovating targeted solutions for complex emissions such as VOCs.

Figure 3.

Time series of NO₂, PM_2.5, PM₁₀, SO₂, and VOC emission (unit: mol/yr) in the BTH, YRD, PRD, FWP, and CC regions from 2010 to 2020

The black line denotes the GDP value during 2010–2020

The YRD, one of the China’s most economically advanced regions, demonstrated a notable decoupling of economic growth from environmental pressures between 2010 and 2020 (Figure 3B). Quarterly GDP rose by 140%, from 250 billion to 600 billion yuan (corresponding to an annual growth rate of 9.2%)—the highest among the five megacities, while most pollutants showed substantial reductions: SO₂ emissions fell by 78.6% (from 140,000 to 30,000 tons/year), NO_x by 20% (from 200,000 to 160,000 tons/year), and PM₁₀ and PM_2.5 both by approximately 50%. However, the 25% increase in VOC emissions (from 200,000 to 250,000 tons/year) reflects unique challenges in this advanced manufacturing and high-tech hub, suggesting that current VOC control measures have not kept pace with the rapid industrial upgrading of the chemical and electronic sectors. This contrast underscores the need for region-specific pollution control strategies tailored to the distinct emission profiles of knowledge-intensive industries.

In comparison, the PRD experienced sustained GDP growth alongside significant reductions in pollutant emissions between 2010 and 2020. As shown in Figure 3C, the region’s quarterly GDP increased from 2 trillion to 3.8 trillion yuan, peaking at 3.9 trillion yuan in 2018 before a modest decline, with an average annual growth rate of approximately 6.6%. During this period of economic expansion, emissions of key air pollutants declined markedly: NOx decreased by 28.6% (from ∼70,000 to ∼50,000 tons/year), SO₂ fell by 83.3% (from ∼30,000 to ∼5,000 tons/year), while PM₁₀ and PM_2.5 emissions dropped by 66.7% (from ∼30,000 to ∼10,000 tons/year) and 60% (from ∼25,000 to ∼10,000 tons/year), respectively. VOC emissions remained relatively stable at approximately 100,000 tons/year, showing only a slight downward trend. Notably, even after 2018, when economic growth slowed, the PRD region continued its trajectory of emission reductions, demonstrating that its pollution control mechanisms had matured and were no longer strictly tied to reducing emissions, indicating that its pollution control mechanisms had matured and become decoupled from economic performance.

The FWP exhibited substantial economic growth alongside gradual progress in emission reductions between 2010 and 2020. As shown in Figure 3D, the region’s quarterly GDP increased from 900 billion yuan to 2.4 trillion yuan, corresponding to an average annual growth rate of approximately 10.3%. Despite rapid economic development, emission reductions in the FWP progressed relatively slowly due to the region’s initially high pollution baseline. SO₂ emissions declined substantially by 81.6% from 380,000 tons/year to 70,000 tons/year, yet remained higher than those of urban megacity clusters in 2020. Similarly, NO_x emissions declined by 31.3% (from 320,000 to 220,000 tons/year), while PM₁₀ and PM_2.5 emissions decreased by approximately 50% (from 200,000 to 100,000 tons/year and 150,000 to 70,000 tons/year, respectively). VOC emissions exhibited a modest 12% reduction from 250,000 to 220,000 tons/year. The emission reduction trajectory followed distinct phases: a pronounced acceleration during 2013–2014 coinciding with the implementation of national environmental policies, including the APPCP and central environmental inspections, followed by a period of gradual decline or stabilization between 2016 and 2020.

As shown in Figure 3E, the CC urban cluster experienced rapid economic expansion, with quarterly GDP rising from 600 billion to 1.8 trillion yuan, corresponding to an annual growth rate of 11.6%, the highest among China’s five major urban agglomerations. This period of rapid development featured divergent emission trajectories: while substantial reductions were observed for conventional pollutants—including an 84.2% decline in SO₂ emissions (from ∼190,000 to ∼30,000 tons/year), 62.5% and 66.7% reductions in PM₁₀ and PM_2.5 emissions (from ∼80,000 to ∼30,000 and ∼60,000 to ∼20,000 tons/year, respectively), and a modest 20% decrease in NOx emissions (from ∼100,000 to ∼80,000 tons/year)—VOC emissions exhibited a contrasting 16.7% increase (from ∼120,000 to ∼140,000 tons/year) before stabilizing after 2016. Temporal dynamics reveal a distinct transition pattern, with accelerated reductions in conventional pollutants beginning around 2013, whereas VOC emissions continued to rise until plateauing in the latter half of the decade.

The YRD and PRD regions have entered a post-industrial stage. The tertiary sector now dominates both regional economies. In the YRD, the service sector contributed 74.1% of Shanghai’s GDP in 2022, with shares in Jiangsu and Zhejiang provinces ranging between 53% and 55% (Shanghai Statistical Yearbook, https://tjj.sh.gov.cn/tjnj/nj23.htm?d1=2023tjnjen/E2306.htm). Similarly, the tertiary sector constituted 65.2% of the PRD’s GDP in 2022, reaching 71.5% in Guangzhou and 62.9% in Shenzhen (Guangdong Statistical Yearbook, https://stats.gd.gov.cn/gdtjnj/index.html).

Concurrently, the engine of economic growth has shifted decisively. It has moved from traditional, emission-intensive industries such as steel and chemicals toward modern services such as finance and information technology, and advanced manufacturing sectors including integrated circuits and biomedicine. This transition aligns with the post-industrial shift from factor-driven to innovation-driven growth, as noted in Zheng et al.,¹⁰ and was actively facilitated by policies such as Guangdong’s “Double Transfer” strategy.

The study demonstrates a clear divergence between economic growth and emission trajectories. As shown in Figures 3B and 3C, during 2010–2020, the YRD and PRD maintained robust annual GDP growth of approximately 9.2% and 6.6%, respectively, while emissions of key conventional pollutants (i.e., SO₂, NO_x, PM₁₀, PM_2.5) exhibited a significant, monotonic decline.

To formally assess this relationship, we calculated decoupling indices (Table 1). The results indicate that the pollutant emission-economic growth nexus in both regions corresponds to a state of “strong decoupling” within the Tapio decoupling model framework (Tapio, 2005). This provides quantitative proof that their economic growth is no longer reliant on increasing pollution emissions.

Table 1.

Comparative summary of EKC characteristics, turning points (TPs), tapio decoupling elasticity indices, and economic losses (yuan/person) across five megacities

Region	Pollutant	EKC Shape	Turning Points	Tapio decoupling elasticity indices	Economic losses (PM2.5)		Economic losses (O3)
					Health	Fuel	Health	Fuel	Agriculture
BTH	NOX	inverted-U	9.32	−0.268	1200.58	442.61	1240.36	70.82	26.56
	PM10	inverted-U	9.32	−0.562
	PM2.5	inverted-U	9.32	−0.540
	SO2	inverted-U	9.31	−0.756
	VOC	inverted-U	9.51	−0.069
YRD	NOX	monotonic decrease	–	−0.198	798.39	557.59	619.60	89.21	33.46
	PM10	monotonic decrease	–	−0.394
	PM2.5	monotonic decrease	–	−0.374
	SO2	monotonic decrease	–	−0.751
	VOC	inverted-U	10.05	0.127
PRD	NOX	monotonic decrease	–	−0.295	2375.59	977.55	1653.55	156.41	58.65
	PM10	monotonic decrease	–	−0.443
	PM2.5	monotonic decrease	–	−0.377
	SO2	monotonic decrease	–	−0.733
	VOC	inverted-U	10.54	0.003
FWP	NOX	inverted-U	9.85	−0.244	172.95	949.91	219.24	151.99	56.99
	PM10	inverted-U	9.89	−0.448
	PM2.5	inverted-U	9.89	−0.424
	SO2	inverted-U	9.85	−0.594
	VOC	inverted-U	10.26	−0.051
CC	NOX	inverted-U	9.07	−0.092	1695.11	412.71	439.64	66.03	24.76
	PM10	monotonic decrease	–	−0.257
	PM2.5	monotonic decrease	–	−0.241
	SO2	inverted-U	8.84	−0.409
	VOC	inverted-U	9.56	−0.010

A double dash indicates that no turning point was identified within the observed data range.

Our finding is consistent with previous research. For instance, Zheng et al.¹⁰ note that the YRD achieved industrial “lightweighting” through phasing out backward capacity and cultivating strategic emerging industries. Chen et al.¹⁷ also confirmed substantial emission reduction potential and a strong decoupling effect in the YRD. For the PRD, Zhou et al.¹⁸ showed a significant slowdown in CO₂ emission growth post-2017, with decoupling from GDP growth already occurring.

In summary, the YRD and PRD not only exhibit post-industrial economic structures but have also empirically achieved strong decoupling between major pollutant emissions and economic growth. This underscores that environmental governance in these regions has entered a new stage, necessitating differentiated policy focus, such as on complex secondary pollutants such as VOCs.

The comparative analysis reveals distinct regional patterns in economic growth and emission characteristics across China’s five major megacity clusters during 2010–2020. Economically, the CC and FWP clusters demonstrated the most rapid expansion with quarterly GDP growth rates of 11.6% and 10.3%, respectively, reflecting the accelerated catch-up development of China’s central and western regions. The YRD sustained robust growth at 9.2% despite its larger economic base, while the PRD exhibited a more mature development trajectory, with a growth rate of 6.6%. In contrast, the BTH recorded the lowest growth rate of 4.6%, reflecting substantial challenges related to structural economic transformation.

From an environmental perspective, the FWP’s emission profile reflects its heavy industry-dominated economic structure. The YRD and CC regions are facing emerging environmental challenges, as indicated by elevated and rising VOC emissions associated with industrial upgrading. Notably, the PRD has maintained relatively low emissions across all pollutant categories, demonstrating the environmental benefits of its advanced industrial structure. These regional disparities underscore the complex interplay among development stage, industrial structure, and environmental outcomes in China’s ongoing urbanization.

A comparative analysis of China’s five megacity clusters reveals a complex, nonlinear relationship between economic development and pollutant emissions, highlighting distinct regional trajectories shaped by industrial composition and development phase. Figure 4 illustrates the EKC framework for characterizing the relationship between per capita GDP and pollutant emissions. In the BTH region, all five major pollutants (NO_x, PM₁₀, PM_2.5, SO₂, VOCs) exhibit clear inverted U-shaped EKC patterns with per capita GDP. Furthermore, Figure S4 shows the impact of different policies on the EKC was analyzed by establishing the following sensitivity scenarios: (1) excluding data from 2020 to isolate the impact of the COVID-19 pandemic; (2) focusing on the period before 2016 to examine the relationship between emissions and GDP prior to the implementation of large-scale VOC controls post-2017; and (3) analyzing the relationship specifically during the implementation phase of the APPCP from 2013 to 2019.

Figure 4.

Non-linear EKC relationships between GDP and anthropogenic emissions in China’s five megacities during 2010–2020

As shown in Figure 4 and Figure S4, the shape of the EKC curves for the five megacities remains consistent after excluding data from 2020. This indicates that the EKC relationships established in this study primarily reflect the long-term structural linkage between anthropogenic emissions and GDP over the 2010–2019 period, with negligible distortion from the abrupt, pandemic-induced emission anomaly in 2020.

The comparative analysis of the pre-2017 period (before stringent VOC controls) reveals its most pronounced impact on the VOC EKC. During this earlier phase, VOC emissions across all five clusters exhibited a monotonic increase with no observable turning point. This result confirms both the later policy focus on VOC mitigation compared to other pollutants and the significant effectiveness of China’s dedicated VOC control efforts initiated after 2017.

Finally, focusing on the APPCP implementation period, we observe that for all pollutants except VOCs, the EKC curves show a monotonic decline. This suggests that the APPCP has successfully driven key conventional pollutants past the peak of their EKC, facilitating a win-win phase of environmental improvement and economic development. However, VOC emissions have not yet fully transitioned past their EKC turning point. Given VOCs’ crucial role as common precursors to both O₃ and secondary PM_2.5, sustained and enhanced efforts toward VOC emission reduction are essential. This is critical for achieving the co-control of O₃ and PM_2.5 and for supporting China’s overarching strategy of synergistic pollution reduction and carbon mitigation.

For these five megacities, we calculated goodness-of-fit metrics for the estimated EKC functions, including R², Akaike Information Criterion (AIC), and Bayesian Information Criterion (BIC). As shown in Table S4, our results indicate that the majority of the models exhibit high explanatory power, with most R² values exceeding 0.9, and the significance tests for the key coefficients are generally statistically significant at the level of p < 0.05. It is worth noting that the p values for VOC are relatively high, which may be attributed to the greater inherent uncertainty in VOC emissions. Additionally, to validate the optimality of the selected functional form, we systematically compared four candidate models using the AIC and BIC. As presented in Table S5, the chosen specification consistently yields the lowest AIC and BIC values among all alternatives, demonstrating its statistical superiority in balancing parsimony and explanatory power.

As a typical secondary pollutant, O₃ is formed through a series of photochemical reactions involving NOx and VOCs under solar radiation, exhibiting highly nonlinear behavior. Extensive studies have confirmed that the O₃-precursor sensitivity (OPS) in China’s five major megacities predominantly falls within VOC-limited or transitional regimes.^19,20 For instance, in the PRD region, sustained NOx reduction efforts have shifted its OPS toward a transitional or even NOx-limited one.^21,22 Furthermore, the dominant regime exhibits seasonal and regional heterogeneity. Elevated biogenic VOC (BVOC) emissions in summer, coupled with temperature-dependent increases in emission rates for certain anthropogenic VOCs (AVOCs) such as those from gasoline and volatile chemical products,²³ can alter the local OPS, necessitating dynamic control strategies.

The relatively late turning points observed in the VOC EKC are closely associated with structural adjustments in emission sources. Since the implementation of the APPCP, the PM_2.5-focused strategy has effectively reduced emissions of NOx, PM_2.5, PM₁₀, and SO₂. However, limited progress has been made in controlling VOC emissions, which originate from diverse and highly reactive sources. As shown in Figure 3, VOC emissions in China’s five megacities exhibited no consistent decline and even exhibited a slight increase during 2010–2020. This trend reflects a “policy lag” where national awareness and regulatory focus on different pollutants have prioritized the control of other species over VOCs. Furthermore, unlike the point-source characteristics of SO₂ and NOx emissions, VOCs are released from numerous dispersed sources, making their mitigation more challenging. Additionally, key VOC-emitting sectors, such as automobile manufacturing, electronics, furniture, and printing, have been economically significant in many regions, leading to a positive correlation between VOC emissions and GDP growth in earlier development stages. As regional economies transition toward low-VOC service sectors and high-end manufacturing, their VOC emissions begin a progressive decline, with a later-occurring turning point. Due to the limited coverage of monitoring data prior to 2015, there remains some uncertainty in assessing the magnitude of air quality improvements during the early policy implementation phase (2013–2014). However, when considered alongside emission trends derived from inventories and the rapid decline in pollutant concentrations observed in monitoring data after 2015, it can be reasonably inferred that the APPCP produced significant effects following its full-scale implementation starting in 2015.

To further investigate the impact of uncertainty in emission inventories on the shape of EKC curves, relevant studies indicate that the uncertainty for VOC emissions at 68–76%, considerably higher than the ranges reported for SO₂ (12–40%), NO_x (31–35%), or PM (30–94%).^{24,25,26,27,28} Based on the reported uncertainty ranges for pollutant emissions, we conducted sensitivity analyses to examine the potential impacts of emission data deviations under upper and lower bound scenarios. Considering the evolution of control policies, improvements in monitoring techniques during 2010–2020, and the uncertainty ranges reported in the literature, we constructed low- and high-emission scenarios by applying time-varying adjustment coefficients to the baseline emission inventory, as listed in Table S6. Under both scenarios, we re-fitted the EKC models to assess the robustness of the estimated emission-GDP relationships to emission inventory uncertainties.

As shown in Figure S5, the turning points for SO₂, NOx, and PM exhibit relatively minor variations under different emission scenarios. This stability stems largely from the sustained and intensive reduction efforts implemented since the introduction of the APPCP in 2013, which have enabled these pollutants to enter a phase of synergistic progress in both emission abatement and economic growth. Moreover, the relatively high accuracy of current monitoring systems for these pollutants reduces the uncertainty associated with emission inventories, thereby contributing to the robustness of their estimated turning points.

In contrast, VOCs display greater sensitivity to inventory uncertainty in shaping their EKC turning points. This is primarily due to the later initiation of dedicated VOC control measures, the complex speciation and reactivity of VOC components, the limited number of monitoring sites, and the substantial challenges in developing accurate emission inventories. These factors collectively introduce greater uncertainty in VOC estimates. Under the high-emission scenario, the VOC turning point tends to occur earlier, suggesting that under a less constrained emission trajectory, structural transformation in the energy system becomes necessary at an earlier stage to sustain economic growth while mitigating environmental pressures. Conversely, under the low-emission scenario, where the energy mix is relatively cleaner, the turning point is delayed or may not yet have emerged within the study period, implying that industrial restructuring can be undertaken at a later phase. Our results underscore that the timing of the turning point is shaped by the interplay between emission levels and structural transitions, and that proactive reduction efforts remain essential to accelerate the decoupling process.

The differences in the slopes of the EKC observed across the five megacities in this study can be attributed to variations in their OPS as well as their distinct energy mixes and industrial structures. Regional heterogeneity in OPS, shaped by underlying energy and industrial structures, significantly influences emission patterns and control intensities. The subsequent transition between these regimes, driven by abatement actions, is a key factor causing regional disparities in the ultimate effectiveness of pollution reduction measures.

The BTH's economy features a mix of heavy industry and rapidly expanding advanced manufacturing. Prior to reaching its turning point, economic development drove a rise in VOC emissions. Subsequently, structural adjustments, achieved through industrial consolidation, relocation, and enhanced end-of-pipe controls, enabled a rapid decline in emissions after the turning point was passed. The post-turning point decline in VOC emissions is markedly gradual in both the YRD and PRD regions. This trend is a reflection of their advanced post-industrial stage, wherein vital sectors such as electronics and automotive manufacturing drive the economy and maintain a tight coupling with VOC emissions. For the FWP, its resource-dependent economy, heavily reliant on coal, heavy chemicals, and traditional manufacturing, results in a strong coupling between economic output and pollutant emissions, thus delaying the emergence of a distinct EKC turning point. In the CC region, the combination of traditional heavy industry and VOC-intensive emerging manufacturing (e.g., automotive, electronics) maintains a strong growth-emissions link, thus explaining the relatively flat decline in its VOC EKC slope.

To explicitly quantify the uncertainty, we have computed and now report the 95% confidence intervals (CIs) for each identified turning point. These CIs, derived from the standard errors of the regression coefficients, provide a clear statistical range for our estimates, as listed in Table S7.

Additionally, we also performed a robustness check by re-estimating the turning points using a robust regression method, which minimizes the potential influence of outliers on the key parameters that determine the turning points, labeled as “Robust TP.”

The calculated 95% CIs indicate that the estimated turning points are statistically well-defined and fall within reasonable economic and policy-relevant ranges for their respective regions and pollutants. For all regions and pollutants where a turning point was identified, the Robust TP values are nearly identical to the original OLS TPs. This robust consistency confirms that the identified turning points are not artifacts of a few influential data points but represent stable features of the underlying regional emission-GDP relationships. The convergence between OLS and robust estimates, coupled with the statistically significant CIs, strongly reinforces the validity of the differentiated EKC pathways described above.

The decoupling analysis quantitatively confirms that all five megacity clusters achieved strong decoupling between economic growth and emissions of conventional pollutants such as SO₂, NOx, and particulate matter, during 2010–2020, as evidenced by consistently negative elasticity indices. This aligns with their progression along the EKC. However, the decoupling state for VOCs is more heterogeneous. While strong decoupling is observed in BTH, FWP, and CC, the slightly positive indices in the YRD and PRD suggest that their rapid, service-led growth remains partially coupled with VOC emissions. This critical nuance underscores the need for advanced, region-specific policies targeting VOC precursors, especially in post-industrial coastal regions.

As illustrated in Figure S7, the strong decoupling status (e < 0) of pollutants, including SO₂, NOx, PM_2.5, and PM₁₀, remains highly stable across all rolling windows in all five megacity clusters. This consistency demonstrates that the strong decoupling observed over the full 2010–2020 study period reflects a robust long-term trend rather than an artifact of any specific sub-period, persisting even under the intensive policy interventions implemented after 2013. For VOCs, a broadly similar trajectory is observed across all clusters in each rolling window, where emissions exhibit an upward trend in earlier windows followed by a decline thereafter. With the exception of the PRD, where minor fluctuations are observed, the remaining four megacity clusters all show declining VOC emissions beginning in the 2013–2017 window. Notably, the BTH and FWP regions shifted into a state of strong decoupling for VOCs following this period, and by the final 2016–2020 window, all five clusters achieved strong decoupling for VOCs. This sequential shift indicates the effectiveness of the VOCs control measures implemented in 2017, which progressively strengthened the decoupling process across these megacities. Although the full-period decoupling indices for the YRD and PRD remain positive, their underlying trajectories are trending toward decoupling, further suggesting that recent policy interventions have begun to take effect.

Modeling the relationship between anthropogenic emissions and economic development

The model presented in this study aims to explore the multivariate statistical relationship between the characteristics of multi-pollutant emissions and GDP levels. It is not intended for causal inference but rather to identify potential associations between economic development and emission profiles, which may reflect underlying industrial structure patterns. Given China’s ongoing nationwide “Dual Carbon” strategy, which promotes region-specific, multi-pollutant synergistic reduction plans, the model established in this work can serve as a useful tool for scientifically assessing the intrinsic linkage between emission reduction effectiveness and economic development across different regions.

In this study, we employed ordinary least squares (OLS) regression to analyze the relationship between MEIC emission inventory data and corresponding GDP statistics from 2010 to 2020, and LP as the base-10 logarithm of the per capita value. The econometric model establishes functional relationships between the base-10 logarithm of per capita GDP (LPGDP) and the base-10 logarithm of per capita emissions of various pollutants.

The pollutant emission-GDP functional relationship models established for the BTH, YRD, PRD, FWP, and CC regions are expressed as follows:

$${LPGDP}_{BTH}=-61.90\ast LPNOx++78.57\ast LP{PM}_{10}-35.22\ast LP{PM}_{2.5}-13.10\ast LPS{O}_2-49.30\ast LPVOC+2.31\ast {\left(LPNOx\right)}^3-7.91\ast {\left(LP{PM}_{10}\right)}^3+7.50\ast {\left(LP{PM}_{2.5}\right)}^3+0.54\ast {\left(LPS{O}_2\right)}^3\begin{array}{c}+1.89\ast {\left(LPVOC\right)}^3+181.55\end{array}$$ (Equation 1)

In the BTH region, reductions in all pollutants except PM₁₀ are positively correlated with GDP growth. Moreover, the cubic-term coefficients for these pollutants are all positive, suggesting a potential decline in the marginal benefits of emission reduction at advanced stages when pollutant levels are significantly reduced.

$${LPGDP}_{YRD}=7.74\ast {\left(LPNOx\right)}^2-15.29\ast {\left(LPPM10\right)}^2+16.81\ast {\left(LPPM2.5\right)}^2-0.17\ast {\left(LPS{O}_2\right)}^2-1.52\ast {\left(LPVOC\right)}^2-2.31\ast {\left(LPNOx\right)}^3+9.99\ast {\left(LP{PM}_{10}\right)}^3-18.10\ast {\left(LP{PM}_{2.5}\right)}^3+0.16\ast {\left(LPS{O}_2\right)}^3\begin{array}{c}+0.37\ast {\left(LPVOC\right)}^3+3.78\end{array}$$ (Equation 2)

In the YRD region, a complex nonlinear relationship between pollutants and GDP was observed. The positive quadratic and negative cubic term coefficient of NOx indicate an inverted U-shaped relationship with GDP, characterized by an initial increase followed by a subsequent decline. For PM₁₀, the negative quadratic term coefficient and positive cubic term coefficient contrast with PM_2.5’s positive quadratic term and negative cubic term coefficients, suggesting fundamentally different impact mechanisms between coarse and fine particulate matter on GDP. These differences may stem from variations in emission sources, control challenges, and sectoral linkages. In contrast, SO₂ and VOCs show relatively small coefficient magnitudes, implying limited influence on GDP in this region.

$${LPGDP}_{PRD}=-21.45\ast {\left(LPNOx\right)}^2+14.98\ast {\left(LP{PM}_{10}\right)}^2-1.30\ast {\left(LPS{O}_2\right)}^2-86.84\ast {\left(LPVOC\right)}^2+6.88\ast {\left(LPNOx\right)}^3-9.19\ast {\left(LP{PM}_{10}\right)}^3\begin{array}{c}+0.06\ast {\left(LPS{O}_2\right)}^3+21.59\ast {\left(LPVOC\right)}^3+245.61\end{array}$$ (Equation 3)

In the PRD region, pollutants and GDP also exhibit a complex nonlinear relationship. Notably, VOCs show a strongly negative quadratic term coefficient and a substantially positive cubic term coefficient, which may be attributed to the region’s developed manufacturing, chemical, and service sectors—the primary sources of VOC emissions. For NOx, the negative quadratic term coefficient combined with the positive cubic term coefficient suggests relatively higher GDP levels at both low and high NOx emission stages. After long-term emission reduction efforts, the PRD has achieved compliance standards for particulate matter and SO₂. Consequently, the current focus of emission control strategies has shifted primarily toward the synergistic reduction of NOx and VOCs.

The functional model of the PRD exhibits highly nonlinear characteristics, suggesting that the region has transitioned from industrialization to post-industrialization, with pollutant emissions largely decoupled from economic growth. This model further implies that the PRD region may have already passed the EKC turning point, entering a stage where pollution decreases alongside sustained GDP growth.

$${LPGDP}_{FWP}=1.49\ast LPNOx+7.3953\ast LP{PM}_{10}-10.22\ast LP{PM}_{2.5}+0.14\ast LPS{O}_2+2.71\ast LPVOC-0.47$$ (Equation 4)

In contrast to other regions, the relationship between pollutants and GDP in the FWP exhibits a highly linear correlation, reflecting its resource-based economy with a dominant traditional industrial sector, where pollutant emissions remain strongly coupled with economic activity. The positive and relatively large coefficient for PM₁₀ contrasts with the negative and even larger absolute coefficient for PM_2.5, indicating fundamentally different impacts of coarse and fine particulate matter on GDP, with PM_2.5 potentially exerting greater inhibitory effects on economic activity. The positive coefficients for NOx, SO₂, and VOCs suggest that economic development in the FWP remains positively correlated with emissions of these pollutants.

The functional model of the FWP reveals a simple linear relationship between pollutants and GDP, characteristic of the region’s traditional industrial and energy base. This pattern further indicates the region’s relatively singular industrial structure and extensive economic growth pattern.

$${LPGDP}_{CC}=5.25\ast LPNOx-7.36\ast LPPM10+2.28\ast LPPM2.5+0.77\ast LPS{O}_2+51.79\ast LPVOC-0.14\ast {\left(LPNOx\right)}^3+0.38\ast {\left(LPPM10\right)}^3-0.24\ast {\left(LPPM2.5\right)}^3-0.04\ast {\left(LPS{O}_2\right)}^3\begin{array}{c}-2.09\ast {\left(LPVOC\right)}^3-92.64\end{array}$$ (Equation 5)

In the CC region, VOCs demonstrate a substantial primary-term coefficient but a negative cubic-term coefficient, indicating a strong positive correlation with GDP at low emission levels that gradually weakens as emissions increase. As an emerging industrialized region, VOC emissions in the CC primarily originate from industrial production and urbanization, and remain closely tied to GDP growth. NOx shows a similar but weaker pattern, with a positive primary term and a negative cubic term coefficient. The relatively small absolute coefficient for SO₂ suggests a limited impact on GDP in the CC.

The complexity of the functional model for the CC reflects its role as an economic growth pole in western China that is currently in the mid-industrialization stage, where pollutant emissions and economic growth remain partially coupled. The dominance of primary term coefficients indicates persistent linear effects, while the presence of cubic terms reveals emerging nonlinearity—suggesting that the region’s economy-environment relationship may soon transition from “economic development driving pollution increases” to “economic growth accompanied by pollution reduction.”

A counterfactual scenario analysis was also conducted to evaluate the economic impact of air pollution control policies, as shown in Figure S4. This scenario assumes no policy intervention after 2013, projecting anthropogenic emissions based on their linear growth trend from 2010 to 2012. Utilizing our established GDP-emission model, we simulated the corresponding economic trajectory under this “no-policy” baseline. The analysis reveals that, for all regions except the YRD, the counterfactual GDP path generally aligns with the actual trend observed during the APPCP period but exhibits a higher growth rate. This divergence indicates a measurable economic trade-off associated with China’s efforts to improve air quality. Interestingly, the YRD region presents a distinct case in which the counterfactual GDP follows an inverted U-shaped curve. This pattern suggests that while unchecked emissions might stimulate short-term economic growth, they would ultimately become a drag on the economy, highlighting a potential long-term growth penalty from severe pollution. These findings collectively underscore the necessity of sustained anthropogenic emission control, not only for environmental and health benefits but also as a strategic investment to secure a win-win future of enduring economic growth coupled with pollution reduction.

Quantification of the PM_2.5- and O₃-related economic loss

The economic losses attributable to PM_2.5 pollution from 2015 to 2020 reveal pronounced regional heterogeneity in both magnitude and composition (Figure 5). A significant downward trend was observed in the BTH region, where per capita health-related losses declined by approximately 27.2%–1006 yuan, while per capita fuel-related costs remained stable within the range of 380–480 yuan. In stark contrast, the PRD experienced a distinct upward trend in per capita health losses, which peaked at 2721 yuan. The YRD exhibited considerable volatility in per capita health losses alongside a sharp 53% increase in per capita fuel-related costs, reaching 664 yuan. The FWP presented a unique loss structure, characterized by substantial and growing per capita fuel-related losses, which increased by 44%–1104 yuan, far exceeding its relatively low health costs. Finally, the CC region recorded the most rapid growth in per capita health-related economic losses, correlating with its pace of urbanization and industrialization; although per capita fuel-related losses also grew, their relative share within the total economic burden decreased.

Figure 5.

The economic losses attributed to PM_2.5 pollution in the (A) BTH, (B) YRD, (C) PRD, (D) FWP, and (E) CC

As shown in Figure 6A, O₃ pollution in the BTH resulted in substantial economic losses, predominantly from health impacts. Health-related economic losses increased markedly by approximately 125.8 percent, rising from about 698 yuan/person in 2015 to a peak of 1689 yuan/person in 2019, before a slight decline to 1576 yuan/person in 2020. In contrast, losses associated with fuel consumption remained relatively stable, ranging from 60 to 80 yuan/person. Agricultural losses were negligible, consistently below 30 yuan/person, indicating a minimal effect on this sector.

Figure 6.

The economic losses and health damage attributed to O₃ pollution in the (A) BTH, (B) YRD, (C) PRD, (D) FWP, and (E) CC

The YRD, illustrated in Figure 6B, exhibited a similar pattern where health costs constituted the primary economic burden. Health-related losses generally increased from approximately 436 yuan/person in 2015 to about 785 yuan/person in 2020. Losses associated with fuel consumption averaged around 90 yuan/person, showing a slight upward trend likely connected to the region’s developed manufacturing and energy-intensive industries. As a comparison, agricultural losses remained low and stable.

In the PRD, health-related economic losses initially rose from roughly 1367 yuan/person in 2015 to 1895 yuan/person in 2019, then stabilized with a marginal decrease to 1775 yuan/person in 2020 (Figure 6C). Fuel-related losses were stable between 130 and 170 yuan/person, reflecting consistent regional energy consumption patterns. Agricultural losses were minor, consistent with the sector’s small share in the local economy. Health impacts overwhelmingly dominated the total economic cost of O₃ in the PRD.

While per capita health-related economic losses in the FWP were lower than in other regions, they followed a fluctuating upward trajectory, increasing from approximately 125 yuan in 2015 to 256 yuan in 2019 before declining to 238 yuan in 2020, as detailed in Figure 6D. Notably, the FWP exhibits a distinct loss structure characterized by relatively high per capita fuel-related losses, averaging between 130 and 170 yuan, and per capita agricultural losses that exceed those in other clusters, averaging 46 to 66 yuan. This pattern underscores the pronounced influence of the region’s specific energy profile and agricultural landscape on its economic vulnerability to O₃ pollution.

Finally, the CC region experienced the most rapid growth in per capita health-related economic losses among all clusters, as illustrated in Figure 6E. These losses surged by approximately 231.9 percent, from 188 yuan in 2015 to 624 yuan in 2020, a trend closely correlated with the area’s rapid urbanization and industrial expansion. Meanwhile, per capita losses from fuel consumption, ranging from 50 to 80 yuan, and agriculture remained stable and relatively low. The steep rise in health costs highlights the escalating economic burden associated with O₃ pollution amid accelerated development. It is important to note that this study quantifies the economic losses associated with O₃ pollution rather than directly quantifying losses from VOCs. This approach is taken due to the inherent complexity of VOCs, which comprise numerous species with vastly different chemical reactivities, making direct economic valuation highly challenging. As VOCs serve as key precursors to O₃ through photochemical reactions, which in turn cause measurable economic loss, quantifying the economic impact of O₃ provides a reasonable and indirect proxy for assessing the economic burden attributable to VOC emissions.

It should be noted that the economic loss estimates for PM_2.5 and O₃ in the five megacities were derived from a comprehensive literature review, providing representative order-of-magnitude ranges of regional GDP shares rather than precise welfare estimates. This approach is subject to uncertainties arising from differences in parameter selection and spatiotemporal representativeness across studies. Therefore, the economic losses presented in this study are primarily intended as order-of-magnitude approximations to reveal the relative economic burdens among the five megacities, shaped by disparities in industrial structure, pollution levels, and population exposure.

Discussion

The constructed pollutant emissions-GDP functional models reveal distinct patterns: rapidly developing regions (i.e., the YRD and PRD) predominantly exhibit nonlinear structures. In contrast, resource-dependent areas such as the FWP show simple linear relationships, while regions in mid-industrialization stages, such as the CC and BTH, exhibit hybrid structural characteristics.

Notably, all five megacity clusters display an inverted U-shaped relationship between VOC emissions and GDP, suggesting that China should intensify its ongoing efforts to reduce VOC emissions. This can be achieved by adopting low-VOC raw materials, optimizing production processes, and enhancing end-of-pipe VOC treatment technologies. While these patterns provide critical insights into the co-evolution of economic development and emission trajectories, the analysis is based on observational data and reduced-form modeling. Therefore, the identified relationships should not be interpreted as formal causal effects of a specific policy intervention, as such identification would require different methodological approaches to account for confounding factors.

This study utilizes air quality monitoring data provided by the China National Environmental Monitoring Center, which undergoes rigorous quality control procedures. It should be noted, however, that inherent methodological limitations in the monitoring process can introduce systematic biases into pollutant observations. For example, NO₂ concentrations measured by the chemiluminescence detection are commonly overestimated. This occurs due to the non-specific conversion of total reactive nitrogen species to NO by molybdenum oxide catalysts within the analyzer.²⁹

The MEIC is constructed using a bottom-up methodology that combines statistical activity data (e.g., energy consumption, industrial production, vehicle fleet) with technology-specific emission factors.^5,30 These uncertainties primarily stem from its bottom-up methodology, which relies on compiled statistical data on activity levels (e.g., energy consumption, industrial production) and emission factors.^31,32 Potential inaccuracies in underlying energy statistics, along with emission factors that may not fully represent the diversity and rapid technological evolution of China’s emission sources, can introduce systematic biases. Furthermore, the spatial and temporal allocation of emissions, which often depends on proxy datasets, adds another layer of uncertainty to the final gridded products.

While the inventory incorporates available measurements and local parameters where possible, its pre-2015 values are not derived from systematic, nationwide real-time monitoring networks, which were sparse and inconsistent at the time, but are instead estimates based on statistical reporting, fuel-use data, and modeled spatial-temporal allocation.

The scarcity of independent, high-quality observational data prior to 2015 makes direct validation of these emission estimates particularly challenging. Ground-based monitoring of criteria pollutants (e.g., SO₂, NO₂, PM) was limited in spatial coverage and standardization, while speciated VOC measurements were even rarer. Consequently, pre-2015 emission trends in MEIC inevitably carry greater uncertainty compared to the post-2015 period, when dense monitoring networks and satellite retrievals provided strong constraints on inventory accuracy, particularly for the years before and during the implementation of China’s stringent APPCA Plan. This data gap hinders a definitive evaluation of whether the inventory fully captures the pace and magnitude of emission changes driven by early policy interventions and economic shifts.

Additionally, VOC data are highly uncertain due to dispersed sources and complex chemical compositions. First, compared to the well-established monitoring networks for SO₂ and PM, high-resolution observational sites for VOCs remain limited, with variations in monitored species and methodologies. VOCs encompass numerous species with vastly different chemical reactivities (e.g., OH reaction rates) and O₃ formation potentials (OFPs).^15,33 Current commonly used monitoring techniques (such as online GC-FID/MS or offline sampling) cannot fully cover all relevant species. Second, VOC emission sources are extremely complex and fragmented, including industrial solvent use, vehicle exhaust, oil and gas evaporation, and biogenic emissions. The development of emission inventories for these sources heavily relies on activity data and emission factors, resulting in significantly higher uncertainty in VOC emission inventories compared to SO₂, which primarily originates from stationary combustion sources (e.g., power plants).^34,35 Third, ambient VOC species exhibit high chemical reactivity. During atmospheric transport from emission sources to receptor monitoring sites, they undergo substantial photochemical degradation, leading to varying degrees of depletion before detection. For instance, highly reactive species such as isoprene have atmospheric lifetimes of only a few hours, making accurate observational quantification particularly challenging and introducing additional uncertainty into emission inventory estimates.^8,36,37

It is important to clarify that while the functional relationship between emissions and GDP established in this study is derived directly from emission inventory and economic data without meteorological interference, the resulting economic losses associated with O₃ and PM_2.5 pollution are indeed subject to substantial meteorological influence. Specifically, meteorological conditions characterized by high temperatures, low humidity, and intense solar radiation significantly promote the photochemical formation of O₃ and secondary PM_2.5, thereby amplifying secondary pollution levels. Given the deteriorating O₃ pollution in China, its negative impacts on human health and economic costs warrant particular attention, especially against the backdrop of global warming. Numerous studies have confirmed that climate change has significantly increased the frequency of extreme heat events, providing favorable meteorological conditions for O₃ photochemical reactions and leading to further elevation of O₃ concentrations, thereby amplifying associated socioeconomic losses.^38,39 Furthermore, evidence indicates that O₃ pollution can impair the carbon sequestration capacity of terrestrial ecosystems, potentially intensifying the greenhouse effect.¹² Therefore, persistently advancing deep emission reductions of O₃ precursors, notably NOx and VOCs, will not only help curb the rising trend of O₃ concentrations but also contribute to enhancing carbon sink capacity, ultimately realizing the co-benefits of simultaneous pollution control and carbon mitigation. More critically, such events can directly inflict substantial economic losses. Furthermore, much research confirms that future meteorological conditions, and their subsequent impact on factors such as labor productivity, agriculture, and energy demand, will diverge significantly under different emission pathways, thereby indirectly influencing future GDP trajectories.^40,41

Therefore, while our current methodological approach provides a clear assessment of past decoupling based on anthropogenic drivers, we fully concur that achieving synergistic economic growth and emission reduction under climate change necessitates integrating meteorological and climatic considerations into forward-looking, integrated assessment and policy design.

Research in countries such as the United States and Europe has also validated the EKC, showing an inverted U-shaped relationship between economic development and environmental pollution (Pata, 2021). Many countries experienced a delayed turning point in VOC control, and there was even a phase in which O₃ concentrations increased rather than decreased. The United States has been managing anthropogenic VOCs for over 70 years; since the 1950s, it has progressively enacted various air pollution-related laws, culminating in the Clean Air Act (CAA). Subsequently, different control strategies were implemented based on regional needs. The US Environmental Protection Agency (US EPA) requires states to report on sources of air pollution emissions and develop related documents for pollution control. Since 1997, the US EPA has introduced multiple emission standards to regulate various pollution sources and has published detailed guidelines for alternative materials and process improvement technologies across several industries. Additionally, market-based mechanisms such as emissions trading, taxation, and green finance have been adopted to control pollution emissions. This multifaceted approach encompasses policy regulation, source substitution of processes, technological improvements, and end-of-pipe treatment strategies for pollution management. Drawing on the advanced experiences of Europe and the United States, China can enhance its environmental policies by implementing synergistic control of VOCs and NOx, establishing emission standards for VOCs and other pollutants across various industries, particularly focusing on high-pollution sectors such as chemical manufacturing, coating, and printing. In light of the regional transport characteristics of O₃, it would be beneficial to adopt control strategies from different U.S. regions to create a regional collaborative mechanism.

Based on the distinct emission-GDP trajectories observed across five megacity clusters, region-specific policy recommendations are proposed. For the YRD and PRD regions, which have entered a post-industrial stage and lead in decoupling from conventional pollutants, O₃ remains a persistent challenge. Policy should therefore shift toward precision management of multi-pollutant interactions, enforcing source-specific controls on VOCs from dominant sectors such as electronics, automotive, and services. In the resource-dependent FWP, where a strong linear coupling between emissions and economic growth persists, the fundamental policy direction must be energy system transformation. This involves phasing out coal-intensive industries and fostering a transition toward emerging industrial clusters such as advanced equipment manufacturing, electronics, and biomedicine, taking lessons from structural transformation experiences like those of Germany’s Ruhr region. For the rapidly industrializing CC area, the strategy should focus on green leapfrogging during industrial transfer, implemented through stringent environmental market access and investment in green infrastructure. Meanwhile, in the BTH region, which is in a transitional phase with preliminary but not yet fully stabilized decoupling, policy must deepen industrial upgrading and sustain intensive reduction of pollutants such as PM₁₀, while piloting coordinated regional prevention and control mechanisms informed by experiences from regional joint air quality management frameworks in Europe and the United States.

The decoupling indices for VOCs across all megacities are significantly higher than those for other pollutants. This can be primarily attributed to the generally later policy emphasis on VOC control. Furthermore, VOC emissions originate from highly dispersed sources, resulting in higher marginal abatement costs. Considering that OPS in China’s megacities is predominantly a VOC-limited regime, the marginal benefit of VOC reduction is substantial in the early stages of control, indicating significant potential for post-turning-point declines. In contrast, for regions such as the PRD, which may be transitioning toward NOx-limited regimes, the marginal benefit of VOC reduction diminishes, explaining the more gradual decline observed after the EKC turning point. Therefore, when formulating emission reduction strategies, it is essential to account for regional differences in OPS and design localized, dynamic mitigation pathways, prioritizing VOC reduction initially and gradually shifting the focus to NOx control in later stages. Given the complexity of VOC speciation and the significant variation in chemical reactivity among different species, it is recommended to refine and complete speciated VOC emission inventories. During the implementation of coordinated VOC and NOx control, more efforts should prioritize the targeted reduction of highly reactive VOC species. Furthermore, to address the uncertainties in emission inventories, top-down constraints from satellite observations can be employed to optimize them and enhance their spatiotemporal resolution. This improvement in data quality will increase the accuracy of the GDP-emission relationships characterized by EKC, thereby providing essential data support for developing scientifically sound O₃ mitigation strategies and improving the precision of air quality forecasting.

Limitations of the study

Several limitations of this study should be acknowledged. First, the MEIC emission inventory used for the pre-2015 period relies heavily on statistical modeling rather than direct observational constraints due to the limited coverage of real-time monitoring data, which may introduce uncertainties, particularly for VOC emissions. Second, the economic loss estimates for PM_2.5 and O₃ are derived from a synthesis of existing literature rather than direct health or agricultural impact modeling using our own data, which limits the precision of absolute welfare estimates. Third, the EKC relationships identified in this study are empirical associations based on historical data and should not be interpreted as formally identified causal effects, as unobserved confounding factors may influence the results. Future research should aim to integrate high-resolution exposure-response functions, satellite-constrained emission inversions, and integrated assessment models to further reduce uncertainties and enhance the causal interpretability of the observed decoupling pathways.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Kaihui Zhao (khzhao@ynu.edu.cn).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •Data: All data reported in this paper will be shared by the lead contact upon reasonable request.
• •Code: All analysis and plotting codes have been archived and made publicly available: https://github.com/xdnn369/code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This work was supported by the National Key R&D Program of China (no. 2023YFF0805504), the National Natural Science Foundation of China (no. 42465008), the National Natural Science Foundation of China (no. 42105164), the Yunnan Science and Technology Department Project (no. 202501AT070239), the Yunnan Science and Technology Department Youth Project (no. 202401AU070202), the Xianyang Rapid Response Decision Support Project for Ozone (no. 2022A1515011078), and the Xianyang key research and development program (no. L2022ZDYFSF040).

Author contributions

D.X.: methodology, investigation, software, validation, formal analysis, and writing - original draft. Y. Ma: supervision, validation, formal analysis, and writing - review and editing. Z. Fan: supervision, conceptualization, and methodology; B. Wang: software and validation. H. Zhang: methodology and validation. N. Wang: validation and formal analysis. X. Zhang: supervision and methodology. K. Zhao: methodology, formal analysis, and writing - review and editing.

Declaration of interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Software and algorithms
Air quality monitoring data	National Urban Air Quality Real-time Publishing Platform	http://www.cnemc.cn/sssj/
Multi-resolution Emission Inventory for China	Tsinghua University	http://meicmodel.org.cn
GDP and utilized population census data	National Bureau of Statistics of China	https://www.stats.gov.cn/

Experimental model and study participant details

This study does not involve experimental models or study participants. It is based on ambient atmospheric monitoring data and computational modeling.

Method details

Study area and data

This study focuses on five megacities in China: the Beijing-Tianjin-Hebei region (BTH), Yangtze River Delta (YRD), Pearl River Delta (PRD), Fenwei Plain (FWP), and Chengdu-Chongqing (CC) urban cluster, as illustrated in Figure S1 and Table S1.

Air quality monitoring data were sourced from the National Urban Air Quality Real-time Publishing Platform (http://www.cnemc.cn/sssj/), operated by China National Environmental Monitoring Center. This authoritative platform has provided real-time monitoring data since 2015 from environmental monitoring stations across major Chinese cities, including hourly resolved Air Quality Index (AQI) values and concentration measurements of six criteria pollutants: PM_2.5, PM₁₀, SO₂, NO₂, CO, and O₃. To ensure data integrity and consistency, this study implemented rigorous procedures for handling missing values and resolving data inconsistencies. Temporal gaps in the observational dataset were addressed using linear interpolation for short consecutive gaps, while longer periods of missing data were filled based on historical hourly averages from the same calendar date across multiple years.

The emission data used in this study were obtained from the Multi-resolution Emission Inventory for China (MEIC, http://meicmodel.org.cn), a high-resolution anthropogenic emission database for atmospheric pollutants and greenhouse gases developed by Tsinghua University. This study utilized the MEIC emission inventory for the period 2010 to 2020. The dataset provides monthly gridded emissions at a 0.25 by 0.25° resolution for key pollutants including NO_x, PM₁₀, PM_2.5, SO₂, and VOCs, covering major anthropogenic sectors such as power generation, industry, residential activities, transportation, and agriculture.

The MEIC emission inventory employed in this study spans the period 2010–2020, whereas nationwide real-time air quality monitoring data from the China National Environmental Monitoring Center are only available from 2015 onward. Consequently, for the analysis of air quality changes during the early policy implementation phase (2013–2014), we primarily rely on estimated trends derived from the emission inventory. Although pre-2015 monitoring data were not directly used, a recent study by Geng et al.⁵ demonstrates that the emission trends of five major air pollutants in China during 2013–2020 are broadly consistent with satellite observations and ground-based measurements. Overall, the trends in key pollutant emissions reflected in the MEIC inventory align well with the directional changes observed from remote sensing and regional surface observations, thereby enhancing the credibility of the trend analysis for the early study period.

Population sources and calculation method

The GDP and utilized population census data from National Bureau of Statistics of China (https://www.stats.gov.cn/) to normalize the indicators. The population figures used for the study regions are as follows: BTH at 112 million, YRD at 230 million, PRD at 86.44 million, CC at 91 million, and FWP at 55.54 million.

Per-capita emissions were calculated by dividing the total anthropogenic emissions (derived from the MEIC inventory) by the total resident population of the respective regions. The population data were obtained from the annual data published by the National Bureau of Statistics of China (https://www.stats.gov.cn/). Consequently, the calculation and alignment were performed on an annual basis, matching the yearly resolution of both the emission inventory and the population statistics. The specific calculation formula is as follows:

$$E_p=\frac{E_{meic}}{p}$$ (Equation 6)

E_p denotes the per capita pollutant emissions(tons/person); E_meic represents the annual total pollutant emissions in the MEIC inventory for the region(tons); Pindicates the total resident population of the region in the corresponding year.

Environmental Kuznets Curve (EKC) analysis

The Environmental Kuznets Curve (EKC) hypothesis examines the relationship between economic development and environmental quality (Dinda, 2004). This theoretical framework suggests an inverted U-shaped pattern: during the early stages of economic development, environmental degradation tends to increase with rising per capita income, primarily due to industrialization and resource-intensive growth. However, upon reaching a critical income threshold, continued economic development leads to environmental improvement through structural economic transitions, technological advancements, and strengthened environmental regulations, resulting in the characteristic inverted U-curve.

The core equation of the EKC theory can be expressed as

$$\begin{array}{c}\ln \left(E\right)={\beta }_0+{\beta }_1\ln \left(GDP\right)+{\beta }_2{\left[\ln \left(GDP\right)\right]}^2+\epsilon \end{array}$$ (Equation 7)

Here, E denotes per capita pollutant emissions, GDP denotes per capita gross domestic product, β₀, β₁, and β₂ are regression coefficients, and ℇis the random error term. When β₁ > 0 and β₂ < 0, the relationship between pollution and economic development follows an inverted “U”-shaped pattern, consistent with the EKC hypothesis. The turning point of the inverted U-shaped curve indicates the level of economic development at which emissions peak, calculated in logarithmic form as – β₁ / (2 β₂).

In more complex scenarios, the EKC may exhibit an “N”-shaped pattern, suggesting that environmental degradation could re-emerge at advanced stages of economic growth. To capture such nonlinear dynamics, the quadratic model can be extended with a cubic term, yielding the following enhanced model formulation:

$$\begin{array}{c}\ln \left(E\right)={\beta }_0+{\beta }_1\ln \left(GDP\right)+{\beta }_2{\left[\ln \left(GDP\right)\right]}^2+{\beta }_3{\left[\ln \left(GDP\right)\right]}^3+\epsilon \end{array}$$ (Equation 8)

Tapio decoupling elasticity indice calculation

The Tapio model combines absolute and relative changes, employing elasticity analysis over specific time periods to represent the decoupling relationships between variables(Tapio, 2005). Tapio proposed the concept of decoupling elasticity and further refined the classifications of decoupling states, allowing for a more precise analysis of the relationship between pollutants emissions and economic growth. In this study, the Tapio decoupling model is used to quantitatively analyze the relationship between economic growth and pollutant emissions in the five megacities.

The Tapio decoupling elasticity indice e describes the percentage change in pollutant emissions corresponding to a 1% increase in GDP, calculated as follows.

$$e=\Delta Emi/\Delta CDP$$ (Equation 9)

where ΔEmi represents the percentage change in emissions and ΔGDP denotes the percentage change in GDP.

The decoupling state can be classified into distinct categories: strong decoupling (economic growth with emission reduction, e < 0), weak decoupling (economic growth with low emission increase, 0 < e < 0.8), expansive decoupling (economic growth with emission increase, 0.8 < e < 1.2), and expansive negative decoupling (economic growth with rapid emission increase, e > 1.2).

To enhance the robustness of the decoupling analysis, we conducted a rolling window analysis with a five-year window length. Specifically, decoupling elasticity indices between pollutant emissions and GDP growth were calculated sequentially for seven overlapping time windows: 2010–2014, 2011–2015, 2012–2016, 2013–2017, 2014–2018, 2015–2019, and 2016–2020. This procedure was applied individually to each pollutant across all megacities.

Identification of the PM_2.5- and O₃-related economic loss

As shown in Table S2 and S3, the economic losses attributable to PM_2.5 pollution were estimated by many studies, which quantify impacts primarily in two dimensions: health and fossil fuel consumption.^{16,42,43,44,45,46,47,48,49,50} Health-related economic losses vary considerably across the five megacities. The PRD exhibits the highest proportion at 6.1%, followed by the BTH, YRD, and CC regions, all within the range of 2.6%–5.0%. In contrast, the FWP shows a notably lower health loss proportion, between 0.37% and 0.55%. These estimates are drawn from recent peer-reviewed studies applying localized exposure-response functions and valuation parameters. Farrow et al.⁴⁵ estimated that economic losses attributable to PM_2.5 pollution from fossil fuel combustion accounted for 2.5% of GDP.

Many studies have estimate the economic losses attributable to surface O₃ pollution across the five megacities.^{14,16,42,43,44,45,46,47,48,50,51,52,53} As summarized in Table S3, health-related economic losses, expressed as a percentage of regional GDP, show significant spatial variation. The estimated loss is highest in the PRD at 4.2%, followed by the BTH (2.7%–4.9%) and the YRD (2.4%–3.2%). In contrast, the FWP and CC regions exhibit substantially lower health loss proportions, ranging from 0.41%–0.75% and 0.55%–1.1%, respectively. Previous studies have also estimated that O₃-induced economic losses account for approximately 0.15% of GDP in the agricultural sector and 0.4% from fossil fuel combustion.

Quantification and statistical analysis

All statistical analyses were performed using MATLAB (R2021b) and NCL (version 6.6.2). The Environmental Kuznets Curve (EKC) relationship between per capita emissions and GDP was estimated using polynomial regression, with model selection based on AIC and BIC criteria. Turning points were calculated from significant cubic terms (p < 0.05), and 95% confidence intervals were derived using the delta method with bootstrap validation (1,000 replications). Tapio decoupling elasticities were computed with a five-year rolling window analysis to assess temporal stability, and robust regression was employed to mitigate outlier effects. Sensitivity analyses for VOC emission uncertainties were conducted using low- and high-emission scenarios based on literature-reported uncertainty ranges (68–76%). Population-weighted exposure indicators were calculated using gridded census data. Further statistical details, including R² values, sample sizes (n = 11 per region), and confidence intervals, are provided in the corresponding tables and figures.