A Multi-Agglomeration Assessment of Air Quality Responses to Top-Down NO_x Emission Changes: Insights from Trends in Surface NO₂ and O₃ Across Urban China (2014–2021)

Abstract

China’s stringent clean air policies have substantially reduced nitrogen oxides (NO_x) emissions, leading to a general decline in nitrogen dioxide (NO₂). However, surface ozone (O₃) pollution remains severe, creating a complex challenge due to the non-linear relationship between O₃ and its precursors. To disentangle the drivers behind these trends, this study quantifies the impacts of interannual variations in top-down constrained NO_x emissions on surface NO₂ and O₃ concentrations from 2014 to 2021 across mainland China and five national urban agglomerations. We employed the WRF-CMAQ model with a fixed-emission simulation approach, using an observationally optimized NO_x emission inventory derived from the assimilation of surface NO₂ measurements. Results reveal that NO₂ reductions were predominantly emission-driven (>80% post-2017), with declines most pronounced in winter. A strong linear consistency was found between interannual changes in top-down NO_x emissions and attributed NO₂ concentration variations, validating the methodology. In contrast, O₃ responses to NO_x reductions were spatially and seasonally heterogeneous, reflecting a non-linear photochemical regime. In major urban agglomerations (e.g., Beijing–Tianjin–Hebei (BTH), Yangtze River Delta (YRD), Pearl River Delta (PRD)), NO_x reductions post-2018 showed limited effectiveness in mitigating summertime O₃ and even increased O₃ in spring and autumn, indicating a prevalent VOC-sensitive regime where NO_x reduction can disinhibit O₃ formation. Conversely, certain provinces (e.g., Anhui, Shanxi, Jilin) exhibited O₃ decreases, suggesting a NO_x-sensitive regime. The area benefiting from NO_x reductions expanded steadily in summer after 2017 but not in other seasons. This study confirms the efficacy of NO_x-focused policies for reducing primary NO₂ pollution but highlights that mitigating persistent O₃ requires a strategic shift to synergistic, region-specific control of volatile organic compounds alongside NO_x, informed by local chemical sensitivity.

1. Introduction

Air pollution is one of the leading environmental threats to humanity, with profound impacts on socioeconomic productivity. China has implemented a series of stringent national clean air policies, most notably the Action Plan for Air Pollution Prevention and Control (2013–2017) and the Three-year Action Plan to Win the Battle for a Blue Sky (2018–2020) [1,2]. These policies have targeted substantial reductions in primary pollutant emissions, particularly sulfur dioxide and nitrogen oxides (NO_x) [3], leading to a significant nationwide decline in fine particulate matter (PM_2.5) concentrations [4,5]. Accompanying this progress, ambient nitrogen dioxide (NO₂) levels, primarily emitted from fossil fuel combustion, have also shown a general decreasing trend [6]. However, surface ozone (O₃) pollution has emerged as a new critical air quality challenge, with concentrations persistently high or even increasing in many urban and regional hotspots across China during the same period [7,8]. This phenomenon has been extensively documented through studies of the weekend effect [9,10]. Researchers have observed higher ozone concentrations on weekends than weekdays in many urban areas, despite lower NO_x emissions from reduced traffic and industrial activity [11,12]. Early studies in Los Angeles, New York, and other U.S. cities established that this weekend ozone excess arises from the NO_x-saturated chemistry typical of urban cores [9]. Subsequent research has confirmed similar patterns across Europe, Japan, and China [13,14]. This contrasting trend presents a complex scientific and policy dilemma, as the photochemical production of O₃ is non-linearly related to its precursors, NO_x and volatile organic compounds (VOCs) [15]. Consequently, the effectiveness of NO_x focused control strategies in mitigating concurrent O₃ pollution remains ambiguous and regionally variable, necessitating a thorough investigation into the attribution of observed concentration trends.

Emerging evidence indicates that air pollutants like O₃ and NO₂ serve as leading contributors to air pollution-related acute mortality [16]. Quantifying the drivers behind the trends of NO₂ and O₃ is essential for evaluating policy efficacy and informing future strategies. Interannual variations in observed concentrations are governed by a combination of changes in anthropogenic emissions and fluctuations in meteorological conditions, which influence atmospheric dispersion, chemical transformation, and regional transport [17,18]. Globally, the intricate relationship between NO_x emissions and O₃ constitutes a central challenge in atmospheric chemistry and air quality management. While aggressive emission controls in North America and Europe have successfully reduced peak O₃ levels over recent decades, background O₃ concentrations have increased or stabilized, and some urban areas continue to experience non-attainment [19,20]. In developing regions of South and Southeast Asia, rapidly growing emissions coupled with favorable meteorology have led to widespread and worsening O₃ pollution [21,22]. This global pattern underscores that the response of O₃ to changes in its precursors is not linear but is mediated by the local chemical regime, shifting from NO_x-saturated to NO_x-sensitive conditions depending on the relative abundance of precursors [23]. Therefore, quantifying the impact of emission changes on pollutant trends requires disentangling them from the confounding effects of meteorological variability, a challenge pertinent to air quality studies worldwide.

Substantial research has been conducted globally to attribute air pollutant trends to emissions and meteorology [24,25,26]. A range of statistical modeling tools, from traditional regression to computational frameworks like learning techniques [27,28,29], use long-term observational data to derive meteorology-normalized trends, effectively filtering out the meteorological influence [4,30,31]. While valuable, these methods may not fully capture complex, non-linear chemical interactions. In contrast, chemical transport models (CTMs) provide a process-based, physical framework to simulate atmospheric chemistry and transport [32,33]. A powerful application of CTMs is holding emissions constant at a baseline level while using actual, interannually varying meteorological fields to directly quantify the pollutant variability induced by meteorology alone [6]. The difference between simulations with varying emissions and these fixed-emission runs then isolates the emission-driven signal [34]. Studies in North America and Europe have successfully separated the influences of emission controls and meteorological variability on long-term O₃ and NO₂ trends, highlighting the dominant role of emission reductions [35,36,37]. This method has increasingly been used in Asia [38,39]. However, its application in assessing the impacts of recently implemented, stringent policies across China’s diverse urban agglomerations remains limited, particularly concerning the dual effects on NO₂ and O₃ over an extended, policy-relevant period.

Furthermore, the accuracy of such attribution studies heavily depends on the quality of the emission inventory input. While bottom-up inventories are essential and have been widely used [3,40,41], they are subject to uncertainties and may not fully capture recent, rapid emission changes due to policies [42,43]. Top-down constraints on emissions, derived from assimilating satellite or surface observations into models, offer a valuable approach to reduce these uncertainties and provide more realistic spatiotemporal variations in emissions for trend analysis [44,45,46,47,48]. A synthesis of the fixed-emission simulation method with a top-down optimized emission inventory presents a robust framework for attribution but has not been extensively applied to a multi-year and multi-region analysis in China.

Thus, this study aims to systematically evaluate the impacts of interannual variations in top-down constrained NO_x emissions on the trends of surface NO₂ and O₃ concentrations from 2014 to 2021 across mainland China, especially five typical and economically significant urban agglomerations—Beijing–Tianjin–Hebei urban agglomeration (BTH), Yangtze River Delta urban agglomeration (YRD), Yangtze River Middle Reach urban agglomeration (YRMR), Cheng-Yu urban agglomeration (CY), and Pearl River Delta urban agglomeration (PRD). We employ the Weather Research and Forecasting (WRF) model and Community Multiscale Air Quality (CMAQ) model in conjunction with a fixed-emission simulation approach. A critical foundation is the use of an observationally constrained, top-down NO_x emission inventory derived from the assimilation of hourly surface NO₂ measurements [49], which better represents real-world emission changes. The specific objectives are to isolate and quantify the relative impacts of anthropogenic NO_x emission variations and meteorological factors on interannual NO₂ trends in different regions and seasons; evaluate the consistency between the interannual variations in optimized NO_x emissions and the attributed NO₂ concentration changes; elucidate the spatially and seasonally heterogeneous impacts of NO_x emission changes on O₃ pollution patterns across the nation and within key regions; and discuss the implications of the non-linear O₃ chemistry for future synergistic control strategies. By integrating advanced emission constraints with a process-based attribution method, this work seeks to provide a clearer mechanistic understanding of the drivers behind recent air quality trends in China, offering insights that are relevant for policy evaluation and for air quality management in other regions facing similar complex pollution challenges.

2. Materials and Methods

2.1. Data

2.1.1. Ground Measurement Data

Real-time NO₂ and O₃ concentrations data were obtained from the National Urban Air Quality Real-time Release Platform, maintained by the China National Environmental Monitoring Centre. The dataset, covering 31 provincial-level administrative divisions in mainland China (excluding the regions of Hong Kong, Macao, and Taiwan), sourced from the official website (https://air.cnemc.cn:18007/, accessed on 17 March 2026). The monitoring network exhibits a heterogeneous spatial distribution, with station density higher in the economically advanced eastern coastal regions but markedly lower in the central and especially western areas. All stations, with sampling ports positioned 3–20 m above ground level, provided hourly measurements [50,51,52]. In accordance with the pollutant concentration specifications outlined in China’s Ambient Air Quality Standards 2012, a stringent quality assurance procedure was applied following the methods described in a previous study [6], removing outliers for both species and retaining only monitoring sites with continuous multi-year records.

The performance of the CMAQ model was evaluated by comparing simulated concentrations with ground-based observations. To ensure a consistent comparison, a rigorous spatiotemporal matching protocol was implemented. Spatially, for each monitoring site, the simulated value was extracted from the CMAQ grid cell encompassing its geographic coordinates. Temporally, hourly model outputs were paired with observations from the corresponding hour, and these matched hourly pairs were subsequently aggregated into monthly statistics for further assessment.

2.1.2. Meteorological Data

The meteorological simulations utilized the initial and lateral boundary conditions sourced from the National Centers for Environmental Prediction (NCEP) Final Operational Global Analysis datasets. The datasets are produced by the Global Data Assimilation System, which integrates observational data from multiple platforms, including ground stations, satellites, and aircraft. The resulting products provide global coverage at a 1° × 1° spatial resolution and 6-hourly temporal intervals (available from NCEP at http://rda.ucar.edu/datasets/ds083.2/, accessed on 17 March 2026). The key variables comprise thermal variables (including air temperature, sea surface temperature, and soil temperature), dynamic variables (such as wind speed and direction, surface pressure, and sea-level pressure), and relative humidity, etc.

For validation of the simulated meteorological conditions, observed data from ground stations across China were downloaded from the public FTP repository of the National Climatic Data Center (available from NCDC at ftp://ftp.ncdc.noaa.gov/pub/data/noaa/isd-lite/, accessed on 17 March 2026). The variables include thermal variables (air temperature, dew point), dynamic variables (wind speed, wind direction, atmospheric pressure), precipitation, and cloud cover. The records were incorporated from more than 400 stations, most of which offer data at 3 h intervals in recent years, with a limited number of stations providing measurements at 1 h resolution.

2.1.3. Anthropogenic Emission Data

The NO_x emissions were derived from our earlier optimization of mainland China’s emission inventory constrained by surface hourly NO₂ measurements [49]. These optimized NO_x emissions were inferred by the Regional Air Pollutant Assimilation System (RAPAS), a system designed to quantitatively optimize regional-scale air pollutant emissions [53]. RAPAS is built upon the integrated WRF-CMAQ modeling framework (see Section 2.2) and employs the three-dimensional variational algorithm along with the ensemble square root filter methodology. RAPAS can simultaneously assimilate multiple hourly scale observations of surface air pollutant concentrations to retrieve daily emissions of various air pollutants at regional scales. The overall performance of the optimized emissions has been evaluated by forward simulations of atmospheric pollutant concentrations to indirectly assess the accuracy against ground measurements. Results from our previous studies demonstrate that simulating concentrations of different species using the retrieved emissions leads to significant improvements in site -averaged mean bias, root mean square error, and correlation coefficient [47,49,54].

Multi-resolution Emission Inventory (available from MEIC at http://meicmodel.org/ [3,55], accessed on 17 March 2026) developed by Tsinghua University was used as a priori emission over mainland China, while for other East Asian regions, the MIX inventory was utilized [21]. Biogenic emissions were derived from the Model of Emissions of Gases and Aerosols from Nature (MEGAN) model [56]. Biomass burning emissions were not included based on the lower fire counts in mainland China compared to other Asian regions.

2.1.4. Satellite Data

The satellite-derived ratio of formaldehyde (HCHO) to NO₂ column concentrations, denoted as the formaldehyde-NO₂ ratio, is widely employed to delineate ozone formation sensitivity regimes [57,58,59]. The analysis employed HCHO and NO₂ data acquired from the Ozone Monitoring Instrument (OMI) spanning the years 2014 to 2021, specifically focusing on spring (March, April, May), summer (June, July, August), and autumn (September, October, November) seasons. The HCHO dataset was acquired from the OMHCHOd product (available at https://disc.gsfc.nasa.gov/datasets/OMHCHOd_003/summary, accessed on 17 March 2026), which provides quality-controlled, gridded vertical column densities at a spatial resolution of 0.1° × 0.1°. The NO₂ data were obtained from the QA4ECV product (available at https://www.temis.nl/airpollution/no2col/no2regioomimonth_qa.php, accessed 17 March 2026), providing the monthly mean datasets with a spatial resolution of 0.125° × 0.125° distributed by the Royal Netherlands Meteorological Institute.

2.2. WRF-CMAQ Model

The regional chemical transport model WRF-CMAQ was employed in this investigation, which, developed by the U.S. Environmental Protection Agency, operates as a three-dimensional Eulerian “one-atmosphere” framework designed to simulate complex multi-pollutant processes and address interrelated air quality issues. Meteorological drivers for CMAQ (version 4.7.1) were provided by the WRF (version 3.5) model, a mesoscale numerical weather prediction system widely used for atmospheric research and operational forecasting. The initial and lateral boundary conditions required for the WRF model were derived from the National Centers for Environmental Prediction (NCEP) Final (FNL) Operational Global Analysis data, with a spatial resolution of 1° × 1° and a temporal interval of 6 h. For the CMAQ chemical simulations, the chemical initial and lateral boundary conditions were generated using the default, static clean-air concentration profiles provided by the CMAQ default dataset. Because our modeling domain encompasses the entirety of East Asia (Figure S1), the focal urban agglomerations in central and eastern China are located sufficiently far from the lateral boundaries. Given the short atmospheric lifetime of key precursors like NO_x, this large spatial buffer effectively minimizes the direct influence of boundary transport on the simulated anthropogenic pollution trends within the target regions. To further minimize the impact of initial conditions, each simulation period was preceded by a 20-day spin-up run.

The simulation covered a major portion of East Asia with a horizontal grid spacing of 36 km × 36 km (Figure S1), which is considered sufficient to capture the regional transport and synoptic-scale meteorological features influencing the major urban agglomerations in China [45,49,53]. The CMAQ configuration used 15 sigma-pressure layers vertically, derived by compressing the WRF model’s original 51-layer structure while maintaining the same model top at 50 hPa. Of these 15 layers, 7 are located within the planetary boundary layer, typically extending up to approximately 1–2 km above the ground. The remaining layers are situated in the free troposphere, which spans from the top of the boundary layer to the tropopause, at an altitude of roughly 10–12 km. The simulation employed the Carbon Bond 05 mechanism with updates to toluene reactions (CB05tucl) to represent gas-phase chemistry, and particulate matter was simulated using the sixth-generation aerosol module (AERO6) [60,61]. Additionally, biogenic emissions were derived from MEGAN version 2.1 [56], a modeling framework used to estimate biogenic emissions exchanged between terrestrial ecosystems and the atmosphere. It simulates key known processes controlling biogenic emissions via relatively simple mechanistic algorithms [56,62], and it accounts for environmental influences on these emissions by parameterizing meteorological inputs, which were simulated by the WRF model in this study.

2.3. Method

To separate the contributions of emissions and meteorology to the interannual variation of NO₂ concentrations, two simulation scenarios were conducted using a fixed-emission simulation approach (

Table 1. Simulation under different emission scenarios.

Scenarios	Number of Years	Year of Meteorological Data	Year of NOx Emissions
Base	8	2014–2021	2014–2021
FixedEmis	7	2015–2021	2014

), namely a base (Base) simulation and a fixed-emission (FixedEmis) simulation. For the Base simulation, daily continuous simulations from 2014 to 2021 were performed using annual-varying meteorological conditions and inversion-optimized emissions for each corresponding year. For the FixedEmis simulation, daily continuous simulations from 2015 to 2021 were conducted using annual-varying meteorological conditions while keeping emissions fixed at the inversion-optimized 2014 level. The starting point of 2014 was selected as it provides the first full year of consistent monitoring data following the 2013 network launch and the implementation of the Action Plan for Air Pollution Prevention and Control [1]. The Base simulation, incorporating both annual varying meteorology and corresponding emissions, provides annual NO₂ concentrations for each year. Differences in NO₂ concentrations between adjacent years in the Base simulations arise from the interplay of meteorological variability and changes in emissions. The FixedEmis simulation provides NO₂ concentrations under the meteorological conditions of each year from 2015 to 2021 with emissions held constant at the 2014 level. By comparing NO₂ concentrations between adjacent years in the FixedEmis simulation, the year-to-year differences in NO₂ concentrations attributable to changes in meteorological conditions can be derived. By subtracting the meteorology induced changes obtained from the FixedEmis simulation from these combined changes, the emission related changes in NO₂ concentrations—i.e., the quantitative contribution of emission changes to NO₂ concentration variation—can be isolated. The model domain and the physical and chemical configurations used in the simulations are consistent with those described in Shen et al. (2023) [63].

3. Results

3.1. Model Evaluation

Meteorological processes significantly influence pollutant transport, mixing, and photochemical reactions, making the performance of the WRF model critical for accurate pollutant simulations. In our previous studies [6,49], the performance of the model over the study region was evaluated by comparing simulated 2 m temperature (T2), 2 m relative humidity (RH2), and 10 m wind speed (WS10) against ground-based observations. The results demonstrated that the WRF model successfully reproduces T2 and RH2, exhibiting relatively low bias and high correlation coefficients. Although WS10 is slightly overestimated, a tendency also reported in other WRF-based studies [64,65], the resulting enhanced pollutant dispersion and transport partly offset the potential impacts of biases in T2 and RH2. Following the criteria proposed by Emery et al. (2001) [66], the performance of the WRF model in simulating meteorological parameters in this study is acceptable. Furthermore, the yearly statistical results remain consistent across different years, indicating that the WRF simulations reliably capture interannual variations in weather conditions.

Figure 1 presents the evaluation results of simulated NO₂ concentrations for each urban agglomeration in the Base simulations. Compared to the simulations using prior emissions (MEIC), the posterior simulations utilizing optimized emissions show excellent agreement with ground-based observations. The correlation of monthly mean concentrations between the posterior simulations and observations exceeds 0.95 across all urban agglomerations. According to widely adopted performance benchmarks in regional air quality modeling [67,68], a correlation coefficient (CORR) greater than 0.50 is considered acceptable, while a value exceeding 0.60 to 0.70 satisfy the rigorous criteria for high-fidelity model performance. Therefore, our model effectively captures the interannual variation characteristics of NO₂ concentrations in each region (Figure 1). As shown in Figure 2, the posterior CORR frequently reaches 0.99 for individual agglomerations, markedly higher than prior correlations. While absolute metrics do not have universally fixed thresholds due to their dependence on regional concentration baselines, the site-averaged bias in each urban agglomeration is also significantly reduced relative to the prior simulations, particularly in the Yangtze River Basin and the Pearl River Delta. Similar improvements are evident in the mean bias (BIAS) and the root mean square error (RMSE), with posterior RMSE values consistently lower than the prior counterparts, reflecting analogous reductions and robust model optimization (Figure 2). In terms of annual mean concentrations, NO₂ levels began to decline after 2014, exhibited a slight rebound in 2017, and then continued to decrease thereafter. Both ground observations and posterior simulations demonstrate highly consistent trends in NO₂ concentration changes. The above statistical analysis indicates that the Base simulations reliably capture the trend characteristics of NO₂ concentrations over the study period.

For O₃ concentrations, the simulated results show high consistency with observations across all urban agglomerations and the whole country (Figure S2). The correlation coefficients exceed 0.80 in all urban agglomerations except PRD. Specifically, the correlation coefficient reaches 0.81 in YRMR, while exceeding 0.87 in others. Nevertheless, some discrepancies were noted in the O₃ simulations, with varying magnitudes across regions. The model tended to overestimate observed O₃ levels in certain areas and periods. These biases are likely attributed primarily to uncertainties in O₃ precursor emissions. For instance, the resolution of the emission inventory may inadequately resolve fine-scale spatial variations in urban areas. Additionally, deviations between simulated and observed wind speeds could further contribute to the overestimation of O₃ concentrations.

Monthly average NO₂ observational data and corresponding FixedEmis simulation results were extracted for the whole country and five national urban agglomerations for the evaluation of interannual NO₂ variations driven by different meteorological conditions under the FixedEmis simulations (Figure 3). Detrending analysis was performed on the monthly mean NO₂ observations and FixedEmis simulation data to remove trends derived from linear regression models [69], thereby eliminating the influence of anthropogenic emission controls on interannual NO₂ concentration changes. Monthly anomalies of the detrended observations and FixedEmis simulations relative to the respective monthly averages during 2014–2021 were further calculated to reflect meteorologically driven interannual variations in NO₂ concentrations. The correlation coefficients between the detrended observational anomalies and FixedEmis simulation results ranged from 0.82 to 0.95 across different regions, indicating good agreement between observations and simulations in all areas. These results demonstrate that our FixedEmis simulations effectively capture the meteorologically driven interannual NO₂ variations across all studied regions.

3.2. Impact of Interannual Variations in NO_x Emissions on NO₂ Pollution

3.2.1. NO_x Emission Trends of Five National Urban Agglomerations

Figure 4 illustrates the NO_x emission trends for the five national-level urban agglomerations from 2014 to 2021. BTH exhibited the highest emission levels throughout the study period, peaking at 2900 kt in 2016 before declining sharply to 1760 kt in 2021, with a reduction of 39% from the peak. This substantial decline reflects the region’s stringent emission control measures, particularly in the industrial and transportation sectors [70]. In contrast, YRD and PRD displayed U-shaped patterns, with emissions increasing slightly during 2014–2017 before transitioning to continuous declines after 2017. By 2021, NO_x emissions in YRD had fallen to 1910 kt, while PRD reached nearly 800 kt. YRMR and CY showed more fluctuating trends. The emissions in YRMR ranged between 1910 and 2020 kt during 2014–2018, declined to a low value in 2020, but rebounded in 2021. Similarly, the emissions in CY increased from 2014 to a peak of 1720 kt in 2017, then declined in 2020, followed by a rebound in 2021. Despite these regional differences, a consistent pattern emerges that all five urban agglomerations experienced emission reductions after 2017, coinciding with the implementation of the Three-year Action Plan to Win the Battle for a Blue Sky [2].

3.2.2. The Contributions of NO_x Emission Changes on NO₂ Concentrations

Figure 5 displays the interannual trends of NO₂ concentration ratios relative to 2014 levels from the Base and the FixedEmis simulations during summer (June to August) and winter (December to the following February) across five national urban agglomerations from 2015 to 2021. Compared with the Base simulations, the trends and interannual variations in NO₂ concentrations from the FixedEmis simulations were substantially weaker across all urban agglomerations. In summer, the simulated NO₂ levels with fixed emissions in BTH, YRD, and YRMR remained relatively stable from 2014 to 2021. This stability indicates that NO₂ reductions observed in BTH during 2015, and 2018 onward, as well as the declining trend in YRMR after 2015, were primarily attributable to emission reductions. The decreases in NO₂ from the FixedEmis simulations were much smaller than those from the Base simulations in BTH after 2017 and in YRD and YRMR after 2015, demonstrating concurrent influences from both meteorological variations and emission reductions. PRD showed opposing interannual trends between the two simulation scenarios, suggesting that NO₂ variability here was dominantly controlled by emission changes while still being modulated by meteorological conditions. Winter analyses revealed stable NO₂ levels in the FixedEmis simulations throughout the study period for YRMR, CY, and PRD, confirming emission changes as the principal driver of interannual NO₂ variability. In BTH and YRD from 2014 to 2017, along with CY in 2015, the near-identical trends between the two simulation scenarios implied weaker anthropogenic contributions, with meteorology being the predominant factor. However, from 2017 to 2021, dynamics showed dramatically weaker variability in the FixedEmis simulations compared to the Base simulations across BTH, YRD, YRMR, and PRD. The particularly pronounced NO₂ increase observed in 2020 Base simulations for YRD, YRMR, CY, and PRD provided compelling evidence for substantial emission-driven contributions. These findings collectively demonstrate that interannual NO₂ variability is predominantly governed by emission changes, though with varying degrees of meteorological influence across regions and seasons.

Figure 6 and Figure 7 present the quantitative contributions of meteorological condition and anthropogenic emissions variations to the interannual changes in NO₂ concentrations in summer and winter across five national-level urban agglomerations. The results demonstrate that, overall, emission changes exerted a greater influence on NO₂ concentration variations than meteorological factors, particularly during winter when NO₂ concentrations were higher and emission-driven contributions became more pronounced.

In BTH, anthropogenic emission reductions led to NO₂ concentration decreases of 2.8, 4.9, and 6.8 μg·m⁻³ during summer 2015, 2018, and 2021, respectively. Winter meteorological effects were more significant in BTH than in other urban agglomerations, altering NO₂ concentrations by +5.0, −7.4, +5.8, +1.9, and −6.2 μg·m⁻³ annually from 2016 to 2020. YRMR exhibited an 84% attribution of its total NO₂ reduction to emission controls during summer 2015. YRMR recorded significant dual-season reductions in 2015 (−7.0 μg·m⁻³ in summer, −6.7 μg·m⁻³ in winter), with reductions in winter 2018 approaching −10 μg·m⁻³. Similarly, in YRD, emission reductions accounted for 91% (summer 2016) and 86% (winter 2018) of NO₂ declines, with reductions in winter 2019 reaching −11.8 μg·m⁻³. In CY, emissions increased NO₂ concentrations before 2017. From 2018 onward, the benefits of emission reductions became apparent, with NO₂ concentrations decreasing by 6–8 μg·m⁻³, though emissions caused an increase of 8.6 μg·m⁻³ in NO₂ concentrations in 2020. In PRD, the impact of emission changes on NO₂ concentrations was also evident. Emission reductions in summer 2019 and 2021 each lowered NO₂ concentrations by more than 3 μg·m⁻³. However, in 2021, the reduction from emission control was offset by unfavorable meteorological conditions that increased NO₂ concentrations. A similar pattern occurred in winter 2021, where over half of the emission-driven decrease was counteracted by worsened meteorological conditions. In winter, emission reductions were particularly effective, with associated NO₂ decreases reaching 10.5, 8.2, and 9.4 μg·m⁻³ in 2015, 2018, and 2021, respectively. Nevertheless, emission control in PRD was ineffective in some years, with anthropogenic emissions leading to NO₂ increases, for example, by 1.7 μg·m⁻³ in summer 2017, and by 8.8 and 5.8 μg·m⁻³ in winter 2016 and 2020, respectively.

Overall, summer NO₂ concentrations displayed gradual declines across all urban agglomerations from 2014 to 2021, though the rate of decrease was relatively slow. Winter trends were more dynamic, characterized by an initial decline and subsequent rise during 2014–2017 (2014–2016 for BTH), before transitioning to a fluctuating downturn after 2018. These patterns largely mirror changes in emissions, consistent with the post-2014 trend in NO_x emissions reported in the inventory, rising initially and then declining [49].

3.3. Impact of Interannual Variations in NO_x Emissions on O₃ Pollution

3.3.1. Seasonal Variation in O₃ Concentrations over the Years

The spatial patterns of summer O₃ concentrations in China from 2014 to 2021 were presented in Figure S3, simulated based on optimized NO_x emissions in Base simulations. The O₃ pollution patterns consistently exhibited urban agglomeration characteristics, with elevated concentrations predominantly clustered in BTH, YRD, and CY metropolitan regions. North China Plain, Central China, and East China progressively merged into a contiguous high-O₃ zone, forming an extensive polluted area encompassing the Bohai Rim, Guanzhong Plain, Central Plain, YRD, YRMR, and CY. Eastern coastal areas also exhibited relatively high O₃ concentrations, particularly in the Shandong Peninsula and the coastal regions of Jiangsu province. The core area of O₃ pollution was mainly located in southeastern Shanxi province from 2014 to 2017, gradually expanding and shifting westward, with pollution intensifying annually and peaking in 2017. This is consistent with previous studies indicating an increasing spatial agglomeration of high O₃ concentrations from 2015 to 2017 [71]. After 2018, O₃ pollution in the Central Plain regions gradually decreased, although northern Shanxi and Shaanxi remained areas of high O₃ concentrations. The O₃ pollution in Central and Eastern China also showed some mitigation. Overall, the spatial distribution of summer average O₃ concentrations across China exhibited a clearer pattern of higher levels in the north and lower levels in the south.

Summer is the peak season for O₃ pollution, while spring, as temperatures gradually rise, also exhibits notable O₃ pollution issues (Figure S4). Compared to summer, the center of high O₃ pollution shifts westward in spring, transitioning from the North China and Central Plain toward the southwest into the Chengdu-Chongqing region, particularly affecting Sichuan province and surrounding areas. Consistently elevated O₃ concentrations indicate severe ozone pollution across these regions annually. Notably, the highest O₃ levels predominantly encircle the Sichuan Basin, forming a distinct ring-shaped distribution along its periphery. Recent interannual trends suggest mitigation, with peak O₃ concentrations demonstrating a decreasing pattern. Furthermore, Central and South China also exhibit relatively high O₃ levels, especially in 2019 when O₃ pollution in provinces such as Hubei and Hunan intensified compared to 2018. By 2021, nationwide spring O₃ pollution showed further improvement, with only western Sichuan Basin maintaining notably high O₃ concentrations. While localized O₃ pollution remains substantial, the overall national trend is moving to a favorable direction.

In autumn, declining temperatures coincide with reduced O₃ pollution intensity (Figure S5). Most regions across the country exhibit relatively low ozone concentrations, especially in northeastern and eastern North China, where seasonal averages frequently fall below 50 μg·m⁻³. However, the western part of the Sichuan Basin remains an area with relatively high O₃ levels; for instance, from 2014 to 2019, ozone concentrations in these regions still exceeded 100 μg·m⁻³. Moreover, similar to spring, Southern China experienced prominent ozone issues in autumn 2019, emerging as both the most concentrated and highest-magnitude O₃ hotspot nationwide.

3.3.2. The Impact of NO_x Emission Changes on O₃ Concentrations

Figure 8 illustrates the differences in summer mean O₃ concentrations between the Base and FixedEmis simulations during 2015–2021, representing the impact of NO_x emissions variations relative to 2014 levels. In 2015, NO_x emission modifications led to reduced O₃ concentrations in North China, Liaoning province, and central provinces of East China, while intensifying O₃ pollution in the Northwest and Southwest regions. Additionally, aggravated O₃ pollution was observed across Central China, South China, southern parts of East China, and the central-northern region of Northeast China. With further NO_x emission increases in 2016 and 2017, O₃ pollution intensified correspondingly. In 2016, O₃ reductions were limited only to northern Liaoning, northwestern Hebei, northeastern Shanxi, and eastern Anhui and Henan provinces, whereas western regions continued experiencing worsening pollution. The Shandong Peninsula and coastal areas of YRD also exhibited moderate O₃ increases. In 2017, when NO_x emissions peaked in recent years, O₃ pollution worsened further, with only sporadic areas showing improvement. Notably severe pollution occurred in eastern Gansu, southern Shaanxi, and the Sichuan Basin. By 2018, the regions with reduced O₃ pollution began to expand, extending southward from Northeast and North China, and further expanding to northern Central China and central East China by 2019. However, western Shandong Peninsula maintained persistently high O₃ levels. In 2020, the nationwide O₃ pollution level remained largely comparable to that in 2019. By 2021, sustained NO_x emission control measures yielded substantial improvements, with marked O₃ reductions across most central-eastern regions except the Bohai Rim area. The declines of O₃ concentrations were also achieved in the Northwest and eastern Southwest China.

The analysis of nationwide summertime O₃ concentrations (Figure S6) and their deviations from simulations using fixed 2014 NO_x emissions (Figure 9a) reveals a distinct post-2018 decline in O₃ levels across the national urban agglomerations. However, compared to scenarios maintaining 2014 NO_x emission intensities, O₃ concentrations remain elevated in recent years, indicating limited effectiveness of NO_x reduction measures in mitigating O₃ pollution within economically developed regions. Similar situations have been documented in other Asian countries. For instance, stringent NO_x emission controls implemented in South Korea to reduce PM_2.5 levels paradoxically increased O₃ concentrations in the Seoul metropolitan areas [72]. This occurs because these regions have entered a NO_x-saturated regime, where initial emission reductions temporarily exacerbate O₃ formation. Consequently, the observed O₃ increase represents a transitional threshold before further NO_x cuts enable progression into the NO_x-controlled regime, wherein continued reductions ultimately suppress O₃ production. Notably, NO_x abatement has demonstrated substantial efficacy in controlling O₃ pollution in several provinces, such as Anhui, Shanxi, and Jilin (Figure 9b). After 2018, sustained and progressively stronger reductions in O₃ levels were observed, suggesting that NO_x mitigation strategies are achieving meaningful impacts in these areas. Additionally, Inner Mongolia, Liaoning, Hubei, and Hunan province exhibited marked O₃ decreases after 2018 compared to pre-2017 levels, further evidencing the gradual emergence of NO_x reduction benefits for O₃ control.

The impact of NO_x emission reductions on springtime O₃ concentrations exhibited distinct patterns compared to summer effects. During 2016–2017, emission changes primarily contributed to significant O₃ decreases in southern North China, northern Central China, and northwestern East China. In contrast, during 2020–2021, O₃ increases were pronounced in East China and North China, with concentrations rising by over 18 μg·m⁻³ in areas such as the western Shandong Peninsula and parts of East China (Figure 10). Among the urban agglomerations, apart from some alleviation of O₃ pollution in BTH and YRD in 2016 and 2017, other agglomerations also showed clear O₃ increases (Figure 11). At the provincial level, Anhui, Shanxi, and Jilin province exhibited notable O₃ decreases, similarly to the summer pattern. Additionally, NO_x emission changes played a role in mitigating spring O₃ pollution in Henan and Jiangsu provinces (Figure S7).

Autumn responses to NO_x reductions largely resembled spring patterns, with Central and North China (especially Henan, Anhui and Jiangsu provinces) showing substantial O₃ decreases (Figure 12). However, urban agglomerations maintained severe O₃ pollution with minimal improvement (Figure 13). Provincially, besides the improvements seen in Anhui, Shanxi, Henan, and Jiangsu during spring, Shaanxi and Heilongjiang also began to show favorable trends in O₃ pollution in autumn (Figure S8). A statistical analysis of the areal extent of O₃ concentration reduction across spring, summer, and autumn revealed that the area with decreasing O₃ concentrations in summer began to expand steadily from 2017 onward, indicating effective implementation of control measures. In contrast, the area of O₃ reduction in spring and autumn did not show significant interannual expansion (Figure S9).

4. Discussion

This study quantitatively assessed the relative contributions of interannual variations in anthropogenic NO_x emissions and meteorological conditions to surface NO₂ and O₃ concentrations across five major Chinese urban agglomerations from 2014 to 2021. The core methodological approach, employing a pair of fixed-emission simulations, effectively isolated the emission-driven signal from the confounding influence of meteorological variability. Our results demonstrate that, while meteorology modulates concentrations at seasonal and interannual scales, anthropogenic emission changes are the predominant driver of long-term trends in NO₂ and a critical, though complex, factor influencing O₃ pollution patterns.

4.1. Consistency Between NO_x Emission Variations and Associated NO₂ Concentration Changes

Changes in NO₂ concentrations are driven by both meteorological conditions and anthropogenic emissions, with the latter being the dominant factor. Consequently, variations in NO_x emissions significantly impact trends of NO₂ concentrations. Therefore, the contributions of emission changes to NO₂ concentrations were isolated through systematic comparison between the annual Base simulations and the FixedEmis simulations (Table 1). Seasonal consistency between these derived contributions and documented NO_x emission changes was then rigorously evaluated (

Table 2. Variations in NO₂ concentration related to anthropogenic NO_x emission in five national urban agglomerations in summer and winter during 2014–2021 (unit: μg·m⁻³).

Year	BTH		YRD		YRMR		CY		PRD
	Summer	Winter	Summer	Winter	Summer	Winter	Summer	Winter	Summer	Winter
2015–2014	−2.8	1.6	−0.6	1.1	−7.0	−6.7	1.8	1.0	−1.5	−10.5
2016–2015	0.4	3.1	−3.2	1.1	−0.1	3.1	−2.8	3.8	1.4	8.8
2017–2016	−0.9	−8.5	2.1	3.8	−1.0	3.3	3.1	4.2	1.7	2.6
2018–2017	−4.9	−3.3	−1.7	−7.4	0.0	−9.6	−0.6	−6.6	0.4	−8.2
2019–2018	−1.7	−10.4	−1.4	−11.8	−1.2	−4.9	−3.5	−7.5	−3.1	−2.2
2020–2019	−2.6	5.3	−3.7	9.7	−1.6	8.6	−2.3	7.0	−0.8	5.8
2021–2020	−6.8	−2.6	−1.6	−4.5	−1.9	−0.6	−1.6	−4.2	−3.5	−1.0

, Figure 14).

Table 2 presents the statistical results of NO₂ concentration changes attributed to anthropogenic NO_x emission variations during summer and winter from 2014 to 2021 for the five national-level urban agglomerations. Correspondingly,

Table 3. Interannual NO_x emission variations in five national urban agglomerations in summer and winter during 2014–2021 (unit: kt).

Year	BTH		YRD		YRMR		CY		PRD
	Summer	Winter	Summer	Winter	Summer	Winter	Summer	Winter	Summer	Winter
2015–2014	−44.4	31.3	−2.3	−32.2	−55.8	−47.3	45.7	18.3	−13.5	−56.1
2016–2015	−13.5	21.8	−67.2	6.1	5.9	26.9	−49.6	20.2	12.7	51.5
2017–2016	3.0	−146.3	40.6	45.5	−14.8	24.9	61.0	42.8	19.2	11.4
2018–2017	−90.4	−54.6	−35.4	−67.2	−6.8	−84.6	−25.3	−88.7	−9.2	−55.7
2019–2018	−10.9	−122.7	−32.1	−92.1	−16.4	−24.5	−24.7	−45.0	−25.0	13.6
2020–2019	−37.3	18.4	−34.6	73.8	−31.4	69.7	−32.9	42.6	−11.9	9.6
2021–2020	−61.6	−77.1	−41.2	−3.5	−32.3	−62.2	−15.4	−48.9	−29.2	−30.8

shows the interannual changes in NO_x emissions for each agglomeration. A regression analysis between the two sets of changes reveals a high consistency across the urban agglomerations (Figure 14). In summer, the correlation coefficient reaches 0.82 in BTH, while in the other four agglomerations it exceeds 0.90.

The annual changes in NO_x emissions were normalized relative to the change in the preceding year to facilitate a more intuitive comparison of their trends, resulting in the histograms shown in Figure 15. Overall, the trends show strong agreement, particularly in YRD, CY, and PRD. However, opposite trends occur in certain years, for example, in BTH in 2016 and 2017, in 2016 in YRMR, and in 2018 in PRD. Further analysis indicates that these opposing trends generally occur during periods with relatively small changes in both concentration and emissions, whereas higher consistency is observed when changes in concentration and emissions are larger. Winter exhibits stronger consistency between NO_x emission changes and NO₂ concentration variations across most urban agglomerations (except PRD) compared to summer, with correlation coefficients reaching 0.94 in BTH and CY. This seasonal contrast is particularly evident in comparative histograms, though exceptions occur, specifically, YRD (2015) and PRD (2019) display anomalous opposing trends.

4.2. Dominance of Emission Controls on NO₂ Trends

The pronounced decline in NO₂ concentrations across all studied urban agglomerations post-2014, particularly the sharp decreases observed after 2017/2018, aligns temporally with the rigorous nationwide implementation of emission reduction policies and measures. Our quantitative attribution confirms that emission reductions were the principal contributor to these declines. The high consistency between the year-to-year changes in top-down optimized NO_x emissions and the corresponding emission attributable NO₂ variations (Section 4.1) robustly validates the effectiveness of the optimized emission inventory and the attribution methodology. This strong linear relationship underscores a direct, regionally coherent link between policy-driven emission abatement and observed air quality improvements for primary pollutants like NO₂.

Notably, the relative importance of emissions versus meteorology exhibited seasonal and regional nuances. Winter NO₂ concentrations showed larger emission-attributable changes than summer, consistent with higher absolute emission levels and more stagnant meteorological conditions in winter amplifying the impact of emission changes. Furthermore, the efficacy of emission controls varied regionally. For instance, PRD, an early adopter of stringent air quality measures, displayed complex dynamics where meteorological counteractions occasionally offset emission-driven benefits (e.g., winter 2021). In contrast, regions like YRMR and CY showed particularly strong emission-driven responses in certain periods, potentially reflecting the more recent and intensive application of control measures in central and western China.

4.3. Complex and Non-Linear Impacts of NO_x Reductions on O₃ Pollution

The impact of NO_x emission changes on O₃ concentrations revealed a more intricate picture, characterized by strong non-linear chemistry and regional disparities. The nationwide increase in summer O₃ concentrations from 2014 to 2017, concurrent with initially rising or peaking NO_x emissions [49]. In several agglomerations, particularly the YRD and PRD, O₃ showed slight increases or plateaued during the study period, consistent with the weekend effect phenomenon, whereby urban areas experience higher O₃ on weekends despite lower NO_x emissions. The subsequent limited decline of O₃ despite substantial NO_x cuts after 2018, point to China’s widespread transition into or persistence within a NO_x-saturated (VOC-sensitive) photochemical regime in key urban agglomerations [57,73]. In such a regime, reducing NO_x can paradoxically increase O₃ production by slowing the titration of O₃ by NO and shifting chemical pathways, as evidenced by the positive O₃ anomalies under reduced-emission scenarios (Figure 8 and Figure 9).

This interpretation is strongly supported by the spatial and temporal patterns of the satellite-derived formaldehyde-to-nitrogen dioxide ratio (HCHO/NO₂ ratio, FNR), a robust indicator for O₃ formation sensitivity [57,58]. Our analysis for summer from 2014 to 2021 shows persistently low FNR values (less than 1.5) over the major emission hotspots of Eastern China (e.g., North China Plain, YRD), consistent with a VOC-sensitive regime (Figure 16). Notably, the observed increase in FNR over these regions from 2014 to 2017, and its subsequent stabilization after 2018, aligns with the documented shift from VOC-limited towards transitional regimes driven by aggressive NO_x controls. Concurrently, the higher FNR values (more than 4) observed in parts of Central and South China (e.g., Jiangxi, Guizhou, Guangxi province) suggest a prevailing NO_x-sensitive regime in these less developed areas.

However, the O₃ response to NO_x changes is not uniform. We identified provinces such as Anhui, Shanxi, and Jilin where NO_x reductions consistently led to O₃ decreases across multiple seasons, suggesting these areas may reside in a NO_x-sensitive regime. This spatial divergence underscores the critical importance of region-specific chemical regimes. Furthermore, the seasonal modulation of control effectiveness is evident. The expansion of areas showing summer O₃ reductions between the Base and FixedEmis simulations after 2017, a pattern not mirrored in spring or autumn (Figure 9, Figure 11 and Figure 13), implies that precursor balances, solar radiation, and background conditions seasonally alter the chemical sensitivity. The pronounced O₃ increases in eastern China in response to NO_x cuts during the 2020–2021 spring and autumn periods reinforce this seasonal complexity, likely due to lower biogenic VOC emissions and altered photochemistry that exacerbate the disinhibition effect from NO_x reduction.

The spatial patterns of FNR and the divergent O₃ responses necessitate a differentiated, synergistic control strategy. In major urban agglomerations like the BTH, YRD, and PRD, where FNR values are typically low and regimes are VOC-sensitive or transitional, effective O₃ reduction requires concurrent, stringent cuts in anthropogenic VOC emissions alongside NO_x controls. The observed plateauing of FNR increases after 2018 in these regions suggests that further O₃ gains may now be critically dependent on accelerating VOCs emission reductions (Figure 16). Conversely, in provinces demonstrating a NO_x-sensitive response (e.g., parts of Central and South China with higher FNR), sustained NO_x reduction remains the primary and effective strategy for O₃ control.

4.4. Progress Toward WHO Air Quality Guidelines: The Role of Emissions and Meteorology

Benchmarking against the World Health Organization (WHO) Air Quality Guideline 2021 reveals both progress and persistent challenges in air quality management [74]. By 2021, the national annual mean NO₂ concentration had reached the WHO interim target II of 30 μg·m⁻³, with YRMR, CY, and PRD also attaining this level (Figure 1). Regarding PM_2.5, concentrations declined from 46 μg·m⁻³ in 2015 to 30 μg·m⁻³ in 2021, thereby meeting the WHO interim target I, with BTH exhibited the most substantial emission-driven reduction during this period [75]. However, the trend for O₃ presents a stark contrast. Since 2014, over 90 percent of days across China have recorded maximum daily 8 h average O₃ concentrations exceeding the WHO guidelines of 100 μg·m⁻³, with similarly high exceedance rates observed in PRD and YRD [76]. Furthermore, since 2017, more than 98 percent of Chinese cities have experienced at least one day with O₃ exceedance, indicating widespread acute health risks despite the long-term reductions in NO_x emissions [76]. These contrasting patterns highlight that while emission controls have successfully reduced primary pollutants, secondary pollution remains a major challenge when assessed against health-based standards.

As mentioned above, anthropogenic emission changes are the primary driver of long-term trends in NO₂, while O₃ exhibits a more complex response due to its nonlinear photochemistry. However, meteorological conditions modulate pollutants at seasonal and interannual scales, explaining variability not captured by emission trends alone. Wintertime accumulation of NO₂, as well as PM_2.5, in northern agglomerations (e.g., BTH) is exacerbated by shallow boundary layers and weak winds, which trap pollutants near the surface [77]. In southern regions, precipitation enhances wet removal, contributing to interannual fluctuations. For O₃, high summer temperatures and strong solar radiation drive photochemical production, with the record-high O₃ in 2017 coinciding with eastern China heatwaves [78]. Regional transport further links agglomerations. For example, southerly monsoon winds carry O₃ and its precursors from PRD and YRD to inland areas such as YRMR, explaining synchronized pollution episodes [79]. The increase in O₃ after 2017, despite reductions in NO_x, reflects a phenomenon known as ozone penalty, a nonlinear response under VOC-limited regimes that is exacerbated by warming trends [80]. The strong influence of meteorology on pollutant behavior also has implications for how progress toward WHO guidelines is perceived. Interannual meteorological variability can temporarily mask or amplify the benefits of emission reductions, potentially leading to over-optimistic or overly pessimistic interpretations of short-term trends.

4.5. Limitations and Future Perspectives

While this study advances understanding of emission-driven changes in NO₂ and O₃ trends, several methodological considerations and associated limitations should be acknowledged. The fixed-emission approach employed here effectively isolates the net effect of total emission changes but does not resolve sector-specific contributions (e.g., transportation versus industry). The scenario design focuses primarily on NO_x variations and does not explicitly isolate the potential co-variation in other precursors such as VOCs that could influence O₃ formation pathways. The assumption of constant 2014-level VOCs emissions within the FixedEmis simulation, while necessary for isolating certain drivers, may not reflect actual temporal trends and could influence simulated O₃ responses. Additionally, the use of a 36 km horizontal resolution, though suitable for regional analysis, may not fully resolve finer-scale chemical gradients within urban areas. Furthermore, we employed static default chemical boundary conditions rather than dynamic outputs from global models. While the extensive modeling domain effectively buffers interior target regions against boundary inflows of short-lived species, this simplification may introduce uncertainties in estimating background concentrations of longer-lived species like O₃. Finally, because our primary focus was on capturing long-term trends and interannual variability, we did not assess the modeling performance for short-term air quality. It should be noted that modeling short-term pollutant behavior is inherently more challenging than modeling mean monthly concentrations due to high-frequency fluctuations in meteorological conditions and short-term emission profiles.

Future work would benefit from more targeted sensitivity experiments (e.g., fixing NO_x or VOCs emissions individually) to better disentangle the roles of different precursors. Higher resolution modeling and the integration of sector-specific emission inventories would enhance the mechanistic understanding and support more targeted policy design. Coupling regional models with global chemical transport models to provide time-varying boundary conditions would also improve the assessment of intercontinental background O₃ transport. Long-term monitoring and analysis are also needed to determine if and when key urban agglomerations will transition into a NO_x-sensitive regime, allowing NO_x reductions to yield consistent O₃ benefits. Finally, while this study relied on monthly mean values to evaluate broad O₃ trends, health impacts are most closely associated with the maximum daily 8 h average (MDA8) concentrations. Therefore, analyzing the trends in the number of exceedances of the MDA8 O₃ standard represents a critical direction for future study to provide a more rigorous assessment of the associated acute health risks.

5. Conclusions

This study systematically evaluated the impacts of interannual variations in top-down constrained NO_x emissions on surface NO₂ and O₃ concentration trends across five national urban agglomerations in China during 2014–2021, using WRF−CMAQ model with a fixed-emission sensitivity approach. Key findings are summarized as follows.

NO₂ concentration trends were predominantly driven by emission changes. Emission contributions were quantitatively isolated via the Base and FixedEmis simulations, revealing that NO_x controls accounted for over 80% of NO₂ declines in several regions, especially after 2017, with significant reductions observed in winter across BTH, YRD, YRMR, and PRD. Meteorological conditions could cause interannual fluctuations, occasionally offsetting emission-driven improvements, but did not alter the long-term downward trend dictated by emission controls. Regression analyses confirmed strong consistency between the interannual changes in top-down NO_x emissions and the attributed changes in NO₂ concentrations (r > 0.80 in summer, r > 0.90 in winter), validating the RAPAS-assimilated emission inventory and the attribution method.

O₃ responses to NO_x reductions were spatially and seasonally heterogeneous. In economically developed urban agglomerations (e.g., BTH, YRD, PRD), NO_x reductions since 2018 have shown limited effectiveness in mitigating summertime O₃ pollution and have even led to O₃ increases in some seasons (spring and autumn). This indicates a disinhibition effect where reduced NO_x titration slows O₃ loss. Certain provinces (e.g., Anhui, Shanxi, Jilin) exhibited consistent O₃ decreases in response to NO_x reductions, highlighting the necessity for region-specific control strategies. Seasonal differences in O₃ response were pronounced. The areal extent benefiting from NO_x reductions (i.e., showing O₃ decreases) expanded steadily in summer after 2017 but showed no similar clear expansion in spring or autumn, indicating seasonal differences in chemical sensitivity and the influence of varying meteorological and background conditions.

In conclusion, China’s stringent NO_x emission control policies have been highly effective in reducing primary NO₂ pollution. However, to combat the persistently high and complex O₃ pollution, a shift towards integrated, multi-pollutant control strategies is imperative. This involves implementing stringent VOC controls alongside NO_x reductions in VOC-sensitive urban agglomerations, while continuing NO_x-focused measures in NO_x-sensitive regions. Future efforts should refine emission inventories to better resolve sector-specific contributions and account for spatial and seasonal disparities in photochemical regimes. This will enable the design of scientifically grounded, integrated policies that effectively leverage NO_x-VOC synergies to improve overall air quality.