Emission Control and Sensitivity Regime Shifts Drive the Decline in Extreme Ozone Concentration in the Sichuan Basin During 2015–2024

Highlights

What are the main findings?

• O₃ pollution in the SCB became more widespread but less intense.
• Shift from VOC-limited to NO_x-limited O₃ formation regimes observed across the SCB.

What is the implication of the main finding?

• Reduced NO_x emissions contributed to lower extreme O₃ levels.

Abstract

In recent years, ozone (O₃) pollution has become a prominent air quality concern in the Sichuan Basin (SCB). Based on surface O₃ measurements from 22 cities between 2015 and 2024, this study investigates the evolution of extreme O₃ pollution events and their underlying causes. While the average O₃ concentration, the number of affected cities, and the total O₃ pollution hours have all increased during the past decade, extreme O₃ concentrations have shown a significant decline since 2020. These trends suggest that O₃ pollution in the SCB has become more spatially extensive and less intense. Decomposition analysis attributed ~75% of the post-2020 decline in extreme O₃ concentrations to precursor emission reductions, with meteorological variability explaining the remaining ~25%. Satellite observations of formaldehyde (HCHO) and nitrogen dioxide (NO₂) column densities indicate a regional shift in O₃ formation regimes across the SCB, with many areas transitioning from VOC (volatile organic compound)-limited to transitional or NO_x (nitrogen oxide)-limited conditions. This shift likely contributed to the broader spatial extent and longer duration of O₃ pollution in recent years. Model sensitivity simulations and Integrated Reaction Rate (IRR) analysis demonstrate that reductions in precursor emissions, particularly NO_x, directly weakened daytime photochemical O₃ production and disrupted NO_x-driven radical propagation under transition and NO_x-limited conditions, collectively driving the observed decline in extreme O₃ concentrations.

1. Introduction

As a ground-level atmospheric pollutant, elevated concentrations of ozone (O₃) can have harmful effects on human health, crop growth, ecosystems, and economic development [1,2,3]. Since the implementation of the Action Plan for Air Pollution Prevention and Control in 2013, China’s air quality has significantly improved, particularly with a sustained decline in particulate matter pollution [4,5,6]. However, in many regions, O₃ pollution has shown a marked upward trend, particularly before 2019 [7,8,9,10]. The implementation of the Three-Year Action Plan to Win the Battle for a Blue Sky (2018–2020) resulted in continued reductions in O₃ precursors, contributing positively to the mitigation of O₃ pollution in China [10,11].

The formation of ground-level ozone involves complex photochemical reactions between nitrogen oxides (NO_x) and volatile organic compounds (VOCs), which compete for reactions with hydroxyl radicals (OH) [12,13]. Ozone production depends on the relative abundance of NO_x and VOCs, categorized into three regimes: NO_x-limited, VOC-limited, and transitional [14]. In the NO_x-limited regime, where the VOC-to-NO_x ratio is high, OH mainly reacts with VOCs, and reducing NO_x emissions decreases O₃ levels. In the VOC-limited regime, with a low VOC-to-NO_x ratio, the OH + NO₂ reaction dominates, and reducing NO_x emissions may paradoxically increase O₃ concentrations. The transitional regime occurs between these extremes, where ozone production is equally sensitive to both species, and the reaction rates of OH with VOCs and NO₂ are balanced at a specific VOC:NO₂ ratio, at which the maximum amount of ozone is produced. However, this ratio depends on the particular VOC or VOC mixture, as the reaction rate constants for OH with each VOC differ [12].

Therefore, trends in ozone concentrations are largely influenced by changes in the emissions of its precursors. From 2013 to 2017, the decline in NO_x and PM_2.5 emissions was responsible for the O₃ increases in China due to VOC-limited regimes in most urban areas [7,11,15]. Since 2018, the synergistic control of VOCs and NO_x has initially worked to mitigate O₃ pollution [11]. In the future, mutual reduction in VOCs and NO_x emissions may be an effective strategy for improving O₃ air quality. Furthermore, meteorological conditions play a significant role in the long-term variations in O₃, particularly after 2017, when the contribution of meteorological factors to O₃ has significantly increased [7,16].

The Sichuan Basin (SCB), one of the most densely populated and industrially developed regions in China, has been experiencing increasingly severe O₃ pollution [11,15]. Bounded by the Tibetan Plateau to the west, the Qinling and Loess Plateaus to the north, the Wu Mountains to the east, and the Yunnan-Guizhou Plateau to the south, the SCB is geographically enclosed, which limits the transport of anthropogenic pollutants from surrounding regions [17,18,19]. This unique topographical setting provides an ideal situation for examining the impacts of local anthropogenic emissions on O₃ pollution.

Recent studies have characterized O₃ dynamics in the SCB using observational, satellite, and modeling analyses. Wu et al. [15] documented a sustained O₃ increase during 2013–2020, driven by precursor emissions and meteorological variability, highlighting the basin’s sensitivity to synoptic patterns and thermal conditions. Similarly, Wang et al. [20] linked rising O₃ in Chengdu to evolving NO₂ and HCHO columns, while Ren et al. [21] quantified heterogeneous O₃ responses across the Chengdu–Chongqing urban agglomeration, attributing NO₂ declines to emission controls. At broader scales, Wang et al. [22] identified amplified O₃ burdens during heatwaves, synergistically enhanced by biogenic VOCs in the SCB’s topographically enclosed environment. Complementary regime analyses [7] further emphasized the prevalence of VOC-limited conditions in the SCB, with seasonal transitions toward NO_x-limited regimes. However, most of these studies mainly focused on long-term trends up to 2020, while the evolution of O₃ after 2020, especially the changes in extreme O₃ events, remains poorly understood. In particular, the impacts of recent emission reductions on O₃ formation regimes during extreme episodes have rarely been explored. To address these gaps, this study integrates ground-level O₃ observations, satellite-derived formaldehyde (HCHO) and NO₂ data, and model simulations to investigate the evolution of O₃ pollution and its formation regimes in the SCB from 2015 to 2024, as well as the role of anthropogenic emission changes in driving extreme O₃ events.

2. Materials and Methods

2.1. Ozone Observation Data

Since 2013, the Ministry of Ecology and Environment (MEE) of China has established air quality monitoring stations across cities nationwide. By January 2015, the air quality monitoring network covered all 22 cities in the Sichuan Basin. This study collected hourly O₃ data from 156 stations in these cities between 1 January 2015, and 31 December 2024. According to China’s technical regulation on ambient air quality standards [23], air quality is classified as pollution if the hourly average O₃ concentration exceeds 200 μg m⁻³ or the 8-h moving average O₃ exceeds 160 μg m⁻³. Note that O₃ concentrations reported by the China National Environmental Monitoring Center (CNEMC) were in units of μg m⁻³ under standard state (273 K, 1013.25 hPa) prior to September 2018, but were changed to reference state (298.15 K, 1013.25 hPa) thereafter [23,24]. To ensure data consistency from 2015 to 2024, O₃ concentrations after September 2018 were converted to standard state conditions.

The conversion between the two states follows the ideal gas law (PV = nRT), where gas mass concentration is expressed as

$$C={\displaystyle \frac{PM}{RT}}$$

(1)

where P denotes atmospheric pressure, M represents the molar mass of the air, R is the universal gas constant, and T is the absolute temperature. For the standard (std) and reference (ref) states, the ratio of O₃ concentrations becomes

$${\displaystyle \frac{C_{std}}{C_{ref}}}={\displaystyle \frac{P_{std}}{P_{ref}}}·{\displaystyle \frac{T_{ref}}{T_{std}}}$$

(2)

Given constant atmospheric pressure (1013.25 hPa), the partial pressure of O₃ remains unchanged (i.e., P_std = P_ref). T_ref and T_std are 298.15 K and 273 K, respectively. As a result,

$$C_{std}=C_{ref}·{\displaystyle \frac{T_{ref}}{T_{std}}}=C_{ref}·1.092$$

(3)

To study extreme ozone pollution events in the SCB, hourly O₃ concentration data were used to extract exceedance instances (hourly O₃ concentration exceeded 200 μg m⁻³) in all cities from 2015 to 2024. The hourly standard was chosen rather than the 8-h moving average standard because the hourly threshold is more difficult to reach and thus better reflects extreme O₃ pollution. The total duration of pollution using the hourly standard is only about one-third of that based on the 8-h standard.

2.2. Satellite HCHO and NO₂ Data

The ratio of HCHO to NO₂ column concentrations (referred to as FNR) derived from satellite observations is commonly used to identify ozone formation sensitivity regimes [20,21,25,26]. This study investigates the temporal evolution and spatial distribution of these regimes over the SCB, based on HCHO and NO₂ data from Ozone Monitoring Instrument (OMI) and TROPOspheric Monitoring Instrument (TROPOMI) for the period 2015–2024.

OMI operates in a sun-synchronous polar orbit and crosses the equator at approximately 13:40 local time each day, providing global daily observations of trace gases with a nadir spatial resolution of approximately 13 km × 24 km [27]. TROPOMI is a nadir-viewing spectrometer designed for atmospheric composition monitoring [28]. Operating in a sun-synchronous orbit with a local overpass time of approximately 13:30, it collects daily global observations across four spectral ranges: ultraviolet, visible, near-infrared, and shortwave infrared. TROPOMI features high spectral and spatial resolution, with typical ground pixel sizes around 7.2 km × 3.6 km (improved to 5.6 km in the along-track direction since August 2019), and a wide swath of about 2600 km. This configuration enables near-complete global coverage on a daily basis, with enhanced sensitivity and accuracy compared to previous satellite instruments [29].

The OMI HCHO data are based on the OMHCHOd product, which offers quality-controlled, gridded vertical column densities at 0.1° × 0.1° resolution. Retrievals influenced by high solar zenith angles, elevated cloud fractions, or instrumental anomalies, such as the row anomaly, are excluded to ensure data quality. The HCHO dataset used in this study covers the period from January 2015 to June 2022 [30]. The TROPOMI HCHO data are obtained from the daily Level 3 product with the same spatial resolution of 0.1° × 0.1°. The dataset is processed using advanced retrieval algorithms developed under the German Aerospace Center (DLR) InPULS project, with strict quality filtering applied to exclude scenes with unfavorable observational conditions, such as high cloud cover or low signal quality. The dataset used in this study spans the period from May 2018 to December 2024. To ensure consistency and continuity between the OMI and TROPOMI HCHO datasets, monthly grid-based adjustment factors were applied to the earlier OMI data (January 2015 to April 2018). These factors were calculated as the ratio of monthly mean TROPOMI HCHO to monthly mean OMI HCHO within each grid cell during their overlapping period (May 2018 to June 2022).

The OMI NO₂ data are obtained from the QA4ECV product, which applies an improved retrieval algorithm incorporating enhanced spectral fitting techniques and a data assimilation-based stratospheric background estimation. These improvements enhance the separation of stratospheric and tropospheric components and refine air mass factor calculations [31]. In this study, QA4ECV NO₂ data from January 2015 to March 2021 were used, and regridded to 0.125° × 0.125° spatial resolution after applying standard quality screening. The TROPOMI NO₂ data are based on the latest version 2.8.0 of the operational product, which includes updates to the cloud algorithm, surface elevation processing, and quality assurance filtering [32]. These enhancements contribute to more accurate retrievals, particularly over snow-covered, cloud-free regions. In this study, TROPOMI NO₂ data from February 2018 to December 2024 were used. The original data were also regridded to a spatial resolution of 0.125° × 0.125° following standard quality control procedures. To further harmonize the two datasets, monthly adjustment factors were also applied to the earlier OMI NO₂ data (January 2015 to January 2018), following the same approach as for HCHO. These factors were computed as the ratio of monthly mean TROPOMI to monthly mean OMI NO₂ concentrations within each grid cell during their overlapping period (February 2018 to March 2021).

To quantify the uncertainty introduced by the harmonization, OMI and TROPOMI data over the Sichuan Basin during the overlapping period were compared (Figure S1). For HCHO, the unadjusted OMI data showed a weak correlation with TROPOMI (R = 0.39) and a strong negative normalized mean bias (NMB = −38%). After adjustment, the correlation coefficient increased to 0.55, the bias was largely removed (NMB = −5%), and the root mean square error (RMSE) decreased. The mean adjustment factor during the overlap period was 1.58, and as shown in Figure S2a, the adjustment effectively eliminates the offset between the two products and ensures continuity in the long-term HCHO record. For NO₂, the correlation improved from 0.51 to 0.78, while the bias decreased from 15.4% to 3.7%. The mean adjustment factor was 0.89, and Figure S2b shows that only minor changes were introduced, suggesting that OMI and TROPOMI NO₂ retrievals were already largely consistent.

2.3. CESM and Decomposition Method

Meteorologically induced variations in O₃ were investigated using Version 1.0.4 of the Community Earth System Model (CESM), incorporating the Community Atmospheric Model version 5 (CAM5). CESM employs a horizontal resolution of 1.9° × 2.5° (latitude × longitude) with 56 vertical layers extending from the surface up to 4 hPa. The model simulates concentrations of various trace gases based on the MOZART-4 (Model for Ozone and Related chemical Tracers, version 4) chemical mechanism [33].

In our simulations, anthropogenic emissions were held constant at 2010 levels, while meteorological variables, including horizontal wind, air temperature, land surface temperature, surface pressure, heat fluxes, and wind stresses, were prescribed according to actual conditions. To ensure realistic meteorology, CAM5 simulations were nudged toward the MERRA-2 reanalysis dataset (Modern Era Retrospective-analysis for Research and Applications, Version 2) at a 3-h temporal resolution (see also https://rda.ucar.edu/datasets/d313003/; last access 8 September 2025).

To quantitatively separate the contributions of variations in meteorology and anthropogenic emissions on extreme O₃ concentrations, we employed a decomposition method developed by Kang et al. [34]:

$$C_{met}={\displaystyle \frac{O_i-M_i}{M_i(O_i-1)}}\times 100\%$$

(4)

$$C_{emis}={\displaystyle \frac{O_i{(M}_i-1)}{M_i(O_i-1)}}\times 100\%$$

(5)

where C_met is the contribution of meteorology, C_emis is the contribution of anthropogenic emission, M_i refers to the normalized annual O₃ from simulations, and O_i refers to the normalized annual extreme O₃ from observations.

2.4. Model Description and Configuration

To investigate the impact of anthropogenic emission changes on extreme O₃ pollution, we selected a representative episode that occurred from 10–12 July 2017. This event affected 13 cities across the SCB, with the highest observe O₃ concentration reaching 335 µg m⁻³. The model simulation was initialized at 00:00 UTC on 4 July 2017, with a six-day spin-up period (4–9 July) to minimize the influence of initial conditions. This episode was chosen due to its spatial coverage and intensity, providing a suitable case for evaluating the effects of emission evolution on extreme O₃ levels.

The Weather Research & Forecasting (WRF) model version 4.5 was used to simulate the meteorological fields, which were used to drive the Community Multiscale Air Quality (CMAQ) model, in the O₃ pollution episodes. The model is set up with two nested domains centered at the SCB and with horizontal grid numbers (grid space) of 220 × 220 (9 km) and 220 × 220 (3 km) (Figure 1). The vertical direction of the model is divided into 39 sigma layers, with the lowest 20 layers being below 2 km to better resolve the boundary layer. The boundary and initial conditions are derived from the 0.25° NCEP FNL dataset [35], which was also used to analyze the s of temperature and relative humidity in the SCB from 2015 to 2024. The “CONUS” physics suite [36] was chosen for physics schemes.

The CMAQ model version 5.4 was applied to simulate O₃ and its precursors in the SCB with the same grid settings as the WRF model. The Carbon Bond version 6 (CB6) mechanism and three-mode Aerosol version 7 (AERO7) scheme were used for gas-phase and aerosol chemical simulations, respectively. The initial and boundary conditions for the outer domain of CMAQ simulation were derived from the Community Atmosphere Model with Chemistry (CAM-chem) simulation. The anthropogenic emissions in China were obtained from the Multi-resolution Emission Inventory for China (MEIC, http://www.meicmodel.org/, last accessed: 20 July 2024), which provides monthly gridded emissions at 0.25° × 0.25° resolution. To ensure compatibility with the modeling system, the MEIC emissions were interpolated to the WRF–CMAQ domains at 9 km and 3 km resolutions. In addition, temporal allocation was applied by introducing sector-specific weekly and diurnal variation factors, allowing the monthly totals to be distributed into hourly emission fluxes. Biogenic emissions were generated from the Model for Emissions of Gases and Aerosols from Nature (MEGAN) version 2.1 [37]. Biomass burning emissions were derived from the Global Fire Emissions Database (GFED) version 4 [38].

Process Analysis (PA) technique is used to quantify the contributions of different physical and chemical processes to changes in pollutant concentrations in CMAQ simulations. The PA includes two components: Integrated Process Rate (IPR) analysis and Integrated Reaction Rate (IRR) analysis. IPR tracks changes in species due to physical processes like advection, diffusion, emissions, deposition, and aerosols, as well as the overall effect of chemical processes. IRR focuses on the rates of individual chemical reactions or specified combinations of reactions and species cycling.

2.5. Model Evaluation Metrics

The Pearson correlation coefficient, NMB, and normalized mean error (NME) were calculated to assess the model’s performance in simulating ozone concentrations in the Sichuan basin. The NMB and NME were determined using the following formulas:

$$NMB={\displaystyle \frac{{\sum }_{i=1}^N\left(M_i-O_i\right)}{{\sum }_{i=1}^N{O}_i}}\times 100\%$$

(6)

$$NME={\displaystyle \frac{{\sum }_{i=1}^N\left|M_i-O_i\right|}{{\sum }_{i=1}^N{O}_i}}\times 100\%$$

(7)

where Mi represents the hourly model simulations, Oi represents the hourly observations, and N is the number of data pairs.

3. Results

3.1. Long-Term Variation in Extreme O₃ Pollution in Sichuan Basin

Based on the boxplot analysis of O₃ concentrations in 22 cities across the SCB from 2015 to 2024, distinct trends were observed between hourly and exceeded-hour O₃ pollution (>200 μg m⁻³). The boxplot of hourly O₃ concentrations, along with the corresponding annual mean values, shows a clear and consistent increase in the median, mean, and upper whisker values over the decade (Figure 2). This indicates a gradual rise in both background and high-end ambient O₃ levels throughout the basin. In contrast, the boxplot for exceeded-hour O₃ concentration reveals a different temporal pattern, the median, mean and upper whisker values all peaked around 2019–2020, followed by a decline trend in subsequent years. The notable downward trend of the upper whisker suggests a weakening in the intensity of the extreme O₃ pollution in the SCB. A similar temporal feature was also identified when using the 8-h exceedance O₃ (Figure S3).

From 2015 to 2024, O₃ pollution in the SCB showed an overall pattern of spatial expansion and increasing frequency in the earlier years, followed by slight decline after 2020. The number of cities experiencing O₃ pollution hours increased significantly from 2015 to 2020, then remained relatively stable at around 19–20 cities (Figure 2), indicating that O₃ pollution had reached a broad spatial coverage over the basin. By 2024, only Aba and Ganzi Prefectures, located on the eastern margin of the Tibetan Plateau, had not reported any O₃ pollution, likely due to their geographic remoteness and low anthropogenic precursor emissions.

The total number of O₃ pollution hours across all cities in the SCB increased significantly from 490 h in 2015 to a peak of 1549 h in 2022, before showing a declining trend thereafter. This upward trend before 2020 is consistent with previous findings [15] and was mainly driven by a shift in the O₃ formation regime, from VOC-limited to transitional or NO_x-limited conditions, as a result of substantial reductions in NO_x emissions in the SCB [15,21]. Notably, 2021 recorded a sharp drop in O₃ pollution hours, primary due to unfavorable meteorological conditions. The interannual variation in average surface temperature and relative humidity in the SCB during the O₃ pollution season (April to September), when over 95% of O₃ exceedances occurred, indicates that 2021 had the lowest temperature and highest humidity (Figure 3), which are meteorological conditions least favorable for O₃ formation [39,40]. In contrast, the exceptionally high pollution hours in 2022 can be attributed to extreme summer heat [22]. When excluding data from August, however, the number of exceedance hours in 2022 was slightly lower than in 2023 and comparable to 2021, highlighting the role of extreme weather events in shaping annual variability in O₃ pollution.

To further quantify the relative roles of meteorology and emission control measures in the long-term evolution of extreme O₃, we applied a fixed-emissions CESM simulation and compared the CESM (meteorology-only) series to observations. Both CESM-simulated and observed extreme O₃ concentrations were first normalized relative to 2015 (Figure 4). The 2015-based series shows that both observation and simulation exhibited minor fluctuations before 2019, but a pronounced decline emerged after 2020, particularly in the observations, suggesting that the recent downward trend cannot be explained by meteorological variability alone.

To further isolate the variations after 2020, the normalized time series was re-based to 2020. Observed extreme O₃ concentrations decreased ~10% from 2020 to 2024, while the CESM simulated O₃ only declined ~2.6%. Following the decomposition method, this implies that about 75% of the observed decline can be attributed to emission reductions, with the remaining ~25% linked to meteorological variability. Year-by-year decomposition further indicates that the 2021 dip was mainly meteorology-driven (62%), whereas in 2022–2023 adverse meteorological conditions would have increased O₃, so the observed decreases in those years must have been overwhelmingly driven by precursors emission reductions.

3.2. Spatiotemporal Variation in O₃ Control Regimes

The trends described above, including the continued expansion and increased frequency of O₃ pollution events since 2015 and the decline in extreme O₃ concentrations after 2020, highlight the need to investigate the underlying mechanisms. One key factor potentially influencing this shift is the spatiotemporal variation in O₃ formation regimes. As the chemical sensitivity of O₃ formation depends strongly on the relative abundance of VOCs and NO_x, understanding the prevailing control regime is essential. In this study, we use FNR as an indicator of O₃ formation sensitivity, enabling a detailed analysis of the regional and temporal dynamics of O₃ production across the SCB.

Figure 5 presents the spatial distributions of average HCHO, NO₂, and FNR over the SCB during the O₃ pollution season (April to September) from 2015 to 2024, based on OMI and TROPOMI satellite observations. Elevated HCHO concentrations are observed over the central and southern parts of the basin (Figure 5a), which coincides with the densely populated and industrialized areas. The NO₂ columns exhibit two distinct hotspots centered over Chengdu and Chongqing (Figure 5b), consistent with intensive vehicular and industrial NO_x emissions in these metropolitan regions.

Due to elevated NO_x concentrations in urban areas, the FNR is significantly lower in cities than in surrounding rural regions (Figure 5c), with Chengdu and Chongqing emerging as two prominent low-ratio centers. Based on the threshold values proposed for the Sichuan–Chongqing region (2.5–4.2) [26] and for Chinese cities more broadly (2.3–4.2) [25], our multi-year averages indicate that Chengdu and Chongqing consistently fall within the VOC-limited regime during early afternoon hours in the O₃ pollution season. Areas surrounding the two megacities, along with most other urban centers, predominantly lie within the transitional regime, while more remote regions with limited anthropogenic influence are generally NO_x-limited. This spatial pattern aligns well with other studies in the SCB [20,26].

The interannual variation in HCHO and NO₂ column density during the O₃ pollution season in the SCB is shown in Figure 6. From 2015 to 2024, HCHO levels exhibit an overall increasing trend, with a more pronounced rise observed after 2020. The HCHO values during 2015–2019 show relatively large fluctuations, which may be partly attributed to data quality issues in the OMI product, including frequent data gaps over the SCB. In contrast, NO₂ column densities show a continuous declining trend before 2020. A slight increase in column NO₂ levels is observed in 2021, followed by a continued decrease thereafter.

The annual evolution of FNR distributions during the O₃ pollution season over the SCB is shown in Figure 7. From 2015 to 2017, urban areas exhibited markedly lower FNR values than the surrounding rural and remote mountainous regions. Most cities recorded FNR values below the VOC-limited threshold of 2.5. In particular, the urban cores of Chengdu and Chongqing showed values below 1.0, indicating a strongly VOC-limited regime. With the continued reduction in NO_x emissions and the associated decline in NO₂ concentrations [15], the extent of VOC-limited regions gradually shrank. By 2024, many small- to medium-sized cities in the SCB had transitioned from VOC-limited regimes to transitional or NO_x-limited regimes, with only Chengdu and Chongqing remaining VOC-limited.

3.3. Model Simulation of Extreme O₃ Pollution Events

3.3.1. Description of Extreme O₃ Pollution Events

To investigate the mechanisms driving extreme O₃ pollution in the SCB, a systematic identification of the most severe annual O₃ episodes was conducted for the period 2015–2024 based on surface observations from 22 cities. The selection criteria prioritized both spatial extent and pollution intensity, with the primary focus on events exhibiting the widest spatial coverage of O₃ exceedances (>200 μg m⁻³). Specifically, for each year, the dominant episode was defined as the continuous period during which the largest number of cities simultaneously experienced O₃ exceedances for at least two consecutive days. In cases where multiple events had similar spatial extents, the event with the higher average exceedance O₃ concentration was selected to ensure representation of high-intensity pollution. The identified extreme O₃ pollution episodes are summarized in

Table 1. Extreme O₃ pollution episodes in each year from 2015 to 2024.

Year	Dominated Period	Mean Exceeded O3 (μg m−3)	Max O3 (μg m−3)	No. of Affected Cities
2015	27–29 April	227	270	9
2016	23–25 August	222	278	9
2017	10–12 July	233	335	13
2018	15–16 May	228	285	16
2019	15–18 August	233	305	17
2020	26–28 August	228	327	14
2021	30 July–3 August	225	343	18
2022	5–8 July	219	323	16
2023	15–18 July	221	333	17
2024	20–23 August	220	258	19

.

Based on the annually selected extreme O₃ pollution events during the 2015–2024 period, this study further analyzed the meteorological conditions during these events to identify their common features and interannual variability. Overall, most extreme O₃ events occurred under the control of high-pressure systems, characterized by persistent subsidence in the boundary layer, sparse cloud cover, and weak surface winds (typically <3 m s⁻¹ in the afternoon). These stagnant meteorological conditions hindered pollutant dispersion and removal, creating favorable conditions for ozone accumulation. Figure 8 illustrates the spatial distribution of the 500 hPa geopotential height field, surface wind patterns, and total cloud water content during the July 2017 event, which clearly highlights the typical atmospheric circulation configuration and low-cloud characteristics of such episodes. Notably, the 2023 extreme event exhibited slight differences, influenced by peripheral typhoon cloud systems (Figure S4). Localized cloudiness or transient overcast conditions emerged in the basin, which, to some extent, weakened the radiation-enhancing effects compared to other years, making it an exception to the otherwise stable synoptic pattern.

The impact of chemical regimes on O₃ formation was also examined using satellite-derived FNR. Results demonstrated that basin-wide FNR values during most extreme O₃ events generally exceeded 2.5 (Figure 9), indicating transitional regimes where O₃ production is jointly sensitive to VOCs and NO_x. In contrast, megacities like Chongqing and Chengdu exhibited relatively lower FNR values (<2.0), suggesting VOC-limited conditions. Additionally, since 2020, FNR values in these megacities have shown an upward trend, implying a gradual weakening of VOC limitation and an increasing importance of NO_x control in O₃ mitigation.

Considering the representativeness of meteorological conditions, data availability of emission inventories, and evolving chemical regimes, this study selected the July 2017 extreme O₃ event as a typical case for subsequent model simulations. Although the 2021 event featured slightly higher peak O₃ concentrations, the lack of complete official anthropogenic emission inventories beyond 2020 constrained accurate assessment of emission-driven contributions and effective model validation. In contrast, the 2017 case not only exhibited typical stagnant meteorological conditions and high O₃ levels but also benefited from detailed and reliable emission data, making it more valuable for in-depth analysis.

To further explore the factors contributing to the recent decline in extreme O₃ concentrations, we conducted a case study using the WRF-CMAQ model to simulate the 2017 extreme O₃ pollution event. Three experiments were designed to assess the influence of anthropogenic emission changes on peak O₃ levels: a control simulation using July 2017 emissions (CTL), sensitivity experiment 1 using July 2015 emissions (EMIS_2015), and sensitivity experiment 2 using July 2020 emissions (EMIS_2020). This approach allows us to isolate the impact of emission evolution on the intensity of extreme O₃ events.

3.3.2. Model Evaluation

The model performance was assessed against observed hourly O₃ concentrations from 22 cities across the SCB during the 10–12 July 2017 episode, yielding 1584 paired samples in total. (Figure 10). The model reproduced the temporal variations reasonably well, with correlation coefficients in most cities ranging from 0.6 to 0.9 (Figure 10a). In approximately two-thirds of the cities, the model overestimated O₃ concentrations, with an average NMB of 18.2%. The scatter plot of observed versus simulated O₃ concentrations (Figure 10b) reveals that the model performed better at higher concentrations (above 100 μg m⁻³), while it tended to underestimate O₃ levels below 50 μg m⁻³. This bias is mainly due to underestimation of nocturnal boundary-layer turbulent mixing [41], which enhances NO titration, with additional contributions from uncertainties in anthropogenic emissions (especially due to lack of high-resolution inventories) and the representation of fine-scale meteorology [42]. The NME was below 60% in most cities, with an average of 40.9%.

To further assess the model’s ability to reproduce the spatial distribution of O₃, Figure 10c presents the distribution of simulated O₃ concentrations at 14:00 LT on 11 July, which corresponds to the time of the highest O₃ levels and widest spatial coverage of O₃ pollution during the simulation period. The simulated O₃ field is overlaid with observational data from 156 monitoring stations across the SCB. In most urban areas, the modeled high O₃ concentrations closely match the observed values, as indicated by the similar colors between the filled contours and the overlaid scatter points. This spatial agreement demonstrates the model’s strong capability in capturing both the magnitude and spatial extent of the extreme O₃ pollution event.

3.3.3. Impact of Emissions Variation on Extreme O₃

Figure 11a compares the hourly O₃ concentrations averaged across all cities in the Sichuan Basin on 11 July 2017, as simulated by the three model experiments. Among the three cases, the control simulation (CTL) using 2017 emissions produced the highest O₃ levels throughout the day, especially around midday, with concentrations exceeding those in the two sensitivity experiments by 20–30 μg m⁻³. The two sensitivity simulations, EMIS_2015 and EMIS_2020, showed relatively similar diurnal patterns, with EMIS_2015 producing slightly higher O₃ concentrations than EMIS_2020 in the afternoon hours. The two megacities, Chengdu and Chongqing, exhibited similar daytime patterns to the SCB city average (Figure 11a,b). However, during nighttime, the EMIS_2015 simulation showed significantly lower O₃ levels compared to the other two experiments.

During this extreme O₃ pollution episode, FNR values across 22 cities in the SCB in the afternoon (13:00 LT) were 6.5 in the CTL experiment, and 4.4 and 4.8 in the EMIS_2015 and EMIS_2020 experiments (Figure 12a), respectively, all indicating a NO_x-limited regime [25,26]. However, both Chengdu and Chongqing consistently fell within the VOC-limited regime in all three experiments. This was particularly evident in Chengdu, where the FNR values remained below 0.5 across all scenarios, indicating strong sensitivity to VOCs.

According to the MEIC emission inventory, although total NO_x emissions in the SCB declined continuously from 2015 to 2020 (Figure 13), the urban NO_x emission flux in 2017 (CTL experiment) was 7% and 18% higher than in 2015 (EMIS_2015) and 2020 (EMIS_2020), respectively. Meanwhile, urban VOC emission fluxes in CTL were 19% and 26% higher than those in EMIS_2015 and EMIS_2020, respectively. These precursor differences led to enhanced O₃ production in the CTL simulation, as also indicated by the CMAQ process analysis (Figure 12b), which showed the strongest chemical production of O₃ in urban areas under the 2017 emissions scenario.

To further elucidate the underlying chemical mechanisms, an IRR analysis was conducted to explicitly quantify the key reaction pathways controlling O₃ formation and loss under different emission scenarios (Figure 14). The daytime (10–14 LT) O₃ budget (production, chemical loss, and titration) indicates that the CTL case exhibited the largest fluxes in all terms, consistent with its highest simulated O₃ concentrations (Figure 14a). Both EMIS_2015 and EMIS_2020 showed systematically weaker production and loss, reflecting the strong sensitivity of O₃ chemistry to precursor emissions. Similar patterns were also found at night, confirming the robustness of these results.

In addition, the reaction rates of anthropogenic VOC (aVOC) oxidation by OH (aVOC → RO₂), RO₂ + NO → NO₂ (the dominant propagation step linking NO_x to O₃ formation), and OH + NO₂ → HNO₃ (the major NO_x sink) were examined (Figure 14b). The CTL experiment again showed the strongest fluxes in all three reactions, demonstrating that both VOC-derived radical production and NO-driven propagation were maximized in 2017, thereby sustaining the highest O₃ levels under favorable meteorological conditions. In contrast, the EMIS_2020 case exhibited substantially weakened radical production and propagation, highlighting the role of sustained emission reductions in driving the observed decline in extreme O₃ concentrations after 2020.

4. Discussion

The long-term changes in O₃ pollution in the SCB since 2015 reveal a clear evolution in its spatial characteristics, intensity, and chemical formation regimes. The simultaneous persistence of high spatial extent and frequency of O₃ pollution since 2020, alongside a marked decline in the intensity of extreme concentrations, suggests a fundamental shift in the characteristic of O₃ pollution in the region. Decomposition analysis reveals that ~75% of the decline stemmed from precursor emission reduction, while meteorological variability accounted for only ~25%. The persistence of declining extreme O₃ levels under adverse meteorological conditions in 2022–2023 further underscores the dominance of anthropogenic mitigation measures in curbing regional O₃ pollution.

Satellite observations of HCHO and NO₂ offer valuable insights into this transformation. The rising HCHO levels and declining NO₂ concentrations observed between 2015 and 2024 have significantly altered the chemical environment of O₃ formation. These trends are consistent with prior findings in both the SCB and other heavily polluted regions in China. For instance, a similar upward trend in HCHO has also been reported in this region [26] and over North China [43]. In contrast, NO₂ column densities exhibited a continuous downward trend before 2020, again in agreement with earlier studies [15,26]. These simultaneous changes in precursors led to increasing values of the FNR. The upward trend in FNR indicates a shift in chemical regimes from VOC-limited to transitional or NO_x-limited conditions in many cities across the SCB, particularly after 2020. This is in line with the threshold criteria for chemical regimes in Chinese cities proposed by other studies [25,26].

This regime transition has substantial implications for O₃ photochemistry. Under VOC-limited conditions, reducing NO_x has limited efficacy and may even worsen O₃ levels. However, once the regime shifts to NO_x-limited or transitional, ongoing reductions in NO_x can effectively suppress O₃ production [12]. In the SCB, many small- to medium-sized cities have already transitioned into such regimes, while Chengdu and Chongqing remain VOC-limited due to high NO_x emissions. The regime shift also reduces NO titration effects and enables more efficient photochemical O₃ production under favorable meteorological conditions, such as high temperatures and intense solar radiation [44,45].

While our long-term FNR analysis indicates a shift in O₃ formation sensitivity across the Sichuan Basin, it is important to recognize that HCHO represents only one class of oxygenated VOCs (OVOCs) and does not fully capture the speciated VOC mixture that controls local ozone production. HCHO is a useful, satellite-retrievable proxy for VOC reactivity because it is both directly emitted and produced during VOC oxidation. Nevertheless, observational studies and ozone formation potential analysis in the SCB show that alkenes, aromatics, and OVOCs often contribution disproportionately to ozone formation potential (OFP) relative to their mixing ratios [46,47]. These species can therefore dominate local O₃ production even when bulk HCHO suggests a given regime.

It should also be noted that the spatial resolution of satellite products introduces additional uncertainty into city-scale FNR estimates. Previous work has shown that coarse-resolution sensors such as OMI can lose more than 10% of spatial information compared with TROPOMI, and large pixels may fail to capture fine-scale urban variability, particularly for localized NO₂ hotspots [48]. Such representation errors can bias FNRs, as NO₂ tends to be more spatially heterogeneous than HCHO. Nevertheless, in urban areas with stronger column signals, the relative errors are generally moderate (<50%), whereas rural regions may experience much larger uncertainties. Since our analysis focuses mainly on urban centers such as Chengdu and Chongqing and relies on the higher-resolution TROPOMI product in recent years, the impact of spatial resolution limitations on our basin-scale conclusions is expected to be limited.

In addition to changes in precursor emissions and chemical regimes, meteorological conditions also played an important role in shaping extreme O₃ pollution across the SCB. Interannual variations in temperature, humidity, and cloudiness significantly influenced the intensity of extreme events. For instance, the sharp decline in O₃ exceedance hours in 2021 coincided with unusually low seasonal temperature and high humidity, conditions that suppress photochemical O₃ formation by reducing actinic radiation, slowing temperature-dependent reactions, and enhancing removal through wet deposition [39,40]. Conversely, the exceptionally high O₃ pollution hours in 2022 were largely driven by extreme summer heatwaves, illustrating how meteorological anomalies can amplify pollution episodes. However, the decomposition analysis shows that the overall decline in extreme O₃ concentrations after 2020 cannot be explained by meteorology alone. Adverse meteorological conditions in 2022–2023 should have driven higher O₃ levels; however, observed concentrations continued to decline. This contrast provides clear evidence that sustained reductions in precursor emissions were the dominant driver of the post-2020 decline. These results underscore both the strong modulation of extreme O₃ by meteorology and the effectiveness of emission controls in mitigating peak concentrations.

Beyond year-to-year differences, typical extreme O₃ events shared common meteorological characteristics. Extreme O₃ pollution episodes were often associated with persistent high-pressure systems, clear skies, and stagnant surface winds (<3 m/s), which together facilitated the accumulation of O₃ precursors and enhanced photochemical production. These features are indicative of strong solar radiation and poor dispersion conditions, consistent with previous findings [39].

Model simulations further confirm the key role of anthropogenic emissions in shaping extreme O₃ events. The CTL simulation, which used 2017 emissions, produced the highest O₃ concentrations among the three experiments. This result is consistent with its higher levels of NO_x and VOC emissions, along with elevated FNR values, which together created more favorable conditions for photochemical O₃ formation.

Interestingly, although the EMIS_2015 and EMIS_2020 simulations yielded similar average FNR values across urban areas in the SCB, their different NO_x levels led to distinct O₃ behaviors. In the higher-NO_x EMIS_2015 experiment, daytime O₃ formation was slightly stronger due to enhanced photochemical reactions under NO_x-limited conditions, while nighttime O₃ levels were lower because of increased titration by NO. These results emphasize the dual role of NO_x in NO_x-limited regimes: it enhances O₃ formation during the day but suppresses O₃ concentrations at night through titration processes [39].

Altogether, the persistent increase in average O₃ concentrations and the expansion of O₃ pollution coverage in the SCB appear to be driven by a combination of rising VOC availability (e.g., HCHO), decreasing NO_x, and shifting chemical regimes. At the same time, the recent decline in extreme O₃ concentrations since 2020 can be attributed to continued NO_x emission reductions and increased prevalence of NO_x-limited conditions across the basin.

5. Conclusions

This study examined the evolution of extreme ozone pollution in the SCB from 2015 to 2024 based on surface observations. The changes in O₃ formation regimes were further investigated using satellite-retrieved column densities of HCHO and NO₂. To explore the underlying mechanisms driving the variations in extreme O₃ pollution, a typical episode was simulated using the WRF-CMAQ model. The key findings are summarized as follows.

From 2015 to 2024, average O₃ concentrations in the Sichuan Basin increased, alongside a rise in both the number of cities affected and the total O₃ pollution hours. While the number of polluted cities stabilized after 2020, pollution hours peaked in 2022 before declining. Notably, extreme O₃ concentrations have been steadily decreasing since 2020. These trends suggest that O₃ pollution in the region has become more widespread but less intense. Decomposition analysis show that precursor emission reduction contributed ~75% of the post-2020 decline in extreme O₃ concentrations, meteorological variability explained the remaining ~25%.

Satellite observations of HCHO and NO₂ from 2015 to 2024 reveal a clear spatial and temporal evolution of O₃ formation regimes across the SCB. Chengdu and Chongqing consistently remained in the VOC-limited regime, while surrounding urban and rural areas exhibited a gradual transition from VOC-limited to transitional or NO_x-limited regimes, driven by increasing HCHO levels and decreasing NO₂ concentrations. The corresponding rise in FNR values indicates a regional shift in O₃ sensitivity, which likely contributed to the broadening spatial extent and persistence of O₃ pollution during this period.

Model simulations of a representative extreme O₃ pollution episode demonstrate that stronger anthropogenic emissions across the cities in the SCB in 2017 led to higher O₃ concentrations, particularly under NO_x-limited conditions. Despite similar FNR patterns, the EMIS_2015 case with higher NO_x levels enhanced daytime O₃ formation but increased nighttime titration, leading to lower nocturnal O₃ compared to EMIS_2020. Over time, emission reductions, especially in NO_x, have weakened the intensity of extreme O₃ events, and the observed regime shift toward NO_x-limited conditions has further limited peak O₃ buildup. IRR analysis further revealed that emission reductions substantially weakened both radical production and NO_x-driven propagation, directly contributing to the decline in extreme O₃ concentrations after 2020. These findings underscore the critical role of emission control and chemical regime transitions in shaping long-term O₃ pollution trends.