Attributing Increases in Ozone to Accelerated Oxidation of Volatile Organic Compounds at Reduced Nitrogen Oxides Concentrations

Abstract

Introduction

Ground-level ozone (O₃) is an air pollutant detrimental both to human health (1) and to terrestrial vegetation (2). In the mid-20th century, Haagen-Smit et al. proved that O₃ is generated by the photochemical reaction of volatile organic compounds (VOCs) and nitrogen oxides(NO_x) in sunlight (3). It is now well established that the chemistry of O₃ formation is highly nonlinear and pertains to the coupling of the “NO_x cycle” and the “RO_X (RO_X = OH + HO₂ + RO₂) radical cycle” (4–6). The NO_x and RO_X cycles are terminated by the cross-reaction of NO_x and/or RO_X. From the perspective of O₃ pollution reduction, if HO₂ self-reaction and its reaction with RO₂ constitute the dominant termination reaction (known as the “NO_x-limited” regime), reductions in NO_x emissions serve the greatest benefit. Meanwhile, if the reactions of NO₂ with OH and RO₂ form the dominant termination reaction (known as the “VOC-limited” regime), reductions in VOC emissions prove the most effective (5, 7). In the context of China, most city clusters [e.g. the Pearl River Delta, Yangtze River Delta (YRD) and Northern China Plain] are found to be VOC-limited (5), with anthropogenic VOCs dominating, whereas most rural areas are NO_x-limited in this regard (8).

China has been confronting serious air pollution for several years (9, 10). In an attempt to improve air quality, China has implemented strict pollution control policies since 2013. However, ground-level O₃ concentrations have not improved following NO_x abatement (11, 12). Recent studies have shown that VOCs and NO_x should be controlled simultaneously to reverse O₃ increases (13, 14). These results concur with those of Pusede (15) and Roberts et al. (16), who found that simultaneous VOC and NO_x abatement could considerably mitigate the peak of O₃ production and prevent surface air quality deterioration.

The large-scale emission reductions seen during the COVID-19 lockdown are unprecedented opportunities for improving our understanding of O₃’s response to the control of its precursor emissions. The Chinese government adopted countrywide measures from January 23 to February 13, 2020, to prevent the propagation of the disease. According to satellite and surface observations, NO₂ concentrations in China decreased by >50% during the lockdown period compared with those in the pre-lockdown period, and other pollutants such as PM_2.5 and CO also decreased significantly (17–19). However, ground-level O₃ concentrations in parts of China increased substantially (20–22). The increase in O₃ during the lockdown has been tentatively attributed to weakened titration (23, 24), O₃ production (25), and decreased PM_2.5 (19, 24). However, the changes in the chemical mechanisms connecting O₃ levels with the abatement of its precursors are rarely discussed.

Here, we use detailed field observations during the Shanghai lockdown combined with an observation-based model (OBM) to show evidence for the increased photochemical formation of O₃ from VOC oxidation with decreasing NO_x. We demonstrate that this mechanism can explain the observed O₃ increase. This also helps explain why O₃ concentrations in China increased in recent years as NO_x emissions have greatly decreased while VOC emissions have remained unchanged (26, 27). We further show that a reasonable ratio of annual NO_x and VOC reduction decreases O₃ photochemical production, with important implications for developing O₃ pollutant control strategies to decrease O₃ in China.

Results and Discussion

Worsening O₃ air quality from 2014 to 2020

Fig. 1A to F shows the spatial distribution of the winter and summer mean maximum daily 8-h average O₃ (MDA8 O₃) concentrations around the YRD (Figure S1) in 2014 and 2020 (maps for 2015 to 2019 are presented in the Figure S2). In both years, high levels of MDA8 O₃ concentrations were observed in the eastern YRD, especially in Shanghai and surrounding regions, where the emission intensity was the highest (28). As illustrated in Fig. 1C and F, a slight increase in MDA8 O₃ concentrations occurred across the YRD from 2014 to 2020 even with considerable reductions in the emissions of major air pollutants (Figures S3 and S4, and Tables S1 to S3). According to an estimate of the temporal trend (Fig.   1G and H), the winter (summer) YRD concentration of MDA8 O₃ increased by 0.43% yr⁻¹ (1.25% yr⁻¹) from 2014 to 2020. The titration effect caused by the 2.3% yr⁻¹ (4.1% yr⁻¹) NO_x decreases along with unchanged VOCs was found to be a significant cause of such increases. Meanwhile, the 6.2% yr⁻¹ (9.2% yr⁻¹) reductions in PM_2.5 may also be major factors as aerosol particles scavenge hydrogen oxide (HO_x) and nitrogen trioxide (NO₃) radicals, which would otherwise produce O₃ (29). Similar increases in O₃ simultaneously recorded with the progressive implementation of clean-air actions were generally observed in various areas of China, including the North China Plain and Pearl River Delta (30, 31). Such results point to potential increases in oxidation capacity in the YRD. Although NO_x reduction has traditionally been considered important for controlling O₃, recent studies have shown that the co-abatement of VOCs and NO_x is more important for suppressing O₃ photochemical production (32, 33). In this regard, COVID-19 represents an opportunity that informs our understanding of how VOCs and NO_x should be effectively controlled to reverse O₃ increases.

Fig. 1.

Variations in winter and summer YRD’s MDA8 O₃ concentrations from 2014 to 2020. (A and B) Distributions of winter mean MDA8 O₃ concentrations in YRD in 2014 and 2020. (D and E) Distributions of summer mean MDA8 O₃ concentrations in YRD in 2014 and 2020. (C and F) Reductions of winter and summer mean MDA8 O₃ concentrations between 2014 and 2020 (negative values). (G and H) Interannual variation (Error bars: SD) of NO₂, MDA8 O₃ and PM_2.5 in the YRD during winter and summer.

Open in new tabDownload slide

The national lockdown policy in response to COVID-19 beginning on 2020 January 24, impelled a rapid drop in transportation and industrial emissions, compounded by the Chinese Spring Festival holiday (19, 34). For the present work, the “first period” (FP) comprised the normal emissions from January 1 to January 11, and the “second period” (SP) included the lockdown from January 24 to February 3 when lockdown restrictions were at their height (34). Fig. 2 and Figure S5A show 40% to 80% observed decreases in surface NO₂ levels and tropospheric NO₂ columns from the TROPOspheric Monitoring Instrument (TROPOMI) satellite, suggesting rapid falls in NO_x emissions in the YRD (the measurement of sites changes are discussed in SI Appendix A, Figure S6 and S7, and Tables S4 to S7). TROPOMI Formaldehyde (HCHO) observations suggest a weaker reduction in VOC emissions (Fig. 2E). For example, TROPOMI HCHO decreased by 12% in the western YRD (red squares), whereas it increased by 38% in the eastern area (blue squares). The surface NO₂ concentrations came from the Ministry of Ecology and the Environment (MEE) monitoring site network. For the eastern YRD, as laid out in Fig. 2E, the average falls in the course of SP were 40% for surface NO₂ and 60% for TROPOMI NO₂. Conversely, the O₃ concentrations quantified at MEE sites and TROPOMI satellite increased in the course of the SP (Figure S6B and Fig. 2F). Ground-level MDA8 O₃ in the YRD increased from 18 ± 5 ppb during the FP to 40 ± 5 ppb during the SP.

Fig. 2.

Air quality changes in YRD in response to the COVID-19 lockdown from January 1 to February 20, 2020. (A and C) The red lines show observed concentrations of 24-h average NO₂ and MDA8 O₃ at YRD. The NO₂ and O₃ observations are averages across the monitoring sites of the MEE. Also shown are observations results sampled at the background site and the corresponding correlation coefficients (r). (B) VOCs observation data from background site. (D, E, and F) Relative percentage change [(FP-SP)/FP] in tropospheric NO₂, HCHO, and O₃ column densities measured by the TROPOMI satellite instrument during FP and SP. The TROPOMI aggregation is on a 5.5 × 3.5 km grid.

Open in new tabDownload slide

We utilized a machine learning model to predict O₃ levels in a “business-as-usual” scenario (Materials and Methods section). Figure S8 shows that the O₃ concentration during the lockdown significantly increased compared with that in the business-as-usual scenario. The increase in O₃ might well be explained by unfavorable meteorological conditions and the co-abatement of O₃ precursors. Therefore, we prioritized the use of multiple linear regression (MLR) models to assess the meteorological short-term effects on MDA8 O₃ before and after the lockdown (2020 January 1 to 2020 February 29) in the YRD (35). The coefficient of determination (⁠|$R^{ 2}$|⁠), for the MLR model in fitting MDA8 O₃ anomalies in the YRD was 0.52 (regression coefficients are summarized in Table S8). We then used the Lindeman, Merenda, and Gold method to quantify the relative importance of each meteorological variable in explaining that the atmosphere-driven O₃ increases (Figure S9). A decrease in fractional total cloud cover during the lockdown (Table S9) was the most significant driver of the O₃ increase seen in the YRD with a contribution of 41%. Less cloud cover favors strong solar radiation, which in turn promotes O₃ production (36). We found that 850-hPa relative humidity (RH850) was negatively correlated with O₃ (regression coefficient: −0.3) because wet aerosols formed under high humidity reduce ultraviolet actinic flux and thereby inhibit photochemical reaction rates (37). Meanwhile, the weakening of the 10-m zonal wind (U10, 7%) and 10-m meridional wind (V10, 3%) caused less downwind flux of O₃, resulting in more O₃ remaining in situ in the YRD (Table S9) (38). These findings further demonstrated the importance of meteorological short-term effects on O₃ (39). However, VOC and NO_x reduction effects on O₃ formation mechanisms must be further considered to effectively understand the impact of the lockdown on O₃ production.

Evidence for O₃ formation by accelerated photochemistry

The cycle between OH and HO₂ is highly pertinent to O₃ formation, influencing the chain reaction propagation (40). Thus, the levels and reaction routes of free radicals are significant factors in O₃ generation. The simulated diurnal patterns (3-day average) of free radicals at urban and background sites are displayed in Fig. 3A and B, and Figure S10A and B. The highest levels of free radicals were modeled around 12:00 to 13:00 at two sites. The daily peak of OH at the urban site increased by ∼0.69× 10⁶ cm⁻³ in the SP relative to the FP but decreased by 0.86 × 10⁶ cm⁻³ at the background site. Meanwhile, the daytime-averaged OH concentration (08:00 to 16:00) at the urban site (0.89 × 10⁶ cm⁻³) was comparable with that at the background site (0.67 × 10⁶ cm⁻³) during the SP. The daytime-averaged OH level simulated in this research was comparable with that (1.5 ± 0.9 × 10⁶ cm⁻³), modeled on a Beijing winter (41).

Fig. 3.

Relevant simulation results (3-day averaged) for urban site during FP and SP. (A and B) Simulated average diurnal variations in OH and HO₂. (C and D) Daytime (08:00 to 16:00 LT) average OH-HO₂-RO₂ budget. Units: ppb h⁻¹. (E and F) Model-simulated average chemical budgets of O₃.

Open in new tabDownload slide

In comparison, the daily peak of HO₂ at the urban site increased by ∼0.95 × 10⁷ cm⁻³ but decreased by 0.49 × 10⁷ cm⁻³ at the background site. The daytime-averaged HO₂ concentration at the background site (1.02 × 10⁷ cm⁻³) during the FP was more than double that at the urban site (0.41 × 10⁷ cm⁻³), whereas the HO₂ concentrations were similar at both sites during the SP, which was more likely because there were less VOCs at the background site and thus less HO₂ produced through photochemical reactions. The daytime-averaged HO₂ concentration at both sites during the SP was higher than that (0.52 ± 0.23 × 10⁸ cm⁻³) in Beijing during typical winter pollution episodes (the atmospheric oxidation capacity and the OH reactivity are discussed in SI Appendix B and Figures S11 and S12) (42). Table S10 compares the OH and HO₂ levels for different Chinese cities. The differing levels of free radicals in various studies may be attributed to the different air pollutants and weather conditions at each locale sampled (40). The peak OH and HO₂ concentrations responded differently at the two sites, highlighting the complex effects of the levels of NO_x and VOC reductions on photochemistry (23, 43).

Because OH–HO₂–RO₂ is to a great extent responsible for the oxidation capacity of the atmosphere, calculating its radical sources and sinks offers an insightful understanding of the effects of the co-abatement of NO_x and VOCs on atmospheric chemical processes. The average production and loss rates for OH–HO₂–RO₂ radicals at the urban and background sites are shown in Fig. 3C and D, and Figure S10C to D. The reaction HO₂ + NO was the main OH production source, with reaction rates reaching 1.27 and 2.71 ppb h⁻¹ for urban and background sites during the FP, respectively. However, this reaction increased to 1.67 ppb h⁻¹ at the urban site and decreased to 0.84 ppb h⁻¹ at the background site during the SP. Next, came the photolysis of HONO. The OH production rate from this reaction weakened by 48% and 66%, respectively, for the two sites. The main mechanism of OH consumption was the production of HNO₃ and HO₂. During the FP, the high OH + NO₂ reaction rate (to produce HNO₃) was for the most part due to the relatively high daily average NO_x level at the urban (39.1 ppb) and background (26.6 ppb) sites (44). With NO_x abatement, this reaction fell by 41% and 59% at the urban and background sites, respectively. Unlike the OH + NO₂ response, the rate of HO₂ production from OH rose by 57% at the urban site but fell by 58% at the background site during the SP, probably due to different levels of NO_x and VOC co-abatement in the two locations.

In addition to OH, RO and oxygenated VOCs (OVOCs) were also primary sources of HO₂. These two chemical pathways decreased by 18% and 32%, respectively, at the urban site during the SP but decreased by 73% and 74%, respectively, at the background site. The major HO₂ removal mechanism was its reduction reaction with NO, which produced OH radicals. Some research has also revealed that HO₂ radicals can be taken up by aerosols (45). The RO₂ formation rates of OH and OVOCs did not change significantly for the urban site but decreased by 36% and 67%, respectively, at the background site. Substantial VOCs reductions are important factors in the decline of the background site RO₂ formation rates. We keenly point out that the reactions NO + HO₂ and RO₂ + NO are the two major pathways for radical cycling at the two sites during the FP and SP, which are common in NO_x-rich environments (46, 47). Obviously, our results show that the OH–HO₂–RO₂ cycling at urban stations remained at high levels during the lockdown, with particularly enhanced cycling between OH and HO₂. Meanwhile, significant weakening was observed at the background sites, which may have depended on the levels of VOC abatement.

We further examined the O₃ formation mechanisms for the lockdown. O₃ was mainly produced through the oxidation of NO by HO₂ and RO₂ (48). For the O₃ loss, the predominant scavenging pathways were the photolysis of O₃ and the NO₂ + OH reaction. Fig. 3E and F, and Figure S10E and F show the detailed O₃ chemical budgets before and after the lockdown at the urban and background sites. The major photochemical production of O₃ was clearly demonstrated, with net daytime O₃ average generation rates of 1.5 and 3.8 ppb h⁻¹ for the urban and background sites, respectively, during the FP. The O₃ production intensity at the two sites was lower than that derived from an urban site in Wuhan autumn (8.8 ppb h⁻¹) (44) and comparable with that in Beijing’s clean episode (2 ppb h⁻¹) (41). During the SP, the net daytime O₃ average generation rates increased to 1.8 ppb h⁻¹ for urban sites but decreased to 0.8 ppb h⁻¹ for background sites. This may have had something to do with the O₃ production rate increase from NO_x reduction exceeding the decrease rate associated with VOC reduction at urban sites, causing an increase in net O₃ production rate and, thus, a substantial O₃ increase (Figure S13A and B). Conversely, the NO_x reductions at the background sites had less impact on O₃ than the control of the VOCs, leading to the decrease in O₃ production rates (Figure S13C and D). The O₃ production rate peaks around midday (literally at around 12:00 LT), only to be succeeded by a major drop during the afternoon, consistent with the overall findings from other works that reported noon or afternoon peaks (49, 50).

The links between O₃ and precursor species are elucidated more by the relative incremental reactivity (RIR) process using the OBM (Materials and Methods section). As can be seen in Figure S14, the simulation results for the two sites were the same during the FP and SP. The RIR values of each VOC category were positive for all scenarios, and only NO_x was negative, indicating that O₃ formation was consistently in the VOC-limited regime (51). During the FP, the two sites had high RIR for aromatics and alkenes, which showed that they control the production of O₃ to a larger extent. Notably, we found that O₃ production at both sites was more sensitive to alkenes and CO during the SP, suggesting that future targets of O₃ control should emphasize reducing not only the active species of VOCs but also CO emissions.

Implications for air pollution control

O₃ formation was determined to be in the VOC-limited regime at the background and urban sites. However, how much VOCs should be controlled to achieve the optimal O₃ reduction was nonetheless uncertain, particularly in a future emission scenario where VOCs and NO_x tend to be controlled simultaneously. In the pursuit of this objective and to offer detailed information on the requisite decreases in VOCs and NO_x, we simulated the maximum O₃ production rate (MOPR) increment with the reduction of both VOCs and NO_x (Fig. 4A and B). We found that at the background (urban) site, when the mixing ratios of VOCs were lowered from 0% to 95%, the appropriate reduction percentages of NO_x tended to fall between 0% and 78% (85%) or in the range of 94% (96%) to 100% for the zero MOPR increment. Admittedly, reducing NO_x by 94% to 100% may prove impracticable. Therefore, this section looked primarily at the range 0% to 78% (85%) in NO_x reduction to produce the best possible VOC and NO_x control measures. The researchers found that the MOPR could be reduced when the NO_x/VOC co-abatement ratio was <1.5 for the background site, resulting in the suppression of O₃ production, and 1.25 for the urban site (i.e. the cutting ratios of NO_x/VOCs where the curves meet the MOPR zero increment line). However, the simulation results for OH and HO₂ radicals at both sites showed that the radical concentrations can be effectively reduced when NO_x/VOCs <1 (Fig. 4C and D and Figure S15). This indicates that the weakening of OH and HO₂ experiences lags relative to the MOPR, which may influence atmospheric PM_2.5 formation (52). Similar simulation results for the two sites indicate the consistency of future co-abatement strategies for Shanghai and the YRD.

Fig. 4.

Simulation in background and urban sites according to the emission changes of NO_x and VOCs, with first period (yellow pentagon) as the simulation scenario. (A and B) MOPR increment in background and urban site, the black line represents zero MOPR increment. (C and D) Isopleths of daytime OH concentration in background and urban sites. (E) In a 4.5% annual NOx reduction scenario, simulation of MOPR under five ΔNO_x/ΔVOCs reduction ratios from 2021 to 2030. (F) MOPR is simulated for alkenes, aromatics, and alkenes and aromatics at an annual reduction rate of 9% for the scenario.

Open in new tabDownload slide

We obtained a reasonable ratio above by controlling the total abatement of NO_x and VOCs. However, to achieve the National Ambient Air Quality Standard for O₃, NO_x and VOC emissions must be reduced to a large degree (14, 53), which cannot be accomplished in the short-term. Thus, reasonable annual reduction plans should be developed to ensure their enforceability. Accordingly, NO_x emissions are assumed to annually decrease by 4.5% (ΔNO_x = 4.5%). The reduction rate in NO_x was referenced to the trend of winter NO₂ concentration in Shanghai, Nanjing, Hangzhou, and the YRD region from 2014 to 2020 (Figure S16). Working with this postulate, we looked at the associated benefits of various VOC emission reduction rates for 2021 to 2030. As presented in Fig. 4E, we designed five control scenarios with different VOC control rates. For every scenario, the annual emission reduction rate was the same during the emission control period. The results show the additional lowering of the MOPR with faster VOC emission control than those under lower VOC falls. Taking the annual VOC reduction rates (ΔVOCs) of 4.5%, 9%, and 18% as examples, the MOPR reduction rates were 4.5, 6.7, and 8.5 ppb h⁻¹, respectively, compared with the zero VOC reduction (0%) in 2030. Apparently, controlling VOCs to large extents can guarantee the effectiveness of NO_x control in lowering O₃. Such reduced O₃ levels arising from extensive VOC control strategies can introduce other positive effects through the reduction of their negative effects on humans and nature (14).

The above results indicate that the MOPR decreases yearly when annual ΔVOCs ≥4.5%. Considering the lag of free radical decrease relative to MOPR (Figure S17) and economic feasibility, we considered annual ΔVOCs of 9% to be reasonable. Meanwhile, previous results showed that alkenes and aromatics have stronger control effects on O₃ formation. To highlight the role of reactive VOCs in future controls, we conducted sensitivity simulations by removing the effects of other VOCs (Fig. 4F). We found that the reduction of alkenes and aromatics alone can only offset the negative impact of NO_x reduction relative to zero VOC reduction conditions. For the condition with both reductions, we see from Fig. 4F that the weakening of MOPR is also significant relative to the condition with all VOCs reduced. This implies that future VOC plans should prioritize the reduction of alkenes and aromatics (54).

In summary, we have shown evidence from field observations in the Shanghai lockdown that reveal the increasing photochemical formation of O₃ from VOC oxidation under decreased NO_x emission to promote O₃ production. The evidence suggests that the O₃ increase by the NO_x reduction counteracted the O₃ decrease by the VOC emission reduction in the VOC-limited region. Therefore, O₃ control requires more aggressive VOC reduction under current NO_x reduction conditions, particularly for alkenes and aromatic VOCs. What we present thus explains the O₃ increase in the lockdown despite the reductions in its precursors. This also helps explain why O₃ has increased from 2014 to 2020 in response to NO_x emission reduction even as VOC emissions have remained unchanged. Although the national lockdowns are not sustainable, O₃ response to the precursor emission reductions could be used as a reference for future VOC and NO_x control. The reduction associated with O₃ formation may be important for aerosol climate effects and in understanding pollution–weather feedback loops, most particularly the mitigation of PM formation through VOC reduction (55, 56).

Materials and Methods

Observations

The YRD, located in eastern China adjacent to the East China Sea, is one of the biggest city conglomerates in the world. Its location places it in a subtropical monsoon climate. Although the YRD covers only about 1.1% of China’s total land area, it is home to ∼7.6% of China’s total population and accounted for 16.5% of the country’s Gross Domestic Product in 2016 (http://data.stats.gov.cn/). The high population density and widespread heavy industries lead to intensive energy consumption and emission of pollutants. Therefore, the YRD region has experienced severe O₃ pollution and is considered the key prevention and control area for O₃ pollution in China (57, 58). In this study, 26 cities in the YRD were considered (Figure S1A): Shanghai, Nanjing, Wuxi, Changzhou, Suzhou, Nantong, Yancheng, Yangzhou, Zhenjiang, Taizhou, Hangzhou, Ningbo, Jiaxing, Huzhou, Shaoxing, Jinhua, Zhoushan, Taizhou, Hefei, Wuhu, Maanshan, Tongling, Anqing, Chuzhou, Chizhou, and Xuancheng. The above data of routine air pollutants (PM_2.5, NO₂, CO, O₃) at ground level for winter (December, January, and February) and summer (June, July, and August) 2014 to 2020 in 26 cities were obtained from the China National Environmental Monitoring Center (CNEMC). Monitoring data were collected using continuous automated equipment. All equipment passed the applicability test by the CNEMC, and all monitoring data were subjected to validity audits.

The field experiment was conducted at the background (121.021°E, 31.238°N) and urban site (121.507°E, 31.329°N) in YRD region, China. The background site (Figure S1B) is surrounded by Dianshan Lake and several villages with low population density and is about 40 km away from the built-up center of Shanghai. The region is frequently defined as a reference for the study of air pollution in the YRD (59, 60). The urban site is surrounded by a concentration of residential areas, commercial areas, schools and hospitals, which was under strong influence by urban pollutant emissions.

At the background and urban observation sites, NO_x (NO/NO2) and 60 VOCs were measured. To be specific, hourly concentrations of 60 VOC species were monitored by an online gas chromatography analyzer (Focused Photonics, GC955-615), VOC species are detailed in Table S6. Ambient pollutants were monitored by Thermo Environmental Instruments. NO, NO₂, and NO_x were detected by the NO-NO₂-NO_x analyser (Model 42i). For routine gaseous pollutant concentration (PM_2.5, O₃, NO₂, SO₂, and CO) monitoring, we used monitoring data from the CNEMC sites nearest to the background and urban site. In addition, meteorological data (temperature, humidity, etc.) are collected hourly from the ERA5 reanalysis dataset.

Satellite observations of O₃, NO₂ and HCHO columns from the TROPOMI instrument were accessed from https://s5phub.copernicus.eu/dhus/. TROPOMI provides daily global coverage with 5.5 × 3.5 km² pixel resolution. TROPOMI data have been applied in tracking anthropogenic emission changes during the lockdown. In this study we used the TROPOMI level 2 observations (Table S11).

The LightGBM model

LightGBM is a state-of-the-art gradient-boosting framework that uses tree-based learning algorithms. It is designed with faster training speed, lower memory usage, better accuracy and capability of handling large-scale data (61). We trained the LightGBM model with 17 features and ozone labels from 1 January 2019 year to 23 January 2020 (Table S12). To achieve better performance, we performed grid tuning of the parameters by a randomly crossvalidated search on parameter settings. We finally selected the following hyperparameters: max depth = 16, num estimators = 127, reg alpha = 5, learning rat = 0.1. Notably, the emission levels for the “business-as-usual” scenario in the model were calculated using the methodology of shi et al (25).

Observation-based model (OBM)

The OBM for the investigation of atmospheric oxidative capacity and photochemistry (OBM), as a zero-dimensional box model, has been extensively used to simulate atmosphere chemical processes (62–65). It is based on the latest version of the Master Chemical Mechanism (MCM v3.3.1) (66), which contains more than 5800 chemical species and 17 000 reactions, especially, over 200 reactions about VOCs oxidation by chlorine radical. Expect for gas-phase reactions, several heterogeneous processes and physical processes are considered in simulation (67, 68). However, the impact of changes in aerosol loading on HO₂ and O₃ production is not parameterized in the model. The latest study by lvatt et al. (45) shows that a novel “aerosol suppression” regime is suppressing surface O₃ production in China and suggests removing the uptake of HO₂ into aerosols increases O₃ concentrations in the North China Plain by 20% to 30%. This result is based on the 2014 emission scenario. PM_2.5 concentrations in the YRD region are significantly decreased in 2020 compared to 2014 (Fig. 1G and H), therefore we suggest that the sensitivity of surface O₃ to HO₂ uptake in China has currently been reduced. This is in line with lvatt et al., who consider that O₃ sensitivity to HO₂ uptake had reduced in Europe, North America and Japan as policy interventions had reduced particulate-matter concentrations.

In this study, we used the field-measured VOCs, O₃, NO, NO₂, SO₂, and CO, and physical parameters to constrain the boundary of the model and input them into the model for calculation. The photolysis frequencies data were obtained from the Tropospheric Ultraviolet and Visible (TUV, v5.3.1) model combined with the European Centre for Medium-Range Weather Forecasts (ECMWF) products ERA5 reanalysis dataset and observed data. Moreover, the missing VOCs data were interpolated using the MLR with other hour data of the day or average the close hour data. If the data miss more than three hours in a day, the average value of the corresponding hour during this period will represent the hourly value of that day.

OBM is able to simultaneously quantify the O₃ production rate, atmospheric oxidation capacity (AOC), OH reactivity, and the primary production, recycling, and termination rates of OH–HO₂–RO₂ radicals (62). Here, we focus on AOC, O₃ budget, and OH–HO₂–RO₂ budget. Before each simulation, the model starts at 00:00 LT and pre-run eight days under constraints of input data to reach a steady-state for unmeasured species in the sampling field.

Relative incremental reactivity (RIR)

Multiple linear regression (MLR)

We next use the LMG (Lindeman, Merenda, and Gold) method to quantify the relative importance of each of the ten selected meteorological variables in explaining the atmospherically driven increase in O₃ in Shanghai over pre-lockdown and COVID-19 lockdown. The LMG method can divide the total |${R}^2$| explained by the MLR model into each non-negative individual |${R}^2$| contribution of each relevant regressor. This approach has been used in many previous studies to investigate the relative importance of model predictors in PM_2.5 concentrations (72), clouds, and changes in radiative forcing (38).

Funding

This work was supported by the National Natural Science Foundation of China (No. 42061134006, 21876029 and 22276038) and Deutsche Forschungsgemeinschaft [“Coupling and Abatement of atmospheric Ozone and PM in the Chinese Yangtse River Delta (PMO3)” under HE3086/46–1 and Project No. 448587068].

Authors' Contributions

Z. Z., X. L., J. C., and H. H. designed research; Z. Z. and J. J. performed research; Z. Z., B. L., and X. M. analyzed data; Z. Z., and X. L. wrote the paper.

Data Availability

All study data are included in the article and Supplementary Material.

Notes

Competing interest statement: The authors declare no competing interest.

References