Introduction

China is suffering from ground-level ozone (O3) pollution, with the ozone level comparable to the historical levels witnessed in the United States during the 1980s1,2,3,4,5. As a major urban conurbation, the Pearl River Delta (PRD) has been populated with over 78 million residents. Two megacities, Guangzhou and Shenzhen, have over 18 and 17 million residents nowadays, respectively. The health effects of ozone pollution have caused widespread concern in the city clusters. Therefore, uncovering the causes of ozone pollution in the region holds significant benefits for the scientific community and the public. Surface ozone concentration is driven by local photochemical formation, transport, and background levels, and hinges on various factors, including meteorological and climatic impacts and chemical precursor emissions6,7,8,9. Ozone pollution presents considerable complexities for mitigation strategies due to the non-linearity formation related to its precursors10,11,12. In addition to anthropogenic emissions, meteorological conditions intricately modulate the transport processes of ozone and its precursors, regulate the rate of photochemical reactions, and steer biogenic emissions, which are also a source of ozone precursors, crucially impact ozone concentration variations9,13,14,15. Meteorological drivers have been estimated to be an important factor for ozone pollution, especially in southern China and the Yangtze River Delta16.

Considerable studies have been dedicated to examining the phenomenon of elevated ozone pollution during the warm season in the PRD, where ozone concentrations frequently surpass national standards17,18,19,20,21,22,23,24,25,26,27,28. For example, the identification of key meteorological factors, such as low humidity and specific wind patterns, has provided insights into how these conditions can exacerbate O3 pollution through increased photochemical production23,24. The influence of large-scale weather systems, including subtropical highs and typhoons, on altering these conditions and precipitating severe O3 episodes has been elucidated19,24,27,29. The examination of O3 diurnal variations has underscored the importance of meteorological processes in the temporal dynamics of O3 concentrations18. In addition, the role of biogenic volatile organic compounds (BVOCs), enhanced by extreme heat events, has been recognized as a critical factor in the formation of high O3 levels, which highlights the interplay between BVOCs and meteorological condition26,28. Overall, these findings advance our understanding of the underlying mechanisms responsible for the pronounced O3 pollution and also underscore the complexity of O3 pollution in the warm seasons in the PRD region.

Recent studies showed that ozone pollution not only occurs in the warm season but also becomes a year-round concern across China, along with a shift in the seasonal characteristics17,30,31. The North China Plain, known for its pronounced summer ozone pollution, is now experiencing ozone exacerbation, notably between February and May from 2013 to 201932. Similar patterns have emerged in Eastern China, where ozone exceedance events have been reported from March and extending as late as November30. However, the investigations about the long-term evolution and driving mechanism of spring ozone variation in the PRD region are still lacking. Here, we quantify and interpret the spring surface ozone (March to May) trend from 2013 to 2022 at 57 national monitoring sites across the PRD (Methods). We quantify meteorological contributions to the rising spring ozone levels and identify key meteorological factors, determining the ozone variations and synoptic weather types linked to ozone pollution. Given the complex and nonlinear relationship between surface ozone and meteorological variables33, we further interpret the spatial response of spring ozone to meteorological fields and prevailing atmospheric conditions.

Results

The trend of springtime surface ozone in the PRD region

From 2013 to 2022, springtime ozone concentrations across the entire PRD region have increased fast. 84% of the sites showed a notable increasing trend in MDA8 O3, with growth rates between 0.22 and 3.23 ppbv a−1 (Methods). Only 16% of the sites observed declining trends (−0.01 to −1.74 ppbv a−1), most of which had minor decreases ranging from −0.01 to −0.5 ppbv a−1 (Fig. 1a). This pattern echoes the distinctive trend observed at the Tung Chung (TC) site in Hong Kong, where an increase in springtime MDA8 O3 by 1.8 ± 0.48 ppbv a−1was noted during 2013–201734. We confirm that a similar, widespread surge in springtime ozone has been manifesting across a broader region within the PRD over an extended period. Over the past five years (2018–2022), an elevated prevalence of springtime ozone levels, marked by a significant increase in concentrations, has been observed. The median springtime MDA8 O3 concentration has increased from 32.4 ppbv during 2013–2017 to 39.8 ppbv in 2018–2022. Additionally, the peak of the MDA8 ozone probability distribution functions averaged from 57 sites, has shifted from 25.0 ppbv in the earlier period to 31.7 ppbv in the latter period (Fig. 1b). Furthermore, the steepest O3 trends have been identified in the central clusters of the PRD and their southwestern areas, whereas less pronounced or even declining O3 trends have been noted at some inland and coastal sites spatially. This distribution patterns of springtime is similar with the year-round ozone trends from 2013 to 2019, with ozone hotspots consistently expanding from the central areas toward the southwest17,35.

Fig. 1: 2013–2022 springtime MDA8 O3 variations across 57 sites in the PRD.
figure 1

a Spatial distribution of trends based on monthly MDA8 O3 anomalies following the method adopted by Cooper et al. 68. The color of the circles indicates the range of values, while the size of each circle represents the magnitude of the value within its respective range. b Probability density functions of spring daily MDA8 ozone. The thick blue and red lines represent 2013–2017 and 2018–2022 values, respectively.

Full size image

Quantitative contributions of meteorology and anthropogenic emissions

Regionally, the springtime MDA8 O3 concentrations increased by an average of 1.15 ppbv a−1 from 2013 to 2022 (Fig. 2a), exceeding the year-round annual growth rate of 1.11 ppbv a−1 at the same 57 sites from 2013 to 201917. To isolate the roles of changes in anthropogenic emissions from meteorological factors in driving ozone trends, we utilized the same methodology as Li et al. 31. Meteorologically driven factors accounted for 0.88 ppbv a−1 of the springtime O3 growth, whereas anthropogenic emissions contributed 0.27 ppbv a−1 (Methods). This emphasizes the dominant role of meteorological conditions in springtime ozone levels in the PRD, constituting 77% of the increase. In contrast, the rapid increase in ozone during the cold season in the NCP is primarily attributable to the uncoordinated anthropogenic emissions reduction31. This indicated the distinct drivers of ozone growth observed in the cold season between the PRD and the NCP.

Fig. 2: Comparison of meteorological and anthropogenical trends of monthly MDA8 O3 anomalies in springtime in 2013–2022.
figure 2

a Region-scale MDA8 O3 anomalies averaged from all available sites. b City-scale MDA8 O3 anomalies averaged from all available sites for each city. The black frame represents the observed trend. The red and blue bars indicate meteorological and anthropological contributions, respectively. c Site-scale MDA8 O3 anomalies caused by the meteorological change. d Same with (c) but for the anthropogenic driven pattern. Statistically significant trends above the 90% confidence level are marked with black dots. Trend values for each site are detailed in Supplementary Table 2.

Full size image

For the city scale, MDA8 O3 trends in nine major cities in PRD spanned from 0.38 to 2.05 ppbv a−1 (Fig. 2b, detailed in Supplementary Fig. 1). Meteorological factors, shown to be the uniform and positive contributors to the increments in springtime ozone levels, have driven increases ranging from 0.59 to 1.09 ppbv a−1 with minor differences across the nine cities. On the other hand, anthropogenic emissions exhibit a diverse impact on urban MDA8 O3 trends, spanning from −0.44 to 1.02 ppbv a−1. The most striking growth rates were exhibited in Guangzhou, Foshan, Dongguan, Zhongshan, and Jiangmen, notably with Jiangmen, at a growth rate of 2.05 ppbv a−1. Besides the meteorological change, anthropogenic emissions contributed significant in Guangzhou, Foshan, and Jiangmen. Meanwhile, coastal cities such as Shenzhen, Zhongshan, and Zhuhai see only a marginal increase. Conversely, cities including Zhaoqing, Dongguan, and Huizhou illustrate a dampening effect of anthropogenic change (between −0.44 and −0.3 ppbv a−1), with the changes in anthropogenic emissions effectively counteract the meteorological increase in ozone.

Delving further into the smaller scale, ozone trends driven by meteorological conditions or human activities are zoomed in for each monitoring site (Fig. 2c, d). The positive contributions to ozone increase, driven by meteorological factors, are evenly distributed across nearly all sites. In contrast, the contribution of anthropogenic emissions to ozone trends varies significantly across different sites, ranging from −1.56 to 2.12 ppbv a−1. Positive contributions are typically observed at central urban sites, whereas peripheral zones exhibit negative impacts, with values ranging from −1.56 to −0.08 ppbv a−1(Supplementary Table 2). The positive or negative ozone trends driven by human activities can be attributed to different photochemical mechanisms controlling ozone formation. Since 2013, extensive emission reduction policies have significantly decreased NOx emissions and slightly reduced VOC emissions within PRD, resulting in a continuous decline in the NOx/VOC ratio36,37,38,39, indicated an expansion of transitional ozone formation zones, while VOC-limited zones have been shrinking from 2013 to 201940. Supplementary Fig. 2 shows in the spring of 2019 to 2022, relatively low NO₂ and high HCHO columns, with peak levels in the industrial, densely populated central PRD, most of central PRD areas were still primarily VOC-limited ozone formation regions, while peripheral areas fell within transitional ozone production zones, this is consistent with previous studies21,35,41,42. In PRD urban centers areas, the anthropogenic ozone growth trend may be largely driven by significant NOx reductions in VOC-sensitive regions under emission policies (see Fig. 2d). This suggests prioritizing VOC reductions could better mitigate springtime ozone pollution under climate change. Meanwhile, reductions in NOx emissions in transitional zones have resulted in the ozone decrease, highlighting the effectiveness of these policies in addressing spring ozone growth.

Key meteorological factors inducive to springtime ozone enhancement

As shown in Fig. 3 and Supplementary Table 3, although the influence of meteorological changes on ozone levels is uniformly positive across the region, the key contributing factors display distinct spatial distribution characteristics. These can be categorized into coastal, inland, and transitional zones (Methods). Specifically, U10 is identified as the most critical driver in coastal areas, followed by SSR and V850. V850 is the primary factor in transitional zones, with U10 and SSR also playing significant roles. In inland regions, SSR and V850 are the dominant meteorological influences.

Fig. 3: Key meteorological factors influencing spring MDA8 O3 for each site, 2013–2022.
figure 3

a The 1st most important meteorological variable. Circles and squares represent meteorological variables positively and negatively impacting ozone growth. b Same with (a) the 2nd most important meteorological variable. c Same with (a) the 3rd most important meteorological variable.

Full size image

Supplementary Fig. 3 shows that in March, coastal areas predominantly experience easterly and southeasterly winds, while inland areas are characterized by northerly and northeasterly winds. Near the urban cluster center, these wind patterns converge. Influenced by the summer monsoon from South China Sea in April and May, the dominant surface winds shift gradually from southeasterly to southerly17,43. Higher wind speeds along the coast enhance atmospheric dispersion, reducing pollution build-up. This characteristic contrasts with the inland areas, which are more prone to pollutants accumulation despite the continually increasing wind speed from March to May. U10, the principal impact meteorological factor along the coast, means that atmospheric dispersion is the most significant determinant of ozone levels in these areas (Fig. 3a). Moreover, the negative correlation between U10 and ozone indicates that strengthening in the westerly component of U10 corresponds with decreased ozone concentrations, particularly evident in March’s more frequent easterly winds that tend to elevate coastal ozone levels compared to April-May, corroborating previous results17,43. The predominant wind at 850 hPa is southwesterly, representing the mid to upper part of the boundary layer (Supplementary Fig. 3).

V850 negatively correlates with ozone growth, indicating that increased northerly airflow leads to higher ozone levels (Supplementary Fig. 4). Notably, in the southwest of the PRD, northerly winds potentially transport pollution precursors from densely populated and industrial upwind areas, culminating in the convergence of polluted air masses35,43. Here, V850 is also identified as the leading meteorological factor, suggesting that the northerly flow within the boundary layer is a key mechanism behind the increased springtime ozone exceedances (Methods) in the southwest.

SSR, indicating the intensity of daytime photochemical reactions of ozone formation44, is the leading meteorological factor in inland areas and positively correlates with ozone growth (Fig. 3a). As shown in Supplementary Fig. 3, the PRD experiences a substantial increase in net surface solar radiation from March to May, with higher SSR in coastal areas but a more pronounced increase inland. Previous studies noted that the photochemical ozone formation capacity in the PRD inland areas might exceed that of the coast17. The positive contributions of anthropogenic emissions to ozone growth in almost all inland areas continue to be quantified and verified in this study (Supplementary Table 2), which affirms that local photochemical reactions play a decisive role in ozone variations of the inland.

Overall, U10, V850, and SSR collectively emerge as the paramount meteorological factors influencing springtime ozone concentrations in the PRD. The variability in these key factors among different sub-regions suggests diverse ozone change mechanisms, with their collective changes driving the overall springtime ozone trends. As shown in Supplementary Fig. 4, the trendency in U10 were extremely minor (0.002–0.023 m/s a−1) during spring from 2013 to 2022, in contrast to the notable decrease in V850 (−0.144 to −0.139 m/s a−1) and increase in SSR (0.175–0.183 MJ/m2 a−1). In coastal areas, secondary and tertiary meteorological factors mainly influence the increase in springtime ozone. At the same time, in inland and transitional zones, it predominantly results from the primary factor, leading to slightly lower ozone growth in coastal cities, especially Shenzhen and Zhuhai, compared to other urban areas, as indicated in Fig. 2b. Therefore, intensified photochemical reactions and increased northerly flows within the boundary layer are identified as the critical meteorological drivers of the increased springtime ozone in the PRD. A recent study points out that under the current global warming scenario, increased summer solar radiation in the PRD leads to more frequent and prolonged ozone pollution events28. Moreover, sea surface temperature anomalies in the central equatorial Pacific and the Philippine Sea were found to be contributed to the O₃-favorable meteorological conditions modulated by early South China Sea summer monsoon (SCSSM) onset, primarily due to stronger solar radiation enhancing the emission of biogenic VOCs45. This leads to increased ozone concentrations during the late spring period (May) in southern China. These insights indicate that the variability of the South China Sea monsoon and the resulting increased solar radiation, potentially exacerbated by global warming, are key factors influencing springtime ozone levels in this climate-sensitive areas.

Typical weather pattern for springtime ozone pollution

Meteorological factors, predominantly governed by synoptic patterns and reflecting local atmospheric conditions46,47,48, could be correlated with actual weather patterns to elucidate their specific roles in driving upward trends in ozone formation across sub-regions. Utilizing the Lamb Weather Types (LWTs) classification method (Methods), we determined individual daily weather types for each site from 2013 to 2022. Within the period, eight distinct weather types occurred, including Low Flow (LF), Northeasterly (NE), Easterly (E), Southeasterly (SE), Southerly (S), Southwesterly (SW), Cyclonic (C), and Anticyclonic (A) types, with six playing crucial roles in ozone-polluting episodes. As depicted in Fig. 4, LF, E, and AC types led to ozone exceedances across almost all monitoring stations in the PRD. NE type is associated with a higher likelihood of ozone exceedances in the urban center and its southwestern and coastal areas. The northwest is susceptible to occurring ozone pollution under C and SE types. Specifically, the highest ozone exceedances are associated with LF, AC, and NE weather types, with probabilities for sites with recorded exceedances ranging between 20% and 40%. The probability for sites in the Dongguan area under AC weather-type conditions and for sites in the Jiangmen area under NE weather type is higher, concentrated in the 40% to 60% range. The LF type presents the broadest spatial extent of sites experiencing ozone exceedances across the region. Additionally, the LF weather type occurs more frequently on a regional average basis, with a frequency of 52.2 days a−1, compared to 3.4 days a−1 for the AC type and 0.8 days a−1 for the NE type. Supplementary Fig. 5 demonstrates that nearly all stations across the region experienced the LF weather type far more frequently than other key weather types during spring, highlighting its widespread prevalence.

Fig. 4: Key synoptic circulations influencing spring O3 exceedances, 2013–2022.
figure 4

The conditional probability of an O3 exceedance event under the following weather types for each site: a Low flow, b Easterly, c Cyclonic, d Southeasterly, e Anticyclonic, and f Northeasterly. The color of the circles indicates the range of probability values, while the size of each circle represents the magnitude of the specific probability value within its respective range. The “a” value denotes the average annual occurrence of a specific weather type across the PRD region, calculated by averaging the annual occurrence totals from each monitoring site over multiple years. The “b” value is defined as the regional-averaged total number of days per year that stations experience ozone exceedances, categorized by each weather type. The specific calculation formulas for the “a” and “b” values are provided in the Methods section.

Full size image

As shown in Fig. 5a, the regional ozone exceedance totals for six key weather types have increased annually from 2013 to 2022. The LF weather type has been identified as the most influential in causing spring ozone exceedances in the PRD with a contribution of 57.5% on the decade average. Between 2017 and 2019, the regional ozone exceedance totals were predominantly concentrated under the LF type, ranging from 66% to 100%. During 2019–2022, the regional ozone exceedance totals for the LF weather type have shown a significant upward trend, and the ozone exceedance totals under the AC and the E weather conditions also have been increasing, nearly matching those observed under LF conditions. Furthermore, the ozone exceedance totals in 2022 reached the highest level in a decade. Based on the regional pollution patterns associated with AC, E, and LF weather conditions, it is evident there has been a continuous expansion in the geographic scope of ozone exceedances during spring.

Fig. 5: The ozone exceedance magnitude evolution under LF weather type.
figure 5

a The annual regional ozone exceedance totals for six key weather types from 2013 to 2022. The values tagged on the bars represent the percentage of regional ozone exceedance totals under Low Flow weather type out of the totals for all weather patterns each year. The annual regional ozone exceedance totals for the given weather type are detailed in the Methods section. b Trends in the annual total occurrences of the Low Flow weather pattern at each site. c Trends in the annual total ozone exceedances under the LF weather pattern at each site. Statistically significant trends above the 90% confidence level are marked with black dots.

Full size image

We should note that the ambient O₃ concentrations are also influenced by regional background O₃. A substantial increase in SSR can boost natural emissions, raising background ozone levels (Supplementary Fig. 3). Additionally, previous study indicates that during A-type pollution episodes, cross-regional transport, particularly from the Yangtze River Delta (YRD) region, significantly increases background ozone concentrations27. The strong systematic winds with A-type weather systems facilitate the long-range transport of ozone and its precursors, contributing to elevated background ozone levels in the PRD. However, the infrequent occurrence of A-type weather systems during spring limits their overall impact on long-term ozone trends compared to the most frequent and influential LF-type weather systems. The potential meteorological mechanisms driving the increased background ozone concentrations mentioned above vary across different weather systems. This may explain the differing ozone growth mechanisms between the PRD and the NCP during the cold season.

As shown in Fig. 5b, the total frequency of the LF weather type has increased across most sites in the region. This trend may be influenced by interannual climate variability, particularly ENSO events, as studies have shown that weakened southerly flows during El Niño can lead to more static atmospheric conditions, which are conducive to ozone accumulation49. Such events may partially explain the increased frequency of LF weather patterns in the PRD. Figure 5c further reveals that the total number of ozone exceedances associated with the LF weather type has steadily increased from 2013 to 2022 at nearly all monitoring sites across the region, with the trend being especially pronounced in the urban center and southwestern parts of the PRD. The observed increase in ozone exceedances under LF conditions highlights the critical role that LF weather patterns play in driving springtime ozone pollution, particularly in densely populated and industrialized areas.

Regional springtime ozone pollution mechanism in low flow weather

We further defined the springtime typical ozone pollution as Regional Low Flow Exceedance Events (RLFEs) by the following criteria: firstly, when 47 or more out of 57 stations in the Pearl River Delta are under the LF weather type, it is classified as a Regional Low Flow (RLF) weather type day; In this context, if there are five or more stations with ozone exceedances, it is defined as an RLFE event, namely a regional ozone-polluting event under the regional Low Flow weather type day (for the threshold setting of 47 and 5 stations, Supplementary Fig. 6). Between 2013 and 2022, a total of 147 RLFEs occurred during the spring season (147 out of 920 days), with these regional ozone pollution events predominantly taking place in inland areas. During RLFEs, significant increases in MDA8 O3 concentrations were observed both inland and along the coast, indicating a continuous worsening of regional ozone pollution that warrants increased attention (Supplementary Fig. 7a).

Meteorological conditions during RLFEs in the Pearl River Delta markedly differ from typical spring conditions (Fig. 6, Supplementary Fig. 8 and Fig. 9). During these events, the region experiences an increase in SSR by 3.1 MJ m⁻² and Tmax by 2.04 °C, alongside a decrease in RH by −3.51%, all of which enhance local photochemical ozone production50,51,52,53. These meteorological anomalies are particularly pronounced in inland areas, emphasizing the importance of local photochemical reactions in driving ozone growth in these regions, potentially leading to concentrated outbreaks of ozone pollution in inland areas during RLFEs (Fig. 6a–c). The dominant wind direction at 10 meters is southeast, and during RLFEs, wind speeds typically decrease (by −0.56 m s⁻¹), resulting in calm and light winds53,54 (Fig. 6d and Supplementary Fig. 9a, c). This reduction in wind speed is even more significant in coastal areas, although wind speeds there still much exceed those in inland areas. Urban centers also report a significant drop in wind speed, which further supports the accumulation of ozone due to decreased dispersion and increased local photochemical activity. At the 850hPa level, the prevailing wind direction switches to southwesterly, with a notable reduction during RLFEs (by −2.51 m s⁻¹) (Fig. 6e and Supplementary Fig. 9b, d). The combined weakening of system winds and surface wind fields leads to poorer atmospheric dispersion conditions, greatly facilitating the occurrence of ozone pollution.

Fig. 6: Meteorological anomalies during RLFEs, in contrast to the average spring values from 2013 to 2022.
figure 6

a For daytime (11:00–18:00 local time) surface net solar radiation (SSR). b For the daily maximum value of 2-m air temperature (Tmax). c For daily 24-h averaged surface air relative humidity (RH). d For daytime (11:00–18:00 local time) 10-meter wind speed. e For daytime (11:00–18:00 local time) wind speed at 850hPa.

Full size image

From 2013 to 2022, we observed a consistent increase in the proportion of RLFEs occurring during regional low flow weather type days (RLF) each spring, with an annual rise of 0.89% (Fig. 7a). However, the year 2019 exhibited an anomalous percentage of RLFE occurrences within RLF; excluding this year reveals an even sharper increase of 1.32% a−1 (p < 0.05) (Supplementary Fig. 10a). Concurrently, the number of monitoring stations recording ozone exceedances during these events has been steadily growing by 1.04% annually (Fig. 7b). Without the data from 2019, this expansion rate is more pronounced at 1.22% a−1 (p < 0.05) (Supplementary Fig. 10b). The escalating frequency of RLFEs and their widening impact likely result from evolving local dynamic and thermal conditions, particularly under stagnant atmospheric conditions.

Fig. 7: The evolutionary characteristics of the Regional Low Flow Exceedance Events (RLFEs) from 2013 to 2022.
figure 7

a The annual variation in the proportion of RLFEs relative to the total regional Low Flow weather type days (see text). b The annual variation in the proportion of sites experiencing ozone exceedances within the RLFEs out of all the 57 sites in PRD each year.

Full size image

Horizontal wind speed serves as a more significant indicator of local dynamic characteristics than turbulent diffusion on a regional scale. During RLFEs, the instances of low wind speed (LWS, less than 2 m s−1) incrementally increase from coastal to inland regions (Supplementary Fig. 11). From 2013 to 2022, there has been a steady rise in the proportion of LWS hours during springtime RLFEs, with notable increases in areas like Jiangmen, Guangzhou, and Foshan (Supplementary Fig. 12a). The annual proportion of sites experiencing Low Wind Speed Events (LWEs), as depicted in Supplementary Fig. 12b, has also demonstrated an upward trend, particularly evident in the past five years. These changes in local dynamic factors suggest an increase in atmospheric stagnation during RLFEs, leading to a continuous decline in the capacity for horizontal transport of ground-level pollutants. This scenario fosters an environment more conducive to the local photochemical production and accumulation of ozone.

During RLFEs, the diurnal variation in thermodynamic conditions is distinctly pronounced, with the most significant temperature differentials observed between 13:00 and 17:00 local time (Supplementary Fig. 13). In this period, temperatures gradually decrease from the urban centers towards the peripheral areas of the city clusters, with the coastal suburbs near the Pearl River Estuary being the cooler regions (Supplementary Fig. 14), demonstrating a pronounced Urban Heat Island (UHI) effect55,56. Previous studies have noted that under the influence of Urban Heat Island (UHI) effects, observations typically include elevated ground temperatures, decreased wind speeds, increased boundary layer heights, and intensified convergence of surface wind fields55, all of which imply enhanced photochemical pollution. In this study, the UHI Intensity was characterized by the difference in 2-meter air temperatures between the Guangzhou City Monitoring Station, which represents high-temperature zones, and the Meisha Station in Shenzhen, indicative of low-temperature areas57,58,59. The diurnal peak of the UHII index, recorded at 17:00, is a reliable indicator for effectively reflecting the regional thermal conditions (Supplementary Fig. 15a). From 2013 to 2022, the UHII within the PRD increased annually (Supplementary Fig. 15b). Such changes could potentially lead to an intensification of local photochemical ozone production, thereby possibly exacerbating the worsening of ozone pollution during RLFEs. Overall, over the past decade, changes in local dynamic and thermal factors during RLFEs have resulted in the continuous deterioration of ozone pollution and the expansion of its impact across the Pearl River Delta.

Conclusion and discussion

We show that the PRD has experienced a comprehensive increase in springtime ozone concentrations over the last decade (2013–2022). Meteorological factors are found to be the primary drivers for this increase, contributing to 77% of the regional ozone growth. Specifically, SSR, U10, and 850 V have been identified as key contributors. The LF weather type emerged as crucial in causing widespread ozone pollution in the region. During LF weather events, anomalies in meteorological conditions have led to fast growth in ozone pollution in inland areas, while changes in local thermodynamic and dynamic meteorological conditions have further exacerbated the deterioration and expansion of ozone pollution during these events. In addition, we show that the different mechanisms in driving cold-season ozone growth in the PRD compared with those in the NCP31, underscoring the significance of regional disparities in ozone pollution. We highlight that the meteorological-induced springtime ozone pollution deserves our full attention, as spring ozone pollution control in the global climate-sensitive regions, for example, the PRD region28,45, will more challenge towards climate change, this means that we need to implement more stringent emission reduction policies to avoiding the spring ozone pollution.

Methods

Data source

The PRD is in South China’s coastal region, encompassing nine cities: Guangzhou, Shenzhen, Huizhou, Dongguan, Zhaoqing, Foshan, Jiangmen, Zhongshan, and Zhuhai. The sub-regions discussed in the results section are defined as follows: Coastal areas primarily include Shenzhen and Zhuhai, adjacent to the South China Sea. Transitional zones encompass the area from southern Huizhou through the Shenzhen-Dongguan junction, extending to large parts of Zhongshan, Jiangmen, and Zhuhai, covering the southwest of the Pearl River Delta and the eastern area between the coast and inland. All remaining areas are classified as inland regions. Hourly surface ozone concentrations from March to May (MAM) from 2013 to 2022 were sourced from the China National Environmental Monitoring Center (CNEMC) network, which includes 57 national sites in the PRD. The QA/QC has been conducted and detailed in prior studies3,17,60. Meteorological fields for 2013–2022 were obtained from the ERA5 reanalysis dataset by the European Centre for Medium-Range Weather Forecasts (ECMWF)61,62. The nearest neighbor interpolation matched meteorological variables from ERA5 to the locations of the 57 monitoring sites, enabling the acquisition of corresponding meteorological data for each site. In the Multiple Linear Regression (MLR) model analyses, eight ERA5 meteorological variables were considered ozone covariates (Supplementary Table 4), in line with prior research14,32,34,63, including daily maximum 2 m air temperature (Tmax), 10 m zonal wind (U10) and meridional wind (V10), surface net solar radiation (SSR), planetary boundary layer height (PBLH), sea level pressure (SLP), relative humidity (RH), and 850 hPa meridional wind (V850). Each variable maintains an hourly temporal resolution but possesses unique spatial resolutions. Specifically, Tmax, U10, V10, and RH have a spatial resolution of 0.1°, while SLP, PBLH, and V850 have a resolution of 0.25°. Moreover, the following meteorological parameters were also utilized: hourly mean sea level pressure (MSLP) for weather type classification, and hourly vertical velocity, eastward (U) and northward (V) wind components at 850 hPa, for the wind field analysis. The data have a resolution of 0.25° (Supplementary Table 4). We have collected satellite observations of NO₂64 and HCHO65 column densities using the TROPOMI instrument, which can be accessed at https://s5phub.copernicus.eu/dhus/. TROPOMI provides daily global coverage with a pixel resolution of 5.5 × 3.5 km² and offers data from March 1 to May 31 for the years 2019 to 2022. The data from TROPOMI has been used to monitor anthropogenic emissions66,67. This study utilized TROPOMI Level 2 observations, selecting data with quality assurance values greater than 0.75 for NO₂ and greater than 0.5 for HCHO.

Trend analyses

In this study, we report trends based on monthly mean MDA8 O3 anomalies, which can be more accurate than the trend estimated from monthly observed ozone due to its strong seasonal cycle60,68. The monthly anomalies of each ozone metric are derived as the difference between the individual monthly mean values and the 2013–2022 monthly average. We calculated linear trends using the generalized least-squares regression method, with p-values calculated under a two-tailed Student’s t-test. We report trend values directly and do not treat p < 0.05 as a “bright line” to label a trend as statistically meaningful68,69. For each city, the monthly ozone metric anomalies are computed as the mean of the corresponding values across all sites within the city.

Stepwise multiple linear regression (MLR) model

The MLR model, widely used in quantifying the influence of meteorological and anthropogenic factors on ozone trends, has been robustly established and extensively harnessed in a multitude of studies spanning North America, Europe, and China32,63,70,71,72. We use a stepwise MRL model to quantify the role of meteorology in driving 2013–2022 ozone trends. The model is of the form:

$$y={\beta }_{0}+\mathop{\sum }\limits_{k=1}^{8}{\beta }_{k}{x}_{k}+{\rm{interaction\; terms}}$$

Where \(y\) is the daily MDA8 ozone concentration and (\({x}_{1}\), …, \({x}_{8}\)) are the eight meteorological variables (Supplementary Table 4). The interaction terms are up to second order. The \({\beta }_{k}\) are the regression coefficients. The regression is done stepwise for each site to add and delete terms based on Akaike Information Criterion (AIC) statistics to obtain the best model fit73,74. The coefficient of determination (R2) quantifies the fraction of variance of MDA8 O3 that can be accounted for with the MLR model. To avoid overfitting, only the three locally key meteorological variables are regressed onto monthly MDA8 O3 to fit the effect of 2013–2022 meteorological variability on ozone.

The trend in regressed ozone reflects the meteorological contribution, and the residual is then taken to reflect the presumed anthropogenic emission contribution. The effect of biogenic VOCs on ozone trends depends on meteorological and land cover drivers. Meteorological drivers, particularly temperature, have been accounted for in the MLR model63. The effect of land cover changes is expected to be negligible over the past ten years75. Similar statistical decomposition of anthropogenic and meteorological contributions to air pollutant trends based on such MLR model approach has been extensively employed by previous studies31,32,63,76,77,78. In addition, we identify the critical meteorological drivers for each city based on the occurrence frequency and proportion of the top three drivers across all sites within the city. This approach allows for a more thorough understanding of the key factors contributing to ozone levels in each city, considering the frequency and the magnitude of influence each driver has on the observed ozone variations.

Gridded lamb weather types (LWTs) classification

The objective scheme developed by Jenkinson and Collinson of the Lamb weather types (LWTs) has been widely used to classify daily atmospheric circulation for many regions in mid-latitudes, such as continental Europe, USA, China, or Russia for many different purposes46,79,80,81,82,83,84,85. Based on the automated LWTs classification method, the daily weather circulation pattern is described using the location of the center of high and low pressure that determines the direction of the geostrophic airflow86. The only input that the scheme requires is MSLP. The detailed procedure can be found in Trigo and Dacamara81. Here, we follow the novel approach proposed by Otero et al. by adapting the traditional scheme point-by-point and obtaining gridded LWTs over the PRD domain (16.5°–24.5°N and 111.0°–115.0°E) for the given day87, taking each grid point as the central point surrounded by the 16 points that would describe the synoptic situation for the given central point88. According to the scheme, 27 different weather types are obtained: eight directional types, two types that describe atmospheric rotation, 16 hybrid-combined types, and one “Unclassified” type representing weak or chaotic flow (Supplementary Table 5). The hybrid types count equally as a half occurrence to each of their major types; thus, we derive 11 weather types (Supplementary Table 5), described in Otero et al. and Jones et al. in detail82,88. The gridded LWTs classification implementation allows us to classify the time-step synoptic pattern at each grid cell to assess the impact of the key weather types on any variable during the study time89,90.

Impact of LWTs on spring ozone exceedance events

Our study averaged hourly MSLP data from the ERA5 reanalysis to calculate daily LWTs based on the predefined scheme. By aligning the nearest grid results with each monitoring site, we ensured accurate weather-type identification for sites positioned centrally. Through this approach, we could examine the manifestation of synoptic-scale circulation to the site-scale meteorological conditions, which has important implications for air quality6. Thus, the direct link between synoptic circulations and ozone exceedances was established by calculating the ozone exceedance frequencies under specific weather types at each monitoring site. This methodology enabled us to identify key synoptic patterns driving the increase in springtime ozone levels. We aim to elucidate the typical weather type contributing to spring ozone pollution and characterize the regional ozone pollution patterns associated with these weather types. The data processing, definitions and detailed calculation methods for these indices are as follows.

Ozone exceedances are defined as the number of days with the ozone concentration exceeding the Chinese grade II national air quality standard, defined as MDA8 O3 > 75 ppbv or MDA1 O3 > 93 ppbv for each monitoring site.

The “a” value value analyzed in Fig. 4 is calculated as the following formula:

$$a=\frac{1}{10}\mathop{\sum }\limits_{y=2013}^{2022}\left(\frac{1}{57}\mathop{\sum }\limits_{s=1}^{57}{T}_{{sy}}\right)$$

Let \({T}_{{sy}}\) denote the total number of occurrences of a specific weather type at monitoring site \(s\) during year \(y\). Here, the total number of monitoring sites is 57, and the number of years under study amounts to 10 years, ranging from 2013 to 2022.

The “b” value value analyzed in Fig. 4 is calculated as the following formula:

$$b=a\times \left(\frac{1}{V}\mathop{\sum }\limits_{s=1}^{V}{P}_{s}\right)$$

Here, \({P}_{s}\) is the probability of ozone exceedance at monitoring site s for the specific weather type. V is the count of valid monitoring sites for the specific weather type, where a valid site is one that has recorded at least one instance of ozone exceedance (\({P}_{s}\) > 0) during the weather type in question.

The annual regional ozone exceedance totals for a given weather type in year \(y\) (\({E}_{y}\)) analyzed in Fig. 5 are computed using the formula:

$${E}_{y}=\frac{1}{V}\mathop{\sum }\limits_{s=1}^{V}{E}_{{sy}}$$

Here, \({E}_{{sy}}\) is the count of ozone exceedance days at monitoring site s during year y, under the specific weather type. A value of 0 indicates no exceedances for that site and year. \(V\) is the number of valid monitoring sites for the specific weather type, which is the count of sites where \({E}_{{sy}}\) > 0 for at least one year in the study period.