To enhance air quality monitoring and climate change forecasting in East and Southeast Asia, the National Institute of Environmental Research (NIER) of South Korea developed the new-generation and first GEMS aboard the Geostationary (GEO) Korea Multi-Purpose Satellite 2B (GEO-KOMPSAT-2B). This satellite was successfully launched in February 2020. GEMS operates at a spectral resolution of 0.6 nm, covering a wavelength interval of 0.2 nm across the 300–500 nm range (Kim et al., 2020). It provides hourly observations across Asia, from 5° S to 45° N and 75 to 145° E, with a spatial resolution of 7×8 km. Compared with other satellites, GEMS offers significant advantages, with its superior temporal resolution allowing for eight daily observations from 00:45 to 07:45 UTC (Kim et al., 2020). The instrument primarily tracks key atmospheric constituents, including NO₂, O₃, HCHO, and aerosols. The GEMS L2 HCHO and NO₂ products are retrieved using the Differential Optical Absorption Spectroscopy (DOAS) method. First, the slant column densities (SCDs) of the target gases are obtained through DOAS fitting of the ultraviolet–visible spectra. The SCDs are then converted to tropospheric vertical column densities (VCDs) by applying air mass factors (AMFs) derived from a radiative transfer model. The main sources of uncertainty include spectral fitting errors, AMF calculation errors, and assumptions in the a priori profiles. Detailed retrieval principles and product validation can be found in Kim et al. (2020) and Lange et al. (2024). As shown in Fig. 1, GEMS covers most of China from 08:45 to 15:45 LST, with a 1 h interval, enabling hourly analyses. Fig. 1a and 1b correspond to the first and last daily overpasses, respectively, illustrating the observational time boundaries and spatial coverage underlying the diurnal analysis in this study. LST was used for all study times in this paper if not otherwise stated. The study area for GEMS eight observation periods is defined as 17–43° N, 97–127° E, with a spatial resolution of 0.07°×0.08°. Additionally, the observation window has been adjusted to 09:00 to 16:00, crucial for monitoring photochemical activities and urban emissions in major Chinese cities (BTH, YRD, PRD, SC, in Fig. 1c. GEMS level-2 tropospheric NO₂ and HCHO data from 2021 to 2023 are free to download at https://nesc.nier.go.kr/en/html/cntnts/91/static/page.do (last access: 6 July 2026).

In addition, this study utilizes tropospheric HCHO and NO₂ column data from the Sentinel-5Precursor TROPOMI for 2021–2023. The products, provided by the European Space Agency (ESA), were obtained in Level-3 format via the Google Earth Engine platform and subjected to quality control filtering (qa _value > 0.75) to remove cloud contamination and low-quality observations. Both TROPOMI and GEMS retrieve gas column densities using the DOAS method; however, they differ substantially in orbital configuration and temporal resolution. TROPOMI is a polar-orbiting satellite with approximately one overpass per day, whereas GEMS is a geostationary satellite capable of providing hourly observations. TROPOMI data have been shown to exhibit a high correlation (R= 0.9) with GEMS NO₂ measurements, confirming their reliability and accuracy in capturing atmospheric composition (Oak et al., 2024). In this study, TROPOMI data are primarily used to assist in validation and spatial distribution comparisons, while analyses of ozone formation sensitivity and its diurnal variations are conducted mainly using GEMS data.

The China National Air Quality Monitoring Network provides hourly surface O₃ measurements through an extensive network of over 1400 ground-based stations in more than 330 cities across China. This invaluable data can be accessible through the official website at https://air.cnemc.cn:18007/ (last access: 6 July 2026). Our analysis includes data from 642 stations located in 210 cities, with all O₃ concentrations recorded in micrograms per cubic meter (µg m⁻³). Many studies have used satellite-derived tropospheric NO₂ and HCHO concentrations in combination with near-surface ozone measurements to investigate O₃–NO_x–VOCs sensitivity. Accordingly, this study employs the aforementioned GEMS data and applies a k-dimensional tree (KDTree) nearest-neighbor approach to match ground stations to their nearest grid points, thereby extracting the corresponding hourly satellite observations. Additionally, this study incorporates hourly ERA5 reanalysis data, provided by the European Centre for Medium-Range Weather Forecasts (ECMWF). The ERA5 dataset includes critical meteorological variables such as 2 m temperature (T2m), relative humidity (RH), surface solar radiation (SSR), and boundary layer height (BLH). These variables provide comprehensive insights into the meteorological conditions affecting ozone formation and its diurnal variations across the study area, enhancing the understanding of O₃–NO_x-VOCs interactions.

This study employs locally estimated scatterplot smoothing (LOESS) to characterize the nonlinear relationship between O₃ and FNR. LOESS is a nonparametric regression method that fits low-order polynomials to localized subsets of data, requiring no predetermined global functional form. Specifically, the LOESS fitting at each hour from 09:00 to 16:00 LST was performed independently. Only the O₃ concentrations and corresponding FNR samples at the same hour were used for each fitting, without incorporating information from preceding or subsequent hours, thereby obtaining the functional relationship O3=f(FNR).

To identify high-response regions for ozone formation, the peak of the LOESS curve (denoted as f_max) was first determined. The FNR interval satisfying f(FNR)≥0.9× f_max was then defined as the transitional (high-response) regime. Based on the endpoints of this interval (FNR_start and FNR_end), ozone formation sensitivity was classified into three regimes: VOC-limited: FNR < FNR_start, Transitional: FNR_start ≤ FNR ≤ FNR_end, NO_x-limited: FNR > FNR_end.

To evaluate the robustness of the fitting results, a bootstrap resampling method (1000 iterations) was applied to the LOESS curves to quantify uncertainties and estimate the corresponding 95 % confidence intervals. Compared with high-order polynomial fitting, however, LOESS provides a more robust representation by reducing the influence of local fluctuations and satellite observational noise on the overall fitting results, making it more suitable for identifying the overall variation trends and high-response regions of ozone formation sensitivity at the regional scale. In addition, the ozone sensitivity classification in this study was based on high-response intervals rather than a single peak point; therefore, small variations in local peak positions are unlikely to substantially affect the classification results.

Figure 2, based on GEMS satellite data and ground monitoring observations, shows the daily variations in atmospheric pollutants across China during the warm seasons (April–September) from 2021 to 2023, including the tropospheric column concentrations of HCHO and NO₂, as well as ground-level O₃ concentrations. At 09:00, higher initial concentrations of HCHO relative to NO₂ are apparent, likely from reduced nighttime dispersion and increased morning emissions (Wang et al., 2017). The majority of HCHO is produced through photo-oxidation of VOCs, while relatively little is emitted directly (Sun et al., 2021). As the day progresses, increases in temperature and radiation intensity promote the conversion of VOCs to HCHO (Eq. 1), significantly increasing HCHO levels, especially in densely populated urban areas (Fig. 1) with larger concentrations of VOCs such as BTH, YRD, PRD, and SC. Supplement Figs. S1–S4 indicate that the BTH, YRD, PRD and SC regions generally exhibit higher T2m, SSR and BLH, except where regional climate and topography exert influence, providing favorable conditions for photochemical reactions. Statistical analysis (Table S1 in the Supplement) shows that HCHO is significantly positively correlated with T2m (ρ= 0.531, p<0.01) and SSR (ρ= 0.415, p<0.01), indicating that HCHO concentrations are influenced by photochemical processes. Previous studies have also shown that HCHO concentrations increase with enhanced photochemical activity, peaking in the afternoon when solar radiation is strongest (Wu et al., 2023). Although the reaction of HCHO with OH radicals serves as a consumption mechanism (Eq. 2), in regions with high VOC emissions, the continuous conversion of both biogenic and anthropogenic VOCs, driven by increasing temperature and radiation, may result in a net production rate of HCHO that surpasses its consumption rate, thereby leading to a sustained increase in HCHO concentrations during the afternoon period (Javed et al., 2019). Overall, HCHO concentrations increase steadily throughout the day, reaching a maximum at 16:00, approximately 1.3 times the 09:00 level. Meanwhile, NO₂ concentrations display pronounced diurnal fluctuations. Due to traffic and industrial emissions, concentrations are higher in the morning, while photochemical loss is minimal under low solar radiation (Fig. 3) (Xie et al., 2016). As solar radiation and temperature increase, enhanced photochemical activity promotes NO₂ depletion (Eqs. 3–4), leading to a gradual decrease in concentrations over the course of the day. The most rapid NO₂ reduction occurs between 13:00 and 14:00, primarily due to fast photochemical conversion coupled with strong mixing associated with the maximum boundary layer height (Fig. 3). Statistical analysis (Table S1) further indicates that NO₂ is significantly negatively correlated with T2m (ρ=-0.556, p<0.01) and BLH (ρ=-0.201, p<0.01), reflecting the combined effects of enhanced photochemical loss and dilution by boundary layer mixing. However, after 16:00, as photochemical activity diminishes and traffic increases in some cities due to school dismissals and varying work shifts, NO₂ levels rise again (Zhang and Batterman, 2009). This trend is consistent with observations by Kim et al. (2020) and Elise and Tracey (2020). NO₂ peaks in the morning and then gradually decline, reaching around 70 % of their initial level by 15:00, followed by a slight rebound in the late afternoon. Although NO₂ is also regenerated during this process (Eqs. 5–7), increased photolytic activity and the conversion of NO₂ to nitric acid (Eq. 8) typically decrease NO₂ concentrations throughout the daytime (Tan et al., 2019). The peak of O₃ occurs between 15:00 and 16:00, with concentrations gradually increasing throughout the day, starting from a morning average of 56.29 µg m⁻³ and reaching a peak of 122.32 µg m⁻³ around 15:00. Liu and Wang (2020) and Xia et al. (2021) also reported that high O₃ concentrations during the warm season peak around 15:00 in major city clusters in China. In addition, tropospheric ozone columns retrieved from the geostationary satellite GEMS exhibit a consistent diurnal pattern, with ozone gradually increasing throughout the morning and peaking in the afternoon (around 15:00) (Kim, 2023), showing good agreement in diurnal phase with surface observations.

Overall, the spatial distribution of HCHO, NO₂, and O₃ concentrations reveals significant regional differences across China. In the morning, high concentrations of HCHO and NO₂ are primarily concentrated in urban clusters, reflecting the impact of industrial activities and traffic emissions on the precursors of O₃. Meanwhile, O₃ concentrations are predominantly concentrated in the northern regions. In the afternoon, both HCHO and ozone concentrations increase significantly, with their distribution range notably expanding compared to the morning, particularly in high-population-density urban areas. However, high concentrations of NO₂ remain primarily concentrated in northern China, but due to the influence of local meteorological conditions, industrial emissions, and the involvement in O₃ formation, NO₂ concentrations gradually decrease. 1VOC+OH→RO2+other by-products (e.g., HCHO)2HCHO+OH+O2→H2O+HO2+CO3NO2+hν(UV)→NO+O4O+O2+M→O3+M5O3+NO→NO2+O26RO2+NO→NO2+RO7HO2+NO→NO2+OH8NO2+OH+M→HNO3+M9HCHO+hν(UV)→2HO2+CO

Here, VOCs react with hydroxyl radicals (OH), producing RO₂ and by-products like HCHO. RO₂ and HCHO can further react with nitrogen monoxide (NO), forming NO₂. NO₂ undergoes photolysis (hν), breaking down into NO and atomic oxygen (O), which combines with molecular oxygen (O₂) to form O₃. O₃ can react with NO to regenerate NO₂, maintaining the NO_x cycle. Meanwhile, HCHO can photolyze into carbon monoxide (CO) and more hydroperoxyl radicals, accelerating O₃ formation. NO₂ can also react with OH and a third body (M) to form nitric acid (HNO₃), reducing NO_x levels and limiting further O₃ production.

Figure 4 displays the daily relationship between HCHO and NO₂ concentrations and ground-level O₃ utilizing LOESS fitting method to highlight trends and high O₃ concentration events. The distributions of HCHO and NO₂ at 09:00 are dispersed. The concentration of O₃ is relatively low in Fig. 4a, so the photochemical reaction required for O₃ formation has not yet started completely because of the low temperature and radiation (Fig. 3). At this time, data points (NO₂ > 20 × 10¹⁵ molec cm⁻²) are concentrated in the upper left corner, suggesting a NO_x-saturated state with VOCs (represented by HCHO) as the limiting factor. In addition, a small amount of O₃ reacts with NO to form new NO₂ at this time (Eq. 5), but its contribution is smaller than the impact of direct emissions and the effect of nighttime accumulation (Romer et al., 2018). From 10:00 to 12:00, as sunlight and temperature rise, data points move towards the lower right, significantly increasing O₃ concentration. During this period, three typical states of O₃–NO_x-VOCs chemical processes can be identified: the VOC-limited regime, the transition zone, and the NO_x-limited regime. By the afternoon (Fig. 4e–h), the data points consistently cluster in regimes characterized by elevated HCHO and depleted NO₂, indicating a progressive shift in the photochemical environment. Under these conditions, O₃ concentrations concurrently reach their diurnal maximum. The photochemical reaction of VOCs is accelerated with increasing solar radiation (peak at 13:00) and temperature, generating more free radicals and intermediates (e.g., HCHO). The increase of these free radicals will further promote the photolysis of NO₂ and accelerate the generation of O₃. At the same time, the photolysis of NO₂ not only produces O₃ directly but also reduces the concentration of NO₂, reducing the O₃ consumption reaction (Yang et al., 2021). In addition, the lower relative humidity reduces the involvement of water vapor, which usually consumes O₃, and this reduction favors the accumulation of O₃ (Zhang et al., 2022). Since the reaction of VOCs and NO₂ photolysis are mutually reinforcing, O₃ production in this environment shows a nonlinear enhancement, which is particularly evident under conditions of abundant VOCs but low NO₂.

Although there are some differences in the concentration and spatial distribution of HCHO and NO₂ between the GEMS and TROPOMI satellite data (Fig. 5), they all exhibit similar distribution characteristics of high values in the four major urban agglomerations (BTH, YRD, SC, and PRD). Specifically, the areas of high concentrations of HCHO and NO₂ are the same in both data, while the HCHO / NO₂ ratio also exhibits a similar spatial pattern. This similarity is further supported by the difference values (Fig. 5i), which show relatively small discrepancies in FNR over major urban agglomerations. In addition, the quantitative comparison in Fig. S5 indicates positive correlations between GEMS and TROPOMI observations for both HCHO and NO₂, with correlation coefficients of 0.506 and 0.838, respectively. The mean biases (MB) are 0.185×1015 molec cm⁻² for HCHO and -2.509 × 10¹⁵ molec cm⁻² for NO₂. Overall, NO₂ exhibits higher consistency between the two datasets, whereas HCHO shows relatively weaker agreement. This discrepancy likely reflects known uncertainties in HCHO retrievals, such as a lower signal-to-noise ratio and a stronger dependence on a priori profiles, and is also influenced by differences in satellite overpass times (GEMS: hourly daytime observations; TROPOMI: ∼ 13:30 LST). As a ratio, FNR partially cancels systematic biases, yielding a correlation coefficient of 0.783 with an MB of 1.172, and thus exhibits strong agreement and similar overall trends between the two datasets. Previous studies have likewise reported good agreement between GEMS and TROPOMI for HCHO and NO₂, with correlation coefficients reaching 0.80 and 0.62, respectively, and overall correlations ranging from 0.59 to 0.85, with most regional differences within 20 % (Fu et al., 2024; He, 2025).

Therefore, FNR derived from both datasets can effectively capture the nonlinear relationships between ozone precursors relevant to ozone formation processes. Based on this, we further investigated the diurnal characteristics of the quantitative relationship between the probability of high-ozone events and the FNR derived from GEMS satellite observations, as shown in Fig. 4i–p. Compared to cubic polynomial fitting (Fig. 6), we found that LOESS fitting performed better in identifying peak probabilities of high O₃ events with minimal uncertainty. During the morning hours (from 09:00 to 12:00), O₃ concentrations gradually increase within the transition ranges of [0.94, 1.46] to [1.36, 2.07], with fitting curves that are steep and closely aligned to the y-axis, indicating a rapid shift from a VOC-limited to a regime significantly influenced by HCHO and NO₂. As solar radiation intensifies and emissions from traffic and industrial sources continue, O₃ gradually accumulates near the ground. Note that the maximum FNR position gradually shifts rightward, indicating the growing importance of NO₂ in O₃ formation. In the afternoon (after 12:00), the fitted regimes, with broader transition zones spanning [1.30, 2.22] to [1.36, 2.28], more accurately captured the nonlinear relationship of O₃–NO_x-VOCs (R2>0.88) than those in the morning. The afternoon typically meets peak radiation and temperatures (Fig. 3), accelerating chemical reactions, especially photochemical reactions, thus enhancing O₃ production (Coates et al., 2016). Besides, the stable emissions and atmospheric chemical processes lead to consistent O₃ sensitivity curves from 13:00 to 16:00. Throughout the day, the maximum high-probability O₃ values from GEMS LOESS fitting increase, peaking at 1.97 at 15:00, before declining at 16:00. The O₃ sensitivity thresholds at 13:00 in this study align with previous findings, such as [1.0, 2.0] by Jin and Holloway (2015), and [1.5, 2.3] by Chang et al. (2016). Furthermore, GEMS data further revealed that the hourly variability in O₃–NO_x-VOCs sensitivity throughout the day was more complex and dynamic compared to previous studies, which primarily reported diurnal changes limited to a single time period, as observed by polar-orbiting satellites.

Figure 7 illustrates the diurnal characteristics of ozone formation regimes over the study region from April to September 2023. Spatially, the region is predominantly NO_x-limited, followed by transitional regimes, with VOC-limited and transitional areas mainly concentrated in the northern part. Further analysis indicates that VOC-limited regimes are particularly pronounced in urban centers, whereas surrounding suburban and rural areas are more often classified as transitional or NO_x-limited (see Figs. S6–S7). This spatial differentiation is consistent with the findings of Ren et al. (2022) and Li et al. (2024).

In terms of diurnal variation, ozone formation regimes exhibit significant dynamic transitions. As the day progresses into the afternoon (after approximately 13:00), many urban areas that are VOC-limited in the morning gradually shift toward transitional regimes. For example, in the BTH, YRD and SC regions, urban areas predominantly maintain a VOC-limited state in the morning but clearly transition toward transitional regimes in the afternoon. In contrast, the PRD shows a different pattern, being mainly transitional in the morning and further shifting to NO_x-limited in the afternoon. This behavior may be associated with higher temperatures and humidity in the region, which facilitate the photochemical reactions between VOCs and NO_x, thereby altering ozone formation sensitivity. This hypothesis is consistent with Wang et al. (2024), who reported enhanced photochemical reaction rates in the PRD. Moreover, recent air quality policies in the PRD have significantly reduced VOC and NO_x concentrations, which may also contribute to the observed shift from VOC-limited to NO_x-limited regimes (Li et al., 2024). Figure 8 (analogous to Fig. 7 but for 2021 and 2022) further indicate a trend in recent years in the PRD toward transitional and NO_x-limited regimes in the morning. In contrast, the SC region maintains a primarily VOC-limited regime during the day, likely due to unique geographic features that restrict regional circulation and form temperature inversions, partially suppressing pollutant dispersion (Yang et al., 2020). Notably, the intensity of the VOC-limited regime in the SC region weakens in the afternoon and is not particularly pronounced in the morning, which may be influenced by solar radiation intensity and atmospheric dynamics. The above results indicate that ozone formation regimes exhibit pronounced spatial heterogeneity and diurnal variation at the regional scale. However, such large-scale patterns may obscure differences in precursor evolution and meteorological drivers among individual cities. Therefore, further analysis at the city scale is necessary to elucidate the physicochemical processes governing ozone formation.

Four representative cities – Beijing, Nanjing, Chengdu, and Guangzhou – were selected to represent the BTH, YRD, SC and PRD regions, respectively. A spatial masking was performed between the administrative boundaries of each city and the aforementioned ozone formation sensitivity regimes, thereby quantifying the fractional characteristics of the ozone formation regimes within each city. As shown in Fig. 9a–d, except for Guangzhou, the other three cities exhibit a high proportion of VOC-limited regimes during the daytime. Notably, the VOC-limited fraction in Beijing and Chengdu peaks around noon (12:00). The changes in ΔHCHO and ΔNO₂, along with their rate-of-change ratio (|ΔHCHO / ΔNO2|), serve as a robust indicator for tracking the dynamic response of relative precursor conversion intensities. With this analytical framework, we further combine the hourly meteorological conditions (Fig. 9e–h), the rates of change in ΔHCHO and ΔNO₂ (Fig. 9i–l), and the correlations between pollutants and meteorological factors in each city (Tables S2–S5) to conduct a comparative analysis of ozone formation mechanisms and their diurnal evolution.

Beijing is characterized by the coexistence of VOC-limited and transitional regimes between 09:00 and 16:00, reflecting a NO₂-abundant environment. More specifically, Beijing remains predominantly VOC-limited from 09:00 to 13:00, after which the fractions of the NO_x-limited and transitional regimes gradually increase. Notably, taking 12:00 as a turning point, the VOC-limited fraction increases hourly before noon while the other regimes decrease, whereas after 12:00 the VOC-limited fraction declines sharply and reaches a minimum at 16:00, accompanied by a concurrent increase in the other two regimes, which peak at the same time. Strong SSR, particularly between 12:00 and 13:00, together with a rapidly increasing BLH, substantially enhances NO₂ photolysis and weakens the titration of O₃ by NO. As shown in Table S2, the negative correlation coefficients of NO₂ with SSR and BLH are -0.384 and -0.219, respectively. Even so, under high-NO_x emission conditions, ozone formation is still unlikely to enter a NO_x-limited regime and is instead more strongly constrained by VOCs availability. Previous studies have shown that under NO_x-rich conditions, ozone production is more sensitive to VOCs, and that boundary layer development and enhanced radiation further amplify this sensitivity (Ren et al., 2022). Consistently, Fig. 9i shows a relatively large net loss rate of NO₂, whereas the increase in HCHO is comparatively limited, leading to low values of ΔHCHO / ΔNO₂ in the morning and further indicating VOC-limited ozone formation under a NO_x-abundant background. In the afternoon, ΔHCHO / ΔNO₂ begins to increase, suggesting a transition toward a transitional regime with emerging NO_x-limited characteristics.

O₃ formation in Nanjing exhibits a more complex diurnal evolution, characterized by a predominantly transitional regime jointly controlled by VOCs and NO_x, alongside a non-negligible contribution from the VOC-limited regime. During 09:00–11:00, the fractions of the NO_x-limited and transitional regimes decrease, whereas the VOC-limited fraction increases. Taking 13:00 as a turning point, the former two regimes show an increase followed by a decrease, while the VOC-limited regime displays the opposite pattern. Compared with Beijing, the growth rate of the BLH in Nanjing is substantially weaker, whereas T2m and RH are considerably higher. Such meteorological conditions may affect the efficiency of HO₂ / RO₂ chain reactions, rendering ozone production more nonlinearly responsive to precursor changes and thereby leading to more complex dynamic behavior (Lu et al., 2019b). The statistical results (Table S3) further show that O₃ is significantly positively correlated with both SSR and BLH (ρ= 0.507 and 0.387, respectively), whereas the correlation between NO₂ and BLH is relatively weak (ρ= -0.017), indicating that ozone formation in Nanjing is jointly regulated by photochemical activity and complex boundary layer evolution. Consistent with this interpretation, the rates of change in ΔHCHO and ΔNO₂ (Fig. 9j) increase synchronously after 12:00, resulting in a transient rise in ΔHCHO / ΔNO₂; however, this elevated ratio is not sustained. This indicates that Nanjing does not stably enter a single-precursor-limited regime but instead remains largely within a transitional state characterized by joint VOC-NO_x control, reflecting strong photochemical activity under a NO₂ -rich background.

In contrast, Chengdu is dominated by a VOC-limited regime, followed by the transitional regime, while the NO_x-limited regime accounts for the smallest fraction. Before 12:00, the fractions of the NO_x-limited and transitional regimes decrease hourly, accompanied by a continuous increase in the VOC-limited fraction. After 12:00, although the VOC-limited fraction decreases slightly, it remains substantially higher than the other two regimes. Ozone formation in Chengdu is strongly influenced by basin-induced meteorological conditions: the enclosed topography restricts atmospheric dispersion and suppresses photochemical intensity, while sustained traffic and industrial emissions maintain relatively high NO_x levels, making ozone production more strongly constrained by VOC availability (Lu et al., 2019a). As shown in Table S4, the negative correlation between HCHO and RH (ρ=-0.185) is substantially stronger than that between NO₂ and RH (ρ=-0.014), indicating that under basin conditions, high humidity exerts a more pronounced suppressive effect on HCHO accumulation, thereby further enhancing the sensitivity of ozone formation to VOCs. The rate-of-change characteristics of ΔHCHO and ΔNO₂ further support this interpretation. Compared with the other three cities, Chengdu exhibits the smallest variability in both ΔHCHO and ΔNO₂. During the morning (09:00–11:00), NO₂ continues to increase, whereas HCHO does not show a corresponding enhancement, limiting the amplification of radical chain reactions and thereby suppressing rapid ozone formation. Even in the afternoon, when NO₂ experiences sustained net loss and HCHO increases concurrently, ΔHCHO / ΔNO₂ remains persistently low, indicating that ozone formation in Chengdu is consistently dominated by a VOC-limited regime.

In Guangzhou, ozone formation is dominated by the transitional regime before 12:00, followed by the NO_x-limited and VOC-limited regimes. After 12:00, the VOC-limited regime disappears, and ozone formation gradually shifts toward a pattern dominated by the NO_x-limited regime, with the transitional regime as secondary. Among the four cities, Guangzhou exhibits the highest T2m and RH, which may enhance photochemical reaction rates and contribute to the observed shift toward NO_x-limited regimes in the afternoon. The statistical results further show (Table S5) that O₃ is significantly positively correlated with T2m (0.475) and SSR (0.578), whereas NO₂ exhibits a negative correlation with SSR (-0.180), indicating that strong photochemical conditions promote NO₂ photolysis and ozone formation. In addition, the continuously increasing anthropogenic and biogenic VOC emissions and oxidation processes in southern coastal regions maintain relatively high HCHO concentrations, which further contributes to the persistently low fraction of VOC-limited regimes (Wei et al., 2023). After 12:00, the net loss rate of NO₂ weakens rapidly, whereas HCHO maintains stable positive production, leading to a pronounced short-term enhancement in the marginal sensitivity of ozone formation to NO_x and a progressive transition toward a NO_x-limited regime. Overall, Guangzhou is characterized by an HCHO-abundant regime driven by strong biogenic and anthropogenic VOC emissions. These features are consistent with previous understandings of ozone formation mechanisms in eastern and southern Chinese cities (Wang et al., 2021).

Therefore, effective ozone pollution control should avoid single-precursor strategies and instead implement coordinated and time-dependent reductions of VOCs and NO_x tailored to different regions and meteorological conditions to achieve stable and effective ozone mitigation.