CMAQ version 5.4 with a modified SAPRC-11 photochemical mechanism was applied with a horizontal resolution of 36 km × 36 km. The WRF version 4.2.2 was utilized to generate meteorology inputs, driven by initial and boundary conditions from National Centers for Environmental Prediction (NCEP) Final Reanalysis Data (FNL) with a high resolution of 0.25 ° × 0.25 ° (https://rda.ucar.edu/datasets/ds083.2/, last access: April 2024). Anthropogenic emissions were projected for 2021 and 2022 by adjusting the 2020 Multi-resolution Emission Inventory for China (MEIC) (http://www.meicmodel.org, last access: April 2024) with updated traffic and industrial emission data (Zheng et al., 2018) (Sect. S1 in the Supplement). Biogenic emissions including BVOC and SNO were generated using the Model of Emissions of Gases and Aerosols from Nature (MEGAN) version 2.1 (Guenther et al., 2012). Based on the MEGAN estimates, the relative proportions of isoprene, terpenes, and sesquiterpenes are highly consistent between the 2021 baseline and 2022 heatwave conditions (Fig. S1 in the Supplement). They account for 81.8 %, 15.5 %, and 2.6 % of the total BVOC emissions in 2021, and 82.8 %, 14.7 %, and 2.5 % in 2022, respectively. Overall, the simulation was acceptable for the analysis compared with the observations, and a detailed model validation is presented in Sect. 3.1.

The process analysis tools embedded in the CMAQ model were applied in the following analysis. The Integrated Process Rate (IPR) analysis was employed to quantify the contributions of key physical and chemical processes to O₃ formation and to identify those most responsible for its increase (Byun and Schere, 2006). Seven processes in the IPR contributed to O₃ formation, including horizontal diffusion (HDIF), horizontal advection (HADV), vertical diffusion (VDIF), vertical advection (ZADV), gas-phase chemistry (CHEM), dry deposition (DDEP), and cloud processes (CLDS). In this study, the sum of VDIF and ZADV was classified as vertical transport (VTRA). The sum of HADV and HDIF was classified as horizontal transport (HTRA).

The summer of 2022 was selected for its record-high temperatures, while the summer of 2021 served as a baseline representing typical meteorological conditions (Lu et al., 2023). Two base cases, BASE21 and BASE22, were configured with default settings for 2021 and 2022 (Table 1). To isolate the effects of meteorology, BVOC, and SNO emissions, three sensitivity cases MET22, BVOC22, and SNO22 were conducted. These cases simulated 2022 using 2021 meteorology (MET22), 2021 BVOC emissions (BVOC22), and 2021 SNO emissions (SNO22), respectively. These simulations allowed the total difference between the two years (BASE22 – BASE21, termed “22-21”) to be partitioned into three components: (1) the direct meteorological effect (BASE22 – MET22, “MET”), (2) the indirect effect from climate-driven BVOC emissions (BASE22 – BVOC22, “BVOC”), and (3) the indirect effect from SNO emissions (BASE22 – SNO22, “SNO”). Four major Chinese city clusters including the North China Plain (NCP), the Yangtze River Delta (YRD), the Chongqing-Sichuan region (CY), and the Pearl River Delta (PRD) were selected for detailed analysis (Fig. 1a).

To better quantify the response of secondary pollutants to meteorology and biogenic emissions, a best-fit slope method was applied. This method, previously applied to the O₃-Tmax⁡ relationship, was extended to SOA and biogenic emissions in this study (Wang et al., 2021a; Ma et al., 2019). Linear regression slope (k) and correlation coefficient (r) between secondary pollutants (MDA8 O₃ and SOA) and Tmax⁡, daily average temperature (T), BVOC and SNO emission rates were calculated, respectively. The equations are shown below (Su et al., 2012): 1k=∑i=1nxiyi-nx‾y‾∑i=1nxi2-nx‾22r=∑i=1nxi-x‾yi-y‾∑i=1nxi-x‾2∑i=1nyi-y‾2 where n is the total number of grids in the area, and x represents Tmax⁡, average temperature, BVOC and SNO emission rates, respectively. y denotes the concentrations of MDA8 O₃ and SOA. x‾ and y‾ are the average of x and y. The sensitivity of MDA8 O₃ is expressed in ppb °C⁻¹ for temperature and ppb s mol⁻¹ for emissions. The sensitivity of SOA is expressed in µg m⁻³ °C⁻¹ for temperature and µg s (m⁻³ mol⁻¹) for emissions. Because the calculated sensitivities to temperature and emissions possess incomparable units, these regression slopes were not normalized.

The simulation of meteorological conditions and pollutants was generally acceptable for the following analysis. Simulated 2 m temperature (T2), relative humidity (RH), 10 m wind speed (WS), and 10 m wind direction (WD) were compared with observations (Table S1 in the Supplement). The benchmarks were provided by Emery et al. (2017). For T2, despite the gross error (GE) being slightly higher than the benchmark (≤ 2), the mean bias (MB) values ranging from -0.2 to 0.0 K fully met the benchmark (≤ 0.5). WS was slightly overestimated, indicated by positive MB values of 0.6 to 0.8 m s⁻¹, while the GE (1.4 to 1.6) and root mean square error (RMSE, 1.8 to 2.1) well met the benchmarks (≤ 2.0). Additionally, CMAQ simulations were compared with hourly observations from China National Environmental Monitoring Centre (CNMEC). Table S2 showed the model performance statistics of O₃-1 h, PM_2.5, and NO₂. Simulated O₃ and PM_2.5 concentrations fell within the recommended standards. Specifically, although O₃-1 h was slightly overestimated, its mean normalized bias (MNB) values ranged from 0.08 to 0.15 and successfully met the ≤ ± 0.15 benchmark. Similarly, while PM_2.5 was underestimated, the corresponding MFB values ranged from -0.17 to 0.06, fully satisfying the ≤ ± 0.6 benchmark. Overall, the WRF and CMAQ model performance in this study was acceptable and comparable with previous studies, giving robust results for the subsequent process attribution analysis (Ji et al., 2024; Lyu et al., 2023).

However, a direct evaluation of SOA is currently unfeasible due to the absence of a nationwide monitoring network in China. Furthermore, previous studies reported that air quality models generally underestimate SOA concentrations because of missing precursor emissions, such as semivolatile/intermediate VOCs, and unrepresented multigenerational oxidation pathways (Zhao et al., 2022; Li et al., 2026). Consequently, the actual amplification of SOA caused by high temperatures is likely more severe than the model estimates.

The temperature in China was higher in the summer of 2022 than in 2021 based on observations (Fig. 1b and c), with an overall increase of ∼ 1 °C. In particular, the warming was most pronounced in the YRB, with the CY region experiencing an increase of over 4 °C in August. In contrast, the temperatures slightly decreased in the PRD region, consistent with previous studies (Yuan et al., 2023). Model simulations also reproduced the observed high-temperature pattern in summer 2022 (Fig. S2). Besides, these elevated temperatures were accompanied by an increase in planetary boundary layer (PBL) (∼ 300 m) and reductions in RH (-15 %) and soil moisture (SM) (-10 %), comparable with previous studies (Ding et al., 2021; García-García et al., 2023).

Correspondingly, the observed MDA8 O₃ concentration in summer 2022 was higher than that in 2021 (Fig. 1d). The most notable increase of O₃ was also found in the YRD and CY regions, with an increase of more than 10 ppb in July and August, aligned with the greatest warming (Fig. 1b and d), indicating that the occurrence of secondary pollution such as O₃ was closely linked to high temperatures (Chen et al., 2024). Along with the high temperatures, simulated biogenic emission rates also increased during summer 2022 (Fig. 2). Notably, BVOC and SNO emissions increased in the YRD and CY regions, coinciding with the areas of most pronounced warming. In July, the BVOC emission rate in the YRD increased by 3 mol s⁻¹, whereas the SNO emission rate in NCP saw the most significant increase of over 1 µmol s⁻¹ in June. To understand how high temperatures affect secondary pollution, the following analysis separated their influence into direct meteorological effects and indirect effects mediated by biogenic emissions.

Model results showed that the MET22 case (direct meteorological impacts of high temperatures) contributed the most to secondary pollution increases (Fig. 3a and b). For O₃, the MET22 case accounted for over 80 % of the total increase and exhibited a spatial pattern consistent with the BASE2022 case. The most pronounced increases were concentrated in the YRD and CY regions. For SOA, the BVOC22 case also made a substantial contribution, particularly in the CY region. In contrast, the SNO22 case contributed the least to secondary pollutant formation (mostly <3 % for SOA) and even caused a slight SOA reduction in the CY region.

IPR analysis of the total difference between summers 2022 and 2021 identified enhanced chemical production (CHEM) and horizontal transport (HTRA) were the primary processes responsible for the O₃ rise (Fig. S3). Over the YRD and CY regions, where O₃ increased markedly, both CHEM and HTRA increased by more than 1.0 ppb h⁻¹ in 2022. Furthermore, more than 75 % of the increases in these two processes were attributed to the MET22 case. High-temperature conditions were typically accompanied by intensified solar radiation, which accelerated chemical reaction rates and further enhanced the CHEM process (Tian et al., 2024). Such conditions were often associated with persistent high-pressure systems that favor stable and intensified synoptic-scale flows, subsequently promoting O₃ advection and its net influx into downwind areas (Tian et al., 2024).

Diurnal variations of secondary pollutants (O₃ and SOA) across regions exhibiting significant increases (NCP, YRD, and CY) were also examined (Figs. 4 and S4). In general, the MET22 case contributed significantly to secondary pollutants during both daytime and nighttime compared to other cases. Notably, its contribution to O₃ varied regionally. The MET22 simulation showed a clear daytime contribution in the YRD and CY, whereas its contribution in the NCP was more pronounced at night. For SOA, the MET22 contribution was nearly equal between day and night across these regions.

To understand this regional divergence in the diurnal MET22 response, the atmospheric oxidation capacity (AOC) was also discussed. In this study, daytime AOC is indicated by hydrogen oxide radical (HO_x = OH + HO₂) concentrations, while nighttime AOC was indicated by nitrate radical (NO₃) concentrations (Sillman, 1995; Zhu et al., 2021; Zhu et al., 2023; Liu et al., 2012). In general, the AOC enhancement driven by the MET22 case accounted for over 90 % of the total increase in AOC, which was closely linked to the formation of secondary pollutants (Fig. 4) (Coates et al., 2016; Xing et al., 2017). For O₃, the daytime increase in the MET22 was primarily driven by enhanced HO_x concentrations, which showed net increases of 4.5 and 2.7 ppt in the YRD and CY regions, respectively (Liu et al., 2022). While daytime chemistry dominated under high-temperature conditions, concurrent high-pressure system promoted stable atmospheric conditions at night (Nasong et al., 2025). This stability led to nocturnal surface cooling, which facilitated the formation of a temperature inversion layer. This inversion inhibited the vertical diffusion of O₃, resulting in its accumulation at ground level and a subsequent increase in NO₃ concentration. This mechanism explained the larger contribution of the MET22 case to nocturnal O₃ elevation. For SOA, both daytime formation via thermally enhanced emissions and OH oxidation and nighttime formation through NO₃/O₃ oxidation were promoted under high temperatures. Consequently, SOA contribution from the MET22 remained similar throughout the diurnal cycle.

In addition to the MET22 case, the BVOC22 case (weather-driven changes in BVOC emissions) also played an important role in O₃ elevation, particularly in the NCP region (28 %). Although the increase in BVOC emissions was small there (+0.15 mol s⁻¹), the resulting O₃ increase was greater than in other regions (Fig. 5). This could be explained by the VOC-limited regime in the NCP, which makes O₃ highly responsive to VOC emission changes (Fig. S5). The best-fit slope method quantifies this sensitivity at 4.03 ppb s mol⁻¹ in the NCP (Fig. 6), which was an 8-fold enhancement over the national mean and a 5-to-6-fold increase compared to the YRD. These findings highlight the urgent need for enhanced VOC emission regulations within the NCP.

Moreover, the impact of the BVOC22 on NCP O₃ exhibits a diurnal reversal, from negative in the daytime to positive at night (Fig. 4). During the day, an increase in BVOCs enhanced the oxidation of NO to NO₂, and further facilitated the formation of O₃. However, the concurrently increased NO from NO₂ photolysis also intensified the O₃ loss via titration (O₃ + NO). When the titration process became slightly dominant, this could lead to a net negative daytime contribution to O₃. Conversely, at night, the suppressed PBL height limits vertical mixing and traps pollutants (Fig. S6). Within this stable layer, increased BVOCs promoted NO₃ radical chemistry, which is verified by the net decrease of nighttime NO₃ concentrations (Fig. 4b). This process consumed NO and thus reduced the nocturnal O₃ titration, facilitating residual O₃ accumulation.

These estimated impacts of biogenic emissions in the NCP likely represent a conservative estimate. Specifically, coarse vegetation data and simplified fertilizer parameterizations lead the MEGAN model to underestimate urban BVOCs and SNO emissions, respectively (Zhang et al., 2025; Oikawa et al., 2015; Huang et al., 2023). As the NCP features dense urban clusters interspersed with intensive agricultural lands, the actual emissions of both precursors and their subsequent contributions to secondary pollution during the heatwave are expected to exceed the current simulations.

Based on best-slope analysis, O₃ in urban areas was more sensitive to temperature changes (including both Tmax⁡ and T) compared to the national average level (+50 % to 290 %, Fig. 6). Especially in the YRD and PRD, the sensitivity of MDA8 O₃ to temperature reached 0.66 and 0.95 ppb °C⁻¹, respectively. This was partly due to the urban heat island (UHI) effect, which elevated daytime temperatures and favored pollutant accumulation. This effect could be further intensified during heatwaves (Clinton and Gong, 2013; Delcloo et al., 2018; Wang et al., 2018).

Notably, despite a slight decrease in summer O₃ and temperature in the PRD in 2022, the region exhibited the highest O₃-temperature sensitivity, which heightened local risks and necessitated targeted mitigation. As defined in Sect. 2.4, this positive slope represents the spatial linear regression across PRD grids rather than interannual temperature changes. This spatial variability heightens local risks and necessitates targeted mitigation. However, the correlation between O₃ and temperature was higher in the YRD than in the PRD (Table S3). This was because precursor emissions in the PRD were generally lower and affected by multiple factors such as sea-land breezes, which may weaken the temperature-O₃ linkage. NCP O₃ showed remarkable sensitivity to emission changes. During heatwaves, both BVOCs and SNO emissions increased. This may accelerate radical-initiated oxidation cycles and enhance OH regeneration, resulting in amplified O₃.

For SOA, the sensitivity to temperature showed pronounced regional variability, with the highest values found in the YRD and CY regions (0.28 to 0.30 µg m⁻³ °C⁻¹, Fig. 6), roughly twice the national average. This heightened sensitivity in the CY region is primarily driven by weather-driven biogenic emissions (the BVOC22 scenario), which contribute more significantly to SOA formation than the direct meteorological impact (MET22) (Fig. 5). In the extensively vegetated CY region, high temperatures markedly increase the emissions of BVOC, such as isoprene and monoterpenes (Cao et al., 2022; Guenther et al., 2012). Simultaneously, the enhanced AOC during heatwaves (as discussed in Sect. 3.4) rapidly oxidizes these precursors into low-volatility products, facilitating SOA yields. Alongside BVOCs, SOA is also highly sensitive to SNO emissions in the YRD and CY areas (0.63 and 0.58 µg m⁻³ °C⁻¹, respectively, Fig. 6). However, the SNO22 scenario reports that reducing SNO emissions only leads to a slight decrease in overall SOA concentrations (Fig. 5). Total NO_x emissions in these regions are still dominated by anthropogenic sources (Huang et al., 2023). Although heat-induced SNO provides additional NO_x to promote BVOC oxidation, fluctuations in SNO emissions have a limited impact on the total SOA concentration.

In contrast, temperature and SNO have minimal influence on SOA in the PRD. The limited contribution of heat induced SNO is largely constrained by low precursor emissions resulting from strict regional controls (Ou et al., 2016). Furthermore, elevated temperatures increase the saturation vapor pressure of semi-volatile organic compounds, driving their thermodynamic partitioning back into the gas phase and thereby suppressing SOA yields (Donahue et al., 2006). In the NCP, SOA is highly sensitive to BVOC emissions. Driven by high NO_x emissions and AOC alongside lower overall vegetation coverage, additional biogenic precursors in the NCP are rapidly oxidized into SOA, leading to this pronounced sensitivity.