The locations of the monitoring sites, the YRD urban boundaries, and the NO2 gradient are shown in Fig. 1.

The observational analysis uses surface air-quality measurements from Shanghai and Lishui as two contrasting environments within the YRD. Shanghai represents a densely urbanised megacity environment with strong anthropogenic precursor influence, while Lishui provides a lower-NOx environment with stronger influence from surrounding vegetated areas. These two cases are used to examine how O3-temperature sensitivity differs across contrasting precursor and land-surface settings.

To place the selected monitoring sites within the broader YRD context, AQMS locations were compared with the 2018 Global Urban Boundary (GUB) dataset (Li et al., 2020). The GUB dataset delineates physical urban extents at 30 m resolution using Global Artificial Impervious Area (GAIA) data. The 2018 boundaries were used to identify whether individual AQMS sites were located within or outside mapped urban areas.

The YRD-wide comparison included 250 valid AQMS sites in JJA 2022. Validation defined by hourly data in June–August (JJA) 2022 with a threshold of 75 %. The selected Shanghai sites are located within GUB urban settlement polygons and occupy the high-NO2 part of the YRD AQMS distribution, with a mean NO2 concentration of 20.58 µgm-3, corresponding to approximately the 87th percentile among valid YRD sites.

The selected Lishui sites occupy a lower-NO2 part of the YRD AQMS distribution, with a mean NO2 concentration of 11.46 µgm-3, corresponding to approximately the 18th percentile among valid YRD sites. Two of the three selected Lishui sites fall within GUB urban settlement polygons, indicating populated urban/suburban receptor locations within a broader lower-NOx and more biogenically influenced regional setting.

As far as population density and land-use type are concerned, Shanghai is a megacity with a population density of 3900 km⁻² (Shanghai Municipal Statistics Bureau, 2024), and construction land covered 42.4 % of its total area (Gong et al., 2022). Lishui is an area where 81.7 % of the land is covered with forest (Tang et al., 2022). The population density of Lishui is 140 km⁻², and the urban district accounts for about 8.6 % of its total area (Lishui Municipal Statistics Bureau, 2025).

We collected ground-level hourly O3 observation data from the China National Environmental Monitoring Center (CNEMC). The AQMS are in the busiest areas of Shanghai (30.67–31.88° N, 120.85–122.2° E) and in the populated centre area of Lishui (27.44–28.78° N, 118.70–120.41° E).

The analysis focuses on JJA 2022, when the YRD experienced severe and persistent high-temperature conditions. This period provides a well-observed recent heatwave case with routine surface O3 measurements from the national monitoring network. Earlier major heatwave summers in the region, such as 2003 and 2013, predate the availability of constant O3 observations (see https://quotsoft.net/air/, last access: 14 July 2025). The 2022 data are therefore used to examine O3-temperature sensitivity under a heat extreme setting.

We assessed O3 and meteorological data for June–August 2022 (JJA 2022). Eight AQMS and two Meteorological Monitoring Sites (MMS) in Shanghai, and three AQMSs and one MMS in Lishui met this criterion (spatial distribution see Fig. 1). For each retained site, we extracted hourly surface O3 and then formed two aggregated series: an urban mean (average across the 8 Shanghai sites at each hour) and a rural mean (average across the 3 Lishui sites at each hour). Meteorological observations originate from the China Meteorological Monitoring System (CMMS) stations; we accessed the archived data via the NOAA National Climatic Data Center (NCDC).

The meteorological and air-quality monitoring sites retained for the analysis are summarised in Table 1 below.

The AQMSs in Lishui (1267A–1269A; Table 1) are located in a populated environment that is best characterised as urban or suburban. The Lishui emission inputs used in the box model (explained later in Sect. 2.2), however, represent regional mean conditions with relatively low NOx and stronger biogenic influence. The resulting chemical regime therefore characterises the regional photochemical environment influencing Lishui, rather than the immediate atmosphere at the monitoring sites.

In this study, a heatwave is defined as a period of at least 3 consecutive days during which the daily maximum temperature exceeds the 90th percentile of the reference period (2000–2019), as used by Zhao et al. (2019). The 90th percentile threshold has been widely used in studies examining heatwave patterns, health impacts, and interactions with air quality (Engdaw et al., 2022; Fischer and Schär, 2010; Sun et al., 2020). To remove day-to-day variation, a 15 d smoothing window centred on the investigated day was applied to threshold calculation (based on the reference period) (García-Martínez and Bollasina, 2021). The years 2020 and 2021 were excluded from the reference period calculation to eliminate any effects resulting from the COVID-19 lockdowns in China.

For the ozone-temperature analysis, we use daily maximum surface O3 (O_3max) and daily maximum air temperature (Tmax) because climate-penalty assessments are conventionally based on daily peak metrics (e.g. daily max 1 h O3 versus daily Tmax), which target the photochemical period, reduce confounding nocturnal NO-titration and inter-site timing offsets (Nolte et al., 2021; Otero et al., 2021). We stratify by environment type (urban Shanghai vs. rural Lishui) and by period (all JJA days vs. heatwave days), thus yielding four subsets of ozone-temperature analysis: urban-all, urban-HW, rural-all, and rural-HW.

The XJH site (Shanghai; 31.12° N, 121.26° E) reports hourly air temperature and dew-point temperature. The BS site (Shanghai; 31.24° N, 121.27° E) and the LS site (Lishui; 28.27° N, 119.55° E) provide three-hourly records of air temperature and dew-point temperature. Daily maximum air temperature values at sites XJH and BS were averaged to obtain the representative Shanghai temperature conditions. To evaluate the joint influence of surface air temperature and humidity on ground-level O3, we aligned the temporal resolution by applying a 3 h mean to the hourly O3 series, ensuring that O3, temperature, and humidity share the same 3-hourly meteorological sampling times.

To separate the effects of temperature, humidity, diurnal temperature variation, and temperature-dependent isoprene emissions, we defined 16 sensitivity configurations, summarised in Table 3.

Each of the 16 configurations was repeated at seven target temperatures from 25 to 40 °C at 2.5 °C increments, while all other settings were held according to the experiment definition. These follow the naming convention <location>_<RH>_<process>:

Location. SH = Shanghai, urban background; LS = Lishui, rural background.

Two binary switches modify the model processes: D and ft. D (diurnal temperature enabled) applies a realistic diurnal cycle to the target-temperature (daily maximum) day; when D is off, air temperature is held constant at the target value, an extreme condition in which daily minimum values equal daily maximum values. The temperature-dependent isoprene emission factor ft enables the temperature response of isoprene.

The temperature response of isoprene followed the equations below: 1isoprene emissions=local summer average emissions⋅ft The ft is the isoprene temperature dependence factor (Pacifico et al., 2011): ft=e0.1⋅T-30when e0.1⋅T-30<2.3(T in °C)ft=2.3when e0.1⋅T-30≥2.3(T in °C) Where T is temperature in centigrade.

In these experiments, isoprene emissions increase with temperature until they reach their plateau at 38.3 °C (as shown in Fig. 3); when ft is off, isoprene emissions are temperature-independent.

The reference conditions were set at 30 °C and 70 % RH (mass mixing ratio of water (q) = 0.0184 kg kg⁻¹), representing the average environmental conditions in the 2022 JJA summer in the YRD based on observed values. Unless stated otherwise, pressure and boundary-layer height are held at the baseline values (Pressure = 1000 hPa, PBLH = 590 m), which are the averaged YRD summer conditions based on observations and ECMWF ERA5 reanalyses in 2022. Photolysis varies diurnally with the solar zenith angle at the site's latitude/longitude, so it shows a sinusoidal variation peaking at midday and being zero at night. The clouds were ignored, and all simulations are under clear-sky conditions. The location settings determine initial species levels.

The aim of UKCA Box sensitivity experiments is to quantify the local chemical and emission-related response to prescribed temperature and humidity conditions. The results should be interpreted as contributions within this controlled chemical framework, rather than as a full decomposition of all meteorological controls on the observed O3-temperature relationship.

The observed relationships between daily maximum O3 and daily maximum temperature in Shanghai and Lishui are shown in Fig. 4.

Across both data subsets (JJA and HW days), and at both locations (Shanghai and Lishui), O_3max increases with Tmax, resulting in positive correlations. Averaged over eight Shanghai sites, the summertime O3 fractional change per unit temperature change is 7.27 µgm-3°C-1 with a 95 % confidence interval (CI) [5.13, 9.41] and p<0.01. The corresponding mean sensitivity across three Lishui sites is 3.36 µgm-3°C-1 (95 % CI [2.30, 4.41], p<0.01).

This urban-rural contrast is consistent with differences in the chemical regime. As reported by Gu et al. (2020), Shanghai exhibits a strong VOC-limited regime in O3 formation, where temperature increases enhance BVOC emissions while NOx remains abundant, leading to a relatively strong ozone-temperature relationship. By contrast, the rural areas in YRD generally show higher VOC abundance, and the effect of temperature is less pronounced than in VOC-limited areas because BVOC changes here have a negligible impact (Jin and Holloway, 2015). Those differences in chemical regimes are also supported by model simulation results in this study (Sect. 3.2; Fig. 6). Thus, the doubled O3 sensitivity to temperature in Shanghai compared to Lishui is consistent with stronger NOx availability and generally higher pollution levels in Shanghai, and may relate to the capacity for PAN thermal decomposition in the YRD (detailed PAN analysis see Fig. 10) (Steiner et al., 2010).

Regressions during HWs indicate a different behaviour at the two locations. In Lishui, the heatwave-period O_3max–Tmax sensitivity is 4.05 µgm-3°C-1 (95 % CI [0.203, 7.899], p<0.05), indicating a significant positive association between O_3max and Tmax during HW periods. By contrast, in Shanghai, the Ordinary Least Squares (OLS) slope estimated from heatwave days is 3.49 µgm-3°C-1 with 95 % CI [-1.464, 8.441] (p=0.16, n=41), so a statistically significant linear association is not supported. The range of 95 % CI includes zero, indicating substantial uncertainty in the direction and magnitude of ozone's temperature response during the 2022 summer heatwaves in Shanghai. Including more historical heatwave data may help diagnose whether this loss of significance is due to insufficient data or suggest that additional heatwave co-factors (e.g., stagnation, humidity, precursor variability, or chemical regime shifts) could modulate the O3-T linkage in Shanghai.

To examine the joint influence of temperature and humidity, Fig. 5 shows the observed daytime O3 distribution across T-RH space in Shanghai and Lishui.

In both regions, RH decreases systematically with warming. In Shanghai, the typical RH spread within a given temperature bin is about 20 % (e.g. 45 %–65 % at 35 °C), with larger variability at lower T and progressively narrower ranges at higher T (Fig. 5c). Lishui shows a similar tendency, with most samples concentrated between 32–37 °C and 35 %–55 % RH (Fig. 5d). Across the observed T-RH space, O3 increases with temperature and decreases with RH, indicating a preference for warm, dry conditions.

Linear regression of RH on T yields slopes of -2.429 % °C⁻¹ for Shanghai and -3.191 % °C⁻¹ for Lishui (p<0.01 in both cases). The weaker slope in Shanghai, together with its narrower T range, is consistent with the moderating influence of sea-land thermal and moisture exchanges on coastal boundary-layer conditions, whereas the steeper slope in Lishui reflects a greater propensity for hot-dry excursions in a forested inland setting. The RH-T co-variation is much tighter in Shanghai (R2=0.83) than in Lishui (R2=0.57): in Shanghai the highest-density region (Fig. 5c) and the strongest O3 increases (Fig. 5a) lie close to the regression line, implying that the linear T-RH relationship provides a good description of typical warm-season daytime conditions and of the pathway along which O3 responds to combined warming and drying. In Lishui, by contrast, the main density maximum lies above the regression line and does not coincide with the dominant O3-increase pathway, indicating that the fitted slope is influenced by a smaller number of hot, anomalously dry episodes. This behaviour is consistent with previous studies reporting that Lishui RH conditions are dominated by strong evapotranspiration from vegetation and soils, which is negatively correlated with temperature over the YRD in summer (Wang et al., 2023). In addition, stomatal closure under low-humidity conditions, which always corresponds to high-temperature conditions, would reduce O3 dry deposition and further enhance O3 levels (Kavassalis and Murphy, 2017), contributing to the high-O3 tail at the hot, dry end of the Lishui distribution.

In Fig. 5c and d, we compare relative humidity from the VRH model configuration used in the subsequent simulations – defined by a conserved absolute humidity equal to the regional JJA 2022 mean (q=0.0184 kg kg⁻¹) – with the observed T-RH distribution. The VRH curve follows the locus of highest observational density in both Shanghai and Lishui, indicating that this configuration provides a realistic representation of the humidity-temperature conditions sampled in 2022 and is therefore suitable as the baseline humidity scenario for the box-model experiments.

Overall, Fig. 5 demonstrates that hot, dry conditions are favourable for elevated O3 at both sites, but that the joint T-RH-O3 variability differs between the coastal megacity and the forest-surrounded inland region: Shanghai exhibits a stronger, more coherent O3-T-RH relationship in summer 2022 than Lishui, reflecting differences in local meteorology, surface characteristics, precursor emissions and chemical regime.

To interpret these observed O3-temperature-humidity relationships in terms of process contributions and mechanisms, we next use an observation-constrained box model to quantify the extent to which the observed high-temperature O3 response can be explained by changes in local chemical kinetics, precursor abundances, temperature diurnal cycles, and humidity conditions. Furthermore, the mechanisms underlying process contributions will be explored through budget-level simulations.

This section evaluates whether the UKCA Box model provides chemically plausible Shanghai-like and Lishui-like environments for the subsequent temperature-sensitivity experiments. The purpose is not to perform a formal model evaluation against observations or to use UKESM as an observational benchmark. Instead, we use three complementary diagnostics: (i) local O3 responses to perturbations in NOx and VOC emissions, (ii) contextual comparison of selected short-lived chemical species with available observations, and (iii) comparison of Ox production and loss pathways between the box model and the same-chemistry UKESM-AMIP simulation.

To establish the precursor-sensitivity regimes underlying the subsequent O3-temperature analysis, Fig. 6 examines the O3 response to NOx and VOC changes using O3-NOx-isoprene response surfaces at 30 °C and ±10 % precursor perturbations across 25 to 40 °C.

Figure 6 summarises the box-model sensitivity experiments used to diagnose precursor regimes in Shanghai (Fig. 6a and c) and Lishui (Fig. 6b and d). The baseline experiments, where there are no other processes but chemical reaction rates and pathways changing with temperature (marked as Ori), are consistent with the observed conditions in the 28.75–31.25 °C temperature bin. In Shanghai, the daily-mean O3 is 63.7 µgm-3 in the model versus 69.5 µgm-3 in the observations. In Lishui, the corresponding values are 35.8 µgm-3 (model) and 52.5 µgm-3 (observations), a low bias that is consistent with Lishui being NOx-limited (see below) and suggests that the grid-cell-average NOx levels in the box configuration are lower than those sampled by the monitoring site.

Panels (a) and (b) show the simulated O3 response surfaces in NOx-isoprene concentration space at 30 °C for Shanghai and Lishui, respectively. The Shanghai baseline lies within a VOC-limited regime, where O3 increases with isoprene and decreases with increasing NOx. In contrast, the Lishui baseline lies within a NOx-limited regime, where O3 responds positively to NOx and only weakly to additional isoprene.

The local ±10 % perturbations show consistent precursor responses across the investigated temperature range. In Shanghai (Fig. 6c), O3 decreases with higher NOx emissions and increases with higher VOC emissions from 25 to 40 °C, indicating a VOC-limited regime (Pusede et al., 2015). This VOC-limited behaviour is consistent with the high-NOx, anthropogenically dominated environment of Shanghai and agrees with Liu et al. (2021b)'s study on the effectiveness of emission control strategies in China's major urban areas. In contrast, Lishui (Fig. 6d) exhibits the opposite tendency: O3 increases with NOx enhancement and is only weakly affected by VOC perturbations, signifying a NOx-limited regime (Utembe et al., 2018). This behaviour is characteristic of rural or suburban photochemical environments influenced by regional BVOC emissions and relatively low NOx levels (Sillman, 1999).

The signs of these responses remain unchanged between 25 and 40 °C, indicating no regime reversal within the tested perturbation range. These experiments therefore establish contrasting VOC-limited and NOx-limited baseline environments for interpreting the subsequent O3-temperature and Ox budget analyses.

The simulated PAN, OH, HO2, and C5H8 concentrations are compared with the available observational references in Table 4.

The model values in Table 4 are from baseline experiments (Ori) at 30 °C. The simulated PAN and C5H8 concentrations are consistent in order of magnitude with the reported observations. Direct summertime OH and HO2 measurements are not available for urban Shanghai, so the winter Shanghai measurements provide the closest direct urban reference. Same-site measurements in other cities show that daytime OH is approximately 2–5 times higher in summer than in winter, while the summer-to-winter difference in HO2 ranges from little seasonal change to a factor of approximately 5 (Heard et al., 2004; Kanaya et al., 2007; Ren et al., 2006). The Shanghai observations therefore provide a well-defined urban reference for the magnitude of HOx, while the Taizhou measurements provide the corresponding warm-season YRD comparison.

The corresponding O_x production and loss pathways in the UKCA Box model and UKESM-AMIP are compared in Fig. 7.

Figure 7 compares the fractional Ox production and loss pathways in the UKCA Box model with those simulated by UKESM-AMIP using the same UKCA chemical mechanism. The UKESM-AMIP simulation is used as a three-dimensional context for interpreting the box-model Ox budget, because it includes physical processes that are absent from the 0-D box configuration, such as transport, boundary-layer mixing, convection, and meteorological variability.

The Box Model reproduces the rank order of production pathways in the nudged UKESM-AMIP model. In Shanghai, the total Ox production flux is 1.62×10-9 kgm-2s-1 in the Box model versus 2.03×10-9 kgm-2s-1 in the UKESM model. And in Lishui, this Ox production flux is 4.10×10-10 kgm-2s-1 in the Box model, while this value is 8.40×10-10 kgm-2s-1 in the UKESM.

In both cities, peroxy-NO reactions (HO2+NO and RO2/CH3O2+NO) dominate the chemical production fraction in the box model, matching the pattern in UKESM and consistent with summertime O3 photochemistry described by Ma et al. (2022b). This agreement in fractional chemistry indicates that the box model captures the regime (NOx-VOC-OH coupling) rather than simply tuning totals. Chemical loss through O(1D)+H2O→2OH and radical-ozone reactions (HO2+O3, OH+O3) is represented with similar sign and relative importance in both frameworks.

The UKCA Box simulation shows a larger fractional contribution from dry deposition to the Ox sink (Fig. 7a–b). In the 3-D UKESM simulations, this sink is explicitly divided into deposition and dynamical export (dynamical export refers to physical transport and mixing, including horizontal advection, vertical entrainment with the growing boundary layer, shallow/deep convection, and ventilating outflow; not included in Fig. 7). In contrast, the box model treats the air parcel as a closed, well-mixed volume with no transport, so all non-chemical removal must appear as a lower-boundary deposition term. According to the monthly mean O3 levels from June to August (UKESM) and the trend of daily mean O3 levels (UKCA box model), both UKESM and UKCA Box produce nearly constant surface O3, indicating that the net Ox tendency is close to zero. In UKESM, this large residual (“UKESM sum”) reflects the contribution of dynamical export. In the box model, the production-loss balance is achieved by having chemical production intentionally offset by the parameterised dry-deposition term, leaving only a small net production or loss. The larger deposition fraction in UKCA Box should therefore be interpreted as representing the combined effect of true surface uptake and the export processes, which are explicitly resolved in UKESM.

Overall, Fig. 6, Table 4, and Fig. 7 support the use of the UKCA Box model as a controlled chemical-process diagnostic. The model reproduces the expected contrast between a VOC-sensitive Shanghai-like environment and a NOx-sensitive Lishui-like environment, and its dominant Ox production and loss pathways are broadly consistent with the same-chemistry UKESM simulation, while the differences illustrate the impact of absent physical drivers in the Box model. At the same time, the comparison does not constitute a formal validation of absolute O3 concentrations, especially for Lishui, where monitoring-site observations and the regional box-model environment represent different spatial scales. The following sections therefore interpret the box-model experiments primarily in terms of relative process sensitivity and chemical-regime behaviour, rather than as direct simulations of monitored O3 concentrations.

The observed and simulated O_3max–Tmax relationships are compared in Fig. 8.

Figure 8 compares observed JJA O_3max–Tmax relationships in the VRH set of experiments with the UKCA Box Model: Ori (reference – fixed temperature and isoprene emissions across all hours of day), D (diurnal temperature cycle only), ft (temperature-dependent isoprene only), and Dft (both). The main purpose of these sensitivity experiments is to evaluate the impact of those processes on O3's temperature responses. In the box model simulations, the VRH experiments exhibit a trend in RH-temperature that resembles reality; thus, these experiments are considered reasonably good evaluators of the box model's performance when compared with observations.

Shanghai (urban; Fig. 8a). Above 30 °C, the ft and Dft simulations exhibit significant coordination in both ozone concentrations and trend, indicating that the temperature dependence of isoprene emissions explains most of the observed sensitivity in this regime. Relative to Ori and D, both ft and Dft lift O3 by about 20 to 40 µgm-3 at 32.5–37.5 °C and reproduce the monotonic increase with temperature. At T < 30 °C, observations align more closely with Ori/D, whereas ft/Dft are low-biased. Two factors likely contribute: (i) wider uncertainty at the cool end (grey 95 % CI band) owing to fewer low-temperature days in summer 2022, and (ii) meteorological controls (e.g., mixing depth, advection, cloudiness) that covary with temperature but are not explicitly represented in the box model, which isolates chemical responses. At the upper tail, the simulated slope weakens because the applied isoprene function saturates at ∼38.3 °C, reducing the Dft (and ft) response from 9.3 µgm-3°C-1 between 35–37.5 °C to 2.1 µgm-3°C-1 between 37.5–40 °C; the observations maintain a steeper gradient (≈ 12 µgm-3°C-1). The different isoprene emission threshold might be the cause, since suppression of isoprene emissions was not observed in the 10–40 °C range in Shanghai (Gu et al., 2022). And additional heat-linked processes (e.g., boundary-layer dynamics, anthropogenic VOCs, or NOx variability) may continue to amplify O3 at extreme temperatures.

Lishui (rural; Fig. 8b). The simulations reproduce the shape of the observed O3-T curve, but the simulated O3 is systematically low-biased by about 50 µgm-3. This offset is consistent with site representativeness (i.e. the observations exhibit a populated area in Lishui where NOx emissions are higher than the box model simulations, which cover the whole Lishui region, most of which are rural areas). Importantly, introducing temperature-dependent isoprene (ft/Dft) yields only minor changes relative to Ori/D, indicating that Lishui operates largely in a NOx-limited regime where additional VOC reactivity produces only modest O3 increases, i.e. including T-dependence for isoprene leads to higher VOC levels, but this doesn't increase O3 in Lishui, whereas it does in Shanghai. The small influence of the D option confirms that, after binning by Tmax, imposing a diurnal temperature cycle alone does not substantially alter daily O_3max in this case.

Overall, the VRH experiments show that the UKCA Box captures the observed temperature sensitivity of O3 when the isoprene-temperature coupling is represented (Shanghai), while the weak ft/Dft effect at Lishui is consistent with NOx limitation. Discrepancies at low temperatures and in the highest bins highlight the role of meteorology and extreme-heat processes that lie outside the box configuration, motivating targeted sensitivity tests (e.g., boundary-layer depth, deposition, NOx variability) in future work.

The resulting changes in O_3max between 30 and 40 °C are quantified in Table 5.

Table 5 presents the increase in O_3max from 30 to 40 °C for the observations and the box model experiments. Because the experiments differ only in the process switched on, we quantify each process's contribution to the O3 warming response by comparing the simulated 30 to 40 °C change. For each experiment X, we define ΔO3X = O3X(40 °C) - O3X (30 °C), and similarly, ΔO3obs from the observations. All metrics below are expressed as fractions of ΔO3obs and therefore dimensionless. The net chemical response is fchem=ΔO3VRH_Ori/ΔO3obs, i.e., the fraction of the observed 30 to 40 °C increase that can be explained by temperature-dependent gas-phase chemistry alone under VRH conditions. The additional effect of each process is then quantified as the incremental fraction of ΔO3obs gained when that process is switched on: for the diurnal cycle, fdiurnal=(ΔO3VRH_D-ΔO3VRH_Ori)/ΔO3obs; for the isoprene temperature dependence, fiso=(ΔO3VRH_ft-ΔO3VRH_Ori)/ΔO3obs; and for humidity, fhumid=(ΔO3FRH-ΔO3VRH)/ΔO3obs. Positive values indicate that the process amplifies the increase in O3 with temperature, whereas negative values indicate damping. This normalisation allows us to directly compare the extent to which the observed O3-temperature sensitivity can be attributed to each process at the two sites. Notably, due to the nature of the 0-D box-model (no transport, clouds, or physical feedbacks), this attribution should be interpreted as under a controlled chemical framework instead of a full decomposition of all meteorological controls.

In Shanghai, decomposition of the Ox budget shows that the net chemical term accounts for 18.7 % of the observed 30 to 40 °C increase, while the diurnal temperature cycle alone has a negligible effect on the net chemistry. By contrast, the temperature dependence of isoprene emissions explains 69.1 % of the observed O3 rise, identifying this pathway as the dominant driver of the warm-season O3-T slope in urban Shanghai. When isoprene's temperature sensitivity is enabled, the diurnal cycle yields an additional 14.3 % O3 increment, consistent with enhanced photolysis and radical propagation during warmer, sunnier hours reported in YRD (Xia et al., 2022).

Humidity acts as an important modulator of the O3-temperature response. In our experiments, increasing RH by about 30 % at 40 °C (FRH: q=0.0320 kg kg⁻¹, RH = 70 %; VRH: q=0.0184 kg kg⁻¹, RH = 40.2 %), while keeping the temperature diurnal cycle and isoprene temperature dependence switched on (VRH_Dft settings), leads to a 21.9 µgm-3 reduction in O_3max at 40 °C. Expressed as a fraction of the observed 30 to 40 °C change, this contribution is (ΔO3FRH_Dft-ΔO3VRH_Dft)/ΔO3obs. This humidity-driven decrease reflects the role of water vapour in enhancing Ox loss via O(1D)+H2O pathway: in the FRH case, the vertically integrated O(1D)+H2O flux (kg m⁻² s⁻¹) increases by about 42 %, strengthening the O3 decomposition pathway and offsetting part of the thermally driven O3 enhancement.

In the NOx-limited regime (i.e. Lishui), the process attribution gives different results. The chemical response itself explains 80.4 % of the observed 30 → 40 °C change, implying that enhanced photochemistry at higher temperatures – through faster reaction rates and stronger actinic flux – dominates the O3 increase (Archibald et al., 2020b). The isoprene temperature dependence contributes only 10.4 %, consistent with a NOx-limited environment where additional VOC reactivity produces comparatively small changes in O3 production (Steiner et al., 2006). The diurnal temperature cycle makes only a minor contribution to the 30 → 40 °C O3 increase, whereas the humidity perturbation is non-negligible. When RH is raised from 40.2 % to 70 % at 40 °C (VRH: q=0.0184 kg kg⁻¹; FRH: q=0.0320 kg kg⁻¹) under the VRH_Dft settings, the resulting change in humidity accounts for 9.1 µgm-3 of O_3max reduction. Ox budget diagnostics show that this damping arises from two linked effects of higher water vapour: a reduction in Ox production via the HO2+NO channel (≈15.3 % reduction in FRH scenarios compared with that in VRH scenarios), and an enhancement of Ox loss via O3 photolysis followed by O(1D)+H2O (≈36.9 % compared with VRH scenarios). Both mechanisms are consistent with Li et al. (2021), who showed that elevated relative humidity suppresses O3 generation by inhibiting the OH/HO2-NO2 chain and promotes O3 loss through the O(¹D) + H₂O pathway.

Taken together, Fig. 8 and Table 5 demonstrate that the UKCA Box model produces a physically interpretable decomposition of the O3-temperature relationship: isoprene-temperature coupling is decisive in Shanghai, while baseline chemical partitioning and humidity dominate in Lishui. These site-specific attributions align with the contrasting chemical regimes and underscore the need to represent both biogenic VOC responses and meteorologically linked processes when projecting O3 extremes under warming.

The temperature dependence of the O_x budget is examined through its diurnal evolution in Fig. 9 and the corresponding daily integrated production and loss terms in Table 6.

Figure 9, Tables 6 and 7 summarise how Ox production and loss terms from the UKCA Box Model respond to temperature. In Shanghai (Fig. 9a–d; Table 6), all major production pathways increase with temperature, except for organic-nitrate photolysis (labelled ORG NITR PHOT in Table 6), which has a minor impact on total production. From Fig. 9a–d, the reaction HO2+NO→OH+NO2 is the dominant and most temperature-sensitive source of Ox production. Although its marginal sensitivity (change in Ox production per °C) weakens from 25 → 40 °C, it remains the single largest contributor. The second key source, RO2+NO→RO+NO2, displays the opposite behaviour – its temperature sensitivity strengthens with warming. From 35 to 40 °C, the contribution of RO2+NO to total Ox production increases by 207.7 t Ox d⁻¹ in Shanghai, nearly doubling the increase from 25 to 30 °C, underscoring the growing role of peroxy-radical chemistry at high temperatures. This growth is consistent with high-temperature partitioning of the OH-HO2-RO2 cycle – most notably the HO2+NO channel – and with the well-documented increase in effective VOC reactivity under hot conditions due to both faster kinetics and temperature-sensitive VOC emissions (e.g. isoprene) (Meng et al., 2023). Even with reduced temperature sensitivity, the HO2+NO reaction accounted for more than 50.1 % of total chemical production, confirming the importance of the HO2-NO pathway in the urban regime.

For chemical loss in Shanghai, O(¹D) + H₂O → 2OH is the leading term. Its temperature sensitivity is 55.01 t Ox d⁻¹ °C⁻¹ between 25–35 °C, decreasing to ∼36.74 t Ox d⁻¹ °C⁻¹ from 35–40 °C, consistent with partial saturation of midday photolysis at the hottest conditions. In the total budget, however, deposition dominates loss. O3 dry deposition exhibits the stronger temperature sensitivity, which maximises between 25–35 °C, whereas NO_y deposition varies less with temperature but remains the largest absolute sink at all temperatures, controlling the sign of the net Ox tendency. As deposition in the UKCA Box Model serves both as deposition and as physical loss, the magnitude comparison is not so useful.

In Lishui (Fig. 9e–h; Table 7), both production and loss magnitudes are substantially smaller than in Shanghai, consistent with lower precursor loadings in rural southwestern Zhejiang and the prevalence of NOx-limited chemistry outside urban cores in the YRD (An et al., 2021; Ren et al., 2022). HO2+NO is still the largest single production term, but at T>30 °C its strength becomes comparable to CH3O2+NO and RO2+NO, indicating a more distributed peroxy-NO contribution in the regional background. For chemical loss, O(¹D) + H₂O dominates from 25–30 °C. At temperatures higher than 30 °C, O3 + ALKENE reactions' contribution grows rapidly with temperature and becomes comparable in magnitude to O(¹D) + H₂O, highlighting the role of thermally enhanced VOC-ozone chemistry outside the urban core where the O3+C5H8 reaction plays a crucial role. In the depositional budget, O3 dry deposition is relatively more important in Lishui than in Shanghai, consistent with the much larger vegetated land fraction – which elevates deposition velocity of O3 via stomatal and surface uptake – and with weaker in-situ NO_y production under rural low-NOx conditions (Clifton et al., 2020a).

Overall, the Box model yields a coherent, process-level explanation for the observed O3-temperature behaviour. In Shanghai, warming amplifies Ox primarily through HO2+NO chemistry while deposition – especially NO_y deposition – limits the net build-up. In Lishui, the warming response is shared more evenly among peroxy-NO sources, and O3 deposition plays a larger role in the sink, with O3-alkene loss emerging at the hottest conditions.

To identify the role of PAN-mediated NOx recycling, Fig. 10 shows the temperature dependence of PAN concentration and thermal decomposition at both sites.

Figure 10 contrasts the diurnal evolution of peroxyacetyl nitrate (PAN) concentrations (dashed curves, right axis) with the thermal decomposition flux PAN → CH₃C(O)O2 (PA) + NO2 (solid curves, left axis) at four imposed temperatures in Shanghai (panel a) and Lishui (panel b). In both locations, the decomposition flux increases monotonically with temperature and peaks around local noon-early afternoon, with daytime values far exceeding nighttime values. This behaviour is consistent with the strong temperature dependence of the PAN stability (PAN thermally decomposes rapidly at higher temperatures) and with daytime heating that accelerates the decomposition reaction (Fischer et al., 2014).

Despite the broadly similar flux patterns, concentration responses differ by site. In Lishui (Fig. 10c–d), PAN concentrations decrease with warming across the full range. As the temperature rises, thermal loss outpaces daytime production (due to low ambient NOx), so the reservoir is progressively depleted. The result is a clean negative PAN-T relationship.

In Shanghai, the response is non-monotonic. In Fig. 10b, from 25 to 35 °C, the PAN concentration increases, even as the thermal decomposition of PAN accelerates as shown in Fig. 10a. This implies that, over this range, enhanced PAN production – via larger peroxyacetyl radical supply and abundant NO2 – offsets the faster loss, allowing concentrations to rise. However, at 40 °C, the pattern reverses after midday: the decomposition flux jumps sharply relative to 35 °C, while the PAN concentration drops in the afternoon. Two explanations, which are not mutually exclusive, are consistent with the curves. First, PAN's thermal instability means the PA + NO2⇌ PAN equilibrium shifts strongly toward PAN dissociation at high temperature (Cleary et al., 2007); thermal loss is the primary PAN sink in the lower troposphere and accelerates through the warmest hours, producing the observed afternoon drawdown in Fig. 10a. Second, extreme heat coincides with lower afternoon PAN production efficiency, e.g. due to the reduction in biogenic precursor supply, e.g., C5H8 as an indirect source of PA encountered its threshold at 38.3 °C, or enhanced radical competition (Liu et al., 2021a; Ma et al., 2022a; Phillips et al., 2013), so low peroxyacetyl radical (PA) levels limit the reaction between PA + NO2 to form PAN. This PAN concentration drop may imply a threshold of PAN thermal decomposition above 40 °C in Shanghai, which occurs above 39 °C in California as described by Steiner et al. (2010).

A further notable contrast is the lower temperature behaviour (25 °C). Both the thermal decomposition flux and the PAN concentration are lower in Shanghai than in Lishui at this temperature, indicating that, under cooler conditions, PAN acts as a less effective NOx reservoir near the urban site.

Overall, Fig. 10 shows that warming consistently strengthens PAN thermal decomposition at both sites, but the net concentration response is site-specific: monotonic decreases in Lishui, and increases followed by depletion in Shanghai at the highest temperature. These distinctions have practical implications: during heat extremes, PAN-mediated transport of NOx is likely reduced because rapid afternoon decomposition limits PAN lifetime in VOC-limited urban areas like Shanghai, and the effect can be less significant in Lishui.

The associated temperature responses of HOx cycling and the HO2+NO pathway are shown in Fig. 11.

Figure 11 presents the diurnal evolution of the HO2+NO production flux (solid orange), the aggregate HOx-termination (“sink”) flux (green dashed; e.g. OH+NO2+M,HO2+RO2/HO2), and the HOx concentration (blue dotted; right axis) at four imposed temperatures. As shown in Fig. 9, the HO2 and NO reaction is the dominant O3 chemical production pathway in both environments. HOx sink reactions terminate O3 formation, while HOx is widely used as a proxy for the atmosphere's oxidising capacity, with OH controlling daytime oxidative processing (Feng et al., 2021); moreover, net ozone formation scales with the peroxy-NO reactions, so higher HOx – especially HO2 – correlates with stronger photochemical O3 production (Archibald et al., 2020b; Pusede et al., 2015).

In both panels, chemical activity is tightly bound to daylight: fluxes are negligible at night, rise sharply after sunrise, peak near local noon-early afternoon, and collapse after sunset, reflecting the photochemical control of radical production.

In Shanghai (Fig. 11a–b), warming mainly strengthens the HOx-driven Ox production. With increasing temperature, the daytime HO2+NO production flux (solid curves) increases strongly, and the peak HOx concentrations (blue dotted curves, right axis) also increase, indicating more efficient radical recycling and a faster O3 production cycle at higher T. The HOx termination flux (green dashed curves) also grows with temperature, but the enhancement remains much smaller than the HO2+NO pathway throughout the day, so the temperature increase shows a more substantial impact on including more HOx in Ox formation instead of facilitating the HOx sink pathway. At T<30 °C, both HOx levels and the sink are low, implying that primary radical production is weak and the O3 production cycle can be slow.

In Lishui (Fig. 11c–d), the behaviour is significantly different. Across the temperature range, the HOx sink plays a much larger role, and at T≥30 °C, its magnitude exceeds the HO2+NO flux. Interpreted in the context of the O3 chemical cycle, this means a larger fraction of the available HOx is channelled into termination rather than propagation. Even so, HOx concentrations remain high from 25 to 40 °C, indicating efficient radical production; the high sink simply prevents those radicals from translating into proportionally larger HO2+NO rates.

Together, the panels show that warming enhances radical activity in both environments, but the partitioning between propagation and termination is site dependent: Shanghai exhibits production-dominated HOx cycling that favours O3 formation at high temperature, whereas Lishui exhibits sink-dominated cycling, consistent with background conditions where O3 production is constrained by NOx rather than HOx availability.

The present UKCA Box configuration represents only a limited subset of emitted AVOCs. In particular, some alkanes, alkenes, and aromatic compounds associated with temperature-dependent evaporative emissions are not available as explicit emitted tracers in the current chemical mechanism. Incorporating the full range of temperature-dependent AVOCs would therefore require an extension of both the chemical mechanism and the emission mapping, and these species could not be included directly in the present box-model experiments.

To assess the potential importance of temperature-dependent AVOC emissions, we reproduced the temperature-dependent emission-scaling framework of Qin et al. (2025) using the JJA 2022 CEDS emissions for Shanghai. The calculation included selected AVOC categories associated with volatile chemical products (VCPs), including industrial and residential solvent use, and gasoline evaporation. Both temperature functions were normalised to 30 °C, which represents the average baseline condition used in the box-model experiments. From 30 to 40 °C, the selected VCP-related emission flux increased by 58.6 %, from 4.83×10-10 to 7.66×10-10 kgm-2s-1, while the gasoline-evaporation flux increased by 33.3 %, from 4.17×10-11 to 5.56×10-11. The combined selected temperature-dependent AVOC flux consequently increased from 5.25×10-10 to 8.22×10-10 kgm-2s-1, corresponding to an additional 56.6 % increase.

We further examined whether this estimated AVOC enhancement could be represented explicitly in the UKCA Box experiments by applying the CEDS-to-StratTrop lumping approach of Archer-Nicholls et al. (2021). Comparison of the relevant CEDS species and sectors with the emitted StratTrop AVOC tracers showed that the overlap is very limited. On the source-species mass basis used in the mapping, these species account for only 3.56 % of the selected gasoline-evaporation emissions, while none of the selected VCP-related solvent emissions is represented. Most of the temperature-dependent butanes, pentanes, higher alkanes, other alkenes and aromatics are therefore absent from the emitted-tracer mapping.

Qin et al. (2025) found that temperature-driven AVOC emission enhancements increased MDA8 O3 by 4.6 ppb and peak O3 by 7.5 ppb during a Shanghai heatwave. Their results indicate that an AVOC enhancement of the magnitude estimated here can produce a measurable additional O3 increase under hot urban conditions. Temperature-dependent AVOCs can therefore increase anthropogenic VOC availability under hot urban conditions, and their resulting O3 contribution will be examined using an expanded VOC mechanism in future work.

Because temperature-dependent isoprene emissions strongly amplify simulated O3 in Shanghai, we examined how this effect depends on the emission scheme. We compared the JULES iBVOC formulation used in the main experiments with the temperature activity factor implemented in MEGAN v2.1. The iBVOC scheme used with JULES represents the short-term temperature response as fT=min⁡[exp⁡(0.1(T-303.15)),2.3]. The response therefore increases exponentially above the 30 °C reference condition until reaching a fixed upper limit of 2.3 at approximately 38.3 °C, after which the emission factor remains constant (Pacifico et al., 2011). By contrast, the MEGAN temperature activity factor accounts for both the instantaneous temperature and thermal acclimation through the mean temperatures over the preceding 24 and 240 h, T24 and T240. These prior-temperature terms determine the magnitude and temperature of the emission maximum rather than imposing a fixed threshold (Guenther et al., 2012).

We examined the influence of these formulations in Shanghai by normalising both temperature factors to the same isoprene emission at 30 °C and applying them over the 30–40 °C experimental range. The JULES temperature factor increased from 1.0 at 30 °C to its maximum value of 2.3 above 38.3 °C. The MEGAN factor increased continuously throughout the investigated range, reaching 2.68 at 40 °C; under the T24 and T240 conditions (30 °C; assumed from JJA YRD conditions and model setting) applied here, its emission maximum therefore occurred above 40 °C and outside the experimental range. The corresponding isoprene flux at 40 °C was 2.92×10-10 kgm-2s-1 using JULES and 3.40×10-10 kgm-2s-1 using MEGAN, compared with the common value of 1.27×10-10 kgm-2s-1 at 30 °C. These values correspond to increases of 130.0 % and 167.9 %, respectively.

The differences in calculated isoprene emissions propagated directly into the simulated evolution of daily maximum O3. At 30 °C, both schemes produced the same daily maximum O3 of 116.1 µgm-3. At 40 °C, daily maximum O3 reached 199.3 µgm-3 with JULES and 212.3 µgm-3 with MEGAN. The resulting 30–40 °C increases were therefore 83.3 and 96.2 µgm-3, respectively. Differences between the schemes remained relatively small between 32.5 and 37.5 °C, but became more pronounced once the JULES temperature factor approached its fixed upper limit. At 40 °C, MEGAN produced 12.9 µgm-3 more O3 than JULES. Thus, both schemes confirm a large isoprene-driven amplification of urban O3 at high temperature, while the greater high-temperature response in MEGAN indicates that the fixed JULES threshold provides a more conservative estimate near and above 40 °C.