The NO and CO gases were injected into the system accompanied by N2 as a balancing gas. To ensure gas purity, the NO (100 ppmv in N2) was passed through a ferrous sulfate heptahydrate (FeSO4⋅7H2O) filter to remove trace NO2 impurities. The flow rate of this NO mixture was regulated at 120 mLmin-1, achieving a final concentration of 8 ppmv in the reactor. Similarly, high-purity CO gas (99.99 %) was passed through an activated carbon column to eliminate residual carbonyl compounds before entering the reactor at 120 mLmin-1, corresponding to a constant concentration of 8 % (v/v) during amplification mode. The total system flow was maintained at 1500 mLmin-1 using mass flow controllers, corresponding to a residence time of approximately 3 s in the reaction zone.

To enable automated switching between amplification and background modes in the single-channel configuration, an automatic valve-switching module driven by a programmable timer was developed. The flow paths of CO and N2 were controlled using four two-way solenoid valves, with two valves assigned to each gas line. During the amplification mode, the timer directs CO to the front inlet to initiate the chain reaction with NO, while N2 is routed to the rear inlet. Conversely, in the background mode, the valve states are toggled so that N2 is introduced at the front and CO is directed to the rear. By synchronizing the switching of these four valves, the system achieves stable and automated mode transitions for real-time detection of peroxy radicals.

Humidity plays a critical role in determining the efficiency of chemical amplification reactions. To control water vapor levels, a Nafion tube dryer (model MD-700-12-F-3, total length 40 cm, effective length 30 cm, ID 1.78 cm) was used to remove water vapor and serve as the chemical amplification reaction chamber in the meantime. To ensure efficient drying and maintain a low-humidity sample stream, a continuous flow of dry purge gas was introduced along the outer side of the Nafion tube. In this work, the dry purge gas was generated by a pump-driven system connected in series with a drying tube filled with indicating silica gel. The gas was first deeply dried through the silica gel column to ensure extremely low moisture content before being circulated from the lower inlet to the upper outlet of the Nafion tube, forming a closed-loop flow. A humidity sensor was installed at the outlet of the sample stream to continuously monitor the relative humidity inside the reaction tube. Through this precise humidity control, the relative humidity during the chemical amplification process was maintained below 20 %, ensuring system stability and improving the reliability of the experimental results.

As described in the principle of the chemical amplification method (Eq. 1), the CL is a key parameter determining the accuracy of peroxy radical measurements. In this work, CL was determined using on-site peroxy radical calibration sources based on water vapor photolysis. A pen-type Hg lamp was used to photolyze H2O in an N2 carrier gas (mixture of dry and humidified N2), producing H atoms and OH radicals (Reaction R5). Immediately downstream of the lamp, a mixture of synthetic air and reagent gases (CO/N2 or CH4/N2) was introduced to allow rapid mixing with the photolysis products and reduce radical wall losses. In this configuration, H atoms react with O2 in the synthetic air to form HO2 (Reaction R6), while the reagent gases control the radical composition. When CO is used as the reagent gas, OH radicals are quantitatively converted to HO2 (Reaction R4), providing a HO2 calibration source without interference from OH and RO2 radicals. When CH4 (150 ppm in N2) is used instead of CO, OH reacts with CH4 to form CH3O2 (Reaction R7), yielding a mixed peroxy radical source containing approximately 50 % HO2 and 50 % CH3O2 under typical calibration conditions. In all cases, the relative humidity in the calibration flow was adjusted to ∼10 %, yielding a stable peroxy radical concentration of ∼1–3 ppbv. R5H2O+hv→OH+HR6H+O2+M→HO2+MR7OH+CH4+O2+M→CH3O2+H2O+M

The CL was determined using four distinct NO2 signals detected by the CEAS. For the pure HO2 source, the Hg lamp was switched on and CO was introduced immediately behind the H2O photolysis zone, with NO at the inlet of Nafion reactor. Under these conditions, both H atoms and OH from H2O photolysis were promptly converted to HO2, then titrated by NO to form NO2. The resulting NO2 signal is denoted as S1 (amplification mode). In the background mode, the lamp remained on while CO was substituted by N2. The corresponding NO2 signal, S2, represents the background mode when HO2 is only produced by H+O2 pathway. With the lamp off, signal S3 and S4 correspond to the amplification and background signals in the absence of radicals, respectively. Thus, (S2–S4) represents the HO2 only produced by H+O2 pathway, whereas (S1–S3) corresponds to the chain-amplified NO2 signal from HO2 produced via both the H+O2 and OH+CO pathways. The CL is then obtained from these four signals using Eq. (2). The mixed HO2–CH3O2 standard was generated by introducing CH4 only during the S1 (amplification) measurement and was used to verify the instrument response to organic peroxy radicals relative to HO2 under otherwise identical conditions. 2CL=ΔNO2[HO2]=S1-S32(S2-S4)

The CL is influenced by several factors, including the concentrations of reactive gases, reaction time, relative humidity, and wall losses (Liu and Zhang, 2014). To investigate the response of CL to these parameters in our PERCA-CEAS system, we measured the CL under a series of controlled conditions using a pure HO2 radical standard source, specifically by varying NO concentrations and RH levels. The CL measurement was repeated at least five times for each condition. For the subsequent experiments, the CO concentration was fixed at 8 %, which is sufficiently high to sustain efficient HO2 production while remaining within safe limits of CO (explosion range: 12.5 %–74.2 %).

To assess how NO concentration influences CL, we conducted a series of experiments under varying NO levels while keeping the relative humidity fixed at 10 %. As shown in Fig. 2a, CL increases steadily with increasing NO concentration and reaches its maximum when NO is 8 ppm. Considering both measurement efficiency and system economy, 8 ppm was chosen as the optimal NO concentration for instrument operation.

At high RH, water vapor on the reactor walls enhances the heterogeneous uptake of RO2 and HO2 and promotes HOx-terminating reactions, thereby suppressing chain propagation and shortening the CL (Mihele et al., 1999; Yang et al., 2019). To quantify the effect of RH on the CL, a custom-built humidifier was used to control the flow rate of the humidified gas, thereby generating HO2 sources at different RH levels (10 %–50 %). The relative humidity of the sample gas was monitored by a temperature-humidity sensor installed downstream of the Nafion reaction tube. The experimental results, presented in Fig. 2b, demonstrate a substantial decrease in CL as the RH increases. Specifically, when the relative humidity (RH) of the sample gas passing through the Nafion dryer reaches 50 %, the chain length drops by more than 50.1 %, a reduction that severely compromises the sensitivity required for accurate field RO2* measurements. Consequently, to ensure optimal measurement performance, the RH of the airflow entering the amplification zone is strictly controlled to remain below 20 % during field campaigns, corresponding to 45 % RH in ambient air.

The optimal CL for the PERCA-CEAS measurement system was determined to be 30 ± 2 under the conditions of RH=10 %, NO=8 ppm, and CO=8 % based on the pure HO2 calibration. Calibration using a mixed radical source consisting of 50 % HO2 and 50 % CH3O2 yielded a CL of 24 ± 1 under the same conditions. As summarized in Table 1, the obtained CL is lower than values reported in most previous PERCA studies, which were typically above 50 and occasionally exceeded 200 (George et al., 2020; Wood and Charest, 2014). The differences in CL are primarily attributed to wall loss of HO2 radicals due to reactor material, reactor design and working pressure. Although the obtained CL is relatively low compared with these PERCA instruments, the results demonstrate that the developed PERCA-CEAS system provides stable and reliable sensitivity for atmospheric RO2* measurements.

In the PERCA–CEAS system, the concentration of RO2* is determined by converting it into measurable NO2 through chain reactions with excess NO. Consequently, the overall detection limit of the system depends on both the detection limit of NO2 and the CL, as described by Eq. (3). The LOD for the CEAS-NO2 detector was determined under laboratory conditions by continuously measuring high-purity zero-air for over 5 h and calculating the standard deviation of blank noise (Zheng et al., 2024). At a 3 min time resolution, the LOD of NO2 was determined to be 11.4 pptv, corresponding to a detection limit of 0.38 pptv (1σ, 3 min) for RO2*, based on a CL of 30 derived from the pure HO2 calibration.

The overall measurement uncertainty of the system was quantified using the Gaussian error propagation (Eq. 4). Four major sources contribute to the total uncertainty: (1) NO2 measurement uncertainty (6 %), which includes uncertainties in the absorption cross-section (4 %), mirror reflectivity calibration (5 %), effective cavity length (0.5 %), and pressure measurement (0.1 %). (2) CL calibration uncertainty (6.7 %), derived from repeated calibration results (30 ± 2). (3) Uncertainty in radical partitioning. This arises from variability in the relative contributions of HO2 and CH3O2 to total peroxy radicals. Based on the reported HO2/RO2 (0.8–1.2) and CH3O2/RO2 (0.85–0.95) ratios from Stone et al. (2010), the mole fraction of CH3O2 in total peroxy radicals (f) is estimated to be 0.41–0.54. Under these conditions, the use of an HO2-based calibration would introduce a systematic bias of 16 %–22 % in the derived RO2* concentration. (4) PAN interference. The thermal decomposition of PAN represents a potential source of interference, as it releases NO2 and peroxy radicals that can trigger additional chemical amplification within the reactor (Liu and Zhang, 2014; Wood and Charest, 2014). Kinetic simulations using the F0AM model (Framework for 0-D Modeling; Wolfe et al., 2016) indicate that 1.0 ppbv PAN results in an equivalent RO2* signal of 1.95 pptv under our experimental conditions (26.5 °C and an amplification reaction time of 0.15 s). This accounts for 1.3 %–9.8 % of the typical RO2* levels (20–150 pptv) observed during the campaign. Based on these factors, the overall uncertainty in ROx measurement was estimated to be 18.4 %–25.7 %, demonstrating the robustness and reliability of the PERCA–CEAS system for RO2* measurement. 3LOD(RO2∗)=LOD(NO2)CL4σRO2∗[RO2∗]=σNO2[NO2]2+σCLCL2+σPartPart2+σPANPAN2

The instrument has been deployed in a field campaign to evaluate the performance of instrument under real atmospheric conditions. All instruments are placed on the roof of a six-story building on the campus of Sun Yat-sen University (22°21′06′′ N, 113°35′42′′ E), which is located on the western shore of the Pearl River Estuary and approximately 15 km from downtown Zhuhai. The observation site mainly received an airmass mixed with vehicle emissions from a traffic artery (∼300 m) and nearby residential areas. There are no major industrial facilities in the surroundings.

The field observation campaign was conducted from 26 October to 15 November 2024, producing more than 15 d of valid continuous RO2* radical data. For PERCA-CEAS operation, both the chemical amplification mode and the background mode were set to 90 s in this observation, resulting in a time resolution of 3 min. Ambient temperature and relative humidity were continuously monitored using a temperature-humidity probe (HC2A-SH, ROTRONIC, Switzerland). The concentrations of NO and NO2 were measured by a commercial chemiluminescence analyzer (Model 42i-TL, Thermo, USA), while O3 concentrations were obtained using a UV photometric analyzer (Model 49i, Thermo, USA). The mirror reflectivity of the CEAS detector was calibrated before and after the campaign, and the results remained consistent.

Figure 3a–c presents the time series of meteorological parameters and major pollutants during the observation period. Ambient temperature was consistently above 25 °C, with an average value of 28.44 ± 1.5 °C. Due to the coastal meteorological conditions, the RH was relatively high and showed clear diurnal variations – lower during the day and higher at night – ranging from 25 % to 85 %, with an average of 55.98 % ± 12.53 %. Low NO and NO2 concentrations were observed with an average of 0.34 ± 0.81 ppbv and 13.21 ± 4.72 ppbv, respectively, indicating generally weak NOx emissions from traffic at this site. The daily maximum of O3 concentrations ranged from 54 to 116 ppbv, while the overall mean concentration was 44.86 ± 25.03 ppbv, suggesting a moderate level of photochemical pollution during the observation period.

Compared to previous observations in PRD region, moderate levels of RO2* were observed with the daily peaks ranging from 48 to 87 pptv (Fig. 3d). The RO2* exhibited a pronounced daytime peak, with an average of 51.3 pptv at around 13:00 LT, and decreased to around 20 pptv at night (Fig. 4a). The relatively high level of RO2* at night indicates the presence of a nighttime peroxy radical source possibly due to NO3 chemistry oxidation. We found that the average NO3 production rate (P(NO3)) at night during the campaign was approximately 1.4 ppbvh-1, higher than the reported warm-season mean of 1.07 ± 0.38 ppbvh-1 over China (Wang et al., 2023a), indicating active nighttime NO3 chemistry at this site. The daytime average of RO2* (31.11 pptv) was 1–3 times higher than those reported in most suburban areas, while it was lower than those observed in urban areas (Tan et al., 2019; Wang et al., 2023b).

In addition, our observed RO2* exhibited nonlinear response to NO variation (Fig. 4b). At low NO levels, RO2* increases with NO by promoting the RO2*–HOx propagation cycle. Once NO exceeds about 1 ppbv, the excess NO rapidly converts peroxy radicals into termination products (e.g., organic nitrates), thus interrupting the radical chain cycle and leading to a decline in RO2* concentrations (Thornton et al., 2002). The turning point of NO response is higher than that reported in previous studies (Ma et al., 2022; Wei et al., 2023; Yang et al., 2022). This behavior is consistent with the elevated VOC reactivity inferred for studies conducted near our observation site. Studies at nearby coastal sites (e.g., Da Wan Shan Island) have reported high oxidative capacity associated with strong VOC consumption and enhanced radical cycling (Sun et al., 2024), while the dense vegetation in the Pearl River Delta contributes substantial biogenic VOC emissions (e.g., isoprene) (Situ et al., 2013; Wang et al., 2023c). Collectively, these factors contribute to sustaining propagation cycles of RO2 radical and shift the turning point of the NO dependence of RO2* concentration higher at our site.

The instantaneous ozone production rate is driven by peroxy radical chain propagation involving the reaction of NO and RO2* (Griffith et al., 2016). Therefore, we calculated P(O3) based on measured RO2* using Eq. (5). Since the rate constants of HO2 + NO and RO2+NO reactions are similar (the difference between kNO+HO2 and ki is less than 10 %) (Orlando and Tyndall, 2012), an effective rate constant keff is adopted for the calculation, which is typically approximated by the value of kNO+HO2 (Anderson et al., 2019). The photochemical loss of ozone D(O3) is calculated using Eq. (6), including the photolysis, reactions with OH and HO2, and the reaction of NO2 with OH (Tan et al., 2021). Here, D(O3) was derived using kinetic rate constants together with observed mean concentrations (and representative literature values for unmeasured species) under 298 K and 1 atm. Using the typical conditions (j(O1D)=1.0×10-5 s-1, OH=1.0×106 cm-3, NO=10 ppbv, O3=40 ppbv, HO2≈15 pptv), D(O3) was estimated to be ∼0.70 ppbvh-1, accounting for only ∼5 % of P(O3). Therefore, P(O3) can be approximated as P(O3)net. Considering uncertainties associated with RO2* measurements (18.4 %–25.7 %), the approximation of keff (∼10 %), and upper limit of D(O3) level (5 %), the overall uncertainty in P(O3)net was estimated to be 21.5 %–28.0 %. 5P(O3)=kNO+HO2[HO2][NO]+[NO]∑iki[RO2]i=keff[RO2∗][NO]6D(O3)=fH2Oj(O1D)[O3]+kOH+O3[OH][O3]+kHO2+O3[HO2][O3]+kOH+NO2[OH][NO2]7P(O3)net=P(O3)-D(O3)

Substantial variability in P(O3) was observed during the campaign, with daily peak values ranging from 14 to 111 ppbvh-1 and a daytime average of 14.41 ± 17.04 ppbvh-1 (Fig. 3e). After sunrise, with increasing solar intensity and NO concentrations, photochemical reactions became more active, leading to a rapid increase in P(O3), which peaked around 10:00 (∼27 ppbvh-1) (Fig. 5a). Subsequently, P(O3) gradually decreased in the afternoon with the weakening solar radiation. We found a sustaining increase of P(O3) with NO concentration (Fig. 5b), suggesting that ozone formation fell within the NOx-limited regime under typical daytime conditions at this site. Similar positive P(O3)–NOx relationships have been reported in both urban and suburban environments – such as the Nashville plume study (Thornton et al., 2002), rural North China field campaigns (Tan et al., 2018a), and recent observations in Hefei (Yu et al., 2023). The consistent increase of P(O3) and RO2* within low NOx levels demonstrates that the chain propagation for peroxy formation can be enhanced by NO increase and jointly accelerate instantaneous P(O3).

We summarized previously reported observations of RO2*, P(O3), and related parameters in Table 2. The P(O3) observed in this study was generally higher than those at low-NO sites such as Wangdu (10.44 ppbvh-1), San Antonio (4.2 ppbvh-1), and Hefei (5.91 ppbvh-1) except for a suburban site in northern China (19.5 ppbvh-1). Even high-NO sites like Paris (15.8 ppbvh-1) and Heshan (18.1 ppbvh-1) showed comparable or slightly higher P(O3) than our results. This comparison suggests that, despite the moderate RO2* levels at our site, the instantaneous photochemical ozone production remained considerably strong, pointing to an efficient RO2*–NOx catalytic cycle in Zhuhai.

Based on our measurements, we can quantitatively distinguish the relative contributions from local photochemical production and regional transport to surface O3. Considering the rapid equilibrium between NO2 and O3, the variation of odd oxidants (d[Ox]/dt, where Ox=O3+NO2) is used to reflect the combined influence of local photochemical and transport on surface ozone. Therefore, the transport impact can be expressed by R(O3), which is calculated from the difference between d[Ox]/dt and P(O3)net as defined in Eq. (8): 8R(O3)=dOxdt-P(O3)net Here, the sign of R(O3) indicates the different influences from transport. When R(O3)>0, regional transport enhances local O3 concentrations, suggesting that O3 could be contributed by transport from surrounding regions. Conversely, R(O3)<0 implies that transport acts as a dilution or export role in regulating the surface ozone.

The contribution from local photochemical production and transport to surface O3 exhibited distinct patterns under different pollution conditions (Fig. 6). We define the polluted days as the maximum O3 concentration exceeding the national Class I standard of 160 µgm-3, while the clean periods corresponded to days with a maximum hourly O3 concentration below 120 µgm-3. A total of 14 d and 4 d are categorized into polluted and clean periods, respectively. The daytime increase in ozone concentration in Zhuhai was primarily driven by local photochemical production, whereas regional transport generally exhibited an export effect. However, the magnitudes and temporal patterns of their contributions differed significantly between clean and pollution scenarios. The P(O3) exhibited similar diurnal variations among both polluted and clean days, while polluted periods showed stronger ozone formation than clean periods. Specifically, the peak P(O3) on polluted days reached ∼30 ppbvh-1, which was approximately 50 % higher than that on clean days (∼20 ppbvh-1). In contrast to P(O3), the behavior of R(O3) on polluted days diverged significantly from that on clean days. Around midday, R(O3) on polluted days rose rapidly from a negative value to zero and became slightly positive, indicating a suppressed airmass dispersion or even slight import from regional O3. The distinct transport patterns during polluted days are attributed to frequent coastal wind-field convergence that weakened ventilation. Such weakening export in the midday facilitated the rapid accumulation of locally produced ozone and contributed to the higher O3 peaks during polluted days compared to clean days. As a result, the elevated surface ozone on polluted days was resulted from the combined effect of stronger photochemical production and reduced O3 export at midday, whereas the P(O3) was almost offset by the transport effect of O3 on clean days, leading to flat variation of O3.