TD is implemented using heated quartz-tube reactors. Each TD inlet consists of a quartz tube (60 cm length; 1/4 in. outer diameter; 3.9 mm inner diameter) wrapped with 20 cm Ni–Cr heating wire and insulated with thermal cotton to minimize heat loss. A K-type thermocouple positioned at the center of the heated region monitors the tube outer-wall temperature and is regulated by a temperature controller (stability ±2 °C). The thermocouple accuracy is ±1 °C over -20–1000 °C.

Sample air is delivered through 1/4 in. PTFE tubing with PTFE straight/tee fittings (Swagelok). During operation, the sampling flow is 1 L min⁻¹, corresponding to a residence time of 0.048 s in the heated region. This short residence time is designed to promote rapid dissociation of thermally labile ONs while minimizing post-dissociation chemistry within the heated section. Because the thermocouple measures the tube wall temperature rather than the gas temperature, the reported TD setpoints are expected to be higher than the corresponding gas temperature. Therefore, TD setpoints reported here should be interpreted as operational setpoints for this specific reactor design and are not necessarily directly comparable with other TD geometries.

Thermally generated NO₂ is detected by CEAS near 405 nm. In CEAS, an optical cavity provides a long effective absorption pathlength through multiple reflections between high-reflectivity mirrors, enabling sensitive absorption-based NO₂ measurements while retaining the selectivity of optical spectroscopy. NO₂ has strong, structured absorption in the 405 nm region, allowing retrieval via spectral fitting. A single CEAS NO₂ spectrometer is used for all five inlet channels. The CEAS instrument used here has been characterized previously (Zhou et al., 2022). In the present configuration, the CEAS reports NO₂ mixing ratios for each channel segment, and all TD-derived quantities are obtained by differencing of channel measurements (Sect. 2.5).

The five-channel TD-CEAS combines two TD temperatures (250 and 450 °C) with two filter placements (upstream versus downstream) to enable separation of gas and particle phases (Fig. 1). Ambient air is sampled through a PM_2.5 cyclone prior to entering the inlet manifold. A multi-channel inlet selection unit sequentially routes each channel to the shared CEAS detector. When the multi-channel inlet unit is in operation, only the sample flow from the currently selected channel is allowed to enter the analyzer. Since there is a transition period during which the NO₂ signal stabilizes after switching channels, the data from the first 15 s following the channel switch should be discarded, as shown in Fig. S2 in the Supplement.

The instrument samples five channels plus an ultra-high-purity N₂ reference segment. The ultra-high-purity N₂ is supplied from cylinders and fed directly into the CEAS zero-air inlet, the N₂ flow rate matches the sampling. Each channel is sampled for 1 min, followed by a 1 min N₂ segment, yielding a 6 min measurement cycle. The N₂ segment is used as a reference for instrument baseline and stability (Sect. 2.6).

We denote “F” as a particle filter placed upstream of the TD (gas-only enters TD), and “UF” as an inlet without an upstream filter (gas + particle enter TD), with a particle filter placed downstream of the TD reactor to prevent particles from entering the CEAS cavity. The filters are made of PTFE and has a diameter of 47 mm, with retention efficiency of 99.5 % or higher for 0.3 µm particles. The filter is replaced once a week during field campaigns, or earlier if the analyzer's flow rate decreases and in the event of air pollution. Table S1 summarizes the channel definitions and the corresponding quantities measured as NO₂ mixing ratios by CEAS.

According to a study by Rollins et al. (2010), particulate transmission experiments using non-volatile NaCl particles revealed significant loss of small-particle-size (≤ 100 nm) particles, while the transmission efficiency for particles larger than 100 nm exceeded 85 % (represent a lower limit to the transmission and detection efficiency for semi volatile particles). As for volatile particles, they rapidly evaporate and thermally decompose into NO₂ within the TD. In addition, Garner et al. (2020) performed a cross-validation of NaNO₃ aerosols using TD-CRDS and SMPS; the correlation coefficient between NaNO₃ and the NO₂ produced by its quantitative conversion was 0.98 ± 0.03, which further demonstrates the high transmission efficiency of the particles. Although there may be transmission losses for small particles, these losses are generally acceptable. One line of evidence suggesting that wall losses of organic nitrates (ONs) are negligible is that the uncertainty in the recovery of 2-EHN is below 4 % (Fig. 4). This indicates that ANs rapidly decompose into NO₂ during transport through the TD inlet, and that no significant back-reaction loss of RO radicals with NO₂ occurs. Of course, for high-molecular-weight ONs, wall loss effects are largely dependent on compound volatility, tubing material, and temperature. In this study, we used PTFE tubing and ambient-temperature sampling to minimize such effects.

For each 1 min segment, the CEAS reports an NO₂ mixing ratio for the selected channel. Let CRTF, C250F, C450F, C250UF, and C450UF denote the NO₂ mixing ratios (pptv or ppbv) measured by CEAS in the corresponding channels. Gas- and particle-phase ON classes are retrieved by channel differencing: 1gΣPNs=C250F-CRTF2pΣPNs=C250UF-C250F3gΣANs=C450F-C250F4pΣANs=C450UF-C250UF-C450F-C250F These expressions correspond to the field configuration in which the RT/250/450 channels are operated with upstream particle filtration for gas-phase retrieval, while paired 250/450 channels without upstream filtration (but with downstream particle removal) provide total (gas + particle) ON classes. Sobanski et al. (2016) noted that heated quartz tubes packed with frit and glass beads exhibit a small but significant loss of NO₂, which may be attributed to the packing material increasing the collision area for NO₂, resulting in a small loss of NO₂. No NO₂ transmission losses were observed in heated quartz tubes without packing material (Thieser et al., 2016; Lin et al., 2024).

Because channels are sampled sequentially, each derived quantity is timestamped at the midpoint of the corresponding 1 min segment. When longer averaging periods are applied (e.g., hourly means), within-cycle timing offsets are typically small compared with the averaging window.

Detection limits are reported as 1σ precision inferred from Allan deviation analyses with ultra-high purity N₂ continuously introduced to each channel. For the RT, 250, and 450 °C channels, the 1σ NO₂ detection limits at 1 s integration are 34, 35, and 33 pptv, respectively; at 10 s integration they are 12, 13, and 12 pptv.

Derived ON quantities are calculated as differences between channels. Assuming independent noise among channel measurements, the uncertainty for a difference A-B is propagated as: 5σA-B=σA2+σB2 Using the 1 s NO₂ precision for relevant channels, the resulting 1σ (1 s) detection limits for derived gas-phase ΣPNs and ΣANs are: 6σgΣPNs=σ250F2+σRTF2≈352+342=49pptv7σgΣANs=σ450F2+σ250F2≈332+352=48pptv And for particle-phase ONs are: 8σpΣPNs=σ250UF2+σ250F2≈49pptv9σpΣANs=σ450UF2+σ250UF2+σ450F2+σ250F2≈68pptv These values correspond to the reported field-operation performance. Table 1 compares the detection limits of the present TD-CEAS with those reported for representative TD-based techniques. Overall, the detection limits achieved here are within the range of recent TD instruments, while the present configuration additionally enables sequential retrieval of gas- and particle-phase PNs and ANs using a single NO₂ spectrometer. Because published detection limits are reported at different integration times and for different instrument configurations, the comparison is intended as a general benchmark rather than a strict ranking of sensitivity.

The uncertainty of the reported ONs classes includes detector calibration, channel differencing, and inlet-related artifacts. For the underlying NO₂ measurement, the CEAS calibration uncertainty is approximately 6 % (Zhou et al., 2022), and the repeatability for continuous 10 ppbv NO₂ passed through the TD-CEAS configuration is 1.4 %. These values describe instrumental accuracy and stability for simple NO₂ inputs. They do not include composition-dependent TD chemistry, aerosol transmission losses, or the representativeness of surrogate standards, which may dominate the uncertainty of ambient ΣPN and ΣAN retrievals. Accordingly, the quantitative results reported below are best interpreted as operational ON measurements for the present inlet geometry and correction framework.

A zero-dimensional box model was implemented to reproduce the correction-factor lookup table reported by Li et al. (2021) and to compare that approach with the regression-based correction proposed in this study. The chemical mechanism followed previous work and was based on the Master Chemical Mechanism (MCM) v3.3. The wall loss rate constants for RO₂, HO₂, and OH radicals were set to 0.3, 0.5 and 5.4 s⁻¹, respectively, following Fuchs et al. (2008), Wooldridge et al. (2010), Thieser et al. (2016), and Li et al. (2021). These values are used as ballpark and do not necessarily reflect the wall losses of the inlets since they vary with inlet geometry and operating conditions such as flow and temperature. The model was used only for method comparison and not for interpretation of ambient observations.

To define operational TD temperature for PNs and ANs, we measured temperature-dependent NO₂ yields (“thermal decomposition profile”) for representative organic nitrates. PAN was used as a surrogate acyl peroxy nitrate, isopropyl nitrate and isobutyl nitrate were used as surrogate alkyl nitrates, and 2-ethylhexyl nitrate (2-EHN) aerosol was used as a surrogate particulate organic nitrate. Sodium nitrate aerosol was used to evaluate potential interference from particulate inorganic nitrate.

For each compound, the TD setpoint was stepped across the temperature range and the NO₂ enhancement in the heated channel (relative to room temperature) was used to infer dissociation completeness. As shown in Fig. 2, PAN reaches a stable plateau in NO₂ yield at temperatures above ∼ 250 °C, indicating near-complete conversion at and above this temperature. In contrast, isopropyl nitrate and isobutyl nitrate show minimal dissociation below 250 °C (< 10 % of the high-temperature plateau), increase sharply near ∼ 300 °C, and reach stable plateau by ∼ 350 °C. The particulate 2-EHN shows a similar trend, stabilizing near ∼ 400 °C. Particulate inorganic nitrate behaves differently: the NO₂ yield remains < 10 % below 400 °C and does not reach a stable plateau until ∼ 750 °C, implying that particulate inorganic nitrate from NaNO₃ contributes minimally to the NO₂ signal at 450 °C for this TD geometry and residence time.

Based on these measured thermal decomposition profiles, we selected 250 °C as the operational setpoint for ΣPNs and 450 °C as the operational setpoint for ΣANs. These setpoints are specific to the present TD reactor design (Sect. 2.1) and correspond to wall temperatures measured at the tube exterior. Because they were derived from surrogate standards, they should not be interpreted as compound-independent thresholds for all ambient ONs; in particular, partial dissociation of some multifunctional ANs in the lower temperature channel cannot be excluded (Dewald et al., 2021).

Instrument precision was assessed using Allan deviation analysis with ultra-high-purity N₂ introduced to the inlet. The 1σ (1 s) NO₂ detection limits are 34 pptv (RT), 35 pptv (250 °C), and 33 pptv (450 °C), improving to 12–13 pptv at 10 s integration (Fig. 7). Propagating these channel precisions through the channel-differencing retrieval (Sect. 2.6) yields 1σ (1 s) detection limits of 49 pptv for gΣPNs, 49 pptv for pΣPNs, 48 pptv for gΣANs, and 68 pptv for pΣANs, consistent with values used for field reporting.

Table 1 compares detection limits reported for representative TD-based techniques. The present TD-CEAS achieves detection limits comparable to the lower end of reported TD methods, while providing simultaneous gas/particle partitioning using a single, compact NO₂ spectrometer framework. In terms of temporal resolution, single-cavity configurations are limited by channel switching, resulting in a lower time resolution. Multi-cavity designs therefore represent a promising direction for future instrumental development. Various strategies have been explored to mitigate measurement interferences, including the use of glass bead packing (which reduces but does not fully eliminate interference), correction factor lookup tables, and post-inlet ozone addition. In this study, we propose a correction model based on the NO / NO₂ ratio that accounts for the coupled effects of both interferents. Existing TD setups for particulate-phase measurements mainly employ either denuders or filter. Denuders require careful monitoring of absorption efficiency and saturation, whereas filter membranes only need to be replaced periodically, making them more convenient to use.

The five-channel TD-CEAS was deployed on the rooftop (∼ 15 m a.g.l.) of an office building at the Guangzhou Institute of Geochemistry (GIG), Guangzhou, China (113°21^′49^′′ E, 23°8^′46^′′ N), as shown in Fig. S6. The site is located in a densely populated urban area influenced by nearby traffic corridors (∼ 350 m). Two short intensive observation campaigns were conducted during 4–7 September and 1–2 October 2025 under clear, rain-free conditions.

When measuring ambient air, the air passes through a PM_2.5 cyclone separator (Fig. S6) into a 1/2 in. stainless steel main sampling tube with a high-flow pump used to draw air at the end. The TD-CEAS system collects samples via a bypass line (PTFE, ∼ 2 m) connected to the main sampling tube. Ancillary reactive nitrogen measurements were used to support evaluation of TD-CEAS performance and to enable post-processing correction of TD artifacts. NO was measured with a chemiluminescence NO_x analyzer (Thermo Scientific 42i-TL). During the October deployment, PAN was measured concurrently by a GC–ECD system for intercomparison with the TD-CEAS gas-phase PN signal. Unless otherwise stated, times are reported in local time (UTC+8).

Ambient air was sampled through a PM_2.5 cyclone upstream of the inlet manifold. The instrument was operated in the five-channel configuration described in Sect. 2, with paired inlet/filter arrangements used to separate gas-only and total (gas + particle) signals. The gas-phase channels (RT-F, 250-F, 450-F) used an upstream particle filter (PTFE, retention efficiency of 99.5 % or higher for 0.3 µm particles) to remove particles prior to the TD reactors and CEAS, thereby providing NO₂, gΣPNs, and gΣANs by channel differencing. The particle-inclusive channels (250-UF, 450-UF) sampled air without upstream filtration so that both gas and particles entered the TD reactors, while a downstream filter removed particles immediately before the CEAS cavity. These channels therefore provided total (gas + particle) ΣPNs and ΣANs, and particle-phase quantities (pΣPNs and pΣANs) were obtained by differencing the total and gas-only channels.

Each channel was sampled for 1 min, after all five channels have been measured, a 1 min ultra-high-purity N₂ segment was introduced, yielding a 6 min measurement cycle. One-minute mean NO₂ values were calculated for each channel segment, and the derived ON products were reported at the 6 min cadence of the cycle. For clarity in time-series presentation, hourly means were calculated from the 6 min data (Sect. 4.3).

Because thermally labile nocturnal species such as N₂O₅ and ClNO₂ can decompose in the TD inlets and bias gas-phase ON retrievals, the quantitative discussion in this work focuses on daytime gas-phase ONs, when such interferences are expected to be less important. Nighttime gas-phase ON data are shown for completeness and should be interpreted with caution. This is because the presence of other reactive nitrogen oxides at night, such as ClNO₂, N₂O₅, and HONO, as described in Sect. 3.3.4 (which form at night and photolysis during the day), can significantly interfere with ON measurements. In contrast, particle-phase ONs were retained for both daytime and nighttime because the differencing approach reduces gas-phase interferences that are common to the gas-only and total configurations, although it may not remove them completely.

Laboratory-derived NO / NO₂ interferences in the ΣPN channels (Sect. 3.3.1) are expected to be most relevant under urban, high-NO_x conditions. Accordingly, the correction model obtained from PAN-based laboratory experiments was applied to the 250 °C channel measurements using simultaneously measured NO from the chemiluminescence analyzer and NO₂ from the RT channel. The corrected 250 °C data were then used to derive gΣPNs and pΣPNs. No additional correction was applied to ΣANs beyond the 450–250 differencing described above, and the resulting ΣAN values are interpreted operationally.

Figure 8 shows hourly mean time series for NO, NO₂, gΣPNs, pΣPNs, gΣANs, and pΣANs during the two intensive observation periods. NO and NO₂ exhibited typical urban variability, with higher concentrations at night and lower concentrations during the day. The mean NO and NO₂ mixing ratios were 1.5 ± 2.1 and 14.8 ± 9.3 ppbv (mean ±1σ) in September, and 1.4 ± 1.3 and 18.2 ± 14.4 ppbv in October.

Daytime gΣPNs are elevated in both campaigns, with mean values of 2.8 ± 1.2 ppbv in September and 2.1 ± 1.2 ppbv in October and corresponding maxima of 8.4 and 5.8 ppbv, respectively. Particle-phase ΣPNs were smaller (e.g., 1.1 ± 0.5 ppbv in September) and showed less distinct diel structure. gΣANs displayed clear daytime maxima, with a mean of 1.0 ± 0.5 and 1.2 ± 0.5 ppbv in September and October, respectively. In contrast, pΣANs (1.0 ± 0.4 ppbv in September; 0.9 ± 0.5 ppbv in October) tended to peak at night, consistent with enhanced nighttime nitrate chemistry contributions and the expected nocturnal partitioning/production of low-volatility ONs (Fry et al., 2013).

The trends in daytime gΣPNs and temperature are positively correlated, as shown in Fig. 9. The fractions of gΣPNs (gΣPNs / (gΣPNs + pΣPNs)) increases from less than 60 % in the morning (08:00, ∼ 29 °C) to nearly 80 % in the afternoon (16:00, ∼ 34 °C), which is qualitatively consistent with warmer conditions favoring the gas phase. We therefore regard the daytime evolution of gΣPNs as robust in a relative sense. By contrast, the absolute magnitude of pΣPNs should be interpreted more cautiously because pΣPN is obtained by channel subtraction and has not been directly validated with a particulate PN standard. The residual particulate signal may reflect a combination of less volatile PN components, temperature-dependent partitioning, and subtraction uncertainty. Relative humidity may also influence ΣPNs through PAN hydrolysis and heterogeneous uptake on particles (Salas et al., 2020; Sun et al., 2022). The nighttime increase in gΣPNs is not interpreted quantitatively because ClNO₂, N₂O₅, and HONO can contribute to the gas-phase channels after dark. For ΣANs, the particle-phase fraction increases as temperature decreases, which is consistent with partitioning of lower-volatility organic nitrates towards the condensed phase during nighttime conditions.

During the October campaign, corrected gΣPNs were evaluated against independently measured PAN from the GC-ECD system. The two measurements agreed well in temporal variability, and linear regression yielded R2 = 0.83. The regression slope indicates that PAN accounted for approximately 78 % of daytime gΣPNs (Figs. S7, S8). This proportion ranges from 48 % to 78 % at the temperate forest site in Paris, France (Andersen et al., 2025). This supports the quantitative performance of the TD-CEAS gas-phase ΣPNs retrieval under urban conditions and provides an observational constraint on the extent to which PAN dominated the measured gΣPNs pool during the deployment. Based on the comparison of raw ΣPNs and PAN (where gΣPNs are, at certain times, unreasonably lower than PAN), it is necessary to correct the raw data because the uncorrected values occasionally imply a physically implausible negative inlet bias. At the same time, the correction remains PAN-anchored. In the real atmosphere, ΣPNs comprise a mixture of peroxy nitrates, whose thermal dissociation products may not all behave identically to the PA radical. The corrected daytime gΣPNs values should therefore be interpreted as operational ΣPN estimates constrained by PAN, rather than as composition-independent absolute totals.

These field observations demonstrate the capability of the five-channel TD-CEAS to provide continuous measurements of NO₂ and operational ON classes in both gas and particle phases at a 6 min cadence, with performance evaluation supported by independent PAN measurements and correction for NO / NO₂-dependent artifacts relevant to high-NO_x environments.