The sampling campaign was concurrently conducted at both the inlet and at the outlet of the Zhujiang tunnel (23∘6′ N, 113∘14′ E), which is located in urban Guangzhou, southern China (Fig. S1), on three weekdays (14–16 October 2019) and two weekend days (13 and 19 October 2019). A detailed description of the Zhujiang tunnel can be found in our previous studies (Liu et al., 2014; Zhang et al., 2016, 2017, 2018). IVOCs were collected by a sorption tube (Tenax TA/Carbograph 5TD, Marks International Ltd., UK) using an automatic sampler (JEC921, Jectec Science and Technology, Co., Ltd., Beijing, China). A Teflon filter was installed before the tube to remove particles in the air flow. The sampling flow rate was set at 0.6 L min-1, and hourly samples were collected from 05:00 to 24:00 LT on each sampling day. In order to compare SOA production from IVOCs and VOCs, hourly VOC samples were collected on 13 October 2019 with stainless-steel canisters at a flow rate of 66.7 mL min-1 using a model 910 pressurized canister sampler (Xonteck, Inc., CA, USA). Two-hour quartz filter samples were also collected by a high-volume PM2.5 sampler (Thermo Electron, Inc., USA) at the outlet and inlet sampling sites from 13 to 19 October. Trace gases were measured by online analyzers (CO, model 48i, Thermo Electron Inc., USA; NOx, model 42i, Thermo Electron Inc., USA). A video camera was installed at the inlet to record the vehicle flow during the campaign. After sampling, the video data were used to count the passing vehicles and classify vehicles into different fuel types.

Sampled sorption tubes were analyzed by a thermal desorption (TD) system (TD-100, Markes International Ltd., UK) coupled to a gas chromatography and mass selective detector (GC/MSD; 7890 GC/5975 MSD, Agilent Technologies, USA) with a capillary column (HP-5MS, 30m×0.25mm×0.25µm, Agilent Technologies, USA). Deuterated standards (C12-d26, C16-d34, C20-d42, naphthalene-d8, acenaphthene-d10, and phenanthrene-d10) were injected into the sorption tubes to determine their recoveries before analysis. The sampled sorption tubes and field blanks were thermally desorbed at 320 ∘C for 20 min, and the desorbed compounds were carried by high-purity helium into a cryogenic trap at -10 ∘C; then the trap was rapidly heated to transfer them into the GC/MSD system. The initial temperature of the GC oven was set at 65 ∘C, held for 2 min, then increased to 290 ∘C at 5 ∘C min-1, and kept at 290 ∘C for 20 min. The MSD was used in the SCAN mode with an electron impacting ionization at 70 eV.

Individual speciated IVOCs were quantified with the calibration curves by using authentic standards. The total IVOC mass was determined using the approach developed by Zhao et al. (2014, 2015, 2016), and a detailed description is provided in the Supplement (Sect. S1). Briefly, the total ion chromatogram (TIC) of IVOCs was divided into 11 bins based on the retention times of C12–C22 n-alkanes. Each bin centered on the retention time of an n-alkane. The start time and end times of the bin were determined by the average retention time of two successive n-alkanes. For example, the start time of bin 16 (B16) was calculated as the average retention time of n-C15 and n-C16, and the end time of B16 as the average retention time of n-C16 and n-C17. The IVOC mass in each bin was quantified by the response factor of n-alkane in the same bin. The total IVOC mass was the sum of IVOC masses determined in the 11 bins. The mass of unresolved complex mixtures of IVOCs (UCM-IVOCs) was determined by the difference between the total IVOCs and speciated IVOCs in each bin. The UCM-IVOCs were further classified into unspeciated branch alkanes (b-alkanes) and cyclic compounds (Zhao et al., 2014) (Sect. S1). The analysis of VOCs can be found elsewhere (Zhang et al., 2018). The POA emission was estimated as 1.2 times that of organic carbon that was measured in quartz filter samples (Zhao et al., 2015), which were analyzed by an OC/EC analyzer (DRI model 2015, Nevada, USA) (Li et al., 2018).

Before their use for field sampling, sorption tubes were conditioned at 320 ∘C for 2 h in an oxygen-free nitrogen flow and then sealed at both ends with brass storage caps fitted with PTFE ferrules. About 15 % of conditioned tubes were selected randomly to be analyzed in the same way as normal samples to check if any targeted species existed in the tubes. The batch of sorption tubes were certified as clean if speciated IVOCs were not found or presented in levels below the method detection limits (MDLs). Before and after sampling, the flow rates of samplers were calibrated by a soap-membrane flowmeter (Gilian Gilibrator-2, Sensidyne, USA). During the sampling, 10 field blanks (five at the inlet and five at the outlet) were collected by installing a sorption tube onto the sampler each day but with pump off at both the inlet and the outlet. The speciated IVOCs were not detected or presented in levels below their MDLs in the blanks. MDLs for all the speciated IVOCs, including n-alkanes and polycyclic aromatic hydrocarbons (PAHs), were below 8 ng m-3, such as 5.8 ng m-3 for n-C12, 5.9 ng m-3 for n-C16, and 4.7 ng m-3 for n-C22. To check if any breakthrough occurred during the sampling, prior to the field campaign two sorption tubes were connected in series to sample at the tunnel outlet station in the same way. IVOCs detected in the second tube only accounted for 2.6±1.4 % of the total in the two tubes, indicating negligible breakthrough during the sampling. To check the recoveries during thermodesorption, selected sampled sorption tubes were analyzed twice by the TD-GC/MS system, and the desorption recoveries, calculated as the percentage of IVOCs in first analysis, were 96.7±3.2 % on average. Duplicated samples revealed less than 15 % differences for all the speciated IVOCs.

The vehicular EF of IVOCs can be calculated by the following equation (Pierson and Brachaczek, 1983; Zhang et al., 2016, 2017, 2018): 1EF=ΔC×Vair×T×AN×l, where EF (mg km-1 veh.-1) is the fleet-average emission factor of a given species during the time interval T (1 h in this study); ΔC (mg m-3) is the inlet–outlet incremental concentration of IVOCs; Vair (m s-1) is wind speed parallel to the tunnel measured by a 3-D sonic anemometer (Campbell, Inc.); A (m2) is the cross-sectional area of the tunnel; N is the number of vehicles traveling through the tunnel during the time interval T; and l (km) is the distance between the outlet and the inlet.

Figure 1 shows diurnal variations of vehicle numbers and vehicular IVOC emission factors (EFIVOCs) during the campaign. Traffic flow in the tunnel varied between 571 and 2263 vehicles per hour during the campaign, and gasoline vehicles (GVs) dominated the vehicle fleets with a share of 76.3 % on average; diesel vehicles (DVs) only accounted for 4.0 %, and other types of vehicles, including liquefied petroleum gas vehicles (LPGVs) and electrical vehicles (EVs), had an average percentage of 18.7 % (Fig. S2). As LPGVs and EVs are considered to have no IVOC emissions (Stewart et al., 2021), only GVs and DVs are responsible for the inlet–outlet incremental concentrations of IVOCs. Based on the above Eq. (1), fleet-average EFIVOCs (GVs + DVs) ranged from 13.29±5.08 mg km-1 veh.-1 to 21.40±5.01 mg km-1 veh.-1, with an average of 16.77±0.89 mg km-1 veh.-1 (average ±95 % CI, confidence interval) (Fig. 1). The average EFIVOCs for DVs and GVs could be further derived through linear regression as below (Ho et al., 2007; Kramer et al., 2020): 2EFIVOCs=EFDV×α+EFGV×1-α, where EFIVOCs represents the fleet-average emission factor measured during a time interval; EFDV and EFGV are the average EFIVOCs for DVs and GVs, respectively; and α is the fraction of DVs in the total IVOC-emitting DVs and GVs traveling through the tunnel. Based on the regression results (Fig. S3), the average EFIVOCs for DVs (62.79±18.37 mg km-1 veh.-1) was ∼4.5 times that for GVs (13.95±1.13 mg km-1 veh.-1).

The mileage-based EF can be converted to fuel-based EF with the fuel density and fuel efficiency (Sect. S2) (Zhang et al., 2016). Thus, we could obtain an average fuel-based EFIVOCs of 239.5±19.5 mg kg-1 for GVs and 984.9±288.2 mg kg-1 for DVs. Zhao et al. (2015, 2016) measured IVOC emissions from DVs and GVs in the US by dynamometer tests. As shown in Fig. 2, the average EFIVOCs for DVs measured in our study was significantly lower than that for DVs without any diesel particulate filter (DPF) in the US but over 4 times higher than that with DPF. It is worth noting that the EFIVOCs for DVs from this study was comparable to that for ships and non-road construction machines (NRCMs) with diesel-fueled engines in China (Fig. 2) (Huang et al., 2018; Qi et al., 2019). As a matter of fact, China III or lower emission standard DVs accounted for ∼40 % of China's total in-use DVs in 2019 (http://www.mee.gov.cn/xxgk2018/xxgk/xxgk13/202012/t20201201_810776.html, last access: 1 July 2021), and like the non-road engines, they are not equipped with any after-treatment facilities. Although the after-treatment systems are installed in the China IV and China V DVs, their working performance might be not so satisfactory (Wu et al., 2017). This may explain why the DVs in this study had IVOC EFs comparable to non-road engines. The EFIVOCs for GVs from this study fell into the ranges of that for GVs in the US but was at the high end of the tested values (Fig. 2). A recent study revealed a significantly lower EFIVOCs of 83.7 mg kg-1 for a China V gasoline vehicle (Tang et al., 2021), implying that upgrading the emission standard could help reduce emissions of IVOCs from GVs, as China IV and China III GVs still share a much larger portion than the China V and China VI ones in China's on-road fleets (http://www.mee.gov.cn/hjzl/sthjzk/ydyhjgl/201909/P020190905586230826402.pdf, last access: 1 July 2021).

Figure 3 shows the EFs and compositions of the vehicular IVOCs in each retention-time-based bin (Table S1). Similar to previous studies (Zhao et al., 2015, 2016; Huang et al., 2018; Qi et al., 2019; Tang et al., 2021), the unspeciated cyclic compounds dominated the IVOCs, accounting for 59.07±1.06 %, followed by unspeciated b-alkanes (25.27±0.75 %) and speciated IVOCs (15.66±0.60 %). Among the speciated IVOCs (Table S1), naphthalene dominated the quantified PAHs, accounting for 56.82±1.21 % of total PAH emissions. The distribution of IVOCs in retention-time-based bins presented a significant decreasing trend with bin numbers. Previous studies have reported that more than 50 % of IVOCs concentrated in higher-volatility bins like B12, B13, and B14 in gasoline exhaust, while much broader volatility distributions were found in diesel exhaust (Zhao et al., 2015, 2016; Tang et al., 2021). The IVOCs in B12 measured in this study was also the most abundant as the GVs previously tested in the US (Fig. S4). This was reasonable since the GVs dominated the vehicle fleets during our tunnel experiments (Fig. S2). As shown in Fig. S5, the IVOCs determined in each volatility bin well correlated with those in the volatility bins close to them, and the total IVOCs have stronger correlations with IVOCs in the higher-volatility bins like B12, B13, and B14. In addition, the n-alkanes, as displayed in Table S2, were found to be significantly correlated to the total IVOCs that determined in the same volatility bins except for B20 and n-C20. The mass ratios of IVOCs to the n-alkane in the bins ranged from 9.0 to 15.8 (Table S2). As n-alkanes can be more easily and routinely quantified, the relationships of IVOCs and n-alkanes in each volatility bin might be used to estimate total IVOCs from on-road vehicles. However, vehicles types should be taken into consideration when using these ratios, as the results here were obtained for a fleet dominated by GVs.

Emissions of IVOCs from vehicles are often estimated by assuming a ratio of IVOCs to other species such as POA or NMHCs (Shrivastava et al., 2008; Pye and Seinfeld, 2010; Gentner et al., 2012; Murphy et al., 2017; Wu et al., 2019). However, these ratios might be highly variable with fuel types, operation conditions, and engine performance (Lu et al., 2018). As demonstrated in Fig. S6a and b, IVOCs correlated well with NOx (R=0.63, p<0.05) and CO (R=0.58, p<0.05), with an average IVOC-to-NOx ratio of 0.039±0.004 and an average IVOC-to-CO ratio of 0.033±0.015. The measured IVOC-to-POA ratio was 3.35±1.79 (Fig. S6c), which is comparable to that of 3.0±0.9 for GVs previously measured in dynamometer tests simulating arterial and freeway cycles but much higher than that of 1.5 previously used for estimating vehicle emissions in models (Robinson et al., 2007; Hodzic et al., 2010). As shown in Fig. S6d, the average IVOC-to-NMHC ratio measured in this study was 0.36±0.09, which is lower than that previously measured for diesel vehicle exhaust (0.6±0.1) (Zhao et al., 2015) but higher than that previously measured for gasoline vehicle exhaust (<0.2) (Zhao et al., 2016; Tang et al., 2021). It is worth noting that the IVOCs did not present significant correlations with POA or NMHCs from this study for on-road vehicle fleets (Fig. S6c and d). This would cast uncertainty over the emission estimates of IVOCs based on their ratios to POA or NMHCs.

SOA formation potentials of IVOCs from the on-road vehicle fleet as measured in this tunnel study can be estimated as 3SOAFP=∑EFi×Yi, where SOAFP is the SOA formation potential from the gaseous precursors; EFi represents the emission factor of precursor i; and Yi is the SOA yield of precursor i under high NOx at OA concentration of 20 µgm-3 (Zhao et al., 2015; Huang et al., 2018; Qi et al., 2019; Tang et al., 2021) (Table S3). As shown in Fig. 4, the SOA formation potentials from vehicular VOCs and IVOCs totaled 8.24±0.68 mg km-1. The SOA-to-POA ratio was 2.41±1.45, which was comparable to that of GVs tested in China (1.8–4.4) (Tang et al., 2021) and that of GVs (3.6) (Zhao et al., 2016) and high-speed DVs (3.2±1.7) without DPF in the US (Zhao et al., 2015). Our previous chamber studies simulating SOA formation from vehicle exhausts revealed the SOA-to-POA ratios of 2.0 for DVs and 3.8 for GVs when cruising at 40 km h-1 (Deng et al., 2020; Zhang et al., 2020), which is near the average driving speed of vehicles in the tunnel. Among the vehicle-emitted SOA precursors, similar to previous studies (Zhao et al., 2015, 2016; Huang et al., 2018; Qi et al., 2019; Tang et al., 2021), IVOCs produced significantly higher SOA (7.19±0.62 mg km-1), which is ∼7 times that from traditional VOCs (1.04±0.30 mg km-1). Previous smog chamber studies found that SOA formed during photochemical aging of vehicle exhaust could not be explained by traditional VOCs, especially for vehicles cruising at higher speeds (Robinson et al., 2007; Deng et al., 2020; Zhang et al., 2020). If this SOAIVOCs-to-SOAVOCs ratio of 7 from this study is used to re-estimate the SOA formation from exhaust for vehicles cruising at 40 km h-1 in our previous chamber studies (Deng et al., 2020; Zhang et al., 2020), the VOC plus IVOC precursors could explain 91 %–98 % SOA formed for GVs and 31.2 %–48.2 % SOA formed for DVs. Zhao et al. (2015, 2016) reported significantly higher SOAIVOCs-to-SOAVOCs ratios for diesel vehicle exhaust than gasoline vehicle exhaust. Furthermore, we also resolved the SOAIVOCs-to-SOAVOCs ratios for DVs and GVs via liner regression (Sect. S3). As shown in Fig. S7, although the correlation between SOAIVOCs-to-SOAVOCs ratios and DV fractions was not significant, the DVs did present a much higher average SOAIVOCs-to-SOAVOCs ratio (54.9) than that of GVs (6.82). Thus, SOAIVOCs-to-SOAVOCs ratio of 7 obtained from this study in a tunnel dominated by GVs would underestimate SOAIVOCs from DVs, consistent with higher NMHCs to IVOCs ratios in gasoline exhaust than in diesel exhaust (Zhao et al., 2015, 2016; Huang et al., 2018; Qi et al., 2019; Tang et al., 2021). Overall, the observed vehicular IVOCs as SOA precursors can help achieve mass closure between predicted and measured SOA.