The National Centers for Environmental Prediction and National Center for Atmospheric Research reanalysis meteorological data set (https://ready.arl.noaa.gov/archives.php) was input into National Oceanic and Atmospheric Administration Hysplit (http://ready.arl.noaa.gov/HYSPLIT.php) software to calculate mixing layer depths (MLDs) during the measurement period.

The maximum incremental reactivity (MIR) method was used to calculate the OFP of each VOC component to evaluate the contribution of VOCs to O₃ generation. The OFP calculation formula is as follows: O F P i = M I R i × [ V O C ] i (1) where MIR is the maximum incremental reactivity, and the value of MIR is obtained from the California Code of Regulations, 2010 (Carter, 2010); [VOC] _i is the concentration of each VOC component in ppb.

The OH radical loss rate (L _OH ) of VOC components was calculated using the following formula: L O H i = k i O H × [ V O C ] i (2) where k ^OH is the reaction rate constant of VOCs and hydroxyl radicals, and the data originate from reference (Atkinson and Arey, 2003).

Positive matrix factorization (PMF) model was first proposed to factor analysis in 1994 (Paatero and Tapper, 1994) and it has been widely used recently as a tool to resolve VOCs source-receptor relationship and to evaluate the different sources contributions. PMF is a receptor model which quantifies contributions from various sources by mathematical statistics according to the different chemical composition or source fingerprint. In this paper, the EPA PMF 5.0 software with a multiple linear engine version 2 (ME-2) platform (Norris et al., 2014) is used for source analysis. Based on Eq. 3, PMF calculates the following parameters: 1) the number of VOCs source factors p, 2) the chemical composition of each source factor f, 3) contribution g of each source factor to the sample, and 4) residual e _ij of each species to each sample. The constraint of PMF is that the contribution g and source factor f in the equation should be non-negative to have physical significance. X i j = ∑ k = 1 p g ik f kj + e ij (3) where, the data matrix X _ij represents the concentration of VOCs, i represents the observed sample, j represents the species.

The target of PMF is to minimize the functional residual Q in the Eq. 4, where n and m are the number of samples and the number of species respectively, and u _ij is the uncertainty of X _ij . Q = ∑ i = 1 n ∑ j = 1 m [ ∑ k = 1 p g ik f kj + e ij − ∑ k = 1 p g ik f kj u i j ] (4)

The input file consists of two matrices, observed species concentration (X _ij ) and uncertainty (u _ij ).

Uncertainty calculation formula is: U n c = { ( E r r o r   F r a c t i o n × C o n c . ) 2 + ( 0.5 × M D L j ) C o n c . > MDL 5 6 × C o n c . C o n c . ≤ M D L 4 × m e d i a   c o n c . M i s s i n g   d a t a (5) Q e x p e c t e d . = n × m − p × ( n + m )

Based on the extensively used principles, missing rates greater than 25% or concentrations lower than MDL are excluded (Han et al., 2021). Finally, 21 species were selected and input into the PMF model calculation. Three to nine factors were estimated to obtain the best resolution based on the Q_true/Q_expected and Q_true/Q_robust value (Li et al., 2020). In this study, six factors were determined with regard to the Q_true/Q_robust = 1.06 and Q_true/ Q_expected = 1.09. Around 95% of the scaled residuals were between −3 and 3. Fpeak with the value of −1 to 1 at intervals of 0.1 was determined to investigate the free rotation (Song et al., 2018). The results showed that Fpeak values had no significant difference, indicating a good fit of the modeled results.

To quantitatively describe the relationship between the human exposure dose and adverse health reaction, according to the carcinogenicity of substances, health risk assessments are generally of two types: carcinogenic and noncarcinogenic risk assessments (Xia et al., 2014; Cheng et al., 2019). In this study, the assessment indexes were calculated using Eqs. 6–9. E C = C × E T   × E F   × E D A T (6) R i s k = I U R × E C (7) H Q = E C / R f C (8) H I = ∑ i = 1 n H Q i   (9) where EC is the exposure concentration in μg/m³, C is the environmental concentration of pollutants in μg/m³, ET is the exposure time, EF is the exposure frequency, ED is the exposure duration (in years), AT is the average time, RfC is the inhaled carcinogenic risk concentration in mg/m³, IUR is the US Environmental Protection Agency (EPA) inhalation unit risk in μg/m³, Risk is the value of carcinogenic risk, HQ is the noncarcinogenic risk hazard quotient (HQ), and HI is the noncarcinogenic risk index. EF was obtained from the US EPA Integrated Risk Information System (IRIS, 365 d/y; http://www.epa.gov/iris), and ET, ED, and AT were 3.68 h/day, 74.8 years, and 74.8 × 365 × 24 h, respectively (Cheng et al., 2019).

As detailed in Table 1, the hourly mean mixing ratio with ±1 σ of total VOCs in our measurement was 21.5 ± 18.6 ppb (median: 17.8 ppb), which was lower than the annual average VOC mixing ratio of 37.4 ppb (sum of alkanes, alkenes and aromatics) obtained in Lhasa city during 2019 through Summa canister offline sampling (Yu et al., 2022). Leaving their difference in sampling time and frequency aside, it indicated that there have months with high VOCs at Lhasa than that in May when we measured. The concentration was also lower than those reported in large cities such as Beijing, Tianjin, Nanjing, and Los Angeles (see Table 2) but higher than those reported in clean rural sites such as Gonggar Mountain. The alkane volume mixing ratio was 12.4 ± 11.2 ppb, accounting for 57.7% of the total VOCs, whereas the ratios of aromatics and alkenes were 4.6 ± 3.5 and 4.5 ± 2.8 ppb, accounting for 21.4% and 20.9% of the total VOCs, respectively. The relative contribution of the four types of VOCs here was similar to other cities, inferring comparable emission inventory structure of VOCs. Much lower VOCs might be related to the relatively lower emission strength in the city of Lhasa. According to mixing ratios, the top 10 VOC components were ethane, isobutene, isobutane, n-butane, benzene, toluene, n-undecane, 2,2-dimethylbutane, propylene, and n-hexane, respectively (Table 3).

BTEX is a general term for benzene, toluene, ethylbenzene, and three xylene isomers (m-xylene, p-xylene, and o-xylene); it represents a key group among VOCs. BTEX mainly originates from solvent use, fossil fuel and biomass combustion, and it plays a pivotal role in atmospheric photochemical smog and particulate pollution (Zhang et al., 2016; Yao et al., 2017).

Supplementary Figure S1 depicts the time series variations in benzene, toluene, ethylbenzene, m/p-xylene, and o-xylene mixing ratios from May 1 to 31 May 2019. BTEX mixing ratios exhibited day-to-day fluctuations of 1.8–20.2 ppb. The average benzene, toluene, ethylbenzene, m/p-xylene, and o-xylene mixing ratios with ±1 σ were 1.65 ± 0.8, 1.21 ± 1.02, 0.29 ± 0.25, 0.56 ± 0.58, 0.49 ± 0.48 ppb, respectively, and BTEX accounted for 19.5% of the total VOCs.

Figure 3 provides a comparison of BTEX mixing ratios in Lhasa with the ratios recorded in other cities. The BTEX mixing ratios in Lhasa were much lower than those in Beijing (January 2014) (Li et al., 2015), Shanghai (January 2007–March 2010) (Cai et al., 2010), Guangzhou (November–December 2009) (Zhang et al., 2013), and the Dinghushan rural site in the Pearl River Delta (April 2005) (Tang et al., 2007). However, it is worth noting that the benzene to BTEX ratio was higher in Lhasa (39%) than that in Beijing (24%), Shanghai (19%), and Guangzhou (30%). The BTEX mixing ratio in Lhasa was higher than that at the rural site in Changdao, Shandong province (March–April 2011) (Yuan et al., 2013) and Waliguan, Tibet Plateau (April–May 2003) (Xue et al., 2013).

The main emission sources of benzene are combustion processes, including biomass, biofuel, or coal burning (BM/BF/CB). The proportion of benzene in BM/BF/CB sources is higher than that of toluene, whereas the proportion of benzene in motor vehicle emissions is generally lower than that of toluene (Barletta et al., 2005). In this study, a ternary diagram of benzene, toluene, and ethylbenzene mixing ratios was used to determine the predominant emission sources (Zhang et al., 2016), and the results are presented in Figure 7. In the diagram, the benzene, toluene, and ethylbenzene mixing ratios are normalized. The distribution of benzene, toluene, and ethylbenzene mostly occurred in the biomass burning and transportation emission areas, and the BM/BF/CB areas accounted for a larger proportion of the emissions than the areas with industrial and solvent and traffic emissions. However, most the scattering dots did not fall within the wireframes, which indicated different emission spectrums on the plateau from that in the area with low altitudes. Studies on aerosols and other air pollutants in Lhasa have demonstrated the considerable impact of biomass burning related to religious activities and regional transport (Cui et al., 2018; Duo et al., 2018). As displayed in Figure 8, during rush hours (7:00–9:00 and 17:00) and nighttime peak periods (21:00), which were expected to be associated with higher traffic-related emissions, the emission levels did not differ with those of other time periods. Because the patterns of benzene, toluene, and ethylbenzene distribution between BM/BF/CB and traffic emissions should differ, traffic emissions were expected to have been a dominant influence on ambient VOCs in Lhasa since disorderly incense and biomass burning activities have been greatly restrained in protecting the environment in recent years.

The ratio of distinct VOC species can be used to determine the predominant pollution sources on the basis of the difference in the characteristic species of VOCs in the corresponding sources. The toluene-to-benzene (T/B) ratio has often been used to identify the sources of pollutants (Niu et al., 2012; Y. Zhang et al., 2013). The T/B ratio of ≤2 indicates that the main sources are vehicle exhaust emissions or biomass burning. A T/B ratio of >2 indicates that the pollutants were emitted from other sources, and a T/B ratio of 4.25–10.00 strongly implies the influence of industrial emission sources (Barletta et al., 2008; Niu et al., 2012; Kumar et al., 2018). A study on pollution sources revealed that the T/B ratio for biomass combustion was approximately 0.5 or even lower, whereas that for motor vehicle emissions was nearly 1.7 (Barletta et al., 2005).

The correlations between toluene and benzene mixing ratios in distinct periods are depicted in Figure 8. All the correlations were significant (R ² > 0.6 and p < 0.05), indicating that the pollutants were emitted from the same or similar sources. The slopes (T/B ratio) of the fitting curves were 1.52 at 7:00–9:00 and 1.85 at 21:00, both of which are higher than the T/B ratio values of 1.24 at 17:00 and 1.12 during other periods. The emissions from motor vehicles were associated with higher T/B ratios than biomass combustion (Zhang et al., 2013); thus, the results reflected more traffic-related emissions during the periods of 7:00–9:00 and 21:00.

M/p-xylene and ethylbenzene usually have a common source, but the reaction rate of m/p-xylene with hydroxyl radicals is nearly three times higher than that of ethylbenzene, and the ratio of m/p-xylene to ethylbenzene (X/E) gradually decreases as the photochemical reaction proceeds; hence, the X/E can be used as an indicator of the degree of air mass aging (Huang et al., 2015; Han et al., 2020).

Figure 9 displays the average diurnal variations in benzene, toluene, m/p-xylene, and ethylbenzene mixing ratios, as well as in T/B and X/E ratios. Benzene, toluene, ethylbenzene, and m/p-xylene mixing ratios had peak values and high T/B and X/E ratios were found from 7:00 to 9:00 in the morning, which coincided with the morning rush hours. Similar patterns were observed during other times of peak traffic at approximately 17:00 and 21:00. Trough values were observed at 13:00–14:00, as well as low T/B and X/E ratios. The peak and trough values of toluene and m/p-xylene mixing ratios were more obvious than those of benzene and ethylbenzene mixing ratios because toluene and m/p-xylene have higher photochemical reactivity. The average diurnal variation in T/B ratios was 0.5–0.9, which is close to the values obtained at the rural site of Shangdianzi (Han et al., 2020). The low T/B ratio indicates the crucial influence of biomass combustion. The average diurnal variation in the X/E ratio was 0.8–2.0, which was higher than the ratios obtained at Shangdianzi and Beijing (Han et al., 2020).

The result of VOCs source apportionment by PMF model analysis was shown in Figure 10 (left panel), in which six factors’ resolutions with a stable Q value were identified as the dominant source of VOCs in urban Lhasa city. The average diurnal variations in each source concentration were shown in Figure 10 (right panel).

The first factor was characterized by the strong presence of ethane, ethene, n-butane, butene, 2-methylpentan, 3-methylpentan, aromatics like toluene, benzene, xylene. These species are mainly from traffic emissions and gasoline evaporation (Li et al., 2020). Therefore, this source is identified as gasoline automobile emissions. As can be seen in Figure 10 (right panel), the average diurnal variation in gasoline automobile emissions presented peaks at 7:00–9:00, 17:00 and 21:00, which peaks reflected the influences of traffic rush hours.

Factor 2 was distinguished by a high percentage of alkanes with high carbon number, such as n-decane and n-undecane, which were abundant in diesel vehicle exhaust (Zhang et al., 2021). Ethane, ethene, propene and benzene, ethyl benzene and xylene were also rich in this factor with high percentages. These species are associated with diesel vehicular emissions. Thus, Factor 2 was featured as diesel vehicle exhaust. Due to restrictions in freight vehicles in the daytime, the VOCs emissions contribution from diesel vehicles might be lower during the day than at night. In addition, lower MLD also increased the level of the corresponding VOCs mixing ratio at night (Figure 10, right).

Factor 3 was characterized by high levels of aromatics (xylene, styrene, ethylbenzene, and benzene). Toluene, ethylbenzene, xylene, and 1,3,5-trimethylbenzene are used as organic solvents in various industries and can be regarded as indicator species for solvent volatilization. Alkanes like methylpentane, methylheptane, n-decane, n-undecane, are also widely used as nonpolar solvents (Yuan et al., 2010; Song et al., 2018). Therefore, source 3 was identified as solvents usage. Because solvent usage often occurred in daytime and the solvents have a more volatility under higher temperature, VOCs concentrations from solvents usage contribution were higher during the day than at night. The change of BTEX and aromatics mixing ratios were significantly and positively correlated with air temperature in Supplementary Table S2 also supported this conclusion.

Factor 4 was distinguished by high percentages of ethane, ethene, butane and butene which could be emitted from incomplete combustion of biomass (Ling et al., 2011; Li et al., 2020). In addition, benzene accounts for a high load with a contribution of 44%. The concentration of benzene was higher than toluene in this factor, which is consistent with the characteristics of biomass combustion. There was high level of VOCs from biomass burning in the morning (Figure 10, right), which may partly be related to the more activities in incense burning in the forenoon. There had a distinct feature with a flat curve and with relatively high concentrations for its diurnal change in this factor, which indicates that biomass burning in Lhasa is still common since biomass fuel is the main traditional fuel in the plateau. The less variation in diurnal changes might also tell us that the influence of biomass burning there is of regional characteristics.

Factor 5 was distinguished by high percentages of C₂-C₄ alkanes, propane, i-butane, and n-butane are regarded as the major compounds of liquid petroleum gas (LPG) and natural gas (NG) (Shao et al., 2016; Han et al., 2021). LPG and NG both play important roles in domestic catering and traffic. Thus, source 5 is assigned to LPG & NG usage. The diurnal variation of this source was similar to motor vehicles, except no obvious peak at 21:00. Public transportation in Lhasa has promoted the fuel conversion from oil to natural gas since 2014. Since this diurnal variation showed a similar pattern with factor 1, it reflected more contribution from traffic emissions than others.

The source profile of factor 6 was dominated by isoprene, which contributed 71% to the total isoprene. Isoprene is a typical tracer for plant emission, but traffic emissions and combustion also emit a small amount of isoprene (Borbon et al., 2001). Plants emit more isoprene during the day with solar radiation and higher temperature than at night. Therefore, the diurnal variation in plant emission showed typical feature with high values in the daytime and low at night (Song et al., 2018; Li et al., 2020). Our result in factor 6 was consistent with this natural law. Accordingly, factor 6 was identified as plant emission.

Figure 11 showed the contribution of each source to the concentration, L _OH and OFP. Traffic emissions contributed 44.5%, including gasoline automobile emissions (30.3%), diesel vehicle exhausts (14.2%), to TVOCs; 41.6% to L _OH and 47.4% to OFP. In addition, LPG&NG related emissions contributed 5.7% to TVOCs, including cooking fuel emissions, gas station volatilization, natural gas bus and taxi emissions, and so on. Biomass burning contributed 41.3% to TVOCs, 26.4% to L _OH and 29.7% to OFP. Overall, air quality in urban Lhasa city was strongly affected by motor vehicle emissions now, exceeding the biggest contribution of biomass burning from the result of Yu et al. (2022). Plant emissions contributed only 1.5% to TVOCs, but relatively high proportions to L _OH (14.6%) and OFP (4.8%), due to the high reactivity of isoprene. The contribution of plant emissions was similar with the other studies (Yu et al., 2001; Yu et al., 2022).