Effectiveness of replacing catalytic converters in LPG-fueled vehicles in Hong Kong

. Many taxis and public buses are powered by liqueﬁed petroleum gas (LPG) in Hong Kong. With more vehicles using LPG, they have become the major contributor to ambient volatile organic compounds (VOCs) in Hong Kong. An intervention program which aimed to reduce the emissions of VOCs and nitrogen oxides (NO x ) from LPG-fueled vehicles was implemented by the Hong Kong govern-ment in September 2013. Long-term real-time measurements indicated that the program was remarkably effective in reducing LPG-related VOCs, NO x and nitric oxide (NO) in the atmosphere. Receptor modeling results further revealed that propane, propene, i -butane, n -butane and NO in LPG-fueled vehicle exhaust emissions decreased by 40.8 ± 0.1, 45.7 ± 0.2,


Introduction
Vehicular exhaust is a major source of air pollutants such as particulate matter (PM), volatile organic compounds (VOCs) and trace gases including carbon monoxide (CO), carbon dioxide (CO 2 ), sulfur dioxide (SO 2 ) and nitrogen oxides (NO x ) (Xie et al., 2003;Barletta et al., 2002;Ruellan and Cachier, 2001).Apart from primary pollutants, secondary pollutants formed from vehicular exhausts, i.e., ozone (O 3 ) and secondary organic aerosols (SOAs), are of high concern due to their detrimental effects on atmospheric environment and human health (Ciccone et al., 1998).
Fuel substitution is a universal and effective method to constrain vehicular emissions.As relatively clean energy, liquefied petroleum gas (LPG), has been widely used either alone or combined with petrol and diesel (Lau et al., 2011;Lai et al., 2009;Gamas et al., 1999), for the benefit of emission reduction of CO 2 , fine PM and VOCs (Chikhi et al., 2014;Myung et al., 2012).However, one must bear in mind that maintenance status, conversion method, composi-tion and driving speed all influence the emissions of air pollutants (Schifter et al., 2000;Gamas et al., 1999).Therefore, regular on-site measurements are necessary to monitor traffic emission profiles and rates.Although LPG is a relatively clean fuel, the emission of LPG-related VOCs (i.e., propane, propene and nand i-butanes) from evaporation and incomplete combustion cannot be neglected.For example, source apportionment analysis indicated that LPG accounted for a considerable percentage of ambient VOCs in Guangzhou (8-16 %) (Liu et al., 2008) and Hong Kong (32.6 ± 5.8 %) (Ou et al., 2015).Furthermore, the LPG-related VOCs can contribute to O 3 formation and cause photochemical smog.Blake and Rowland (1995) emphasized the importance of LPG leakage in OH reactivity and O 3 formation in urban Mexico.Farmer et al. (2011) reported that the increase of LPG-related VOCs led to O 3 increments even though the total VOCs (TVOCs) were reduced, as the peroxyl radicals (RO 2 ) generated from VOCs with low vapor pressure had a higher branching ratio leading to NO 2 formation from NO, rather than reserving nitrogen in the form of organic nitrates.On the other hand, NO emitted from LPG combustion may titrate O 3 .Therefore, it still remains undecided as to whether the LPG usage is contributive or destructive to O 3 formation, especially in Hong Kong.
Hong Kong has experienced rapid development during the past decades, and increasingly suffers from crowded traffic and population, with nearly 0.7 million vehicles traveling on the total road length of 2.1 × 10 3 km and 7.2 million people living in a territory 1.1 × 10 3 km 2 in size.The road usage rate in Hong Kong is among the highest in the world, transporting approximately 1.2 million passengers per day.As such, traffic emissions significantly influence the air quality in Hong Kong.Apart from local emissions, air quality in Hong Kong is also affected by regional transport.In autumn and winter, prevailing northerly and northeasterly winds bring anthropogenic emissions from mainland China to Hong Kong, enhancing local pollution burden.In summer, clean oceanic air masses dilute the levels of air pollutants under the influence of southerly winds (Ling et al., 2013;Guo et al., 2009;Wang et al., 2005;So and Wang, 2003;Chan and Chan, 2000).Ling et al. (2013) quantified the contribution of regional transport as 0-61 % for summer O 3 and 0-56 % for autumn O 3 on high O 3 days in Hong Kong.Guo et al. (2006) found that over 45 % of VOC samples collected from August 2001 to December 2002 at Tai O, Hong Kong, were influenced by regional transport from inland China.Based on numerical simulations, Wang et al. (2006) quantified regional contributions of 40-90 % to O 3 formation in Hong Kong, and Huang et al. (2006) indicated that the regional contribution was < 50 % when a trough was over the South China Sea (SCS), and it increased to > 50 % in the presence of an anticyclone over mainland China and/or a tropical cyclone over SCS.Furthermore, Wang et al. (2006) revealed that the regional contribution was lower during daytime than that at night, while Lam et al. (2005) found that the contribution decreased from > 50 % during an O 3 episode to 30-50 % after the episode.In addition, Jiang et al. (2008) reported that the contribution of horizontal transport to surface O 3 in Hong Kong was negligible during a typhoon-induced O 3 episode.However, all the studies focused on the regional contribution at non-roadside sites, and the O 3 chemistry at roadside sites in Hong Kong was seldom reported.Given high NO x levels, tall buildings and low wind speeds along the main roads in Hong Kong, it is expected that the regional impact on air pollutants in a roadside environment is relatively minor, especially during daytime hours on non-O 3 episode days.
To improve the air quality, the Hong Kong Environmental Protection Department (HKEPD) launched a series of programs including a stepwise LPG replacement program for taxis and light buses from 1997.By the end of 2010, > 99 % of the registered taxis (i.e., 17 950 vehicles) and 51 % of the registered public and private light buses (i.e., 3280 vehicles) were powered by LPG (HKCSD, 2010).Due to the usage of LPG as vehicle fuel, the ambient VOC profile in Hong Kong changed substantially, with the most abundant species switching from toluene to propane and n/ i-butanes (Ou et al., 2015;Guo et al., 2007;Ho et al., 2004).In addition, studies consistently confirmed an increased contribution of LPG-fueled vehicle emissions to ambient VOC levels, which was 11 % at an urban site in 2001 (Guo et al., 2004), 15 % at a suburban site during 2002-2003(Guo et al., 2007) ) and 26.9 ± 0.6 % and 40.8 ± 0.8 % at an urban site during 2002-2003and 2006-2007, respectively (Lau et al., 2010)).Moreover, O 3 has been experiencing an increasing trend in Hong Kong over the last 2 decades (Wang et al., 2009;Guo et al., 2009).As O 3 formation was generally VOC-limited in Hong Kong (Cheng et al., 2010;Zhang et al., 2007), the contribution of LPG-fueled vehicle emissions to O 3 was not negligible, in view of the dominance of propane and n/ i-butanes in ambient air and the high contribution of LPG-fueled vehicle emissions to ambient VOC levels.As such, an intervention program which aimed to reduce VOCs and NO x emitted by LPG-fueled vehicles was initiated in September 2013.This subsidy program promoted the replacement of aging catalytic converters in LPG-fueled vehicles, and the catalytic converters in ∼ 75 % of LPG-fueled vehicles were replaced by the end of May 2014.Although laboratory studies identified a high removal efficiency of the new catalytic converter on VOCs and NO x emissions (Table S1 in the Supplement), the impact of the program on roadside air quality, which is more complex, required investigation.
In this study, the effectiveness of the intervention program on the reduction of LPG-related VOCs and NO x , on O 3 production and on the photochemical reactivity driving the O 3 formation was evaluated.Furthermore, the sensitivity of O 3 variation (i.e., the difference of net O 3 production by LPG before and during the program) relative to the reduction percentage of LPG-related VOCs and NO x was studied, and subsequently a maximum NO x reduction percentage and a reasonable reduction ratio of VOCs / NO x were proposed for achieving the best O 3 reduction.

Site description
Hong Kong, located on the coast of the South China Sea (SCS), constitutes the greater Pearl River Delta (PRD) together with Macao and nine mainland cities in Guangdong province (i.e., Guangzhou, Shenzhen, Zhuhai, Foshan, Dongguan, Zhongshan, Huizhou, Jiangmen and Zhaoqing).Hong Kong has a typical coastal hilly terrain, with a large area of water (∼ 60 %) and large vegetation coverage (∼ 70 % of land area).A marine subtropical monsoon climate dominates, and the prevailing winds are north and northeast in winter, east in spring and autumn and south and southwest in summer.
This study involves three roadside sites (Mong Kok (MK, 22.32 • N, 114.17 HT is a remote site at the tip of southeastern Hong Kong, ∼ 25 km from the urban core.It receives air masses from SCS in summer and from continental areas including Hong Kong and PRD in autumn/winter.Therefore, air masses at HT are a mix of marine and continental background air.Indeed, HT has been serving as a background site and is extensively used for air quality studies in Hong Kong (Wang et al., 2009;Ho et al., 2006;Lee et al., 2002).Figure 1 shows the geographical location of the sampling sites.

Trace gases
Five trace gases (i.e., CO, SO 2 , NO, NO 2 and O 3 ) were continuously measured at all the sites from 2011 to 2014.The analyzers were deployed in the air quality monitoring stations, with a height of 3.0 m (MK and CWB), 4.5 m (Central), 11.0 m (TM) and 20.0 m (HT) above ground level.Air samples were drawn through a perfluoroalkoxy (PFA) Teflon tube with an outside diameter (OD) of 12.7 mm and an inside diameter (ID) of 9.6 mm.The inlet of the tube was approximately 0.5 m above the rooftop of the monitoring station, and the outlet was connected to a PFA manifold with a bypass pump, drawing air into the analyzers at a rate of 5 L min −1 .Details about the analyzers are provided elsewhere (HKEPD, 2014;So and Wang, 2003).

VOCs
Thirty C 2 -C 10 non-methane hydrocarbons (NMHCs), including 11 alkanes, 10 alkenes and 9 aromatics, were continuously measured at MK and HT from 2011 to 2014.The VOC samples were collected and analyzed every 30 min with an online analytical system (Syntech Spectra GC 955, Series 600/800, the Netherlands).The data were averaged into hourly values.The detection limits varied by VOC species and were in the range of 0.002-0.787ppbv.
Table S2 summarizes average mixing ratios of the main VOCs and trace gases, and their OH reactivity at the roadside (MK) and background site (HT) before (October 2012-September 2013) and during the program (October 2013-May 2014) in Hong Kong.The much higher OH reactivity at MK (before: 86.6 ± 6.0 s −1 ; during: 77.9 ± 6.1 s −1 ) than at HT (before: 5.0 ± 0.3 s −1 ; during: 4.9 ± 0.1 s −1 ) meant that the OH lifetime was much shorter at MK (∼ 0.01 s) than at HT (0.2 s), and even shorter than that at an urban site in New York (0.05 s) (Ren et al., 2003), indicating that reactive species were extremely abundant in the roadside environment of Hong Kong.The LPG-related VOCs (i.e., propane and n/ i-butanes) contributed 56.5 ± 2.7 and 46.0 ± 3.1 % to total VOCs, and 18.7 ± 1.2 and 15.1 ± 1.1 % to the total OH reactivity of VOCs at MK, before and during the program, respectively.These fractional contributions suggest the importance of LPG-related VOCs in the budgets of VOCs and O 3 in Hong Kong.Consistent with the decreases of fractional contributions once the program started, the mixing ratios (29.4 ± 3.3 and 21.0 ± 2.2 ppbv for before and during the program, respectively) and OH reactivity (1.12 ± 0.04 and 0.79 ± 0.02 s −1 before and during the program, respectively) of LPG-related VOCs significantly declined at MK (p < 0.05), while at HT, their mixing ratios increased from 1.5 ± 0.1 to 1.9 ± 0.04 ppbv, and the OH reactivity increased from 0.06 ± 0.004 to 0.07 ± 0.003 s −1 .The minor variation of background LPG-related VOCs relative to the decreases at MK (4.8 % in mixing ratios and 3.0 % in OH reactivity of VOCs) indicated that the influences of the background concentrations on the variations of LPG-related VOCs and O 3 at MK were minimal and can be neglected.

Quality assurance and control (QA/QC)
To guarantee the quality of the data acquired from the online monitoring systems, the instruments and QA/QC procedures for trace gases measurements were identical to those in the US air quality monitoring program (http://epic.epd.gov.hk/ca/uid/airdata).The measurements have been accredited by the Hong Kong Laboratory Accreditation Scheme (HOK-LAS).All the instruments were regularly calibrated, tested and audited by standards with known traceability for measurements of trace gases.According to HOKLAS criteria, a quality system was established to control the accuracy and precision within the limits of ±15 and ±20 %, respectively (HKEPD, 2014).For VOC analysis, a built-in computerized program, including auto-linearization, auto-calibration and calibration with span gas, was adopted to control the quality.The accuracy and precision of VOC measurements were 1.0-10 % and 2.5-20 %, respectively.The accuracy was based on weekly span checks, monthly calibration and annual auto-linearization using the National Physical Laboratory (NPL) span gas.The precision was based on quarterly precision check results (the 95 % probability limits for the integrated precision based on weekly precision check results of the latest 3 months).Moreover, the online measurements of VOCs were regularly compared with whole-air canister samples collected and analyzed by University of California at Irvine (UCI).Good agreements were identified for the alkanes (e.g., R 2 = 0.95 and 0.85, slope = 1.14 and 0.97 for propane and butanes, respectively), while the agreements for the more reactive alkenes and aromatics were also reasonable (e.g., R 2 = 0.64 and 0.94, slope = 1.34 and 0.86 for propene and toluene, respectively).

Primary emissions of VOCs and NO x
The measured concentrations of VOCs and trace gases were composed of the backgrounds, primary emissions and secondary formation (applicable to O 3 and NO 2 ).To obtain the concentrations of LPG-related VOCs and NO x emitted from primary sources at the roadside sites, the background concentrations and the concentrations elevated by the secondary formation (if applicable) in this study were excluded using Eqs.(1)-( 5) (Takekawa et al., 2013): (1) (2) (3) where [xx]  (3) 8 h averages of O 3 and NO x showed good correlations (R 2 = 0.75 and 0.57 for O 3 and NO x , respectively) between TM and HT, and not all the data were available at HT during the entire study period.

Impact of regional and super-regional air masses
Since MK was the only roadside site with VOC and trace gas data, this study mainly focused on the data analysis of this site.As a roadside site, MK was expected to be strongly influenced by fresh vehicular emissions.VOC ratios can indicate the relative ages of air masses and regional transport of air pollutants.For example, the ratios of VOCs with higher reactivity to those with lower reactivity (e.g., toluene / benzene and xylenes / ethylbenzene) imply more fresh air when values are higher, which has been extensively used in previous studies (Guo et al., 2007;Ho et al., 2004).Therefore, ratios of toluene / benzene and xylenes / ethylbenzene were calculated here to roughly estimate the age of air masses at MK. Table 1 summarizes the ratios at MK (roadside site), Tung Chung (TC) and Yuen Long (YL) (both general ambient sites) and HT (background site) in Hong Kong from October 2012 to May 2014.It is noteworthy that ambient VOCs at all the four sites were simultaneously measured.The ratios of toluene / benzene and xylenes / ethylbenzene at MK were significantly higher than those at the other three sites (p < 0.05).Furthermore, these ratios at MK were even higher after the background values were deducted.The results suggested that air masses at MK were dominated by fresh emissions.However, regional and super-regional transport cannot be completely eliminated only with the VOC ratios, and the influence of regional/super-regional air should be further examined when studying the emissions of local LPG-fueled vehicles.Using the propane / CO ratio method proposed by Guo et al. (2006), i.e., the ratio range of 5.0-300 pptv ppbv −1 for air masses in Hong Kong, the influence of regional/superregional air masses on local air was evaluated.Figure 2 shows the ratios of propane / CO at MK from 2011 to 2014.The propane / CO ratios of ∼ 96 % air masses were between 5.0 and 300 pptv ppbv −1 .Further inspection of the values of LPG-related VOCs and NO x with ratios of 5.0-300 showed insignificant differences (p > 0.1) to those with ratios beyond 5-300, suggesting that the air masses at MK were seldom influenced by regional/super-regional air.Moreover, wind speeds were lower than 2.0 m s −1 for more than 95 % of the time (i.e., calm or light air according to the Beaufort Wind Scale) (http://www.spc.noaa.gov/faq/tornado/beaufort.html), and the values of LPG-related VOCs and NO x in all air masses were not different to those in the air masses with wind speeds lower than 2.0 m s −1 (p > 0.1), confirming a minor impact of regional/super-regional pollution on air at MK. Indeed, the prevailing wind direction was from the east, where local urban areas are located.

PMF model
Positive matrix factorization (PMF) is a receptor model for source apportionment and has been extensively used in many fields (i.e., PM, VOCs and sediment) (Brown et al., 2007;Lee et al., 1999).Based on the principle of mass balance, it decomposes the matrix of measurement (X) into the matrices of factor contributions (G) and factor profiles (F) in p sources, as shown in Eq. ( 6) (Paatero, 1997;Paatero and Tapper, 1994): where x ij is the measured concentration of j species in i sample, g ik represents the contribution of kth source to i sample, f kj indicates the fraction of j species in kth source and e ij is the residual for j species in i sample.
The matrices of G and F are obtained in the case of the minimum of objective function Q, as shown in Eq. ( 7): where m and n represent the number of species and samples, respectively, and u ij is the uncertainty of j species in i sample.Q (robust) is automatically calculated by excluding the points not fit by the model, and the run with lowest Q (robust) value is selected as the optimum solution by the model.

PBM-MCM model
The photochemical box model incorporating the master chemical mechanism (PBM-MCM) is a computing model based on the chemical reactions of observed species.It has been introduced and successfully applied in many previous studies in reproducing the observed values, describing the production and destruction mechanisms and simulating the photochemical reactivity (Ling et al., 2014;Lam et al., 2013;Cheng et al., 2010).Briefly, the observations of VOCs, trace gases and meteorological parameters are input to construct and constrain the model, which consists of 5900 reactions and 16 500 species in the latest version (MCM 3.2).However, it is noteworthy that the heterogeneous reactions are not included, and the model does not consider the transport processes (i.e., horizontal and vertical transport).
In this study, CO, SO 2 , NO, NO 2 , O 3 , 27 VOCs and two meteorological parameters (i.e., temperature and relative humidity) were used to construct and constrain the model.Hourly data were used as input (i.e.24 h per day) and the outputs were from 07:00 to 19:00 by each hour.To better describe the photochemical reactions in Hong Kong, the model was modified; i.e., the photolysis rates were calibrated using the photon fluxes from the Tropospheric Ultraviolet and Visible Radiation (TUV-v5) model (Madronich and Flocke, 1997), according to the location of Hong Kong and the modeling period, and the height of the mixing layer was set as 300-1400 m.The concentrations of some species in the free troposphere, which may influence the mixing ratios of air pollutants in the lower troposphere with the development of boundary layer, were set according to the real conditions in Hong Kong (Lam et al., 2013  A base case was established to simulate O 3 and the photochemical reactivity (i.e., the formation and loss pathways of OH and HO 2 listed in Table 2).The observed VOCs and trace gases were input to construct the base case, and the simulated O 3 was compared with the observed levels to validate the model and check the influence of regional transport during daytime hours.At low wind speed (1.3 ± 0.01 m s −1 ) and high NO x (223.2 ± 1.8 ppbv) at MK, the majority of regional O 3 would be completely consumed before being detected by measurement instruments.In fact, the PBM-MCM model considered the regionally transported concentrations of O 3 precursors, as the observed concentrations input into the model included the regional fractions.Moreover, since O 3 levels were higher at night (8.1 ± 0.2 ppbv) than those during daytime (5.9 ± 0.1 ppbv) at MK, it suggests the existence of nocturnal regional transport.Hence, the observed O 3 at 07:00 was input into the model to initiate the simulation for daytime hours; i.e., the nocturnal regional transport of O 3 was also considered.
To simulate the contribution of LPG source to O 3 formation, the following approach was adopted to avoid altering the real ambient environment for atmospheric chemistry.Briefly, the concentration of species X in the whole air and in the assigned LPG source was defined as X whole air and X LPG , respectively.Hence, the concentration of X whole air − X LPG was the input for the constrained cases, while X whole air was the input for the base case.The three constrained cases were as follows: (i) only VOCs in the LPG source were excluded from the whole air; (ii) only NO x in the LPG source was excluded; and (iii) both VOCs and NO x were removed simultaneously.As such, the differences of the outputs between the base case and the three constrained cases were the contributions of (i) VOCs, (ii) NO x and (iii) VOCs + NO x in the LPG source to the O 3 formation, respectively.The changes of these contributions from the period before the program to during the program represented the impact of the intervention program on O 3 formation.Table S3 in the Supplement lists the configurations of the model input for the base case and the three constrained cases.

Relative incremental reactivity (RIR)
RIR, initially proposed by Carter and Atkinson (1989), has been extensively used to describe the relationship between O 3 and its precursors, i.e., VOCs, NO x and CO.Although the observation-based models incorporating carbon bond mechanisms were often utilized to simulate the O 3 production rate (Ling et al., 2011;Zhang et al., 2007;Martien et al., 2003), a more explicit PBM-MCM model was applied in this study.The RIR and RIR (weighted concentration) were calculated using Eqs.( 8) and ( 9) (Ling et al., 2011): where P S O 3 −NO (X) and P S O 3 −NO (X − X) represent the original O 3 production rate, and that in the scenario, with a hypothetical change ( X) (10 % in this study) in source/species X, respectively, both of which considered O 3 titration by NO.S(X) is the change in the concentration of X (S(X)).The concentration refers to the observed or PMFextracted concentration of source/species X.Table 3 shows the mixing ratios of the primarily emitted LPG-related VOCs and NO x before and during the program.
The LPG-related VOCs and NO x were reduced significantly from the period before the program to during the program (p < 0.05).To investigate the meteorological influences on these reductions, we looked into the average geopotential height (HGT) and wind field on 1000 hPa for East Asia for the matched period and the period during the program, as shown in Fig. S1 in the Supplement (the matched period (October 2012-May 2013) was defined as the same time span as that during the program, but in different years.)Noticeably, the meteorological conditions were fairly similar between the two periods; i.e., the pressure decline (indicated by the decrease of HGT) from northern China to SCS led to the prevailing northeasterly winds in Hong Kong, and the differences of HGT and wind speed between the two periods for Hong Kong were only 0-2 gpm and less than 1 m s −1 , respectively.Ground monitoring data also indicated insignificant differences of temperature (matched: 21.9 ± 0.5 • C; during: 21.2 ± 0.7 • C) between the two periods (p > 0.05).However, the mixing ratios of LPG-related VOCs and NO x during the matched period were comparable to those before the program, and absolutely higher than those during the program (p < 0.05) (see Table 3).Given the similar meteorological conditions between the matched period and during the program, the significant decreases of LPG-related VOCs and NO x were caused by the interventional program, rather than meteorological variations.
To further understand the effectiveness of the program, the monthly reductions of LPG-related VOCs after the commencement of the converter replacement were calculated and compared with those before the program (Fig. 3).Since the program was initiated in October 2013, the averages of VOC species in September in each year were taken as the base-lines for the calculation of monthly reduction.Briefly, the monthly reductions of LPG-related VOCs were the differences between their averages in September and those in the other months.Hence, positive and negative values indicate reductions and increments of the corresponding species, respectively.It was found that the monthly averages of LPGrelated VOCs consistently decreased from September 2013 to May 2014 except for n-butane in October 2013 when the program was just initiated.Compared to those before the program (i.e.September 2013), the mixing ratios of propane, i-butane and n-butane decreased 3.2 ± 0.2, 2.8 ± 0.2 and 4.9 ± 0.2 ppbv by May 2014, respectively, when 99.2 % of catalytic converters participating in the program had been exchanged.Furthermore, the monthly reductions correlated well with the cumulative converter replacements (R 2 = 0.92, 0.93 and 0.89 for propane, i-butane and n-butane, respectively).In contrast, no consistent reduction was observed from September 2012 to May 2013 for LPG-related VOCs, suggesting the effectiveness of the program on the reduction of LPG-related VOCs.On the other hand, although the average mixing ratios of NO x decreased significantly during the program as shown in Table 3, no consistent reductions were found for their monthly averages.This might be due to the fact that NO x emitted from LPG-fueled vehicles was minor compared to gasoline-and diesel-fueled vehicles (∼ 4.0 % from emission inventory, and 1.1-7.3% from source apportionment.See Sect.3.2.2).

Temporal variations of primary LPG-related
VOCs and NO x The abnormally low alkane levels in late June 2013 might be caused by the anticyclone over southern China and a tropical storm over the SCS, leading to extremely high temperature on 18-20 June (i.e., favorable for photochemical reactions), and stronger winds (i.e., conducive to atmospheric dispersion) on the following days, respectively.In contrast, Hong Kong was strongly influenced by an active ocean flow and a low pressure trough over the SCS in late July 2013.The consequently heavy rain and low temperature suppressed the photochemical reactions and caused the unusually high levels of VOCs (Fig. S2 in the Supplement illustrates these events).Generally, the LPG-related VOCs experienced a significant (p < 0.05) reduction throughout the year.The daily changing rates of LPG-related VOCs in two periods, i.e., September-December and January-May, in different years at MK are shown in Table 4.It is noteworthy that the two periods were selected based on data availability each year, and the minimized influence of meteorological parameters in the same month of different years.The LPG-related VOCs decreased from September to December in 2011 and 2013, but there was no significant difference in 2012.In general, the levels of VOCs in the atmosphere are associated with source emissions, photochemical reactions and regional transport.As the regional influence was excluded in this study, source emissions and photochemical reactions became the main factors determining the ambient concentrations of VOCs.The decreasing and unchanged trends in 2011 and 2012 respectively might be related to the integrated influence of reduced photochemical degradation (which increases ambient VOC levels) and temperature decrease (which reduces evaporative emissions of VOCs) from September to December.It is noticeable that the LPG-related VOCs had the highest decreasing rates in September-December 2013, compared to those in previous years.The higher decreasing rates of LPGrelated VOCs in September-December 2013 implied the possible effectiveness of the interventional program on VOCs reduction.For the period of January-May, the LPG-related VOCs increased in January-May 2013, except for propane which did not show significant change from January to May.However, significant decreasing trends (p < 0.05) were observed for propane and n/ i-butanes during the same period in 2014, which was likely owing to the continuous replacement of catalytic converters on LPG-fueled vehicles.By comparison, the reduction rates of LPG-related VOCs in January-May 2014 were all lower than those in September-December 2013.This was mainly attributable to the fact that a large portion (i.e.∼ 69 %) of the converter replacements were completed by the end of December 2013, and there were much fewer replaced converters (i.e.∼ 31 %) in January-May 2014.
Figure 5 shows the temporal variations of primary NO x , NO, NO 2 and secondary NO 2 at MK from June 2013 to May 2014.The primary NO x and NO mixing ratios decreased significantly (p < 0.05), implying the possible effectiveness of the intervention program on NO / NO x reduction.However, the secondary NO 2 levels increased, while there was no significant change (p > 0.05) for primary NO 2 .To explore the reasons of NO-NO 2 -NO x variations, the changing rates of primary NO x , NO, NO 2 and secondary NO 2 during the same period (i.e., June in the previous year to May in the following year) in different years were compared.Table 5 shows the statistics of changing rates of NO x -NO-NO 2 at the roadside sites (i.e., MK, CWB and Central) from June 2011 to May 2014.Taking MK as an example, the primary NO x and NO decreased much faster (i.e., 91 and 94 pptv day −1 , respectively) from June 2013 to May 2014 than during the same period in 2011-2012 and 2012-2013, suggesting the effectiveness of the program in reducing NO x and NO.Compared to the decreasing rate in June 2012-May 2013 (i.e., 13 pptv day −1 ), no significant change (p > 0.05) was observed for primary NO 2 from June 2013 to May 2014.However, it cannot be concluded that the program caused the increase of primary NO 2 , in view of the same insignificant variation in June 2011-May 2012 and the fact that LPG-fueled vehicles emit negligible NO 2 (see Sect. 3.2.2).In contrast, secondary NO 2 significantly increased (with a rate of 13 pptv day −1 ) from June 2013 to May 2014, which was also observed in June 2011-May 2012 (i.e., increasing rate of 20 pptv day −1 ).Since secondary NO 2 is formed by NO reacting with O 3 , inspection of the O 3 production would provide more comprehensive interpretation on the secondary NO 2 increment during the program.Similar variations were observed at CWB and Central, where primary NO x and NO showed significant decreasing trends, whereas the variations of primary and secondary NO 2 were insignificant (p > 0.05).
In summary, the LPG-related VOCs were significantly lowered during the program with monthly reductions of 3.2 ± 0.2, 2.8 ± 0.2 and 4.9 ± 0.2 ppbv for propane, i-butane and n-butane by the end of this study, respectively.Continuous decreasing trends were observed for LPG-related VOCs, and the reduction rates were almost unprecedented, e.g., 39 pptv day −1 for n-butane.Furthermore, the mixing ratios of NO and NO x decreased as well during the program.The reduction rates of NO and NO x during June 2013-May 2014 at the three roadside sites were much higher than those in previous years.Overall, the field measurement data indicated that the program was effective in reducing emissions of LPGrelated VOCs and NO x .

Source identification
To investigate the change of the contributions of LPG-fueled vehicles to VOCs and NO x , the online data of 15 VOCs and three trace gases at MK before (i.e., 8753 samples during October 2012-September 2013) and during the intervention program (i.e., 5833 samples during October 2013-May 2014) were separately applied to PMF for source apportionments.It is noteworthy that the whole-air ambient concentrations of VOCs and trace gases, rather than those with the backgrounds deducted were used here, in order to keep consistency with the simulations of whole-air ambient O 3 and photochemical reactivity.Figure 6 shows the source profiles of the four factors that best reproduced the concentrations of the input species before and during the intervention program, respectively.The first factor had high loadings of C 3 -C 5 hydrocarbons and toluene, and was dominated by CO, NO and NO 2 , which are all main components of vehicle exhaust (Guo et al., 2011a;Ho et al., 2009).The n/ ipentanes and toluene indicated emissions from gasoline vehicles, while the high NO x loading was a signature of diesel exhaust.Since propane and n/ i-butanes were not prominent, this factor was assigned as gasoline and diesel-fueled vehicle exhaust.Factor 2 was closely associated with LPG-fueled vehicle exhaust, with the dominance of propane, n/ i-butanes and propene, the major components and combustion product of LPG (Guo et al., 2011b;Blake and Rowland, 1995).Factor 3 was distinguished by NO 2 and the long-lived species, i.e., ethane, ethyne, benzene and CO.The long-lived species were the indicators of aged air masses, in which NO 2 was accumulated due to photochemical reactions.Hence, this factor represented aged air masses.Indeed, this profile highly coincided with the aged air in Hong Kong identified by Lau et al. (2010).The last factor explained most of the TEX compounds (toluene / ethylbenzene / xylenes), the tracers of solvent usage (Guo et al., 2007;Borbon et al., 2002).Therefore, factor 4 was identified as solvent usage.

Source contribution
Based on the VOC loadings in each source, Table 6 summarizes the mass and percentage contributions of the sources to VOCs at MK. LPG-fueled vehicle exhaust was the greatest contributor to VOCs at MK, with the mass contribution of 114.2 ± 0.1 µg m −3 (51.51 ± 0.05 %) and 64.8 ± 0.1 µg m −3 (39.07 ± 0.05 %) before and during the program, respectively.These contributions were higher than those quantified at other sites in Hong Kong, i.e., suburban TC (32.6 ± 5.8 %) (Ou et al., 2015), urban site Tsuen Wan (TW) (21 ± 2 %) (Ling and Guo, 2014) and suburban YL (15 %) (Guo et al., 2007).It is noteworthy that factors such as study period, chemical species, source profiles and models used all influence the source apportionment.Bearing these factors in mind, the higher contributions of LPG-fueled vehicle exhaust in this study were likely due to the fact that MK was a roadside site closer to the emission sources.On the other hand, though gasoline/diesel vehicles emitted considerable VOCs (i.e., 60.5 ± 0.1 and 56.8 ± 0.2 µg m −3 before and during the program, respectively), they were significantly lower than those emitted by LPG-fueled vehicles (p < 0.05), particularly before the replacement of the catalytic converters.While LPG-fueled vehicles only accounted for a small number of the registered vehicles (∼ 3.1 %) in Hong Kong (Table S4 in the Supplement), it was not unreasonable that LPG   made the highest contribution to VOCs, in view of low vapor pressure of the LPG component and high-density flow of the LPG-fueled vehicles at MK. Apart from vehicular exhaust, the aged VOCs and solvent usage also contributed ∼ 24 % to VOCs at MK.
Table 7 lists the reduction of individual species from LPGfueled vehicles due to the intervention program.Propane (before: 21.15 ± 0.02 µg m −3 ; during: 12.53 ± 0.01 µg m −3 ), n-butane (before: 57.96 ± 0.01 µg m −3 ; during: 30.27 ± 0.02 µg m −3 ) and i-butane (before: 27.20 ± 0.02 µg m −3 ; during: 17.50 ± 0.01 µg m −3 ) all decreased remarkably throughout the study period (p < 0.05).Meanwhile, the concentrations of CO and NO were also reduced from 248.4 ± 1.3 and 18.2 ± 0.1 µg m −3 to 228.9 ± 0.6 and 2.08 ± 0.04 µg m −3 , respectively.However, NO 2 apportioned in LPG exhaust was extremely minor (i.e., 0.6 ± 0.2 µg m −3 and nil before and during the program, respectively), which might explain the insignificant decrease or even increase of NO 2 .During the study period, the LPGrelated VOCs and NO increased by ∼ 1.4 and ∼ 2.9 µg m −3 , respectively, at the background site, only respectively accounting for 3.0 % and 1.3 % of the decreased LPG-related VOCs (∼ 46.1 µg m −3 ) and NO (∼ 226.8 µg m −3 ) in LPGfueled vehicle exhaust.Therefore, it is believed that the re- To quantitatively evaluate the performance of the model, the index of agreement (IOA) was introduced to test the agreement between the simulated and observed O 3 .Ranging from 0 to 1, higher values of IOA indicate better agreement between the simulation and observation.The IOA was calculated using Eq. ( 10) (Huang et al., 2005): where O i and S i are the hourly observed and simulated values, and Ō represents the average of observations.In this study, the IOA reached 0.75, and the accuracy of the simulation was 16.7 ± 2.1 %, suggesting good performance of the model in O 3 simulation.Bearing in mind the uncertainty of the model, the good agreement between the simulated and observed O 3 implied a minor regional contribution at this roadside site.3.4 Photochemical reactivity 3.4.1 OH, HO 2 and their formation/loss rates in whole air As the "detergent" of atmosphere, OH initiates the oxidation of air pollutants including VOCs, CO and NO x , leading to O 3 formation, and the cycling between OH and HO 2 accelerates the propagation of the chain reactions.Thus, the budget of OH and HO 2 is an important parameter of a photochemical system.Figure 9 presents the average daytime patterns of OH and HO 2 before and during the program.Typical bellshaped patterns were observed for OH and HO 2 , which increased from morning to noon, peaked at 12:00-13:00 and then decreased to low levels in the afternoon.The average simulated concentrations of OH and HO 2 were 1.64 ± 0.78 × 10 5 and 2.49 ± 1.28 × 10 5 molecules cm −3 before the program, and 1.80 ± 0.85 × 10 5 and 4.18 ± 2.03 × 10 5 molecules cm −3 during the program, respectively.Compared to those modeled at an urban site in Hong Kong (2.3-3.6 × 10 6 molecules cm −3 for OH and 3.4-4.4× 10 8 molecules cm −3 for HO 2 ) (Ling et al., 2014), and those measured at a VOC-rich site in PRD (15 × 10 6 and 18 × 10 8 molecules cm −3 for OH and HO 2 , respectively) (Lu et al., 2012;Lou et al., 2010), the OH and HO 2 levels were much lower at MK.This is not surprising because much OH and HO 2 were consumed by high levels of VOCs and NO x at the roadside MK (Fig. S3 in the Supplementary material shows the simulated OH and HO 2 at different sites with the PBM-MCM model).In addition, this study covered different seasons of the study years, rather than the O 3 episodes in Ling et al. (2014) and summer only in Lu et al. (2012) and Lou et al. (2010) when photochemical reactivity was usually stronger.Apart from the increase of OH and HO 2 from the period before the program to during the program, the ratio of HO 2 / OH was significantly higher during the program (i.e., 3.7 ± 0.5) than that before (i.e., 1.9 ± 0.3) (p < 0.05), which might partly explain the increase of O 3 throughout the study period.Nevertheless, since the OH, HO 2 and HO 2 /OH were simulated using the whole-air concentrations of O 3 precursors, whether their increases were mainly caused by the intervention program requires further investigation, as shown below.
Since the changes in the formation/loss rates from the period before the program to during the program  were minor for each pathway of OH and HO 2 , Fig. 10 presents their average pathway-specific formation and loss rates over the whole study period.The total formation/loss rates of OH and HO 2 were 5.8 ± 2.4 × 10 6 and 2.6 ± 1.0 × 10 6 molecules cm −3 s −1 , respectively.The reaction of HO 2 with NO, HONO photolysis, O 3 photolysis and ozonolysis of alkenes were the main sources of OH, with contributions of 56.7 ± 11.6, 31.7 ± 10.7, 6.6 ± 3.3 and 5.0 ± 1.7 %, respectively.On the other hand, OH was consumed by reaction with NO (36.5 ± 5.5 %), NO 2 (35.1 ± 4.6 %), VOCs (14.8 ± 1.2 %) and CO (13.6 ± 1.0 %).For HO 2 formation, the reaction between RO 2 and NO was the most predominant pathway (54.8 ± 8.8 %), followed by the reaction of OH with CO (23.3 ± 3.9 %), HCHO photolysis (13.1 ± 4.6 %), ozonolysis of alkenes (7.9 ± 2.1 %) and the reaction of OH with HCHO (0.8 ± 0.2 %).Meanwhile, HO 2 was almost exclusively consumed by reacting with NO.

Influence of the program on photochemical reactivity
The differences between the simulation outputs of the scenarios with and without LPG VOCs and NO x inputs determined the contributions of LPG-fueled vehicle exhaust to the formation/loss rates of OH and HO 2 , as summarized in Table 8.Generally, the formation/loss rates of OH and HO 2 contributed by the LPG source decreased from the period before the program to during the program, which was caused by the reduction of VOCs and NO x in the LPG source.Furthermore, since the sum of the formation rates was lower than the loss rates for OH for both before and during the program, but for only HO 2 before the program, the net effects of the LPG source to OH and HO 2 were destructive.However, the HO 2 formation rate exceeded its loss rate for the period of during the program, indicating a net production of HO 2 by the LPG source.
Figure 11a shows the net effects of the LPG source on the production of OH and HO 2 .From before to during the program, the destruction rate of OH decreased, while the destruction of HO 2 switched to production.These variations led to the increases of OH and HO 2 from the period before the program to during the program, as shown in Fig. 11b.Different from the increases of OH and HO 2 in the whole air as shown in Fig. 9, the increases here were caused by the intervention program.The OH and HO 2 levels increased by 6.9 ± 1.1 × 10 3 molecules cm −3 and 3.4 ± 1.2 × 10 4 molecules cm −3 , respectively.The higher increase of HO 2 than OH led to a higher ratio of HO 2 / OH during the program, resulting in a consequent O 3 increment.
3.5 Improvement of the intervention program 3.5.1 O 3 -VOCs-NO x sensitivity in the whole air O 3 -VOCs-NO x sensitivity can reflect the O 3 variation relative to the change of VOCs and NO x , from which VOClimited regimes and NO x -limited regimes in O 3 formation can be indicated.Figure 12a and b show the O 3 -VOCs-NO x sensitivity in the air at MK between 40 and 100 and 0.5 and 40 % of the observed average NO x , respectively.The ranges were selected according to the O 3 levels and behaviors responding to the variations of VOCs and NO x (details were given below).This sensitivity diagram was obtained with the aid of the PBM-MCM model, and based on the average diurnal profiles of air pollutants in the atmosphere before the intervention program.The abscissa and longitudinal coordinates represent the percentages of NO x and VOCs relative to the real average values measured at MK (i.e., 100 % NO x = 235.6 ppbv; 100 % VOCs = 51.2ppbv).In other words, they reflect the reduction percentages.For example, 80 % NO x or VOCs means NO x or VOCs was cut Net HO 2 formation −(7.0 ± 3.1) × 10 4 (7.8 ± 3.1) × 10 3 by 20 %. O 3 was simulated in 220 cases (i.e., 10 VOCs × 22 NO x ), and the maximum O 3 in each case was extracted.
It was found that within the cutting range of 0-60 % of NO x (Fig. 12a), O 3 increased with the increase of VOCs and the decrease of NO x , indicating a VOC-limited regime in O 3 formation.The black curve, perpendicular to the isopleths of O 3 , demonstrated the most sensitive response of O 3 to VOCs and NO x .Namely, O 3 reduction could be achieved most efficiently towards the abscissa.Using the absolute value of RIR (i.e., |RIR|) as a measure of the O 3 -VOCs-NO x sensitivity, it increased with the reduction of VOCs and NO x .For example, |RIR| for VOCs and NO x increased from 0.09 and 0.22 in the scenario of 90 % VOCs and 90 % NO x , to 0.25 and 0.90 in the scenario of 50 % VOCs and 50 % NO x , respectively.From before to during the program, VOCs and NO x decreased ∼ 12 and ∼ 15 %, respectively (i.e., from point A to B), causing a slight O 3 increment as described in Sect.3.3.2.The red and green curves in the lower right corner of Fig. 12a show the O 3 production as a function of VOCs cut before (NO x = 100 %) and during the program (NO x = ∼ 85 %), respectively.With the reduction of VOCs, O 3 decreased.Since O 3 -VOCs-NO x sensitivity increased during the program, O 3 decreased by 45.5 and 67.6 pptv with 10 % cut of VOCs before and during the program, respectively, which means that O 3 reduction could be achieved more efficiently by further cutting VOCs during the program.
With the continuous reduction of NO x , it was expected that the O 3 -VOCs-NO x sensitivity might change substantially due to the dual role of NO x in O 3 formation and titration.Figure 12b shows the O 3 -VOCs-NO x sensitivity in the cutting range of 60-99.5 % of NO x .It is noticeable that in the cutting range of ∼ 60-89.5 % of NO x , O 3 increased with the increase of VOCs and decrease of NO x , similar to that in the cutting range of 0-60 % of NO x .However, a transition area appeared when NO x was further cut, where O 3 stayed relatively stable with NO x variations, and decreased with VOC reductions.This transition area changed from ∼ 5.5-10.5 % (i.e., VOCs = 100 %) to ∼ 2.5-6 % of NO x (i.e., VOCs = 10 %).The appearance of the transition area implied that the titration of O 3 by NO x reached the minimum level, and further cutting of NO x might actually cause O 3 reduction.As expected, O 3 decreased with the reduction of NO x when NO x was reduced to lower than ∼ 2.5-5.5 % (i.e., ∼ 2.5 % and ∼ 5.5 % for 10 % and 100 % VOCs, respectively), and responded weakly to VOC variations.This means that O 3 formation switched to a NO x -limited regime.It should be emphasized that this is the first attempt on the study of O 3 -VOCs-NO x sensitivity at roadsides in Hong Kong, which could be a reference for the formulation and implementation of future air pollution control strategies in Hong Kong.Figure 13 shows the net O 3 increment (i.e., positive and negative values indicate the increase and decrease of O 3 , respectively) as a response of the reduction of VOCs and NO x in LPG-fueled vehicle exhaust.It was found that the net O 3 increment uniformly increased with the increase of NO x cut and the decrease of VOCs cut.To ensure that O 3 would not increase during the program, the reduction of VOCs and NO x should be within the highlighted area; that is, the highest cutting percentage of NO x should be less than 68.9 % (i.e., point A).Furthermore, when the cutting percentage of VOCs increased from 20 to 100 %, the maximum cutting percentage of NO x for zero O 3 increment was between 18.2 and 68.9 %, and the cutting ratio of VOCs / NO x increased from 1.1 to 1.45, suggesting that the cutting ratio of VOCs / NO x should be > 1.1 in order to maintain zero O 3 increment.During the program, VOCs and NO x in LPG-fueled vehicle exhaust were cut by ∼ 43 and ∼ 89 % (i.e., point B), respectively.According to the cutting ratios of VOCs / NO x (∼ 1.45) in the high cutting range of NO x , the minimum reduction percentage of VOCs should be ∼ 129 % when NO x was cut by ∼ 89 %.In other words, O 3 would inevitably increase in this case.Indeed, NO 2 experienced an overall increase from 1999 to 2013 at the roadsides in Hong Kong (HKEPD, 2014).According to this study, secondary NO 2 might be more responsible for the increase.Considering nil emission of NO 2 (Sect.3.2.2) and low emission of NO from LPG usage (i.e., 4.0 and 1.1-7.3% based on the emission inventory and source apportionment, respectively), an alternative scheme was proposed; namely, reasonable cutting ratio of VOCs / NO x was used to fulfill zero O 3 increase when VOC and NO x were cut.Although the O 3 production increase was minor (i.e., 0.4 ppbv or 5.6 %) in this study, this scheme could be applicable in future programs.Moreover, since the reactivity and concentration of VOCs influenced their O 3 formation potential, the relative incremental reactivity (RIR) and RIR (weighted concentration) of LPG-related VOCs were calculated.Table 9 shows the RIR and RIR-weighted concentration of propane, propene and n/ i-butanes.Propene had the highest RIR (i.e., 5.21 × 10 −2 ), suggesting the highest sensitivity of O 3 production to propene.However, n-butane in LPG was found to have the highest RIR (weighted concentration) (2.61 µg m −3 ), indicating that cutting n-butane in LPG source was optimal for O 3 pollution control.
In this study, an intervention program, initiated in September 2013, and which aimed to reduce emissions of VOCs and NO x from LPG-fueled vehicles in Hong Kong, was evaluated.The results indicated that LPG-related VOCs and NO x decreased significantly during the program, when higher reduction rates were observed for LPG-related VOCs and NO x compared to those in previous years.Source apportionment revealed that propane, n/ i-butanes and NO in LPG-fueled vehicle emissions were significantly lower during than before the program.It was therefore concluded that the program was remarkably effective in reducing VOCs and NO x .To evaluate the program more comprehensively, the variation of O 3 production was simulated with the PBM-MCM model.It was found that LPG-fueled vehicular emission was generally O 3 -destructive at the roadside MK site, and the O 3 destruction decreased from 0.47 to 0.06 ppbv due to the intervention program, causing an hourly average O 3 increase of 0.40 ppbv (∼ 5.6 %).The LPG-fueled vehicle exhaust generally made negative contributions to the production of OH and HO 2 .During the program, the destructive effect weakened and even turned to a positive contribution to HO 2 production, resulting in the increases of OH, HO 2 and HO 2 / OH.This was in line with the fact that O 3 increased slightly during the implementation of the program.To improve the program for future application, an O 3 -VOCs-NO x sensitivity analysis was conducted for ambient air that is not partitioned to sources, and the individual source of LPG-fueled vehicle exhaust.The NO x -limited regime in O 3 formation was only found when NO x was reduced to less than 5.5 %.Furthermore, for the emission reductions in LPG-fueled vehicle exhaust, the maximum NO x cutting percentage of 68.9 % and the lowest cutting ratio of VOCs / NO x (i.e., 1.1) were determined to maintain zero O 3 increment.
The Supplement related to this article is available online at doi:10.5194/acp-16-6609-2016-supplement.
• E), Causeway Bay (CWB, 22.28 • N, 114.19 • E) and Central (22.28 • N, 114.16 • E)), a general site (Tap Mun (TM, 22.47 • N, 114.36 • E)) and a background site (Hok Tsui (HT, 22.22 • N, 114.25 • E)).MK, CWB and Central are in mixed residential/commercial areas with heavy traffic and are surrounded by dense tall buildings.TM is a rural site surrounded by country parks, and is upwind of Hong Kong in autumn/winter seasons.These four sites have air quality monitoring stations deployed by Hong Kong Environmental Protection Department (HKEPD) (more details are provided at http://www.aqhi.gov.hk/en/monitoring-network/air-quality-monitoring-stations9c57.html?stationid=81).

Figure 1 .
Figure 1.Geographical location of the sampling sites.The yellow line represents the border between Hong Kong and mainland China.

Figure 4
Figure4presents the temporal variations of propane and n/ ibutanes at MK from June 2013 to May 2014 covering the periods of both before (i.e., June-September 2013) and during the intervention program (i.e., October 2013-May 2014).The abnormally low alkane levels in late June 2013 might be caused by the anticyclone over southern China and a tropical storm over the SCS, leading to extremely high temperature on 18-20 June (i.e., favorable for photochemical reactions), and stronger winds (i.e., conducive to atmospheric dispersion) on the following days, respectively.In contrast, Hong Kong was strongly influenced by an active ocean flow and a low pressure trough over the SCS in late July 2013.The consequently heavy rain and low temperature suppressed the photochemical reactions and caused the unusually high levels of VOCs (Fig.S2in the Supplement illustrates these events).Generally, the LPG-related VOCs experienced a significant (p < 0.05) reduction throughout the year.The daily changing rates of LPG-related VOCs in two periods, i.e., September-December and January-May, in different years at MK are shown in Table4.It is noteworthy that the two periods were selected based on data availability each year, and the minimized influence of meteorological parameters in the same month of different years.The LPG-related VOCs decreased from September to December in 2011 and 2013, but there

Figure 3 .
Figure 3. Monthly reductions of LPG-related VOCs at MK before (blue bar) and during (red bar) the program.(The average value of VOC species in September is the baseline.The values along the dot line were the cumulative numbers of converters replaced.The bars above and below the x axis refer to the reduction and increase of VOC mixing ratio, respectively.)

Figure 4 .
Figure 4. Variation trends of the daily average propane and n/ ibutanes at MK from June 2013 to May 2014.In the equation, label x has a unit of days.

Figure 5 .
Figure 5. Variation trends of the daily average primary NO x , NO and NO 2 and secondary NO 2 at MK from June 2013 to May 2014.In the equation label x has a unit of days.

Figure 6 .
Figure 6.Source profiles resolved by PMF before (red bar) and during (green bar) the program.The standard errors are estimated with the bootstrap method in the model.

Figure 8 .
Figure 8. Net O 3 variation as a function of VOCs, NO x and VOCs + NO x emitted from LPG-fueled vehicles.

Figure 9 .
Figure 9. Daytime patterns of OH and HO 2 (a) before and (b) during the program.

Figure 10 .
Figure 10.Average formation and loss pathways of OH and HO 2 at MK during the study period.

Figure 11 .
Figure 11.(a) Net production of OH and HO 2 by LPG-fueled vehicle exhaust (negative values indicate net destruction); and (b) increases of OH and HO 2 from the period before the program to during the program.

Figure 12 .
Figure 12.(a) O 3 -VOCs-NO x sensitivity within the cutting range of 0-60 % of NO x .The black curve demonstrates the most sensitive response of O 3 to VOCs and NO x .Points A and B represent the O 3 -VOCs-NO x relationship before and during the program, respectively.The red and green curves in the small legend show the O 3 production as a function of VOC cuts before and during the program, respectively.(b) O 3 -VOCs-NO x sensitivity within the cutting range of 60-99.5 % of NO x.

Figure 13 .
Figure 13.Net O 3 increment as a function of the reduction percentages of VOCs and NO x in LPG-fueled vehicle exhaust.The highlighted area indicates zero O 3 increment.

Table 1 .
Ratios of toluene / benzene and xylenes / ethylbenzene at MK and other sites in Hong Kong from October 2012 to May 2014 (unit: ppbv ppbv −1 ).

Table 2 .
Formation and loss pathways of OH and HO 2 considered in this study.

Table 3 .
Mixing ratios of LPG-related VOCs and NO x during the period before the program, the matched period and the period during the program (unit: ppbv).

Table 4 .
Changing rates of propane and n/i-butanes in September-December and January-May in different years (ppbv day −1 ).

Table 5 .
Changing rates of primary NO x , NO, NO 2 and secondary NO 2 (ppbv day −1 ).Site Species Jun 2011-May 2012 Jun 2012-May 2013 Jun 2013-May 2014 MK [NO x ] prim −0.004 0.040 * * The changing rate is insignificant (p > 0.05); the bold numbers are changing rates during the implementation of the program.

Table 6 .
Mass concentration and percentage contribution of the four sources to VOCs at MK before and during the program.

Table 7 .
VOCs and trace gases emitted from LPG-fueled vehicles before and during the intervention program (unit: µg m −3 unless otherwise specified).The standard errors are estimated with the bootstrap method in the PMF model.The bold fonts demonstrate significant reductions of LPG related VOCs and NO x .

Table 8 .
Contributions of LPG-fueled vehicle exhaust to the formation and loss of OH and HO 2 before and during the program.The bold fonts demonstrate the sum of OH/HO2 formation/loss rates or the net formation rate of OH / HO 2 .