Source apportionment vs. emission inventories of non-methane hydrocarbons (NMHC) in an urban area of the Middle East: local and global perspectives

. We applied the positive matrix factorization model to two large data sets collected during two


Introduction
Fast urbanization expansion, high population density, industrialization, intensive transport networks, and negligent pollution control have led to a substantial degradation of air quality in the Middle East (MEA) region (UNEP, 2006). Nowadays, air quality is taken as a major aspect of the quality of life leading to sustainable development in many areas of the world.
In Lebanon, a developing country in the MEA, which is located in Western Asia on the eastern shore of the Mediter-ranean Sea, air quality has continuously deteriorated leading to high pollutant levels exceeding the World Health Organization (WHO) recommended values (WHO, 2005); for instance, Afif et al. (2009) reported an annual average concentration of nitrogen dioxide (NO 2 ) in Beirut from December 2004 to June 2006, of 67 µg m −3 which is higher than the WHO annual recommended value of 40 µg m −3 (WHO, 2005). In addition, high levels of particulate matter, PM 10 and PM 2.5 , were obtained with annual concentrations of 64 and 20 µg m −3 respectively (Massoud et al., 2011), exceeding WHO guideline values of 20 and 10 µg m −3 . Although these measurements provide valuable information on air pollution, they are scarce and limited to a few pollutants. In the absence of adequate environmental policy and action, Lebanon is facing today significant urban air pollution problems; the cost of degradation of air quality was last estimated in 2001 at 1.02 % of the annual gross domestic product (GDP) (Sarraf et al., 2004).
Among air pollutants, non-methane hydrocarbons (NMHCs) play an important role in urban areas since they contribute to the formation of photochemical oxidants such as ozone and peroxyacetylnitrate (PAN) and they are also involved in the formation of secondary organic aerosols (SOA) (Seinfeld and Pandis, 2006). Moreover, some species are associated with adverse health effects or are even carcinogenic (e.g., benzene, 1,3-butadiene, etc.) (WHO, 2000).
NMHCs are emitted mainly from anthropogenic sources in urban areas. Major anthropogenic sources include vehicle exhaust, fuel evaporation, solvent use, emissions of natural gas and industrial processes (Friedrich and Obermeier, 1999;Sauvage et al., 2009). While urban emission sources of NMHC are usually well identified, their relative importance is still under debate and can be region-dependent (Yuan et al., 2013).
Considering the diversity in chemical reactivity among NMHCs, their different ozone and SOA formation potentials (Carter, 1994;Derwent et al., 2010) and the speciation profiles of the various emission sources, there is a clear need to identify and quantify NMHC emission sources, and their associated speciation by the development of regional emission inventories in order to apply efficient control strategies and air pollution management.
Moreover, emission inventories are used as input data for air quality models; therefore, their accuracy is also important for air quality impact assessment. However, recent in situ observations have revealed persistent and high uncertainties in emission inventories in urban areas of post-industrialized countries. Borbon et al. (2013) found large discrepancies between observed emissions ratios and those calculated from the inventories in Los Angeles and Paris. Moreover, Gaimoz et al. (2011) showed that the current emission inventory for Paris strongly overestimates the volatile organic compounds (VOCs) emitted from solvent use and needs to be corrected. Niedojadlo et al. (2007) also revealed some inconsistency be-tween emission inventory in Germany and source apportionment results, especially regarding emissions related to solvent use. Those studies have revealed that source-receptor approaches could be a relevant alternative to emission inventories where provided observations are available.
For the MEA region, emissions uncertainties could be even higher. At present, local inventory, emission data and observations for their evaluation are sparse and not speciated, potentially resulting in higher uncertainty in modeling approaches Waked et al., 2013). In particular, there is a paucity of data on NMHC levels and their emission sources in the Middle East region. At the global scale, several emission inventories exist but very few of them are speciated or exhibit fine temporal resolution (ECCAD database, 2015, http://eccad.sedoo.fr). For Lebanon, a recent work has led to the development of an atmospheric emission inventory of anthropogenic and biogenic sources considering CO, NO x , SO 2 , NMVOC, NH 3 , PM 10 , and PM 2.5 , for a base year of 2010 according to the 2009 EMEP/EEA guidelines (EMEP/EEA 2009;. The aim of this work is to identify major NMHC sources and quantify their contribution to NMHC concentrations on a seasonal basis by applying the source-receptor positive matrix factorization model (PMF) to the observations collected at a suburban site in the urban area of Beirut, Lebanon, and evaluate its consistency with existing emission inventories. The receptor modeling techniques are numerous; many of them have been previously used in NMHC source apportionment worldwide including principal component analysis (PCA), positive matrix factorization (PMF), chemical mass balance (CMB) and UNMIX like in Europe (Badol et al., 2008;Sauvage et al., 2009), the USA (Leuchner and Rappenglück, 2010), and in China (Guo et al., 2007;Yuan et al., 2013). The performance of four receptor models (PCA, PMF, CMB and UNMIX) was evaluated by Miller et al. (2002) by applying them to the same artificial data set of VOCs. They found that PMF extracted factor profiles that most closely represented the major sources used to generate the simulated data, and concluded that PMF is the most appropriate model to explain the results (Miller et al., 2002).
A large set of speciated NMHC have been continuously measured during 2-week periods in summer 2011 and winter 2012 in the frame of the Emission and Chemistry of Organic Carbon in the Eastern Mediterranean (ECOCEM) experiment. The consistency of the source composition obtained by the PMF on a seasonal basis is evaluated with regards to NMHC source profiles established by canister sampling in the vicinity of emission sources (Salameh et al., 2014). 2 Experimental procedures

Sampling site: location and surroundings
The site was chosen in a way to be far enough from strong direct emissions (industrial, road transport) since they can hide the emissions from distant sources. At the same time, the site is appropriately located in order to receive the air masses coming from the Greater Beirut Area (Fig. 1). The measurements were conducted on the roof of the Faculty of Sciences building of Saint Joseph University (33 • 87 N, 35 • 56 E) from 2 to 18 July 2011 in summer and again from 28 January to 12 February 2012 in winter. The site is located in the eastern suburb of the city of Beirut (6 km southeast of Beirut downtown) at an altitude of 230 m a.s.l. (above sea level). The site is surrounded by a forested pine area and a high density of residential premises. Beirut International Airport is located 8 km southwest of the site and one fuel storage facility is located at 3.6 km north (Salameh et al., 2014).

Material and methods
NMHCs were continuously analyzed by on-line Thermal Desorption Gas Chromatography with a Flame Ionization Detector (TD-GC-FID) provided by Perkin Elmer described elsewhere (Salameh et al., 2014). The on-line measurement was performed hourly, covering 30 min of ambient air sampling. Sixty seven NMHCs from C2 to C9 belonging to alkanes (29), alkenes (19), alkynes (2) and aromatics (17) were identified and quantified. The detection limits were around 40 pptv for the targeted compounds except for ethane and ethylene (90 pptv). A certified NPL (National Physical Laboratory) standard of NMHC mixture (∼ 4 ppbv) was used to determine and check the stability of the GC carbon response during both measurement periods. The data have been published in Salameh et al. (2015).
Additional trace gases including carbon monoxide (CO), NO x and O 3 were measured on-line on a 1 min basis using trace gas analyzers. Basic meteorological parameters (wind speed and direction, temperature, relative humidity and atmospheric pressure) were measured on a 1 min basis during both campaigns.
3 Source apportionment by positive matrix factorization (PMF)

PMF model description
The USEPA PMF v3.0 was applied for NMHC source apportionment analysis. This method is described in details elsewhere (Paatero and Tapper, 1994;Paatero, 1997). NMHCs which are not identified during both seasons (like tetrachloroethylene) or their concentrations were frequently below the detection limits were not considered in the PMF analysis. According to the PMF, an observed concentration at a receptor site can be viewed as a data matrix X of i by j dimensions in which i number of samples and j chemical species were measured.
where g ik is the mass contribution of the kth source to the ith sample, f kj is the j th species mass fraction from the kth source, p is the number of independent sources, and e ij is the residual associated with the concentration of the j th species in the ith sample. The solution of the equation is obtained by minimizing the residual sum of squares Q given by where f kj ≥ 0 and g ik ≥ 0 and where n is the number of samples, m the number of considered species, and s ij an uncertainty estimate for the j th species measured in the ith sample. The results are constrained so that no sample can have significantly negative source contributions. The goal is to identify the number of factors p, the species profile f of each source, and the amount of mass g contributed by each factor to each individual sample by using measured source profile information, meteorological parameters analysis, and additional species.

PMF model implementation
One of the advantages of the PMF is to consider the uncertainty of each individual concentration. Special attention was paid to the estimation of the uncertainty. For each measured concentration, an uncertainty was calculated following 3598 T. Salameh et al.: Source apportionment vs. emission inventories the method developed within the ACTRIS (Aerosols, Clouds, and Trace gases Research InfraStructure) network (Hoerger et al., 2015, see their Supplement). It includes the systematic errors in the measurement and the precision which reflects the random errors. Thus, the total uncertainty of a measurement is obtained with Eq. (3): x 2 unc = x 2 precision + x 2 systematic. ( The precision is derived from a series of measurements of the NPL standard gas. It covers the random errors of the sampling and analytical system. Possible systematic errors include the calibration gas uncertainty, systematic peak integration errors, sample volume determination, and potential blank value (Hoerger et al., 2015, see their Supplement). Therefore, the overall systematic error is described as follows: The systematic integration error can be neglected because the chromatograms (peak shapes of the considered species and the form of the baseline) were quite similar for air samples and for the calibration gas.
The systematic volume error is neglected since δV sample = δV calibration .
For the expanded uncertainty, the total uncertainty calculated is multiplied by the coverage factor k = 2.
The relative expanded uncertainty ranged between 4 and 27 % in summer for all the compounds except for octane which was 38 %. In winter, the relative expanded uncertainty was between 5 and 35 % for most of the compounds except for trimethylbenzenes, isoprene, octane and heptane which had an uncertainty between 40 and 60 %.
The missing data were replaced by the median of the concentrations measured the other days at the same hour; and in these cases, the uncertainty was set as four times the concentration. Values reported as below DL were replaced by half of the DL and the corresponding uncertainty was set as 5/6 of the DL value. Paatero and Hopke (2003) have introduced the signal-tonoise ratio (S/N) which takes into consideration the concentration (x) and the uncertainty (s) of the species (j ) (Eq. 5). This ratio indicates whether the variability in the measurements is real or within the noise of the data. A compound characterized by low concentrations or by a large number of observations (i) associated with relatively high uncertainties will have a low S/N ratio. If the S/N ratio is less than 0.2, the species is excluded and if the ratio is greater than 0.2 but less than 2, the uncertainty is multiplied by four.
For a reliable identification of the sources, the PMF requires a large number of samples. The large data sets collected during the two campaigns, 298 samples in summer and 179 samples in winter, were lumped separately for PMF source apportionment seasonal analysis. By taking into consideration the different existing reactivities, mainly with OH radical within the NMHC species, the profile of each source may change temporally during transport of emission plumes. Nevertheless, we have examined the factors influencing the distribution of NMHCs (Salameh et al., 2015) and we concluded that NMHCs were freshly emitted from local sources in winter but also in summer, and the chemical losses of NMHC species, even for the most reactive species, were likely not significant.

Determination of the optimal solution
Several base runs were performed with a different number of factors from 3 to 12. The diagnostic parameters include the value of Q, IM which is the maximum individual column mean and IS which is the maximum individual column standard deviation defined by Lee et al. (1999). Q, IM, and IS are then plotted against the number of factors (from 3 to 12). The number of factors chosen corresponds to a significant decrease of Q, IM, and IS. In order to control the rotation and to optimize the selected solution, the F peak parameter was used. Six factors were extracted from winter data set and five from summer data set. Quality indicators from the PMF application are summarized in Table 1.

Meteorological conditions
Lebanon is characterized by a narrow coastal strip located on its western part and by the Bekaa valley located between the western and eastern Lebanese mountain chains. The coastal region has a Mediterranean climate with land-sea breeze circulation. During the summer measurement campaign, the temperature at the Saint Joseph University site ranged from 20 to 29 • C with an average of 25 • C. The average wind speed was 2 m s −1 with a maximum wind speed of 10 m s −1 during the day coming mostly from the southwest and from the north on 7, 8 and 9 July. The wind direction was mostly from the northeast and slightly from south, southwest at night. During the winter measurement campaign, the temperature ranged from 7 to 22 • C with an average of 13 • C. The average wind speed was 2 m s −1 and the wind direction was mostly coming southeasterly and easterly (Salameh et al., 2015).

Identification of PMF profiles
In order to attribute PMF factors to emission sources, a comparison between extracted PMF profiles and speciated profiles established in the vicinity of emission sources by  Salameh et al. (2014) was carried out, when possible. In addition, the diurnal cycle and time series of the contributions of modeled sources were analyzed with independent parameters like meteorological data and specific tracers like CO and NO x .

Winter PMF factor identification and contribution
In winter, six sources were identified corresponding to combustion mainly related to regional traffic, combustion related to local traffic, gas leakage, gasoline evaporation related to traffic, gasoline evaporation as an episodic point source and a source characterizing the urban background. The composition of the source profiles and the diurnal variations of the source contribution are reported in Figs. 2 and 3, respectively.

Background
Long-lived species are dominant in factor 1. The ethane and propane variability explained by this factor is 76 and 42 % respectively, even though, toluene, butane and acetylene are also present in this profile. This factor is, therefore, categorized as urban background (Lanz et al., 2008;Sauvage et al., 2009). The diurnal profile of this factor does not show any significant peak and instead maintains the same contribution independently of the hour of the day (Fig. 3). The average relative contribution of this factor is 18 % in winter.

Combustion mainly related to regional traffic
Factor 2 has significant amounts of toluene, m,p-xylenes, ethylbenzene, ethylene, acetylene, 1,2,4-trimethylbenzene and decane (co-eluted), and benzene which are typical combustion products. Almost 30 % of the variances of ethylene, propene, acetylene and benzene are explained by this factor. This profile is very similar to the profile determined in the vicinity of the emission sources in Beirut which consists of exhaust and evaporative running losses from vehicles (Salameh et al., 2014). The factor contribution is well correlated with the concentrations of CO and NO x which are combustion tracers (R = 0.8 and 0.7, respectively). The daily time series of the contribution of this source are character- ized by a first significant increase of concentrations at around 07:00-08:00 LT (UTC/GMT+2 h), the second one at 14:00-15:00 LT, another increase at 17:00-18:00 LT and a small peak is observed at 20:00-21:00 LT in the evening. According to , the diurnal variation profile for on-road mobile source in urban areas in Lebanon shows two significant increases in weekdays, a morning peak at around 07:00-10:00 LT and an evening peak from 16:00 to 18:00 LT consistent with the important peaks observed for factor 2. In addition, in winter, additional sources related to domestic heating should be considered since the measurement site is surrounded by residential area. The average relative contribution of this factor is 19 % in winter. Factor 3 is also characterized by some combustion tracers (ethylene, acetylene, toluene, benzene, propene, and m,pxylenes). Nearly 50 % of the variances of ethylene as well as propene are explained by this factor. The factor contribution strongly correlates with CO and NO x (R = 0.8 and 0.7 respectively). This factor is associated to "combustion related to local traffic" since it does not show a significant contribution during the weekend which is due to the limited local activity nearby the measurement site (schools, universities, residential area). The daily variation of this factor is characterized by an initial increase at 07:00-08:00 LT, a peak at 18:00-19:00 LT and the last peak in the evening at 20:00-21:00 LT. These enhancements are far less than those of factor 2. The average relative contribution of this factor is 10 % in winter.

Gasoline evaporation related to traffic
Isopentane, butane, isobutane, toluene and 2-methylpentane are the main compounds of factor 4. Most of isopentane (53 %) and around 36 % of butane and isobutane are explained by this factor. This strongly suggests that this profile is related to fuel evaporation. Additionally, the profile of this factor, taking into account the larger NMHC contributors, is compared to three gasoline evaporation profiles established by near-field measurements including the fuel storage facilities, hot soak (a type of evaporative emissions which occurs when a warmed-up vehicle is stationary and the engine is stopped) and refueling cars (Salameh et al., 2014) in Fig. 4. The fingerprint is consistent within all the gasoline evaporation source profiles confirming the nature of this source. In Lebanon, light-duty vehicles operate on gasoline while only heavy-duty vehicles are allowed to run on diesel (4 % of fleet in Beirut) . The diurnal profile as well as the time series in Fig. 5 display the same aforementioned peaks in factors 2 and 3 especially in the morning and in the evening assuming that the gasoline evaporation is linked to traffic. The average relative contribution of this factor is 22 % in winter.

Gasoline evaporation
Factor 5 is also characterized by the C4-C5 alkanes, toluene and 2-methylpentane. Nearly 42 % of the isopentane and pentane variability and 48 % of the isobutane and butane variability are explained by this factor. The fingerprint of the chemical composition is consistent with the ones of typical gasoline evaporation profiles established by near-field measurements (Fig. 4). The diurnal profile of this source contribution is marked by a night minimum and an increase in the midday. The time series presented in Fig. 5 shows that this factor (F5) is primarily episodic with high contribution originating from the north wind sector as demonstrated in the pollution rose in Fig. 6. The extremely high levels of the source contribution come from the north sector where the fuel storage facilities are located (Fig. 6). A collinearity between both gasoline evaporation factors (F4 and F5) is observed around noon when the wind blows from the north sector since the highest contributions are detected in this sector making the apportionment difficult. The average relative contribution of this factor is 21 % in winter.

Gas leakage
58 % of propane variability is explained by factor 6. The composition of this source is mainly impacted by high loadings of propane, butane and isobutane which are the signatures of LPG (liquified petroleum gas) (Blake and Rowland, Atmos. Chem. Phys., 16, 3595-3607, 2016 www.atmos-chem-phys.net/16/3595/2016/ 1995). Hence, this source is attributed to gas leakage, mainly LPG, from residential use. LPG is the main fuel for cooking in households in Lebanon and it is used, among others, for domestic heating. The diurnal profile of this factor does not show significant peaks, only some small peaks are observed in the morning, at noon and in the late evening (Fig. 3). The average relative contribution of this factor is 10 % in winter.

Summer PMF profile identification and contribution
One of the limitations of PMF application for summer data could be the effect of the photochemistry (Yuan et al., 2012). The effect of photochemistry has been assessed in Salameh et al. (2015), and the results showed a limited influence of the photochemistry in summer. The five sources identified by the PMF analysis in summer correspond to combustion related (so-called nocturnal and diurnal) to traffic, gasoline evaporation related to traffic, gasoline evaporation as an episodic point source and a source characterizing the local background. The composition of the source profiles and the diurnal variation of the source contribution are reported respectively in Figs. 7 and 8.

Local background
64 % of the variability of isoprene, commonly used as the tracer of biogenic emissions, is explained by the factor 1. The profile of this factor consists largely of toluene, ethane, isopentane, m,p-xylenes, acetylene, butane and propene which are related to combustion. Therefore, "local background" is attributed to this factor including the regional background, the combustion from the surrounding sources, as well as the biogenic emissions from the forest nearby the measurement site. It is not possible to separate this type of source even by increasing the number of PMF factors. The diurnal profile shows higher contributions during the day starting 07:00-08:00 LT and lower contributions during the night. The average relative contribution of this factor is 10 % in summer.

Traffic-related sources
Two profiles (factor 2 and factor 3) related to traffic have been extracted with different contributions between day and night.
The profile of factor 2 is essentially composed of aromatic compounds toluene, m,p-xylenes, o-xylene, ethylbenzene, 1,2,4-trimethylbenzene and decane (co-eluted), ethylene, acetylene and isopentane which are included in the traffic related sources (gasoline evaporation and combustion, Fig. 9) and are present in the composition of factor F2 in winter (especially the aromatics). The diurnal variation shows high loadings during the day (Fig. 8). A significant first increase is observed at 07:00-08:00 LT (UTC/GMT+3 h), two other small peaks are observed during the afternoon (14:00-15:00 LT) and in evening (20:00-21:00 LT) comparable to the peaks observed in the combustion related to traffic during winter (winter F2). This factor is linked to traffic emissions occurring during the day and coming mainly from the southwest sector. The average relative contribution of this factor, combustion related to diurnal traffic, is 17 % in summer.
Typical combustion tracers are the major contributors to the profile of the factor 3. The variability of acetylene (34 %), ethylene (43 %), propene (45 %) and benzene (30 %) is mostly explained by this factor. The daily variability of this factor displays high contributions during nighttime characterizing the combustion related mainly to nocturnal traffic. At night, the wind direction is mostly from the northeast, the south, and southwest where the highway network is densely occupied at night in summer and the vehicles run slowly. When the vehicles speed is low, especially in the case of vehicles running on gasoline without a functioning catalytic converter , the NMVOC emissions increase (Fontaine, 2000). The average relative contribution of this factor is 27 % in summer.

Gasoline evaporation related to traffic
Factor 4 encloses a high percentage of butane, toluene, propane, m,p-xylenes and isopentane which strongly suggests that this factor is related to the gasoline evaporation. This factor exhibits also high loadings during nighttime alike the combustion related to nocturnal traffic factor 3 confirming the traffic as the origin of these two factors (F3 and F4).
By comparing both combustion (F2 and F3) and gasoline evaporation (F4) factors related to traffic profiles with the road transport profile established in near-field measurements (Salameh et al., 2014), which consists of exhaust and evaporative running losses from vehicles (Fig. 9), we notice the same overall fingerprint for the majority of the species, nevertheless, propane exhibits a high percentage in the gasoline evaporation profile and 75 % of its variability is explained by this factor (F4), suggesting the liquefied petroleum gas leak- age as an additional source to the gasoline evaporation. The average relative contribution of this factor is 30 % in summer.

Gasoline evaporation (episodic point source)
Factor 5 is characterized by C4-C5 alkanes, toluene and 2methylpentane similar to the factor 5 composition in winter. This factor is gasoline evaporation showing a high contribution in the middle of the day. The pollution rose, as in winter, shows high contributions coming from the north sector, where the fuel storage facilities are located, as shown in Fig. 10 which corresponds to episodic events having occurred on 7, 8 and 9 July. The average relative contribution of this factor is 16 % in summer. Figure 11 summarizes the average relative contribution in weight % of each factor in winter and in summer. We found several common factors in winter and summer but their relative contribution was slightly different. For instance, the contribution of traffic-related sources including the combustion and the gasoline evaporation from traffic accounted for 51 % in winter and 74 % in summer. Nevertheless, the episodic gasoline evaporation source contributed to 21 % in winter and 16 % in summer. This source was a major contributor to winter's NMHC concentrations reflecting the importance of gasoline evaporation in winter. If we consider the gasoline evaporation sources (episodic and related to traffic), we obtain a significant contribution of 43 % in winter and 46 % in summer. Additional factors were found in winter but absent in summer: for instance, gas leakage due to the use of LPG in res-Atmos. Chem. Phys., 16, 3595-3607, 2016 www.atmos-chem-phys.net/16/3595/2016/ Figure 9. Source composition profiles (wt %) of combustion related to diurnal and nocturnal traffic (F2 and F3) and gasoline evaporation (F4) in summer compared to the on-road transport profile established by near-field measurements (Salameh et al., 2014). idential heating. Another factor is characterizing the urban background highly contributed to winter's NMHC concentrations (18 %) mainly because of the meteorological conditions and the absence of photochemistry in winter (Salameh et al., 2015). A source characterizing the "local background" including the biogenic emissions was exclusively found in summer since the meteorological conditions are favorable. Nevertheless, this "local background" also includes regional background and combustion emissions. Based on the PMF analysis, we conclude that the main contributors to the NMHCs are the combustion and gasoline evaporation sources in summer as well as in winter. These sources are of local origin confirming what has been demonstrated previously (Salameh et al., 2015) that there is no significant influence of long-range transport of air masses on the NMHC concentrations.

Comparison of PMF results to emission inventories
Our study provides the first comprehensive observational constraints for a quantitative evaluation of anthropogenic emission inventories at local, national  and global levels for a Middle Eastern area. While such a comparison is somewhat limited as the basis for comparison is different (NMHC speciation, number of NMHCs, temporal and spatial resolution), it still provides relevant insights on uncertainties and perspectives of improvement.

The annual national emission inventory (NEI) for Lebanon
The contribution of the factors discussed above is dependent on the seasons and associated meteorological conditions, therefore, care must be taken when comparing these results to the annual national emission inventory (NEI) for Lebanon. The NEI is not speciated, and was established for a base year of 2010 within the EEA/EMEP guidelines, by , covering major anthropogenic and biogenic sources in the region with 5 km spatial resolution for Lebanon and 1 km spatial resolution for its capital city Beirut and its suburbs. The total VOC emissions in Lebanon were estimated to be 115 Gg for the year 2010. According to this inventory, transport is the main source of NMVOCs with a relative contribution of 67 % of total emissions of VOCs in Beirut and its suburbs as well as all over Lebanon. PMF results give nearly the same contribution of this sector in winter 51 and 74 % in summer. According to , the gasoline evaporation contribution was evaluated to be responsible for 8 % of total NMVOC emissions in Beirut and its suburbs and 17 % of total NMVOC emissions at the national level based on gasoline consumption whereas the PMF gasoline evaporation (related to traffic) sources contributed to 22 % in winter and 30 % in summer.
The comparison held here is subject to several limitations. On one hand, the national emission inventory is not speciated, and it needs further improvements regarding the input data and the Lebanon-specific emission factors need to be established . On the other hand, the measurement site is qualified as suburban and is located far from some sources identified in the inventory, like for instance large forests representing biogenic sources, and industrial plants. These sources have not been identified in the PMFresolved profiles since they were not major sources during this study and some of them are located in the north and south of Lebanon far away from the measurement site. Additionally, it was hard to discern the biogenic source since the key tracers of this source such as terpenes and oxygenated volatile organic compounds (methanol, acetone. . . ) were not included in the PMF analysis. Furthermore, the emission inventory data used for comparison are provided on annual statistical data, on a national scale whereas the data here are obtained from a single measurement site during a shorter time regarding a non-exhaustive list of species. Therefore, longterm and continuous studies integrating more than one measuring site and more specific tracers are of great interest in order to provide more reliable information.

Emission inventories for road transport
The objective of this last section is to estimate NMHC emissions from the road transport sector from PMF results and carbon monoxide observations and to compare these values to the ones derived from various emission inventories.

National emission inventory for Lebanon -NEI-
The annual emissions of air pollutants originating from onroad mobile sources in 2010 in Beirut specifically and in Lebanon were investigated by . At a national level, the estimated annual emissions for CO were 522 and 48 Gg for NMVOCs by using the EMEP methodology . In parallel, we calculated separately, for each season, the ratio of the sum of the 59 NMHC contributions (in µg m −3 ) coming from the PMF factors related to road transport (combustion and gasoline evaporation) and of the mean of the CO measured (in µg m −3 ) during each measurement campaign according to Eq. (6). Then, we used the estimated annual emissions for CO (Q) in Gg to generate the emissions of the 59 NMHCs based on the winter ratio on one hand, and on the summer ratio on the other hand, in order to give a range of annual emissions levels.
With NMHC = Sum of 59 measured species during ECO-CEM (in Gg). The mean concentration of CO was equal to 470 ± 195 µg m −3 during the summer campaign, while it was equal to 388 ± 228 µg m −3 during the winter cam-paign. The measured NMHC to CO ratio was estimated to be 0.15 in summer and 0.12 in winter.
By this method, the annual emissions for Lebanon are estimated to be 78 Gg from the summer ratio and 62 Gg by the winter ratio, which are higher than the annual emissions of 48 Gg. In both cases, the emissions from the national road transport inventory are lower with a reasonable difference ranging from 22 % for winter-based emissions to 39 % for summer-based emissions (Fig. 12). By applying the same calculation method for Beirut and its suburbs, the annual emissions obtained using the summer ratio are estimated to be 29 Gg higher than the annual emissions calculated based on winter ratio (23 Gg) (Fig. 12).
The measurements we conducted are more representative of Beirut and its suburbs (Greater Beirut Area). The emissions from the inventory regarding the road transport for Greater Beirut Area are also lower with a difference ranging from 20 % for winter-based emissions to 37 % for summerbased emissions.
The discrepancies for Lebanon as well as for Beirut and its suburbs using winter and summer ratios are comparable since the road transport is the dominating sector in the whole country and especially in the capital.

Global emission inventories
The objective here is to assess the PMF estimated emissions to the global emission inventories for the road transport sector for 59 NMHCs and, where possible, for a common combustion tracer "ethylene" which is also an ozone precursor due to its reactivity towards the OH radical (Atkinson and Arey, 2003).
We estimated the annual total NMVOCs emissions for Lebanon by using a mean flux of total NMVOCs of eight grids covering Lebanon surface area (ECCAD database) from the transportation sector from the global historical emission inventory ACCMIP (Atmospheric Chemistry and Climate Model Intercomparison Project) (Lamarque et al., 2010) for 2000 at a spatial resolution of 0.5 • . The total annual NMVOCs transportation emissions from the ACCMIP are estimated to be 8 Gg, which is a factor of 6 to 10 lower than the annual emissions obtained via the PMF results (62 to 78 Gg) and the road transport national inventory emissions (48 Gg). This gap is expected to be a lower limit regarding the fact that PMF only considers 59 NMHCs. Nevertheless, when we apply the same method to another global inventory EDGARv4.2 (Emissions Database for Global Atmospheric Research) at a spatial resolution of 0.5 • , the annual transportation emissions are estimated to be 23 Gg, which is a factor of 2 to 3 lower than the road transport national inventory emissions and the annual emissions obtained via the PMF results. This comparison also reveals that global inventories are not consistent between each other. Additionally, the global inventories do not describe the methodologies applied Atmos. Chem. Phys., 16, 3595-3607, 2016 www.atmos-chem-phys.net/16/3595/2016/ nor the data used (emission factors, NMVOC species considered. . . ). For a more detailed evaluation of speciation accuracy, we compared the emissions of an individual NMHC, ethylene, obtained by the PMF results according to Eq. (6), where NMHC represents ethylene, to the emissions from a speciated inventory having a monthly temporal resolution, over the same period, called MACCity. The MACCity global emission inventory has been developed within two European Commission funded projects (MACC and CityZen) as an extension of the ACCMIP and RCP8.5 (Representative Concentrations Pathways) emission inventories (Granier et al., 2011) at a spatial resolution of 0.5 • . Based on the PMF results, the emissions of ethylene during July 2011 are estimated to 0.30 and 0.25 Gg during January 2012 for Lebanon. The MACCity global emission inventory displays very low ethylene emissions from the transportation sector (ECCAD database) during July 2011 (0.05 Gg) and during January 2012 (0.04 Gg) for Lebanon (Fig. 12). The difference with the PMF results is by a factor of 6.

Comparison of the PMF results with the NEI and the global emission inventories
The comparison of the PMF results to the NEI for Lebanon as well as for Beirut and its suburbs suggests that the inventory underestimates the road transport emissions in a reasonable way (20 to 39 %). Whereas the comparison of the PMF results to the global emission inventories shows significant differences reaching a factor of 10. Taking into account uncertainties, when comparing PMF results to emission inventory, is not easily achievable due to the multiple sources of uncertainties. First, the uncertainty of the global emission inventory itself and, in particular, of the estimated annual emissions of CO (Q) (Eq. 6), is unknown. To get a sense of uncertainties one can compare emission amounts provided by different emission inventories which cannot be applied here. Second, the confidence intervals of the PMF results were not assessed. However, when comparing PMF to emission inventory, one could consider the standard deviation and the coefficient of variation (CV) of each PMF factor contribution in order to get an idea of the variability of the PMF factors contributions in each season.
Our calculation shows a high variability of the PMF factors contributions in both seasons including the transport-related factors (coefficients of variation around 100 %). We can rely on this variability to test whether differences between PMF and emission inventory are significant or not. On the one hand, for the Lebanese NEI, the differences are not significant (lower than 100 %). For the global emission inventory, differences are higher by a factor of 6 to 10 which is significant and higher than 100 %.

Conclusions
Measurements of 67 gaseous organic carbon have been performed at a suburban site in Beirut during two intensive field campaigns in summer 2011 and in winter 2012 within the framework of the ECOCEM project, resulting in a large and unique database, in order to improve our understanding of their major emission sources in the area.
The USEPA PMF v3.0 was applied to the winter and summer data sets in order to identify and quantify the source contribution. Six factors were then extracted in winter and five factors in summer. Source profiles established close to the main potential emitters by near-field measurements (Salameh et al., 2014) were used to identify the PMF factors.
In winter, the six NMHC emission factors consist of traffic-related sources including combustion mainly related to regional traffic, combustion related to local traffic and gasoline evaporation from vehicles accounting for 51 %. In summer, the combustion and gasoline evaporation related to regional traffic contributed to 74 % of which 57 % are nocturnal. Additionally, an episodic gasoline evaporation source originating principally from the fuel storage facilities was identified in winter and in summer contributing to 21 and 16 % respectively. The gas leakage, mainly LPG, contributes to 10 % and the urban background to 18 % of the ambient NMHC levels during winter. A factor characterizing local background was identified in summer contributing to 10 %. Having similar species composition profiles, we found a difficulty in discerning the combustion related to power generators, running on diesel oil, within the combustion sources, hence, additional species (e.g., PM 2.5 species measured with high temporal resolution) may have helped to extract this source.
Overall, the gasoline evaporation contribution (related to traffic and the episodic point source) was found to be a large contributor to the NMHC ambient concentrations: 43 % in winter and 46 % in summer.
Control measures should be developed to mitigate the NMHC emissions resulting from traffic-related sources (combustion and gasoline evaporation).
Verification of the reported emission inventory for Lebanon and Greater Beirut Area can assist in the evaluation of future control strategy effectiveness. The comparison of the modeled results to the emission inventory shows an agreement regarding the load of the on-road mobile source, by considering the vehicle exhaust and gasoline evaporation linked to traffic, in both seasons. Nevertheless, the gasoline evaporation related to traffic is lower in the inventory (17 %) by considering PMF summer results (30 %) suggesting a slight underestimation of the inventory. We assessed also the accuracy of several global inventories for the transportation sector and high discrepancies up to a factor of 10 still remain between the emission inventories.
To maximize current resources, continuous measurement of NMHC species and the implementation of ground-based monitoring networks for ambient air quality monitoring and assessment are essential in order to conduct source apportionment which would be of less uncertainty and of greater help in establishing effective NMHC, ozone and PM abatement strategies in Lebanon.