Measurement report: Summertime and wintertime VOCs in Houston: Source apportionment and spatial distribution of source origins

Abstract. The seasonal variations of volatile organic compounds (VOCs) was studied in the Houston metropolitan area in the summertime and wintertime of 2018. The analysis of hourly measurements obtained from the automated gas chromatograph (auto-GC) network showed the total VOC average concentrations of 28.68 ppbC in the summertime and 33.92 ppbC in the wintertime. The largest contributions came from alkane compounds, which accounted for 61 % and 82 % of VOCs in the summer and winter, respectively. We performed principal component analysis (PCA) and Positive Matrix Factorization (PMF) and identified seven factors in the summertime and six factors in the wintertime, among which alkane species formed three factors according to their rate of reactions in both seasons: (1) the emissions of long-lived tracers from oil and natural gas (ONG long-lived species), (2) fuel evaporation, and (3) the emissions of short-lived tracers from oil and natural gas (ONG short-lived species). Two other similar factors were (4) emissions of aromatic compounds and (5) alkene tracers of ethylene and propylene. Summertime factor 6 was associated with acetylene, and one extra summertime factor 7 was influenced by the biogenic emissions. The factor 6 of wintertime was affected by vehicle exhaust. Higher nighttime and lower daytime values of the ethylene/acetylene ratios during the summertime indicated the stronger impacts of ethylene photochemical degradation. Also, the exploration of the photochemical processes of the VOCs showed that the ethylene and propylene had the highest contributions to the summertime and wintertime ozone formation as well as the emissions of the isoprene, which showed a high impact on summertime ozone. Our results acknowledged that ethylene and propylene continue to be significant emissions of VOCs, and their emissions control would help the mitigation of the ozone of Ship Channel. Based on trajectory analysis, we identified main VOC emission sources in Houston Ship Channel (HSC) local industrial areas and regions south of the HSC. Ambient VOC concentrations measured at the HSC were influenced by the emissions from the petrochemical sectors and industrial complexes, especially from the Baytown refinery and Bayport industrial district next to the HSC and Galveston Bay refineries at the south of the study area.



80
Atmospheric VOCs originate from emissions of either anthropogenic or natural biogenic sources. On a global scale, natural emissions contribute more than 80% of total emissions. Biogenic VOCs emissions originate from biomass burning and complex gas-phase exchange processes in land-based plants dependent on temperature and light (Forkel et al., 2006;Fuentes et al., 2000;Karl et al., 2003;Kesselmeier and Staudt, 1999). In urban areas, however, numerous anthropogenic sources emit VOCs into the atmosphere and likely exceed biogenic emissions in these areas 85 (Rappenglueck et al., 2005;Winkler et al., 2002). The main sources of anthropogenic emissions are industrial processes, including those of crude oil and liquefied petroleum gas (LPG), hereinafter defined as oil and natural gas (ONG), gasoline transport and storage, vehicle combustion, and the manufacturing production of commercial goods (Piccot et al., 1992). A number of studies have identified these VOC sources, their emission strengths and contributions in photochemical processes in several urban areas, including Houston (Bi et al., 2021;Buzcu and Fraser, 2006;Czader 90 et al., 2008;Czader and Rappenglueck, 2015;Diao et al., 2016;Jobson et al., 2004;Leuchner and Rappenglueck, 2010;Pan et al., 2015). These studies reviewed the effectiveness of control strategies and policy regulations and links between VOC concentration levels and surface ozone production as explored by Kleinman et al. (2002).
Houston has been the site of a number of studies characterizing the emissions of organic compounds and their impact on ozone formation in urban areas. The Houston metropolitan area has some of the largest anthropogenic emission 95 sources of atmospheric pollutants in the United States (Song et al., 2021). The Houston Ship Channel area, for example, is affected by high rates of numerous pollutants from petrochemical and industrial facilities (Port Houston, 2019). Buzcu and Fraser showed that refinery and petrochemical activities are the dominant sources of VOCs at three sites in the Ship Channel area of Houston (Buzcu and Fraser, 2006). Building upon VOC source apportionment for six sites in the Houston's Ship Channel area from August to September of 2006, (Leuchner and Rappenglueck, 2010) 100 expanded that analysis further by looking into diurnal variations and including wind directional dependencies for a https://doi.org/10.5194/acp-2021- 565 Preprint. Discussion started: 26 July 2021 c Author(s) 2021. CC BY 4.0 License. receptor site outside the industrial region of the Ship Channel which was also exposed to urban emissions, and attributed the contributions of anthropogenic and biogenic emissions to ambient VOCs on the site.
This study aims to explore the characteristics of VOCs (such as concentration levels, diurnal variations, and emission features) and examine their seasonal variations for the emission factors of the Houston Ship Channel based on the 105 measurements of the summertime and wintertime 2018. In the present paper, we document the study domain and measurement datasets as well as the methods of our study (Sect. 3.1), which includes an integrated source-receptor relationship study for the Houston Ship Channel region using Principal Component Analysis (PCA) and Positive Matrix Factorization (PMF) models to characterize the emissions of VOCs and their temporal variation in this industrial sector of the Houston metropolitan area in summer-and wintertime (Sect 3.2). We also discuss the removal 110 process of the VOC and their contributions to ozone formation (Sect. 3.3). Leuchner and Rappenglueck (2010) provided a categorical view about the wind direction frequency for emission sources, but this paper goes beyond by studying the geographical origins of the air masses coming from the larger Houston metropolitan area (Sect 3.4).

115
For this study, we used in-situ measurements of VOCs for the summertime and wintertime of 2018 in Houston industrial regions from Photochemical Assessment Monitoring Stations (PAMS) operated by the Texas Commission on Environmental Quality (TCEQ) auto-GC system. These datasets were reported by Texas Air Monitoring Information System (TAMIS), and the analytical method and procedure for method detection limits (MDLs) are available at EPA technical air pollution resources. The summer season is defined as June, July, and August (JJA) and 120 the winter season as December, January, and February (DJF). Among the auto-GC sites, data from Channelview, HRM #3 Haden Road, Clinton, and Lynchburg Ferry was compiled for this study (Fig. 1).
The Channelview site is located on the northeastern curve of the Houston Ship Channel at the east of Downtown Houston. Haden Road site is on the top of Houston Buffalo Bayou at the east of Channelview. Clinton site is located near Port of Houston surrounded by both residential and industrial areas, and the Wallisville Road site is near Baytown 125 city on the road to an exurb northeast of Houston. The Lynchburg Ferry site is located in a park and recreation area on a peninsular stretching into the HSC, exposed to a complex mixture of various emission sources across the Houston Ship Channel. For experimental details, see Leuchner and Rappenglueck (2010)  Out of 48 compounds measured by the auto-GC system, seven species 2-methyl-2-butene, n-heptane, cyclohexane, 2methylhexane, 1-butene, n-undecane, and 1.2.3-Trimethylbenzene had to be excluded from the datasets due to their 140 smaller signal/noise ratio or they had > 25% of missing values.
A list of alkane, alkene, alkyne, and aromatic tracers for this study is shown in Table 1, here exemplary for the Lynchburg Ferry site. Based on the average values of the summertime and wintertime concentrations in this Table, these measurements were dominated by alkanes, followed by alkenes and aromatics.
The concentrations of each VOC class changed from station to station (Fig. 2). The total concentration of summertime 145 VOC level was highest at Lynchburg Ferry and HRM #3 Haden Road. But the measurements of alkane, alkene, and aromatic compounds show higher concentrations during the wintertime. At almost all sites, alkanes were significantly higher in wintertime compared to summertime. The plots of alkenes and aromatics, however, showed less seasonal differences in concentrations than alkanes, albeit they have a shorter atmospheric lifetime. The alkane concentrations differences between the summertime and wintertime could be explained by the longer lifetimes of ethane and propane, 150 the two most abundant alkanes. Assuming limited seasonal changes in emissions, the more rapid reaction of the alkenes and aromatics with radicals would cause their seasonal differences less variant. Because of the higher average concentrations of alkenes and aromatic compounds at the Lynchburg Ferry station and the higher number of PCAresolved factors during both summer and winter 2018 (Table S2), along with the strategic location of the site amidst a crucial petrochemical region in the United States, we analyzed and characterized the measured VOCs at that site.

160
A multivariate statistical technique, PCA transforms an original group of inter-correlated variables into a new group of independent and uncorrelated variables (Henry and Hidy, 1979;Sadeghi et al., 2020). To perform the PMF, we used the PCA method to reduce the original number of samples into the required number of factors. For the factor analysis, we used the Kaiser-Meyer-Olkin (Kaiser, 1970) and Bartlett's tests to evaluate and confirm the adequacy and sphericity of the measured dataset.

Positive matrix factorization (PMF)
Receptor modeling is a source apportionment technique that elicits information about the sources of air pollutants (Hopke, 2003). Receptor models are based on the application of a mass balance analysis on measured mass concentrations (Hopke, 2016). Based on mass conservation, the PMF receptor model is an advanced multivariate factor analysis tool commonly used to identify and quantify the contributions of primary sources of ambient 170 measurements. The mathematical theory and principles of the model were extensively elaborated by Paatero and Tapper (1994).
This statistical model consists of decomposing the matrix of measured ambient data into two factor profile matrices and contributions. Eq (1) briefly defines this principle:

175
where represents ambient data concentrations, in which the number of samples and chemical species were measured, is the factor contribution of the k th factor to the i th sample, is the factor profile of the j th species in https://doi.org/10.5194/acp-2021-565 Preprint. Discussion started: 26 July 2021 c Author(s) 2021. CC BY 4.0 License.
the k th factor, and represents the residuals that the model cannot fit. The goal of PMF is to find the non-negative matrices of the factor contribution and profile that lead to the minimum value of the objective function Q, defined as Eq (2), where is an estimate of the uncertainty in chemical species of the sample (Paatero, 1999): (2) A weighted least square method of PMF minimizes the differences between the measurement values and the modelcalculated data. This algorithm constrains all of the matrix components of the factor contribution and profiles to ≥ 0.
We have used the EPA PMF 5.0 to identify the sources of VOCs measurements (Norris et al., 2014).
For the uncertainty of the species, we compared the concentration values to the method detection limit (MDL). If the 185 concentrations were lower than or equal to the MDL, uncertainty calculation was based on Eq. (3): (3) If the species concentration were higher than the MDL, based on the concentration, MDL, and the error fraction (Norris et al., 2014), the uncertainty would be estimated by Eq. (4):

Ozone formation potential
Tropospheric ozone production depends on the availability of NOx and VOC. The contribution of individual VOCs to ozone formation largely depends on their concentration and their reactivity with the hydroxyl radical, which both cover a wide range (Table 1). The ozone formation potential (OFP) is an ozone sensitivity indicator that describes the relative effects of VOCs on ozone formation in the troposphere and identifies key sources and species for ozone 195 formation. OFP scales have been widely used to evaluate the roles of different organic compounds in ozone formation (Kumar et al., 2020;Li et al., 2019;Marvin et al., 2021;Xie et al., 2008). In this study, we used two methods, i.e., the propylene-equivalent weighted concentration and maximum incremental reactivity MIR-weighted concentration, to evaluate the O3 formation potential by VOCs kinetic reactivity and mechanism reactivity, respectively (Duan et al., 2008). The propylene-equivalent method estimates the OFP of each VOC based on its kinetic reactivity with hydroxyl 200 radical normalized against the corresponding reactivity of propylene (Chameides et al., 1992); reaction mechanisms between VOCs, peroxide radicals, and NO are ignored (Zou et al., 2015). The calculation of the propylene-equivalent concentration for each VOC is shown in Eq. (5): where , − denotes the propylene-equivalent concentration (ppbC) of species , ( ) represents the 205 carbon atom concentration (ppbC) of species , ( ) and ( ) denotes the chemical reaction rate (cm 3 molecule -1 s -1 ) of species and propylene with OH radical at 298 K, respectively ( concentration, proposed by Carter (1994), considers the impacts of different reaction mechanisms and VOCs/NOx ratios on ozone formation. Therefore, it represents both the reactivity of individual VOCs to oxidation by OH radicals and the capacity of these VOCs to ozone formation as follows (Atkinson, 2000): where , is the ozone production potential for species , represents the molecular mass of ozone, and represents the relative molecular mass of species in the VOCs. The is the maximum incremental reactivity coefficient of j th VOC. These values were estimated by selecting the MIR value for each of the VOCs from modeled scenarios conducted for Los Angles in 1980 (Carter, 1994;Dodge, 1984). Here, we use the most recent MIR values 215 for VOCs provided by Carter (2010), defined as grams of ozone formed per gram of VOC emitted. The OH reaction rate constants and MIR coefficients for 41 VOCs are shown in Table 1.

Determination of the concentration weighted trajectory (CWT)
To localize potential regional sources of measured hydrocarbons at the auto-GC sites in the Houston Ship Channel, we used the CWT algorithm, which is a function of VOC concentrations reported every 2 hours, and the residence 220 time of the trajectory crossing at each grid cell. We prepared one trajectory every measurement hour for the summertime and wintertime of 2018. Prior to using the CWT, we acquired back trajectories for each two hour measurement of VOCs by the National Oceanic and Atmospheric Administration (NOAA) Hybrid Single-Particle Lagrangian Trajectory (HYSPLIT) model (Stein et al., 2015). The 12 back trajectories with two hour intervals, starting from 00:00 to 22:00 (Central Standard Time), were derived from the Global Data Assimilation System wind field data 225 archive from the National Weather Service National Center for Environmental Prediction. To investigate the impacts of turbulence created by hot emission plumes at the site, we considered varying boundary layer heights. As shown in Fig. S4, CWT results did not change for different altitudes. Thus, the arrival height was set as 100 meters above ground levels, the height at which the impact of turbulence created by hot emission plumes is minimized. In the following equation (5), each grid cell received a weighted concentration by averaging the sample concentrations of each of the 230 PMF-resolved factors with associated trajectories that passed that grid cell as follows: where C represents the weighted average concentration in a grid cell ( , ), C is the measured concentration at the sampling site during day , and τ is the residence time of the 24 backward trajectories corresponding to the day in grid cell ( , ) (Liu et al., 2021).  showed the greatest differences in VOC concentrations between the seasons. These two alkanes together formed 27% and 52% of total VOCs in the summertime and wintertime, respectively.
With regard to the differences between the concentrations of measured compounds in summer and winter, there are 245 several factors such as emission sources, convective mixing, and photochemical reactions which caused the seasonal variations. With regard to the emission sources, an annual peak reflecting fluctuations in energy demand occurs during the summer months with the increased use of electricity for air conditioning. Higher daytime convective boundary layer due to turbulence likely led to reduced summertime concentrations due to vigorous turbulent mixing and dilution compared to wintertime conditions. With regards to the photochemical reactions, stronger solar radiation and higher 250 temperatures accelerate the photochemical reaction processes of VOCs, mostly by hydroxyl radicals (Atkinson et al., 1997). These processes result in stronger removal of VOCs during the day in the summertime. Altogether, these factors contribute to lower VOC concentrations in the summer than in the winter.

2. Factor profiles by the PMF analysis
The PCA model for the Lynchburg Ferry site indicated seven and six factors, in the summer and winter, respectively.

255
The results of the PMF modeling reveal the sources of measured VOCs in the Houston Ship Channel in 2018 (Fig. 3).

260
The PMF algorithm treats the measured organic compounds as inert species, which adds some limitations to the interpretation of the results. These limitations are due to the simplification of processes such as mass production and atmospheric removal, e.g., loss of organic compounds through a chemical reaction, formation of secondary organic aerosols, and dry deposition. The method, however, provides supportive and reliable information concerning the local  As shown in Fig. 3, three resolved factors of VOCs turned out to be shaped by the high presences of alkanes. Factors 1 and 3 were consistent with the emissions of ONG-related hydrocarbons. The first factor, ONG long-lived species, 275 was more dominated by less-reactive species, whereas the third factor, with more highly-reactive species, was identified as ONG short-lived species. The factor ONG long-lived species was comprised of the high mixing ratios of simple alkane compounds. This factor contains ethane (69.0% and 66.3%) and propane (73.9% and 58.2%) for the summertime and the wintertime, respectively. Because of the high abundances of ethane and propane, this factor represents anthropogenic fossil fuel emissions, which are the primary contributors of these two hydrocarbons,

280
according to most global inventories (Dalsøren et al., 2018;Koppmann, 2008). Previous studies have indicated the contributions of fossil fuel emissions to elevated ambient levels of ethane and propane in Houston (Buzcu and Fraser, 2006;Jobson et al., 2004). Our findings indicate that the ONG long-lived species played a significant role in both seasons, and it is unlikely that other anthropogenic sources contributed to ethane and propane emissions at the Lynchburg Ferry site.

285
In the summertime, the VOC composition of factor 2, which includes n-butane (51.1%), isobutane (26.7%), n-pentane (62.8%), and iso-pentane (52.0%) of their total percentages, hints at fuel evaporation in line with earlier findings (Leuchner and Rappenglueck, 2010). By comparison, in the wintertime, n-butane, iso-butane, n-pentane, and iso-  (Berger and Anderson, 1981;Warneke et al., 2014). While the alkane-based factors, -ONG long-lived, fuel evaporation, and ONG short-lived factors-are associated with the emissions related to petrochemical industries and natural gas, these three PMF-resolved factors 295 could be classified and distinguished in terms of their photochemical lifetimes. In the summertime, the contributions of these factors were 27.2%, 19.2%, and 6.7% of total measurements for ONG long-lived, fuel evaporation, and ONG short-lived species, respectively. In the wintertime, the contributions of these factors were higher (50.9%, 24.2%, and 4.8%, respectively).
Therefore, compared to C2-C5 isomers, the C8-C10 isomers would be removed about 2.5 times faster on average than 305 C2 and about 40 times on average faster than C5.
The diurnal average of concentrations of PMF resolved factors appear in Figs. 4 and 5. The pattern of diurnal variations of the three ONG species factors is consistent with their degradation rate of reaction so that more reactive species cause more variant diurnal profiles of the mixing ratios than less reactive species. In the wintertime, ONG short-lived species and fuel evaporation decreased because of the effects of lower temperature; the ONG long-lived 310 species factor, however, underwent only a moderate change. The summertime diurnal patterns of these three alkanebased factors could also be explained by the degradation rate of alkanes, except for ONG long-lived species as with longer lifetimes physical processes such as transport and dilution would have a greater impact on their ambient mixing ratios. While these factors include tracers, which are less reactive to OH, i.e., ethane and propane, they also showed lower concentrations during the daytime hours (Fig. 4). A potential reason for this decrease could be the contribution with incomplete fuel combustion in urban areas (Koppmann, 2008). The high percentage of 1,3-butadiene (61.8%) is 325 most likely due to industrial releases (Czader and Rappenglueck, 2015).  Factor 5 of the summertime was dominated by two compounds: ethylene (82.1%) and propylene (90.8%) (Fig. 3). The Although the factors of one to five showed similar characteristics in both the summertime and the wintertime, factors 6 and seven for the summertime and factor 6 for the wintertime presented different features. Summertime factor 6 was 345 exclusively shaped by the highest percentage of acetylene (90%). Acetylene is known as a tracer for the combustion processes of fossil fuels, agricultural and domestic burning, and wildfires (Nicewonger et al., 2020). Acetylene  Fig. 3). It is well-known that deciduous trees emit isoprene, with emission rates critically depending on solar 355 radiation and temperature (Guenther et al., 1994;Sanadze, 2004;Sharkey et al., 2008;Sharkey and Singsaas, 1995).

Source signature of emitted organic compounds
Ratios of VOCs provide helpful insights into the characterization and photochemical processes of emission sources (Wilde et al., 2021). Due to their similar OH reactivity of isobutane and n-butane, their ratio is independent of air mass 360 dilution and mixing and can indicate what sources they have been presumably emitted from, as long as not different emission sources overlap and reactions with Cl are negligible. As seen in Section 3.3, emissions of butane isomers could be associated with natural gas, LPG, vehicular emissions, and biomass burning (Zheng et al., 2018). The ratios of isobutane to n-butane usually vary according to the specific source: 0.2-0.3 for vehicular exhaust emissions, 0.4-0.5 for LPG, and higher than 0.6 for natural gas (Buzcu and Fraser, 2006;Russo et al., 2010). Figure 6 shows that this 365 ratio is equal to the slope of a linear two-sided fit of a correlation plot. The ratios of iso-butane to n-butane were within the range of reported emissions from LPG (0.51 in the summertime and 0.44 in the wintertime, respectively), which was similar to the results of PMF in Fig. 3. Iso-pentane and n-pentane, two other positional isomers, have similar physical and chemical features of reactivity with OH radicals, and their ratios can provide additional information about the source signature of their emissions (Gilman et al., 2013) provided that reactions with NO3 are negligible (Stutz et al., 2010). Studies have reported that iso-pentane to n-pentane ratios of 0.8 to 0.9, ~2.2 to 3.8, 1.5 to 3.0, and 1.8 to 4.6 could correspond to the origins of NG, vehicular exhausts, liquid gasoline, and fuel evaporations, respectively (Gilman et al., 2013;McGaughey et al., 2004;Watson et al., 2001). An iso-pentane to n-pentane slope of 1.3 (summertime) and 1.6 (wintertime) in this study suggested that the measured pentane tracers likely stemmed from mixed sources of fuel evaporation and emissions from natural gas (Fig. 6). This assumption is consistent with the high 375 percentages of both compounds of iso-pentane and n-pentane in the source profile of factor 2, fuel evaporation (Fig.   3).

Photochemical reaction of organic compounds in the Ship Channel and the effects of VOCs on ozone formation
In addition to source signatures, the ratios of organic tracers can indicate the photochemistry of VOC compounds.
Benzene, toluene, ethylbenzene, and xylenes (BTEX) are among the organic compounds known to play a significant role in the chemistry of the atmosphere (Wallace et al., 2018). BTEX is a unique group, as they exclusively react with urban areas (Rappenglueck et al., 1998;Rappenglueck and Fabian, 1999;Winkler et al., 2002) and to estimate OH concentrations as well as the primary OH production rate (Rappenglueck et al., 2000). Consequently, we assumed that 390 the ratios of benzene to toluene and ethylbenzene to m.p-xylene would be higher in the summertime. Indeed, Figure   6 supports our assumption of the seasonal variation of benzene to toluene. The figure shows the general photochemical behavior in the Ship Channel. The ratio of benzene to toluene notably increased from the wintertime (0.95) to the summertime (3.10). The ethylbenzene/m.p-xylene ratio, however, changed only slightly between the two periods, suggesting that the impacts of source emissions or downwind air plumes affected seasonal variation. The analysis of 395 the nighttime data of the m.p-xylene to ethylbenzene ratio (4.3-4.9) show some proximity to ranges reported for near crude oil refineries (Baltrėnas et al., 2011).

400
In a further investigation of the photochemistry of ethylene and propylene, we examined ratios of the measured ethylene and propylene to acetylene concentrations. Depending on their emission strengths, ethylene and propylene, two highly reactive species with considerable reaction rates with OH and O3, influence the atmospheric chemistry of industrial areas. As acetylene, however, has a lower rate of reaction, the ratio of the diurnal concentrations of ethylene to acetylene could explain their influences on photochemical reaction processes. Figure 7 shows the average diurnal  (Table S2). Calculations were based on the mean mixing ratios listed in Table 1., which would also better reflect the occurrence of transient high values relevant to ozone exceedance events compared to the median data.

420
Among the top 10 reactive species, seven species were the same between these two methods in both seasons, differing only in the order of their rank. In the summertime, these top 10 species, as shown in Table 2, together accounted for 87.3% and 90.4% of the total ozone formation potential using propylene-equivalent and MIR methods. In the wintertime, the top reactive species of propylene-equivalent weighted concentration and method methods together accounted for 76.6% and 85.5% of total ozone formation potential in propylene-equivalent and MIR methods, 425 respectively. The results of Table 2 also show that six compounds out of the top 10 potential species to the ozone formation are similar in summer and winter. These six species include propylene, ethylene, iso-pentane, n-pentane, n-butane, and 430 m.p-xylene that together contribute to (62.8% and 52.2%) of summertime ozone formation and (54.6% and 67.8%) of wintertime ozone formation using propylene-equivalent and MIT methods, respectively. Propylene and ethylene showed the highest contributions to the formation of ozone in summer and winter. This result acknowledges the earlier findings (Buzcu and Fraser, 2006;Leuchner and Rappenglueck, 2010;Zhao et al., 2004)

4. Use of backward trajectory to identify the geographic origins of VOCs over the Ship Channel
To investigate the origins of VOCs in the region, we applied the CWT method, a widely used receptor-based model that spatially reflects the concentration levels of trajectories and explores the potential geographic origins of identified source locations (Pouyaei et al., 2020;Stein et al., 2015). Since the receptor location was inside the Ship Channel, an 440 area responsible for substantial air pollution in the region, we focused on studying local sources rather than on regional transport (Fig. 8).
During summertime 2018, back-trajectory plots revealed that the dominant cluster of trajectories originated south and southeast of the receptor location (Fig. 9). CWT results also showed that the majority of highly weighted concentration values covered the southern regions of the receptor. Upon further investigation, however, we found evidence of specific source locations in the study region. The CWT provided spatial information on the influences of VOC emissions from various industrial sectors and residential areas on the receptor site in the Ship Channel. For factor one,

450
ONG long-lived species, we identified ship traffic and fuel tanks in the area as primary source locations (Fig. 9).

485
An overall comparison of the CWT results for the two seasons showed similar potential sources for most of the distributed VOCs. The geography of source emissions for some factors (i.e., ethylene and propylene, and vehicular exhaust) might have varied due to seasonal changes in terms of the wind direction. The ONG long-lived species factor in both seasons indicated both Baytown and Galveston Bay refineries as potential origins; the potential source of the ONG short-lived species in each season, however, differed. The results of factor 2 of fuel evaporation, factor 3 of 490 ONG short-lived species, factor 4 of aromatic, and factor 5 of ethylene and propylene show a similar spatial distribution of the source emissions between two seasons, although the seasonal discrepancies could be due to different wind patterns, affecting the transport of the VOCs over the Ship Channel.

4-Summary
For two sampling periods, summertime and wintertime of 2018, we investigated the characteristics of the VOCs over 495 the Houston Ship Channel using receptor modeling and the CWT dispersion model. We evaluated the levels and compositions of VOC compounds, identified their emission sources, and identified their possible origins. Our results found that average VOCs concentrations were 28.68 ppbC in the summertime and 33.92 ppbC in the wintertime at the Lynchburg Ferry site. In both seasons, the dominant groups of VOCs were the alkane compounds which contributed to the 61% and 83% of total VOC concentrations in summertime and wintertime, respectively. Studying the source 500 profiles and diurnal variations of the PMF-resolved factors identified three types of ONG emission factors, which differ in their reaction rates. The source apportionment identified two other factors, -aromatics, along with ethylene and propylene-that were present in both summertime and wintertime. Acetylene and biogenic emission factors were found to be present for the summertime, and vehicular exhaust was shown to be effective for the wintertime. In terms of reactivity-based concentration of VOCs, alkene had the highest ozone formation potential in summertime propylene 505 https://doi.org/10.5194/acp-2021-565 Preprint. Discussion started: 26 July 2021 c Author(s) 2021. CC BY 4.0 License.
We also examined the geographical location origins of the emissions sources of VOCs compounds using trajectory analysis. During both seasons, the measurements of VOCs at the Lynchburg Ferry site were influenced by emissions 510 from the petrochemical sector and industrial complexes within the Houston industrial area, especially the Baytown refineries and Bayport industrial district. The results also showed parts of the VOCs at the site originated from the Galveston Bay refineries. The CWT analysis results also indicated parts of the measured aromatic compounds were affected by the plumes of air masses transported from the south of the receptor where the Bayport industrial district is located.

515
The findings of this study suggest two different policy approaches: (1) targeting the alkane emissions sources regarding their higher contributions to the total organic compounds or (2) focusing on other groups of VOCs (e.g., aromatics and alkenes), which are less relevant to the energy utilization policies and are more linked to the industrial productions. Findings from prior studies for the Houston area have led to significant emission reductions in highly reactive VOCs over the last 15 years. However, the results of this study admit the importance of the ethylene and 520 propylene emissions which continue to be an issue for the mitigation of ozone.

Declaration of competing interest
The authors declare no competing financial and/or non-financial interests in relation to the work described.

Acknowledgments
This study was partially supported by the High Priority Area Research Seed Grant of the University of Houston.