Principal component analysis/absolute principal component scores (PCA/APCS) and positive matrix factorization (PMF2), an advanced factor analysis technique were employed to apportion the sources influencing the PM2.5 levels measured during 2003 through 2005 at a rural coastal site located within the Corpus Christi urban airshed in South Texas. PCA/APCS identified five sources while PMF2 apportioned an optimal solution of eight sources. Both PCA/APCS and PMF2 quantified secondary sulfates to be the major contributor accounting for 47% and 45% of the apportioned PM2.5 levels. The other common sources apportioned by the models included crustal dust, fresh sea salt and traffic emissions. PMF2 successfully apportioned distinct sources of fresh and aged sea salt along with biomass burns while PCA/APCS was unsuccessful in identifying aged sea salt and biomass burns; however it successfully identified secondary organic aerosols from photochemical oxidations and also emitted by petrochemical refineries. The influence of long range transport was noted for sources such as secondary sulfates, biomass burns and crustal dust affecting the region. Continued collection of speciation data at the rural and urban sites will enhance the understanding of local versus regional source contributions for air quality policy makers and stakeholders.
Epidemiological studies conducted over the past decade have provided ample confirmation of the adverse effects of PM2.5 on human health and welfare, and thus reiterating the need for source identification and quantification. Scientific studies over the past decade focussed on multiple techniques of source apportionment of PM2.5 measured in urban and rural locations with wide variations in geographic, climatic and emission conditions. Chemical mass balance (CMB), one of several traditional source receptor models requires input of specific source profiles and thus is regarded to be highly subjective. On the other hand, statistical models including principal component analysis/absolute principal component scores (PCA/APCS), UNMIX and positive matrix factorization (PMF) are data driven and need no prior knowledge of site specific source profiles and thus have been widely applied in several air quality studies spread across the globe [1-4].
Principal component analysis (PCA)/absolute principal component analysis (APCS) [
Prior studies have demonstrated the spatial variability in the chemical composition of PM2.5 measured at urban and rural monitoring sites [19-22]. Researchers thus have employed the source receptor models not only to identify and quantify local anthropogenic sources but also to study the impact of long-range transport from highly polluted regional sources [4,23,24].
Corpus Christi, located along the Gulf of Mexico in South Texas, is a fast growing industrialized urban region. A more recent comprehensive source apportionment study using PMF2 on the PM2.5 speciation data measured during 2003 through 2008 at an industrialized urban site in Corpus Christi showed the influence of both local anthropogenic sources and long-range transport from regional sources [
The Corpus Christi urban airshed is the eighth largest Consolidated Metropolitan Statistical Area (CMSA) in Texas and is home to a cluster of energy related Industries adjacent to the sixth largest port in USA. TCEQ operates and maintains about sixteen continuous ambient monitoring stations measuring ozone, particulate matter, meteorological parameters and other pollutants including sulphur dioxide (SO2), hydrocarbons, and oxides of nitrogen (NOx) within this urban airshed.
Continuous ambient monitoring station (CAMS) 314 is situated in a rural location at 20420 Park road (27˚25'N, 97˚17'W) within the Padre Island National Seashore park and is classified as a coastal rural site. TCEQ has collected twenty four hour averaged PM2.5 filter samples once every six days at CAMS 314 during January 2003 through August 2005 for chemical characterization purposes.
The samples were collected on pre-conditioned and pre-weighed 47-mm diameter Whatman Teflon and quartz filters. Gravimetric analysis of the Teflon filters was performed to measure the mass of PM2.5 collected. The filters were then stored at 4˚C and shipped to Research Triangle Institute (RTI), North Carolina for further chemical analysis. Filter samples acquired were preconditioned to room temperature and were analyzed for elements (energy dispersive x-ray fluorescence: ED-XRF), water soluble ions (ion chromatograph: IC) and carbon species (total optical transmittance: TOT). Speciation data along with method detection limits and analytical uncertainty were then reported to TCEQ. For the purpose of this study, the speciation data set measured at the rural site (CAMS 314) during January 2003 through August 2005 was acquired from TCEQ.
PCA is a statistical model which identifies principal components or factors to explain the variance in the provided
raw or correlated data [
PMF is an advanced factor analysis technique developed for source apportionment analysis [
The speciation dataset acquired from TCEQ consisted of concentrations of 55 species where some species typically recorded below method detection limit (MDL). Traditional signal-to-noise (S/N) ratio approach applied by other researchers was employed to select key species for further analysis [22,29]. Species with S/N ratio greater than 2.0 were classified as strong and those with S/N ratios ranging from 0.2 to 2.0 were identified as moderately weak variables and only these species were used in the analysis as shown in
used, and for those below MDL the value was set to 5/6 of the MDL and for those with missing values, it was set to four times the arithmetic mean. With the processed concentration and uncertainty files as input the model was run in a robust mode to reduce the impact of any outliers in the data. Iterative runs of PMF2 were performed to identify the solution with optimal objective function value “Q” (which is defined as Q = No. of species × No. of samples). Further iterative runs with varing FPEAK were performed to identify the optimal solution with a minimum rotational ambiguity [3,22,29,30].
Conditional probability function analysis is the probability measure of the concentrations in a wind sector being above a certain threshold level. It is characterized as the ratio between the number of samples in a wind sector above the given threshold to the total number of samples in that sector [25,30]. The prevailing meteorological conditions including the resultant wind speed and resultant wind direction measured during the sampling days were acquired from TCEQ’s website for CAMS 314. The hourly wind speed and wind direction data were then coupled with the APCS and PMF2 apportioned concentrations and sorted into wind sectors of 30˚ bins. The 75th percentile of apportioned concentrations was then used as the threshold to estimate the probability associated with each sector for a source [25,30].
Corpus Christi urban airshed is currently in compliance with both the primary and secondary NAAQS for PM2.5, however a gradual increasing trend has been observed over the years. The annual average filter mass concentrations of ambient PM2.5 (based on the total mass from filter samples) measured at CAMS 314 during 2003 through 2005 were observed to be 7.19 µg×m−3, 8.06 µg×m−3 and 10.77 µg×m−3, respectively.
Filter samples were analyzed for a total of 55 PM2.5 species. Statistical analysis was performed on the speciation data to identify the major chemical constituents of PM2.5 and is shown in
Sulfates were identified to be the largest chemical component in the measured PM2.5 accounting for approximately 33.9% of the total filter mass. Organic carbon and its constituents including OC1, OC2, OC3, and OC4 were calculated to be the second largest component totaling 25.9% of the measured PM2.5. The other key species included sulfur (11.7%), ammonium ion (10.5%), chloride ion (5.3%), sodium ion (4.5%), EC (4.1%), and nitrate ion (2.3%). Trace species such as Ba, Br, As, Hg and Pb accounted for the remainder (1.8%).
Seasonal variations in the chemical compositions were also observed at these sites indicating an impact of both local anthropogenic emissions and long range transport during regional pollution events as noted in earlier studies [25,31]. Data driven factor analysis models including PCA/APCS and PMF2 were employed to apportion the various sources influencing the rural site. The seasonal variations in the apportioned mass and the influence of local emissions versus transported levels of pollution affecting the measured PM2.5 concentrations within the study region were further analyzed.
PCA identified five principal component scores with eigen values greater than 1.0 (Kaiser criterion) explaining 84% of variance in the measured PM2.5 levels. Source profiles were classified using the factor loadings of species greater than 0.5 as employed in earlier source apportionment studies [6,32]. The factor loadings, eigen values along with source classification and percent contribution are shown in
As shown in
marily sulfates contributed by local anthropogenic sources such as petrochemical refineries and also associated with long-range transport during regional haze events, have been identified and quantified as the major contributor of PM2.5 in this study. Similar findings were also reported in earlier studies and this factor was classified as “Secondary sulfates” [32,33]. APCS identified secondary sulfates as a major contributor to the observed PM2.5 concentrations and it accounted for 47% of the apportioned mass.
PC2 with significant presence of elements such as Ca, Fe, Si and K, as shown in
Secondary organic aerosols are produced primarily by photochemical oxidation of volatile organic compounds from petrochemical complexes, oil refineries, mobile emissions and biomass burning. Thus, PC3 characterized by OC, OC1, OC2, OC3 and OC4 (
PC4 with high factor loadings of Na+ and Cl as shown in
Traffic sources have typically been characterized with OC, EC, Cu, Zn, and contributed by the vehicle exhaust along with minor contribution of wear and tear in brake line and tires with composition of Cu, Zn and Cr [7,19]. PC5 with high factor loadings of along with the presence of OC and EC with 6% of variance were classified as “Traffic emissions” which accounted for 9% of the PM2.5 mass as shown in (
PMF2 identified an optimal solution of eight sources with minimum rotational ambiguity at FPEAK 0.0 influencing the ambient PM2.5 levels measured at the rural coastal site (CAMS 314).
Secondary sulfates were identified to be the major contributor of PM2.5 concentrations measured at the rural coastal site as reported by other researchers [29,36,37]. PMF2 apportioned two sources of secondary sulfates. Source 1 rich in S, , and P was classified as “Secondary sulfates I” which accounted for 12% of the apportioned mass while source 2 contributing 33% with significant composition of S, , and was classified as “Secondary sulfates II (
Seasonal variations were noted in secondary sulfates-I with higher concentrations during summer months while elevated concentrations of secondary sulfates-II were recorded during fall suggesting the influence of two separate sulfate sources. Source 3 accounting for 13% of the apportioned mass was identified to be the second largest source influencing PM2.5 levels measured at the coastal rural site as shown in
Air borne soil and crustal dust sources with elemental composition including Ca, Fe, Ti and Si has been identified by several studies influencing the urban and rural air quality [17,29]. Thus, source 7 in the current study with similar chemical composition as shown in
Heavy oil usage for combustion in the compressors has been identified as the primary source of metals includeing V and Ni along with OC and its components [32,33]. Oil and natural gas exploration activities have been reported on nearby Padre Island along with off shore operations in the Gulf of Mexico.
Thus, source 7 with rich composition of V, OC and its components including OC1, OC2, OC3 and OC4 was classified as “Combustion source”, which accounted for 4% of the apportioned PM2.5 mass.
The CPF analysis of the sources apportioned by PCA/ APCS and PMF2 influencing the PM2.5 levels measured at CAMS 314 are shown in
The CPF analysis of traffic emission source apportioned by both PMF2 and PCA/APCS at CAMS 314 showed the influence of the northeast and northwest wind sectors. Vehicular emissions from a nearby recreational vehicle (RV) park located to the northeast of the monitoring site along with the heavy oil combustion in offshore oil and natural gas exploration activities were identified to be the probable contributors of this source. PMF2 apportioned combustion source and PCA/APCS apportioned secondary organic aerosol consisted of significant loadings of organic carbon and its components along with V. As shown by the CPF analysis (
The CPF analysis of the two unique sources apportioned by PMF2 including aged sea salt and biomass
burns along with secondary sulfates I are shown in
PMF2 apportioned two distinct sources of secondary sulfates contributed by anthropogenic emissions and photochemical formed aged aerosols while PCA/APCS aggregated both into one single source. The study area is located along the coast and thus significantly influenced by marine aerosols including both fresh and aged. PCA/APCS could not distinguish between fresh and aged sea salt, however PMF2 was successful in apportioning
distinct sources for both aerosol types. In addition, the PCA/APCS was unsuccessful in apportioning the biomass burns. PCA identified secondary organic aerosol source with significant loadings of organic carbon components which was apportioned by APCS to be the second largest source (20%) influencing PM2.5 levels measured at CAMS 314, while PMF2 apportioned a similar source contributing only 4% of the mass. Wood combustion also results in organic carbon thus, PMF2 apportioned significant levels of organic carbon and its components into the biomass burns source resulting in a variation in naming the sources and their distinct contributions.
A larger speciation dataset (July during 2003 through December 2008) at the industrialized urban site was used in an earlier source apportionment study conducted by Karnae and John [
The primary limitation of the current study was the lack of availability of speciation data points beyond August 2005 at the coastal rural site due to discontinuation of sampling by TCEQ. Continued collection of speciation data at the rural site will most certainly enhance our understanding of local versus regional source contributions of PM2.5.
PCA/APCS apportioned five sources explaining 84% of variance in the PM2.5 concentrations measured at the rural coastal monitoring site (CAMS 314) adjacent to the Corpus Christi urban airshed. PMF2, the advanced factor analysis model, apportioned an optimal solution of eight sources at CAMS 314. Secondary sulfates were apportioned to be the major contributor influencing the PM2.5 levels measured at CAMS 314 accounting for 47% by PCA/APCS. PMF2 also identified secondary sulfates to be the major contributor however it was successful in identifying two distinct sources including “Secondary sulfates-I” contributed by local emissions (12%) and “Secondary sulfates-II” contributed by the photo chemically aged aerosols (33%). PCA/APCS apportioned seconddary organic aerosols with higher factor loadings of organic carbons contributed to 20% of the apportioned PM2.5 mass. PMF2 also apportioned secondary organic aerosols accounting for 4% of the mass. PMF2 apportioned significant levels of organic carbon and it’s components into the biomass burn source resulting in lower percentage of contribution by secondary organic aerosol source as compared to PCA/APCS. PMF2 apportioned traffic emissions source accounting for 10% while PCA/ APCS quantified it as 9%. The major contributors of the source were a nearby RV park along with the vehicular traffic on the beach. Crustal dust source was apportioned by both models accounting for 7% and 9% of the apportioned mass. PCA/APCS was unsuccessful in apportioning distinct sources of fresh and aged sea salt at both the sites, while PMF2 apportioned the two sources successfully accounting for 11% and 8%, respectively. PMF2 apportioned a unique source of biomass mass accounting for 13% of the apportioned mass contributed by local residential wood combustion and major biomass burn events in Mexico and Central America during the spring months. Using the data available as demonstrated by this study, PCA/APCS has successfully apportioned secondary sulfates and organic aerosols, however was unsuccessful in identifying natural sources including fresh and aged sea salt and the contribution of biomass burns. Both techniques have unique strengths and some identifiable weaknesses. While, the PMF2 resolved a larger number of sources, PCA/APCS resolved fewer unique sources. A hybrid approach to source apportionment will possibly enhance our understanding of the impact and influence of sources affecting the measured ambient PM2.5 concentrations.
The authors would like to thank the Office of Compliance and Monitoring Division at the Texas Commission on Environmental Quality (TCEQ) for providing the PM2.5 speciation data used in this study.