Source Apportionment Using Positive Matrix Factorization on Daily Measurements of Inorganic and Organic Speciated PM(2.5)

Particulate matter less than 2.5 microns in diameter (PM(2.5)) has been linked with a wide range of adverse health effects. Determination of the sources of PM(2.5) most responsible for these health effects could lead to improved understanding of the mechanisms of such effects and more targeted regulation. This has provided the impetus for the Denver Aerosol Sources and Health (DASH) study, a multi-year source apportionment and health effects study relying on detailed inorganic and organic PM(2.5) speciation measurements.In this study, PM(2.5) source apportionment is performed by coupling positive matrix factorization (PMF) with daily speciated PM(2.5) measurements including inorganic ions, elemental carbon (EC) and organic carbon (OC), and organic molecular markers. A qualitative comparison is made between two models, PMF2 and ME2, commonly used for solving the PMF problem. Many previous studies have incorporated chemical mass balance (CMB) for organic molecular marker source apportionment on limited data sets, but the DASH data set is large enough to use multivariate factor analysis techniques such as PMF.Sensitivity of the PMF2 and ME2 models to the selection of speciated PM(2.5) components and model input parameters was investigated in depth. A combination of diagnostics was used to select an optimum, 7-factor model using one complete year of daily data with pointwise measurement uncertainties. The factors included 1) a wintertime/methoxyphenol factor, 2) an EC/sterane factor, 3) a nitrate/polycyclic aromatic hydrocarbon (PAH) factor, 4) a summertime/selective aliphatic factor, 5) an n-alkane factor, 6) a middle oxygenated PAH/alkanoic acid factor and 7) an inorganic ion factor. These seven factors were qualitatively linked with known PM(2.5) emission sources with varying degrees of confidence. Mass apportionment using the 7-factor model revealed the contribution of each factor to the mass of OC, EC, nitrate and sulfate. On an annual basis, the majority of OC and EC mass was associated with the summertime/selective aliphatic factor and the EC/sterane factor, respectively, while nitrate and sulfate mass were both dominated by the inorganic ion factor. This apportionment was found to vary substantially by season. Several of the factors identified in this study agree well with similar assessments conducted in St. Louis, MO and Pittsburgh, PA using PMF and organic molecular markers.     

The accuracy of two- and three-way positive matrix factorization models: Applying simulated multisite data sets

The application of three-way data sets (combined multisite data sets) for source apportionment has become common, but its influence on the performance of receptor modeling techniques has not yet been explored systematically. To study the influence of site-to-site correlations of source contributions and the spatial variability of source profiles on two- and three-way positive matrix factorization (PMF), simulated three-way data sets were constructed and modeled by different applications of PMF (PMF2 for each site individually, PMF2 for data sets combining all sites together, and PMF3 for all sites). In addition, the performance of PMF was evaluated under conditions of collinearity and different source categories at two sites. The results indicated that if the sites were contributed by sources with identical profiles, the site-to-site correlations of source contributions would not influence the PMF2, and the three-way blocks could be used by PMF2. However, the PMF2 using three-way data sets was sensitive to the spatial variability of source profiles. For the three-way model, PMF3 could perform well only when all of the sources exhibited strong site-to-site associations among all sites, and at the same time, the spatial variability of source profiles were sufficiently small. It might due to the algorithm that, for each source, PMF3 produces the same source profile and the same temporal variation in daily contributions among all sites.

Implications:?The application of multisite data sets for source apportionment has become common. However, limited work investigated the accuracy of two- and three-way PMFs when using multisite data sets. If the application of PMFs using multisite data sets were not appropriate, the results would be unreasonable. The unreasonable results would supply confused information for PM control strategies. In this work, simulated multisite data sets were modeled by different applications of PMFs. The effort to assess and compare the performance of two- and three-way PMFs using multisite data sets is very limited. The findings could provide information for multisite source apportionment.     

Comparison between sample-species specific uncertainties and estimated uncertainties for the source apportionment of the speciation trends network data

In order to use the US Environmental Protection Agency's speciation trends networks (STN) data in source apportionment studies with positive matrix factorization (PMF), uncertainties for each of the measured data points are required. Since STN data were not accompanied by sample-species specific uncertainties (SSU) prior to July 2003, a comprehensive set of fractional uncertainties was estimated by Kim et al. [2005. Estimation of organic carbon blank values and error structures of the speciation trends network data for source apportionments. Journal of Air and Waste Management Association 55, 1190–1199]. The objective of this study is to compare the use of the estimated fractional uncertainties (EFU) for the source apportionment of PM_2.5 (particulate matter less than 2.5 μm in aerodynamic diameter) measured at the STN monitoring sites with the results obtained using SSU. Thus, the source apportionment of STN PM_2.5 data were performed and their contributions were estimated through the application of PMF for two selected STN sites, Elizabeth, NJ and Baltimore, MD with both SSU and EFU for the elements measured by X-ray fluorescence. The PMF resolved factor profiles and contributions using EFU were similar to those using SSU at both monitoring sites. The comparisons of normalized concentrations indicated that the STN SSU were not well estimated. This study supports the use of EFU for the STN samples to provide useful error structure for the source apportionment studies of the STN data.     

Source identifications of airborne fine particles using positive matrix factorization and U.S. Environmental Protection Agency positive matrix factorization

The widely used source apportionment model, positive matrix factorization (PMF2), has been applied to various air pollution data. Recently, U.S. Environmental Protection Agency (EPA) developed EPA positive matrix factorization (PMF), a version of PMF that will be freely distributed by EPA. The objectives of this study were to conduct source apportionment studies for particulate matter less than 2.5 microm in aerodynamic diameter (PM(2.5)) speciation data using PMF2 and EPA PMF (version 1.1) and to compare identified sources between the two models. In the present study, ambient PM(2.5) compositional datasets of 24-hr integrated samples collected at EPA Speciation Trends Network monitoring sites in Chicago, IL, and Portland, OR, were analyzed. Both PMF2 and EPA PMF extracted eight sources for the Chicago data and 10 sources for the Portland data. The model-resolved source profiles were similar between two models for both datasets. However, in several sources, the average contributions did not agree well and the time series contributions were not highly correlated. The differences between PMF2 and EPA PMF solutions were caused by the different least-square algorithm and the different nonnegativity constraints. Most of the average source contributions resolved by both models were within 5-95% uncertainty provided by EPA PMF, indicating that the sources resolved by both models were reproducible.     

The application of positive matrix factorization in the analysis, characterisation and detection of contaminated soils

Multivariate factor analytical techniques are widely used for the approximation, in terms of a linear combination of factors, of multivariate experimental data. The chemical composition of soil samples are multivariate in nature and provide datasets suitable for the application of these statistical techniques. Recent developments of multivariate factor analytical techniques have led to the approach of Positive Matrix Factorization (PMF), a weighted least squares fit of a data matrix in which the weights are determined depending on the error estimates of each individual data value. This approach relies on more physically significant assumptions than methods like Principal Components Analysis which is frequently used in the analysis of soil datasets. In this paper we apply PMF to characterise the pollutant source in a set of geographically referenced soil samples taken within a 200 m radius of a site characterised by a high concentration of heavy metals. Each sample has been analysed for major and minor elements (using wavelength-dispersive X-ray fluorescence spectrometry), carbon, hydrogen and nitrogen (using a CHN elemental analyzer) and mercury (using cold-vapour atomic absorption spectrometry). Analysis of the soils using PMF resulted in a successful partitioning of variances into sources related to background soil geochemistry, organic influences and those associated with the contamination. Combining these results with a geostatistical approach successfully demarcated the main source of the combined organic and heavy metal contamination.     

Assessment of source apportionment by Positive Matrix Factorization analysis on fine and coarse urban aerosol size fractions

This study was conducted in order to investigate the differences observed in source profiles in the urban environment, when chemical composition parameters from different aerosol size fractions are subjected to factor analysis. Source apportionment was performed in an urban area where representative types of emission sources are present. PM₁₀ and PM₂ samples were collected within the Athens Metropolitan area and analysed for trace elements, inorganic ions and black carbon. Analysis by two-way and three-way Positive Matrix Factorization was performed, in order to resolve sources from data obtained for the fine and coarse aerosol fractions. A difference was observed: seven factors describe the best solution in PMF3 while six factors in PMF2. Six factors derived from PMF3 analysis correspond to those described by the PMF2 solution for the fine and coarse particles separately. These sources were attributed to road dust, marine aerosol, soil, motor vehicles, biomass burning, and oil combustion. The additional source resolved by PMF3 was attributed to a different type of road dust. Combustion sources (oil combustion and biomass burning) were correctly attributed by PMF3 solely to the fine fraction and the soil source to the coarse fraction. However, a motor vehicle's contribution to the coarse fraction was found only by three-way PMF. When PMF2 was employed in PM₁₀ concentrations the optimum solution included six factors. Four source profiles corresponded to the previously identified as vehicles, road dust, biomass burning and marine aerosol, while two could not be clearly identified. Source apportionment by PMF2 analysis based solely on PM₁₀ aerosol composition data, yielded unclear results, compared to results from PMF2 and PMF3 analyses on fine and coarse aerosol composition data.     

Positive matrix factorization as source apportionment of soil lead and cadmium around a battery plant (Changxing County,China)

Chemical compositions of soil samples are multivariate in nature and provide datasets suitable for the application of multivariate factor analytical techniques. One of the analytical techniques, the positive matrix factorization (PMF), uses a weighted least square by fitting the data matrix to determine the weights of the sources based on the error estimates of each data point. In this research, PMF was employed to apportion the sources of heavy metals in 104 soil samples taken within a 1-km radius of a lead battery plant contaminated site in Changxing County, Zhejiang Province, China. The site is heavily contaminated with high concentrations of lead (Pb) and cadmium (Cd). PMF successfully partitioned the variances into sources related to soil background, agronomic practices, and the lead battery plants combined with a geostatistical approach. It was estimated that the lead battery plants and the agronomic practices contributed 55.37 and 29.28 %, respectively, for soil Pb of the total source. Soil Cd mainly came from the lead battery plants (65.92 %), followed by the agronomic practices (21.65 %), and soil parent materials (12.43 %). This research indicates that PMF combined with geostatistics is a useful tool for source identification and apportionment.     

Sources of Fine Particulate Species in Ambient Air over Lake Champlain Basin,VT

Abstract

This study is a part of an ongoing investigation of the types and locations of emission sources that contribute fine particulate air contaminants to Underhill, VT. The air quality monitoring data used for this study are from the Interagency Monitoring of Protected Visual Environments network for the period of 2001–2003 for the Underhill site. The main source-receptor modeling techniques used are the positive matrix factorization (PMF) and potential source contribution function (PSCF). This new study is intended as a comparison to a previous study of the 1988–1995 Underhill data that successfully revealed a total of 11 types of emission sources with significant contributions to this rural site. This new study has identified a total of nine sources: nitrate-rich secondary aerosol, wood smoke, East Coast oil combustion, automobile emission, metal working, soil/dust, sulfur-rich aerosol type I, sulfur-rich aerosol type II, and sea salt/road salt. Furthermore, the mass contributions from the PMF identified sources that correspond with sampling days with either good or poor visibility were analyzed to seek possible correlations. It has been shown that sulfur-rich aerosol type I, nitrate aerosol, and automobile emission are the most important contributors to visibility degradation. Soil/dust and sea salt/road salt also have an added effect.     

Source apportionment of Phoenix PM2.5 aerosol with the Unmix receptor model

The multivariate receptor model Unmix has been used to analyze a 3-yr PM2.5 ambient aerosol data set collected in Phoenix, AZ, beginning in 1995. The analysis generated source profiles and overall average percentage source contribution estimates (SCEs) for five source categories:gasoline engines (33 +/- 4%), diesel engines (16 +/- 2%), secondary SO4(2-) (19 +/- 2%), crustal/soil (22 +/- 2%), and vegetative burning (10 +/- 2%). The Unmix analysis was supplemented with scanning electron microscopy (SEM) of a limited number of filter samples for information on possible additional low-strength sources. Except for the diesel engine source category, the Unmix SCEs were generally consistent with an earlier multivariate receptor analysis of essentially the same data using the Positive Matrix Factorization (PMF) model. This article provides the first demonstration for an urban area of the capability of the Unmix receptor model.     

High occurrence of BTEX around major industrial plants in Kaohsiung,Taiwan

The Houjing River, which flows by large industrial complexes in southwestern Taiwan, has been seriously polluted with benzene, toluene, ethylbenzene, and xylene (BTEX). Using Spearman’s analysis of BTEX concentrations measured at different sites along the river, we identified the main sources of this pollution to be the Dashe and Renwu Industrial Parks. Maximum concentrations of benzene and toluene (402 and 143.19 µg/L, respectively) were considerably higher than those reported in similar studies and regulatory limits. We compared these findings with those of positive matrix factorization (PMF) modelling. PMF also identified the two industrial parks as being sources of BTEX, most likely originating from petrochemical activities that occur there. This study can serve as an important reference for future watershed management and pollution control plans for Kaohsiung, the most industrialized city in Taiwan. PMF proved to be a reliable computer modelling program for source apportionment and environmental forensic studies.     

Sources of fine particulate species in ambient air over lake Champlain Basin, VT

This study is a part of an ongoing investigation of the types and locations of emission sources that contribute fine particulate air contaminants to Underhill, VT. The air quality monitoring data used for this study are from the Interagency Monitoring of Protected Visual Environments network for the period of 2001-2003 for the Underhill site. The main source-receptor modeling techniques used are the positive matrix factorization (PMF) and potential source contribution function (PSCF). This new study is intended as a comparison to a previous study of the 1988-1995 Underhill data that successfully revealed a total of 11 types of emission sources with significant contributions to this rural site. This new study has identified a total of nine sources: nitrate-rich secondary aerosol, wood smoke, East Coast oil combustion, automobile emission, metal working, soil/dust, sulfur-rich aerosol type I, sulfur-rich aerosol type II, and sea salt/road salt. Furthermore, the mass contributions from the PMF identified sources that correspond with sampling days with either good or poor visibility were analyzed to seek possible correlations. It has been shown that sulfur-rich aerosol type I, nitrate aerosol, and automobile emission are the most important contributors to visibility degradation. Soil/dust and sea salt/road salt also have an added effect.     

Utilizing the Chemical Mass Balance and Positive Matrix Factorization models to determine influential species and examine possible rotations in receptor modeling results

Data from two of the United States Environmental Protection Agency's Speciation Trends Network fine particulate matter sites within Chicago, Illinois were used to examine the influence that the results and profiles of the Chemical Mass Balance (CMB) receptor model have on the source contributions and profiles of the Positive Matrix Factorization (PMF) model. This was accomplished using the target shape technique, which utilizes a priori information from the CMB source profiles inputted into the PMF model. The target shape methodology involves inputting specific information for the source profiles into the PMF model as it is resolving source profile and contribution matrices. The target shape results demonstrated it is possible to determine in both the CMB and PMF source profiles those species, which do not influence the solutions of either model.A second method utilizing information from the CMB results was used to impose a condition where the Motor Vehicles source never had a zero contribution as was applied to the CMB model. This involved utilizing an edge rotation to rotate the PMF results to yield a different solution without worsening the fit of the original results. The purpose of this work is to achieve a rotation, which produced a PMF solution where all of the Motor Vehicles contributions were greater than zero. Comparing the rotated Motor Vehicles and Sulfates source contributions in PMF to those obtained from CMB showed a better correlation between the PMF Motor Vehicles contributions to the original CMB results than those prior to rotation.     

Source apportionment and location by selective wind sampling and Positive Matrix Factorization

In order to determine the pollution sources in a suburban area and identify the main direction of their origin, PM_2.5 was collected with samplers coupled with a wind select sensor and then subjected to Positive Matrix Factorization (PMF) analysis. In each sample, soluble ions, organic carbon, elemental carbon, levoglucosan, metals, and Polycyclic Aromatic Hydrocarbons (PAHs) were determined. PMF results identified six main sources affecting the area: natural gas home appliances, motor vehicles, regional transport, biomass combustion, manufacturing activities, and secondary aerosol. The connection of factor temporal trends with other parameters (i.e., temperature, PM_2.5 concentration, and photochemical processes) confirms factor attributions. PMF analysis indicated that the main source of PM_2.5 in the area is secondary aerosol. This should be mainly due to regional contributions, owing to both the secondary nature of the source itself and the higher concentration registered in inland air masses. The motor vehicle emission source contribution is also important. This source likely has a prevalent local origin. The most toxic determined components, i.e., PAHs, Cd, Pb, and Ni, are mainly due to vehicular traffic. Even if this is not the main source in the study area, it is the one of greatest concern. The application of PMF analysis to PM_2.5 collected with this new sampling technique made it possible to obtain more detailed results on the sources affecting the area compared to a classical PMF analysis.     

Source apportionment of indoor, outdoor, and personal PM2.5 in Seattle, Washington, using positive matrix factorization

As part of a large exposure assessment and health-effects panel study, 33 trace elements and light-absorbing carbon were measured on 24-hr fixed-site filter samples for particulate matter with an aerodynamic diameter <2.5 microm (PM2.5) collected between September 26, 2000, and May 25, 2001, at a central outdoor site, immediately outside each subject's residence, inside each residence, and on each subject (personal sample). Both two-way (PMF2) and three-way (PMF3) positive matrix factorization were used to deduce the sources contributing to PM2.5. Five sources contributing to the indoor and outdoor samples were identified: vegetative burning, mobile emissions, secondary sulfate, a source rich in chlorine, and a source of crustal-derived material. Vegetative burning contributed more PM2.5 mass on average than any other source in all microenvironments, with average values estimated by PMF2 and PMF3, respectively, of 7.6 and 8.7 microg/m3 for the outdoor samples, 4 and 5.3 microg/m3 for the indoor samples, and 3.8 and 3.4 microg/m3 for the personal samples. Personal exposure to the combustion-related particles was correlated with outdoor sources, whereas exposure to the crustal and chlorine-rich particles was not. Personal exposures to crustal sources were strongly associated with personal activities, especially time spent at school among the child subjects.     

Fine particulate source apportionment using data from the USEPA speciation trends network in Chicago,Illinois: Comparison of two source apportionment models

Data from two of the United States Environmental Protection Agency's speciation trends network fine particulate matter sites within Chicago, Illinois were analyzed using the chemical mass balance (CMB) and positive matrix factorization (PMF) models to determine source contributions to the ambient fine particulate concentrations. The results from the two models were compared to determine the similarities and differences in the source contributions. This included examining the differences in the magnitude of the individual source contributions as well as the correlation between the contribution values from the two methods. The results showed that both models predicted sulfates, nitrates and motor vehicles as the three highest fine particle contributors for the two sites accounting for approximately 80% of the total. The PMF model attributed a slightly greater amount of fine particulate to the road salt, steel and soil sources while vegetative burning contributed more in the CMB results. Correlations between the contribution results from the two models were high for sulfates, nitrates and road salt with very good correlations existing for motor vehicles and petroleum refineries. The predicted PMF profiles agreed well with measured source profiles for the major species associated with each source.     

Causes of haze in the Columbia River Gorge

Visibility impairment in the Columbia River Gorge National Scenic Area is an area of concern. A field study conducted from July 2003 to February 2005 was followed by data analysis and receptor modeling to better understand the temporal and spatial patterns of haze and the sources contributing to the haze in the Columbia River Gorge in the states of Washington and Oregon. The nephelometer light scattering and surface meteorological data at eight sites along the gorge showed five distinct wind patterns, each with its characteristic diurnal and spatial patterns in light scattering by particles (bsp). In summer, winds were nearly always from west to east (upgorge) and showed decreasing bsp with distance into the gorge and a pronounced effect of the Portland, OR, metropolitan area on haze, especially in the western portions of the gorge. Winter often had winds from the east with very high levels of bsp, especially at the eastern gorge sites, with sources east of the gorge responsible for much of the haze. The major chemical components responsible for haze were organic carbon, sulfate, and nitrate. Positive matrix factorization (PMF) using chemically speciated Interagency Monitoring of Protected Visual Environments data indicated seven source factors in the western gorge and five factors in the eastern gorge. Organic mass is a large contributor to haze in the gorge in all seasons, with a peak in fall. The PMF analysis suggests that approximately half of the organic mass is biomass smoke, with mobile sources as the second largest contributor. PMF analysis showed nitrates (important in fall and winter) mainly attributed to a generic secondary nitrate factor, with the next largest contributor being oil combustion at Mt. Zion, WA and mobile sources at Wishram, WA. Sulfate is a significant contributor in all seasons, with peak sulfate concentrations in summer.     

Improving source apportionment of fine particles in the eastern United States utilizing temperature-resolved carbon fractions

The objectives of this study were to examine the use of carbon fractions to identify particulate matter (PM) sources, especially traffic-related carbonaceous particle sources, and to estimate their contributions to the particle mass concentrations. In recent studies, positive matrix factorization (PMF) was applied to ambient fine PM (PM2.5) compositional data sets of 24-hr integrated samples including eight individual carbon fractions collected at three monitoring sites in the eastern United States: Atlanta, GA, Washington, DC, and Brigantine, NJ. Particulate carbon was analyzed using the Interagency Monitoring of Protected Visual Environments/Thermal Optical Reflectance method that divides carbon into four organic carbons (OC): pyrolized OC and three elemental carbon (EC) fractions. In contrast to earlier PMF studies that included only the total OC and EC concentrations, gasoline emissions could be distinguished from diesel emissions based on the differences in the abundances of the carbon fractions between the two sources. The compositional profiles for these two major source types show similarities among the three sites. Temperature-resolved carbon fractions also enhanced separations of carbon-rich secondary sulfate aerosols. Potential source contribution function analyses show the potential source areas and pathways of sulfate-rich secondary aerosols, especially the regional influences of the biogenic, as well as anthropogenic secondary aerosol. This study indicates that temperature-resolved carbon fractions can be used to enhance the source apportionment of ambient PM2.5.     

Sources of organic aerosol: Positive matrix factorization of molecular marker data and comparison of results from different source apportionment models

This paper presents results from positive matrix factorization (PMF) of organic molecular marker data to investigate the sources of organic carbon (OC) in Pittsburgh, Pennsylvania. PMF analysis of 21 different combinations of input species found essentially the same seven factors with distinctive source-class-specific groupings of molecular markers. To link factors with source classes we directly compare PMF factor profiles with actual source profiles. Six of the PMF factors appear related to primary emissions from sources such as motor vehicles, biomass combustion, and food cooking. Each primary factor contributed between 5% and 10% of the annual-average OC with the exception of the cooking related factor which contributed 20% of the OC. However, the contribution of the cooking factor was sensitive to the specific combinations of input species. PMF could not differentiate between gasoline and diesel emissions, but the aggregate contribution of primary emissions from these two source classes is estimated to be less than 10% of the annual-average OC. One factor appears related to secondary organic aerosol based on its substantial contribution to biogenic oxidation products. This secondary factor contributed more than 50% of the summertime average OC. Reasonable agreement was observed between the PMF results and those of a previously published chemical mass balance (CMB) analysis of the same molecular marker dataset. Individual PMF factors are correlated with specific CMB sources, but systematic biases exist between the two estimates. These biases were generally within the uncertainty of the two estimates, but there is also evidence that PMF is not cleanly differentiating between source classes.     

Combined source apportionment,using positive matrix factorization–chemical mass balance and principal component analysis/multiple linear regression–chemical mass balance models

The methods of positive matrix factorization–chemical mass balance and principal component analysis/multiple linear regression–chemical mass balance were studied in this paper, for combined source apportionment. Due to the high similarity among the source profiles, several problems would raised when only one receptor model was applied. For example, the collinearity problem would result in the negative contributions when applying CMB model; certain sources would not to be separated out when applying PCA or PMF model. In this study, PCA/MLR–CMB model and PMF–CMB were attempted to resolve the problem, where the combined models were applied to study the synthetic and ambient datasets. In synthetic dataset, there were seven sources (six actual sources from real world, and one unknown source). The results obtained by the combined models show that the combined source apportionment technique is feasible. In addition, an ambient dataset from a northern city in China was analyzed by PCA/MLR–CMB model and PMF–CMB model, and these two models got the similar results. The results show that coal combustion contributed the largest fraction to the total mass.     

Discussion of “Sensitivity of a molecular marker based positive matrix factorization model to the number of receptor observations” by YuanXun Zhang,Rebecca J. Sheesley,Min-Suk Bae and James J. Schauer

There is no specific number of samples that ensure a satisfactory PMF analysis. The statement made in this paper with respect to a specific number of samples is only applicable to this data set and should not be applied to any other data set.