Grantee Research Project Results
Final Report: Advancing ATOFMS to a Quantitative Tool for Source Apportionment
EPA Grant Number: R831083Title: Advancing ATOFMS to a Quantitative Tool for Source Apportionment
Investigators: Prather, Kimberly A. , Hopke, Philip K.
Institution: University of California - San Diego , Clarkson University
EPA Project Officer: Chung, Serena
Project Period: October 1, 2003 through September 30, 2006 (Extended to September 30, 2007)
Project Amount: $450,000
RFA: Measurement, Modeling, and Analysis Methods for Airborne Carbonaceous Fine Particulate Matter (PM2.5) (2003) RFA Text | Recipients Lists
Research Category: Air , Air Quality and Air Toxics , Particulate Matter
Objective:
The objective of this project involved using the aerosol time-of-flight mass spectrometry data to ascertain if single particle mass spectrometry can: 1) quantitatively measure the carbonaceous component of the ambient aerosol including organic and elemental carbon and to ascertain if it is possible to develop a quantitative and universal calibration that provides results comparable to time resolved OC/EC measurements; 2) provide key markers that distinguish among sources of carbonaceous aerosol including diesel and spark-ignition vehicles, mobile and stationary sources, fossil fuel sources versus biomass burning, and primary biological, and/or secondary organics; 3) provide insights into atmospheric processes that can then be better represented in air quality models such as the relationship of secondary OC with primary particle types.
Summary/Accomplishments (Outputs/Outcomes):
We have established unique mass spectral ion marker combinations (i.e. fingerprints) in ATOFMS data for primary and secondary carbonaceous species including elemental carbon (EC) and organic carbon (OC). Unique markers in the mass spectra of these particle types allow one to distinguish between EC, OC, and OC on EC cores at the single particle level. These markers are consistently observed in ambient field studies, as well as source combustion studies of spark ignition and diesel vehicle emissions, with extremely reproducible ion intensities. We have used the mass spectral data to establish a correlation between the measured relative ion intensities and the mass fractions of OC and EC in individual particles as a function of size, which represents a significant advance in our ability to be quantitative for these species with ATOFMS (Spencer, 2006). In addition, because individual particle compositions are measured, we are able to directly assess the mixing state (or chemical associations) between EC and OC and other (primarily) secondary species such as ammonium, nitrate, and sulphate. The EC particle trends measured using ATOFMS in the ultrafine size range show strong correlations with the temporal evolution of gas phase CO, a species indicative of fresh vehicle emission in a study conducted near a major freeway. In the area of source apportionment, EC signatures from cars and trucks are almost identical; however the single particle associations with other chemical species such as calcium, phosphate (both indicators of engine oil observed in diesel exhaust), and certain metals vary from source to source. The overall fingerprints used for apportioning single particles is based on the presence of a combination of ion markers for EC, OC, calcium, phosphate (and/or lack of some of these species) and do show distinctions between these carbonaceous species from different sources . Using the results from the flow tube lab studies conducted as part of this project, the quantitative OC/EC measurements were used to quantify the amount of transformations occurring on particles after passing through an ultrafine particle concentrator system used for health effects studies as part of the University of Rochester PM Center project. Finally, Spencer developed a method for determining the densities of ambient EC particles. The method couples an SMPS to the ATOFMS. A quantitative approach for deriving effective densities of EC was established (Spencer, 2007).
The ATOFMS is a number based detection method that can determine the relative fractions of aerosol particles from specific sources. However, to directly compare to PM2.5 regulatory standards, ideally ATOFMS data must be converted into mass concentrations. Qin, et al. (2006b) showed it is possible to convert number concentrations to mass fractions using an aerodynamic particle sizer. Scaling procedures have been developed to transform ATOFMS particle counts acquired in measurements made at the Fresno Supersite to mass concentrations. These scaled mass concentrations compare well (R2 ~ 0.8) with standard PM2.5 mass concentration measurements of a beta attenuation monitor (Qin, 2006b). Scaled size- and temporal-resolved concentrations of particles as a function of single particle composition are derived using these scaling factors for an urban location in Fresno and compared with ATOFMS data acquired at a rural location in central California (Angiola) located outside of Fresno. This study was conducted in the winter and thus the Fresno particles were dominated by biomass burning aerosols. In contrast, the Angiola PM concentrations are lower and show ion markers indicating they have undergone fog processing and gas-to-particle conversion upon transit to the Angiola site. The results obtained to date show great promise for using ATOFMS as a tool for performing time- and size-resolved source apportionment of ambient aerosols which will be important for establishing control strategies in order to meet compliance. Also, characterizing the composition changes on shorter timescales allow comparisons to be made between ATOFMS and other gas and particle phase measurements which will allow us to better understand aerosol processing in the atmosphere.
Source Apportionment using Factor Analysis of ATOFMS Data
There have been several publications using positive matrix factorization to apportion particle number concentrations to sources (Owega et al., 2004; Pekney et al., 2006). Positive Matrix Factorization decomposes the observed time series (the data) as a positive sum of fewer time series (the factors). The reconstituted series approximates the data but cannot always capture all of the original details. This result raises the question of whether the factors that PMF finds measure real phenomena, such as time series of the source emissions or of air mass displacements, or are ad hoc mathematical solutions without reflecting the underlying physical or chemical phenomena. Real ATOFMS aerosol time series from the ACE-Asia campaign were analyzed with PMF and factored successively with different numbers of assumed sources. Two methods are introduced for testing the resulting factors: (1) comparing the residuals of the factorization to residuals obtained from control data made by shuffling the original data, and (2) examining the relative difference between actual and low band pass filtered factors. In addition the factors are compared to the original time series of the overall particle types. All of these methods show that several factors can be retained: Factorization residuals are substantially lower than residuals from the same series randomized, factors differ much less from their low band pass filtered version than random variables would, and the smoothest factors coincide with the time series of the most easily explained chemical classes. Besides their immediate application to PMF of ATOFMS aerosol counts, these results suggest that these validation methods can be used in other blind source separation problems, and in particular, that frequency content of the data can be used to select good from bad factors in blind source separation. A manuscript describes this in detail in Analytica Chimica Acta (Zhao, 2005).
Calibration and quantification of ATOFMS
As one approach to use single particle aerosol time-of-flight mass spectrometry (ATOFMS) data to provide a quantitative estimation of chemical compositions of ambient aerosols, a multivariate calibration approach can be applied. In an initial study (Fergenson et al., 2001), the possibility of developing a calibration model to predict chemical compositions from ATOFMS data was demonstrated, but because of the limited number of samples (only 12), the ability of the calibration model was not fully realized. Zhao et al (2005) used 50 samples were created to further test the prediction ability of the calibration model. The conceptual framework in these studies is to relate the mass concentrations of the particles in the identified particle type classes to the average aerosol compositions for each sampling time interval. Particles were classified using the ART-2a neural network (Song et al., 1999; Phares et al., 2001; Bhave et al., 2002). Multivariate calibration models can then be used to provide a predictive relationship between the class mass concentrations and measured species concentrations. Because there may be nonlinearity between cluster mass concentrations and ambient species concentrations because of measurement errors, the scaling equations used to estimate particle mass, and various assumptions required for building the model, PLS regression was integrated with radial basis functions (RBF-PLS) to obtain better prediction results. These results were compared to partial least square (PLS) regression alone. Compared with the earlier study, these results provide better and a more convincing demonstration of the ability of the calibration model to estimate the chemical compositions from ATOFMS data. The results also suggest that the model would be able to provide carbon data and thus substitute for thermal optical reflectance (TOR) measurements. Additionally, the calibration model based on RBF-PLS showed more accurate predictions in the cases with some degree of nonlinearity in the underlying relationships.
Development of Improved Classification of Particles based on ATOFMS data
As described in the section on estimation of the composition of the atmospheric aerosol, the first step in the process is the sorting of particles into homogenous classes based on each particle’s mass spectral characteristics. The earliest work on this problem had involved fixed classifiers such as cluster analysis (Kim et al., 1987) and rule-building expert systems (Buydens et al., 1988; Kim and Hopke, 1988). These approaches are applied to a fixed set of data and cannot adjust to new classes as additional objects are analyzed. Thus, a dynamic classification system is needed. Following our initial work on Adaptive Resonance Theory neural networks (Song et al., 1999), ART-2a has been used by various groups for the analysis of ATOFMS data (Phares et al., 2001; Bhave et al., 2001).
Recently, a novel cluster analysis method based on a different grouping principle, called density based spatial clustering of application with noise (DBSCAN), was introduced to the cluster analysis of ATOFMS data and compared with ART-2a (Ester, et. al., 1996; Zhou, et al. 2006). However, without any information on particle origins, the explanation of the clustering results was mainly based on the physical interpretability of each cluster center. Therefore, it was difficult to make a full comparison of these two methods, particularly because of the similarity of the mass spectrometry data of the particles from similar sources and the ambiguity of the categorization of these particles.
Zhou et al. (2007) used a set of benchmark ATOFMS data (i.e., with a priori known source indexes) of the ambient aerosols from six sources (gasoline emission, diesel emission, biomass burning, coal combustion, sea salt and soil dust) to compare clustering by these two methods. The ART-2a based approaches show better initial clustering results than the usage of DBSCAN alone. The territory expansion principle of DBSCAN largely prevented the overlap of generated cluster spaces, but it does not show its specific advantages in this study because of the complex and entangled distribution spaces of the six sources. A proper vigilance factor can produce a reasonable ART-2a clustering result, but an overly fine clustering result for ART-2a with a high vigilance factor can be recovered by a post-processing strategy. DBSCAN seems to be more effective and robust in post-processing than other regrouping analyses. Thus, a recommended approach is to begin the analysis with ART-2a with a relatively high vigilance factor (>0.7) and regroup the clusters using DBSCAN.
Source apportionment of OC and EC particles using ATOFMS: A key goal of this project involved using ATOFMS to apportion carbonaceous aerosols. As described above, a number of approaches have been taken to achieve this goal. In addition the methods above, we also tested using single particle mass spectral fingerprint libraries created using ART-2a (Song, 2001) to "match" ambient particles and apportion particles (Toner, 2007; Toner, 2006; Shields, 2006). In addition, we have studied the impacts of biomass burning on urban emissions in Fresno (Qin, 2006). The ATOFMS has now been shown to be able to distinguish between major combustion sources including biomass burning, aged organic carbon, cars, trucks, and ships.
A strong correlation (R2 = 0.88) was shown between ATOFMS scaled mass concentrations and co-located PM2.5 mass concentrations acquired with a beta attenuation monitor (BAM).
As part of the source apportionment project, Rebotier et al. (2007) also compared various clustering algorithms to one another to determine which would be most appropriate for "on-the-fly" apportionment (Rebotier, 2007). In this study, ART-2a showed the most promise and as a result, a new source apportionment tool was developed. The ATOFMS is now capable of real-time source apportionment of ambient particles.
References:
Bhave, P. V., J. O. Allen, B. D. Morrical, D. P. Fergenson, G. R. Cass and K. A. Prather, A field-based approach for determining ATOFMS instrument sensitivities to ammonium and nitrate. Environmental Science & Technology, 36, 4868-4879, 2002.
Buydens, L., Massart, D.L. Hopke, P.K., Evaluation of the Expert System Shells EX-TRAN and TIMM as Rule-Building Tools for Classification Purposes, Chemom. Intel. Lab. Syst. 3:199-204 (1988).
Fergenson, D.P., X.H. Song, Z. Ramadan, J.O. Allen, L.S. Hughes, G.R. Cass, P.K. Hopke, and K.A. Prather, Quantification of ATOFMS data by multivariate methods, Analytical Chemistry, 73 (15), 3535-3541, 2001.
Kim, D.S., Hopke, P.K., The Classification of Individual Particles Based on Computer-Controlled Scanning Electron Microscopy Data, Aerosol Sci. Technol. 9:133-151 (1988).
Kim, D.S., Hopke, P.K. Massart, D.L., Kaufman, L. Casuccio, G.S., Multivariate Analysis of CCSEM Auto Emission Data, Sci. Total Environ. 59:141-155 (1987).
Owega, S., Khan, B.U.Z., D'Souza, R., Evans, G.J., Fila, M., Jervis, R.E., 2004. Receptor modeling of Toronto PM2.5 characterized by aerosol laser ablation mass spectrometry ,Environmental Science & Technology 38 (21): 5712-5720.
Pekney, N.J., Davidson, C.I., Bein, K.J., Wexler, A.S.., Johnston, M.V., 2006. Identification of sources of atmospheric PM at the Pittsburgh Supersite, Part I: Single particle analysis and filter-based positive matrix factorization, Atmospheric Environment 40: S411-S423 Suppl. 2.
Phares, D.J., K.P. Rhoads, A.S. Wexler, D.B. Kane, and M.V. Johnston. Application of the ART-2a algorithm to laser ablation aerosol mass spectrometry of particle standards. Anal. Chem. 73, 2338-2344, 2001.
Qin, X., K. Prather, P. Bhave, ATOFMS Measurements at Urban and Rural Locations: Comparison of Single Particle Size and Composition, AAAR PM meeting, February 2005, Atlanta GA.
Qin, X. Y., et al. (2006a), Comparison of two methods for obtaining quantitative mass concentrations from aerosol time-of-flight mass spectrometry measurements, Analytical Chemistry, 78(17), 6169-6178.
Qin, X. Y., and K. A. Prather (2006b), Impact of biomass emissions on particle chemistry during the California Regional Particulate Air Quality Study, International Journal of Mass Spectrometry, 258(1-3), 142-150.
Rebotier, T.P., Prather KA (2007), Aerosol time-of-flight mass spectrometry data analysis: A benchmark of clustering algorithms, Analytical Chim. Acta, 585 (1): 38-54.
Shields, L. G., et al. (2007), Determination of single particle mass spectral signatures from heavy-duty diesel vehicle emissions for PM2.5 source apportionment, Atmospheric Environment, 41(18), 3841-3852.
Song, X.H., P.K. Hopke, D.P. Fergenson, and K.A. Prather, Classification of single particles analyzed by ATOFMS using an artificial neural network, ART-2A, Analytical Chemistry, 71 (4), 860-865, 1999.
Spencer, M. T., and K. A. Prather (2006), Using ATOFMS to determine OC/EC mass fractions in particles, Aerosol Science and Technology, 40(8), 585-594.
Spencer, M. T., et al. (2006), Comparison of oil and fuel particle chemical signatures with particle emissions from heavy and light duty vehicles, Atmospheric Environment, 40(27), 5224-5235.
Spencer, M. T., et al. (2007), Simultaneous measurement of the effective density and chemical composition of ambient aerosol particles, Environmental Science & Technology, 41(4), 1303-1309.
Toner, S. M., et al. (2006), Single Particle Characterization of Ultrafine and Accumulation Mode Particles from Heavy Duty Diesel Vehicles Using Aerosol Time-of-Flight Mass Spectrometry, Environmental Science & Technology, 40(12), 3912-3921.
Toner, S. M., et al. (2007), Using mass spectral source signatures to apportion exhaust particles from gasoline and diesel powered vehicles in a freeway study using UF- ATOFMS, Atmospheric Environment, In Press.
Zhou, W., Hopke, P.K., Prather, K.A., Comparison of Two Cluster Analysis Methods using Single Particle Mass Spectra, Atmospheric Environment (in press).
Zhao WX, Hopke PK, Qin XY, et al. (2005) Predicting bulk ambient aerosol compositions from ATOFMS data with ART-2a and multivariate analysis, Analytical Chim. Acta, 549 (1-2): 179-187.
Journal Articles on this Report : 4 Displayed | Download in RIS Format
Other project views: | All 8 publications | 4 publications in selected types | All 4 journal articles |
---|
Type | Citation | ||
---|---|---|---|
|
Spencer MT, Prather KA. Using ATOFMS to determine OC/EC mass fractions in particles. Aerosol Science and Technology 2006;40(8):585-594. |
R831083 (Final) R827354 (Final) R827354C001 (Final) R832415 (2010) R832415 (2011) R832415 (Final) |
Exit Exit Exit |
|
Spencer MT, Shields LG, Prather KA. Simultaneous measurement of the effective density and chemical composition of ambient aerosol particles. Environmental Science & Technology 2007;41(4):1303-1309. |
R831083 (Final) R827354 (Final) R832415 (2010) R832415 (2011) R832415 (Final) R832415C001 (2006) R832415C001 (2010) R832415C001 (2011) |
Exit Exit Exit |
|
Zhao W, Hopke PK, Qin X, Prather KA. Predicting bulk ambient aerosol compositions from ATOFMS data with ART-2a and multivariate analysis. Analytica Chimica Acta 2005;549(1-2):179-187. |
R831083 (Final) R827354 (Final) R827354C001 (Final) R832415 (2010) R832415 (2011) R832415 (Final) |
Exit Exit Exit |
|
Zhao W, Hopke PK, Prather KA. Comparison of two cluster analysis methods using single particle mass spectra. Atmospheric Environment 2008;42(5):881-892. |
R831083 (Final) |
Exit Exit Exit |
Supplemental Keywords:
PM2.5, OC/EC, secondary organic aerosol, source apportionment, ATOFMS, continuous measurements, sources, Supersite , Air, Ecosystem Protection/Environmental Exposure & Risk, RFA, Scientific Discipline, Analytical Chemistry, Atmospheric Sciences, Environmental Chemistry, Environmental Engineering, Environmental Monitoring, Monitoring/Modeling, Physics, air toxics, particulate matter, aerosol analyzers, aerosol particles, aerosol time of flight mass spectrometry, air quality model, air sampling, atmospheric dispersion models, atmospheric measurements, atmospheric particulate matter, carbon particles, emissions, human exposure, human health effects, mass spectrometry, measurement methods, modeling, modeling studies, monitoring stations, particle phase molecular markers, particulate matter mass, secondary organic aerosols,, RFA, Scientific Discipline, Air, Ecosystem Protection/Environmental Exposure & Risk, particulate matter, air toxics, Environmental Chemistry, Monitoring/Modeling, Environmental Monitoring, Environmental Engineering, atmospheric particulate matter, atmospheric dispersion models, atmospheric measurements, source apportionment, aerosol particles, human health effects, secondary organic aerosols, air quality models, monitoring stations, air sampling, carbon particles, air quality model, emissions, modeling, particulate matter mass, human exposure, secondary organic aerosol, particle phase molecular markers, transport modeling, modeling studies, aerosol analyzers, measurement methodsRelevant Websites:
Progress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.