You are here:
HIGHLY TIME-RESOLVED SOURCE APPORTIONMENT TECHNIQUES FOR ORGANIC AEROSOLS USING THE AERODYNE AEROSOL MASS SPECTROMETER
Impact/Purpose:
The overall objective of this proposed project is to develop, validate, and apply fine particulate matter (PM) source apportionment techniques for measurements made with the Aerodyne Aerosol Mass Spectrometer (AMS). The AMS is the only current real-time instrument that provides quantitative size-resolved organic aerosol data with a time resolution of a few minutes. AMS organic data have less fragmentation and thus are more specific for source apportionment than data from laser-ablation mass spectrometers. We expect a number of instruments being developed to improve organic detection specificity via chemical ionization and/or photoionization. The proposed efforts will focus on AMS data but will provide the foundation for using other such data for source apportionment.
Description:
This project had two major components: (1) the development and application of receptor model techniques to AMS OA data, and (2) the field deployment and field data analysis for several new techniques aimed at obtaining more chemically-specific OA information from the AMS for source apportionment purposes. Our work successfully met and exceeded the original goals of the project, as summarized below for each of the two major components.
(1) Progress in receptor model development and application
Several statistical source apportionment models for apportioning OA were investigated, including custom techniques that take advantage of our understanding of the data (Custom Principal Component Analysis, CPCA, and Multiple Component Analysis, MCA, both developed as part of this project), and a standard multivariate receptor model (Positive Matrix Factorization, PMF). Our work was the first to demonstrate that different organic components could be extracted from AMS field datasets, including their concentrations, mass spectra, diurnal cycles, and size distributions (and more recently elemental composition). The application of these techniques to AMS data has received very intense interest from the scientific community and in fact it has become an active subfield on its own. Results on these analyses have been presented at major national and international conferences, including an invited oral presentation at the Gordon Research Conference on Atmospheric Chemistry and a plenary lecture at the Annual Conference of the American Association on Aerosol Research (AAAR), both delivered by the PI. Sixteen publications on this topics and acknowledging this grant have been published or submitted (listed below), and many presentations have also been given at major meetings such as the AAAR, AGU, EAC, and EGU annual meetings. The papers presenting these results are having a significant impact, and they have already received more than 400 citations in ISI Web of Science (as of May 2009). Seven of the published papers have been classified as “Highly Cited Papers” (Top 1% citation in their journals) by ISI Web of Science, and two have been classified as “Top Cited Papers” by the journal Environmental Science and Technology. One is the most cited paper in Geophysical Research Letters since 2007 (of 3382 papers) and another is the most cited paper in Environmental Science and Technology since 2008 (of 2099 papers).
We first demonstrated all the techniques with the AMS data acquired at the EPA Pittsburgh Supersite in September 2002, resulting in several publications where we reported for the first time the direct identification of hydrocarbon-like organic aerosols (HOA) and oxygenated organic aerosols (OOA), which showed strong correspondence with primary and secondary organic aerosols (POA and SOA), respectively [Zhang et al., ES&T 2005; ACP 2005]. OOA was larger than HOA, contrary to previous results at this site. We also explored whether OOA increased during periods with high aerosol acidity, which could be a sign of acid-catalyzed SOA formation, and found that this increase was at most 25% and likely lower [Zhang et al., ES&T 2007]. We later published [Zhang et al., GRL, 2007] an MCA analysis of 37 highly timeresolved AMS datasets acquired in the Northern Hemisphere mid-latitudes in 11 urban areas, 5 regions downwind of urban areas, and 11 rural/remote locations (representative of high elevation, forested, pristine, and continentally-influenced marine atmospheres). At most sites the MCA retrieved several types of OOA. The results of this analysis strongly indicate that global models largely underestimated the importance of Secondary Organic Aerosols (SOA) and likely overestimated the importance of Primary Organic Aerosols (POA).
In addition to the work on custom source apportionment techniques just described, we also pursued the application of Positive Matrix Factorization (PMF), a source apportionment method commonly used in atmospheric science, to AMS datasets. We performed an in-depth evaluation of the proper specification of errors for AMS data, as well as the interpretation and pitfalls of PMF factorization of AMS data [Ulbrich et al., ACP 2009]. In particular we identified the tendency of PMF-AMS solutions to produce “split” factors which have realistic-looking time series and mass spectra, yet have no physical reality. We have developed the PMF Evaluation Panel (PET), an Igor-based software interface to automate the running and especially the analysis of PMF solutions. The code automatically generates the most useful plots to evaluate the results of the model and diagnostic statistics that help investigators investigate the range of solutions of the model. Systematic investigation of uncertainties is automated with bootstrapping, multiple seeds, and FPEAK variation [Ulbrich et al., ACP 2009]. The PET has been shared with the AMS and the wider research community starting at the AMS Users Meeting in Manchester, UK, in Sep. 2008, and about two dozen groups are using this software tool at present, greatly facilitating the application and consistent interpretation of this powerful but very complex technique.
The identification of spectra from statistical techniques necessitates the use of standard spectra for known sources. For this purpose, we created three public web-based mass spectral databases: the unit-resolution AMS database, the high-resolution AMS database, and the thermal-desorption / thermal denuder database (see e.g. http://cires.colorado.edu/jimenez-group/AMSsd/). These databases are growing as new spectra are added from our group and others in the AMS community, and now contain over 100 spectra or spectra/volatility profiles [Ulbrich et al., ACP 2009].
We have applied PMF and MCA to multiple datasets of our own and from collaborators. PMF was first applied to the Pittsburgh 2002 dataset, largely confirming the previously published CPCA results, but also identifying a small (8%) semivolatile OOA-2 component which correlates with nitrate and chloride while the main OOA-1 component correlates with particle sulfate instead and appears non-volatile [Ulbrich et al., ACP 2009]. Other publications on this area and acknowledging this grant include: (a) Takegawa et al. [JGR, 2006] obtains good agreement between the HOA & OOA estimated with the AMS for Tokyo and the CO-tracer method for SOA estimation; (b) Kondo et al. [2007] concludes that about 90% of the OOA in Tokyo is water soluble while HOA is water insoluble by comparison of AMS CPCA results with PILS-WSOC analysis; (c) Cottrell et al. [JGR, 2008] reports that OOA dominates the aerosol composition in rural New Hampshire, apparently due to SOA of both anthropogenic and biogenic origin; (d) Nemitz et al. [AS&T, 2008] use PMF with a mixed OA concentration / eddy covariance flux AMS dataset from Boulder, CO, to quantify the fluxes OOA-1, OOA-2, and HOA, and reports that while the more regional OOA-I is undergoing net deposition, the more local HOA and OOA-2 have a net emission flux at this location; (e) Johnson et al. [ES&T, 2008] compare the AMS OA with the OA estimated from Proton Elastic Scattering Analysis (PESA) measurements of hydrogen for the 2003 Mexico City Metropolitan Area (MCMA-2003), reporting that 75% of the OA evaporated before PESA analysis, and that the OOA correlates better with the remaining PESA OA, which is consistent with the lower volatility of this component deduced from thermal denuder measurements (below); (f) Fast et al. [ACP, 2009] evaluate the results from a 3-D model (WRF-CHEM) using our results from PMF of highresolution AMS data.
An important part of this project was to compare the results of AMS-PMF with other more established methods for OA source apportionment. We have reported good comparisons with results from chemical mass balance (CMB) of organic molecular markers for the MILAGRO and SOAR-1 datasets [Docherty et al., ES&T 2008; Aiken et al., ACPD 2009]. Good comparisons were also observed with the EC-tracer method [Zhang et al., ACP 2005; Docherty et al, ES&T 2008], the WSOC method [Kondo et al., JGR, 2007; Docherty et al., ES&T, 2008], and the COtracer method [Takegawa et al., JGR, 2006; Docherty et al., ES&T, 2008]. Further comparisons for the SOAR-1 dataset are in progress and will be presented in future publications.
(2) Deployment in Field Studies and Data Analysis of New Techniques
A major accomplishment of this project was the organization and performance of the SOAR-1 and SOAR-2 (Study of Organic Aerosol in Riverside, phases 1 & 2) field experiments. These studies were organized by Profs. Jimenez and Ziemann (UC-Riverside), with participation from Profs. Schauer, Hannigan, and Zhang, with a focus on characterizing organic aerosols with a variety of techniques. They were carried out at UC-Riverside, which is located in a polluted region downwind of Los Angeles that is heavily impacted by both primary and secondary aerosol. Although it was originally intended that this study would include only the UCR, CU, and Wisconsin groups, as word of our intentions spread, many other research groups became interested in making measurements with a variety of complementary methods during this period. As a result, during the SOAR-1 study (~July 15-Aug. 15, 2005) approximately 60 scientists from 17 universities and research institutes and companies participated in what is probably the most complete field study of organic aerosols using advanced techniques to date. During SOAR-2 (Nov. 1-24, 2005) about 20 scientists from 8 groups participated. More detailed information on the participants and measurements can be found at http://cires.colorado.edu/jimenezgroup/ Field/Riverside05/. Eighteen (18) papers have been published so far based on the data acquired in this campaign by all participating groups, and at least 6 more are submitted or in preparation.
AMS results indicate a very reproducible aerosol concentration and composition from day to day, with a similar diurnal cycle. Differences between weekday and weekends are observed, presumably due to changes in emissions. Five methods produce consistent results, indicating that OA during SOAR is overwhelmingly secondary in nature during a period of several weeks with moderate ozone concentrations, and that SOA is the single largest component of PM1 aerosol in Riverside. Average SOA contributions of >80% were observed during mid-day periods while minimum SOA contributions of ~50% were observed during peak morning traffic periods. A similar result holds for Pasadena during SOAR-1. These results are contrary to previous estimates of SOA throughout the Los Angeles Basin which reported that, other than during severe photochemical smog episodes, SOA was lower than primary OA. We have used the SOAR source apportionment results in a study of cloud condensation nuclei and cloud properties for urban aerosol [Cubison et al., ACP, 2008] showing that not just the composition but also the mixing state of OA is critical for successful CCN closure.
During SOAR-1, the Jimenez group deployed several new instrument combinations for organic aerosol analysis: (a) a new high-resolution time-of-flight aerosol mass spectrometer (HR-ToFAMS) was deployed in the field for the first time, to provide additional insight on the aerosol chemical composition by allowing the direct determination of the elemental composition of every peak in the spectrum of the AMS [DeCarlo et al., Anal. Chem. 2006]; (b) a thermal denuder (TD) system was used in front of an AMS and a scanning mobility particle sizer (SMPS) to characterize the coupled chemistry-volatility profiles [Huffman et al., AS&T 2008; ES&T 2009; ACPD 2009]; and (c) a variable-temperature AMS vaporizer was used some of the time in both the ToF-AMS and HR-ToF-AMS to provide additional insight on the thermal properties of the organic aerosol [Docherty et al., AAAR 2008]. These systems were used to continuously analyze ambient particles, and to analyze secondary organic aerosol (SOA) formed in several reactions carried out in the environmental chamber in the Ziemann laboratory.
The high-resolution analysis capability of the HR-ToF-AMS has enabled us to separate the inorganic from organic aerosol spectra, to enhance the spectral differences that are critical to the success of statistical methods such as PMF [Docherty et al., ES&T, 2008; Aiken et al., ACPD, 2009], and to develop a method for the direct measurement of organic elemental composition (O/C, N/C, H/C, and OM/OC) [Aiken et al., Anal. Chem. 2007; ES&T, 2008]. The latter paper applies the elemental analysis method for ambient, chamber, and laboratory biomass burning spectra, with results consistent with those of other techniques. One important result is that the OOA/SOA found in the atmosphere is significantly more oxygenated than the SOA made in traditional chamber experiments [Aiken et al., ES&T, 2008]. This difference decreases at lower precursor / SOA concentrations [Shilling et al., ACP, 2009], although at the expense of decreasing yields that increase the quantitative discrepancies with SOA concentrations in the field.
The thermal-denuder and variable vaporizer temperature data allow the separation of different OA components with different volatility behavior, as they enhance the contrast between them. Results from Mexico City and Riverside confirm that the least volatile OA component is OOA-1 (a surrogate for aged SOA) while HOA and OOA-2 (surrogate for fresh SOA) are more volatile. These results strongly support that all types of OA should be treated as semivolatile in models.
Two laboratory biomass burning experiments (FLAME-1 and FLAME-2 campaigns) that were carried out in 2006 and 2007 at the Missoula, MT Biomass Burning Chamber operated by the US Forest Service. The objective of this deployment was to obtain the first signatures of biomass burning aerosols using the HR-ToF-AMS and the thermal denuder. This experiment allowed us to obtain HR-ToF-AMS and thermal denuder signatures for smoke from 16 different biomasses at realistic dilution levels (1/10,000). During FLAME-2 we demonstrated a new high timeresolution (up to 100 Hz) mode of the ToF-AMS software that allows capturing rapid fluctuations on the emissions and easily separate flaming vs. smoldering emissions during stack burns. Results indicate a wide variability in the organic fraction, composition, and volatility across biomasses and burning conditions (flaming vs. smoldering). Biomass burning organic aerosols (BBOA) are oxygenated, but typically less than SOA. Most BBOA is also more more volatile than real-word SOA, and of similar volatility than urban POA [Aiken et al., ES&T 2008; Huffman et al., ES&T 2009]. The O/C of BBOA is on the range 0.3-0.45 [Huffman et al., ES&T, 2008], which are very similar to the values derived from PMF of ambient measurements [Aiken et al., ES&T, 2008].
An additional experiment was carried out with support from this project April/May 2007 to characterize the high-resolution and thermal denuder spectra from meat cooking, trash burning, and motor vehicle emissions. This experiment was motivated by questions arising from the analysis of the SOAR and MILAGRO campaigns. Mohr et al. [ES&T, 2009] and Huffman et al. [ES&T, 2009] present results of this experiment, which show with high resolution data that OA emitted by combustion engines and plastic burning are dominated by hydrocarbon-like (reduced) organic compounds. Meat cooking and especially paper burning contain significant fractions of oxygenated organic compounds; however, their unit-resolution mass spectral signatures are very similar to mass spectral signatures from hydrocarbon-like OA or primary OA, and very different from the mass spectra of ambient secondary or oxygenated OA (OOA). Thus, primary OA from any of these sources is very unlikely to be a significant direct source of ambient OOA. POA from all of these sources was semivolatile.
Another research topic was the development and application of a method allowing the first realtime determination of PAHs in submicron aerosols from aerosol mass spectrometry data [Marr et al., ACP, 2006; Dzepina et al., IJMS, 2007]. This technique was developed using data from the MCMA-2003 field study, and the AMS measurements showed good correlation with those from two separate techniques, but also identified some very reactive PAHs that are likely being destroyed by reaction artifacts on traditional filter-sampling techniques.
Finally a $15k supplement to this grant funded Prof. Weber of Georgia Tech for deploying two organic aerosol analysis techniques during SOAR-1, which have resulted in two publications [Peltier et al., AS&T 2007; Docherty et al., ES&T 2008].
URLs/Downloads:
Final Progress Report2006 Progress Report
2005 Progress Report
2007 Progress Report
Record Details:
Record Type:PROJECT( ABSTRACT )
Start Date:12/01/2004
Completion Date:11/30/2007
Record ID:
133745