Grantee Research Project Results
Final Report: Evaluation of Regional Scale Receptor Modeling
EPA Grant Number: R832156Title: Evaluation of Regional Scale Receptor Modeling
Investigators: Lowenthal, Douglas H. , Chen, Lung-Wen Antony , Watson, John L. , Koracin, Darko
Institution: Desert Research Institute
EPA Project Officer: Chung, Serena
Project Period: January 1, 2005 through December 31, 2007 (Extended to December 31, 2009)
Project Amount: $436,687
RFA: Source Apportionment of Particulate Matter (2004) RFA Text | Recipients Lists
Research Category: Air Quality and Air Toxics , Particulate Matter , Air
Objective:
The objective of this research was to evaluate multivariate and trajectory-based receptor models for regional source apportionment relevant to the USEPA Regional Haze Rule. This was accomplished by applying receptor models to synthetic data generated with an air quality model for two eastern IMPROVE sites: Brigantine National Wildlife Refuge, NJ (BRIG) and Great Smoky Mountains National Park, TN (GRSM), and by comparing known regional scale contributions to PM and sulfate aerosols to those estimated with the receptor models. In addition, guidance is provided for the application of such models in future applications.
Summary/Accomplishments (Outputs/Outcomes):
This project was a cooperative effort with Drs. Naresh Kumar and Eladio Knipping of EPRI, who provided synthetic IMPROVE data sets generated with the SMOKE/CMAQ/MM5 modeling system. This project has produced a unique data set comprised of hourly multi-species PM concentrations across the entire eastern U.S. during the summer and winter of 2002. The contributions of 7 geographically discrete regions to primary and secondary PM species have been estimated at BRIG and GRSM and throughout the modeling domain. Individual source profiles for 43 source types used to generate the speciated PM concentrations were augmented with tags for primary PM2.5. This will enable post-processing sensitivity tests on source profile composition without having to redo the CMAQ modeling. The value of these synthetic data sets extends beyond the goal of this project, i.e., testing receptor models. The CMAQ modeling is a state-of-the-art emissions-based source apportionment in its own right. It is especially unique because of the large-scale nature of the sources. We expect that these data sets will be used in the future to address issues related to large-scale atmospheric chemistry and transport of PM in the eastern United States.
Model Development, Emissions Inventory, Source Profiles, and Generation of Synthetic IMPROVE Data Sets
EPRI obtained MM5 output files from VISTA modeling studies for 2002. These data were transferred to DRI for use in TMBR and Lagrangian particle modeling and Sonoma Technology, Inc. (STI) for CMAQ/SMOKE/MM5 modeling. EPRI and DRI analyzed NEI inventory data for PM2.5, SO2, NH3, NOx, and VOC, which were used to define emissions source categories and corresponding source profiles for a domain encompassing the eastern half of the country. STI generated synthetic IMPROVE data at BRIG and GRSM for summer (July-September) and winter (January-March) 2002.
Figure 1. Study domain for application of receptor models to synthetic data. The BRIG and GRSM sites are indicated by the heavy black dots in southern NJ and eastern TN.
Primary PM2.5 source profiles for 43 source categories were taken from the EPA’s Speciate and DRI’s PM source profile libraries. The profiles were used in the CMAQ model to produce hourly multi-species IMPROVE-style concentration data. The meteorological input to CMAQ was high-resolution (12 km) data generated with the NCAR Mesoscale Meteorological Model (MM5). Forty-three additional variables (T1-43) were added to each source profile, with unique values equal to the primary PM2.5 emitted by that source. This allowed us to follow each source’s primary PM2.5 contribution to each receptor site. Contributions to PM and sulfate from each of the seven regions were estimated by sequentially running the model with 30% of a given region’s anthropogenic emissions removed. The focus of the analysis to date has been on the summer data set.
Model-predicted and actual IMPROVE sulfate concentrations for the summer of 2002 are compared in Lowenthal, et al. (2010), which shows good correlations at both sites. At BRIG, the major “true” sulfate source was its own northeastern region (MANE-VU). VISTAS (east) and the upper Midwest also were significant contributors. At GRSM, the principal contributor was its own region (VISTAS west). VISTAS east and VISTAS south were the next most significant contributors, followed by the upper Midwest. Note that CENRAP (north and south) was the smallest contributor (<3%) at both sites.
Positive Matrix Factorization (PMF) Results
PMF at both sites was initially done with 6-hour average data. The following species measured in the IMPROVE network were included in the model: Al, Si, Ca, Fe, K, Ti, As, Br, Cl, Cr, Cu, Mn, Mo, Ni, P, Pb, Rb, Se, Sr, V, Zn, Zr, EC, and SO4=. The synthetic concentrations contained no sampling or measurement error. This was in effect an ideal data set with which to test any receptor model. Because PMF requires uncertainties for weighting, these were assigned as 1% of the mean concentration for each species. Thus, the relative importance of all species in the PMF analysis was the same. The number of factors for PMF was based on the “signal to noise” (SN) statistic in the Unmix model. Large values of SN are associated with large singular values, i.e., greater than or equal to two. A choice of seven factors was consistent with SN ~2 and accounted for nearly all of the variance in the data (R2=0.99) at both sites. It represents a clear step-function according to the Unmix number-of-factors-selection criterion.
The 7-factor PMF solution at BRIG provided close fits between measured and calculated species concentrations with r2 > 0.96 for all species except Mn (r2 = 0.89). At GRSM, r2 was > 0.95 for all species except EC (r2 = 0.86). It was immediately apparent that PMF did not resolve regional contributions to sulfate at either site. The interpretation of the PMF factors were based on the correlations of the actual PM2.5 source contributions with the estimated factor contributions. PMF produced a single factor dominated by sulfate at both sites. The correlations (r2) between ambient sulfate concentrations and the sulfate factor contributions were 0.98 and 0.94 at BRIG and GRSM, respectively. This factor is commonly referred to as “secondary sulfate” in the PMF literature. The remaining factors were identified with sources with the highest correlations between their “true” primary PM2.5 contributions and the PMF factor contributions. Sources with the highest correlations with the factors do not necessarily contribute the most PM2.5 to those factors.
The sources of primary PM associated with the PMF factors were similar to those described in previous studies at BRIG at GRSM. Such sources included municipal incineration, industrial manufacturing, ferromanganese production, secondary Al production, steel production, construction dust, paper waste burning, and road dust. Regional sources were not resolved because the regional source profiles were much more similar than were the emissions source profiles used to construct the data. PMF factor contributions to chemical species at BRIG and GRSM were qualitatively similar to the source profiles used to generate the data, e.g., crustal (Al, Si, Ca, and Fe), ferromanganese (Fe-Mn), mobile emissions EC), coal combustion (Se), oil-fired power plant (V, Ni) industrial manufacturing (As, Pb), and residential wood combustion (EC, K)
While the PMF factors corresponded to varying extents with the primary PM2.5 sources used to create the data, they do not associate sulfate with these sources or their regional locations. PMF distributed sulfate mainly to a single factor (84-85%) followed by smaller contributions (5-13%) from coal and oil combustion sources. The latter may be interpreted as more “primary” than the secondary sulfate factor. The true regional sulfate contributions were much more geographically diverse.
A detailed analysis of the relationship between the PMF factors and emission sources is presented by Chen, et al. (2010). In addition to and used to infer the quality of PMF fitting, the interpretability of PMF factors with respect to known primary and secondary sources are evaluated using a root-mean-square difference analysis. PMF factors generally represent imperfect combinations of sources, i.e., the same sources are mixed on multiple factors. The optimal number of factors should be just adequate to explain the input data (e.g., > 0.95). Retaining more factors in the model does not help resolve minor sources, unless temporal resolution of the data is increased, thus allowing more information to be used by the model. If guided with a priori knowledge of source markers and/or special events, rotation of factors can lead to more interpretable PMF factors. The choice of uncertainty weighting coefficients greatly influences the PMF modeling results but it cannot usually be optimized for simulated or real-world data. However, uncertainties in the data divert PMF solutions even when the optimal weighting coefficients and number of factors are used.
Trajectory Mass Balance Regression (TMBR)
HYSPLIT trajectories were calculated every 3 hours using EDAS and high-resolution MM5 wind fields starting at 100, 200, 500, 1000, 1500 and 3000 m above each site. The TMBR model was applied to sulfate concentrations regressed on the number of 1-hour trajectory endpoints over each region. As PMF employs a non-negative solution, a non-negativity constraint was also applied to the TMBR receptor model estimates. TMBR results for BRIG and GRSM were calculated with EDAS and MM5 inputs at six different trajectory starting elevations and for 6- and 24-hour sample averaging times.
TMBR fit the average true regional contributions qualitatively well with differences apparent for trajectory starting elevation and meteorological data source. In most cases, the minor regional contributions were not estimated well. In general, the fits were better for 6-hour than for 24-hour average samples. This is a significant finding since real-world IMPROVE samples are of 24-hour duration. For 6-hr samples, EDAS data produced better fits at BRIG for starting elevations between 100 and 500 m than for higher starting elevations. The corresponding MM5-based trajectories for BRIG produced worse fits for starting elevations between 100 and 500 m. However, while EDAS-based fits were worse above 500 m, the MM5-based fits were largely invariant from 100 to 3000 m starting elevations. At GRSM, the fits were worse than at BRIG and did not vary much between EDAS and MM5 inputs. The fits for the integrated trajectories were always worse than the best fits for discrete starting elevations.
We expected that the best TMBR fits would result from trajectories based on the high-resolution (12 km) MM5 input used to generate the synthetic concentrations. This was not the case at BRIG for starting elevations below 1000 m. The three strongest source regions at each site for 6-hr samples were MANE-VU, VISTAS-E, and MIDWEST at BRIG and VISTAS-W, VISTAS-E, and VISTAS-S at GRSM. TMBR results should be better for these source regions than for all regions because their contributions were not only the largest but they are relatively independent with respect to their directions from the receptors. The TMBR results were better considering only the three strongest source regions except at GRSM with EDAS input for the 1000 m trajectories and with MM5 input for the 1500 m trajectories. Considering starting elevations from 100 to 500 m, the largest improvement in the fit was for BRIG with EDAS data and GRSM with MM5 data. Thus, the TMBR results appear better at GRSM with trajectories based on MM5 data when only the strongest source regions are considered.
Lagrangian Particle Modeling
In order to enhance the back-trajectory analysis, a Lagrangian random particle model was used in inverse mode to estimate the most probable source areas starting from receptor locations. The inverse modeling leads to explicit calculation of a source-receptor relationship as the residence times spent in respective source grid cells by particles released from the receptor.
Eight seven-day episodes were simulated with the Lagrangian model using high-resolution (12-km) MM5 fields as input. In each case, Lagrangian particles were continuously released from the receptor in inverse mode. A region with the greatest residence time was identified and a most probable source area was selected. During the first run (20 July, 2002 backward to 13 July), the most probable source impacting BRIG was identified in Indiana. As a test and a first-guess evaluation, the Lagrangian model was run in forward mode from the Indianapolis urban area. As expected, results for the forward and inverse models were similar. Comparisons were done between sources inferred from HYSPLIT back trajectories and the inverse Lagrangian model. For the July 20, 2002 case, the HYSPLIT results tended to underestimate the contributions of regions near the receptor and overestimate contributions from more-distant region. This was attributed to a rapid passage through the near-receptor area and too much weight on a particular direction limited by the one-dimensional coverage of the back trajectory. The Lagrangian model appeared to capture both of these effects due to multiple particle trajectories and dispersion covering larger area as the transport time increased.
Journal Articles on this Report : 6 Displayed | Download in RIS Format
Other project views: | All 10 publications | 7 publications in selected types | All 6 journal articles |
---|
Type | Citation | ||
---|---|---|---|
|
Chen LW, Watson JG, Chow JC, Magliano KL. Quantifying PM2.5 source contributions for the San Joaquin Valley with multivariate receptor models. Environmental Science & Technology 2007;41(8):2818-2826. |
R832156 (Final) R831086 (Final) |
Exit Exit Exit |
|
Chen L-WA, Lowenthal DH, Watson JG, Koracin D, Kumar N, Knipping EM, Wheeler N, Craig K, Reid S. Toward effective source apportionment using positive matrix factorization:experiments with simulated PM2.5 data. Journal of the Air & Waste Management Association 2010;60(1):43-54. |
R832156 (Final) |
Exit Exit |
|
Koracin D, Panorska A, Isakov V, Touma JS, Swall J. A statistical approach for estimating uncertainty in dispersion modeling: an example of application in southwestern USA. Atmospheric Environment 2007;41(3):617-628. |
R832156 (2008) R832156 (Final) |
Exit Exit Exit |
|
Koracin D, Vellore R, Lowenthal DH, Watson JG, Koracin J, McCord T, DuBois DW, Chen LWA, Kumar N, Knipping EM, Wheeler NJM, Craig K, Reid S. Regional source identification using Lagrangian stochastic particle dispersion and HYSPLIT backward-trajectory models. Journal of the Air & Waste Management Association 2011;61(6):660-672. |
R832156 (Final) |
Exit Exit |
|
Lowenthal DH, Watson JG, Koracin D, Chen L-WA, Dubois D, Vellore R, Kumar N, Knipping EM, Wheeler N, Craig K, Reid S. Evaluation of regional-scale receptor modeling. Journal of the Air & Waste Management Association 2010;60(1):26-42. |
R832156 (Final) |
Exit Exit |
|
Watson JG, Chen L-WA, Chow JC, Doraiswamy P, Lowenthal DH. Source apportionment: findings from the U.S. Supersites Program. Journal of the Air & Waste Management Association 2008;58(2):265-288. |
R832156 (2007) R832156 (2008) R832156 (Final) |
Exit Exit |
Supplemental Keywords:
multivariate receptor modeling, positive matrix factorization, trajectory mass balance regression, Lagrangian particle model., RFA, Scientific Discipline, Air, Ecosystem Protection/Environmental Exposure & Risk, particulate matter, Air Quality, Environmental Chemistry, Monitoring/Modeling, Environmental Monitoring, Atmospheric Sciences, Environmental Engineering, atmospheric dispersion models, atmospheric measurements, model-based analysis, area of influence analysis, source receptor based methods, source apportionment, chemical characteristics, emissions monitoring, environmental measurement, airborne particulate matter, air quality models, air quality model, air sampling, particulate matter mass, analytical chemistry, modeling studies, real-time monitoring, aerosol analyzers, chemical speciation sampling, particle size measurementRelevant 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.
Project Research Results
- 2008 Progress Report
- 2007 Progress Report
- 2006 Progress Report
- 2005 Progress Report
- Original Abstract
6 journal articles for this project