Advanced Receptor Modeling for Source Identification and Apportionment

1
Advanced Receptor Modeling for Source
Identification and Apportionment

• Philip K. Hopke
• Center for Air Resources Engineering and Science
• Clarkson University

2
OUTLINE

• Background
• Air Quality Management
• Receptor Models
• Factor Analysis
• Positive Matrix Factorization
• Applications
• Summary

3
Receptor Modeling

• The management of air quality involves the
  identification of the pollution sources, the
  quantitative estimation of the emission rates of
  the pollutants, the understanding of the
  transport of the substances from the sources to
  downwind locations, and knowledge of the physical
  and chemical transformation processes that can
  occur during that transport.

4
Receptor Modeling
All of those elements can then be put together
into a mathematical model that can be used to
estimate the changes in observable airborne
concentrations that might be expected to occur if
various actions are taken.
5
RECEPTOR MODELING
6
Receptor Modeling

• Thus, other methods are needed to assist in the
  identification of sources and the apportionment
  of the observed pollutant concentrations to those
  sources.

7
Receptor Modeling

• Receptor models are focused on the behavior of
  the ambient environment at the point of impact as
  opposed to the source-oriented models that focus
  on the transport, dilution, and transformations
  that begin at the source and follow the
  pollutants to the sampling or receptor site.

8
(No Transcript)
9
Receptor Modeling

• PRINCIPLE OF AEROSOL MASS BALANCE
• The fundamental principle of receptor modeling
  is that mass conservation can be assumed and a
  mass balance analysis can be used to identify and
  apportion sources of airborne particulate matter
  in the atmosphere.

10
Mass Balance

• A mass balance equation can be written to account
  for all m chemical species in the n samples as
  contributions from p independent sources

Where i 1,, n samples, j 1,, m species and
k 1,, p sources
11
Receptor Modeling

• The question is then what is known a priori to
  solve this equation.

• Divide the problem into two classes
• Source Profiles Known
• Source Profiles Unknown

12
Mass Balance

• A mass balance equation can be written to account
  for all m chemical species in the n samples as
  contributions from p independent sources

Where i 1,, n samples, j 1,, m species and
k 1,, p sources
13
Receptor Modeling

• SOURCES PROFILES KNOWN
• Chemical Mass Balance
• Multivariate Calibration Methods
• Partial Least Squares
• Artificial Neural Networks
• Simulated Annealing
• Genetic Algorithm

14
Mass Balance

• However, generally we do not know source profiles
  and we only have the available ambient
  concentration data. Thus, can we deduce the
  number and nature of the sources and their
  contribution to each sample through an
  appropriate data analysis method?

15
Receptor Modeling

• SOURCES PROFILES UNKNOWN
• Factor Analysis
• Principal Components Analysis
• Absolute Principal Components Analysis
• SAFER/UNMIX
• Positive Matrix Factorization

16
Factor Analysis

• Most factor analysis has been based on an
  eigenvector analysis. In an eigenvector
  analysis, it can be shown Lawson and Hanson,
  1974 Malinowski, 1991 that the equation
  estimates X in the least-squares sense that it
  gives the lowest possible value for

17
Factor Analysis

• Thus, most factor analysis use an unrealistic
  unweighted least-squares fit to the data.

18
Factor Analysis
The problem can be solved, but it does not
produce a unique solution. It is possible to
include a transformation into the equation.
XGTT-1F where T is one of the potential
infinity of transformation matrices. This
transformation is called a rotation and is
generally included in order to produce factors
that appear to be closer to physically real
source profiles.
19
Factor Analysis
20
Positive Matrix Factorization

• Explicit least-squares approach to solving the
  factor analysis problem

• Individual data point weights

• Imposition of natural and other constraints, and

• Flexibility to build more complicated models

21
Positive Matrix Factorization

• The Objective Function, Q, is defined by

where sij is an estimate of the uncertainty in
xij
22
IMPROVE Monitoring Network

• IMPROVE Interagency Monitoring of Protected
  Visual Environments
• IMPROVE aerosol sampler 4 modules

Source http//vista.cira.colostate.edu
23
IMPROVE aerosol monitoring

• Module A PM2.5 on Teflon filter (UC, Davis)
• Gravimetric mass
• Proton Elastic Scattering Analysis (PESA) for
  hydrogen
• Proton Induces X-ray Emission (PIXE) for Na Mn
• X-Ray Fluorescence (XRF) for Fe Pb
• Module B denuder, PM2.5 on nylon filter (RTI)
• Ion Chromatography for sulfate, nitrate,
  nitrite, chloride
• Module C PM2.5 on quartz filter (DRI)
• Thermal Optical Reflectance (TOR) method for 8
  carbon fractions
• Module D PM10 on Teflon filter

24
Carbon Analysis

• As part of the IMPROVE protocol for the
  measurement of organic and elemental carbon
  (OC/EC), individual carbon fractions of OC (OC1,
  OC2, OC3, OC4) and EC (EC1, EC2, and EC3). In
  addition, the pyrolized fraction of the organic
  carbon (OP) is estimated

25
Carbon analysis IMPROVE/TOR method
Source Chow et al., 2001
26
Washington, DC monitoring site

• Roof of the Natl. Capitol Region Park Police HQ
• 3 km NE of Ronald Reagan Washington Natl. Airport
• 2 km SE of Lincoln Memorial

27
Se vs. S
28
Silicon 1

• NOAA HYSPLIT model was used to calculate air mass
  backward trajectories for days with high Si conc.

6/24 7/7, 1993
29
Silicon 2

• Asian dust clouds
• 4/6 developed over Mongolia
• 4/13 start to impact the west coast

4/9 4/22, 2001
30
Fireworks contributions

• July 4 fireworks contributed to high conc. of K,
  Pb, and Cu

7/4/92
7/4/98
7/5/00
K
Pb
Cu
31
OC/EC Fraction Results

• A total of 718 samples collected between August
  1988 and December 1997 and 35 species were used
  in this study.

32
OC/EC Fraction Results
33
Secondary Particles
34
Secondary Particles
35
Sulfate Factors

• When the carbon thermal fractions are added to
  the data set, we have also extracted a third
  sulfate factor in addition to the winter/summer
  factors
• This factor has been seen in data from Atlanta,
  GA, Washington, DC, and Brigantine, NJ.

36
Sulfate-OP Factor
37
Sulfate Factors in Washington, DC study
Summer-highOhio river Valley, eastern
Tennessee, southern Mississippi and Alabama
OP-highCanadian boreal fire, Central American
forest fire
Winter-highNorth Carolina, Midwestern areas,
southeastern Texas Louisiana
38
Sulfate Factor

• With the carbon thermal fractions, the amount
  of carbon in the summer and winter sulfate
  factors drops substantially compared to analyses
  with total OC and EC.

39
Sulfate Factor

• Why is there a covariance between OP and
  sulfate?

• Is this an indicator of secondary organic
  aerosol formation?

• Is the secondary organic aerosol formation
  catalyzed by the acidity of the sulfate particles
  as suggested by Kamens?

40
Combustion Sources
41
Spark-Ignition in Multiple Cities
42
Diesel in Multiple Cities
43
OC/EC Fraction Results
44
OC/EC Fraction Results
45
Conclusions

• We have tools to help analyze the complex
  compositional data being produced by the major
  monitoring networks in the United States.
• These techniques will likely play an important
  role in the development of air quality management
  plans over the next several years.

46
Thanks To

• Eugene Kim and Bilkis Begum for performing the
  analyses presented.
• Environmental Protection Agency and the
  International Atomic Energy Agency for financial
  support.