Title: PM source apportionment using Factor AnalysisMultiple Regression FAMR model: a case study in Bangkok
1PM source apportionment using Factor
Analysis-Multiple Regression (FA-MR) model a
case study in Bangkok urban area
By Hathairatana Garivait Environmental Research
and Training Centre, Department of Environmental
Quality Promotion, Ministry of Natural Resources
and Environment 4 March 2008
2Contents
- Introduction to source impact assessment methods
- of air pollution
- 2. Fundamental of receptor model
- 3. Factor Analysis-Multiple Regression Model
(FA-MR) - Application of FA-MR for PM source apportionment
- in Bangkok urban area in 1996
3Source impact assessment methods for air pollution
Dispersion models
Receptor models
4Dispersion model
- The dispersion models can deal explicitly with
emissions from - single, identifiable sources within the same
source category. - However, the dispersions model must build on
several potentially - uncertain inputs, such as
- emission data
- meteorological data
- the transport-diffusion-transformation-deposition
mechanisms -
5Receptor model
Receptor models start with the measurement of a
specific feature of the aerosol at the receptor
and after the fact calculate the contribution of
a specific source type. Measurable atmospheric
features include particle size distribution, compo
nent identification (organic, inorganic and
radioactive), component chemical state and
concentration, time and spatial variation.
6Air Quality Models(These provide approaches to
identify source impacts)
Curtsey to Prof. Armistead Russell, GIT, USA
7Receptor Models
Microscopic methods
Chemical methods
Optical
Qualitative analysis
S.E.M.
Chemical Mass Balance
Quantitative analysis
Automated S.E.M.
Multivariate data analysis method (e.g.,FA-MR)
8Comparison between the diffusion model and
chemical mass balance model
9(No Transcript)
10Thus, the receptor model and dispersion model are
complimentary in their approach to source
apportionment and using both can reduce the
limitation of each alone.
11Receptor model as a tool for SPM source
apportionment
12Size distribution of aerosols in the atmosphere
13Size ranges that commonly found chemical
constituents in atmospheric aerosol
Source as cited in Sharma (1994)
14Nature of the suspended particle matter
15Relationship between the emission source and
suspended particulate matter (SPM) in the air
elemental composition for 1st source category
16unit for the measurement of suspended particulate
matter
17Fundamental of Receptor Model Chemical Mass
Balance (CMB)
Receptor data (given)
Target to be estimated
Source profile (given)
Sjk
Cik
aij
18Chemical Mass Balance
Ci ambient concentration of species i fi,j
fraction of species i in source j
uncertainty Sj source contribution of source j
effective variance weighted least squares
regression
Curtsey to Prof. Armistead Russell, GIT, USA
19Knowledge of source signatures
20Determination of Source Profiles
Diesel
Biomass
Soil
Automobile
Often use metals in PM as well.
Curtsey to Prof. Armistead Russell, GIT, USA
21Source profile of Okamoto et al. (1990)
(unit )
22Marker elements associated with various emission
source
Important elements for receptor model
23PAH as alternative tracers for motor vehicle
emissions
Automobile traffic BaP, BghiP, Cor
24Limitation of CMB method
- Since CMB method use the statistical properties
of the chemical composition data, we can not know
more than the statistical inference. - If the sufficient number of marker elements was
not analyzed, and/or sampling error and analyzing
error are large, the estimated contribution for
each emission may be questionable. - If we have more than two emission sources, and
they discharge the particulate matters, which
have similar chemical composition, we can not
separate the contributions for there sources.
25Multivariate Methods Factor Analysis
Background 1. Because of the nature of
atmospheric processes and meteorology, the
concentrations of individual air pollutants
will often vary simultaneously and this
occurs irrespective of the sources. 2. The
difficulties in differentiating individual
sources due to similar marker elements
The objective of multivariate methods is to
detect the common variability after the fact and
imply source identity by comparing the elements
with common variability to the elements
associated with specific sources.
26Factor Analysis-Multiple Regression model
FA-MR model has its advantage in identifying the
source categories present at the receptor sites
even when information regarding the emission
sources are insufficient. FA To
identify sources of PM and to select source
emission tracers by grouping the
selected variables according to their common
variations. The principle is to find a minimum
number of factors that explain most of the
variance of the system. MR To obtain a
quantitative relationship between the source
tracers and particle mass
concentration.
27RELATION BETWEEN FACTOR ANALYSIS AND CMB MODELS
- The fundamental relation between the
concentration at a receptor site and source
information can be expressed as - where Cik is the concentration of element i in
sample k and aij is the composition of element i
of source j the fractional abundance of element
i in the j-th source profile, Sjk, is the
contributing mass concentration (contribution) by
the j-th source to sample k. - The basic model for a factor analysis is of the
form
28RELATION BETWEEN FACTOR ANALYSIS AND CMB MODELS
Receptor data (given)
Target to be estimated
Source profile (given)
Sjk
Cik
aij
Factor score
Factor loading
29Application of Factor Analysis-Multiple
Regression for PM Source Apportionment in Bangkok
urban area in 1996
30Objectives
To identify and quantitatively estimate the
source contribution of airborne PM pollution in
Bangkok urban air using a combination of PAHs and
elemental compositions as a useful tracer in
FA-MR model
31study area
32Number of samples 47
33(No Transcript)
34Number of variables 44
35DATA SCREENING
47 samples 44 variables
42 samples 23 variables
N gt 30 (v 3)/ 2
36Correlation matrix
37Concentration Data
FACTOR ANALYSIS
Factor Loading
Factor Score
Calculation of Absolute Factor Score
Estimate probably major source contributed to
the receptor site
Multiple Regression
Source Contribution
Principle of FA-MR quality model, version 1.0
38FA-MR MODEL CALCULATION
- Calculate correlation matrix
- Calculate Eigenvalue
- Calculate Factor Loading
- Calculate Absolute Factor Score
- Regression analysis for estimating the mass
- concentration by using AFS as the predictor
variable - Estimate the new factor loading related to the
- source composition
39Varimax rotated Factor Loading
40Results of Factor Analysis
Factors Prevail variables in Probable source
type Factor loading Factor 1 Al, Sb, Sm,
As, Ca, Fe, Soil road dust
Pb, Mn, SO42- Factor
2 Cu, Fe, BeP, BkF, BaP, Automobiles DBahA,
BghiP Factor 3 Cr, Mn, Na, V, Zn, Industry
Sea salt Cl-, SO42- Factor 4 Zn, Pb, Corg,
Cele Refuse incineration open
burning Factor 5 As, SO42-, DBahA,
Cor Secondary pollutants
41Results of multiple regression analysis
SPM Z1 (AFS)1 Z2 (AFS)2 Z3 (AFS)3 Z4
(AFS)4 Z5 (AFS)5
42Results of multiple regression analysis
SPM Z1 (AFS)1 Z2 (AFS)2 Z3 (AFS)3 Z4
(AFS)4 Z5 (AFS)5
The multiple correlation coefficient 0.79
43Source contribution ()
44Absolute contribution vs wind direction
45CONCLUSIONS
The results of FA-MR model revealed a
consistency of PM pollution sources in Bangkok
urban air in which soil and re-entrained road
dust as well as automobiles were the major
contributors. Besides, the results of Factor
2 revealed that air pollution by airborne PAH in
Bangkok were mostly from automobiles exhaust.
46THANK YOU VERY MUCH FOR YOUR KIND ATTENTION