1. Source Apportionment in Indian Cities: Methodologies and Findings of Recent Studies
Dr. Rakesh Kumar Scientist and Head, Mumbai Zonal Center, National Environmental Engineering Research Institute, 89B, Dr.A.B.Road, Worli, Mumbai , INDIA October 18, 2004

2. Understanding of Air Quality In India
Major emphasis on emission loads
Visible pollution considered the major issue
Earlier : industries
Now: vehicles
Apparent health impacts : Aggravation of asthma

3. Realistic Expectations Public
Realistic Expectations Policy Makers
Realistic Expectations Public
What areas of air pollution will make the biggest Impact ?
PM,
Visual aspects,
SO2, NOx, O3, etc.
Which method or combinations will show immediate improvement ?
How much will it cost?
Will it have mass appeal?
Realistic goals:
What level of pollution reduction is achievable?
What level of pollution reduction is acceptable?
The answers to all these have to come from a combined team of :
Air pollution specialist
Health experts
Environmental economist ,
Policy makers
and
PUBLIC

4. Realistic Expectations Researcher
Complete understanding of all the sources within the city, region, country and elsewhere
Carry out programmes which also support the expectations of the local/national policy maker (CNG/Diesel ??)
Effective validated models for forecasting and continuous upgraded system
More stronger linkages with MET researcher for effective outputs
Others
Source Apportionment is one such tool:
can answer some of the relevant questions

5. Indian Background
First emission law for ambient air was promulgated in the year 1905, called Bengal Smoke Nuisance Act.
First major source inventory was prepared for Mumbai (Bombay) in 1968 by NEERI.
Industrial stack dispersion studies started sometime in early 80’s,
NEERI prepared emission inventory for three cities (Mumbai, Calcutta and Delhi) in the year 1990

6. Why Source apportionment
NEED TO UNDERSTAND SOURCES
When regulatory levels are exceeded: investigations required for sources
Important to know the possible sources
Identify potential sources and meteorological conditions to assist policy makers and modelers in developing control strategies
Plan to know how long the present emission inventories and dispersion models represent the ambient conditions which can be used for prediction and control strategies for future
Action Plan development for all the principal contributors to air pollution
Cost effectiveness of control plan for each of the sources

7. IMPORTANT TO KNOW: LOCAL VS. OUTSIDER
For strategies development, decision maker must know which pollutants set (group) as well as its corresponding quantities belongs to his area.
Multiple Techniques for assessing the difference:
Spatial and temporal analyses (e.g., Are high concentrations observed on a regional basis or only at a few “hot spots”?).
Assessing the age of an air mass accompanied with trajectory analysis.
The use of “tracers-of-opportunity” and species ratios accompanied with trajectory analysis (e.g., using potassium to identify forest fire impact).
The use of satellite information to corroborate transport (e.g., Saharan dust storm impact on U.S. sites).
Model the dependence of PM on ozone to determine a component of PM that is photochemically produced.
Source: Schichtel B.A.(1999)
PMFineAn/ PM_vs_Tran/finalreport/BltPhx_PMvsWnd_Final Report.html

8. UNDERSTANDING LOCAL AND OUTSIDE SOURCE Particulate Matter
First, assess the differences between the concentrations of average urban and nearby rural monitoring data.
Conduct preliminary modelling exercise to assess the PM dependence on wind speed and wind direction to understand the site characteristics (local/outside)
Some developed countries are preparing “regional background” profiles to assist in apportioning PM. These profiles could be used in an attempt to quantify the regional contribution to PM concentrations. (Schichtel B.A.;1999)
It has been noticed that sites are strongly influenced by sources less than 10 km distance
On the other hand even minor sources close to the sampler could overwhelm any outside component in a 24-hr integrated sample (VanCuren, 1998).
However, individual small emitters can have a zone of influence less than 1 km (e.g., Chow et al., 1999).

9. Source Apportionment
Source apportionment in India poses many daunting questions.
Particularly in urban centres, many of these places are actually mixture of industrial, commercial, village, slum and high transport activities.
All of these activities have diverse sources and therefore, complex emission characteristics.

10. The First Step ? Towards SA is Emission Inventory: - sources of the criteria pollutants, - amount of each pollutant emitted, - any processes or control devices utilized Needs of Emission Inventory: - quantifying the sources with a view to know the locations and assess the impact on human health

11. Source apportionment requires addressing following issues :
Major problem is to identify all the possible sources such as:
cow dung burning,
waste (of various types) burning,
resuspended dust (of various types),
unorganized small scale industries,
varying degree of many of such activities,
addition of new sources etc.
Emission profiles of all these sources
Picking up of some of these sources and their emission factor from USEPA (AP-42, AIR Chief)
Methodologies of data generation
Analysis of parameters such as BC and EC/OC as per international norms

12. Emission Factor: What do we have?
The Second Step?
Emission Factor: What do we have?
Emission factor of all the sources (at least the major sources) in the present condition shall have to be developed and used.
One of the major uncertainty in emission factor of major source in urban centres is vehicular sources.
India still has large numbers of vintage technology vehicles, two and three wheelers, other types of utility vehicles.

13. Indian Institute of Petroleum (IIP)
Emission Factor (Contd…) Indian vehicle emission factors are not available for certain vehicle types: particulate matter are not available as shown below :
Vehicle Category
Indian Institute of Petroleum (IIP)
W. B. Literature
Two-Wheelers -
0.21
Three-Wheelers
2.1
Cars/Taxis
0.27
Four Wheelers
Urban Buses
0.275
Trucks
0.450 3
Light Commercial Vehicles
0.100
All values are g/km. The petrol four wheelers no PM emission factor is available. The World Bank literature refers to some estimates used in a study carried out in Dhaka, Bangladesh

14. Dispersion Modelling
Limited work carried out: mainly for point sources
Large scale models not yet applied (MM 5, UAM etc.)
Low scale models: kerb-side data recently started
Problems of data and meteorology for each of the sites
Low cost portable met station

15. Development Planning Urban planning based on air quality goals
Damage control planning based on air quality standards
Health based linkages for urban air quality improvement and study to establish the correlation
YET TO TAKE OFF!

16. Source Apportionment: How We Move ahead?
Limited Data Availability
Limited Need ??
Larger view to take abatement decision

17. Our Situation
SPM values for developed countries are normally less than 100 g/m3
(Rojas et al 1990,
Camuffe &Bernardi, 1996)
India has annual averages : 200 – 500 g/m3
(Sadasivan & Negi, 1990
Sharma & Patil, 1992,
NEERI NAAQM Data )
No source composition available
(Sharma & Patil, 1994)

18. International Studies
Thurston and Spengler, 1985
Quantitative information for apportionment of PM could be found based on characteristics of particulate matter with minimum information about the sources
Assessed the air quality on a regional scale for fine (<2.5 µm) and coarse( µm) fractions
Identified a fugitive source (coal combustion aerosols transported in the area) which accounted for 40% of the mass
Thurston et al. (1985) conducted a case study to demonstrate that quantitative information for apportionment of PM could be found based on characteristics of particulate matter with minimum information about the sources. They assessed the air quality on a regional scale for fine (<2.5 µm) and coarse ( µm) fractions. The characterization of PM2.5 resulted in the identification of a fugitive source (coal combustion aerosols transported in the area) which accounted for 40% of the mass.

19. International Studies (Contd…)
Chan et al., (2000)
PM mass (42%) related to elements from natural sources found in >2.7 µm size range
PM mass (41%) related to human activities found in <0.5 µm size range
The soil and sea salt factors contribute more than 80% of the mass of the >2.7 µm aerosol samples, while the organics and vehicular exhaust factors explained 95% of the mass of the <0.61 µm aerosol samples
Marcazzan et al. (2001)
Reports that the coarse and fine size fractions of the ambient aerosol are dominated by different sources
Coarse particles were dominated by elements of crustal origin, whereas fine particles were enriched with the presence of elements from anthropogenic sources
Chan et al. (2000) sampled PM in Brisbane (Australia) using a high volume cascade impactor. They report that 42 % of the mass of PM in the >2.7 micrometer size range comes from natural sources and 41 % of the mass in the <0.5 micrometer size range was related to human activities. The soil and sea salt factors contribute more than 80% of the mass of the >2.7 µm aerosol samples, while the organics and vehicular exhaust factors explained 95% of the mass of the <0.61 µm aerosol samples

20. India Specific Studies
Sadasivan et al. (1984); Negi et al. (1987 and 1988), Sadasivan and Negi (1990)
All of these were carried out on TSP (SPM)
Crustal source as the major source of PM was reported
Major emphasis during this period was lead
Mostly metals analysis based interpretations
Sampling was done using a vacuum pump with a filter mounted on it
In all these studies, sampling was done using a vacuum pump with a filter mounted on it. They reported crustal source as the major source of PM measured by this method.

21. India Specific Studies (Contd…)
Sharma and Patil (1992 and 1994)
PM < 30mm using a high volume sampler
Factor analysis of the elemental and ionic concentration data resulted in the identification of seven source types
Crustal and marine sources were identified as the highest contributor
Kumar et al. (2001)
Sampled SPM at two traffic junctions in Mumbai, representing a mixed industrial, vehicular and residential site
Road dust and vehicular emissions were found to contribute 40% and 15% to the SPM
Sharma and Patil (1992 and 1994) further studied specifically for PM < 30mm using a high volume sampler at industrial and roadside locations of Mumbai. Factor analysis of the elemental and ionic concentration data resulted in the identification of seven source types. Crustal and marine sources were identified as the highest contributor to the mass in the samples collected.

22. Source Apportionment at Traffic Junction A Case Study, 2001
Why?
High levels of SPM at Traffic Junction
Technique used : factor analysis – multiple regression and receptor modelling
The varimax rotated factor analysis identified five possible sources (qualitative) :
- Road dust
- Vehicle emission
- Marine aerosol
- Metal industries
- Coal combustion

23. Parameters Data Analysis
Cr, Fe, Ca, Al, Mg, K, : Digestion and Flame AAS
Na, Cu, Mn, & Ni
Hg, As : Using hydride generator AAAS
Some part of filter paper : Spectrophotometer
Used for water soluble
fraction
Data Analysis
Factor Analysis – Multiple Regression
(it does not require prior information on source composition)
(Hopke, 2000, Henry, 1984)

24. Source Apportionment at Traffic Junction
Quantitative factor analysis – multiple regression Model indicated
Sakinaka Gandhi Nagar
- Road dust : % %
- Vehicle emission : % %
- Marine aerosol : % %
- Metal industries : % %
- Coal combustion : % %

25. Source Apportionment at Traffic Junction (Contd…)
Due to limitations in source marker elements analysed, about 16% of SPM could not be apportioned.
Observed Pb, about 62% Automobiles
17% Road Dust
11% Metal Industries
07% Coal Combustion
03% Marine Aerosol

26. India Specific Studies
NEERI, PMRAP studies
SPM and PM10 were measured at various locations in Mumbai
Diesel and gasoline vehicle exhaust emissions accounted for 6% to 54% of PM10 at different locations
Industrial sources accounted for 6% to 42% of PM10 at these sites
Other sources were identified as resuspended dust (10% - 20%) and marine aerosols (12% - 14%)
To reduce the PM load in Mumbai, various recommendations and strategies have been reported in a recent study (NEERI, 2004). In this study SPM and PM10 were measured at various locations in Mumbai. Elemental and ionic concentration data along with Black Carbon data were used for source apportionment of PM at four sites representing different activities. The diesel and gasoline vehicle exhaust emissions accounted for 6% to 54% of PM10 at different locations. Industrial sources accounted for 6% to 42% of PM10 at these sites. The other sources were identified as resuspended dust (10% - 20%) and marine aerosols (12% - 14%).

27. NEERI, PMRAP studies 2002-03 (Contd…)
Colaba
Metro Junction
Vile Parle
Mazgaon

28. Kanpur Source Apportionment Study
Industrial
Commercial
Residential
Kerb Side
Auto exhaust -- 16
Auto exhaust and
diesel generating sets 32 22 39
Resuspended dust 24 30 20 31
Secondary aerosol
formation 12 8 10
Earth crust 6 14
Smaller scale industries
Other sources 23 29
Study carried out for a limited period
Major Outputs:
Automobiles contribute in the range of 16-39%
Resuspended dust ranged 20-31%
Other identifiable sources (earth crust, secondary aerosols) : 6-12%
Industries : 8-16 %

29. Major outputs : Kanpur
The method of using factor analysis did not lead to identification of about 24-29% of the sources
Industrial sites were necessarily not impacted by the industries
Impact of automobile and diesel gensets were felt more in residential areas compared to even kerbside
Limited information collected only for PM10 necessarily is not providing the complete details
Fine particles (PM2.5) if monitored can provide more precise information with a simultaneous measurement of organics as well (EC,OC,BC, hopanes steranes etc.)

30. Methodologies used for SA
Principal Component Analysis
PCA can be used without the source profile composition
It can be used to identify the missing sources
It can use tracers which are somewhat reactive
However, it needs large number of receptor samples and also know how many factors to retain
Needs our judgment to identify the factors responsible for the respective sources
It can give negative values that cannot be accounted for any source

31. Methodologies used for SA
Positive Matrix Factorization
Modelling using PMF for source apportionment is comparatively new
PMF is a multivariate modelling, where the source profiles are not needed
It identifies factors and their sources at a place
It does not lead to negative values of chemical components unlike PCA
It can also handle missing or very low values in the input data as also the uncertainties in input measurements
Example of PMF : Yakovleva et al. 1999, Env.Sc.&Tech. 33, pp.3645

32. Use of CMB in India
Sharma et al ( ) used CMB 7 for a limited area in Mumbai
Source profile from the Source Composition Libraries of USEPA Profile
Based on known profile of Bombay, sources selected were: vehicles, combustion processes (industrial and others), resuspended dust, sea salt, ferrous and non-ferrous industrial sources
Model was run using 19 elements and 7 source types
Only 48.5,9.1 and 69.3% of total mass of TSP was accounted for three sites respectively.
Major reasons for non-applicability were:
In-sufficient source profiles
EC,OC and HC were not analysed
Secondary pollutants were also not included in the fit

33. CMB use in World Bank Study: 2002
First time PM2.5 SA carried out in South Asia : Megacities of India [Delhi, Mumbai, Kolkata]
Major sources indicated are:
VEHICLE EXHAUST
RESUSPENDED DUST
SOLID FUEL (BIOMASS BURNING)
MAJOR OUTCOME
It could identify the major crustal sources at all the sites
Marine sources contribution was found to be same at all places
Also knowing nearby sources alone is not good enough
It was difficult to separate the diesel based stationary and mobile
Sources contribution
Road dust was the largest contributor [even for PM2.5]

34. Advantages and Disadvantages of CMB
Suitable for single receptors
Specific identification of well known sources
Trace metals and ionic species can be used
Even stable organic tracer can be used
Need to have complete composition of sources
Can not have missing sources
Non-reactive sources/tracers
Problems of collinear Sources

35. What are the problems ? India Specific Issues
Though receptor modelling appears a better option some of the major issues are:
Limited locations and their results are not/ may not be representative for the whole city
Highly data intensive exercise (large sets of data, large numbers of variables: better results)
Large issues of QA/QC of such data generation
Large scale data collection and its updating a mammoth task
Source profiling is most difficult in India:
Shifting industrial practices
Changing land use pattern
Changes in diffused source combustion etc.
Lack of resources
Single/multiple agency for data generation

36. Possible pathways for Source Apportionment Study plan
Crustal PM contributes to the mass by virtue of its size, and masks the contribution of anthropogenic (toxic) PM in the near 1.0 mm size range
Perception as well as some SA studies in India suggests that background (many times outside) PM need careful assessment before complex and expensive action plans are adopted

37. Use of PM2.5 Sampler and Analysis by EDAX , ICP and Ion Analyser
Ambient Air
Gaseous
High Volume Sampler PM
PM10
Use of PM2.5 Sampler and Analysis by EDAX , ICP and Ion Analyser
Single Stage Impactor
PMx, x = 2.5, 10
Outline of a S-A Study for PM
MOUDI cascade impactor
Validate receptor modeling procedure (SPSS/UNMIX)/ Others
Mass size distribution
Chemical Analysis
The objectives can be viewed graphically in this figure. The ambient air sampling is done for gases and PM. The PM can be measured using different instruments to get PM10, PM2.5 or size fractionated PM. The focus of this study is on PM2.5 and the samples collected will be analyzed for about 35 elements and 5 ionic species. This data will be subjected to receptor modeling for source apportionment and possible source identification. The study will quantify the contribution of major sources of PM in Mumbai, which include vehicular, industrial, crustal and other sources and such data is essential for the formulation of air quality management policies and source control strategies.
Receptor Modeling Techniques
Identification of air quality management policies and strategies
Source Identification and Apportionment

38. Recent Developments Multivariate Statistical Receptor Models
Aerosol compositional data containing a number of species are obtained for each sample
The concentrations of all species for all of the samples form a matrix
This matrix can be decomposed into two matrices representing source contributions and source profiles by using various mathematical techniques
UNMIX model (Henry, 1997 and 2003)
The advantages of this model are
No assumptions about the number or composition of sources
No assumptions or knowledge of errors in the data needed
Automatically corrects source compositions for effects of chemical reactions
These advantages of UNMIX make it a suitable choice as the receptor modeling tool for the present study
The various types of multivariate models are Factor Analysis, Principal component analysis, Positive Matrix Factorization, and Target Transformation Factor Analysis. The most recent development in this area has been UNMIX model. The advantages of UNMIX are that there are No assumptions about the number or composition of sources;
No assumptions or knowledge of errors in the data needed; and
The model Automatically corrects source compositions for effects of chemical reactions.
These advantages of UNMIX make it a suitable choice for using it as the receptor modeling tool for the proposed study.

39. UNMIX – Problem definition
One source
If given with a data set of compositions of many species for many samples
With as few assumptions as possible, find the number of sources, the composition of the sources and the uncertainties
General mixture problem (ill-posed)
Two sources
Now I will discuss the theory of UNMIX receptor model. If we are given a data set of compositions of many species for many samples, then with as few assumptions as possible, we have to find
the number of sources, the composition of the sources, and the uncertainties.
This kind of problem is called as General mixture problem. This problem is ill-posed i.e. there are insufficient number of constraints to define a unique solution. UNMIX solves this problem by using the data set itself as the constraint.

40. UNMIX - Model Output The number of sources Composition of each source
The final output of the model includes:
The number of sources
Composition of each source
The source contributions to each sample
Uncertainties in the source compositions
Apportionment of the average total mass
The final output of UNMIX includes
The number of sources;
Composition of each source;
The source contributions to each sample;
Uncertainties in the source compositions; and
Apportionment of the average total mass.

41. Possible Solutions
Need based data collection ( e.g. meat cooking data may not be relevant)
Emission factor to be developed (dependence on US-EPA should be reduced)
Source profiles for some major activities need development (resuspended dust, refuse burning, etc) to be developed
Link the data collection with census activities or economic survey at least every 5 years

42. Possible Solutions (Contd…)
Source apportionment: Using Principal Component Analysis (a type of factor analysis modelling) or Positive Matrix Factorization
[these techniques also have limitations]
UNMIX could be an another possibilities
Methods which can be easily replicated elsewhere should be used
As sample analysis facilities are easier to locate and used in India:
Till we have better source profiles, We can use PCA, PMF, UNMIX or FA-MR !!

43. THE NEED IS TO START USING RECEPTOR MODELS, AS IT PROVIDES MANY ANSWERS WHICH WERE MISSING EARLIER:
1- These models can prioritize various sources based on their effective  contribution at the receptor points rather than just emission loads. This is better for health linkages.
2- Fugitive sources can be identified only by RM. Its almost like fingerprinting of all sources unlike Dispersion Models
4- RM can generate the actual/live data on pollution levels.
5- RM technique can be used for improvement in emission factors by source inversion methods, tracing deposited material etc.
Conclusions
In most cases, S-A study based on receptor modelling warrants large sets of data
Though SA provides better estimates of what are the contributing sources, its result of one (few) point can not be applied to the whole city
Data creation and its use requires constant updating as also institutionalization

44. What has happened ? Air quality improvement have taken place
Poor quantification and credible reporting
One sided success stories
Limited research and policy intervention success stories
Air quality studies highly biased for Transport sector
Many simpler policies may reduce PM emissions to greater extent.

45. URBS PRIMA INDIS
With teaming crowd pouring into the El Dorado that had come to be in the popular eye, the city, it appeared would burst at the seams at any moment.
Increasing population resulted in shortage of housing space, high rent, bad air quality, poor sanitation facilty, spoilt sea shore -………………………………
Mumbai of 2000’s ??? NO
It is Bombay of 1850’s
Source: Albuquerque, T.; Urbs Prima Indis: An Epoch in the History of Bombay