Model Types
POINT (STACKS)
SOURCE GEOMETRY
LINE (HIGHWAYS)
NON POINT
AREA (CITIES)
CONTINUOUS (PLUME)
RELEASE TIME
INSTANTANEOUS (PUFF)
(EXPLOSIONS, ACCIDENTS)
SIMULATION TECHNIQUE PHYSICAL : LAB SCALE MODELS
(EG. WIND TUNNEL , SMOG CHAMBER)
MATHEMATICAL MODELS
TIME AND SPACE SCALES
STEADY STATE OR DYNAMIC
CONSERVATIVE AND NONCONSERVATIVE
STOCHASTIC OR DETERMINISTIC OR ADAPTIVE
DISPERSION AND RECEPTOR MODELS
***RANGE FROM SIMPLE EMPIRICAL TO COMPLEX METEOROCHEMICAL MODELS
Air Quality Model Types
Air quality models
Dispersion/ Source/ Causal Receptor/ NonCausal
(SR) (RS)
Types: (a) Multi variate statistical analysis
(a) Rollback : AQ α Q (b) Chemical mass balance
C α Q C fQ f Q
i 1 1i 2 2i
(b) BOX: Neglects diffusion
Q Ci f jQ ji
Cα
u i = species
(c) Gaussian dispersion model j = sources
Q Q ji = fraction of i in source j
C α D
u fj = conc. Of source j effecting receptor
(d) Numerical models Eg. two sources:
(Mass Conservation Approach) C i f 1 Q 1i f 2 Q 2 i
So solve equations for f1 & f2
Receptor Models: (Example)
Mass Conservation: Total mass of Pb at any receptor point (R) is sum of contribution to
R of all sources emitting Pb.
CPb (R) = CPb(auto) + CPb(Smelter) + CPb(others)
Now, CPb(auto) = fauto* Qauto(Pb)
Where, Qauto(Pb) = fraction of Pb from total emissions in auto (gm/gm)
fauto = mass conc. of auto effecting R (ug/m3)
it depends on distance, meteorological conditions, etc.
Therefore, CPb (R) = fauto* Qauto(Pb) + find * Qind(Pb) (2 sources say)
or Ci = f1 * Q1i + f2 * Q2i
If two sources then measure another species k,
Ck = f1 * Q1k + f2 * Q2k
Here, f1 and f2 will remain same for i & k.
if Ci ,Ck , Q1i , Q1k ,Q2i and Q2k are known, the simultaneous equation can be solved to determine
f1 and f2 which is % contribution of source 1 & 2 to R.
General Equation: Units:
C = gm/m3
C = C1 + C2 + C3 -----------+ Cn
f = gm/m3
Ci = C1i + C2i + C3i -----------+ Cni
Q = gm/gm
C1i = f1 * Q1i
or Cji = fj * Qji
Therefore, Ci =
where, i are species and j are sources.
if i > j, i.e. no. of species > no. of sources, then the simultaneous equation can be solved to
know f1, f2 and fj
More the species, better the fitting. So, large data sets are needed.
Solution Methods:
Linear Programming, Least Square Method, Back Trajectory, HYSPLIT
Source Model vs. Receptor Models
Known source Estimated
emissions ambient
Source models
concentrations
Known
dispersion
parameters
Some known
source
characteristics
Receptor models Estimated
Known ambient source impacts
concentrations
Some known
dispersion
parameters
Capabilities
Identification of unknown sources
Known Known ?
Unknown
Source
Site 1 Site 2 Site 3 Site 4
Automobiles Iron & Steel Industries
Petroleum Refineries
Power Plants
Apportioning the contributions to monitored ambient
levels of pollutants among various sources
Basic Data Requirement
Complete speciation of particulate matter
•Physical (Size distribution)
•Elements (Heavy metals)
•Inorganic/Organic
10 10 10
15
20
30
15
20
25
25
5
30
30
15
40
Concentration
Uncertainty
Concentration Uncertainty
Signature Compounds
Source Indicator Size fraction
Motor Vehicles Pb, Br Fine
Soil & Road dust Si, Al Coarse
Fuel oil combustion Ni, V Fine
Cement Ca Coarse
Marine aerosols Na, Cl Coarse
Diesel emissions Ba, Pb Fine
Fly ash Se, As Total
Advantages and Disadvantages
Disadvantages of Dispersion model Advantages of Receptor model
• Accuracy depend on EI and MET. • Not needed
Data • Capable of identifying unknown
emissions
• Fugitive/ unknown emission
unidentifiable
Advantages of Dispersion model Disadvantages of Receptor model
• Universal • Site specific
• AQ data needed only for model • Require large data sets
validation • Diagnostic model
• Predictive model • Useful for prioritizing pollutants &
• Useful for planning and sources
management
Rollback Model
AQ or Concentration α Emissions
or C α Q
C Q C B or
2
2 2
C Q C B
1 1 1
C1 ➾ observed or estimated concentration
C2 ➾ standard of pollutants (C2 < C1)
Q1 and Q2 existing & permissible emission resply
B ➾ background concentration
R ➾ % reduction needed in emission
Q Q C C
R 1
100 2
1 2
Q 1
C B
1
Features of Rollback Models
• Control techniques based on roll back are very costly
• All emission sources are changed by same percentages
• Very useful for analysis of global scale problems.
• Has been used by EPA for control of severe problems like
oxidant formation.
MASS BALANCE MODELLING APPROACH :
KEY ELEMENTS
I) A clearly defined control volume
II) A knowledge of inputs and outputs that cross the boundary of the control
volume
III) A knowledge of the transport characteristics within the control volume and
across its boundaries
iv) A knowledge of the reaction kinetics within the control volume
MASS
TRANSPORT IIN
INPUTS
CONTROL
VOLUME (CV)
(SYSTEM BODY)
PHYSICAL, CHEMICAL,
BIOLOGICAL REACTIONS
TRANSPORT OUT
MASS
OUTFLOW
ACCUMULATION = MASS INPUTS – MASS OUTFLOWS REACTIONS
Box Model
• Application: Area sources like cities, small industry,
refuse burning.
z
• Assumption:
Region is enclosed in a box. w
Pollutants are uniformly mixed in a box
H
Pollutants emission rate is constant L
Q = q * Area = q* L* W u
Conc. Of pollutants entering the box is b gm/m3
L - length of box parallel to wind
W - width perpendicular to wind
H - height ( Mixing/ Inversion height )
K Models/ Numerical Models
Mass Conservation Eq. / Diffusion Eq
Ci C C
uj i (k jj i ) R i (C1...Ci ...Cn ) Si (x,t)
t x j x j x j
(Rate of (Wind (Turbulent (Chemical (Source
change trans- Diffusion) Reactions/ term)
of port) Fick s law Removal)
conc.)
where
i = 1,- - - n ; chemical species
j = direction
u = wind speed
k = diffusion co-efficient
Si = source term
• n nonlinear coupled partial differential eq. can include
time varying met condition, chemical reactions etc.
K Models/ Numerical Models
Types: (a) Langrangian
(Trajectory/ Moving cell model)
(b) Eulerian
(Fixed Grid/ Multibox model)
(c) Hybrid particle in cell
Particle in Cell: PIC,
Advection and Diffusion Particle in Cell: ADPIC)
Several sophisticated computer packages on K models
available (DIFKIN, REM, SAI, etc.)
Weaknesses of K model
• Analytical solution only under some assumption
• Equations can be solved numerically (problems of pseudo diffusion
truncation etc.)
• Eddy diffusivity coefficient are empirical parameters not known with
certainty
• Large input data required
• Large computational cost
USAGE
• Also irrespective of math sophistication of AQ models – the correlation
coefficient between observed and predicted value varies between 0.4 to
0.8
• Hence simple models are often preferred especially in developing
countries
Model Uncertainty
Model Uncertainty
= Natural Variability + Input Data Errors + Model Physics
(Stochastic/Inherent) (Instrument/Human) (Formulation/Assumption)
Model Physics
or Simulation
Total Uncertainty Data Errors
Natural or
Stochastic
Uncertainty
Uncertainty
Model Complexity (Number of Parameter)
Models become more realistic as more processes are simulated but reliability does
not necessarily increase
For any model there is an optimum number of parameters for which model
uncertainty is minimum
