Journal of Scientific & Industrial Research

Vol. 65, April 2006, pp. 321-324

Sensitivity study of Gaussian dispersion models

G Sriram1, N Krishna Mohan2 and V Gopalasamy3,*
1
SCSVMV, Deemed University, Enathur, Kanchipuram 631561
2
Department of Mechanical Engineering, 3Department of Civil Engineering, Annamalai University,
Annamalainagar 608 002
Received 05 May 2005; accepted 25 January 2006

Over the last three decades, strict environmental regulations and the availability of personal computers have fueled an
immense use of mathematical models to predict the dispersion of air pollution plumes. This paper discusses how the
propagation of seemingly small errors in the Gaussian model parameters can cause very large variations in the model’s
predictions.
Keywords: Gaussian Plume Model, Over prediction ratio, Sensitivity study
IPC Code: G05B19/04

Introduction Sensitivity Study

Air dispersion modeling has been evolving since The Gaussian models assume an ideal steady state
the 1930s. The present practice for ambient air quality of constant meteorological conditions over long
predictions is through application of Gaussian Plume distances, idealized plume geometry, uniform flat
Model (GPM) and its available variations. The terrain, complete conservation of mass, and exact
application of GPM requires knowledge of emission Gaussian distribution. Such ideal conditions rarely
release rate, atmospheric turbulence, wind speed, occur. A sensitivity study was performed by assuming
dispersion coefficient, effective stack height, mixing reasonable degrees of error in some of the key
height etc. The values of these parameters are often variables used in the Gaussian models and
adopted from other countries without understanding determining the proposed end-result of those errors on
their applicability in Indian context. It has also been the calculated, ground-level pollutant concentrations.
observed that various forms of GPM are used without Several comparative models (Table 2; Figs 1&2) were
providing any reasonable justification in doing so. defined as follows:

Study Area i) Base Model A

The Neyveli Lignite Corporation, a public-sector In the base model A, Gaussian plume equation was
enterprise established in 1956, is situated at Neyveli used to predict ground level pollutant concentration.
in Tamil Nadu. An epitome of Indo-Soviet The concentration (C) of the gas at X, Y, Z from a
collaboration, Neyveli thermal power station-I (TPS-I) continuous source with an effective stack height given
was commissioned with one unit of 50MW in May by GPM is,
1962. Presently, this power station has 6 units of 50
MW each and 3 units of 100 MW each. Thermal C (x,y,z) =
power station-II (TPS-II) has been a major source of
Q  1  y 2   1  Z − he 2 
power to all southern states of India. The 1470 MW exp −    exp −   
capacity power station consists of 7 units of 210 MW 2∏σ yσ zu  2  σ y     2  σ z  
each (Table 1).
 1  Z + he 2  
_________ + exp −    
*Author for correspondence
 2  σ z   
E-mail: [email protected]
322 J SCI IND RES VOL 65 APRIL 2006

Table 1  Details of stacks in TPS-I and TPS-II

Details of stacks TPS-I TPS-II
Stacks 1, 2 Stack 3 Stack 4 Stacks 1, 2, 3 Stacks 4-7

Power production, MW 3×50×2 No. 2×50×1 No. 4×50×1 No. 1×210×3 No. 1×210×4 No.
Number of stacks 2 1 1 3 4
Height of stacks, m 60 each 60 120 170 each 220 each
Stack Diameter at exit, m 5.10 each 5.10 5.10 4.85 each 5.47 each
Exit gas temperature, ºC 158 158 158 150 150
Velocity of flow, m/s 20.2 13.5 27.2 15.33 12.06
Table 2  Calculated ground-level concentrations ((µg/m2)) at downwind distances ranging from 1-10 km (using power law exponent
p=1/7)
Receptor A B C D Over Prediction
downwind Base Model Adjusted Model -1 Adjusted Model -2 Adjusted Model 2 plus ratio
distance, m wind shift A/D
1000 1.65 5.14 0.018 0.004 412
2000 26.82 35.46 4.29 2.77 9.6
3000 43.67 47.59 14.32 11.46 3.75
4000 51.66 52.16 22.76 19.94 2.59
5000 55.47 53.52 28.41 26.05 2.12
6000 56.87 53.14 32.19 30.23 1.88
7000 56.82 51.81 34.84 33.2 1.71
8000 55.91 50.02 36.79 35.38 1.58
9000 54.42 48.03 38.23 37.04 1.46
10000 52.87 45.49 39.27 38.3 1.38

By setting Z = 0 in above equation for ground

level concentration is obtained as follows:

Q  1  y 2   1  he 2 
C (x,y,z) = exp −    exp −   
∏σ yσ zu  2  σ y    2  σ z  

Fig. 1  Calculated ground-level concentrations (µg/m2) at where, Q is the pollutant emission rate (µg/sec), C is
downwind distances ranging from 1-10 km (using power law the pollutant concentration (µg/sec),
y,
z are
exponent p = 1/7). A, B, C and D are four model types dispersion coefficients (m), ū is mean wind speed at
stack height (m/sec), y, z are distance from the plume
axis in the cross wind directions (m), and he is the
effective stack height (m).
A number of important parameters required for
using the above equation are:

Stability Classification
There are number of stability class definitions, the
best known of which is from Pasquill. Six categories
Fig. 2  Calculated ground-level concentrations (µg/m2) at (A-F), based on the use of synoptic meteorological
downwind distances ranging from 1-10 km (using power law data, are distinguished. In this work, Pasquill stability
exponent p = 0.1981). A, B, C and D are four model types is adopted.
322
 SRIRAM et al: SENSITIVITY STUDY OF GAUSSIAN DISPERSION MODELS 323

Table 3  Calculated ground-level concentrations ((µg/m2)) at downwind distances ranging from 1-10 km (using power law exponent
p=0.1981)
Receptor A B C D Over Prediction
downwind Base Model Adjusted Model -1 Adjusted Model -2 Adjusted Model 2 ratio
distance, m Plus Wind shift A/D
1000 3.97 3.21 0.085 0.024 165
2000 36.81 29.79 7.63 5.31 6.93
3000 54.31 44.7 20.44 17.09 3.17
4000 61.54 51.49 29.6 26.63 2.31
5000 63.73 53.96 39.45 33.01 1.93
6000 63.32 54.21 39.34 37.33 1.69
7000 61.6 52.91 41.98 40.36 1.52
8000 59.28 51.15 43.36 42.49 1.39
9000 56.74 49.12 44.89 43.97 1.29
10000 54.16 47.02 45.53 44.94 1.20

Wind Speed at Stack Height ii) Adjusted Model B

Wind velocity at the desired level is calculated The wind velocity is an important parameter in
using the power law expression1,2 as: determining the vertical and cross wind Gaussian
dispersion of stack gas components from continuous,
u Z 
p point-source plumes. The wind velocity used in the
  =  Gaussian dispersion equation (GDE) should be the
 u1   Z1  velocity, which prevails throughout the vertical, and
crosswind spread of the plume. This creates a
where Z1, is usually taken as the height of dilemma in that: a) Deviation of the generalized GDE
measurement, approx 10m, and Z is the height at assumes a constant wind velocity throughout the
which a wind speed estimate is desired. The average plume; and b) Wind velocity is not constant
value of power law exponent p has been determined throughout the plume since it typically varies with
by many measurements around the world to be about altitude4. Thus as a minimum requirement, wind
one seventh. This has led to common reference to velocity used in the GDE should be determined at the
above equation as 1/7 power law equations3. emission height or so called effective stack height5.
So, using power law conversion, the wind velocity at
Dispersion Coefficients and Plume Rise effective height was obtained and the same was used
The Brigg’s interpolation formulae are used to find in the Gaussian equation for predicting the ground
out the dispersion coefficients as suggested in the level concentration6.
guidelines published by Central Pollution Control
Board. Similarly, Brigg’s plume rise equation was iii) Adjusted Model C
used to predict buoyant plume rise as: Most Gaussian models use modifications of
dispersion coefficient derived experimentally by
 2.47Qh1/3 H S 2/3  Paquill. These coefficients are derived based on the
∆h =   assumption that the wind speed and atmospheric
 u  stability class were known exactly. But in actual case,
atmospheric stability class and wind speed are not
where, Qh is the stack gas heat flux (kcal/sec), Hs is known exactly at plume height and they are not
the stack height (m), and Ū is the wind speed at stack constant for the entire distance from the source stack
height (m /sec). to downwind receptor7. Hence in this model, the
323
324 J SCI IND RES VOL 65 APRIL 2006

ground level concentration arrived same as in dispersing plume concentrations within a factor that
model B except that the plume rise increased (20%) may be as high as 10. The model errors, smaller in the
and Pasquill vertical dispersion coefficients decreased far field (>2 km), are due to uncertainty of the
(25%). atmospheric stability class, wind speed, and they are
not constant for the entire distance from the source
iv) Adjusted Model D
stack to the down wind receptors.
Deriving the Gaussian dispersion models assume
the wind speed and direction are constant from source
point to receptors. Such ideal conditions rarely occur Conclusions
(for a wind speed of 2 m/sec and a distance of 10 km, Gaussian model over prediction ratio arrived in this
80 min of constant conditions would be needed). paper can be improved by observing hourly wind
Slight error in the estimation of wind direction can data, which could be used to estimate wind power law
result in tremendous errors of concentration where the coefficients. Similarly, all meteorological parameters
problem is to estimate the concentration at specified are collected only at Central Applied Research and
locations. In this model, the ground level Development (CARD), which is located in the
concentrations were calculated same as in model C premises of Neyveli township at a down wind
except that an assumed winds direction shifts 25 distance of 4.5 km and wind direction sectors of
degrees. southwest with reference to TPS-II. It is suggested to
install some more meteorological stations at Neyveli
Results and Discussion to improve data set, so that both metrological and air
The inputs for the dispersion model consist of pollutants are forecast more accurately.
emission information, meteorological data and
receptor information. These inputs are entries to the
dispersion model, which is a mathematical simulation References
1 Chinnasamy T V & Haridasan T M, Wind energy potential at
of the chemistry and physics of the atmosphere. The Palkalainagar, Renew Ener, 1 (1991) 815-821.
outputs from the model are hourly concentrations at 2 EI- Shobokshy M S & EI- Zayat R E, Site assessment and
each receptors applying Microsoft excel. Minor feasibility analysis for small wind energy conversion
changes in some of the key variables (Table 2) can systems, Int J Amb Ener, 12 (1991) 79-94.
3 Peterson E W, On the use of power laws for estimates of
result in the propagated over-prediction factor (1.38- wind power potential, J App Meteorol, 17 (1978) 390-394.
412). Also, the above steps are performed taking 4 Prakasa Rao G S, Jawal A K & Kumar M S, Effect of
power law exponent p=0.1981 based on 12 years of urbanization on meteorological parameters, Mausam, 55
wind data pertaining to the study area (Table 3). The (2004) 429-440.
over-prediction factor of the Base Model A varies 5 Bruce Turner D, Work Book of Atmospheric Dispersion
Estimates (US EPA Publication AP-26) 1970.
from 1 to 165. 6 Justus C G, Hargraves W H R & Ali Y, Nationwide
It is unrealistic to expect the Gaussian models to assessment of potential output from wind-powered
predict real-world dispersing plume concentrations generators, J Appl Meteol, 15 (1976) 673-678.
consistently by a factor of 2 or 3. It is probably more 7 Milton R Beychok, Fundamentals of Stack Gas Dispersion
(New port Beach, CA 92660) 1995.
realistic to expect consistent predictions of real-world

324