IAPRS & SIS, Vol.34, Part 7, "Resource and Environmental Monitoring", Hyderabad, India,2002 
ADAPTIVE MODEL FOR AIR POLLUTION PREDICTION IN A COASTAL REGION 
N. Manju, A.Rajaraman*, R. Balakrishnan** 
Dept. of Physics, Meenakshi College for Women, Chennai-24. (raghavreghu Q hotmail.com) 
*Dept. of Civil Engineering, Indian Institute of Technology, Chennai-32. 
**Dept. of Physics, Madras Christian College, Tambaram, Chennai-59. 
KEY WORDS: Pollutant concentration, dispersion, modeling, adaptive neural network, meteorology, sulphur di oxide. 
ABSTRACT: 
Air pollution in a coastal region is one of the major concerns due to its impact on public health and living conditions. This paper 
focuses on the prediction of concentration of air borne pollutants using an adaptive neural network model. The meteorological 
parameters that have a profound influence on pollutant transport are identified. The diurnal and daily fluctuations of wind speed, 
wind direction and temperature, which are the critical parameters in pollutant dispersion are analyzed and their relationships with the 
corresponding pollutant levels are studied. Based on this a knowledge based neural network is developed to forecast the 
concentration of the pollutants in an industrial belt located in the coastal region of Chennai. The prediction of Sulphur di-oxide in 
ambient air is made on an hourly, daily and monthly basis. The prediction results indicate that nearly 70% of the values are within 
an average error of +10%. 
INTRODUCTION 
Urban air pollution has emerged itself significantly into a 
problem of global concern owing to its negative impact on 
health and environment. This has forced research in recent 
years to be focused on better means to protect the atmosphere 
and to ensure sustainable development. The first step in the 
process involves the estimation of the pollutant concentration in 
ambient air using suitable monitoring techniques to determine 
the threshold of pollutant level that is considered safe for 
healthy living. This paper attempts to develop a model based 
on real time data of measured pollutant concentration and 
meteorological parameters. 
À plethora of dispersion models (Touma,1995) are available in 
the literature which describe the atmospheric diffusion 
phenomena. These can be categorized into two viz., physical 
and non-physical models. The former describes the dynamics 
of the system using partial differential equations which governs 
the diffusion characteristics of the pollutant. These include 
Lagrangian, Eulerian, Gaussian formulations and 
photochemical models (Seinfeld, 1995). The non-physical 
model use observed data and they are developed based on the 
analysis of those data. 
Predictions made by physical models are less precise as it is 
difficult to obtain sufficient reliable data as demanded by model 
inputs. Meteorological parameters play a dominant role in 
pollutant transport. The diurnal and seasonal variations of 
atmospheric pressure, wind vector, solar insolation, ambient 
temperature and humidity have a drastic effect on ground level 
concentration. In addition source emissions, terrain 
characteristics and building orography, introduce uncertainties. 
Hence the description of the dispersion phenomena using hydro 
-dynamical equations and obtaining a proper solution becomes 
very complex and ambiguous owing to the aforesaid changes, 
which are highly interlinked. All these factors render it 
difficult to conceive an ideal physical model, which will 
incorporate the varied phenomena involved in dispersion 
mechanism to make an accurate prediction possible. Therefore, 
it is preferable to use a non-physical model that is directly 
related to real time observed data rather than to use a 
complicated physical model (Kolehmainen, 2001). It is in this 
context that neural network approach assumes significance as it 
can effectively handle the non-linearities present in the system. 
In this paper a knowledge-based method is used to study the 
impact of meteorological factors on the concentration of the 
pollutant. 
STUDY AREA AND DATA 
The study area is an industrial belt in coastal Chennai 
constituting a large number of industries such as petro- 
chemical, fertilizer units, refineries and thermal power plants. 
Most of these are continuous processing units whose emissions 
have been constantly deteriorating the air quality of the region. 
On site hourly measurements of meteorological parameters 
corresponding to wind speed, wind direction and ambient 
temperature were used for the month of January for the years 
1999 and 2000. The air quality data for SO, measured for this 
industrial zone was procured for the same period from the 
monitoring stations. The meteorological data were also 
compared with the observed values of the Regional 
Meteorological Centre, Chennai. 
DATA ANALYSIS 
The pollutant levels exhibit daily and seasonal fluctuations as 
they get triggered by various dispersal mechanisms such as 
convection, inversion etc. that occur in the atmosphere.