IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India,2002 
  
  
70 
  
65 
351 
50 4 
values with error « 1096 
45 4 
  
  
  
inis 
no: of observations 
408 
Figure 7. Variation of percent values «1096 mean error 
with number of observations 
The number of values which have shown an error of less than 
10% has decreased to 52% for daily values compared to 66% 
for hourly values. Similarly in figure 6, the monthly measured 
and predicted values are presented. The number of values with 
average error less than 10% has fallen to 51%. This indicates 
that the adaptive neural network would yield better results if the 
data used for training in the first phase were large as depicted in 
fig. 7. 
The trained and adaptive network is now used to forecast the 
SO, concentration for the month of January 2000. The results 
show that using four hourly values of inputs, the model 
predictions show 78% of values with average error less than 
10% while with single monthly values 72% indicated less than 
10% average error. 
CONCLUSION 
The adaptive neural network has predicted pollutant 
concentrations fairly better than other conventional models 
used for air quality predictions using a limited number of 
meteorological parameters. The present model can be made 
more adaptive by providing sufficient knowledge to predict 
values for other seasons and topographical conditions. The 
predictions could be further improved if good quality data and 
inputs pertaining to stability and vertical structure of the 
atmosphere are included as these factors greatly influence the 
convection and inversion phenomena involved in atmospheric 
dispersion. 
REFERENCES 
Beale, R., Jackson, T., 1998, Neural Computing: an 
introduction. IOP Publishing Ltd., London. 
Boznar, M., Lesjak., Mlakar, P., 1993, A neural network based 
method for short-term predictions of SO, concentrations in 
highly polluted industrial areas of complex terrain. 
Atmospheric Environment, 27B (2), pp. 221 — 230. 
John H. Seinfeld., 1986, Atmospheric chemistry and physics of 
air pollution. John Wiley & Sons, New York, pp. 526 — 543. 
Kolehmainen, M., Martikainen, H., 2001, Neural networks and 
periodic components used in air quality forecasting. 
Atmospheric Environment, 35 (5), pp. 815 — 825. 
Touma, S., Irwin, S., 1995, A review of procedures for 
updating air quality modeling techniques for regulatory 
programs. 
Journal of Applied Meteorology, 34, pp. 731 — 737. 
White, D.A., Sofge, D.A., 1992, Handbook of Intelligent 
control. Multiscience Press