Correlation of Air Pollution and Meteorological Data Using Neural Networks

Linear regression methods have been applied for decades and are well known and understood (Millionis, A.E. and T.D. Davies, 1994; Robeson, S.M. and D.G. Steyn, 1990; Ryan, W.F. 1995; Shi, J. P. and R.M. Harrison, 1997). However, there are numerous environmental processes that exhibit significant non-linear behaviour. Advances in the field of Artificial Neural Networks (ANN) in the late 1980s popularised non-linear regression techniques like Multi-layer Perceptons (MLP) and self-organising maps (SOM). It is shown that Neural Networks (NN) can be trained to successfully approximate virtually any smooth, measurable function (Hornik, K., M. Stinchcombe and H.White, 1989). NN are highly adaptive to non-parametric data distributions and, whilst other statistical methodologies require a set of assumptions to be fulfilled, the former make no prior hypotheses about the relationships between the variables. NN are also less sensitive to error term assumptions and they can tolerate noise, chaotic components and heavy tails better than most of the others methods. Other advantages include greater fault tolerance, robustness, and adaptability especially compared to expert systems, due to the large number of interconnected processing elements that can be trained to learn new patterns (Lippman, R.P., 1987). These features provide NN the potential to model complex non-linear phenomenon like air pollution (Kolhmainen, M., H. Martikainen and J. Ruuskanen, 2001; Perez, P. and A. Trier, 2001; Chelani, A.B., D.G. Gajghate and M.Z. Hasan, 2002).