Vehicular Pollution Modeling: Comparative Analysis of CALINE and ANN based pollution models

Globally, vehicular pollution is one of the greatest concerns in urban areas. Several studies on air pollution have been conducted using deterministic, statistical, and soft computing methods. However, there has been little research on how soft-computing methods like Artificial Neural Networks (ANN) can help us comprehend the non-linear and highly complex dispersion of vehicular pollution. This study compares the predictive abilities of the California-based Line Source Dispersion Model (CALINE) and the Artificial Neural Network in terms of PM 2.5 concentration. The model has been designed for one of Delhi's busiest traffic intersections, namely the Income Tax Office (ITO) intersection. In the model, PM 2.5 pollutant, background pollution, meteorological, road characteristics, and traffic flow have been taken into account as factors. The results reveal that ANN predicts vehicle pollution far more accurately than the CALINE model. The statistical comparison of CALINE and ANN models revealed r-squared values of 0.62 and 0.74, and normalized root mean square values of 0.44 and 0.35, respectively.