Source Apportionment and Forecasting of Aerosol in a Steel City - Case Study of Rourkela

Abstract

Urban air pollution is one of the biggest problems ascending due to rapid urbanization and industrialization. The improvement of air quality in an urban area in general, constitutes of three phases, monitoring, modeling and control measures. The present research work addresses the requirements of the urban air quality management programme (UAQMP) in Rourkela steel city. A typical UAQMP contains three aspects: monitoring of air pollution, modeling of air pollution and taking control measures. The present study aims to conduct the modeling of particulate air pollution for a steel city. Modeling of particulate matter (PM)
pollution is nothing but the application of different mathematical models in source apportionment and forecasting of PM. PM (PM10 and TSP) was collected twice a week for
two years (2011-2012) during working hours in Rourkela. The seasonal variations study of PM showed that the aerosol concentration was high during summer and low during monsoon.
A detailed chemical characterization of both PM10 and TSP was carried out to find out the concentrations of different metal ions, anions and carbon content. The Spearman rank
correlation analysis between different chemical species of PM depicted the presence of both crustal and anthropogenic origins in particulate matter. The enrichment factor analysis highlighted the presence of anthropogenic sources. Three major receptor models were used for the source apportionment of PM, namely chemical mass balance model (CMB), principal component analysis (PCA) and positive matrix factorization (PMF). In selecting source profiles for CMB, an effort has been put to select the profiles which represent the local conditions. Two of the profiles, namely soil dust and road dust, were developed in the present
study for better accuracy. All three receptor models have shown that industrial (40-45%) and combustion sources (30-35%) were major contributors to particulate pollution in Rourkela. Artificial neural networks (ANN) were used for the prediction of particulate pollution using meteorological parameters as inputs. The emphasis is to compare the performances of MLP and RBF algorithms in forecasting and provide a rigorous inter-comparison as a first step
toward operational PM forecasting models. The training, testing and validation errors of MLP networks are significantly lower than that of RBF networks. The results indicate that both MLP and RBF have shown good prediction capabilities while MLP networks were better than that of RBF networks. There is no profound bias that can be seen in the models which may also suggest that there are very few or zero external factors that may influence the dispersion and distribution of particulate matter in the study area.