Formulation and Assessment of Neural Network and Multiple Linear Regression Models to predict PM10 Levels in Rourkela, India

Abstract

The present study has been performed in residential educational campus located in a steel city, Rourkela with an aim to develop forecasting models using multiple linear regression analysis and neural networks of 8 h peak average values of PM10 concentrations. The concentration of air pollutants in ambient air is governed by the meteorological parameters such as atmospheric wind speed, wind direction, relative humidity, temperature etc. This study analyses the influence of temperature, wind speed, wind direction and relative humidity on ambient PM10 concentrations. 60 numbers of 8 hour average PM10 samples have been utilized to develop the forecasting models and 10 numbers of PM10 samples have been used for validation purpose. The validation of developed models revealed that neural network shows better skills in forecasting PM10 concentrations as compared to the MLR model. The MLR and neural network models could forecast 60% and 81% variance of data, respectively.