Artificial neural network model for ozone concentration estimation and Monte Carlo analysis. (July 2018)

• Record Type: Journal Article
• Authors: Gao, Meng; Yin, Liting; Ning, Jicai
• Is Part Of: Atmospheric environment. Volume 184(2018)
• Page Start: 129
• Page End: 139
• Publication Date: 2018-07
• Subjects: Air pollution -- Artificial neural network -- Monte Carlo simulation -- Uncertainty analysis -- Sensitivity analysis; Air -- Pollution -- Periodicals; Air -- Pollution -- Meteorological aspects -- Periodicals; 551.51
• DOI: 10.1016/j.atmosenv.2018.03.027
• Languages: English
• ISSNs: 1352-2310

Abstract

Abstract: Air pollution in urban atmosphere directly affects public-health; therefore, it is very essential to predict air pollutant concentrations. Air quality is a complex function of emissions, meteorology and topography, and artificial neural networks (ANNs) provide a sound framework for relating these variables. In this study, we investigated the feasibility of using ANN model with meteorological parameters as input variables to predict ozone concentration in the urban area of Jinan, a metropolis in Northern China. We firstly found that the architecture of network of neurons had little effect on the predicting capability of ANN model. A parsimonious ANN model with 6 routinely monitored meteorological parameters and one temporal covariate (the category of day, i.e. working day, legal holiday and regular weekend) as input variables was identified, where the 7 input variables were selected following the forward selection procedure. Compared with the benchmarking ANN model with 9 meteorological and photochemical parameters as input variables, the predicting capability of the parsimonious ANN model was acceptable. Its predicting capability was also verified in term of warming success ratio during the pollution episodes. Finally, uncertainty and sensitivity analysis were also performed based on Monte Carlo simulations (MCS). It was concluded that the ANN could properly predict the ambient ozone level. Maximum temperature, atmospheric pressure, sunshine duration and maximum wind speed were identified as the predominate input variables significantly influencing the prediction of ambient ozone concentrations. Graphical abstract: Image 1 Highlights: We develop a parsimonious ANN model for predicting O3 concentration in urban area. Six meteorological parameters and one temporal covariate are used as predictors. Monte Carlo simulation technique is used for model development and analysis. The relative importance of all inputs in influencing O3 level are identified.