Analiza i modelowanie pól imisji zanieczyszczeń powietrza w dużych miastach : (na przykładzie Łodzi)

Abstract

This work outlines main research directions in air pollution monitoring programs and deals with analysis and modelling methods of air pollution concentration fields in large cities. Studies are carried out on experimental data derived from measurements of air pollution concentrations and meteorological parameters obtained for the city of Łódź. Specific local factors deciding about emissions and air pollutant concentration fields are taken into account. Evaluation of the atmosphere monitoring system in terms of a number and spatial distribution of monitoring stations was carried out by means of artificial neural networks (ANN). Competitive layer network that was trained with the use of the neural gas algorithm was applied for this task. Positive verification of atmosphere monitoring system has justified further studies on modelling air pollutant concentration fields. For a more comprehensive analysis and evaluation of air-sanitary conditions, atmospheric circulation and weather factors influencing air concentration fields of S02 and suspended particulates in the winter season were taken into account. For each of the identified atmospheric circulation type, spatial maps of mean-daily and sub-maximum daily air pollution concentrations were computed. For construction of air pollutant concentration fields an interpolation method using biharmonic Green spline functions was applied. In order to visualise differences between air pollution concentrations characteristic for each of the atmospheric circulation type area histograms and area cumulative distribution functions were calculated. A method for prediction of mean daily air pollution concentrations of S02 and suspended particulates was proposed that uses multilayer perceptron neural networks. Accurate prediction results were obtained thanks to proper selection of the training data, network topology, and gradient descent training algorithm. In particular, the technique of principal component analysis (PCA) applied to vectors of meteorological factors proved very useful in reducing data dimension at the network input. Correlation analysis of mean daily SO2 and suspended particulates concentrations was used for identifying the order of the dynamie process involved in the spread of air pollution concentration at day by day time scale. This study proposes a systematic analysis and modelling methodology of air pollution concentration fields in urban area. The key part of the system is the time series prediction neural based model of daily air pollution concentrations. This model takes into account earlier analyses and simulations. The proposed system and other employed experimental techniques are formulated in generał terms and can be used for other regions and situations in environmental studies. Successful use of the developed models requires, however, careful data preparation and analysis that takes into account local specificity of the modelled environment and factors influencing air pollution concentration fields.