Air Quality Assessment using Stochastic Simulation and Neural Networks
Since the 60’s, there has been a strong industrial development in the Sines area, on the southern Atlantic coast of Portugal, including the construction of petrochemical and energy-related industries. These industries are, nowadays, responsible for substantial emissions of SO2, NOx, particles, VOCs and part of the ozone polluting the atmosphere. The major industries are spatially concentrated in a restricted area, very close to populated areas and natural resources. Their emissions are very similar, making the identification of individual pollutant sources and of their contributions to air pollution difficult.
In this study, the regional spatial dispersion of sulphur dioxide (SO2) is characterized, through the combined use of diffusive tubes (Radiello Passive Samplers) and classical monitoring stations’ air quality data. The objective of this study is to create a regional predictive model of the contribution of different emission sources to the pollutant concentrations captured at each monitoring station.
A two-step methodology was used in this study. First, the time series of each data pair - industrial emission and monitoring station records - was screened, in order to obtain contiguous time periods with a high contribution of that specific industrial emission to the equivalent monitoring-station measurements. For this purpose, an iterative optimisation process was developed, using a variogram between industrial emissions and monitoring-station time series as the objective function. Afterwards, probability neural networks (PNN) were applied to achieve an automatic classification of the time series into two classes: a class of (emission/monitoring station) pairs of highly correlated points and a class of pairs of points without correlation
In a second step, the relationship between time series of emissions and air quality (AQ) monitoring station records - time model - is validated for the entire area for a given period of time, using for that purpose the diffusive samplers measurements. A spatial stochastic simulation is applied to generate a set of equi-probable images of the pollutant, which relationship with the different emissions is calculated using the PNN predictive model.
KeywordsMonitoring Station Pollutant Concentration Ordinary Kriging Probabilistic Neural Network Industrial Emission
Unable to display preview. Download preview PDF.