Joint Space–Time Geostatistical Model for Air Quality Surveillance/Monitoring System

Abstract

Air quality is usually driven by a complex combination of factors where meteorology, physical obstacles and interaction between pollutants play significant roles. The use of models that are able to characterize space–time dispersion of pollutants at fine scales in urban areas (e.g. stochastic and neural networks models) is becoming a common practice. The main objective of this work is to produce an integrated air quality model designed to monitor Lisbon’s metropolitan area. This model, which allows forecasting critical concentration episodes of a certain pollutant by means of a hybrid approach, is based on the combined use of neural network models and stochastic simulations. A stochastic simulation of the spatial component with a space–time trend model is proposed to characterize critical situations at a given present period or for a very near future period, taking into account data from the past and a space–time trend from the recent past. To identify critical episodes in the near future period t + 1, predicted values from neural networks are used at each monitoring station. The neural network model was developed taking into account historical data of pollutants’ concentrations and meteorological conditions measured and also predicted for each monitoring station. First, a joint space–time model is used to build the trend model based on historical data (period ≤ t). Afterwards, stochastic simulation is performed to predict the period t + 1 at any location x, allowing for the local conditional distribution functions characterization and spatial uncertainty assessment. As this approach is performed sequentially in the time domain, the space–time trend is sequentially updated for every new period t + i, i = 1…, N. This spatial-temporal model has been developed and applied to the urban area of Lisbon. An application to the prediction of mean daily NO₂ concentration is presented in this paper.