Hybrid system for urban air pollution management: artificially intelligent and traditional modelling

Two approaches for the calculation of air concentrations in
urban areas - traditional (gaussian model) and a new
technique using artificial intelligence methods. They will
be combined in a hybrid system for air quality prediction.

The emission of pollutants in urban environments is mainly due to traffic and domestic heating in addition to the eventuality of industrial sources in areas not far from the city. Thus, pollutants are emitted at ground level, where severe pollution levels might be found, as well as at elevated sources. However, the transformation of primary into secondary pollutants requires an amount of time which is dependent upon reaction rates, thus secondary pollution is much more diffused and uniformly distributed. In other words, citizens are usually exposed to peak levels of primary pollutants whereas they are exposed continously to secondary pollutants either from ground level or elevated sources. Very few urban areas in the world have reached the stage of applying numerical prediction models to consider how this will change in the future (short-term and long-term) or how the effectiveness of emission-reduction measures and strategies can be tested. The ideal model is one which would accurately predict the air quality at any location at any time for any set of emissions and meteorology conditions. It would provide both short and long-term average predictions and show the impact of each emission source on air quality at each location. The real diffusion models now available fall far short of that goal. The present work will use two approaches for the prediction of air concentrations in urban areas: * a traditional one (Gaussian) and * new techniques using artificial intelligence methods. Since the philosophy behind these two approaches is fundamentally different, the objective is not to identify which one will overpower the other but try to find a way of seeing how they can complement each other so that a hybrid system for air quality prediction can be developed. The traditional approach uses the well known Gaussian model, but takes advantage of the applications already developed for Macau and the Lisbon urban airshed. It is a simple model, well adapted for prognostic simulations using small computers, and can be linked to city's automatic pollution monitors (including those from industries), utilising real-time meteorological data and ambient pollution levels in combination with an emissions inventory and a digitized topographic map as inputs. Even the Gaussian plume approach described above is popular in industrial and urban applications, sound prediction of air concentration with the traditional model requires accurate knowledge of the pollutant mass emission rate, related meteorological factors and the prevailing atmospheric turbulence in the boundary layer, as well as surface roughness; the technique also depends on the distance scales involved in the transport from the source to the receptor or target area. Then, the difficulties in obtaining accurate input data for the traditional model call for an alternate approach. One of the main objectives in air quality modelling is to predict the values of future air concentration based on values of a number of input variables such as the past air concentration record, related meteorology factors and/or the pollutant emission rate. This is basically a regression problem with the goal of finding a suitable function which maps the air concentration to related factors. Artificial Neural Networks (ANN) are currently in vogue for these kinds of modelling applications due to their computation efficiency. In this ANN study, two models will be developed for comparison and selection: (i) To predict the air concentration at a particular location based on the knowledge of only its past history. Since the monitored air concentration data is a time series, the past values may influence future values depending on the presence of underlying deterministic forces; these forces could be the changing meteorological factors which depend on the geographical and topographical features of an area, and the pollutant emission rate. The relationship of these factors to air concentration is highly non-linear and known to be difficult to model directly; nevertheless, these factors may be characterized by trends, cycles and nonstationary behaviour in the air concentration time series. A predictive model which could recognize the aforesaid embedded recurring patterns and non-linear relationships will provide accurate predictions. ANN are models that can be trained to construct non-linear relationships between past and future values of a time series, and thereby extract hidden structures and relationships governing the data and so to give accurate prediction (this only requires the concerned pollutant concentration history so that the cost for initial set-up is low). (ii) To predict the air concentration at a particular location using the nearby meteorological factors (wind speed and wind direction) and pollutant emission rate as independent variables. Since the air concentration is known to depend on the meteorological factors and the pollutant emission rate, this model may reveal more physical interrelationships between these key parameters and so to give not just accurate prediction but also shed some light on the actual physical process as well. Nevertheless, the principles for this ANN model are similar to that for the time series prediction model described in section (i). The outcome of the project will be an hybride system for urban air pollution management, where the models are used in function of the available input variables, providing the authorities and industrials with information about current and predicted air quality throughout the area and so enabling an assessment to be undertaken as to whether pollution levels exceed health-based standards or are likely to exceed them in the future. Keywords: urban air quality, neural networks, modelling.
Project ID: 
1 920
Start date: 
Project Duration: 
Project costs: 
500 000.00€
Technological Area: 
Market Area: 

Raising the productivity and competitiveness of European businesses through technology. Boosting national economies on the international market, and strengthening the basis for sustainable prosperity and employment.