Assessment and prediction of tropospheric ozone concentration levels using artificial neural networks
Introduction
Ozone (tropospheric ozone) can have a negative impact on the environment and public health when present in the lower atmosphere in sufficient quantities. In establishing ambient air quality standards, regulations have been introduced to set limits on the emissions of pollutants in such a way that they cannot exceed prescribed maximum values (Maynard, 1984, EPA, 1999). To achieve these limits, consideration was given to mathematical and computer modelling of air pollution. Ozone, however, is unique among pollutants because it is not emitted directly into the air. It is a secondary pollutant that results from complex chemical reactions in the atmosphere. It results when the primary pollutants nitrogen oxides (NOx) and non-methane hydrocarbons (NMHC) interact under the action of sunlight. Therefore, the primary pollutants NOx and NMHC are referred to as ozone precursors. There are thousands of sources of NMHC and NOx. To track and predict ozone, one must create an understanding of not only ozone itself but also the conditions that contribute to its formation. In addition, ozone concentrations are strongly linked to meteorological conditions. Land–sea breezes also influence ozone concentrations at coastal sites. To predict ozone concentrations, it is necessary to apply a model that describes and understands the complex relationships between ozone concentrations and the many variables that cause or hinder ozone production.
From 1978 to 1997, forecasts were based on the one-hour National Ambient Air Quality Standard (NAAQS) for ozone, which was 0.12 parts per million (ppm) (EPA, 1999). In 1997, the US Environmental Protection Agency (EPA) revised the NAAQS to reflect more recent health-effects studies that suggest that respiratory damage can occur at lower ozone concentrations. Under the revised standard, regions exceed the NAAQS when the three-year average of the annual fourth highest eight-hour average ozone concentration is above 0.08 ppm. More regions will have daily maximum eight-hour ozone concentrations that exceed the level of the revised NAAQS than the old standard, and more agencies may need to model ozone (EPA, 1999). Accordingly, both deterministic and statistical models have been developed to better understand ozone production (Topcu et al., 1993).
Deterministic models (i.e. theoretical or detailed atmospheric diffusion models) are based on a fundamental mathematical description of atmospheric processes in which effects are generated by causes (Zannetti, 1983, Zannetti, 1994). Such models aim to resolve the underlying chemical and physical equations that control pollutant concentrations and therefore require detailed emission data and meteorological conditions for the region of interest. An excellent example is the urban airshed model (UAM) (Zannetti, 1983, Zannetti, 1994, Johnson, 1991). This model can be used to obtain an accurate picture of the factors involved in ozone production. However, the model is highly sophisticated because it requires a high level of human resources and intense data input (Johnson, 1991, Azzi et al., 1995). There are generally severe limitations in both spatial and temporal accuracy of the data. In addition, some input data are not easily acquired by environmental protection agencies or local industries. This means that if these inputs are unknown, then the application of the UAM is problematic. Therefore, it is much more practical to rely on statistical models.
Statistical models are based on semi-empirical statistical relations among available data and measurements. They do not necessarily reveal any relation between cause and effect. They attempt to determine the underlying relationship between sets of input data (predictors) and targets (predictands). Examples of statistical models are correlation analysis (Abdul-Wahab et al., 1996) and time series analysis (Hsu, 1992). However, the complex and sometimes non-linear relationships of multiple variables can make statistical models awkward and complicated (Comrie, 1997). Therefore, it is expected that they will under-perform when used to model the relationship between ozone and the other variables that are extremely non-linear.
Other statistical approaches frequently used include several artificial neural network implementations (Boznar et al., 1993, Ruiz-Suárez et al., 1995, Elkamel et al., 2001). The use of these artificial intelligence-based networks has been shown to give acceptable results for atmospheric pollution forecasting of pollutants such as SO2, ozone and benzopyrene. Ozone in the lower atmosphere is a complex non-linear process. Therefore, the neural network is a well-suited method for modelling this process since it allows for non-linear relationships between variables. Neural networks, by their unique structure, possess the ability to learn non-linear relationships with limited prior knowledge about the process structure. They are therefore useful for evaluating the ozone problem at a particular location. In this paper, neural network modelling was used to predict ozone concentration levels.
Section snippets
Artificial neural network concepts
Artificial neural network (ANN) models are computer programs that are designed to emulate human information processing capabilities such as knowledge processing, speech, prediction, classifications, and control. The ability of ANN systems to spontaneously learn from examples, “reason” over inexact and fuzzy data, and to provide adequate and rapid responses to new information not previously stored in memory has generated increasing acceptance for this technology in various engineering fields
Artificial neural networks and ozone modelling
Neural network models have the potential to describe highly non-linear relationships such as those controlling ozone production. Therefore, the application of artificial neural networks to ozone modelling has recently become available to capture those non-linear features of the relationship that a conventional statistical technique (e.g. regression models) might overlook (Comrie, 1997). Although they are relatively new and not yet widely used for this purpose, neural network models have proved
Area description
The state of Kuwait covers an area of approximately 17,818 km2. It is situated at the head of the Arabian Gulf between latitudes 28° and 30° north and between longitudes 46° and 48° east (Fig. 2). Iraq lies towards the northern and western boundaries of Kuwait, Saudi Arabia lies to the south, while the Arabian Gulf marks the eastern boundary. The terrain is a flat to slightly undulating desert plain. Much of the country is desert. Thus the climate is typically arid with very hot summers and
Results and discussion
A detailed analysis of ozone real-time monitoring data collected by the mobile laboratory indicated that the Khaldiya residential area was occasionally subjected to ambient ozone concentrations exceeding the NAAQS of 80 ppb. The results confirmed that high ozone events occur mainly in summer, which is in line with the results of other investigators (Salop et al., 1983, Bower et al., 1989, Poulid et al., 1991, Varshney and Aggarwal, 1992, Lorenzini et al., 1994).
On the basis of these findings,
Interpreting variable importance
Neural network modelling can also assess the importance of each of the input variables by using the network weights. With this in mind, the method proposed by Garson (1991) for partitioning the connection weights was used. The technique involves partitioning the hidden–output connection weights of each neuron into components associated with each input neuron (Goh, 1995). The results of the calculations are shown in Fig. 9 and Table 4. The results shown in Fig. 9 are displayed as columns
Conclusion
A neural network approach was used to explore the complex relationship between ozone and other variables based on ambient air monitoring measurements. The results offer an insight into the dependence of ozone concentrations on other primary pollutant concentrations and meteorological conditions.
It was found that the models' predictions and the real observations were consistent. The relative importance of the various input variables was also investigated. The results indicated the dependence of
References (30)
- et al.
Prediction of NO2 and O2 and O3 concentrations for NOx plumes photochemically reacting in urban air
Mathematical and Computational Modelling
(1995) - et al.
Description of outdoor air quality in a typical residential area in Kuwait
Environmental Pollution
(1999) - et al.
Surface ozone concentrations in the U.K. in 1987–1988
Atmospheric Environment
(1989) - et al.
A neural network-based method for short-term predictions of ambient SO2 concentrations in highly polluted industrial areas of complex terrain
Atmospheric Environment
(1993) - et al.
Measurement and prediction of ozone levels around a heavily industrialized area: a neural network approach
Advances in Environmental Research
(2001) Back-propagation neural networks for modelling complex systems
Artificial Intelligence in Engineering
(1995)Time series analysis of the interdependence among air pollutants
Atmospheric Environment
(1992)- et al.
Surface ozone in PISA (Italy): a six year study
Atmospheric Environment
(1994) - et al.
Neural networks for the prediction and forecasting of water resources variables: a review of modelling issues and applications
Environmental Modelling and Software
(2000) - et al.
Short-term ozone forecasting by artificial neural networks
Advances in Engineering Software
(1995)
Air pollution modeling in Erzurum City
Environmental Pollution
Ozone pollution in the urban atmosphere of Delhi
Atmospheric Environment
Development of statistical model for prediction of ozone levels in Shuaiba Industrial Area in Kuwait
Environmental Science and Pollution Research (ESPR)
Analysis of ozone pollution in the Shuaiba Industrial Area in Kuwait
The International Journal of Environmental Studies, Section B: Environmental Science and Technology
Cited by (202)
A review on photocatalytic asphalt pavement designed for degradation of vehicle exhausts
2023, Transportation Research Part D: Transport and EnvironmentLong time series ozone prediction in China: A novel dynamic spatiotemporal deep learning approach
2022, Building and EnvironmentChemometric approach in environmental pollution analysis: A critical review
2022, Journal of Environmental ManagementAir quality forecasting with artificial intelligence techniques: A scientometric and content analysis
2022, Environmental Modelling and SoftwareA review of Earth Artificial Intelligence
2022, Computers and GeosciencesCitation Excerpt :AI has been involved to address these challenges, particularly for predicting O3, PM2.5, and nitrogen oxides, a precursor chemical that contributes to the formation of O3 and PM2.5 (Nowack et al., 2018; Wang et al., 2003; Wu et al., 2017; Zhang et al., 2012). Earlier works often utilize neural network methods to improve air quality forecasting (Abdul-Wahab and Al-Alawi, 2002; Kolehmainen et al., 2001; Ruiz-Suarez et al., 1995). Recently, more advanced ML algorithms are used to enhance O3 and NO2 prediction and SVM is better than NN in predicting daily maximum O3 concentrations (Chelani, 2010).