The evaluation on artificial neural networks (ANN) and multiple linear regressions (MLR) models for predicting SO2 concentration
Introduction
Air pollution has a devastating impact on human health and the environment (Barbulescu and Barbes, 2017; Beigzadeh et al., 2019). The World Health Organization (WHO) reported that during the period from 2008 to 2013, urban air pollution has increased globally by 8%. (Emami et al., 2018). Hence, more attention should be given to improving and monitoring the air quality (Barbulescu and Barbes, 2017; Meraz et al., 2015). The previous studies revealed that the total number of deaths attributed to air pollution in Iran increased by approximately 27% during the years 1990–2013. Tehran as the capital of Iran is one of the cities facing serious environmental problems and health issues due to air pollution. Air pollution in Tehran is mainly due to developmental changes in the last few decades including overpopulation, changes in the industry and transportation sectors such as inappropriate patterns of industrial development, improper location of facilitates, development of new industries plants, vehicle malfunction, expansion of intensifying vehicle traffic, and widespread use of fossil fuels. In recent years, Tehran has experienced severe air pollution episodes during the winter month due to strong temperature inversions and accumulation of pollutants over the city that leads to school closures and a high rate of hospital admissions owing to respiratory problems (Torbatian et al., 2020; Janjani et al., 2020; Jamshidi Angas et al., 2020). One of the conventional pollutants is sulfur dioxide (SO2) that is a toxic and highly reactive gas with a pungent, irritating, and rotten odor, which has linked to irritates the eyes and the respiratory system, bronchoconstriction, cardiovascular disease, cancer, and ecological effects on soil, forests and freshwater (Barbulescu and Barbes, 2017; Masoudi et al., 2018). The US Environmental Protection Agency (EPA) declared that short-term exposures to SO2 can lead to an increase in respiratory morbidity (Johns and Linn, 2011). SO2 gas is produced mainly from both natural and anthropogenic sources such as volcanic eruptions, sulfur fossil fuels combustion (coal, oil), and the exhaust of vehicles' engines (Barbulescu and Barbes, 2017). It can persist in the atmosphere from 5 min to 24 h and can be moved great distances (Barbulescu and Barbes, 2017; Masoudi et al., 2018). SO2 is one of the six air pollutants that EPA has designated as criteria pollutants under the US Clean Air Act. The National Ambient Air Quality Standards (NAAQS) for SO2 represent 0.5 ppm as a 3-h level, 0.14 ppm as a 24-h level, and 0.03 ppm as annual arithmetic mean (van Thriel et al., 2010). Therefore, it is important to monitor and predict air pollution. Air pollution prediction using machine learning techniques such as artificial neural networks (ANNs) is one the most popular tools for air pollution predicting and have been preferred linear statistical models owing to their ability to model strongly non-linear dependencies, and their enhanced performances (Shams et al., 2020; Raturi and Prasad, 2018; Masih, 2019). Different categories of ANNs can be used such as Multi-layer perceptron (MLP) for studies on predicting air quality due to their ability and high efficiency in the modeling of nonlinear mechanisms and complex problems (Kalantary et al., 2019; Kalantary et al., 2020; Mosaffaei et al., 2020). The multiple linear regression (MLR) method is a common technique that can be used as the prediction tool in multidiscipline and based on various predictors (Abdullah et al., 2019; Yorifuji et al., 2019; Jahani et al., 2020; Jahani et al., 2019). Hassanzadeh et al. (2009) used statistical models and time series analysis for forecasting SO2 levels in Tehran. The results proved that an autoregressive/moving average (ARMA) model can provide reliable, satisfactory predictions for time series and SO2 emission is considerably high for stationary sources compared to other pollutants (Hassanzadeh et al., 2009). Ebrahimi and Qaderi (2021) evaluated a combination of fuzzy systems and neural networks to select the most suitable scenario for controlling SO2 pollution. In this study, eight variables including wind speed, precipitation, temperature, pressure, humidity, gas oil consumption, gasoline consumption, and urban green space levels have been used as input data. The results showed that the parameters of pressure, urban green space, gasoline consumption, gas oil consumption, temperature, wind speed. and humidity, respectively, had the greatest effect on reducing the SO2 concentration (Ebrahimi and Qaderi, 2021). Shamsoddini et al. (2017) examine the performance of the Random Forest feature selection in combination with multiple-linear regression and MLP methods, to achieve an efficient model to estimate SO2, CO, NO2, PM2.5 levels in the air of Tehran. The estimation accuracy of SO2 emissions was lower than the other air contaminants (R2 = 0.59) (Shamsoddini et al., 2017).
Nature- based solutions such as the presence of trees and urban green spaces in urban areas can be used to a reduction of air pollution and be more cost-effective in the long run than other solutions. This finding is confirmed by previous studies in which green spaces can reduce air pollution and improve air quality owing to trees' ability to remove and absorb pollutants from the atmosphere. The reason for this reduction can be occurred directly by deposition on the tree surface and/or by stomatal uptake of gases such as SO2 (Vieira et al., 2018; Ebrahimi and Qaderi, 2021). According to the literature review, meteorological data, urban traffic data and time parameters has been related to air quality and SO2 concentration (Ebrahimi and Qaderi, 2021; Janjani et al., 2020; Wu et al., 2020). Though these contributions to the urban air quality problems are known, but more information and extensive research are needed to better manage and enhance air purification in urban areas. The aim of this study is the prediction of the concentration levels of SO2 as a factor determining the atmospheric pollution in Tehran city. In this study compared to other similar studies, many environmental variables including urban traffic parameters, urban green space information, time parameters, and meteorological data were used to support model outputs, and finally, we identified the most accurate model for the prediction of the concentration levels of SO2 in Tehran. The result of model sensitivity analysis was illustrated to prioritize model variables.
Section snippets
The study area
Tehran metropolis lies on the south of the Alborz mountains and the northern margin of Iran's central desert, (35°35′–35°50′ N, 51°05′–51°35′ E) at 1280 m above the sea level. Its capital of Iran has a total area of 730 km2 and a population of over 8 million (Hassanpour Matikolaei et al., 2019). The location of Tehran is presented in Fig. 1.
Data collection
For studying the extent of the SO2 pollution, five data groups including the concentration of SO2 as a dependent parameter, urban traffic parameters, urban
MLR analysis
To predict the SO2 concentration in Tehran city, MLR analysis was used. The best result for constants and statistical analysis stepwise- multi linear regression are given in Table 1.
There was a relation between urban traffic parameters, urban green space information, and meteorological data, and the prediction of SO2 values in Tehran. According to the finding of the stepwise-multi linear regression technique, the rainfall, the air temperature, the humidity, the length of sunshine per day, the
Discussion
Air pollution is not a new subject around the world, and atmospheric pollution has a profound impact upon people's work and life. There is a relation between air quality and human health and the environment (Zhu et al., 2019; Yusof et al., 2019; Barati et al., 2017). In this research, MLR and ANN analysis were used to predict SO2 concentrations in the Tehran metropolis. The relation between predicted SO2 and the actual levels of SO2 were described with R2 and RSME values as in the MLR and MLP
Conclusion
This research aimed to compare regression and MLP models in order to develop an accurate, efficient, and easy technique for predicting air pollutants such as SO2. The result indicated that the MLP technique performed much better than the regression model. As well as, the air quality can be improved after suitable decisions for controlling the effective parameters which are the one-day time delay, park indicator, the season of the year, and the total area of parks. Furthermore, the results of
Funding
Not applicable.
Author statement
A.J. designed the experiments, analyzed the data and S.K. wrote the paper. R.S. performed the experimental measurement. A.J., R.S. conceived the mathematical model and ANN Modeling Techniques A.J., M.M., and N.K. conceived the idea and designed the experiment. A.J., M.M., and N.K. supervised research. All authors discussed the results and commented on the manuscript.
Declaration of Competing Interest
The authors have declared no conflict of interest.
Acknowledgments
We would also like to thank the Tehran Air Quality Control Company, the Tehran City Transportation and Traffic Organization, and the Meteorological Organization for their cooperation in providing the information required for this research. May this smallest appreciation to some of their efforts.
References (52)
- et al.
Mathematical modeling of sulfur dioxide concentration in the western part of Romania
J. Environ. Manag.
(2017) - et al.
Occupational exposure to wood dust and risk of nasopharyngeal cancer: a systematic review and meta-analysis
Environ. Res.
(2019) - et al.
Hybrid artificial neural network models for effective prediction and mitigation of urban roadside NO2 pollution
Energy Procedia
(2017) - et al.
Determination of the most effective control methods of SO2 pollution in Tehran based on adaptive neuro-fuzzy inference system
Chemosphere.
(2021) - et al.
Air pollution at Rochester, NY: long-term trends and multivariate analysis of upwind SO2 source impacts
Sci. Total Environ.
(2018) - et al.
Characterizing the effect of traffic density on ambient CO, NO2, and PM2. 5 in Tehran, Iran: an hourly land-use regression model
Transp. Lett.
(2019) - et al.
Aesthetic preference and mental restoration prediction in urban parks: an application of environmental modeling approach
Urban For. Urban Green.
(2020) - et al.
An application of artificial intelligence techniques in prediction of birds soundscape impact on tourists’ mental restoration in natural urban areas
Urban For. Urban Green.
(2021) Review on urban vegetation and particle air pollution–deposition and dispersion
Atmos. Environ.
(2015)- et al.
Characterizing multiple air pollutant indices based on their effects on the mortality in Tehran, Iran during 2012–2017
Sustain. Cities Soc.
(2020)
Statistical persistence of air pollutants (O3, SO2, NO2 and PM10) in Mexico City
Physica A.
Air pollution removal by trees in public green spaces in Strasbourg city, France
Urban For. Urban Green.
Air pollution trends in Tehran and their anthropogenic drivers
Atmos. Pollut. Res.
Sensory and pulmonary effects of acute exposure to sulfur dioxide (SO2)
Toxicol. Lett.
Green spaces are not all the same for the provision of air purification and climate regulation services: the case of urban parks
Environ. Res.
Improving local air quality in cities: to tree or not to tree?
Environ. Pollut.
The high-resolution estimation of sulfur dioxide (SO2) concentration, health effect and monetary costs in Beijing
Chemosphere.
Two-step-hybrid model based on data preprocessing and intelligent optimization algorithms (CS and GWO) for NO2 and SO2 forecasting
Atmos. Pollut. Res.
Development of ozone prediction model in urban area
Int. J. Innov. Technol. Explor. Eng.
The effect of data size of ANFIS and MLR models on prediction of unconfined compression strength of clayey soils
SN Appl. Sci.
Evaluation of relationship between meteorological parameters and air pollutant concentrations during winter season in Elazığ, Turkey
Environ. Monit. Assess.
Evaluation the effect of green space on air pollution dispersion using satellite images and landscape metrics: a case study of Isfahan City
Fresenius Environ. Bull.
Integration assessment of the protected areas using landscape ecological approach (Case Study: Kolah Ghazy National Park and Wildlife Refuge)
J. Town Country Plann.
Forecasting PM10 levels using ANN and MLR: a case study for Sakarya City
Glob. Nest. J.
Statistical models and time series forecasting of sulfur dioxide: a case study Tehran
Environ. Monit. Assess.
Forest landscape visual quality evaluation using artificial intelligence techniques as a decision support system
Stoch. Env. Res. Risk A.
Cited by (84)
Machine learning-based prediction of biological oxygen demand and unit electricity consumption in different-scale wastewater treatment plants
2024, Journal of Environmental Chemical EngineeringIntegrating land use and traffic to spatial prediction of particulate matter
2024, Urban ClimateA hybrid causal machine learning to reveal driving factors responsible coal market: Case of the Chinese industry
2024, Journal of Cleaner Production