Soolmaz Shamsaie; Mozhgan Ahmadi Nadoushan; Ahmad Jalalian
Abstract
Introduction: Industrialization, urbanization, and population growth are considered as the main causes of urban air pollution that is responsible for millions of deaths per year worldwide. Besides, the impact of urban air pollution on health is considerable. Respiratory and lung diseases, and heart attacks ...
Read More
Introduction: Industrialization, urbanization, and population growth are considered as the main causes of urban air pollution that is responsible for millions of deaths per year worldwide. Besides, the impact of urban air pollution on health is considerable. Respiratory and lung diseases, and heart attacks are largely due to urban air pollution. However, there is a lack of air pollution monitoring stations (hereafter stations) in most cities worldwide because of their high expenses, and, thus, access to high spatial and temporal coverage of air pollutants and their distribution is limited. To address this issue, the main purpose of this study was to estimate CO concentration in Isfahan, Iran, based on air pollution monitoring stations and Moderate Resolution Imaging Spectroradiometer (MODIS) data from 2018 to 2019. Material and methods: In the present work, we used adaptive neuro-fuzzy inference system )ANFIS( and Random Forest (RF) algorithms to estimate CO concentrations. To implement the ANFIS algorithm, based on collected air pollution data from the stations and Aerosol Optical Depth (AOD) data from MODIS imagery, the basic fuzzy rules were extracted. Further, with the integration of fuzzy rules and artificial neural network algorithm, ANFIS algorithm was implemented to model the dispersion of CO level in Isfahan city. To model the dispersion of CO using the RF algorithm, air pollution data and AOD data were used. Since the number of trees and the number of variables in each node are two basic parameters in the success of the RF algorithm, a 10-fold cross-validation method was used to identify value for these two variables.Results and discussion: Our findings indicated that the RF algorithm was more efficient and accurate in spatial modeling the dispersion of CO because it achieved better RMSE and MAE results than the ANFIS algorithm. The RMSE error value of the RF and ANFIS algorithms were 0.724 and 0.809 ppm, respectively. Furthermore, the MAE error value of the RF and ANFIS algorithms were 0.636 and 0.792 ppm, respectively. In the case of spatial dispersion of CO pollutants, the ANFIS algorithm showed that the amount of this pollutant varies in the city. For example, the central and northern regions of Isfahan had the most pollution and the eastern and western regions of Isfahan had the least pollution based on the ANFIS algorithm. Regarding the RF algorithm, it was observed that by moving from the southeast to the northwest of Isfahan, the amount of CO pollutant increases, and the northwestern regions of Isfahan had the highest CO pollution. The examination of numerical values obtained from the ANFIS algorithm showed that the lowest amount of CO pollution in Isfahan city was equal to 1.43 ppm and the highest amount was 2.13 ppm. In contrast, obtained results from the RF algorithm showed that the lowest amount of CO pollution in the city was equal to 0.57 ppm and the highest amount was 2.27 ppm.Conclusion: Overall, it can be concluded that since ANFIS and RF algorithms are appropriate and accurate methods in modeling environmental problems due to their nonlinear modeling, the ability to reduce the negative effects of outgoing data, and less sensitivity to the local minimum problem. It should be noted that a significant part of the error observed in the results of ANFIS and RF methods was related to the intrinsic properties of MODIS imagery (i.e., cloud cover and mixed pixel problem due to the coarse resolution of MODIS imagery), point measurements of air pollution data collected from the stations, and recorded data error at the stations.
