SPATIAL MODELING OF AIR POLLUTANT CONCENTRATIONS USING GWR AND ANFIS MODELS IN TEHRAN CITY

Abstract

Today, air quality is a major subject in city regions that have affected human health, the environment, and the city ecosystem. Therefore, government officials, environmental organizations, health organizations, and city managers often need to model the concentration of air contaminants. This study aimed to compare geographically weighted regression (GWR) modeling and neural network (ANFIS) using Segno and Mamdani rules to spatially predict the concentration density of fNO2, CO, and SO2 pollutant indices. And PM 2.5 for the year 2021 in Tehran. The results of the statistical analysis of Sugeno and Mamdani rules revealed that the (RMSE) in evaluating the ANFIS model with the Mamdani method was 0.895 ppm, and with the Sugno method it was 1.004 ppm, whereas the RMSE in terms of Spatial weighted regression model was obtained on digital model with a height of (12.5 m) and a value of 692.0 ppm. The evaluation results showed that Mamdani and Sugno laws do not have the same and desirable accuracy. For Mamdani law, the RMSE level of PM 2.5 pollutant was (0.71 ppm) and according to Sugno law, this level was obtained for CO pollutant (0.81 ppm). While evaluating the geographically weighted regression model for the four air pollution indices the digital altitude model of (12.5 m) had similar results, which statistically for the digital altitude model of (12.5 m) obtained the RMSE for PM 2.5 (0.82 ppm). The findings of this study demonstrated that the weighted geographic regression model and the ANFI neural network have acceptable functionalities for spatial prediction of air pollutants.