Temporal and spatial distribution mapping of particulate matter in southwest of Iran using remote sensing, GIS, and statistical techniques

Abstract

Major items concerning air and weather researches include the amount of aerosols and particulate matter (PM) present in the air. Satellite-retrieved, aerosol optical depth (AOD) is a widely used method for the mapping of particulate matter (PM10) concentrations. Precise estimate and mapping of PM10 depend on the resolution of AOD data and the mathematical model, which considers the spatially non-stationary relationship between PM10 and AOD. Khuzestan province of Iran is deficient in a powerful and validated resolved model of PM10 with high spatial temporal resolution. Therefore, the purpose of this study is to investigate and monitor the concentration of PM10 in 26 air quality monitoring stations along with meteorological data obtained from 21 synoptic stations, aerosol optical depth (AOD) extracted MODIS Terra/Aqua, and their relationship which were analyzed using GIS services, statistical models, and remote sensing throughout Khuzestan province of Iran from January 2008 to December 2018. This research verified the MCD19A2 AOD product and then proved that MCD19A2 could accurately indicate the aerosol distribution in the Khuzestan province of Iran. Analysis of average annual MCD19A2 AOD data revealed 2008 and 2009 as the most polluted years in Khuzestan province during 11 years (2008–2018) of study. The study of dust trends showed a significant increase in spring and summer in the study area. The results of this study indicated that PM10 is influenced by AOD and meteorological parameters. Meteorological data together with simplified aerosol retrieval algorithm-retrieved AOD at 1-km resolution were applied as the predictors for the linear regression (LR), multiple linear regression (MLR), the ordinary least squares (OLS), and geographically weighted regression (GWR) models to predict the spatial distribution of PM10 concentrations. Among all the statistical models, the GWR performed better and had higher accuracy. Also, the investigation of indicators such as root mean squared errors (RMSE), mean absolute error (MAE), Akaike’s information criterion (AICc), adjusted coefficient of determination (R2), normal (Z) scores, and Moran’s I of the model residuals, to evaluate the accuracy of the studied models, showed high accuracy and excellent performance of the GWR model in predicting the amount of particulate matter in the study area. These results suggested that the GWR model could provide a reliable way to predict the spatial distribution of PM10 concentrations over Khuzestan province. Assessment of short- and long-term human exposures and then investigation of the effects of particulate matter will be possible through our model.