Spatiotemporal analysis of PM<inf>2.5</inf> in large coastal domains by combining Land Use Regression and Bayesian Maximum Entropy

Abstract

By combining Land Use Regression (LUR) and Bayesian Maximum Entropy (BME), this study constructed a LUR model based on the parameters of elevation, distance to sea, length of roads and Normalized Difference Vegetation Index (NDVI) to generate a global map of PM2.5 distribution in a large costal area in 2015, china. The Bayesian Maximum Entropy was further introduced in the interpolation of LUR space-time residuals. Because of the introduction of BME, the cross-validation results showed that the R2 increased from 0.36 to 0.85, and the root-mean-square error (RMSE) decreased from 23.53μg/m3 to 11.08μg/m3. The average concentration of PM2.5 in the northern coastal areas was higher than that of the southern areas, and the highest concentration of PM2.5 appeared in the inland area of Beijing, Tianjin, Hebei and Shandong provinces during winter times. The annual spatial distribution of PM2.5 was further integrated with population density in Shandong province for risk exposure analysis. The outcome showed that the outdoor population exposure of PM2.5 decreased from inland to sea, and the highest Per capita outdoor exposure value occurred in the central city, Jinan (85.5μg/m3), while the lowest value occurred in coastal areas of Yantai and Weihai.