Estimating ground PM2.5 concentration using eigenvector spatial filtering regression

摘要

In recent years, air pollution is getting more severe in China, especially fine particulate matter (PM2.5). However, ground monitoring stations are sparse and distributed unevenly, causing difficulty in analyzing the spatial variation of PM2.5 concentration over a large area. More and more research is aimed at estimating PM2.5 using global regression model, ignoring the spatial heterogeneity in PM2.5 distribution. In this paper, we innovatively applied the eigenvector spatial filtering (ESF) method into the PM2.5 modeling, which separates the spatial effects by selecting vectors from the eigenvectors group of a spatial weight matrix and then adds them to the regression model as part of independent variables. The initial model variables includes concentrations of PM₁₀, NO₂, SO₂, CO and O₃, which are related to that of PM2.5. The proposed model was compared with the global multiple linear regression model and the geographically weighted regression model. Experiments results showed that the proposed ESF regression model offers better performance than the other two. Finally, the ESF model was used to map PM2.5 in the study area and help analyze spatial characteristic of the ground PM2.5 concentration.