A new buffer selection strategy for land use regression model of PM_2.5 in Xi'an, China

In order to calculate the spatial distribution of high-resolution air-pollutant levels, the land use regression (LUR) model can be an effective method due to the comprehensive consideration of various factors. Traditional LUR models mostly use predefined buffers, which have the disadvantage of not matching high-resolution data well. In order to get a better-fitting model, a few researches have proposed new buffer selection methods. To solve this problem, we propose a new optimal buffer selection method based on the dichotomy to improve the correlation between predicted variables and pollutant concentration. For some socioeconomic data with high spatial resolution that cannot be obtained, for example, building data is used instead of population density data. Compared with the model with the predefined buffers, the model with our buffer selection strategy explained additional 5% variability in measured concentrations, in terms of the R² of the final model. Our model explained 98% of the samples, and the deviation (1.78%) and root mean square error (5.17 μg/m) were small. It means that the LUR model with our buffer selection strategy can be used as a fit method to better describe spatial variability in atmospheric pollutant levels, which will be conducive to epidemiological research and urban environmental planning.