2022 Volume 57 Issue 1 Pages 1-14
A Land Use Regression (LUR) model was used to assess the health effects of air pollutants. In this study, as the concentration of the air pollutants is significantly affected by meteorological fields, we developed LUR models along with the meteorological model to estimate the monthly average distributions of PM2.5 and NO2. Spatial distributions of the meteorological fields were obtained by using the meteorological model WRF, and used them as candidates for the explanatory variables of the LUR models. We adopted two methods, i.e., the Regression Kriging method and Support-Vector-Regression (SVR) method to develop the regression models. Regarding the completed LUR models, the explanatory variables estimated by the WRF were selected with a high importance in all months, and the estimated distributions showed a relatively high accuracy with R2 values of about 0.7 for both the PM2.5 and NO2. The effect of the introduction of the SVR method to the prediction accuracy was remarkable for NO2, however, that for PM2.5 was not clear. We think that the advantages of introducing the machine learning methods, such as the SVR method, in developing a LUR model becomes clearer by improvement of the WRF settings and the addition of new explanatory variables.