The empirical/statistical models to estimate the ground Particulate Matter(PM) concentration from Moderate-Resolution Imaging Spectroradiometer Aerosol Optical Depth (AOD) product were analyzed for the period of 2011-2015 in Seoul, South Korea.
In the model construction of AOD and PM, two vertical correction methods using planetary boundary layer height (PBLH) and vertical weighting/fraction factor and humidity correction from hygroscopic growth factor of aerosol were applied to the respective models. For the training period of 2013~2014, two linear regression models of AOD-PM10, respectively, using PBLH and vertical weighting/fraction factor showed similar performance; the correlation coefficients between AOD and PM10 were 0.64 and 0.66, respectively. In the multiple linear regression model using all predictors, the correlation coefficient between the observed PM10 and the estimated PM10 was 0.72, showing a similar or better performance compared with the results of previous studies. In the seasonal analysis, the lowest correlation of 0.62 was found in spring season when Asian dust frequently occur, whereas the highest correlation of 0.96 was shown in winter season with low PBLH and stable atmospheric condition. On the other hand, the multiple linear regression models to estimate PM10 and PM2.5 from MODIS AOD were constructed in the same way.
In 2015, the empirical model of PM2.5 showed a better performance than PM10 model; the correlation coefficients were 0.69 and 0.74, respectively. The estimated PM10 tended to be underestimated for PM10 observation and this pattern was clear from February to April in 2015 when severe dust events were observed. Also, the estimated PM2.5 showed somewhat underestimation in comparison with PM2.5 observation and this tendency was obvious in winter season. These results imply that the application of predictor in the model construction to derive PM10 and PM2.5 from satellite AOD can be different and anthropogenic factors such as fossil-fuel combustion, automobiles, and population density play significant roles in the PM2.5 model.