Satellite-based ground PM2.5 estimation using timely structure adaptive modeling

Although ground-level measurement of PM2.5 is relatively accurate, this method is limited in spatial and temporal coverage due to the high costs. Recently, satellite-retrieved aerosol optical depth (AOD), with high-resolution and wide spatial-temporal coverage has been increasingly applied to estimate PM2.5 concentrations. However, these AOD-based PM2.5 concentrations were spontaneously estimated using the structure fixed models across an entire study period. While these 'structure fixed' simplifications greatly facilitated the efficiency of model developments and enhanced their generalizability, they ignored the fact that the 'contributors' of PM2.5 variation are not always coherent with time. For this, we propose a timely structure adaptive modeling (TSAM) method for satellite based ground PM2.5 estimation in this study by considering the timely variations of modeling predictors and magnitude of predictors at respective optimal spatial scales. Meanwhile, the reliability of TSAM for estimating national scale daily PM2.5 concentrations was tested by employing the AOD data from June 1, 2013 to May 31, 2014 over China with other multi-source auxiliary data such as meteorological factors, land use etc. While the fitting degree (R2) of the daily TSAM models is 0.82, the one in 10-fold validation is 0.80, which are relatively higher than previous studies. These results are significantly better than those from structure-fixed models in this study. Additionally, the TSAM simulated PM2.5 concentrations show that the national annual PM2.5 concentration in China during study period is 69.71 μg/m3 with significant seasonal changes. These concentrations exceed the Level 2 of CNAAQS in more than 70% Chinese territory. Therefore, it can be concluded that the TSAM is a promising PM2.5 modeling method that is superior to structure-fixed modeling and could be very useful for air pollution mapping over large geographic areas.