Full-coverage 1-km estimates and spatiotemporal trends of aerosol optical depth over Taiwan from 2003 to 2019

Abstract

Satellite-derived aerosol optical depth (AOD) provides an effective way to investigate global and regional variations in atmospheric aerosols. However, due to cloud cover and surface reflectance, AOD datasets derived from satellite instruments generally have non-random missing values, which introduces additional uncertainty into AOD data and limits its downstream. To remedy this problem, this study used a two-stage approach based on spatial interpolation and a random forest model to fill the gaps in data generated by the Multiangle Implementation of Atmospheric Correction (MAIAC) aerosol retrieval algorithm, which provides the best-available AOD product to the global public. The relationship between ground-level fine particulate matter concentrations and satellite AOD was considered in the modeling. By gap-filling daily 1-km MAIAC AOD data from 2003 to 2019 over Taiwan, the two-stage model achieved comparable accuracy (coefficient of determination = 0.52, root-mean-square error = 0.22) against ground-level AOD measurements to the accuracy that has been achieved by previous studies. Furthermore, it improved daily high-spatial-resolution AOD estimates to 100% of spatial coverage. Comparisons between the full-coverage estimates and MAIAC retrievals showed that the MAIAC AOD dataset generally underestimated monthly/seasonal/annual mean AOD values in Taiwan. We also used the long-term estimates of the daily 1-km AOD dataset with full coverage to explore the spatiotemporal trends in AOD in Taiwan. The practical approach developed in this study is suitable for application in long- and short-term studies of air pollution and its effects on public health.