MODIS Collection 6.1 3 km resolution aerosol optical depth product: global evaluation and uncertainty analysis

Abstract

Abstract Recently, NASA has released the newest Moderate Resolution Imaging Spectroradiometer (MODIS)/Terra Collection 6.1 (C6.1) aerosol product generated at a higher spatial resolution of 3 km (i.e., MOD04_3K) with main updates in the radiation calibration and the Dark Target (DT) aerosol algorithm. This product is of great importance for air pollution studies at small to medium scales but has not been fully evaluated. This paper aims to provide a comprehensive validation and error analysis of the MOD04_3K C6.1 aerosol optical depth (AOD) data set against Aerosol Robotic Network (AERONET) Version 3 AOD measurements at different spatiotemporal scales from 2013 to 2017 over land and ocean. Results suggest that the data quality of the MOD04_3K AOD data set is overall improved at different spatial scales after quality control. In general, the highest-quality 3 km AOD retrievals (quality assurance = 3) and ground-based measurements are highly correlated (correlation coefficient = 0.81), with a mean absolute error of 0.08 and a root-mean-square error of 0.12. About 63% of the data samples fall within the expected error envelope on a global scale. Nevertheless, there is large spatial heterogeneity in the performance at regional and site scales, with the worse accuracy generally observed in areas covered by bright surfaces or dominated by human activities. This is mainly due to the difficulties in estimating surface reflectance and the aerosol-type assumption. In addition, aerosol retrievals are overall overestimated, especially over North America, Europe, and East Asia. Furthermore, the estimation errors and uncertainties are highly related to varying surface and atmospheric aerosol conditions, which become larger with increases in surface brightness, aerosol loading, Angstrom exponent and single scattering albedo. The MODIS 3 km AOD data set is generally less accurate than its coarser-spatial-resolution (10 km) counterpart due to a decrease in the opportunity to discard marginal pixels from the retrieval. Therefore, further algorithm improvements are needed to reduce the estimation uncertainty, especially for heterogeneous urban and bright surfaces.