Climate-specific and global validation of MODIS Aqua and Terra aerosol optical depth at 452 AERONET stations

摘要

Aerosol optical depth (AOD) is a highly influential variable in solar resource assessment and clear-sky radiation modelling. Hence, the accuracy of solar energy estimates ultimately depends on the accuracy of the measured or assumed AOD. Gridded satellite information is often used for solar modelling due to its geographical coverage, and so a global validation of commonly utilised AOD products is imperative. Here, all Level-3 Moderate Resolution Imaging Spectroradiometer (MODIS) daily observations of AOD (at 470, 550 and 660 nm, noted AOD470, AOD550 and A0D660, respectively) from the Aqua and Terra satellites (of 1 degrees x 1 degrees spatial resolution) from 2000 to 02/2018 are compared and validated against all of NASA's ground sensing Aerosol Robotic NETwork (AERONET) V3 Level 2 AOD daily averages from sites that reported at least one year of observations during 2000-2018 (452 sites representing at least 653,000 observations per variable). Furthermore, sub-categorisation by Koppen-Geiger climate regions enables a novel climate-specific validation to ascertain any distinct climatic influence. The results demonstrate significant climatological influences that impact the derived AOD product at all three wavelengths. It is found that blending the two Aqua and Terra products results in a higher accuracy of daily estimates of all AOD products. Each AOD product is validated similarly and separately. AOD550, which is most commonly used in solar resource assessment, is found worst in the equatorial climate (absolute root mean square error (RMSE) of 0.194) and best in the temperate climate (RMSE of 0.126). Globally, the combined Aqua + Terra AOD550 experiences an absolute RMSE of 0.106 and a mean absolute error of 0.109. The most common MODIS AOD retrievals are between 0.01 and 0.25, suggesting that the MODIS daily AOD products may introduce a significant source of uncertainty in modelled irradiance estimates, and that other sources of input data should be used instead whenever their applications demand high accuracy.