1Solar Radiation and Atmosphere Modeling Group, Physics Department, University of Jaén, Jaén, Spain
2Center of Advanced Studies in Energy and Environment, University of Jaén, Jaén, Spain
3Mesoscale and Microscale Meteorology Division, National Center for Atmospheric Research, Boulder, Colorado, USA
4Solar Consulting Services, Colebrook, New Hampshire, USA
Abstract. The Level-3 MODIS aerosol optical depth (AOD) product offers interesting features for surface solar radiation and numerical weather modeling applications. Remarkably, the Collection 5.1 dataset extends over more than a decade, and provides daily values of AOD over a global regular grid of 1°×1° spatial resolution. However, most of the validation efforts so far have focused on Level-2 products (10-km, at original resolution) and only rarely on Level-3 (at aggregated spatial resolution of 1°×1°). In this contribution, we compare the Level-3 Collection 5.1 MODIS AOD dataset available since 2000 against observed daily AOD values at 550 nm from more than 500 AERONET ground stations around the globe. One aim of this study is to check the advisability of this MODIS dataset for surface shortwave solar radiation calculations using numerical weather models. Overall, the mean error of the dataset is 0.03 (17%, relative to the mean ground-observed AOD), with a root mean square error of 0.14 (73%, relative to the same), albeit these values are found highly dependent on geographical region. For AOD values below about 0.3 the expected error is found very similar to that of the Level-2 product. However, for larger AOD values, higher errors are found. Consequently, we propose new functions for the expected error of the Level-3 AOD, as well as for both its mean error and its standard deviation. Additionally, we investigate the role of pixel count vis-à-vis the reliability of the AOD estimates. Our results show that a higher pixel count does not necessarily turn into a more reliable AOD estimate. Therefore, we recommend to verify this assumption in the dataset at hand if the pixel count is meant to be used. We also explore to what extent the spatial aggregation from Level-2 to Level-3 influences the total uncertainty in the Level-3 AOD. In particular, we found that, roughly, half of the error might be attributable to Level-3 AOD sub-pixel variability. Finally, we use a~radiative transfer model to investigate how the Level-3 AOD uncertainty propagates into the calculated direct normal (DNI) and global horizontal (GHI) irradiances. Overall, results indicate that, for Level-3 AODs smaller than 0.5, the induced uncertainty in DNI due to the AOD uncertainty alone is below 15% on average, and below 5% for GHI (for a solar zenith angle of 30°. However, the uncertainty in AOD is highly spatially variable, and so is that in irradiance.