The influence of spatial resolution and three-dimensional radiative effects on the accuracy of global horizontal irradiance retrievals

The current generation of geostationary satellites enables the retrieval of global horizontal irradiance (GHI) down to scales of 500 m.  GHI nowcasting products could benefit from this increase in spatial resolution. However, it also provides challenges. Satellite retrievals are almost exclusively based on one-dimensional (1D) radiative transfer where pixels are assumed to be plane parallel and independent from each other. Towards smaller spatial scales, the validity of the plane parallel assumption is expected to increase whereas the independence of pixels is less guaranteed.  This study quantifies the validity of these assumptions in synthetic satellite retrievals.  We use the newly developed Monte Carlo KNMI (MONKI) three-dimensional (3D) radiative transfer code. With MONKI, we simulate 3D as well as 1D top-of-atmosphere reflectances based on cloud fields modelled with the Large Eddy Simulation (LES) code MicroHH. These top-of-atmosphere reflectances are used to retrieve cloud properties and GHI at spatial resolutions ranging from 0.05 to 6.4 km.  Comparing retrieval results for these resolutions gives an indication of the plane parallel bias, while the validity of the independent pixel approximation is quantified by comparing the 1D and 3D based retrievals. For all retrievals, GHI calculated directly from the LES output with 3D radiative transfer serves as a reference. Preliminary results for a shallow cumulus case show that for spatial resolutions below 400 m, the probability density function of the 3D minus 1D reflectances has a clear bimodal distribution.  The two peaks correspond to the illuminated side of the clouds and the cloud-shaded areas.  At coarser resolutions, the bimodal distribution disappears.