How much must be known about a cloud field in order to accurately compute the reflected and transmitted flux? Given our limited abilities to measure the time-evolving three-dimensional structure of clous, and the high cost of making trhee-dimensional (3D) radiative transfer computations, we would like to determine the accuracy of various approximations used to convert remote sensing observations to domain averaged solar fluxes.

We use highly resolved (50 m, 1 min) clouds fields from large eddy simulations of convective clouds over land to address sampling and radiative transfer issues. As a benchmark calculation we compute the three-dimensional radiative transfer every minute, then average the resultant fluxes over a three-hour window. Time-height cross-sections (such as would be seen by a ground-based radar or lidar) are extracted from the full fields, then extrapolated back to three dimensions using a hierarchy of models. These models range in complexity from simple average profiles to a stochastic 3D model tuned with the cross-section statistics. We compute the fluxes for each appropriate cloud field using different radiative transfer approximations such as plane parallel homogenous with varying overlap assumptions the independent column approximation, and full 3D radiative transfer.