The new parallel SHDOM and the I3RC community Monte Carlo models: Comparisons of 3D radiative transfer efficiency

This poster examines the trade-offs between computational cost and accuracy for two three-dimensional radiative transfer codes: a new parallel implementation of the Spherical Harmonics Discrete Ordinates Method (SHDOM) and a community Monte Carlo model. First, the multiple processor implementation of SHDOM is described and the I3RC community Monte Carlo model is summarized. The monochromatic radiative transfer comparison is made in two high resolution cloud fields: a domain filled with stratocumulus clouds and another containing scattered shallow cumulus, absorbing aerosols, and molecular scattering. Calculations are performed for a range of accuracy parameters and the relationships between accuracy and computation time are compared.

Monte Carlo accuracy depends primarily on the number of trajectories. Several variance reduction techniques are used to reduce the intensity sampling noise due to highly-peaked phase functions, though these techniques can bias the intensities. SHDOM accuracy is controlled by the spatial and angular resolution and must be assessed through convergence tests on a case-by-case basis. SHDOM is hundreds of times more efficient (i.e. faster computation time for comparable error) than Monte Carlo for computing pixel intensities in multiple directions. The two models are roughly comparable for pixel fluxes. Monte Carlo is clearly superior for pixel fluxes and flux divergences with spectral averaging or for spatially averaged fluxes. In some cases memory use, rather than computation time, may limit the resolution of SHDOM calculations.