Validation of three-dimensional radiative transfer in coastal-ocean water systems as modeled by DIRSIG

The radiative transfer equation (RTE) is a mathematical description of radiative gains and losses experienced by a propagating electromagnetic wave in a participating medium. Except for an isotropic lossless vacuum, all other volumes have the potential to scatter, absorb and emit radiant energy. Of these possible events, the global scattering term is the greatest obstacle between a radiative transfer problem and its solution. Historically, the RTE has been solved using a host of analytical approximations and numerical methods. Typical solution models exploit plane-parallel assumptions where it is assumed that optical properties may vary vertically with depth, but have an infinite horizontal extent. For more complicated scenarios that include pronounced 3D variability, a Monte Carlo statistical approach to the radiative transfer solution is often utilized. This statistical approach has been integrated within the Digital Imaging and Remote Sensing Image Generation (DIRSIG) model in the form of photon mapping. Photon mapping provides a probabilistic solution to the in-scattered radiance problem, by employing a two-pass technique that first populates a photon map based on a Monte Carlo solution to the global scattering term, and then later uses this map to reconstruct the in-scattered radiance distribution during a traditional raytracing pass. As with any computational solution, the actual implementation of the technique requires assumptions, simplifications and integration within a cohesive rendering model. Moreover, the realistic simulation of any environment requires several other radiometric solutions that are not directly related to the photon mapped in-scattered radiance. This research attempts to validate raytraced and photon mapped contributions to sensor reaching radiance that can be expected in typical littoral environments, including boundary interface, medium and submerged or floating object effects. This is accomplished by comparing DIRSIG modeled results to those predicted analytically, by comparison to other numerical models, and by comparison to observed field phenomenology. When appropriate, first-order estimates of a computational solution's ability to render a given phenomenon are provided, including any variance or bias that may result as a function of the user-specified solution configuration.