Significance of multiple scattering in remotely sensed images of natural surfaces

To separate reflectivity and surface structure information in remotely sensed images, a practical hybrid radiosity model, first developed for use in natural landscape environments, has been adapted from existing algorithms in the computer graphics and engineering fields. This radiosity model can be used to predict quantitatively the total radiance leaving a surface, including single-scattering and multiple-scattering components. The model operates on digital terrain models (DTMs) of the topography of the surfaces. A central problem in the radiosity model is to estimate the geometric function or "form factor", which determines how much light is received from adjacent surface elements. In order to achieve the high precisions required to predict the multiple-scattering signal from natural surfaces, the new radiosity model balances accuracy and speed of computation efficiently. For an image of 100 x 100 pixels, the radiosities can be estimated with {dollar}{lcub}<{rcub}2%{dollar} error in 1.5 hours on a DEC Alpha 3000 with a 276 Mhz processor, run under Unix.; Using the hybrid radiosity model, multiple scattering (MS) has been quantitatively predicted for natural surfaces at pixel and subpixel scales for the first time. These two scales separate multiple-scattering effects into those resolved by Landsat Thematic Mapper (TM) and those that are unresolved at subpixel scale. The model results have been verified by laboratory measurements. Two experiments have been made in laboratory for two types of surfaces: surface "TA," consisting of parallel triangular prisms and having an accurate DTM, and surface "BC," fabricated by a computer-controlled machine using DTMs of the Bluff Creek watershed in Northern California. Radiosities predicted for the two surfaces using the computational model were compared with CCD camera measurements of radiance from the physical surfaces. The comparisons demonstrate that the practical hybrid radiosity model is reliable, with 98% accuracy.; In this study, I inverted the radiosity model to estimate the surface roughness of geological surfaces from Landsat TM images. I also applied the radiosity model to correct topographic effects in rugged vegetated areas. In both applications, the radiosity model was used to calculate the bidirectional reflectance distribution function (BRDF) for simulated surfaces designed to match the vegetation canopy surface and geological surfaces (such as a gravel bar). Including the multiple-scattering component in the radiosity calculation noticeably improves both the surface roughness inversion and the topographic correction.