The retrieval of surface reflectance parameters from the coupled atmosphere-land surface system: Advanced methods

The need for accurate parameterization of the spatial and temporal variation of land surface characteristics in general circulation models makes it critical to advanced global studies to retrieve the bidirectional reflectance distribution function (BRDF) and albedo of the surface from remotely sensed images. This is the focus of this study. A key step in the retrieval of these parameters is to select a BRDF model that best describes the reflectance properties of the land surface on the global scale. To this end, I compare the current BRDF models and perform a full validation of the Ambrals BRDF model. The results demonstrate that the Ambrals model provides a compact, well-fitting mathematical expression for the anisotropic reflectance of a variety of land covers. Another important step in deriving the surface reflectance properties from the coupled atmosphere and land surface system is to remove the atmospheric effects, which is difficult due to the coupling of atmospheric scattering and surface BRDF and the adjacency effect that is especially significant for off-nadir observations occuring in global studies. I perform a systematic analysis of the sensitivity of atmospheric correction of reflectances to the surface BRDF for the first time, and I show (1) that the surface BRDF must be considered in atmospheric correction and (2) that the two can be coupled in an iteration loop. I also present a new and practical method to derive the first-order analytic approximation of the atmospheric point spread function based on the geometry of the sensor, target pixel and background pixel.; Lastly, I apply the results and approaches obtained above to actual use of the Ambrals BRDF model in the retrieval of surface BRDF and albedo, and in monitoring the phenological change in tropical vegetation with climate change induced by El Nino based on AVHRR GAC images over South America. Results show that surface albedo and normalized difference vegetation index calculated by the Ambrals model improve those obtained by the conventional approaches. The procedure and methodology I develop can also be applied to process the global data obtained by MODIS, MISR and POLDER in the EOS era.