Setup and validation of a new multi-angle multi-spectrum kernel- driven model

摘要

The linear kernel-driven model combines the advantage of empirical models and physical models. With the simple form which is convenience for inversion, it is widely used in bidirectional reflectance and land surface albedo research. At present, the kernels are expressed as function of incident and viewing geometry, not related to wavelength. With the support of spectrum database, we make use of the a priori knowledge of the soil and leaf spectrum to derive the new kernel function by rewriting the LiSparse kernel and the RossThick. After adding wavelength into kernels as driven variable, the weights of kernel or we call it kernel coefficients are independent from wavelength. The weight of kernels or describes the relationship between the pixel and sub pixel and can be used to built forward model to caculate short wave albedo. We retrieved kernel weights with the measured BRDF of winter wheat canopy. Results show that the weights of the kernels are solely related with the canopy structure, and the inversion results with multi-band data are much more stable than that with single band data when numbers of angular samples are limited.