摘要
Kernel-driven model was chosen to calculate global albedo in the project of multiangular remote sensor MODIS. The best kernels were selected by the venerable 'least square' method. The result of this method was very unstable when only a small amount of angular observations is available. A new criterion has been estalished, called 'least variance' for the kernel’s selection. It takes into consideration the effects of model and measurement based on the information inverse theory. Several tests showed that 'least variance' has many advantages. First, it is less sensitive to noises. Second, it operates well in small sample size. Third, it depends less on the sampling position.
Kernel-driven model was chosen to calculate global albedo in the project of multiangular remote sensor MODIS. The best kernels were selected by the venerable “least square” method. The result of this method was very unstable when only a small amount of angular observations is available. A new criterion has been estalished, called “least variance” for the kernel’s selection. It takes into consideration the effects of model and measurement based on the information inverse theory. Several tests showed that “least variance” has many advantages. First, it is less sensitive to noises. Second, it operates well in small sample size. Third, it depends less on the sampling position.
