Validation of parameter-distribution and observation error-distribution of linear kernel-driven BRDF model

Normal distribution is the very kind of distribution that we choose empirically for error-distribution and parameter-distribution when we engage in the work of remote sensing inversion. This assumption does simplify the calculation process, but up to this day, no special work had been done to validate it. In this paper, taking linear kernel-driven BRDF model as an example, we compare which kind of distribution can simulate the distribution of error and parameter better, normal distribution or t distribution, and try to find the optimum distribution and validate its rationality