Confidence interval estimation for DEM mean error

In order to give DEM accuracy an accurate description, this paper developed a non-parameter method to estimate the confidence interval for DEM mean error. ASTERs of six test areas with different terrain topography were employed to comparatively analyze the simulation accuracy of the non-parameter method and the classical student t method. The results indicate that the student t method is clearly influenced by the degree of normality of the DEM error population distribution and the sampling number, i.e, with the increasing of the sampling number, the simulation accuracy is improving; the bigger the degree of the non-normal error population, the less reliable the simulation accuracy. No matter what the error population distribution is and how many the sampling numbers are, the non-parameter method is more accurate than the student t method; when the sampling number is bigger than 10, the confidence interval of the non-parameter method completely satisfies the accuracy requirement. The confidence interval widths of the two methods indicate that with the increasing of the sampling number, the widths become smaller; under the smaller sampling number, the non-parameter needs bigger confidence interval width to satisfy the accuracy requirement. Based on the optimal theorem of confidence interval evaluation, the non-parameter method can be considered as an efficient method for confidence interval estimation.