利用改进非参数估计法的DEM误差置信区间估计

Confidence Interval Estimation for DEM Errors Based on a Modified Non-Parameter Estimating Function

为了降低误差母体分布的非正态性和较少采样数对DEM误差方差置信区间估计精度的影响,发展了改进的非参数估计法(M-NPM)。以机载激光扫描仪获取的6个不同区域的DEM数据为研究对象,基于交叉验证法分别获取了三组近似正态分布和三组非正态分布的DEM误差数据,并将其作为试验母体。从母体中随机采样,基于M-NPM获取方差置信度为95%的置信区间,借助母体进行精度验证,并与NPM结果比较。结果分析表明,当采样数小于40时,两种方法的模拟结果精度均受误差母体分布的影响,但M-NPM受影响的程度小于NPM;相比误差母体正态分布,当误差母体非正态分布时,M-NPM的估计精度明显优于NPM;无论误差母体服从何种分布,采样总量为多少,M-NPM的精度始终高于NPM,但M-NPM较高的精度是以较宽的置信区间为代价的。

A modified non-parameter method（M-NPM） for the confidence interval estimation of DEM errors was developed based on NPM.Six different DEMs obtained by an airborne laser scanner were employed to comparatively analyze the accuracy of M-NPM and NPM.The six DEM error populations with three slightly non-normal and three very non-normal distributions were acquired with an across-validation process.Stochastic sampling from error populations allows us to report that when the sampling number is smaller than 40,both NPM and M-NPM are obviously affected by the degree of normality of the population distribution,but the influential degree of M-NPM is smaller than that of NPM.No matter what the population distribution is,and how many the sampling points are,the results of M-NPM are more robust than NPM,which is attributed to the fact that M-NPM presents wider confidence intervals than NPM.