本文以内蒙古河套灌区内一典型实验区的土壤水盐空间变异性为案例,进行了普通克立格、BP神经网络与稳健克立格估计对比分析。发现对于土壤水分预测,BP法可在一定程度上克服克立格法的平滑效应;在盐分预测中,稳健克立格与BP法对特异值影响估值问题尚未很好解决。BP神经网络技术对原始数据没有参数及严格地采样分布要求,简单而实用。
Based on the spatial variation of soil water - salt at one representative experimental section in Hetao irrigation zone, this paper relatively analysed OK , RK and BP methods , discovered that BP can overcome smooth effect of kriging for soil water, but RK and BP can not solve salt outlier effect. BP neural net technique had no require for parameter and distribution of raw data, it is simple and practical , but it has weak flexibility to sampling design and calculating project.