Neural networks method for computation of autocovariance distance

General regression neural networks(GRNN) are briefly analyzed. The relationship between the parameter spread of GRNN and the parameter r0 of autocovariance function (squared exponential function) is derived, and the method for computation of autocovariance distance and fitting of autocovariance function is presented based on GRNN. The study indicates that with optimum spread, GRNN can reflect the variability of soil property and the characteristics of spatial autocovariance.