具有空间自相关残差的回归模型及其应用

Regression Models with Spatially Correlated Residuals and Its Application

提出了具有空间自相关残差的多变量回归问题,这种残差包括单点、局部和区域化三种尺度.文中提出了两种处理空间自相关的方法:把自相关加入残差协方差作参数的改进估计──相邻相关方法和把残差作区域化最优自相关估计──一般空间自相关方法.用森林对舞毒蛾危害敏感性概率的大范围估计为例,比较了传统回归方法和本文方法间的差异,交叉检验结果表明本文的方法明显优于直接回归方法,这说明加入空间自相关对空间回归问题是必要的.

A multivariate regrssion model with spatially correlated residuals for modeling spatial variation is specified in this paper.Two different ways of modeling spatially correlated errors are considered,of which the neighborhood-correlation method is used to modify the parameters estimation,and the spatial autocorrelation estimation of residuals is proposed to directly estimate the residuals.The prediction of gypsy moth(Lymantria dispar L.)damage is used as an example to compare the differences between the classical regression mode and the modified regression model.The cross-validation results indicated that the new method is better than the old one.