KRIGING AND AUTOMATED VARIOGRAM MODELING WITHIN A MOVING WINDOW
KRIGING AND AUTOMATED VARIOGRAM MODELING WITHIN A MOVING WINDOW
复制标题
DOI：
10.1016/0960-1686(90)90508-k
复制
发表时间：
1990-01-01
期刊：
影响因子：
--
通讯作者：
HAAS, TC
中科院分区：
文献类型：
--
作者：
HAAS, TC
描述并评价了一种基于普通克里格的空间估计方法，该方法仅使用以估计位置为中心的移动窗口内的采样点来建模协方差结构并构造克里格方程。与传统的克里格方法和隐式模型相比，移动窗口仅依靠局部数据来估计空间协方差结构并计算其估计值，受数据空间趋势的影响较小。通过自动拟合球面变异函数模型来估计窗口的协方差结构，该模型是在几个滞后点计算的半方差的无偏估计。自动拟合采用非线性最小二乘回归，并受核块参数非负约束。通过对美国相邻地区NADP/NTN硫酸盐沉积数据的分析，将该估计方法与更标准的普通克里格方法在固定子区域上进行比较。在此分析中，我们发现移动窗口方案提供了受趋势影响最小的局部变异函数模型，并且这种对变异函数集合的使用允许精确建模空间变化的协方差结构。准确的空间协方差模型是酸沉降效应研究人员根据误差（克里格）方差计算可防御沉降置信区间的先决条件。
A spatial estimation procedure based on ordinary kriging is described and evaluated which consists of using only sampling sites contained within a moving window centered at the estimate location for modeling the covariance structure and constructing the kriging equations. The moving window, by depending on local data only to estimate the spatial covariance structure and calculate the estimate, is less affected by spatial trend in the data than conventional kriging approaches and implicity models covariance nonstationarity. The window''s covariance structure is estimated by automatically fitting a spherical variogram model to the unbiased estimates of semi-variance calculated at several lags. The automatic fit uses nonlinear least squares regression constrained by the nugget parameter being nonnegative. This estimation method is compared to the more standard method of ordinary kriging over fixed subregions by using both procedures in the analysis of NADP/NTN sulfate deposition data in the conterminous U.S. For this analysis, we find that the moving window scheme provides local variogram models which are minimally affected by trend, and that also this use of an ensemble of variograms allows the accurate modeling of a spatially changing covariance structure. Accurate spatial covariance modeling is needed by acid deposition effects researchers because it is a prerequisite for the calculation of defensible deposition confidence intervals from the error (kriging) variance.