Sampling and geostatistics for spatial data
2002
Ver Hoef, Jay
The goals of classical statistical sampling (e.g. estimation of population means using simple random sampling stratified random sampling etc.) geostatistics (e.g. estimation of population means using block kriging) can be identical. For example both can be used to estimate the average value or total amount of a variable of interest in some area. The most fundamental difference between classical sampling geostatistics is that classical sampling relies on design-based inference while geostatistics relies on model-based inference. These differences are illustrated with ex mples. Classical sampling usually considers sampling for finite populations but in the spatial context it is easily adapted to infinite populations. Geostatistics has only considered infinite populations but methods for finite populations have been developed recently. To compare classical sampling to geostatistics for both infinite finite populations I consider the following data sets: 1) a fabricated fixed spatial pattern from an infinite population of a spatially-continuous variable; 2) a single fixed real data set from a finite population on a grid of spatial locations; 3) simulated random patterns from an autocorrelated model from a finite population on a grid of spatial locations. For each data setI select samples randomly. Then I use classical sampling estimators geostatistical estimators of the mean values. Results show that both methods provide unbiased estimates have variances confidence intervals that are valid but in general the geostatistical methods are more efficient having estimates closer to the true values.
显示更多 [+] 显示较少 [-]