1: \begin{abstract}

2: We propose a method for estimating the posterior distribution

3: of a standard geostatistical model.

4: After choosing the model formulation and specifying a prior,

5: we use normal mixture densities to approximate the posterior

6: distribution.

7: The approximation is improved iteratively.

8: Some difficulties in estimating the normal mixture densities,

9: including determining tuning parameters concerning bandwidth and

10: localization, are addressed.

11: The method is applicable to other model formulations as long as all the

12: parameters, or transforms thereof, are defined on the whole real line,

13: $(-\infty, \infty)$.

14: %In other words, the method requires the support of each model parameter

15: %to be in one of the following forms:

16: %$(-\infty, \infty)$, $(-\infty, c)$, $(c, \infty)$,

17: %and $(c_1, c_2)$.

18: Ad hoc treatments in the posterior inference such as

19: imposing bounds on an unbounded parameter or

20: discretizing a continuous parameter are avoided.

21: The method is illustrated by two examples,

22: one using digital elevation data and the other using historical soil

23: moisture data.

24: The examples in particular examine

25: convergence of the approximate posterior distributions in the

26: iterations.

27: 

28: \textbf{Key Words:}

29: geostatistics; importance sampling; normal mixture;

30: kernel density estimation; anisotropy; change of support

31: \end{abstract}

32: