|
Activity Number:
|
137
|
|
Type:
|
Invited
|
|
Date/Time:
|
Monday, August 3, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistics and the Environment
|
| Abstract - #302947 |
|
Title:
|
Modeling Soil Type Spatial Distribution Using Markov Random Fields
|
|
Author(s):
|
Gerard B.M. Heuvelink*+ and R. Murray Lark
|
|
Companies:
|
Wageningen University and Rothamsted Research
|
|
Address:
|
P.O. Box 47, Wageningen, International, 6700 AA, The Netherlands
|
|
Keywords:
|
geostatistics ; Brook's lemma ; soil ; Markov random field ; Gibbs sampler ; spatial statistics
|
|
Abstract:
|
Geostatisticians typically rely on indicator geostatistics to model categorical spatially distributed variables such as soil type. This appears to work in practice but there are a number of theoretical problems that call for the development and application of alternative approaches. One of these is the Markov random field approach, which simplifies the problem by assuming that the full conditional distribution equals the local conditional distribution. Main problems with real-world application are parameter estimation and computational complexity. Also, it is not easy to verify that the conditional probability distributions defined are valid and yield a proper joint distribution. In this work we explore the potentials and difficulties associated with application of the Markov random field approach in soil science, among others using a case study taken from the 1:50,000 Dutch soil map.
|
