Mapping depth-to-rock from legacy data, using a generalized linear mixed model

Title
Mapping depth-to-rock from legacy data, using a generalized linear mixed model
Publication Date
2014-01-27
Editor
Editor(s): Dominique Arrouays, Neil McKenzie, Jon Hempel, Anne Richer de Forges and Alex B McBratney
Type of document
Conference Publication
Language
en
Entity Type
Publication
Publisher
CRC Press
Place of publication
London, United Kingdom
Edition
1
UNE publication id
une:1959.11/73535
Abstract

'Depth-to-rock' is a crucial component of the GlobalSoilMap.net project, because soil attributes should only be mapped where soil is physically present. We have used a Generalized Linear Mixed Model (GLMM) to map the probability that depth-to-rock exceeds a depth of interest in the soil profile. The GLMM has fixed effects based on gamma-ray spectrometry, and random effects based on a nugget + exponential covariance function. The model predicts with high accuracy in the shallow soil layers (<30 cm), but accuracy degrades with depth. This is at least partly because gamma-ray spectrometry only measures the top 40 cm of the soil profile. Future work must address the speed of the algorithm, and deal with heteroscedastic variance as depth in the profile increases.

Link
Citation
GlobalSoilMap Basis of the global spatial soil information system, p. 295-299
ISBN
9781138001190
Start page
295
End page
299

Files:

NameSizeformatDescriptionLink