GEOSCAN Search Results: Fastlink


TitleAn application of the new omnibus test for conditional independence in weights-of-evidence modelling
AuthorThiart, C; Bonham-Carter, G F; Agterberg, F P; Cheng, Q; Panahi, A
SourceGIS for the earth sciences; by Harris, J R (ed.); Geological Association of Canada, Special Paper 44, 2006 p. 131-142 Open Access logo Open Access
LinksOnline - En ligne
Alt SeriesEarth Sciences Sector, Contribution Series 20070165
PublisherGeological Association of Canada (St. John's, NL, Canada)
Mediapaper; CD-ROM
RelatedThis publication is contained in GIS for the earth sciences
File formatpdf
NTS41P/01; 41P/02; 41P/07; 41P/08; 41P/09; 41P/10; 41P/15; 41P/16
AreaMontreal Ricer Headwaters; Gowganda
Lat/Long WENS -81.0000 -80.0000 48.0000 47.0000
Subjectsmathematical and computational geology; digital log data; field data methods; data collections; statistical analysis; statistical analyses; statistical methods; statistics; mineral potential
Illustrationsformulae; tables; location maps; graphs
AbstractThe weights-of-evidence (WofE) method is based on the application of Bayes' rule of probability: it is a log-linear version of a general Bayesian model. Known mineral deposits or occurrences are used as a reference or training set to calculate, a prior, logit that a deposit or occurrence occurs per unit area. From the evidential themes (map layers) weights are added to the prior logit to either increase (positive weights) or decrease (negative weights) the posterior logit. The response theme is an output map that displays the posterior logit (generally transformed to posterior probability). The combination of input maps assumes that the maps are conditionally independent (CI) of one another with respect to the reference set. Violation of this assumption causes the posterior probabilities to be overestimated or underestimated. Various tests have been proposed to test for CI: (1) contingency table tests (X2 test for independence), (2) the overall or "omnibus" test, and (3) a "new omnibus test". In this paper, these three methods are compared on a dataset for an area containing gold deposits in Archean greenstone rocks in Ontario. The new omnibus test shows that the CI assumption is violated in a 4-layer model, in agreement with the old omnibus test. It is also demonstrated that the X2 test for CI on pairwise combinations of maps can give misleading results, and the new omnibus test is a more reliable approach. Finally, for the application example, the weights of evidence and logistic regression solutions give posterior probability maps that differ greatly in terms of probability values, although they are strongly correlated and yield similar spatial patterns

Date modified: