Combining weights of evidence analysis with feature extraction - a case study from the Hauraki Goldfield, New Zealand
Feltrin, Leonardo, Motta, João Gabriel, Al-Obeidat, Feras, Marir, Farhi, and Bertelli, Martina (2016) Combining weights of evidence analysis with feature extraction - a case study from the Hauraki Goldfield, New Zealand. Procedia Computer Science, 83. pp. 1262-1267.
|
PDF (Published Version)
- Published Version
Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (695kB) | Preview |
Abstract
In this contribution we combine different image processing and pattern recognition methodologies to map the probability of discovering epithermal mineral deposits in the northern part of the Coromandel peninsula, in New Zealand. The objective of this work is to propose a case-study where the substitution of structural geology GIS themes (commonly developed by humans) with products derived by image processing, computer-based, semi-automatic edge detection analyses, is carried out to reduce subjective input in the prospectivity analysis. Semi-automated lineament extraction results introduced in the mineral favourability statistical modelling can more easily reveal unexpected potentially mineralised target domains, being less subjective. We present initial results of this analysis and explain some of the methodologies adopted. Preliminary results suggest that this approach increases significantly the number of geological discontinuities mapped in the region, with the following implications: (1) prospectivity models are more risk-tolerant and result in an increased number of targets; (2) increments in posterior probability affect the statistical validity of the model due to conditional independence violation, requiring careful assessment of probability overestimation; (3) the feature extraction process identifies numerous lineaments that in some instances represent false positives (lineaments determined by a variety of causes, without geological significance); however, we find that Contrast calculations in the Bayesian analysis tend to penalize these evidential themes, because of the higher number of pixels (cells) containing a positive pattern (lineament existence = 1, being positive). This aspect reduces the overall impact of these predictors on the analysis, mitigating the effect of false positives (lower positive weights of evidence). Despite the limitations, results obtained are encouraging with a clearly superior and more detailed mapping of potential structural sites and their relative probabilities of hosting epithermal deposits.
Item ID: | 47085 |
---|---|
Item Type: | Article (Research - C1) |
ISSN: | 1877-0509 |
Keywords: | cluster analysis, Bayesian learning, weights of evidence, epithermal gold, mineral prospectivity |
Additional Information: | © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/) Presented at the 3rd International Workshop on Machine Learning and Data Mining for Sensor Networks, 23-26 May 2016, Madrid, Spain |
Date Deposited: | 04 Jan 2017 07:42 |
FoR Codes: | 37 EARTH SCIENCES > 3705 Geology > 370510 Stratigraphy (incl. biostratigraphy, sequence stratigraphy and basin analysis) @ 100% |
SEO Codes: | 97 EXPANDING KNOWLEDGE > 970104 Expanding Knowledge in the Earth Sciences @ 100% |
Downloads: |
Total: 987 Last 12 Months: 11 |
More Statistics |