A spatial approach to mineral potential modelling using decision tree and logistic regression analysis

Honarvar, Pauline (2001) A spatial approach to mineral potential modelling using decision tree and logistic regression analysis. Masters thesis, Memorial University of Newfoundland.

[English] PDF (Migrated (PDF/A Conversion) from original format: (application/pdf)) - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (11MB) [English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. (Original Version)

Abstract

Logistic regression analysis and classification methods using decision tree analysis were used to generate two quantitative mineral potential maps for the Lake Ambrose area (NTS 12A/10) of central Newfoundland. The response variable consisted of 47 surface mineral occurrences plus 49 randomly selected sites representing nonmineral occurrences. Mineral deposit models and regional exploration methods were used to choose a set of predictors consisting of geology fault proximity, till and lake sediment geochemistry, and surficial geology. A spatial weighting function predictor was developed to account for the clustering of the mineral occurrences. -- The predictors were analyzed and recoded to derive a set useful in developing the quantitative models. The categorical geology predictor was converted into two binary predictors; felsic volcanics and mafic volcanics. Fault proximity was analyzed by the weights of evidence method to determine the optimal buffer threshold to convert the continuous distance values to a binary measure ‘close to faults’ versus ‘far from faults’. The optimal thresholds were the 400 m and 1000 m buffers. Principal components analysis was applied to the till and lake sediment geochemistry to derive component summary variables. Three component predictors were added to the database: till component 2 (TPC2) representing base metal and gold, lake sediment component 2 (LPC2) representing base metals and lake sediment component 4 (LPC4) representing gold and its pathfinder elements. The till geochemistry predictors (Au, Cu, Pb, Zn and TPC2) were analyzed for spatial autocorrelation and an interpolated surface was derived using kriging techniques. The lake sediment geochemistry predictors (Au, Cu, Pb, residual Zn, LPC2 and LPC4) 2343 converted to a surface by mapping their values on the catchment basins in which they were sampled. -- The decision tree analysis indicated the spatial weighting function, felsic volcanics and the 400 m binary fault proximity predictor were significant predictors of mineral potential. Logistic regression analysis indicated that the spatial weighting function, felsic volcanics, the 1000 m binary fault proximity predictor and copper in till were significant predictors of mineral potential. The agreement, at the 96 sample sites, between these two modelling methods was 84.3%. The decision tree and logistic regression raster mineral potential maps were compared using Yule’s α. A value of 0.54 indicates good agreement between the maps. Both models correctly classified approximately 79% of the 96 mineral/nonmineral occurrences. Due to the sparseness of the dataset, accuracy could not be measured as there were not enough samples to set aside a test dataset. -- Mineral potential reliability maps were generated using the mutually exclusive and exhaustive regions from the decision tree analysis and the joint probability model from the logistic regression analysis. The mineral potential and reliability maps were combined (multiplied) the form a favourability map. The favourability maps from the decision tree and logistic regression analyses were combined to indicate overall zones of high mineral potential and high reliability. Mineral exploration claims cover much of the study area and only a minor part of the high favourability areas were not claimed as of September 2000.

Actions (login required)

View Item