Predictive mapping using GIS to locate epithermal gold deposits at Cabo de Gata (Prov. of Almeria, Spain).

A spatial analysis of exploration data from a Neogene volcanic field, hosting the San Jose and Palai adularia-sericite-type Pb-Zn(-Ag-Cu-Au) and Rodalquilar high-sulphidation Au-alunite(-Cu-Te-Sn) mining districts, was first carried out to identify the statistical and spatial properties of the data including the detection of drifts, local spatial relationships and spatial outliers. Spatial and non-spatial relationships between pairs of variables were analysed. Data integration was carried out using knowledge models, such as weighted index overlay and fuzzy logic, and data-driven models, including weights of evidence, logistic regression and artificial neural networks. Models were developed using the internal language of the ArcGIS software and external programs for statistical analysis and artificial neural networks. More than 30 predictive maps were obtained, all related to one of five exploration criteria: altered acid and intermediate rocks with high K content, intermediate porphyritic intrusions, favourable fracture structures, zoned hydrothermal alteration, and enrichment in Au and associated metals. The results showed that the favourability maps obtained from the different methods were generally similar, although there were some differences. The success rates were only slightly higher for models based on logistic regression and in particular artificial neural networks. The results from the data-driven models indicate that distance from fracture structures, 5/7 ratio Landsat TM bands, residual gravimetric anomalies and PC2 of the geochemistry indicating As, Sb and Sn were the most important predictors, consistent with the deposit model.
A spatial analysis of exploration data from a Neogene volcanic field, hosting the San Jose and Palai adularia-sericite-type Pb-Zn(-Ag-Cu-Au) and Rodalquilar high-sulphidation Au-alunite(-Cu-Te-Sn) mining districts, was first carried out to identify the statistical and spatial properties of the data including the detection of drifts, local spatial relationships and spatial outliers. Spatial and non-spatial relationships between pairs of variables were analysed. Data integration was carried out using knowledge models, such as weighted index overlay and fuzzy logic, and data-driven models, including weights of evidence, logistic regression and artificial neural networks. Models were developed using the internal language of the ArcGIS software and external programs for statistical analysis and artificial neural networks. More than 30 predictive maps were obtained, all related to one of five exploration criteria: altered acid and intermediate rocks with high K content, intermediate porphyritic intrusions, favourable fracture structures, zoned hydrothermal alteration, and enrichment in Au and associated metals. The results showed that the favourability maps obtained from the different methods were generally similar, although there were some differences. The success rates were only slightly higher for models based on logistic regression and in particular artificial neural networks. The results from the data-driven models indicate that distance from fracture structures, 5/7 ratio Landsat TM bands, residual gravimetric anomalies and PC2 of the geochemistry indicating As, Sb and Sn were the most important predictors, consistent with the deposit model.