A quantitative methodology to select training set of coherent deposit-type locations to improve data-driven modeling of mineral prospectivity
|Location||International Geological Congress,oslo 2008|
|Author||Carranza, Emmanuel John|
|Holding Date||15 September 2008|
Selecting training deposit-type locations is a crucial component of data-driven modeling of mineral prospectivity. Dissimilarity or non-coherence of multivariate spatial data signatures of deposit-type locations can undermine quality of data-driven mineral prospectivity maps. Most cases of data-driven mineral prospectivity mapping, however, make use of qualitatively selected deposit-type locations with tacit assumption that they are coherent (i.e., with similar multivariate spatial data signatures). This study shows potential for improving quality of a data-driven mineral prospectivity map by using a quantitatively selected training set of coherent deposit-type locations. Selection of coherent deposit-type locations is performed via 2-stage quantitative methodology. Stage 1 involves analysis and generation of deposit occurrence favourability scores of univariate geoscience spatial data by comparing cumulative frequency distributions of data at deposit-type and non-deposit locations. Stage 2 involves identification of coherent deposit-type locations via logistic regression, using multiple sets of deposit occurrence favourability scores of univariate geoscience spatial data as independent variables and binary deposit occurrence scores as dependent variable. Then, the set of coherent deposit-type locations and three sets of randomly selected deposit-type locations were each used in data-driven mineral prospectivity mapping via application of evidential belief functions. Compared to the mineral prospectivity maps based on individual sets of randomly selected deposit-type locations, the mineral prospectivity map based on the set of coherent deposit-type locations has lower uncertainty, better goodness-of-fit to training deposit-type locations, and better predictive capacity against cross-validation deposit-type locations. This study demonstrates that explicit and quantitative selection of training set of coherent deposit-type locations should be applied in data-driven modeling of mineral prospectivity.