Domaining based on geological information, variogram analysis and clustering algorithms

Domaining plays a significant role when major statistical or geostatistical variations are observed across a site, which can influence the results of spatial modeling. In this paper, hard and fuzzy clustering algorithms are used as a tool for domaining, with the input containing a combination of geological layers (lithology, alteration, and mineral zone), geochemical layers (Cu, Mo, Ag, and Au grade), and coordinates of boreholes from two exploration campaigns in eastern Kahang deposit, central Iran. The algorithms include the K-means, hierarchical, affinity propagation, self-organizing map (SOM), fuzzy c-means, and Gustafson-Kessel. The optimal number of clusters was determined between 3 and 4. The fuzziness and weighting parameters were respectively obtained in the ranges of 1.1-1.3 and 0.1-0.3, by considering and comparing several hard and fuzzy cluster validity indices. Directional variograms were then calculated to define the anisotropy ellipsoid for the data within each domain to detect the model with the highest degree of anisotropic discrimination among domains. Based on the Jensen-Shannon measure, this model was found to be SOM, using a combination of qualitative and quantitative data as input information, and providing three domains as an output.