Development of a Methodology Combining Clustering and Conditional Simulation for the Deinition of Underwater Sampling Models Victor Octavio Tenorio Mining and Geological Engineering Department/University of Arizona, Tucson, AZ, USA Sukumar Bandopadhyay, Debasmita Misra, Sathy Naidu, and John Kelley College of Engineering & Mines/University of Alaska Fairbanks, Fairbanks, AK, USA ABSTRACT: Underwater sampling provides a limited visualization of the distribution of platinum- rich minerals in marine placer deposits. By using clusters for partitioning sectors of the deposit, by grouping them by a common property, such as distance or other more elaborated algorithms, such as K-Means of Fuzzy Means clustering, it is possible to develop a methodology which provides adequate subdivisions of the deposit into more rational sub-sets. he application of Conditional Simulation for developing a spatial model that highlights the most relevant features of the deposit, assigns a more rational degree of conidence of the selected sets for resource estimation. INTRODUCTION Ofshore operations for the extraction of crusts and nodules in marine placers are gradually becom- ing an increasing activity. New technologies allow the extraction of minerals from these new resources with low operating costs and minimum impact in the sealoor ecological balance. Some of the targets contemplated around the world are manganese nodules, polymetallic nodules, gold, tin and platinum. he marine deposits at Goodnews Bay, Alaska consists of an unconsolidated concentration of material, rich in heavy minerals. he quantiication of the resource was obtained from grab samples and drill holes. he parameters used for such quantiication were based on tonnage, densities, phys- ical characteristics and grade and mineral content that were estimated with a reasonable conidence level. he spacing between samples must be adequate for observing geological or grade continuity. he conidence for classiication of resources and reserves is based on qualitative or intuitive levels of precision, adjusted with predicting applications of geostatistical tools or conditional simulation studies with adequate levels of conidence (Dominy, 2002). he objective of the analysis is to predict values for the locations where there are no samples inside the Goodnews Bay study area, with an estimation method that will provide ore reserve (vol- ume and grade) with an acceptable level of conidence. Frequently this has been done using diverse interpolation methods, including Kriging, which have proved to have a limited capacity of provid- ing reliable estimates. he method proposed for this estimation is conditional simulation, and it is expected that the method may provide a grade distribution with a reduced level of uncertainty. 872