SELF-ORGANIZING MAPS APPLIED TO DECLUSTERING IN PREFERENTIAL SAMPLING

Abstract

Sampling processes in mineral exploration often result in preferentially sampled areas, with the formation of clustering. Some factors can cause areas to be preferentially sampled, accessibility conditions, attribute values, and the sampling strategy. Clustering impacts statistical inference of area. The objective of the present paper is to propose a new approach to declustering methods using Kohonen network, Self-Organizing Maps (SOM). SOM are a type of artificial neural network used for unsupervised classification. The methodology assigns each sample a weight to calculate the declustered mean. The assignment of weight to each sample in an area is inversely proportional to the densely sampled in area. The declustered mean is given by the sum of the weight multiplication with the attribute value of each sample. Therefore, the logic of assigning weights is similar to Cell Declustering method, but the delimitation of the densified areas is different. SOM identifies areas with non-linear margins, unlike the Cell Declustering method. A case study is presented, using the Walker Lake data set. The present research is not intended to replace classical declustering methods, but rather to present a new approach to a routine problem in reserve evaluation. Although the mathematics of the applied technique is indeed complex, the results can be promising.