An application of spatial prediction using a fuzzy-neural network

Abstract

Identifying spatial distribution of geology types beneath the earth surface is a problem of considerable importance while undertaking constructions of dams. Normally, borehole drilling and geo-tomography (tomography perfected for subsurface exploration) are used to collect data and infer geological distributions. However, the reliability of inference is generally poor and identifying geological distribution is a human intensive process because: 1) boreholes provide information only at the point of drilling and it is hard to interpolate to other areas, and 2) it is difficult to interpret the geo-tomography images in geological terms due to scarcity of quantitative data. These limitations can be overcome by means of fuzzy-neural networks (NNs) and the prediction accuracy is significantly improved. We present a fuzzy-NN application to a spatial distribution prediction problem. Scarcity of data in the cross borehole region was supplemented by using fuzzy data extracted from man-made maps, as input that summarizes human perception of the region. A brief description of the method and a case study are presented.