Reproducibility of a regional geological map derived from geochemical maps, using data mining techniques : with application to Chungbuk province of Korea

Interpolated geochemical maps contain problems of sample catchments and vacant sample areas. However, the resemblance of these maps with regional geological maps implies that there are strong correlations between them. A conceptual model of the geological units from multivariate chemical elements is constructed by decision tree technique. The reproducibility of the geological map with the conceptual model is about 80% in the Chungbuk province map of Korea. Statistical uncertainties indicate that rock units could be predicted reasonably well by decision tree if the geological map polygon is big enough. Misinterpreted points are due to interpolation problems between samples from adjacent rock units and from different sample catchments. In the Chungbuk example, the latter is less significant. The size of the map polygon is one of the most critical factors, and a polygon size greater than 50km super(2) is reliable for such analyses for the Chungbuk example. Combined lithologies in a single mapping unit are not critically affected by decision tree prediction.