Abstract:

There exists a complex nonlinear relation between ore-bearing possibilities and prospecting evidences. Predicting mineral targets by modeling this complex relation with multivariate nonlinear statistical models is significant for mineral exploration. The authors propose a kernel minimum square error model for mineral target prediction on the basis of kernel function theories and kernel minimum square error. A VC++ program for raster data oriented mineral target prediction with minimum square error algorithm is developed on the basis of GDAL, a C++ library for the input and output of digital image data, and CLAPACK, a linear algebra software package. The model has been applied to the mineral target prediction in Altay, northern Xinjiang. A raster map layer with 100×151 grid cells is generated in MapInfo. 15 evidential raster map layers are transformed into a digital image data cube of size 100×151×15. The discriminant scores derived from kernel minimum error are computed for all the grid cells with the program developed by the authors. It is shown that the areas with high discriminant scores coincide with the known mineral occurrences. Thus, the kernel minimum square error model is feasible for multivariate nonlinear mineral target prediction.

Key words: kernel function, kernel minimum square error, mineral resources, mineral exploration, target prediction