IAMG 2001--Cancún Technical Program--Session F

Home Page Technical Program Session F Intro Session F Papers

Weighted Principal Component Analysis: An Application Using Spatial Weights from a GIS by Graeme F. Bonham-Carter*, Geological Survey of Canada, Canada, Qiuming Cheng, York University, Canada, and Daneshfar Bahram, Tehran Polytechnic University, Iran The results of principal components analysis (PCA) on geochemical data are often difficult to interpret, because the samples belong to mixtures of populations, reflecting the operation of several geochemical processes and factors. One approach to understanding large multielement geochemical surveys is to partition the samples into subsets on the basis of the mapped geology or other geographical criteria (e.g. soil type) before PCA (or other methods). The problem with this approach is that interpretation becomes cumbersome, because each sample subset has a different sets of PCs. Spatial factor analysis is another approach, but again, the results are often difficult to interpret, and the analysis does not consider spatial information known about the region of interest (geology, structure, etc). An alternative approach is to weight the samples according to some independent criterion, such as proximity to mapped features (e.g. known mineral occurrences), and apply PCA to a weighted correlation matrix. Weighted correlation coefficients are calculated using a weighted form of product moment correlation. In GIS, it is straightforward to generate maps showing proximity to a geological feature, such as, for example, a particular geological contact, a system of faults, the locations of mineral deposits. Spatial weights may be calculated on the basis of some suitable function, such as 1/d^n, where d is distance to the feature, and n is a power in the range [0,2]. PCA results can then be based on weighted correlations between geochemical elements calculated on the basis of gridded data or the original data table. Results are evaluated by inspection of the component loadings and maps of the relevant PC scores. A biogeochemical survey (Balsam Fir twigs) from southern-shore Nova Scotia was used to test weighted PCA: (1) to evaluate the effects of changing the value of n in the 1/d^n function for weighting, and (2) to examine the effects of applying weights that reflect proximity to geological features known to be related to Au mineralization. The data analysis was carried out with a new GIS-friendly program (GeoDAS) for spatial analysis of geochemical data sets. GeoDAS has a PCA module that allows spatial weights to be applied. Results showed that combinations of elements known to be pathfinders of Au were enhanced by spatial weighting. The best choice of the power n is data-dependent.