Two ideas for analysis of multivariate geochemical survey data: Proximity regression and principal component residuals

Abstract

Proximity regression is an exploratory method to predict multielement haloes (and multielement 'vectors') around a geological feature, such as a mineral deposit. It uses multiple regression directly to predict proximity to a geological feature (the response variable) from selected geochemical elements (explanatory variables). Lithogeochemical data from the Ben Nevis map area (Ontario, Canada) is used as an example application. The regression model was trained with geochemical samples occurring within 3 km of the Canagau Mines deposit. The resulting multielement model predicts the proximity to another prospective area, the Croxall property, where similar mineralization occurs, and model coefficients may help in understanding what constitutes a good multielement vector to mineralization. The approach can also be applied in 3-D situations to borehole data to predict presence of multielement geochemical haloes around an orebody. Residual principal components analysis is another exploratory multivariate method. After applying a conventional principal components analysis, a subset of PCs is used as explanatory variables to predict a selected (single) element, separating the element into predicted and residual parts to facilitate interpretation. The method is illustrated using lake sediment data from Nunavut Territory, Canada to separate uranium associated with two different granites, the Nueltin granite and the Hudson granite. This approach has the potential to facilitate the interpretation of multielement data that has been affected by multiple geological processes, often the situation with surficial geochemical surveys.