ISSN:
1573-8868
Schlagwort(e):
geochemical exploration
;
multivariate analysis
Quelle:
Springer Online Journal Archives 1860-2000
Thema:
Geologie und Paläontologie
,
Mathematik
Notizen:
Abstract Univariate and multivariate statistical methods were evaluated using published multi-element stream sediment data from southwestern and northern New Brunswick. The statistical distributions of elements do not obey Ahrens' “law of lognormality”; eleven of thirteen elements investigated for the Bathurst-Jacquet River area are not lognormally distributed at the 0.05 level of significance. The distributions are positively skewed and leptokurtic and consist of aggregate populations which represent mineral deposits, bedrock, and many other physiographic factors; some of these populations are normally distributed. The efficiency of the Pearson correlation coefficient varied and was compared to nonparametric correlation. Various methods of factor analysis were evaluated and the structure of the factors was similar to the subjective groupings derived from the correlation matrices. Comparison of correlation coefficients and factor models derived from the log-transformed and untransformed Bathurst-Jacquet River data showed that background associations were enhanced by the log transformation at the expense of associations representing mineralization. Q-mode factor matrices could not be satisfactorily interpreted without recourse to the mapping of the factor loadings. The maps produced were inferior to simple concentration maps. An iterative technique was developed for discriminant analysis to refine the sample training groups representing mineralized and background terrain; repeated discriminant analysis after misclassified samples were eliminated altered the inherent character of the training groups. Trend surface analysis was found to give goodness of fits of the trend equations comparable to the fits expected from random numbers. The method was mathematically inappropriate for the type of data used. The goal of exploration geochemical statistical analysis should be to discriminate and sort populations representing mineralized and background populations by classification or filtering techniques.
Materialart:
Digitale Medien
URL:
http://dx.doi.org/10.1007/BF01032864
Permalink
