Development and Application of Structural Equation Modeling Method For Geochemical Data Analysis

A new Structural Equation Modeling (SEM) approach was proposed and the corresponding algorithms were designed and implemented for model estimation and evaluation in this research. By way of contrast to traditional SEM methods which focus on confirmatory analysis, the new SEM approach is mainly designed for exploratory analysis, which has plenty of applications in geoscience data processing and interpretation. In order to generate an initial model for the new SEM analysis, a constrained variable clustering method was proposed based on a new index representing a type of conditional correlation, which was defined and calculated through SEM. Differently from the conventional conditional correlation coefficient, the new index was designed for measuring the quantity/percentage of the variance existing in two variables related to a response variable, rather than the level of independency of the two variables conditioned by a response variable. It can be used in Principal Component Analysis (PCA) and Factor Analysis (FA) for extracting factors restricted by a response variable. Thereby, these PCA and FA can be considered as constrained PCA and FA. The programs designed for the new SEM are model parameters estimation, conditional correlation coefficient calculation, clustering analysis, and the SEM-based Weights of Evidence (WofE) modeling. The new SEM technology was applied to a lake sediment geochemical dataset to assist for identification of multiple geochemical factors related to gold mineralization in a study area located in Southern Nova Scotia, Canada. The model was further applied in conjunction with the WofE method to integrate geochemical and geological information in mapping mineral potential in the same study area. The results showed that the application of the new SEM method could reduce the effect of the conditional dependency of the evidences involved in WofE.