Application Of Spatial Weighting And Higher-Order Principal Component Analysis In Multivariate Geoscience Information Synthesis

Principal component analysis(PCA)is a commonly used multivariate statistical analysis method.Its purpose is to convert a series of correlated data into a small number of independent principal components,so as to achieve effective dimensionality reduction and system analysis.Each principal component(PC)obtained from principal component analysis reflects the different aspects of information contained in the dataset.In geology,this method is often used to process and analyze massive,multi-source / multivariate geo-data(such as geology,geophysics,geochemistry,remote sensing,etc.),so as to extract target information.The process of principal component analysis is generally based on the correlation coefficient or covariance of the input data.However,the construction of correlation matrix or covariance matrix does not take into account the correlation and structural characteristics of spatial data,which is contrary to the fact that there is more or less a certain genetic relationship between geological bodies caused by long-term evolution.Moreover,the distribution of the original information is mostly concentrated near the average value,which is based on the statistical characteristics of multiple values.For the geological anomalies occurred in the metallogenic process,valuable information is often hidden in a small number of values and ignored.Therefore,although the traditional principal component analysis method is widely used in geological research,it is often used for systematic comprehensive analysis of multi-source geological data,but from the point of view of the composition of the algorithm,there is a lack of constraints on the relationship between geological elements.In addition,as a small probability geological event,the occurrence,development and end of metallogenic events are abnormal in time and space.Accordingly,the geological information generated by it should be far away from the main body or background in the data structure.Considering the algorithm itself and the essential characteristics of metallogenic events,In this paper,the traditional principal component analysis(PCA),spatial weighted principal component analysis(SWPCA)and higher order principal component analysis(HOPCA)are applied to the delineation and evaluation of iron resources in eastern Tianshan mining area in China.The feasibility and advantages ofthree methods in mineral resources prediction are discussed through comparative analysis.The main work of this paper is as follows:(1)According to the geological model of the study area,the regional geochemical data are selected,and the above three methods are used to mine the internal relationship between geological data and all kinds of geological phenomena,so as to realize the quantitative identification of mineralization information and ore-controlling elements.Thus,it can provide intuitive data support for metallogenic prediction and geological exploration.(2)Under the guidance of deposit model,two different spatial weight factors are defined by remote sensing data and geological data to construct spatially weighted principal component analysis model.In order to enhance the relationship information between geological elements which are not included in the traditional principal components.(3)Under the guidance of geological anomaly theory,the high-order principal component analysis method is applied to the inheritance of multivariate geological information,and the information that is ignored in the traditional principal component is enhanced by selecting the optimal order.The nonlinear tensile transformation of the data can identify and highlight a small part or geological anomaly information related to mineralization,and provide ideas for the identification of geochemical anomaly information in this study area.