Study Of Composition Data Processing Methods

Component data is a kind of data with complex properties.Its greatest feature is that the sum of data variables is fixed(for example,100 wt.).Geochemical data is a typical component data.The "closure effect" caused by sum for fixed value will lead to pseudo-correlation of geochemical data,which makes the analysis results of correlation between geochemical elements deviate,and also makes multivariate statistical methods unable to be carried out directly in simple Euclidean space.Previous geochemical data preprocessing work mostly uses direct logarithmic transformation of data,but it can not eliminate the "closure effect" in component data structure.Taking 1:200,000 geochemical data of cuttings in Jianshan-Pingkouxia area of Gansu Province as an example,this paper uses direct logarithmic transformation,additive log-ratio transformation and central log-ratio transformation to preprocess,and then develops unit,binary and multivariate statistical analysis,including statistical parameter analysis,correlation analysis and principal component analysis.Finally,combined with geological knowledge,the results of data processing of different components(including geochemical model of elements and multivariate statistical analysis)are compared and studied,and the similarities and differences of data processing methods of geochemical components are discussed.The main understandings obtained in this paper are as follows:(1)Although the central log-ratio transformation and the additive log-ratio transformation are similar to the direct logarithmic transformation in changing the skewness kurtosis characteristics of the original data,they can make the original data tend to normal distribution,but the logarithmic ratio transformation also takes into account the relative information between the elements.(2)When studying the correlation between elements in geochemical data,additive log-ratio transformation is an effective method to open geochemical component data for constant elements,which can reduce the closure effect in component data and eliminate the pseudo-correlation between variables to a certain extent.The central logarithmic ratio transformation will cause a large number of negative correlations among elements.In addition,due to the low content of trace elements in geochemical samples and the weak influence of closure effect,direct logarithmic transformation is the most simple and effective method when only the correlation analysis of trace elements is involved.(3)There are a large number of intrusive rocks exposed in the study area.Therefore,in addition to the first principal component obtained by additive log-ratio transformation,the principal component elements combination of direct logarithmic transformation and central log-ratio transformation can mostly represent the geological processes related to magmatic intrusion: 1)geological processes related to ultrabasic rocks and copper-nickel deposits;2)Geological processes related to acidic rocks;3)Geological processes related to peralkaline granites;4)Geological processes related to weathering products.By comparison,it can be found that direct logarithmic transformation and central log-ratio transformation have better dimension reduction effect and quality than additive logarithmic ratio transformation.