等距对数比变换及混合分布在区域化探数据分析中的应用  

Regional Geochemical Data Analysis Using Isometric Log-ratio Transformation and Mixture Distribution

作　　者：李柱 张德会[1] 杨帆[3] 刘向冲 LI Zhu;ZHANG Dehui;YANG Fan;LIU Xiangchong(School of Earth Sciences and Resources,China University of Geosciences(Beijing),Beijing 100083,China;Inner Mongolian Geological Engineering Co.,Ltd.,Hohhot,Inner Mongolia 010010,China;Institute of Geophysics and Geochemistry,Chinese Academy of Geological Sciences,Langfang,Hebei 065000,China;Institute of Geomechanics,Chinese Academy of Geological Sciences,Beijing 100081,China)

机构地区：[1]中国地质大学(北京)地球科学与资源学院,北京100083 [2]内蒙古地质工程有限责任公司,内蒙古呼和浩特010010 [3]中国地质科学院地球物理地球化学勘查研究所,河北廊坊065000 [4]中国地质科学院地质力学研究所,北京100081

出　　处：《现代地质》2023年第3期662-673,共12页Geoscience

基　　金：国家自然科学基金项目(41373048,41773030);内蒙古自治区地质勘查基金综合研究项目(2020-KY03)

摘　　要：区域化探数据是典型的成分数据,等距对数比变换(ILR)可以有效构建化探数据的标准正交基,消除其闭合效应,解释数据的组成性质,但是解释ILR转换的变量仍然很困难。为使ILR转换更容易理解,本研究利用地质知识和数据驱动的方法构建可解释的ILR转换变量,并将该方法应用于从大兴安岭中南段水系沉积物地球化学数据中提取地质信息。基于地质知识和层次聚类分析,构建了Sn、W、Cr和Ni元素浓度之间的顺序二元划分(SBP),并经ILR转换后表示为变量b1、b2和b3。此外,还采用了由最小信息长度准则(MML)改进的期望最大化(EM)算法,研究上述变量的混合分布。ILR转换的变量具有镁铁质岩浆作用、Sn-W热液成矿和后期地质作用的信息,服从双正态分布或三正态分布。其中b1、b2和b3的高平均值分组对应于锡钨成矿的异常,综合圈定4个锡钨找矿潜力较高的预测区。本研究表明,ILR转换和MML-EM算法在从区域化探数据中提取地质信息和圈定异常方面是一种很有前途的方法。Regional geochemical exploration data are typical compositional data.Isometric log-ratio transformation(ILR)produces an orthonormal basis of geochemical data,which can eliminate the data closure effect and account for the data compositional nature.However,it is still difficult to interpret ILR-transformed variables.To make ILR transformations easier to understand,geological knowledge and data-driven methods are used to construct the interpretable ILR-transformed variables.This method was applied to extract geo-information from stream sediment geochemical data in the central and southern Da Hinggan Mountains,Inner Mongolia,northern China.Based on these geological information and hierarchical cluster analysis,sequential binary partition was constructed among the Sn,W,Cr,and Ni concentrations,and expressed as variables b1,b2 and b3 by ILR transformation.Furthermore,the expectation-maximization(EM)algorithm modified by a minimum message length criterion(MML)was employed to investigate the variables mixture distributions.The ILR-transformed variables follow either a bi-normal or tri-normal distribution,which were interpreted as fingerprints inherited from mafic magmatic,Sn-W hydrothermal,and later geological processes.The high-average subpopulation of b1,b2 and b3 variables of ILR transformation corresponds to the anomalies of W-Sn metallogenic system,and four areas were predicted to have high Sn-W prospecting potential.This study shows that the ILR transformation and MML-EM algorithm are promising tools to effectively extract geo-information from geochemical data and delineate anomalies.

关 键 词：等距对数比变换 混合分布 成分数据 化探 大兴安岭 

分 类 号：P595[天文地球—地球化学] P611.1[天文地球—地质学] P628