| At the present situation, under the shortage of mineral resources and the sharply increasing of exploration cost and exploration difficulty, how to make full use of massive geological spatial data and the modern information technology for quickly and efficiently carrying out mineral resources assessment and application researches is very important, which will be benefited for understanding regional mineral resources potential, implementing mineral exploration deployment and selecting specific ore prospecting target area. At present, a large number of mineral resource exploration methods and technologies have appeared which are mainly divided into four categories which are geophysical method, geochemistry method, geological factor analysis method and remote sensing technology. The paper focuses on method studying and application related to mineral resource quantitive evaluation and exploration geochemistry. Whether it is for ore district exploration or regional exploration, geochemical analysis methods and quantitative evaluation technologies are important contents for mineral resources quantitative evaluation, which have played huge roles in mineral exploration.The intrinsic relationship among geochemical elements can be characterized by statistical analysis methods. These methods can be used for geochemical process idenfication to adied by infering and revealing the evolving rules and provenance characteristics of geochemical elements. The formation of ore deposits is usually caused by anomalous enrichment or anomalous depletion of multiple geochemical elements, such copper deposits are commonly accompanied by the enrichment of Pb, Zn or other elements. Therefore, multivariate statistical analysis methods, from the point of combination of indicator elements, provides powerful tools for studying of multi-element anomalies. However, tranditional multivariate statistical methods for data processing should meet the prerequisites of non sum constraint of geochemical data, in other words, the geochemical data are compositional data which are constrained by data-sum. Therefore, inappropriated use of tranditional multivariate statistical methods may be lead to spurious correlation among geochemical elements. In recent years, univariate analysis, bivariate analysis and multivariable analysis based on compositional data transformation have been achieved great development and application in exploration geochemistry and environmental geochemistry. In addition, the content distribution of geochemical elements are affected by metal mineralization or human activities, which have caused the content of elements strongly deviate from the normal or lognormal distribution, and thus, traditional statistical methods are not ideal methods for these distributed geochemical data analysis. Another question is the treatment of outliers, the popular methods is robust statistical analysis methods, such as robust correlation analysis, robust principal component robust and robust factor analysis, etc.Exploration geochemistry prospecting is according to the content of geochemical element and the strength of geochemical anomalies for anomaly delineation and evaluation. However, due to buried depth of ore body and the effect of overburden, the geochemical anomalies are often expressed as weak signals and influenced by the superposition of other interfering factors which resulted in the identification of weak and concealed anomalies become very difficulty. How to apply appropriate data processing techniques and mathematical models for recognition of weak and concealed geochemical signals is an important task. Researches show that the natural system is gradually tend to complexity from normal distributions to lognormal distributions, and then to Pareto distributions. Pareto distributions describe the most complex natural system, showing stronger fractal/multifractal characteristics. From the perspective of ore-forming processes, ore formation is the result of complex physical and chemical processes, there are considerable overlaps between igneous and hydrothermal and between sedimentary and hydrothermal. Ore-forming process with nonlinear and complexity characteristics are endowed with self-similarity and the anisotropic features both in space domain and frequency domain. Studies show that the fractal/multifractal theory can identify and separate geochemical anomalies from background.Mineral resources are kind of nonrenewable natural resources which are characterized by concealment, uncertainty and risk of the exploration. Mineral resource assessment not only requires folly understanding the geological models such as ore-forming process model and favorable environment model, but also take the relations between deposits and ore controlling factors into account. Mineral potential evaluation should use a large number of regional geological, geophysical, geochemical data under the condition of the absence of direct information from drilling. The spatial data have the characteristics of multiple sources, multiple types, multiple temporal, and multiple scales, therefore the geologists have to comprehensive treatment of these data. However, how to establish efficient and modern mineral resources assessment system is not only an important scientific problem that is need to be solved, but also the main technical approach for delineating the most mineralized targets. Through comprehensive optimization of multi-source geospatial information based on mineral resource quantitative model, the final predicted map can be acquired and the mineral reserve can be estimated, which play an important role for discovering target prospecting area, and for reducing exploration risk. At present, many mineral resources quantitative evaluation models have been developed, which can be divided into three categories, knowledge-driven (such as fuzzy logic, Boolean operation), data-driven (such as weights of evidence, neural network, support vector machine) and hybrid-drive model (such as fuzzy-weights of evidence, fuzzy-neural network). Each model has their own advantages and disadvantages. According to actual needing, how to choose ideal models to reach the goals of dwindling prospecting area and improve the forecasting precision is an important content.According to the inherent characteristics of geochemical exploration data (such as "sum constraint, different distribution patterns), complementary advantages among multiple-sources geospatial data and the predicted performance among different quantitative evaluation models, the main purposes of the study include the following aspects:(1) Application of univariate analysis, multivariable analysis and compositional data analysis to study the relationships of element combination, geochemical evolving rules, discrimination of material sources and the migration regularity. Multivariate statistical analysis, compositional data analysis methods and robust statistics are used to study the closure effect of geochemical data and outliers derived from experimental results.(2) Application of multifractal theory to simulate geochemical element distribution patterns, metallogenic process and identification of concealed or weak geochemical anomalies. The S-A (fractal filtering) technology is used to decompose multi-elements anomalies. Integrating C-A (area-concentration) model and singularity theory model for characterization of geochemical element distribution patterns at the space domain and frequency domain, and for quantitative identification and separation of geochemical anomalies;(3) Different mineral resources quantitative evaluation models are studied by fusion of multi-source geospatial information, and the key technologies of each model are analyzed, which optimize the models by reducing the exploration risk and effectively identifying the most metallogenic potential area.In view of the above research purpose, at the present study the Nanling metallogenic belt is choose as a demonstration plot to highlight the applied methods are progressiveness, practicability, maneuverability and demonstration, etc. The Nanling belt is characterized by complex geological conditions and is seriously covered by thick vegetation and Quaternary clays. Therefore, the need for a set of geological prospecting theories and effective prospecting methods that satisfy the actual situation in the study area is pressing. In addition, the Nanling metallogenic belt across four provinces or autonomous region, such as Jiangxi, Hunan, Guangdong, Guangxi. However, previous mineral exploration and evaluation were usually implemented within individual province or autonomous region which leads to their studies limited in a small range and lack of macro planning and comprehensive research. Traditional evaluation work is mainly focus on the study of some typical deposits, such as the proposed deposit model, or for empirical or semi quantitative deduction and interpretation aided by available geological data, geochemical and geophysical data, and lack of system research related to exploration geochemistry analysis and mineral resources quantitative evaluation. In the present study, statistical analysis methods, especially compositional data analysis based on multivariate analysis provide powerful tool for detecting geochemical process identification and element combination in the study area. Mesozoic metallogenic characteristics of the Nanling belt, especially the large scale of the Yanshanian ore-forming event within a short time interval means that the mineralization processes exhibit unexpectedness, uniqueness, and complexity which cause the superimposition of geochemical signals.On the basis of fully understanding of regional geological characteristics, metallogenic characteristics and ore-forming regularity in the Nanling belt, in the present study, geochemical analysis methods and quantitative mineral resources assessment models are investigated which are used to identify geochemical anomaly and evaluate mineral potential and reveal geochemical process. The applied geochemical analysis methods include multivariate analysis, compositional analysis, robust analysis and multifractal analysis (e.g., singularity analysis, C-A fractal mode, fractal filtering technique). The quantitative mineral resources assessment models include weights of evidence, logistic regression, data-driven fuzzy logic model and evidence theory model. The main results are obtained as following:(1) k-means clustering is used to partition the data into three groups. The results show strong spatial distribution regular pattern, respectively Cluster 1 mainly corresponds with Precambrian-Ordovician and Triassic-Cretaceous strata; Cluster 2 is mainly related to Yanshanian and Caledonian intrusive rocks; Cluster 3 dominantly associates with Devonian-Permian strata;(2) Compositional data analysis and robust statistics are used for principal component analysis and the factor analysis. There are 39 elements derived from stream sediments geochemical data which are used to study element combination regularity and geochemical evolution process in the Nanling belt. Finally, four types of element combinations are determined, respectively W-Sn-Bi-Mo-Ag-Li-Be (tungsten polymetallic elements combination), Cu-Ni-Sb-As-Cd-Cr (heavy/toxic metal pollution elements combination), La-Y-Th-U-Zr-Nb-K-Na-Al (rare earth element group), and Fe-Co-Ti-V-Cr-Ni-Mg-Mn (basic/ ultrabasic minerals).(3) Fractal/multifactal analysis is applied to study geochemical statistical distribution patterns, simulate metelligenic characteristics and identify weak anomalies. The elements of W、 Sn、Mo、Bi、La and Y closely related to tungsten and REE mineralization are investigated for identification of geochemical anomalies and statistical distribution patterns; (4) Weight of evidence model, fuzzy logic model, logistic regression model and evidence theory model are used for tungsten polymetallic mineral resources assessment in the Nanling belt based on multi-source geo-spatial database. These four mathematical geological model are comparative studied. In order to evaluate the success rate accuracy, the receiver operating characteristic curves (ROC) and area under the curves (AUCs) for the four potential maps are constructed. The results show that the AUCs for the fuzzy logic, logic regression, weight of evidence and evidence theory models are 0.8406,0.7757,0.7828 and 0.8061, respectively. The results show that the prediction accuracy of four models is relatively high, but the performance of the fuzzy logic model is best, and then evidence theory model. |
PDF Full Text Request