Identification Of Geochemical Anomalies Based On Geostatistical Simulation

Exploration geochemistry is one of the most important methods of mineral resources exploration,which contributes a lot to the discovery of many mineral deposits over the past few decades.Extraction of geochemical anomalies is an essencial part of exploration geochemistry,relating directly to the results of mineral resources prediction.Accordingly,it has been a research focus in the field of applied geochemistry for a long time since its establishment.Exploration geochemistry can be reduced to a decision to be made,i.e.,whether certain location should be idenfied to be anomaly indicating potential mineralization occurred in the study area.In general,it is an essencial part to assess the reliability of the final decision made.To make dicision for some geoscientific problems,such as subsurface reservoir modeling,groundwater modeling and soil pollutant disposal,the common procedure is to generate many realizations of the distributions of related variables,from which the statistical distribution of decision variable is then determined.This distribution provides valuable information to support decision made.However,the commonly used procedure for identifying geochemical anomalies rarely paid attention to such reliability measurement,which,however,is a necessary part.On the one hand,geochemical distribution in the surface media is the result of diverse geological processes with complicated properties.As a result,geochemical patterns are endowed with complicated characteristics.But measurement that can be made is usually limited in terms of spatial coverage and resolution,due to a wide range of reasons,such as time,cost and limitions from natural conditions.The two aspects described above jointly result in the uncertainty of the values of unsampled locations estimated from known observations.In other words,the spatial distribution of elemental concentrations is uncertain from the knowledge of observors.The geochemical anomalies obtained are therefore uncertain as well.However,the common procedure to create geochemical distribution patterns is by interpolating the discreted observations into maps,which rarely considered such uncertainty and its effect on the geochemical anomalies obtained thereby.On the other hand,the linear interpolation methods widely used in practice,such as inverse distance weighted(IDW)and kriging,are in nature moving average of values within the local neighborhood,which inevitably results in the local variations being smoothed to some extent.Unfortunately,our interest is the very contrast or variation patterns within the data.Thus,the smoothing effect likely restricts the identification of geochemical anomalies.Furthermore,modeling the local patterns effectively is also important for identifying geochemical anomalies.The commonly used methods,including linear interpolation and variogram-based simulation,mainly have two limitations that make them incapable of charactering complicated spatial structures effectively.One lies in that these methods only focus on the lower-order statistics of distribution patterns,and the other is that most of these methods usually require some assumptions about the distribution type.The recently developed multiple-point statistics(MPS)has proved advantages over traditional geostatistics in a wide range of applications.Its ability in charactering complicated structures,and its nonparametric nature make it flexible in simulating distribution patterns.However,researches related to application of MPS into simulating distribution patterns of geochemical elements were rarely reported.Considering the scientific problem above,the author proposed to identify geochemical anomalies based on simulated geochemical distributions,and a graphical user interface program(GUI)has been developed for this purpose.The method and related ideas were then demonstrated by case studies of identifying geochemical anomalies associated with skarn mineralzation in stream sediment in southwest of China.To be specific,the main research contents and conclusions obtained in the dissertation are follows:(1)The multiple-point statistics of geochemical distribution patternsThe author combined the method of analyzing pattern similarities with hypothesis testing,and constructed a workflow to quantitatively test whether geochemical distribution patterns have multiple-point statistical properties.Case studies about the iron distribution patterns showed that the multiple-point statistics hold for the geochemical distributions.(2)Multiple-point geostatistical simulation of geochemical distribution patternsThe dissertation presented an algorithm named “two-step simulation algorithm” for simulating geochemical patterns based on direct sampling,one of state-of-the-art methods in the fields of MPS.The first step is to simulate patterns on coarse scale grid,and the second step is to simulate patterns on grid at liner scale.The MPS method used here mainly involves three crucial techniques.Firstly,the method used the grid map converted from observations as training data.Secondly,frequency distribution of the values on the simulated location is inferred from sampled values by kernel density estimation(KDE).Then a value is drawn randomly from it.Thirdly,the coarse-scale distribution map is taken as training image for simulating distribution patterns at fine scale.The second one aims to reflect the variability of geochemical patterns more effectively considering the continuous nature of geochemical variables.One major contribution of the dissertation is that a two-step sampling method is derived to effectively and efficiently combine the KDE and direct sampling.The author also presented several aspects to assess the effect of MPS simulation,including lower- and higher-order statistics,multifractal spectra and quantile analysis,etc.Case studies showed that: the MPS simulation method proposed here is feasible and effective;the algorithm takes advantage of the multiple-point statistical properties and showed better results with comparison to sequential Gaussian simulation.(3)Identifying geochemical anomalies based on simulated realizationsThe realizations of geochemical distribution patterns simulated by MPS method can then been used for identifying geochemical anomalies.The dissertation took local singularity analysis(LSA)as an example,and mainly discussed the procedure from three aspects.Firstly,the estimation of singularity exponent is affected by the uncertainties on the unsampled locations within the local neighborhood.Secondly,geochemical anomalies identified should also exhibit uncertainty due to the propogation of uncertainties related to the distribution of geochemical concentrations.Thirdly,geochemical anomalies identified from different realizations can be integrated into a single probability map using thresholding method.Case studies showed that: uncertain geochemical distribution patterns have a significant effect on the geochemical anomalies identified;LSA results of interpolation-based geochemical distribution showed similarities with the trend of LSA results of simulation-based distribution patterns;the method used for determining the optimal segmentation threshold provided result in agreement with that obtained from supervised method,demonstrating its effectiveness;anomaly probability provided by the simulation-based LSA can be seen as the generalized expression of binary value(anomaly or background)from conventional interpolation-based LSA.(4)Identifying multivariate geochemical anomalies based on simulated realizationsConsidering the geochemical association of ore-forming elements,the author further presented method of identifying multivariate geochemical anomalies based on simulated data.One of the most important techniques is to use factor analysis to identify indicator association combined with known geological settings;the other technique is to adopt logistic regression to integrate multiple anomaly probabilities and finally obtain a comprehensive conditional probability indicating the mineralization occurred.Case studies validated the procedure and showed that the multivariate anomaly probability was able to predict unknown mineral deposits.Those areas with high anomaly probability and coinside well with favaroble ore-forming features could provide targets for further mineral exploration.(5)Development of the system used for identifying geochemical anomalies based on geostatistical simulationsThe author followed the idea of software engineering design,developed a software for identifying geochemical anomalies based on geostatistical simulations.The software was coded using the high-level programming language MATLAB,and it has quite a concise architecture,user-friendly interface and the data format being compatible with that of Arc GIS.Thus,it is a practical tool for related data processing.To sum up,the main contributions of the thesis include that(1)it investigated the multiple-point statistics of geochemical distribution patterns and accordingly developed a two-step simulation method suitable for continuous geochemical variables,and(2)further proposed a new procedure for identifying geochemical anomalies based on simulated grid data.Related research and findings above enhance the understandings about identification of geochemical anomalies,and provide general ideas for solving problems related to identification and assessment of patterns of other geoscientific variables.