Study On Spatial Distribution Patterns Of Regional Geochemical Elements

Spatial distributions of geochemical elements are the clustering performance of elements in geological process which has to do with the history of geological evolution in the course of time. Spatial relationships among geochemical elements depend not only on the affinity of elements, but also their related-geological clustering processes conditions. Therefore, geochemical elements in the earth’s crust are not randomly distributed, but structural. The basic information for metallogenic regularity study and mineral resource potential is regional geochemical information. The anomalous characteristics and spatial distribution patterns of major ore-forming elements and associations of elements play an important role in regional metallogenetic potential. In classical inferential statistical analysis, the independence of the observations which is one of the most fundamental assumptions in hypothesis testing is not satisfied, resulting in biased estimation. Spatial statistics is thus developed for quantitative research on spatial variables and theirs distribution structures; spatial auto-correlation, an effective method in spatial statistics that can be used for analysis of spatial structure, spatial clustering and spatial outliers. With the development of GIS technologies, the combination of GIS spatial analysis and spatial autocorrelation theory has been widely used in various areas of earth sciences and plays a more and more important role in evaluation of gechemical anomalies and concealed ore-finding.In combining statistical modeling and GIS-based, spatial auto-correlation theory can be used to study spatial distribution patterns of regional geochemical elements. Variogram and Moran spatial correlogram were employed to describe spatial variability, spatial structure and spatial heterogeneity of geochemical variables. Based on this, local spatial autocorrelation statistics were employed to identify local patterns of an ore-forming element and of three geochemical associations of elements. A total of1482composite samples (1sample/4km2) collected from a regional stream sediment survey at mapping scale of1:200,000representing about5364km2was chosen and analyzed, revealing two metallogenelic series.The spatial variability of three geochemical ore-forming elements; Silver, Gold and Copper (Ag, Au and Cu) and their three associations were analyzed by means of semivariogram and spatial correlogram in spatial statistics. The analysis involved:①Maximum spatial variability and degree of spatial dependence of the data which were determined via a semi-variogram. The data distribution was positively skewed, thus reducing the influence of extreme values and outliers. Cressie robust semi-variogram estimator was employed to estimate the semivariogram②oran spatial correlogram was used to describe spatial structures, spatial heterogeneity in quantitative ways and testing for the presence of spatial autocorrelation (spatial dependence) in data. Box-Cox transformation was chosen before computing a spatial correlogram. The result of spatial variability analysis shows Moran spatial correlograms provides more information than variogram (positive or negative correlation and spaital heterogeneity). On the basis of the results of spatial variability analysis, major works of the dissertation can be summarized as follows:1. Identification of local clusters and outliers of ore-forming elements based on local spatial autocorrelation statisticsA combination of different graphics from Exploratory Data Analysis (EDA) techniques, such as the histogram, the boxplot, cumulative distribution function diagram, Q-Q and P-P plots, and many others, is first employed to study the data structure and the distribution of geochemical ore-forming element of Cu. The spatial distribution patterns of Cu ore-forming element are studied by means of spatial autocorrelation statistics (global, local Moran’s I; statistic and local Gi*statistic). Based on the concentration of Cu element determined at a maximum spatial variability of about12km in investigating the effects of spatial scales of autocorrelation result identification, six different spatial scales (2km,4km,6km,8km,10km and12km) are taken into account in creating a spatial weight matrix. Box-Cox transformation was also chosen to investigate the effects of extreme values and outliers. The concentration-area fractal method, includes the following study contents:①Degree of spatial autocorrelation and spatial similarity (or dissimilarity) observed among neighboring samples over the entire study area, determined by using global Moran’s I statistic at different spatial scales.②Local Moran I, was is used to assess and identify the degree of local spatial instability (non-stationarity) of Cu concentration;③oran scatterplot was used as an exploratory spatial data analysis technique (ESDA) to discover and visualize local patterns of spatial association (local spatial clustering or local spatial outlier), different forms of local spatial instability (non-stationarity), and spatial heterogeneity④Potential non-stationarities and local spatial association (hot spots) were identified by means of the local Gi*statistic when the spatial clustering of high values or low values concentrate are in one sub region⑤Local spatial autocorrelation statistics is used in conjunction with GIS to display and visualize local patterns of spatial association, local non-stationarity and spatial heterogeneity. 2. Identification of local clusters and outliers of geochemical associations of elements based on a combination of Mahalanobis distance and local spatial autocorrelation statisticsThe spatial distribution patterns of three geochemical associations of elements (including association of Cu, Au, Mo, Ag, Pb, Zn, As and Sb elements; association of Mo, Sb, Zn, Ag, V and As elements; and association of Sn, W, Sb, Mo, Zn, Ag and As elements) are studied by means of a combination of the robust Mahalanobis distance and local Moran Ii statistic. Firstly, the robust Mahalanobis distances (RMDs) between pairs of multivariate observations in an association of elements are calculated to measure the degree of similarity between the samples, after that the data structure and the distribution of Cu concentration are studied by means of EDA techniques. Local Moran’s Ii statistic is then used to discover local patterns of spatial distribution of calculated robust Mahalanobis distances, to identify local spatial stability/instability, spatial heterogeneity and thus to identify local multivariate spatial clusters and outliers. On the basis of the results of spatial variability analysis, local patterns are also studied at six different spatial scales (2km,4km,6km,8km,10km and12km) for both the positively skewed raw data (the calculated robust Mahalanobis distance) and the Box-Cox transformed data. The results of identifying of local patterns of spatial association are finally visualized using GIS.The main results and conclusions of the dissertation are as follows:1. The area of spatial clustering and spatial outliers detected by local spatial association statistics is compared with anomalous area separation using GIS-based concentration area fractal method. The results show that:①The distribution of local spatial instability of both ore-forming elements and associations of elements is almost independent of spatial scales;②From comparison of identification of local patterns of spatial association for the local G*statistic, the local Moran’s I; detected a larger area for the high-high spatial clusters, but the local Gi*statistic better detected weak spatial association (weak anomalies).③The number of high-high spatial clusters changes when spatial scales change. This obviously strongly affected the results of local spatial pattern identification.④There was no spatial autocorrelation found at a given scale, which does not mean that autocorrelation may not be found at some other scales, thus different spatial scales should be taken into account in spatial weight matrix construction.⑤Comparison of the positively skewed raw data, shows that high spatial clusters cover a much larger area for the Box-Cox transformed data. This area correlates with the objective reality and has a good conformity with known occurrences. Two important conclusions can be made when applying local Moran Ii statistic, which are:①Local high-high spatial clusters and low-high spatial outliers provide significant information in ore prospecting and in the identification of local spatial instability and spatial heterogeneity caused by the ore-forming process;②local low-low spatial clusters and high-low spatial outliers indicate the presence of local spatial stability and local spatial heterogeneity respectively in the study area, but both of them do not play any significant roles in ore prospecting.2. The results of identification of local patterns of geochemical associations of elements based on local Moran show that local spatial instability, clusters and outliers identified for three geochemical associations of elements basically concur with the objective reality and have a good conformity with known ore deposits(occurrences) at different spatial scales, especially for the Box-Cox transformed data. It can be concluded that:①The distribution of local spatial instability of associations of elements is also almost independent of spatial scales;②The distribution of the calculated Mahalanobis distances are strongly skewed. High values strongly affect the results thus requiring a form of data transformation before calculations.③Spatial scales also affect the results of identification of local patterns of spatial association, thus suggesting that different spatial scales should be taken into account.The results not only provide important roles in ore-forming element potential evaluation, but also can promote the study of the geochemical data processing. The innovations mainly reflect in:1. A new method for identification of local spatial clusters and outliers of regional geochemical ore-forming uni-elements by means of a combination of spatial variability analysis and local Moran’s Ii and Gi*statistics was proposed.On the basis of spatial variability analysis, the method does not only takes the spatial distribution into account, but also tests of significance for local Ii and G*statistics. The results show that high-high spatial clusters and high-low spatial outliers correlate with the objective reality and have a good conformity with known deposits (occurrences).2. A new method for identification of local spatial clusters and outliers of regional geochemical associations of elements by means of a combination of robust Mahalanobis distance, spatial variability analysis and local Moran’s Ii statistic was proposed.The robust Mahalanobis distances (RMDs) between samples in an association of elements are considered. On the basis of spatial variability analysis, the method not only takes the spatial distribution into account, but also tests of significance for local Moran Ii statistic. The results show that high-high spatial clusters and high-low spatial outliers correlate with the objective reality and have a good conformity with known deposits (occurrences).