Factorial kriging of a geochemical dataset for heavy-metal spatial-variability characterization

Characterizing the spatial patterns of variability is a fundamental aspect when investigating what could be the causes behind the spatial spreading of a set of variables. In this paper, a large multivariate dataset from the southeast of Belgium has been analyzed using factorial kriging. The purpose of the study is to explore and retrieve possible scales of spatial variability of heavy metals. This is achieved by decomposing the variance-covariance matrix of the multivariate sample into coregionalization matrices, which are, in turn, decomposed into transformation matrices, which serve to decompose each regionalized variable as a sum of independent factors. Then, factorial cokriging is used to produce maps of the factors explaining most of the variance, which can be compared with maps of the underlying lithology. For the dataset analyzes, this comparison identifies a few point scale concentrations that may reflect anthropogenic contamination, and it also identifies local and regional scale anomalies clearly correlated to the underlying geology and to known mineralizations. The results from this analysis could serve to guide the authorities in identifying those areas which need remediation.     

Estimation of regionalized compositions: A comparison of three methods

A regionalized composition is a random vector function whose components are positive and sum to a constant at every point of the sampling region. Consequently, the components of a regionalized composition are necessarily spatially correlated. This spatial dependence—induced by the constant sum constraint—is a spurious spatial correlation and may lead to misinterpretations of statistical analyses. Furthermore, the cross-covariance matrices of the regionalized composition are singular, as is the coefficient matrix of the cokriging system of equations. Three methods of performing estimation or prediction of a regionalized composition at unsampled points are discussed: (1) the direct approach of estimating each variable separately; (2) the basis method, which is applicable only when a random function is available that can he regarded as the size of the regionalized composition under study; (3) the logratio approach, using the additive-log-ratio transformation proposed by J. Aitchison, which allows statistical analysis of compositional data. We present a brief theoretical review of these three methods and compare them using compositional data from the Lyons West Oil Field in Kansas (USA). It is shown that, although there are no important numerical differences, the direct approach leads to invalid results, whereas the basis method and the additive-log-ratio approach are comparable.     

Cokriging of regionalized compositions

To avoid spurious spatial correlation when analyzing the spatial covariance structure of regionalized compositions, additive-log-ratio transformation can be used. Here, the additive-log-ratio cokriging estimator, derived in a natural way from this transformation, is shown to be invariant under permutation of components of the untransformed regionalized composition. It leads, as expected, to an exact interpolation. As original data, predicted values of the regionalized composition at unknown points add up to the same constant c and lie between 0 and c.     

Multivariate Correlation in the Framework of Support and Spatial Scales of Variability

This paper extends the concept of dispersion variance to the multivariate case where the change of support affects dispersion covariances and the matrix of correlation between attributes. This leads to a concept of correlation between attributes as a function of sample supports and size of the physical domain. Decomposition of dispersion covariances into the spatial scales of variability provides a tool for computing the contribution to variability from different spatial components. Coregionalized dispersion covariances and elementary dispersion variances are defined for each multivariate spatial scale of variability. This allows the computation of dispersion covariances and correlation between attributes without integrating the cross-variograms. A correlation matrix, for a second-order stationary field with point support and infinite domain, converges toward constant correlation coefficients. The regionalized correlation coefficients for each spatial scale of variability, and the cases where the intrinsic correlation hypothesis holds are found independent of support and size of domain. This approach opens possibilities for multivariate geostatistics with data taken at different support. Two numerical examples from soil textural data demonstrate the change of correlation matrix with the size of the domain. In general, correlation between attributes is extended from the classic Pearson correlation coefficient based on independent samples to a most general approach for dependent samples taken with different support in a limited domain.     

Comparison of approaches to spatial estimation in a bivariate context

The problem of estimating a regionalized variable in the presence of other secondary variables is encountered in spatial investigations. Given a context in which the secondary variable is known everywhere (or can be estimated with great precision), different estimation methods are compared: regression, regression with residual simple kriging, kriging, simple kriging with a mean obtained by regression, kriging with an external drift, and cokriging. The study focuses on 19 pairs of regionalized variables from five different datasets representing different domains (geochemical, environmental, geotechnical). The methods are compared by cross-validation using the mean absolute error as criterion. For correlations between the principal and secondary variable under 0.4, similar results are obtained using kriging and cokriging, and these methods are superior slightly to the other approaches in terms of minimizing estimation error. For correlations greater than 0.4, cokriging generally performs better than other methods, with a reduction in mean absolute errors that can reach 46% when there is a high degree of correlation between the variables. Kriging with an external drift or kriging the residuals of a regression (SKR) are almost as precise as cokriging.     

Identifying spatial heterogeneity of coal resource quality in a multilayer coal deposit by multivariate geostatistics

This study presents a new geostatistical approach to characterization of the geometry and quality of a multilayer coal deposit using the data of seam thickness as a geometric property and the contents of ash, sodium, total sulphur, and the heating value as quality properties. A coal deposit in East Kalimantan (Borneo), Indonesia, which has a synclinal geological structure, was chosen as the study site. Semivariogram analysis clarified the strong dependence of heating value on ash content in the top and bottom parts of each seam and the existence of a strong correlation with sodium content over the sub-seams in the same location. The correlations between the geometry and quality of the seams were generally weak. A linear coregionalization model was used to derive the spatial correlation coefficients of two variables at each scale component from the single- and cross-semivariogram matrices. Because the data were correlated spatially in the same seam or over different seams, multivariate techniques (ordinary cokriging and factorial cokriging) were mainly used and the resultant spatial estimates were compared to those derived using a univariate technique (ordinary kriging). A factorial cokriging was effective to decompose the spatial correlation structures with different scales. Another important characteristic was that the sodium content shows distinct segregation: the low zones are concentrated near the boundary of the sedimentary basin, while the high zones are concentrated in the central part. The main component of sodium originates from the abundance of saline water. Therefore, it can be inferred that seawater had stronger effects on the coal depositional process in the central basin than in the border part. The geostatistical modeling results suggest that the thicknesses of all the major seams were controlled by the syncline structure, while the coal qualities chiefly were originated from the coal depositional and diagenetic processes.     

Criteria to Compare Estimation Methods of Regionalized Compositions

The additive logratio (alr) transformation has been used in several case studies to predict regionalized compositions using standard geostatistical estimation methods such as ordinary kriging and ordinary cokriging. It is a simple method that allows application to transformed data all the body of knowledge available for geostatistical analysis of coregionalizations without a constant sum constraint. To compare the performance of methods, it is customary to use a univariate crossvalidation approach based on the leaving-one-out technique to evaluate the performance for each attribute separately. For multivariate observations this approach is difficult to interpret in terms of overall performance. Therefore, we propose using appropriate distances in real space and in the simplex, to improve the crossvalidation approach and, going a step forward, to adapt the concept of stress from multidimensional scaling to obtain a global measure of performance for each method. The Lyons West oil field of Kansas is used to illustrate the impactof using different distances in the performance of ordinary kriging versus ordinary cokriging.     

Regionalized favorability theory for information synthesis in mineral exploration

This paper presents a regionalized method for the estimation of a favorability function through generalization of all relevant variables (explanatory and target) into random functions. The new method allows the use of cross-covariance functions in addition to ordinary covariances for extracting spatial joint information, which is virtually overlooked in the conventional analyses. The optimal weights for a favorability equation are derived from solving a generalized eigen-system established by the maximization of covariances between a favorability function and the principal components of a set of pre-selected target variables. Various correlation coefficients may be computed to assist in interpretation of the favorability estimates. Both favorability functions and correlation coefficients may be estimated for a point or a block. The regionalized favorability theory can be compared to cokriging in that both use the sample-sample covariances to account for the sample-sample relations and the point-sample covariances to account for the point-sample configurations. The new technique is demonstrated on a test case study, which involves the integration of geochemical, airborne-geophysical, and structural data sets for the target selection of hydrothermal gold-silver deposits.     

Mapping Soil Particle Size Fractions Using Compositional Kriging,Cokriging and Additive Log-ratio Cokriging in Two Case Studies

Information on the spatial distribution of soil particle-size fractions (psf) is required for a wide range of applications. Geostatistics is often used to map spatial distribution from point observations; however, for compositional data such as soil psf, conventional multivariate geostatistics are not optimal. Several solutions have been proposed, including compositional kriging and transformation to a composition followed by cokriging. These have been shown to perform differently in different situations, so that there is no procedure to choose an optimal method. To address this, two case studies of soil psf mapping were carried out using compositional kriging, log-ratio cokriging, cokriging, and additive log-ratio cokriging; and the performance of Mahalanobis distance as a criterion for choosing an optimal mapping method was tested. All methods generated very similar results. However, the compositional kriging and cokriging results were slightly more similar to each other than to the other pair, as were log-ratio cokriging and additive log-ratio cokriging. The similar results of the two methods within each pair were due to similarities of the methods themselves, for example, the same variogram models and prediction techniques, and the similar results between the two pairs were due to the mathematical relationship between original and log-ratio transformed data. Mahalanobis distance did not prove to be a good indicator for selecting an optimal method to map soil psf.     

地质统计学及其在第四纪研究中的应用

地质统计学是数学地质领域最为活跃而实用的分支,它是以区域化变量理论为基础,以变异函数为基本工具,研究那些在空间分布上既具有随机性又具有结构性的自然现象的科学。在第四纪研究中的很多特征(变量)均可看成区域化变量进行地质统计学分析。作者在讨论了经典概率论及数理统计方法简单地应用于第四纪研究可能出现的问题后,着重介绍了用于第四纪研究中的若干地质统计学方法及基本理论,同时,对地质统计学方法应用于第四纪研究中的前景进行了分析。     

时空多元协同克立格的理论研究

本文结合地质统计学的最新成果,在空间协同区域化理论的基础上,对时空域中多元信息的协同克立格（ＳＴＣＯＫ）理论进行了较为详细的研究。主要研究内容有：（１）ＳＴＣＯＫ中的互变异函数与互协方差函数;（２）ＳＴＣＯＫ方程组及求解估值权因子的三种方法：①传统普通协同克立格法（ＳＴＴＯＣＯＫ）,②标准普通协同克立格法（ＳＴＳＯＣＯＫ）,③简单协同克立格法（ＳＴＳＣＯＫ）;（３）排列协同克立格（ＳＴＣＯＬＣＯＫ）;（４）指示协同克立格（ＳＴＩＫＣＯＫ）。     

Extending Multivariate Space-Time Geostatistics for Environmental Data Analysis

Environmental studies require multivariate data such as chemical concentrations with space-time coordinates. There are two general conditions related to such data: the existence of correlations among the coregionalized variables and the differences in numbers of data which occur because of insufficient data caused by measurement error or bad weather conditions. This study proposes geostatistical techniques for space-time multivariate modeling that take into consideration these correlations and data absences. These techniques consist of suitable modeling of semivariograms and cross-semivariograms for quantifying correlation structures among multivariables and of extending standardized ordinary cokriging. The tensor product cubic smoothing surface method is used for space-time semivariogram modeling. These methods are applied to the chemical component data of the Ariake Sea, a typical closed sea in southwest Japan. In order to clarify environmental changes in the Ariake Sea, the concentration data of four nutritive salts (NO₂–N, NO₃–N, NH₄–N, and PO₄–P) at 38 stations over 25 years are used as environmental indicators. For each of the kinds of data, there are spaces and times for which there is no data available. The effectiveness of the modeling of space-time semivariograms and the high estimation capability of the extended cokriging are demonstrated by cross-validation. Compared with ordinary kriging for a single variable, multivariate space-time standardized ordinary cokriging can provide a more detailed concentration map of nutritive salts and while elucidating their temporal changes over sparsely spaced data areas. In the space-time models by ordinary kriging, on the other hand, smooth trends are obvious.     

Spatial structure analysis of regionalized compositions

Like compositions in general, regionalized compositions present the problem of spurious spatial correlation. To avoid this problem, this paper uses the additive-logratio transformation of regionalized compositions, following techniques introduced over the last few years for the statistical analysis of compositional data. It leads to an appropriate definition of a spatial covariance structure to describe spatial dependence between regionalized variables subject to constant-sum constraints in the case of weak stationarity. To illustrate stated problems, simulated data are used.     

稳健协同克立格因子分析及其在化探中的应用

稳健协同克立格因子分析集稳健统计、协同克立格和因子分析的优点于一体,可同时研究多个变量不同方向的结构差异及变化性,用该方法研究次生晕指示的结构特征可在一定程度反映原生晕的特征,揭示其深层次信息．对团结沟金矿区次生晕分析表明：在矿产预测中矿床原生晕组合标志并不适用于次生晕及分散流的研究,本方法对大（中）比例尺次生晕、分散流与原生晕的信息转换和关联有独到之处．     

An application of cokriging for estimation of tripartite earthquake response spectra

A tripartite earthquake response spectrum exhibits peak values of acceleration, velocity, and displacement in concert, each as a function of natural frequency of vibration. Because cokriging is a regionalized variable technique for multiple variable estimation, it was employed to jointly estimate peak values of acceleration, velocity, and displacement through a frequency discretization process. The objective of this application was to develop a technique for estimation of response spectra at uninstrumented locations. Because magnitudes of the three response variables differed, these data were normalized prior to cokriging. This process simply involved dividing each type of variable by its maximum value for a given frequency; hence, values of each type of variable ranged between zero and one. This allowed better accuracy in developing cross-variograms. Cokriging proved to be an efficient and accurate technique to use for the estimation of tripartite response spectra.     

Multivariate geostatistical methods of mean annual and seasonal rainfall in southwest Saudi Arabia

A multivariate geostatistics (cokriging) is used for regional analysis and prediction of rainfall throughout the southwest region of Saudi Arabia. Elevation is intruded as a covariant factor in order to bring topographic influences into the methodology. Sixty-three representative weather stations are selected for a 21-year period covering different microclimate conditions. Results show that on an annual basis, there is no significant change using elevation. On the seasonal basis, the cokriging method gave more information about rainfall occurrence values, its accuracy related to the degree of correlation between elevation and rainfall by season. The prediction of spring and winter rainfall was improved owing to the importance of orographic processes, while the summer season was not affected within its monsoon climatology. In addition, fall season shows inverse and weak correlation of elevation with rainfall. Cross-validation and cokriging variances are also used for more accuracy of rainfall regional estimation. Moreover, even though the correlation is not significant, the isohyet values of cokriged estimates provided more information on rainfall changes with elevation. Finally, adding more secondary variables in addition to elevation could improve the accuracy of cokriging estimates.     

Joint Consistent Mapping of High-Dimensional Geochemical Surveys

Geochemical surveys often contain several tens of components, obtained from different horizons and with different analytical techniques. These are used either to obtain elemental concentration maps or to explore links between the variables. The first task involves interpolation, the second task principal component analysis (PCA) or a related technique. Interpolation of all geochemical variables (in wt% or ppm) should guarantee consistent results: At any location, all variables must be positive and sum up to 100 %. This is not ensured by any conventional geostatistical technique. Moreover, the maps should ideally preserve any link present in the data. PCA also presents some problems, derived from the spatial dependence between the observations, and the compositional nature of the data. Log-ratio geostatistical techniques offer a consistent solution to all these problems. Variation-variograms are introduced to capture the spatial dependence structure: These are direct variograms of all possible log ratios of two components. They can be modeled with a function analogous to the linear model of coregionalization (LMC), where for each spatial structure there is an associated variation matrix describing the links between the components. Eigenvalue decompositions of these matrices provide a PCA of that particular spatial scale. The whole data set can then be interpolated by cokriging. Factorial cokriging can also be used to map a certain spatial structure, eventually projected onto those principal components (PCs) of that structure with relevant contribution to the spatial variability. If only one PC is used for a certain structure, the maps obtained represent the spatial variability of a geochemical link between the variables. These procedures and their advantages are illustrated with the horizon C Kola data set, with 25 components and 605 samples covering most of the Kola peninsula (Finland, Norway, Russia).     

Joint simulation of multiple variables with a Markov-type coregionalization model

Many applications are multivariate in character and call for stochastic images of the joint spatial variability of multiple variables conditioned by a prior model of covariances and cross- covariances. This paper presents an algorithm to perform cosimulation of such spatially intercorrelated variables. This new algorithm builds on a Markov-type hypothesis whereby collocated information screens further away data of the same type, allowing cosimulation without the burden of a full cokriging. The proposed algorithm is checked against a synthetic multi-Gaussian reference dataset, then against a multi-Gaussian cosimulation approach using full cokriging. The results indicate that the proposed algorithm perform as well as the full cokriging approach in reproducing the univariate and bivariate statistics of the reference set, yet at less cpu cost.     

Spatial orthogonality of the principal components computed from coregionalized variables

Within the frame of the linear model of coregionalization, this paper sets up equations relating the variogram matrix of the principal components extracted from the variance-covariance matrix to the diagonal variogram matrices of the regionalized factors. The spatial orthogonality of the principal components is investigated in three situations: the intrinsic correlation, two basic structures with independent nugget components, three basic structures with independent nugget components and uncorrelated subsets of variables. Two examples point out that the correlation between the principal components may be nonnegligible at short distances, especially if the correlation structure changes according to the spatial scale considered. For one of the two case studies, an orthogonal varimax rotation of the first principal components is found to greatly reduce the spatial correlation between some of them.     

Spatial and temporal study of nitrate concentration in groundwater by means of coregionalization

 Spatial and temporal behavior of hydrochemical parameters in groundwater can be studied using tools provided by geostatistics. The cross-variogram can be used to measure the spatial increments between observations at two given times as a function of distance (spatial structure). Taking into account the existence of such a spatial structure, two different data sets (sampled at two different times), representing concentrations of the same hydrochemical parameter, can be analyzed by cokriging in order to reduce the uncertainty of the estimation. In particular, if one of the two data sets is a subset of the other (that is, an undersampled set), cokriging allows us to study the spatial distribution of the hydrochemical parameter at that time, while also considering the statistical characteristics of the full data set established at a different time. This paper presents an application of cokriging by using temporal subsets to study the spatial distribution of nitrate concentration in the aquifer of the Lucca Plain, central Italy. Three data sets of nitrate concentration in groundwater were collected during three different periods in 1991. The first set was from 47 wells, but the second and the third are undersampled and represent 28 and 27 wells, respectively. Comparing the result of cokriging with ordinary kriging showed an improvement of the uncertainty in terms of reducing the estimation variance. The application of cokriging to the undersampled data sets reduced the uncertainty in estimating nitrate concentration and at the same time decreased the cost of the field sampling and laboratory analysis. Received: 23 July 1997 · Accepted: 31 March 1998