Geostatistical simulation of rock physical and geochemical properties with spatial filtering and its application to predictive geological mapping
Graphical abstract
Introduction
Understanding geological heterogeneity and predicting the rock properties in the subsurface is a challenging issue in all stages of the exploration of ore deposits, from the identification of prospective areas and/or geochemical anomalies to the modeling of the mineral resources and ore reserves (Delavar et al., 2012; Talebi et al., 2015, Talebi et al., 2016; Afzal et al., 2012, Afzal et al., 2016; Hajsadeghi et al., 2016; Daneshvar Saein, 2017; Wang et al., 2019). Such tasks rely on site-specific knowledge combined with geological and geochemical sampling information from geophysical surveys, core sample logs and assays, which allow the construction of an interpretive geological model, i.e., a 3D representation of the extent of rock type, alteration and/or mineralization domains in the subsurface (Duke and Hanna, 2001; Sinclair and Blackwell, 2002; Knödel et al., 2007; Marjoribanks, 2010; Charifo et al., 2013; Rossi and Deutsch, 2014; Chanderman et al., 2017; Séguret and Emery, 2019; Hajsadeghi et al., 2020; Jin et al., 2020; Mirzaei et al., 2020).
Most studies on constructing interpretive geological models focus on integrating different sources of information for building more accurate models, graphical analyses using related developed software packages to provide a better image of the subsurface and available data or using data inversion methods (Guillen et al., 2008; Lelièvre, 2009). Also, some studies present machine learning or geostatistical clustering algorithms for constructing interpretive geological models (Fouedjio et al., 2018). However, in general, a single representation of the subsurface is constructed and no measure of its accuracy is available. In this context, it is of interest to design quantitative methods that allow assessing the accuracy of a geological model and the uncertainty in the categories assigned to each node or block of this model.
In a recent work, Adeli et al. (2018) propose a geostatistical approach aimed at validating an interpretive geological model and at finding the areas of an ore deposit with a high probability of being misinterpreted. The approach relies on calculating, for each block of the geological model, a measure of the consistency between its interpreted category and the quantitative information brought by the sampling of rock physical and geochemical covariates. It includes the geostatistical modeling and joint simulation of these covariates, followed by a decision-tree classification algorithm to convert the simulated covariates into a geological category, for each target block and each realization. Comparing the prior (ignoring the sampling data) and posterior (conditioned to the sampling data) probabilities of categories for each target block allows identifying the blocks that are most likely to be incorrectly interpreted in light of the information from the quantitative covariates.
The present work builds upon the approach by Adeli et al. (2018) and proposes a methodology to improve the accuracy of the constructed geological model. The methodology relies on the existence of spatial cross-correlations between the geological categories used to define the interpretive geological model and quantitative covariates known through sampling (e.g., drill hole assays). The key idea is to simulate a smoothed version of the covariates, through the filtering of the nugget effect or of the spatial components with a small range of correlation that can be associated with measurement errors or noise, and to use the smoothed simulated values for defining the geological categories. It is believed that the spatial components associated with large-scale structures can better explain geological categories than the measured (noisy) variables.
The outline of the paper is as follows. Section 2 presents the theoretical background of the proposal, with emphasis on coregionalization analysis, the decomposition of coregionalized variables into components acting at different spatial scales and the conditional simulation of specific spatial components. The methodology is illustrated on an iron deposit in Section 3. Discussions are drawn in Section 4 and conclusions in Section 5.
Section snippets
Random field representation of coregionalized variables
Consider p quantitative variables z1, …, zp distributed in a region of space, known at a set of sampling locations x1, …, xn and unknown elsewhere. For the sake of better readability, let us arrange these coregionalized variables into a vector z = (z1, …, zp)T. In the geostatistical formalism, z is viewed as a realization of a vector random field Z = (Z1, …, Zp)T, which opens the door to kriging techniques aimed at predicting z and simulation techniques aimed at quantifying the uncertainty in
Presentation
In the following, the proposed methodology will be illustrated on a data set from a diamond drill hole campaign in an iron ore deposit hosted by banded iron formations, the name and location of which are not revealed due to confidentiality reasons. The data set, already presented by Maleki et al. (2016) and Mery et al. (2017), consist of more than 4000 composite drill hole samples with information of one categorical variable (rock type) and seven quantitative variables (Table 1, Table 2).
Discussion
The output of our proposal is a set of realizations of both the quantitative and categorical variables, which allows assessing the uncertainty in these variables, e.g. through probability maps as in Fig. 6, and predicting their values (e.g., by averaging the realizations of the quantitative variables or, for the categorical variable, by considering the most probable class) at any unsampled location or jointly over several locations. It is therefore an alternative to jointly simulating the
Conclusions
Coregionalized variables with a spatial correlation structure represented by a linear model of coregionalization can be decomposed into spatial components acting at different spatial scales. The components associated with the nugget effect or with the small-scale structures of the coregionalization model represent measurement errors and small-scale variations that can be undesirable for mapping the coregionalized variables. Filtering such components is well-established in applications based on
CRediT authorship contribution statement
Amir Adeli: Conceptualization, Methodology, Software, Validation, Investigation, Writing - Original Draft, Writing - Review & Editing.
Xavier Emery: Conceptualization, Methodology, Software, Writing - Original Draft, Writing - Review & Editing.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
The authors are grateful to two anonymous reviewers for their constructive comments and acknowledge the funding of the National Agency for Research and Development of Chile (ANID), through grants AFB180004 PIA CONICYT (AMTC) and CONICYT/FONDECYT/REGULAR/N°1170101.
References (71)
- et al.
A geostatistical approach to measure the consistency between geological logs and quantitative covariates
Ore Geol. Rev.
(2017) - et al.
Application of power-spectrum–volume fractal method for detecting hypogene, supergene enrichment, leached and barren zones in Kahang Cu porphyry deposit, Central Iran
J. Geochem. Explor.
(2012) - et al.
Delineation of supergene enrichment, hypogene and oxidation zones utilizing staged factor analysis and fractal modeling in Takht-e-Gonbad porphyry deposit, SE Iran
J. Geochem. Explor.
(2016) - et al.
A multivariate approach to study the geochemistry of urban topsoil in the city of Tampere, Finland
J. Geochem. Explor.
(2017) - et al.
Managing borehole samples of unequal lengths to construct a high-resolution mining model of mineral grades zoned by geological units
J. Geochem. Explor.
(2013) - et al.
Delineation of mineralization zones using concentration-volume fractal method in Pb-Zn carbonate hosted deposits
J. Geochem. Explor.
(2012) A turning bands program for conditional co-simulation of cross-correlated Gaussian random fields
Comput. Geosci.
(2008)Iterative algorithms for fitting a linear model of coregionalization
Comput. Geosci.
(2010)- et al.
Geostatistics in the presence of geological boundaries: Application to mineral resources modeling
Ore Geol. Rev.
(2019) - et al.
Enhanced coregionalization analysis for simulating vector Gaussian random fields
Comput. Geosci.
(2012)