You are on page 1of 8

Geostatistical estimation of mineral

resources with soft geological
boundaries: a comparative study
by J.M. Ortiz* and X. Emery*

understanding the characteristics of the
available data and at identifying homogeneous
geological domains within the deposit,
Synopsis according to the spatial continuity of grades
Mineral resource evaluation requires defining geological domains and to geological features such as lithology,
that differentiate the types of mineralogy, alteration and lithology. mineralogy and alteration.1–4
Usual practice is to consider the domain boundaries as hard, i.e. Three problems arise during this process:
data from across the boundaries are disregarded when estimating ➤ The definition of geological domains
the grades within a given domain. This practice may hinder the relies on the subjective interpretation of
quality of the estimates when a significant spatial correlation of the
the mining geologist and on his
grades exists across the domain boundaries.
understanding of the genetic processes
In this paper, several geostatistical methodologies to handle soft
geological boundaries are compared through a case study on a that caused the mineralization. Various
copper mine in Chile. The estimation methods considered are: interpretations are therefore possible.
ordinary kriging using hard boundaries between geological domains; ➤ The delineation of the geological
ordinary kriging omitting the geological boundaries; traditional and domains is always subject to errors,
standardized ordinary cokriging of the grades assayed in different since only fragmentary information is
domains; ordinary kriging within dilated geological domains, that is, available through a finite set of samples
incorporating samples from adjacent domains up to a given radius drilled in the deposit. Delineating the
from the boundary of the domain being considered. The estimations domains must be done carefully,
are performed using a set of exploration data (drill hole samples)
accounting for geological knowledge
and are validated using a set of production data (blast hole samples
about the deposit genesis. Alternatively,
collected for grade control).
Results indicate that kriging with dilated domains performs
one could consider modelling the
better than ordinary kriging using hard boundaries or no boundaries uncertainty in the boundaries layout by
and better than the ordinary cokriging approaches. It therefore resorting to a geostatistical simulation
appears as a simple alternative to global kriging (without technique.5,6
considering the geological domains) and allows accounting for ➤ The boundaries that define the contact
changes in the grade average, dispersion and spatial continuity with between adjacent geological domains are
the geological characteristics of the deposit, via the use of grade seldom ‘hard’, particularly for porphyry
variograms proper to each geological domain. The maximal distance copper mineralization, that is, the grades
for searching samples from adjacent domains should be chosen measured at either side of a boundary
according to the spatial correlation of the grades across the domain
are not independent. Besides, the
boundaries, or via a cross-validation at an early stage of the mine or
boundary may be defined by a change in
a jack-knife if production data are available.
Keywords: Kriging; cokriging; soft boundaries; hard boundaries; the local mean grade, which is usually
geological control; geological domains. gradational rather than abrupt.
Normal practice is to estimate the grades
and to assess the mineral resources within
each geological domain independently.7,8 This
approach implies that:

Introduction
Every mining project requires defining the
mineral resources prior to mine design and
planning. This definition is based on
geological knowledge of the orebody and on
* Department of Mining Engineering, University of
sample information from an exploration
Chile, Santiago, Chile.
drilling grid, usually with infill drilling in the
areas to be extracted during the first years of
the project. The first step in resources
estimation is an exploratory analysis aimed at

grades. Accordingly. For this reason. a geological domain. such as faulting and variogram calculated exclusively from the samples displacements. known in geostatistical applications as the single domain and incorporating auxiliary variables that pseudo-cross variogram14. however. X is a random field representing the grade within a spatial correlation of the grades across the boundary geological domain and Y is a random field for the grade (hard boundary) within a different domain. petrophysical properties of adjacent domains frequently ➤ Case 2—Ordinary kriging of block grades is performed translate into a hard boundary. if so. To address this problem. making it difficult to identify the exact the block grade in a domain is calculated using only layout of the limit. if the porosity disregarding all boundaries and geological information. the question is how to incorporate information from different geological domains. which are fitted by using a such as ordinary kriging. are. so far. a soft boundary shows a transition zone between geological domain by using hard boundaries. of the rock to the mineral bearing fluids is heterogeneous The grade variogram is inferred from all the samples. Methodology Reference cases Hard or soft boundaries? In order to assess the quality of the geostatistical methods A contact between two geological domains. such an approach has not seen 1 a wide application in the mining industry. since both random fields X domains.13 The assumes that the mean grades are the same on both latter is defined as: sides of the domain boundaries. mineralogy or lithology point of view. On the ➤ Case 1—Ordinary kriging is performed within each contrary. The tools usually considered to quantify the spatial ➤ Case 4—Standardized ordinary cokriging of the block correlation are the direct variogram and. that is. they do not show a significant change in the known at the same location. In the following. from one domain to another. the two domains. abruptly.12 variables are considered (see Appendix). Given that the geological domains ➤ grades are assumed stationary within each domain. the grades within (Appendix). is cases are considered: considered hard if there is an abrupt change in mineralogy or grade without a transition at the scale of observation. This cokriging calculated for every location where the grade is being type uses as many unbiasedness constraints as estimated. In many cases. which can be calculated even codify the geological characteristics into the grade estimation when there are no matching samples between X and Y: process. the following approaches are considered: and the other.8.17 This cokriging type implicitly different geological domains). from an that will be proposed for geological control. This case represents the common between geological domains. the grade is likely to change irrespective of the geological domains they belong to. This trending behaviour translates into a correlation between data located on each side of a boundary ➤ Case 3—Traditional ordinary cokriging of the block (generally inferred by an expert geological interpretation). The possible trend in is based on the direct and pseudo-cross variograms of the grades is handled by using robust estimation techniques the different variables. cannot be inferred from the data. . there is no vector h. up to which distance the data across the boundary should be considered for grade estimation in the domain of interest. i.e.15. when dealing with grades.Geostatistical estimation of mineral resources with soft geological boundaries ➤ grades within a geological domain are not influenced where u and u + h are two locations separated by the lag by information across the boundary. which uses a single unbiasedness constraint multiple variables (in this case study. expressed in the same unit. considering the grade within each geological which can be used to improve the estimation of mineral domain as a separate variable16.11 the drill hole samples from the same domain and a Structural control of grades.10 However. the cross variogram. the cross variogram local mean as one gets closer to the boundary.7. if a soft boundary element of interest). two reference alteration. will normally preclude a smooth transition within this domain. Abrupt changes in the practice in the mining industry. several geostatistical visual examination of their plots indicates whether a approaches to geological control are compared through a case significant correlation in the grades across the boundaries study from a copper mine in Chile. several authors suggested considering a another tool.9. we use Alternatively. by definition. changes in the grade behaviour Accounting for information across the boundaries are not so abrupt and one observes a transition zone where To assess the relevance of using data from different the local mean grades vary smoothly between one domain geological domains. one must know when a and Y refer to the same physical variable (the grade of the boundary can be called soft or hard and. X and Y are never simultaneously that is. The cokriging system resources near the domain boundaries. at least at the scale of the cokriging neighbourhood. in which a local mean is implicitly linear model of coregionalization8. within exists. exists or not and. an assumption that 1 seems reasonable when considering soft geological γ XY (h) = E{[ X (u + h) – X (u)] ⋅ [Y (u + h) – Y (u)]} 2 boundaries. γ ∗XY (h) = 2 { E [ X (u + h) – Y (u)] 2 } This paper focuses on the problem of grade estimation in the presence of soft or hard boundaries between geological This formula is meaningful. across this boundary to estimate the grades in a particular Once all the pseudo-cross variograms are calculated.13. In particular. disjoint.

where information from across the these samples. since mineralogy and alteration are not relevant factors for the definition of a within a dilated neighbourhood of the domain volume.Geostatistical estimation of mineral resources with soft geological boundaries T ➤ Case 5—Ordinary kriging within each geological Each composite is assigned a geological domain. molybdenite. these improvements may be counterbalanced by southern areas of the deposit. for the validation of the proposed method- boundary is considered only up to a finite distance. Although n incorporated in the kriging system (provided that they the rock codes were originally six. since southern parts of the deposit. exponential nested structures.05 per cent). holes (1. this situation might change if the grade transitions granodiorite clasts with cement composed by o are smooth near the boundaries and the kriging search tourmaline and sulphides such as chalcopyrite. However. based on the closest of a grade variogram for each geological domain provides an neighbour sample from the drill hole database. pyrite. related to the main alteration-mineralization event. Its emplacement is n parameters (number of samples and search radii) are appropriately defined. The data correspond to 2 376 composites from an exploration diamond-drill hole campaign. for which a single (global) variogram Blast hole samples show a higher mean grade than drill model is used. Although the original rock code was not logged on between Cases 1 and 2. but this is due to their central location where copper grades are higher. Further improvements are expected with Case 3 and This rock has the highest mean grade (about 1. akin to Case 1. The use ologies. This breccia has i However. production data are of coregionalization8. (b) Location map of blast hole samples coded by copper grade for a particular bench ▲ . and some bornite. Case 5 represents an intermediate situation in the mine. ➤ Tourmaline breccia (code 2).20 per cent versus 1. These correspond degrade the quality of grade estimates. advantage over Case 2. t the latter does not account for the geological information. Their emplacement is e simultaneous or more recent than the intrusion of r the greater difficulty in fitting simultaneously the grade tourmaline breccia. namely: a to the domain of interest. one could expect Case 1 to outperform Case 2. This is the host rock where breccias intruded. since the incorporation of secondary information (via cent) and is centrally located in the deposit. It is located in the eastern and c Prior to applying the five methodologies to the case study. to 12 793 samples taken at blast holes over 13 benches of Finally. a the use of covariates) reduces the theoretical estimation ➤ Other breccias (code 3). boundaries. but considering all the data this case relates exclusively to lithology. spaced approximately 40 m in the horizontal plane. All grid. these variograms are modelled with a nugget effect and two completed by infill drills in the central part of the deposit.13. which corresponds to the bench height in the When considering the cokriging approaches (Cases 3 and 4) mine. Part of the drill holes are located on a relatively regular the direct and pseudo-cross variograms are required. This allows data from adjacent domains to be geologically and statistically consistent domains. The Case study: porphyry copper deposit sampling procedures for drill holes and blast holes have been Presentation of the data subject to a careful quality assurance and quality control programme to ensure that there is no systematic bias and Data from a porphyry copper deposit in Chile are now used to that the variance of the errors is within accepted industry compare the performance of the aforementioned approaches ranges. a rock code has been assigned.20 per P Case 4. The summary statistics from all the data sets are to resource estimation in the presence of soft geological shown in Table I. A poor fit of these variograms may available from several benches (Figure 1b). they were grouped into s are close enough to the boundary) as if they belonged three main types (Figure 1a). ➤ Granodiorite (code 1). relocating and diluting it. as shown in Figure 2 and Figure 1—(a) Location map of drill hole samples coded by rock type. They are composed by three different breccia types and outcrop in the western and p variance and therefore provides more precise results. variograms and pseudo-cross variograms with a linear model In addition to the drill hole database. which in r domain. The composites Variogram analysis and modelling are 12 m long.

480 7.604 0. Therefore. .769 0.583 0.110 0.515 0. since there is never a distance of 100 m.578 0. at a lag absent from the cokriging systems.400 0.650 Tourmaline breccia 1 635 1.620 0. Also.552 0.517 0.539 0.709 0.400 Tourmaline breccia 10 028 1.120 0. considered in the cases when the pseudo-cross variograms butions to the sill for the different nested structures are are not used (Cases 1 and 5).625 0.054 0.710 Granodiorite 1531 0.710 Blast holes All data 12 793 1.790 1.423 0.160 0. while the models are presented as lines without dots.990 3. The horizontal variograms are represented by dashed lines and the vertical variograms by solid lines summarized in Table II.445 0.120 0. Sample variograms are shown as lines with dots at the calculated lag distances. which indicates that the correlation becomes very small.710 Other breccias 1 414 0. that is. Min Lower quartile Median Upper quartile Max data (%Cu) (%Cu) of Var.080 1. variograms.783 1.482 0.160 1.620 0.661 0.200 0.200 Figure 2—Direct and pseudo-cross variograms of the copper grade for the different geological domains fitted with a linear model of coregionalization. this distance is chosen as the maximal primary variable is estimated (proof in Appendix).550 7.670 0.991 0.110 0.870 1. search radius to use in Case 5.500 0. Coef. the linear model of coregionalization is relevant within a given geological domain incorporates data from only in the structured portion of the pseudo-cross adjacent domains.645 0.560 0. in which the grade estimation Accordingly.090 1.050 4.730 1.240 Granodiorite 354 0.825 0. Number of Mean Std.920 3. dev.240 Other breccias 387 0. (%Cu) (%Cu) (%Cu) (%Cu) (%Cu) Drill holes All data 2 376 1.140 0.612 0.410 0.196 0.621 0.600 0.310 4. The mathematical consistency of the The direct variograms obtained from the joint fitting are model is checked by verifying that the matrices of contri.870 1. all the pseudo-cross variograms are close collocated secondary sample at the location where the to their sills.170 0.940 1.287 0.330 7.160 0. it should be noted that positive definite.450 7.Geostatistical estimation of mineral resources with soft geological boundaries Table I Basic statistics from drill hole and blast hole data sets. the eigenvalues of the matrices of the nugget components of the pseudo-cross variograms are sill contributions are positive.13 One observes that.

090 0. based on domain. The scatter diagrams contribution of 0. which in reality are but an interpretation from the sample information and are inevitably subject to error. the mean error. The deposit. statistical comparisons are made. horizontal range of 50 m beyond a visual check of the block grade models. The resulting model has a nugget effect of 1 and 3.006 0.100 0. most importantly. One These diagrams indicate that none of the proposed observes that the overall sill is higher than the data variance approaches suffers from a significant conditional bias. one observes that application of hard boundaries to the estimation of mineral resources (Case 1) imposes artefacts that are highly dependent on the geological model and on the layout of the boundaries. According to this table.128 Domain 1 Domain 3 0. The resulting maps (Figure 4) gradational transitions in the mean copper grade. as seen in Cases sample variogram.034 Domain 2 Domain 3 0. the grade estimates are significantly smoother near the boundaries than in the case of traditional ordinary cokriging (Case 3).100 0. If no geological control is accounted for (Case 2). as the (Table I). the map Figure 3—Copper grade variogram disregarding geological boundaries. using the information from the drill hole data set and under study are not ‘hard’ and are associated with an assumed rock type model.056 Case 2 requires the knowledge of the grade variogram as data from adjacent domains to be considered (Case 5) also if no geological boundaries existed. or to the relatively small size fairly close to the first bisector line.110 Domain 1 Domain 2 0.104 0. which produces discontinuities in the grade estimates near the domain boundaries. the ordinary kriging approach with Estimation and validation using jack-knife hard boundaries (Case 1) provides the worst results (highest The prediction of the grade values for blocks over a regular mean absolute and mean squared errors). when the standardized ordinary cokriging is considered (Case 4).340 0.240 0. The the geological boundaries. plane and 180 m in the vertical direction (Figure 3). Regarding the other methods.220 0. the first one with To assess the performance of the different methodologies a contribution to the sill of 0.200 Domain 3 (Other breccias) 0. The same effect can be observed when differentiating the geological domains and incorporating data from across the boundaries as covariates (cokriging approaches). traditional ordinary cokriging assumes that the (unknown) mean grades of the geological domains are not the same. except show clear differences. particularly in the areas where copper maybe the boundary between domain 2 (tourmaline breccia) grades are extrapolated and. This may be due to the effect of combining all the regressions of the true grades upon the estimated grades are domains into a single one. The impact of this higher sill on kriging estimates is statistics such as the correlation between the estimated and not relevant. This stems from the fact that standardized cokriging uses the same local mean at both sides of the boundary and therefore assumes grade continuity. In contrast. This is explained grid with a mesh of 10 m x 10 m x 12 m is done for the five by the fact that the geological boundaries in the deposit cases.200. since kriging weights are independent of the true sample values. The results are presented in Table III . Incorporating the geology by horizontal variograms are represented by dashed lines and the vertical modelling a grade variogram in each domain. (with respect to the correlation range) of the sampled In addition. mean squared error.109 0.02 and two nested exponential structures.Geostatistical estimation of mineral resources with soft geological boundaries Table II Parameters of the linear model of coregionalization fitted to the sample direct and pseudo-cross variograms of the copper grades Head domain Tail domain Nugget effect Exponential model sill Exponential model sill variance Horizontal range = 70 m Vertical range = 100 m Horizontal range = 100 m Vertical range = 160 m Domain 1 (Granodiorite) 0. All drill hole data are frees the resulting estimation map of the strong dependence then pooled together for the calculation and fitting of the upon the assumed layout of the boundaries.005 0.37 and ranges of 70 m in the horizontal between true and estimated grades are shown in Figure 5 . near the and domain 3 (other breccias) in the western part of the assumed boundaries of the geological domains. and the second one with a sill is used against the blast hole data set.100 Domain 2 (Tourmaline breccia) 0.006 0. mean absolute error and scaling of the variogram. but allowing variograms by solid lines ▲ .095 0. jack-knife and vertical range of 90 m. The sample variogram is shown as lines with dots at the calculated lag looks smooth and free of artefacts related to the definition of distances and the model is presented as lines without dots. However. 0.

24450. East (m) 24850. 25100. Case 2 Case 3 25600. 24450. North (m) 25100.0 North (m) North (m) Domain 2 0. 24450. 25600.0 25100. Cu (%) 1. 24450. 24450.5 Domain 3 1. 24450. East (m) 24850. East (m) 24850. East (m) 24850. North (m) North (m) 25100. 25600.5 Domain 1 0. Case 4 Case 5 25600. East (m) 24850. North (m) 25100. 25600.Geostatistical estimation of mineral resources with soft geological boundaries Rock type model Case 1 25600. Figure 4—Plan views of the geological model showing the three geological domains and of the estimated grades obtained with all five methodologies considered Figure 5—Scatter plots comparing the actual blast hole grades from the production data set with the grades estimated by all five methods studied . 25100. East (m) 24850.

C.. Accordingly. Applied mineral inventory estimation.219 0. that the effort needed to perform a cokriging approach brings little improvement over simpler kriging methods. pp. should support the definition of hard boundaries for grade Bettini. and BLACKWELL.H. alternatively. P.317 Mean squared error 0.224 0. A. This can s consistent multivariate variogram model. (ed. it uses a variogram model proper to each geological correlation remains significant. Mineral Resource and Ore Reserve have. A. 381.050 -0.379 0. The maximal distance for searching samples from variables is scarce and the results are still hard to validate.377 0.324 0. resulting in a ordinary cokriging (Cases 3 and 4). tation must be included in the estimation of mineral Melbourne.049 -0.320 0.355 0. CHAMPIGNY. SINCLAIR. unlike global kriging (Case 2).. and MORAGA.. Consequently.242 . Galli.220 0.485 0. as it reduces the mean absolute and mean squared errors with respect to domain.323 0. resources. The choice of the distance to which data from across t the other approaches and improves the correlation coefficient a boundary are used can be determined by examining the i between estimated and true grades.281 0.257 0. This distance can also This research was funded by the National Fund for Science be adapted to the boundary type and be smaller when the and Technology of Chile (FONDECYT) and is part of project grade transitions are sharper. and HANNA. 2002. J. A reason could be the smooth grade transitions from one geological domain to difficulty in representing the spatial correlation of the grades another. be done in a rather simple fashion. C. A. 2001. estimation.. p adjacent domains should be chosen on the basis of the spatial correlation of the grades across geological domains Acknowledgements e (analysis of the pseudo-cross variograms) or via a cross. DUKE. the mineralization is Kluwer Academic. N. G.287 0. necessity of a linear model of coregionalization for cokriging Cambridge. Cambridge University Press. Table III Performance comparison among the five cases studied.246 0. CUADRA.). pp. N. number 1040690. FERRAZ. and Remacre.J. 2002.348 Mean squared error 0. the geological information and interpre- Estimation—The AusIMM Guide to Good Practice.Geostatistical estimation of mineral resources with soft geological boundaries T ordinary kriging disregarding the boundaries (Case 2) or disseminated and the metal is relocated many times by r ‘dilating’ the geological domains (Case 5) outperforms different physical and chemical processes..225 0. 163–176.657 0.543 0. Dordrecht.351 0.J. The case study presented in this paper indicates pp. especially in the areas pseudo-cross variogram (or. ration of geological controls into reserve evaluation: recent examples from It is reasonable to expect that a clear geological reason giant copper mines in Chile.560 0.215 Validation subset Correlation 0.054 -0. the use of the information across the n within and across the geological domains through a boundaries of the geological model is often relevant. Conclusions References The definition of geological domains is highly relevant due to 1. P.670 (12 793 blast holes) True-estimated Mean error -0. (eds. Armstrong. often makes the fitting of the sample direct and cross variograms poorer (and harder) and tends to deteriorate the 3. The Australasian Institute of Mining and Metallurgy...079 Mean absolute error 0. Another reason to go for domain and can therefore account for changes in the grade simpler techniques is that current software for performing P spatial continuity with the geological characteristics of the geostatistical analysis and cokriging with more than two a deposit. This method is plots) between grades sampled on each side of the boundary and by determining the maximal distance for which the n quite simple to implement and.042 -0.081 -0.H. Champigny.102 -0. r validation or a jack-knife procedure. The statistics are given for the entire validation set (12 793 blast holes) and for a subset of it (2 248 blast holes located at less than 25 metres from a geological boundary) Case 1 Case 2 Case 3 Case 4 Case 5 (OK with hard (OK without) (Traditional OCK) (Standardized OCK) (OK with dilated boundaries) boundaries) domains) Validation set Correlation 0. by allowing some samples a On the whole. Edwards.320 0. Successful incorpo- final results. G.662 0.563 (2 248 blast holes close to True-estimated the geological boundaries) Mean error -0. M. o close to the geological boundaries (Table III). the lagged scatter. A.103 -0.655 0. 147–156.500 0.110 -0. Geological Interpretation for Resource the recoveries that ores with distinct geological properties Modelling and Estimation. In many mineral deposits. the use of dilated geological domains of adjacent domains (up to a certain distance from the boundaries) to be used when estimating the grade in a given c (Case 5) gives the most precise grade estimates. A. The 2.665 0.). Geostatistics Rio 2000.044 Mean absolute error 0.

GOOVAERTS. 2003.17 P np multiple rock types using a linear model of coregionalization for soft boundaries. Melbourne. ∑∑λ OCK std i. the cokriging -th domain is defined by a weighted average of all the systems and the resulting cokriging weights are the same available data:8 if one adds nugget effects to the cross-covariances. N. A. The single unbiasedness constraint implicitly assumes that all the 13. linear equations. A. R. Multivariate Spatial Prediction. 2005.. p ( u) = 1 of Mining and Metallurgy. ZHU. A. Pseudo-cross variograms.. H. pp. A. vol. P. pp.. vol.. since there is no matching domain and {upi. Transactions ∑∑λ OCK trad i. and CRESSIE. different geological domains. ( ) Cp0 q u. X.). estimator is the standardized version of it (Case 4) that 11. Berlin. the domain. and least. This P np statement justifies why ordinary cokriging can be performed i . which implies that some secondary data 5.. p ∀p = 1. Edwards. (eds. and JOURNEL... P. G.. 23. D. ◆ . Geological and structural control in kriging. STANDING. nq 10. 1990. 715–739. Accordingly. If one assumes that location u does not coincide with a sample. and Barahona. and ROTH.. the standardized ordinary cokriging Mathematical Geology..A. Henceforth..C.. P 0 [A2] 6. F. 2. Geostatistics Tróia An alternative to the traditional ordinary cokriging ’92 . P. p (u) ⋅ Cpq (uip . C. EMERY. 561.M. The weights of the primary variable pp. no.. p =1 i =1 [A3] pp.. ALABERT..Geostatistical estimation of mineral resources with soft geological boundaries 4. LARRONDO. MYERS. 1998. A. random fields {Zp. available data:7.A. F.8 Mineral Resource and Ore Reserve Estimation—The AusIMM Guide to P np OCK std [A4] Good Practice. 775–790. 1989.E. Journal of the np South African Institute of Mining and Metallurgy. JOURNEL. D.e. no.. 21–42.G. while for each secondary variable the weights addup to 0. pp. no. 1993. hence this type of ordinary cokriging should be Engineers. BYWATER. A. it is seen The grade estimation by ordinary cokriging (Cases 3 and 4) that the cross-covariance terms are never used at the zero uses data from P different geological domains. nq 17. Cobar. 257–268. u qj ) + µq = of the Society for Mining. p ( u) ⋅ Z p ( u i ) Z pOCK (u) = ∑ ∑ λOCK p [A1] without specifying the nugget effects of the cross- o p =1 i =1 structures. The Australasian Institute ∑∑λ p =1 i =1 i.. 923–935. u qj ) + µ = p =1 i =1 [A5] Magri. (ed.. 100. pp. Metallurgy and Exploration. Oxford University Press.13. Estimation of mineral resources using grade In the case of traditional ordinary cokriging (Case 3). Gecamin Ltda. and ORTIZ. In systems [A3] and [A5]. 286. ISAAKS.. Santiago. pp. domains: critical analysis and a suggested methodology. J. Henríquez. 219–240. pp. vol. Heterogeneity in a Complex Turbiditic Such constraints amount to hypothesizing that the Reservoir: Stochastic Modelling of Facies and Petrophysical Variability. expected value.. i = 1 . An Introduction to applied geostatistics. Soares. Multivariate geostatistics: an introduction with random fields {Zp. ∀j = 1. P. SPE Paper Number 20604. ELLIOT. 805–816. Pp.. Indicator Conditioned Estimator. vol. 2000.. primary variable at unsampled locations. µP are Oxford University Press. Springer. the left-hand side members vol. 25. 247–255.) stands for the cross- 8. p (u) ⋅ Cpq (uip . Dordrecht. P} have the same unknown applications. 6. BETZHOLD. p = 1. imposes a single unbiasedness constraint over all the Reconciliation of the McKinnons Gold Deposit. secondary samples are more relevant (their 12. Neither are they for Zp be the random field representing the grade in the p-th the left-hand side member. Society of Petroleum means. Lagrange multipliers:8. positive-definiteness. 1. vol.. Ortiz.. i. J. Apart 15. u qj ∀q = 1. Such an assumption is consistent with the idea of soft geological boundaries. Characterizing the mineralogical variability of a have negative weights: Chilean copper deposit using plurigaussian simulations. no. subject to constraints (A2) leads to the following system of New York. 2001. E. Chile. u qj ∀q = 1.. 1880–1886.. 2. H. C. Mathematical Geology. 1991. The minimization of the error variance 7. J.M. in which Cpq(. ∀j = 1. weights are determined by the following system of linear 16. The grade estimate at location u belonging to the p0 geological domains are disjoint). pp... MININ 2004 International Conference on Mining Innovation. OCK trad pp. GOOVAERTS. P.V.. p = 1. VER HOEF. DOWD. C. P. Kluwer Academic. contain the covariances between pairs of data. 1990.. at the scale of the cokriging neighbourhood) in the cokriging.G.. O.. 1997. 30.. P. Vera. and DEUTSCH. and ROSSI.. while the right-hand side members account for the covariances Appendix between the data and the unknown grade. J. New York. ( ) Cp0 q u.. from Equation [A4]. 1993.C. M. p (u) = δ p . Journal of the there are as many unbiasedness constraints as the number of South African Institute of Mining and Metallurgy. covariance function between Zp and Zq. 387. M.17 P np 9. Knights.V. When do we need a trend model in kriging?. variables considered. and RYBA.. LEUANGTHONG.. 480.. C.. Geostatistics for natural resources evaluation.M.H.E. Mathematical Geology. In this case.M. while µ1. 111–120. Ordinary Cokriging Revisited. 105. that the mean grades are the same (at 14. Grade estimation in equations:8. New South Wales. G. 187–196. no. let distance for the right-hand side member.. 21.. E. P} have different unknown 65th Annual Technical Conference and Exhibition..). S. 7. applied only in case of abrupt grade transitions near the geological boundaries.). average contribution is not zero) when estimating the Mathematical Geology. must addup to one. no. 4. 2004. pp. vol. and MASSONNAT.. pp. ∑λ i =1 i. 1989. SNOWDEN. and SRIVASTAVA.. np} be the sample locations in this samples between two different variables (by construction. WACKERNAGEL. pp. J. (ed.