Lognormal-de Wijsian Geostatistics

for Ore Evaluation

Lognormal-de Wijsian Geostatistics
for Ore Evaluation
D. G. KRIGE

South African
Institute of Mining and Metallurgy
Johannesburg 1981
Published by the South African Institute of Mining and Metallurgy
Kelvin House, 2 Hollard Street, Johannesburg

© South African Institute of Mining and Metallurgy

ISBN: 0-620-03006-2
First published 1978
Second Edition 1981

General Editor: G. S. Baker, D.Phil.

Typeset in the Republic of South Africa by

Northern Cape Printers Ltd., P.O. Box 113, Kimberley 8300
Printed in the Republic of South Africa by
Printpak (Cape) Ltd., Dacres Ave., Epping, Cape
 v

Contents

Pref~ce IV 2.5.2 Outliers in deflection values 29

Introduction V 2.6 The de Wijsian semivariogram 29
2.6.1 Mean of semivariogram within field 29
STATISTICAL CONCEPTS, PROCEDURES AND 2.7 The de Wijsian semivariogram and covariances
APPLICATIONS for different sizes of support and geometric
field 30
1.1 Statistics and probability 1 2.7.1 Variations in size of unit areas (support) 30
1.2 Patterns of variation . 1 2.7.2 Variations in size of field 31
1.3 Brief historical review 1 2.7.3 Variations in sizes of support and field. 32
1.4 Populations and units of measurement 2
1.5 Sampling from populations 3 2.8 The standard data pattern for gold ore reserves 32
1.6 Main distribution parameters 4 2.8.1 The nugget effects 33
1.6.1 Measures of location 4 2.8.2 The population variances 33
1.6.2 Measures of variability 4 2.8.3 The covariances . 33
1.7 The normal frequency distribution 5 2.9 Anisotropic de Wijsian semivariogram and
1.8 The Iognormal frequency distribution 5 covariance patterns 34
1.8.1 Examples 5 2.9.1 Formula for anisotropic semivariogram 34
1.9 The three-parameter lognormal distribution 7 2.9.2 Estimation of the nugget effect 34
1.1 0 Sampling from lognormal populations 8 2.9.3 Form of the anisotropic function f 35
1.11 Lognormal estimators for the mean 12 2.9.4 Estimation of parameters a, b, and e 35
1.11.1 Population variance known a priori 12 2.10 Kriging solutions for normally distributed or
1.11.2 Population variance unknown 12 normalised values 36
1.11.3 Examples 12 2.10.1 Population mean known 36
1.12 Practical considerations for borehole vahiations 13 2.10.2 Population mean unknown or ignored 36
1.13 Selective mining considerations 13 2.11 Lognormal kriging solutions 37
1.14 Quality control of routine underground sampling 13 2.11.1 Population mean known 37
1.15 Quality control of routine mine assaying 14 2.11.2 Solution for mean unknown 37
1.15.1 Example 15 2.11.3 Suboptimum weights 37
1.16 Normal correlation 15 2.12 Practical considerations in applying kriging
1.17 Lognormal correlation 17 procedures 37
1.18 Ore reserve regression curves 18 2.12.1 Size and boundaries of mine sections 38
1.19 Effects of regression on total ore reserves 19 2.12.2 Size of data area 39
1.20 Effective block variance 20
2.12.3 Size of ore block to be valued 39
1.20.1 Grade and tonnage estimates for new
properties 21
1.21 Regression on regional means 22 GEOSTATISTICAL MODELS AND CASE
STUDIES FOR CERTAIN GOLD MINES AND THE
PRACTICAL GEOSTATISTICS FOR THE LOGNORMAL- PRIESKA COPPER MINE
DE WIJSIAN MODEL
3.1 Computer programs 40
2.1 General 23 3.1.1 Gold mines 40
2.2 Geostatistical terms and definitions 24 3.1.2 Prieska 40
2.2.1 Regionalised variable 24 3.2 De Wijsian semivariogram for gold values over a
2.2.2 Population, field 24 large field 41
2.2.3 Support, ore unit 24 3.3 Case studies for sections of the Hartebeestfontein
2.2.4 Point values and regularisation 24 and Loraine gold mines 41
2.2.5 Intrinsic hypothesis - second order 3.3.1 Variances 42
stationarity 24 3.3.2 Semivariogram 42
2.2.6 Semivariogram 24 3.3.3 Practical checks on formulae 43
2.2.7 Average semivariogram value 24 3.3.4 Validity of the model for anisotropy 44
2.2.8 Grading 24 3.3.5 Kriging applications 44
2.2.9 Isotropy, anisotropy 24 3.3.6 Gains from improved procedures for a
2.2.10 Proportional effect 24 low-grade property 46
2.2.11 Linear equivalent 24 3.3.7 Semivariogram on unweighted vs.
2.2.12 Kriging 25 weighted data 46
2.3 The lognormal distribution and de Wijsian
3.4 Example of geostatistical analyses on the Prieska
spatial structure . 25
copper mine 47
2.3.1 The nugget effect 25
3.4.1 Contour surfaces for hanging-wall and
2.3.2 Additivity of variances 26
foot-wall data 47
2.4 Features of the nugget effect, error or extension
3.4.2 Example of a semivariogram analysis for
variances 26
copper grades 49
2.5 The de Wijsian variance - size of area
3.4.3 Metal grades vs. accumulations 49
relationship 27
2.5.1 Block variances for selective mining References 50
estimates 28 Index 51
vi

Preface
In 1962 Dr Danie Krige published a monograph on 'Statistical theory of geostatistics applicable to any regionally correlated
Applications in Mine Valuation' under the aegis of the Insti- variable. What higher acknowledgement could anyone be
tute of Mine Surveyors of South Africa. This publication has paid than to have his name perpetuated in the technical
given yeoman service as a standard text for the teaching of literature of his chosen field? This honour Matheron
mining engineers at the University of the Witwatersrand, and generously bestowed on Dr Krige when he gave the appella-
for others interested or involved in estimating the ore reserves tion kriging to the mathematical procedure for calculating
of a mine or mining prospect. The present monograph is a the best linear unbiased estimator of the grade of an ore
worthy successor which updates further analytical work by block as the weighted mean of surrounding sample values
the author, and correlates it with the theoretical concepts (what Krige originally called the weighted moving average).
developed by Professor G. Matheron of the Centre de The theory developed in this monograph is based essentially
Morphologie Mathematique of Fontainbleau, France. on the application of a de Wijsian spatial model to a log
The significance of this work is primarily due to the fact normally distributed variable. It is of interest to note how
that a mine is very seldom so fortunately placed as to be able closely this model approximates to the observed behaviour of
profitably to mine all the mineralized ore comprising an ore gold values in the South African gold mines, as demonstrated
body. Normally, the distribution of mineralization is such by the numerous practical examples included in the text. In
that the average grade of portions of the ore body is too low addition, Krige indicates the potential for further extensions
to support profitable mining. In these circumstances some of this approach to other ore bodies and other parameters,
form of selective mining must be practised. This requirement for example, the analysis of the undulating foot-wall and
focuses attention on two fundamental problems that can hanging-wall mining limits for the sulphide ore body of the
only be solved effectively by the application of statistical Prieska copper mine.
methods of analysis, namely: The application of geostatistics has now been established
• the need to estimate the 'global percentage payability' on a worldwide basis as a valid method for deriving a statistic-
of the ore body and the average grade of the payable ore, ally sound estimate of mine ore reserves. Development is
that is, to estimate the average grade and total tonnage of the continuously taking place as a result of the efforts of numerous
ore reserves; practitioners and theoreticians. However, the fundamental
• the need to identify the average grade and tonnage of principles established by Krige and Matheron have withstood
individual mining blocks for the purpose of planning selective the test of time. This monograph presents an introduction to
mining and grade control. the subject in the typical clear thinking manner of the
Krige originally developed solutions for both these prob- 'maestro' who has been actively involved in its development
lems by clear-sighted statistical analysis of a carefully assem- from its inception.
bled mass of data derived from continuous sampling records HOWARD M. WELLS, PH.D.,
on the South African gold mines over a long period of time. Professor of Mining Engineering,
These pragmatic procedures were subsequently vindicated by University 0/ Utah,
Professor G. Matheron who evolved the general statistical Salt Lake City.
 vii

Introduction
This monograph is based on the notes prepared for a work- wise obvious choices of, for example, value or distance
shop course for the Australian Mineral Foundation, held in intervals, and would in any case not affect the principles
Adelaide in June 1977. It is intended to provide a brief involved; also, for practical reasons, the unit areas used in
review of the developments in the application of geostatistics geostatistical ore reserve valuations on some of the gold mines
in South African ore valuation, mainly on the gold mines. since the early 1960s have not yet been, and are unlikely to
The applications covered include the earlier statistical be, metricated.
developments which preceded and led to the establishment of This is a revised version of the original monograph pub-
the more 'spatially' orientated statistics now generally known lished in 1978 and incorporates a number of corrections and
as geostatistics. The detailed development of all the relevant improvements.
models and formulae used is not covered; should the reader The permission by Anglo Transvaal Consolidated Invest-
desire a thorough knowledge of these models, this mono- ment Company Limited, and by the other companies con-
graph should be studied together with the listed references. cerned, to publish data and results recorded here, as well as
The metric system has been used throughout except for the opportunities and facilities for research and applications
most of the historical examples of, and references to, earlier generously offered by Anglo Transvaal over a period of some
developments in the field of geostatistics predating the intro- 25 years, are highly appreciated.
duction of metrication in the local mining industry. Metrica- D. G. KRIGE
tion of such examples and references could obscure the other- Johannesburg, 1981