Copyright 2000 by Geostokos (Ecosse) Limited, Scotland. ll rights reser!ed. "#blished sim#ltaneo#sly $orld$ide on C%. Special Request for our Readers &e ha!e a 'a!o#r to ask o' yo#, the readers ( $e $ant yo#r g#idance on $hat changes sho#ld be made in s#bse)#ent editions. &e hope to #pdate this book 'airly 're)#ently as $e are p#blishing this o#rsel!es and ha!e total control o!er $hat it contains, ho$ many copies are printed, etc. *n addition to sending #s )#estions and typos, $e hope that yo# $ill take the time to speci'y $hat ne$ topics $e sho#ld add (or $hat topics sho#ld be e+panded). ,ake s#re Isobel gets a copy o' all s#ch comments since she $ill post all typos and ma-or comments on our errata web page . .hanks to all that help in this e''ort. Practical Geostatistics 2000 is a textbook in Geostatistics which can be used as the basis for undergraduate and Master's level courses or for self-teaching. In an eas-to-read stle! with a "ini"u" of "athe"atics! Practical Geostatistics 2000 continues the traditional of Practical Geostatistics #$%$. &i"ed at non-specialists! PG2000 takes the reader fro" no statistical knowledge through the basic necessar statistical background! inverse distance applications! se"i-variogra" calculation and "odelling to si"ple and ordinar kriging. 'he final chapter gives basic case studies in indicator! universal! lognor"al and rank unifor" kriging. 'he first #0 chapters contain worked exa"ples and exercises for the reader. & separate volu"e of (&nswers to )xercises( will be released in *ctober 2000. 'he +, version contains the book in a hpertext for" plus software and all data sets for exercises and worked exa"ples. -or those who onl bu the book! software and data sets "a be downloaded fro" the .eb CONTENTS Preface Notation #. Introduction Page 1 /. E+pectations 2. .he problem to be sol!ed 0. %ata sets 1. So't$are 2. Why a statistical approach Page 7 /. *n!estigating the sample data 2. ,eas#res o' central tendency 0. ,eas#res o' spread or !ariability 1. Graphical descriptions o' the data 2. 3ther #se'#l descripti!e statistics 4. %iscrete data 5. *nto the #nkno$n 6. &orked e+amples /. Coal pro-ect data, calori'ic !al#es 2. *ron ore e+ample 0. &ol'camp data 1. Scallops, total catch 7. E+ercises 3. Normal (aussian! distri"utions Page 31 /. .he gap bet$een data and pop#lation 2. *s it a 8ormal distrib#tion9 0. Estimating pop#lation parameters /. estimating the pop#lation a!erage 2. estimating the standard de!iation 0. con'idence inter!als 'or standard de!iation 1. con'idence inter!als 'or mean 1. Selection (grade:tonnage) calc#lations 2. S#mmary o' chapter 4. &orked e+amples /. Coal pro-ect, calori'ic !al#es 2. *ron ore e+ample 0. &ol'camp data 1. Scallop data, total catch 5. E+ercises #. $%ognormal distri"utions (and others! Page &7 /. .he lognormal distrib#tion /. estimating the mean o' a lognormal pop#lation 2. con'idence inter!als on the pop#lation mean 2. .he three parameter lognormal 0. Selection (grade:tonnage); calc#lations /. t$o parameter lognormal ((( ree' $idths 2. three parameter lognormal ((( gold grades 1. ,ore comple+ distrib#tions and mi+t#res /. mi+t#res o' 8ormal or lognormal pop#lations 2. &orked e+amples /. Scallops, total ca#ght 2. 3rganic matter in soil 0. Calci#m in limestone 1. Gee!or .in mine, Corn$all 4. E+ercises '. (iscrete distri"utions Page 1)3 /. <e!ie$ o' %iscrete ,oments 2. =erno#lli and =inomial %istrib#tions 0. 8egati!e =inomial and Geometric %istrib#tions 1. "oisson %istrib#tion 2. ,i+t#res o' "oisson %istrib#tions (Compo#nd "oisson) /. 3s$ego >ircon data 2. 3ther e+amples 4. Spatial Considerations 5. Sol!ed "roblems 6. E+ercises &. *ypothesis testing Page 13' /. Single sample tests /. test on sample mean 2. test on sample standard de!iation 2. .$o sample tests /. test on standard de!iations 2. test on means 0. paired sampling 1. test 'or sample distrib#tion 0. &orked e+amples /. ?eights o' st#dents 2. Gee!or tin mine (( de!elopment !ers#s stope 1. E+ercises 7. Relationships Page 1#7 /. Straight line relationships /. )#anti'ying the strength o' the relationship 2. "redicting one !ariable 'rom the other 0. Calori'ic @al#e !ers#s sh Content 1. Calori'ic @al#e !ers#s S#lph#r Content 2. 3ther $orked e+amples /. Gold grade !ers#s ree' $idth 2. Scallops ca#ght 0. pplication ((( ArigeBs <egression E''ect 0. <elationships in!ol!ing more than t$o !ariables /. "redicting S#lph#r 'rom Calori'ic @al#e and sh Content 2. pplication ((( ArigeBs mo!ing a!erage template 0. C#r!ilinear <egression 1. pplication ((( "olynomial .rend S#r'ace nalysis 1. E+ercises +. ,he spatial element Page 1+' /. *ncl#ding location as $ell as !al#e 2. Spatial relationships 0. *n!erse distance estimation 1. &orked e+amples /. Coal pro-ect, calori'ic !al#es 2. *ron ore pro-ect 0. &ol'camp data 1. Scallops ca#ght 2. E+ercises -. ,he semi./ariogram Page 2)7 /. .he e+perimental semi(!ariogram /. *rreg#lar sampling 2. Ca#tionary notes 2. ,odelling o' the semi(!ariogram '#nction /. .he linear model 2. .he generalised linear model 0. .he Spherical model 1. .he e+ponential model 2. Ga#ssian model 4. .he hole e''ect model 5. "addington mi+ model 6. C#dging ho$ $ell the model 'its the data 7. e)#i!alence to co!ariance '#nction /0. the n#gget e''ect 0. &orked e+amples /. Sil!er e+ample 'rom "ractical Geostatistic /757 2. Coal pro-ectD calori'ic !al#es 0. &ol'camp a)#i'er 1. E+ercises 1). 0stimation and 1riging Page 2#7 /. Estimation error /. one sample estimation 2. another single sample 0. t$o sample estimation 1. another t$o sample estimation 2. three sample estimator 2. Choosing the optimal $eights /. three sample estimation 2. the general 'orm 'or the BoptimalB estimator 0. con'idence le!els and degrees o' 'reedom 1. simple kriging 0. 3rdinary kriging /. BoptimalB #nbiassed estimator 2. alternate 'ormD matrices 0. alternate 'ormD co!ariance 1. three sample estimation 1. &orked e+amples /. Coal pro-ect, calori'ic !al#es 2. *ron ore e+ample, ("age72) 0. &ol'camp, resid#als 'rom )#adratic s#r'ace 2. Cross !alidation /. cross cross !alidation 4. E+ercises 11. 2reas and /olumes Page 2-' /. .he impact on the distrib#tion /. *ron ore e+ample, 8ormal e+ample 2. Gee!or .in ,ine, lognormal(ish) e+ample 2. .he impact on kriging /. the #se o' a#+iliary '#nctions 2. *ron ore e+ample, "age 72 0. &ol'camp a)#i'er, )#adratic resid#als 12. 3ther 4riging approaches Page 31' /. Eni!ersal kriging /. &ol'camp a)#i'er 2. Lognormal kriging /. the proportional e''ect 2. the lognormal trans'ormation 0. Gee!or .in ,ine, grades 1. S Gold ,ine 0. *ndicator kriging 1. <ank #ni'orm kriging 2. S#mmary o' chapter 5i"liography Page 33- ,a"les (ata Sets 6nde7 Practical Geostatistics 2000 Isobel Clark and William V Harper
Chapter 1: Introduction 1.1 07pectations =e'ore yo# start reading this book, $e $o#ld like to make it clear e+actly $hat yo# can (and canBt) e+pect 'rom it and $hat $e do (and donBt) e+pect 'rom the reader. .his te+t is based in 26 years o' co#rses ta#ght to mining engineers, geologists, hydrologists, soil scientists, climatologists pl#s the occasional geographer, pattern recognition e+pert, meteorologist, statistician and comp#ter scientist. E!en, on one occasion, an acco#ntant. 3!er those years, $e ha!e endea!o#red to pare a$ay all e+traneo#s mathematics and concentrate on int#iti!e deri!ations $here possible. <eaders interested in rigoro#s mathematical proo's are #rged to stop here and t#rn to the more theoretically based books (c'. <e'erence .e+ts in =ibliography). .his book is not intended to t#rn o#t '#lly 'ledged geostatisticians. *t is intended 'or people $ith problems to be sol!ed $hich can be assisted by a geostatistical approach. .o read this book and bene'it 'rom it yo# need to be 'airly com'ortable $ith basic algebra. .hat is, $ith the notion o' #sing symbols as shorthand 'or longer statements. &e ha!e $orked hard to bring yo# a consistent notation thro#gho#t the book. &here notation is o#t o' o#r control, $e e+plain care'#lly $hat each symbol stands 'or and try not to #se that symbol 'or anything else. .his is not al$ays possible. For e+ample, Student (&illiam Gosset) de!eloped his distrib#tion 'or the mean o' a set o' samples and called it the t distrib#tion. ?erbert Sichel de!eloped an estimator 'or the mean o' a lognormal distrib#tion and called it (s#rprise) t. Calc#l#s ((( di''erentiation and integration ((( is disc#ssed at !ario#s points in the te+t. .he reader is not e+pected to do any calc#l#s (as s#ch) b#t is e+pected to kno$ that the di''erential o' +G is 2+. .he only other complication is the 're)#ent #se o' sim#ltaneo#s e)#ations. &e tend not to #se matri+ algebra in this book b#t $ill gi!e the matri+ 'orm a'ter e+planations ha!e been gi!en in simple algebra. For e+ample, linear regression is easier to #nderstand i' de!eloped $ith algebra, b#t !ery simple to implement in spreadsheets or in packages s#ch as 8at%a"H i' matrices are #sed. *' $e ha!enBt scared yo# o'' yet, be reass#red by the 'act that all the analyses are ill#strated $ith real data sets in '#ll $orked ans$ers. *' yo# ha!e the C%, the data sets are incl#ded along $ith so't$are to reprod#ce the analyses ('or the most part). *' yo# are reading the hard copy, the data sets and so't$are can be do$nloaded 'rom the &eb. .here are e+ercises 'or yo# to try. ns$ers are a!ailable 'or yo# to check yo#r res#lts. ,ost o' these e+ercises ha!e been collected and #sed in classes or e+aminations at Final (Senior) Iear and ,asterBs le!els. *t is o#r o$n '#ndamental regret that this book cannot contain the -okes, anecdotes and sheer fun that $e ha!e on the co#rses. &e do ad!ise yo#, ho$e!er, to keep yo#r sense o' h#mo#r and common sense to the 'ore at all times $hile reading this book. 1.2 ,he pro"lem to "e sol/ed Geostatistics ((( as disc#ssed in this book ((( $as de!eloped in geology and mining. ?o$e!er, the problem $hich it $as de!eloped to tackle is more general than geological applications. .his te+t is intended as a basic introd#ction to statistical and geostatistical analysis o' sample data $hich possesses a location as $ell as at least one obser!ed !al#e. .here is o'ten con'#sion as to the intended ob-ecti!e o' geostatistical techni)#es. &e de'ine them here as t$o'oldD /. to characterise and interpret the beha!io#r o' the e+isting sample dataJ 2. to #se that interpretation to predict likely !al#es at locations $hich ha!e not yet been sampled. .o set the scene 'or the rest o' the book, let #s imagine that there is a (more or less) contin#o#s phenomenon $hich co!ers a st#dy area (or !ol#me). the BrealB phenomenon the a!ailable sample in'ormation Some samples ha!e been taken o!er the st#dy area and their locations noted. ,eas#rements ha!e been made on the samples taken. 3#r ma-or task is to estimate the likely !al#e at a location $hich has not been sampled. .here are many di''erent $ays to tackle this problem. .his book co!ers -#st one approach $hich is based on a $ell de'ined set o' ass#mptions. 3ther ass#mptions lead to other methods. lot o' the criticism $hich is le!elled at geostatistical estimation is 'o#nded on misconceptions abo#t the capabilities and intentions o' the method (c'. section Sceptics in =ibliography). &e $ill tackle those as $e come to them in the te+t. &e $ill also disc#ss the shortcomings o' the techni)#es $hich $ill be de!eloped as and $hen appropriate. .he intention o' this book is to gi!e the reader an #nderstanding o' the statistical and geostatistical techni)#es $hich might be #se'#l, not to lay do$n any la$s and reg#lations on $hat sho#ld and sho#ld not be #sed. .he statistical portions o' this book are intended to lay the gro#nd$ork 'or the geostatistical analysis. ,#ch o' this material can be 'o#nd in 'o#ndation statistics books b#t not in the c#rrent conte+t. .he geostatistical portions o' the book ass#me that yo# ha!e mastered the statistical techni)#es described earlier. *t is not ad!isable to Bskip aheadB on the ass#mption that $hat is being disc#ssed has no rele!ance to yo#r o$n interests. .he de!elopment is e+tremely linear, in that one section leads into another. .here are e+ceptions to this, o' co#rse. For e+ample, i' yo# $ill ne!er ha!e to deal $ith skewed data, yo# can skip the chapter on the lognormal distrib#tion and its !ariants. *' yo# $ill ne!er deal $ith more than one meas#rement per sample, yo# can skip most o' the <elationships chapter. *' yo# ne!er deal $ith data $hich has a trend in the !al#es, yo# can skip all b#t the 'irst 'e$ pages o' that chapter. 1.3 (ata sets .he sort o' applications presented $ithin the book are mainly geological $ith some hydrology and en!ironmental case st#dies. .he potential applications incl#de any 'orm o' meas#rable spatial data and some $hich cannot be gi!en a )#antitati!e meas#re, s#ch as rock type, land #se etc. &e ha!e incl#ded applications o' geostatistical techni)#es in the 'ollo$ing 'ields (so 'ar)D o CoalD a sim#lated set o' data based on a real coal seam in So#thern 'rica. =oreholes drilled into the coal seam are meas#red 'orD thickness o' coal (metres), energy content or Kcalori'ic !al#eB o' coal (,ega-o#les per tonne)J ash content (L) and s#lph#r content (L). .hree co(ordinates in metres are a!ailable 'or the top o' the coal seam $here intersected by the drillhole. o GSD this data set is named 'or the Geostatistical ssociation o' So#th 'rica and $as #sed in an ill#stration o' geostatistical techni)#es at a meeting in pril /765 in Cohannesb#rg. .he sample data are taken 'rom deep boreholes drilled into a typical &it$atersrand type gold ree'. .he meas#rements o' interest are the grade o' the gold in grams per tonne o' rock (parts per million) and the thickness o' the ree' intersection in the borehole (centimetres). .he 25 boreholes lie appro+imately / kilometre apart and constit#te a typical data set 'or the planning and design o' a ne$ &its gold mine. .he !al#es ha!e been disg#ised by a 'actor b#t are other$ise #naltered. Co(ordinates are in metres. o SamplesD this data set is based on a &its type gold mine some decades into prod#ction. .he samples are chipped 'rom the 'ace o' the ree' in a $orking section o' the mine (stope). s the 'ace ad!ances, ne$ chip samples are taken. @al#es $ithin a stope are traditionally estimated #sing the sample !al#es 'rom the 'ace. .his data is totally 'ictitio#s e+cept 'or the locations o' the samples, $hich are taken 'rom a real &its type gold mine. o CopperD a sim#lation based on a stockpile o' mined material in the 'ormer So!iet Enion. =oreholes ha!e been drilled into the d#mp. .he drill core is c#t e!ery 2 metres and assayed 'or copper and cobalt content in percentage by $eight. .his is the only three dimensional set o' t#torial data. Co( ordinates are in metres. o Gee!orD this is sample data 'rom a hydrothermal tin deposit in Corn$all, England. .he mineralisation appears as a contin#o#s !ein $hich is s#b(!ertical. Samples o' aro#nd /kg are chipped across the !ein, $hich a!erages abo#t 21 inches $ide. ,eas#rements are grade o' tin in po#nds o' black tin (Sn32) per ton o' rock. .he thickness o' the !ein or BlodeB is meas#red to the nearest inch. Co(ordinates are in 'eet along section and ele!ation abo!e an arbitrary base le!el. Clark, I., 1979, "Does geostatistics work?", Proc. 16th PC!", #ho$as % !&'eil, (d., Societ) of "ining (ngineers of I"( Inc, 'ew *ork, +1,-++.. o &ol'campD meas#rements o' $ater press#re (potentiometric le!el) in 62 $ater $ells in the .e+as panhandle. .his data set $as part o' a st#dy carried o#t by the 3''ice 'or 8#clear &aste *solation in the mid /760Bs on a potential site 'or a high le!el n#clear $aste repository. .he &ol'camp a)#i'er #nderlies the planned repository. 3ne aspect o' repository planning is to )#anti'y the risks inherent in a breach o' the storage 'acility. Sho#ld radion#clides leak into the local a)#i'ers, the scope and speed o' potential contamination has to be assessed. .he press#re o' 'l#id $ithin the a)#i'er $as one o' se!eral !ariables #sed to determine the tra!el path and speed o' tra!el 'or escaped radion#clides. /eference0 1ar2er, 3.4., and 5urr, %."., 1966. "7eostatistical anal)sis of 2otentio$etric data in the 3olfca$2 8uifer of the Palo Duro 9asin, #e:as", 9"I;!'3I-.67, 2ril, !ffice of 'uclear 3aste Isolation, 9attelle "e$orial institute, Colu$<us, !hio. o ScallopsD Scallop data $ere collected d#ring a /770 s#r!ey cr#ise o'' the east coast o' 8orth merica. Scallop co#nts $ere obtained #sing a dredge. ny scallop smaller than 50 mm $as termed a prerecr#it. .otal catch is the s#m o' prerecr#its and recr#its. ,eas#rements incl#ded in the data 'ile areD o 8ational ,arine Fisheries Ser!ice (8,FS) 1 digit strata designator in $hich the sample $as takenJ o sample n#mber per year ranging 'rom / to appro+imately 120J o location in terms o' latit#de and longit#de o' each sample in the tlantic 3ceanJ o total n#mber o' scallops ca#ght at the sample locationJ o n#mber o' scallops $hose shell length is smaller than 50 millimetersJ o n#mber o' scallops $hose shell length is 50 millimeters or larger. <eference0 (cker, ".D., and 1eltshe, %.5. 199=. "7eostatistical esti$ates of Scallo2 <undance", In, Case Studies in 9io$etr), >ange et al., editors. 3ile), 'ew *ork o %io+inD tr#ck transporting dio+in contaminated resid#es d#mped an #nkno$n )#antity o' these $astes onto a 'arm <oad in ,isso#ri. *n 8o!ember, /760, the E.S. E" collected samples o' the site. *n order to red#ce the n#mber o' samples re)#ired, samples $ere composited along transects. .he transects r#n parallel to the high$ay, and this direction is designated as the M(direction. .he direction perpendic#lar to the high$ay is designated as the I(direction. %ata are .C%% concentration (tetrachlorodibenNo(p(dio+in) in micro grams per kilogram (m#g:kg). Co(ordinates and transect length are gi!en in 'eet. /eference0 ?irsch), %.1., and 1arris, D.%. 1966. "7eostatistical anal)sis of ha@ardous waste site data". %ournal of (nAiron$ental (ngineering, 11+077B-76=. o 3rganicsD %ata are Soil 3rganic ,atter !al#es (in grams per kilogram) deri!ed 'rom soil samples taken in a research 'ield at the Eni!ersity o' 8ebraska &est Central <esearch and E+tension Center near 8orth "latte, 8ebraska, ES. %ata $ere taken as part o' e+periments on !ariable(rate 'ertiliNer technology. Co(ordinates are in metres. /eference. 7otwa), C.. and 1ergert, 7.3. C1997D. EEIncor2orating S2atial #rends and nisotro2) in 7eostatistical "a22ing of Soil Pro2erties&&. Soil Science of $erica %ournal, 610+96-,B9 o @el!etl'D S#bsample o' the n#mber o' !el!etlea' $eeds co#nted in 5 meterG area in a 'ield in 8ebraska. %ata $ere collected by Gregg Cohnson (see 2nd re'erence), as part o' a research program in $eed management at the Eni!ersity o' 8ebraska. <e'erencesD %ata set taken 'romD 7otwa), C.., and Strou2, 3.3. 1997. " generali@ed linear $odel a22roach to s2atial data anal)sis and 2rediction". %ournal of gricultural, 9iological, and (nAiron$ental Statistics, +01.7-176. %ata collected byD %ohnsen, 7.., "ortensen, D..,, and 7otwa), C.. 1996. "S2atial and te$2oral anal)sis of weed seedling 2o2ulations using geostatistics". 3eed Science, ==07B=-71B. ll o' the abo!e case st#dies appear some$here $ithin the te+t. .he data 'iles are a!ailable on the C% and can be do$nloaded 'rom the &eb. ll, e+cept samples and possibly copper, are small eno#gh to tackle at desktop and hand calc#lator le!el. &e strongly recommend that yo# carry o#t each analysis by hand at least once to rein'orce the $ritten te+t. 1.# Soft9are *' yo# ha!e this book on C%, the disk also contains a KdemoB !ersion o' the Geostokos so't$are created specially 'or teaching. .his !ersion has slightly more 'eat#res than the EcoSSe package and rather less than the '#ll Geostokos .oolkit. *t is a &indo$s based package $hich c#rrently operates #nder &indo$s 72:76 and 8.. Follo$ the installation instr#ctions s#pplied $ith the package. ll o' the abo!e data sets are s#pplied on the disk. *' yo# ha!e this book in hard copy, yo# may do$nload the so't$are and data sets 'rom the &eb. Check yo#r deli!ery package 'or c#rrent instr#ctions. F#ll listings o' the data sets (e+cept 'or samples) are gi!en in the ppendi+. .he so't$are is identical to the standard Geostokos EcoSSe and .oolkit so't$are packages e+cept that it $ill only read the data 'iles s#pplied $ith the book.
Geochemical Exploration 1974: Proceedings Of The Fifth International Geochemical Exploration Symposium Held In Vancouver, B.C, Canada, April 1-4, 1974, Sponsored And Organized By The Association Of Exploration Geochemists