Scribd Logo
Upload

Welcome to Scribd!

  • Geoe Stadi Stica

    Uploaded by

    SM_Ing.
    1K views290 pages
    Geología

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document
    Geología

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    1K views290 pages

    Geoe Stadi Stica

    Uploaded by

    SM_Ing.
    Geología

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    of 290
    FOUNDATIONS OF GEOSTATISTICS FOR MINE RESOURCE ESTIMATION DYN Outline for Day 1 = Introduction G5 Short Statsteal Review Probabilities Random Variables ‘Common Random Variables and their Properties Conditional Expectations ©) Resource Estimation Concepts © Random Function and Regionalised Variables © Semi-Variograms ‘Concepts and Theory Practical Implementation and modelling ” Brief Introduction to Sampling Theory lee Introduction A First Glance of the Course... The Lecturer = 1 Who is José Saavedra-Rosas? © A Mathematical Engineer and M.Sc. In Operations Management (University of Chile) © Ph.D. in Natural Resources Engineering (Laurentian university, Sudbury, ON, Canada) © Former Lecturer in Mine Planning, Resource Estimation and Mining ‘Operations Research at WASM, Kalgoorlie 1 Currently a Senior Lecturer at DME, Curtin University, Perth, Australia Whet I have done in the past? ew Pa EEEEPErEETP Se Cellular automata models for gravitational flow © Block caving simulation © Simulation of long-wall coal mining for subsidence prediction (> Operations research applied to mining © Genetic Algorithms ® Block caving planning, open pit design, pushback design © Mixed Integer Programming © Open pit planning = Optimisation of energy use in mine planning and design eq Geostatistics © Use of conditional simulations to incorporate grade uncertainty in ine planning © Optimal Drillhole Placement ee | What | expect from you? fla cern RE TIA @ 7 That you learn and enjoy the short course © That you ask as many questions as you want: oI don't have all the answers but we can figure it out together = There are no stupid questions but probably some stupid answers! (those are on me!) | That you talk to me as if | were a colleague Introduction (RR er eR AEE a EET OID 1H © There are key points in the mining process where the resource model is critical to mining decisions: ility study stage study stage § Stages during mining as additional data becomes available Exploration Feostity What is a Resource Model? 1 As the name indicates, a model is an interpretation e or abstract representation of reality. © There is no way to completely know what is beneath the surface unless we extract all of it 1) We necessarily need to build a theoretical model that can be calibrated using samples collected and that it is going to be used for design/planning purposes 1 The resource can be informally defined as the part under the surface that contains a reasonable mineralisation How to build such a Resource lea Model? ©) Well, there are several ways of doing this, but you have to be aware that: Model building capabilities require experience Models are net perfect, hence the whole Process Produces an output that is just an approximation to the real resource There is a lot of analytical skills required to understand the whole process @ This part of the process has been traditionally in the hands of geologists but the trend is to be more integrated with mining What is the output of the process? hese 1" A block model (in some cases can be more than one) Is any block model equally good? (RE erento eer TE I RE ry | The short answer is definitively NO: 5 The so-called volume-variance effect needs to be taken into account, this effect can make our life difficult (more to come about this) 5 Software packages usually have limited ability to deal with large number of blocks (memory limitations), so in practical terms we obtain only what our tools allow us to obtain 1 Everything depends on the level of information required and its precision Grade-Tonnage Curve RI aR RT i RMI ET IRE Grade-tonnage curve — showing total tonnage and mean grade ‘of reserve above a certain cut-off grade ‘Tomage above atoff grade Volume-Variance Effect (1) 100 tonne blocks 400 tonne blecks e maar Global mear Global meon = 4 Range of vation = 1-8 ange of variation = 3-4 Volume-Variance Effect (2) SY SE a RTT In general, we can represent the deposit using blocks of different sizes (e.g., Vi, V2, Vj or V,). The mean of block values must be identical as we are dealing with the same deposit but only divided into different blocks. The variance of block values, however, will be different. ica@®@ 0) > 0%) > o) > o&% i Impact on Selectivity eet 100 tonne blocke 200 tonnes @ 575% 1200 tonnes @ 4.39% Why should we worry about building ood block models? 1b Well, the decision to invest millions of dollars to develop | a deposit depend on our ability to properly estimate | what lies below the surface! | Usually we don't have that much money and we need to | ask for financing (money doesn’t grow on trees!) There are rules that we have to follow in order to play this game (JORC code) We need to understand the process in order to appreciate the effort required to properly characterize a mining proposition but also in order to understand its limitations 0 = eee aur ease eeeaaiaaa Ore-Waste Misclassification aE RES ee ee There are four possible outcomes when a block is mined: PROFIT os anticipated REDUCED PROFIT. g dmrtot—F poet 3 3 REDUCED PROFIT due 3 10 con oxodeted e é wth ling weste NO ECONOMIC Loss Estinated block grode What is Estimation ——_—— Defi “Use sample information fo predict values in areas we hove not sampled” © Features of an estimation problem: © Do we want a global or local estimate? ©) Do we want estimates of point values or of larger block values? What are the tools of the trade? en (2 TE OEE TE ti There are several tools, each one having its benefits and limitations © Each different mineralisation is better suited for a different kind of tool © There are no universal tools for doing the job © Some of the most popular tools are: 5 Nearest Neighbour Cr) 1 Inverse of the Distance 1 Kriging, ete. at Contents of the course at a Glance... an ESS AAA ETT EEL u DAYI e = Short Statistical Review » Probobiities »» Random Variables ‘= Common Random Variables and their Properties = Conditional Expectations © Resource Estimation Concepts © Random Function and Regionalised Variables © Semi-Variograms = Concepts and Theory & Proctical implementation and modelling © Brief Introduction to Sampling Theory a SESS Contents at a Glance... DAY 2 © Compositing © Simple methods for Resource Estimation = Nearest Neighbour f lnverse of the Distance © Point Kriging = Simple © Ordinary © Block Kriging © Indicator Kriging 5 CoKriging ® Iniroductionto Prieipal Components © Gaussian Anamorphosis . Contents at a Glance... u DAY3 9 Introduction to Simulation ‘© Simulation of Random Variables © Geostatistical Simulation ® Turning Bands Method "= LU Decomposition Method = Gaussian Simulation © Simulated Annealing © Advanced Geostatistical Simulation Methods © Introduction to JORC Code. The very basics Random Variable EN (SE POS IRs Pd * Arandom variable x takes on a defined set of values with different probabilities. * For example, if you roll a die, the outcome is random (not fixed) ‘and there are 6 possible outcomes, each of which occur with probability one-sixth. + For example, if you poll people about their voting preferences, the percentage of the sample that responds “Yes on Proposition 100” is @ alto a random variable (the percentage will be slightly differently every time you poll). * Roughly, probability is how frequently we expect different outcomes to occur if we repeat the experiment over and over (“frequentist” view) ar Random variables can be discrete or continuous 1 Discrete random variables have a countable number of outcomes © Examples: Dead/allve, treatment/placebo, dice, counts, ete. Continuous random variables have an infinite continuum of possible values. © Examples: blood pressure, weight, the speed of a car, the real numbers from 1 to 6, Probability functions FRI a LIT 8 . A probability function maps the possible velues of x against their respective probabilities of occurrence, p(x) 11 p(x) is a number from 0 to 1.0. 4 The area under a probability function is always 1. Discrete example: roll of a die fm it a RRS COLES bid 123466 SP@=1 ax. Probability mass function (jpmf) x P(x) 1 l=1)=1/6 2 P(R=2)=1/6 3 P(x=3)=1/6 4 ple=4)=1/6 5 pI=5y=1/6 6 plx=6)=1/6 1.0 Cumulative distribution function (CDF) 1 P(xs1)=1/6 2 P(xs2)=2/6 3 Plrs3)=3/6 4 P(xc 44/60 5 P(xs5)=5/6 6 P(xs6)=6/6 Continuous case a Pena PY AT TAN] + The probability function that accompanies a continuous random variable is a continuous mathematical function that integrates to 1. For example, recall the negative exponential function (in probability, this is called an “exponential distribution”): Sf(e=e*, x20 This function integrates to 1: Jere = fo dt fede = Jere e*[-ou Continuous case: “probability density function” (pdf; The probability that x is any exact particular valve (such as 1.9976) is 0. We can only assign probabilities to possible ranges of x Example = Probability of x falling between 1 and 2: 2 P(sx <2) [e“de=—e7 i eee ace Example: Uniform distribution I © the uniform distribution: all values are equally likely. f(x)=1, for 12x20 ~ ne =—135 +368 =,23 Pla) We can see it’s a probability distribution because It integrates to 1 (the area under the curve is I): eee 1 face Jode+ fides fode=0+ftde+0= xf =1-0-1 ay a Example: Uniform distribution JR Ge TR SSH POCA: PRETTY What's the probability that x is between 0 and 2? Pes) P(A 2x2 0)= Expected Value and Variance fem 5 RN MIE | c All probability distributions are characterized by an expected value (mean) and a variance (standard deviation squared). Expected value of a random variable fe Ee | ') Expected value is just the average or mean (HJ) of random variable x. © It's sometimes called a “weighted average” because more frequent values of X are weighted more highly in the average. © W's also how we expect X to behave on-average over the long run (“frequentist” view). Expected value, formally ae E(X)= Dx, pl) Continuous case: R(X) = [x f(x) ail E(X) = [1 these symbols are used interchangeably Example: expected value cA RRA RC EEA (- If we have the following probability distribution for @ (discrete) random variable: x NO lo else ld Lew) 14 12 2 ja a | Sap) =10(4)+11(2)+12(2)+ 13+ 14) = 113 Sample Mean is a special case of Expected Value... Sample mean, for a sample of n subjects: = YH, YO 1 n The probability (frequency) of each person in the sample is 1 /n. Variance/standard deviation SEPT CAE 07=Var(x) SE(x-p)2 “The expected (or average) squared distance (or deviation) from the mean” Var(x) = EU(x—n)*]= 10, - 4) pe) all x Variance, continuous r Discrete case: Var(X) = S'(x, — 4)? pos,) a Continuous case?: Var(X)= J (x— mw)? f(x)de ax Var(X)= 0, SD(X) = 6. These symbols are used interchangeably Similarity to empirical variance An eT LT EI] The variance of a sample: s? = w Me-w 2 deh Divan 1 alc feet ht we tov ls "dope 1 feedon” plac of lomaton becae we od 0 ‘inate he tmple mon bere we cold ance the samp vrs Uniform Distribution en SR L. Used to model random variables that tend to occur “evenly” cover a range of values 1 Probability of any interval of values proportional to its width Used to generate (simulate) random variables from virtually any distribution _t asysb ie ee 7oy=]4 Foy=|2=2asysb 1 yob 0 elsewhere Uniform Distribution Density Functions Ome b ae beeex @ [Density Function ‘Cumulative Density Function oe ie Uniform Distribution - Expectations an DS )o-( ef 35 2b-a) ~ 2 ee ae (@ +b? +ab) : oe 70) =a()-[eenp ee) [bee]. =2ab _(b-a)* a0 Ma? +0? ab) 308? +a + 2ab) 12 a = 0.2887(6—a) Exponential Distribution = 1 Right-Skewed distribution with maximum at y=0 5 Random variable can only take on positive valves » Used to model inter-arrival times/distances for a Poisson process y>o elsewhere note ill Exponential Density Functions | Density Functlon Cumulative Density Function ee Exponential Distribution - Expectations fl DCE Ie ERT om =P of Le ly all yeegy = f° rene gy — =D {50" rab L vero [ytertare = grow =0(2-1)!=0 2(Y?)= Degen }o = gh rea =f ytertay = a = 11 6)0 =0°8-1!= 26" =V(Y)=2(?)-[E(f = 26? -(0)? = 6? =0=0 Normal (Gaussian) Distribution istribution with tendency for individuals to clump. Ground the group median/mean © Used to mode! many biological phenomena © Many estimators have approximate normal samping distributions (see Central limit Theorem) TOK Y0 eee H 4 du zd pe oe aa nao( aA gg) Man Flor at Ee Loe) 1 ee ae eee aig fae (SaaS ae =V(z)= E(2*)-[E@)F =1-0 =1 > o=1 Notesity—MGuo?) then Y= Za Brn Bort Pola seta) a oro Sranaverze)-eFZ=s= 2" Lognormal Distribution Y>0,-w0 fO)=; 0 otherwise Note: 7° = In(¥)~ (1,0?) 20) Bl" )o m1 =e) ZL) een a()=2(e"F)=2(")= Mytrenrmonf Hr ZO). eed = V(1)= E(Y?)-[E()f = e%) ene Lognormal Distribution — Density Functions Independence [en SR ET cE TRO ‘Two random variables X and Y are said to be independent if and only if P(ANB) = P(A): P(B) If not, this last expression cannot be calculated as easily as the last one P(ANB) # P(A): P(B) Actually, P(B) needs to be conditioned by the valve that A takes the probability that a 40 year old weights 30Kg if tis 1.90me ©») What is the probability that a 40 year old weights 90Kg if his height is 1.90m? Independent # Mutually Exclusive Events (Aan RE RC IR Pa S| 3 Mutually exclusive events A and ~Aare such e that © P(A) =1- P(~A) »P(An~A) =0 cj However » P(A): P(~A) #0 © Conceptually, once A has happened, ~A is impossible: © Thus, they are completely dependent Bayes’ Rule (1) |e ESSN ROR © Let A and B be two events with P(B) + 0. The conditional probability of A given B is: P(ANB) P(A|B) = PC) & We need to observe that usually this rule is expressed in a slightly different manner P(ANB) = P(A) - P(A|B) 5 Or equivalently (because AN B = BNA): P(A NB) = P(B) - P(BIA) @ bow pre le sinlocel eyo: | Bayes’ Rule (2) i 1 Bayes’ rule is central to geostatistics = What we really want to know is the grade at a certain location conditioned to the actval values measured © Other way of writing Bayes’ rule — P(AnB) _ P(BIA) P(A) PIB) = P(B) —P(BIA)-P(A)+P(B|~A) P(~A) — eee ae | A note on causality (1) ae ER AN EO] \) Causality is generally defined as the relation between cause and effect © A causality relationship between events A and B implies that either event A causes event B or vice versa 5 Typical example of causality: blasting rock causes it to break (but sometimes rock breaks without blasting!) “In the mind of people causality is usually measured through the use of correlations A note on causality (2) Sn ER AE A) «1 Correlation is a statistic that quantifies a relation between two variables = Can be either positive or negative © Falls between -1.0 and 1.00 0 The value of the number (not the sign) indicates the strength of the relation ae A note on causality (3) fi bn Es TERR ORE © Pearson's correlation coefficient — HX = 1d - 1x)} = Ox oy — 2G - Ax) Oi — Br) (n= Dsxsy A note on causality (4) = a Z - \e Ne &- > Oo A note on causality (5) Classical example of bad causality rel ship ‘The Real Cause of Patio! Correlation (almost perfect in this case does not imply causality, i.e., polio is not caused by ice e cream consumption! Spurious Correlations (1) aR eS SS TAROT) = @ Correlation: 0.992082 ‘Source: htps//wwwitylervigee.com/ Spurious Correlations (2) jas @ Correlation: 0.992558 ‘Source: ip //wrwwaylervigensom/ Spurious Correlations (3) ae 9 Correlation: -0.933389 Source: htp://wwwsylervigen.com/ Covariance fen ERE A RES Covariance relates to the correlation coefficient by: _ E(K = ux) ~ Hx)} _ Cov(X¥) ee Oxo, Oxy \) Re-arranging terms we can write: Cov(X,¥) = Pxy * Oxo © Covariance and correlation are similar concepts, as we will see part of the Geostatistical work is to fit a semi- variogram function which in most cases translates to a e spatial covariance function (thus providing us with correlation between points in space) ie Resource Estimation Concepts Estimation (1) = 5 The prediction of the value of a characteristic at a location at which it is unknown from values of the characteristic measured at a number of known locations. , Wot 239 igh a 2 ° @ # ° 06 g/t ° : a 0.1 of ee eerereenanenreeaseeeed Estimation (2) |RATING | 17 In most cases the estimated value will differ from the true value 1 The discrepancy between an estimated value and the corresponding actual valve is called an estimation error. 1 The size of the estimation error will depend on two things: “The variability of the values to be estimated © The ability of the estimation method = Estimation (3) 5 Many traditional ways of estimation in mining or mining geology context include: 5 Inverse distance weighting © Cross-sectional interpolation © Polygonal method | Triangulation method 5 Ete, 4 And more advanced estimation methods © Geostatistics Traditional Methods (A fcc aR ee SNe TS DTS RNY 5 Traditional methods use an implicit model of spatial @ continuity (this is an assumption) © This is usually not explicitly acknowledged © This model is based on beliefs These assumptions may be suitable for applications in a real case setting 5 There is a need for verification of the output of the model in order to assess the appropriateness of the assumptions for the particular mineralization style Geostatistical Methods (RRS Reesor To NTE A 1 Geostatistical methods also use a model for spatial continuity = However, it is not based on beliefs but rather derived from available data 1 Tools are also available is geoestatistics to perform validation of the estimates Resource /Reserve Estimation Cycle fam SE ac NRE c——“ Gollecting Samples “Measviing Characteristics of Somples Inferring cherocterste of the orebody from the ‘measured characteristics of the samples Inferring @ model for estimation Validation of the model Assessing assumptions, models and estimates mn SR SA RN AT Two important criteria On average (in statistical sense), models, concepts, assumptions and estimates are correct mAny local or individual deviations from reality of models, concepts, assumptions and estimates are as small as possible eo Regarding Assumptions mm 25S A NTT TTS) 1 As assumptions increase the risk of error, they ° should, as far as possible, be: Kept as simple as possible Kept to a minimum © Reasonable and logical from every relevant point of view (geological, structural, mathematical); and © Verifiable Statistical Representation (1) (RS SSH OR cee RE LRA TNREN| In geostatistical applications the population is the entire orebody, or some homogeneous section of it, subdivided into sample size volumes © The orebody is viewed as a population of sample-sized volumes. "A relatively small number of these volumes is selected, el these comprise the sample Statistical Representation (2) A RRA AE NT @] © The samples are collected from this population Statistical Representation (3) PR pete saa PANN AST RAE RE a TAY The problem of estimation can now be stated as : 1 Collect samples 2 Measure characteristics (eg, grade) of samples >. From the values of the sample characteristics infer the values of the same characteristics for the rest of the sample size volumes that comprise the orebody 4 Using the values of the sample characteristics predict (estimate) the mean values of the characteristics of the sample size volumes or blocks Statistical Representation (4) 1 | P| \ In mining applications, estimation is usually performed in supports (volume and shape) different from that of the sample support ci The support for which the estimates are produced are usually blocks (which is the usual input for mine planning and other value adding processes) © The group of estimated blocks is called block model © A block can be seen as a set of sample-sized volumes Regionalised or Spatial Variable je reser SnSRRET CARRION REO] © A variable whose value depends on the location at which it is measured. & A variable for which values are averages over a specified volume (e.g., core sample, block) 1 The variability of the values of the variable depend on its volume and, in some cases, on its shape and orientation, e Fundamental idea of Geostatistics .. © The values of characteristics of samples is on average more similar when the samples are close than when the samples are further apart. 41 This relationship is termed Spatial Correlation, and indicates the extent to which samples located at a given distance in a given direction correlate 15 If there is no relationship between samples, they aré called independent Simple Example (1) [RR AMGEN 8 WO AAPA RAEI ROE TREY > Suppose you want to estimate the value of a location that is 2 metres away from a sample value of 13.8 g/t. 13.8 6/t Tne seme value 2m Lb © tinknown von (to he etimated) e Simple Example (2) fa cea RR OTRO] |. Without any other information, to use the value 13.8 e g/t for the estimate is possibly the best that can be done. © How big the error of estimation is going to be? 5 Error = True value ~ Estimated value The answer cannot be calculated if only one sample is know 1 However if we have several samples, we can collect the “statistics” for all those samples that are 5m apart to see what the average difference is, that way we can get an idea of the magnitude of the error if we use 13.8 g/t for the unknown location eee Simple Example (3) 00 Assume for Instance that we have drillhole information that is being “cut” at regular intervals of one metre @ DH DH Simple Example (4) RRA i TTC .. © What we will do is to figure out the error | could expect if assigning the grade 13.8 g/t to the location for which | don't have sample information 5 In order to do so, we need to estimate such error © We consider then all the pairs that are separated by 2m, and calculate the error based on the true grade (which we know) and the estimated grade (which we also know) Simple Example (5) [i Rete (ro TEE NRIOL MRT TER 4 With this as information we can ne actually see how Ba 3 similar/different are {fr aes samples at different or Bo me separations i ous 1 For example at 1m es fo8 & we get fa fob 2 a uo te a fl 67 sa Simple Example (6) Ym ae a A TRE RRR i The scatterplot of the previous pairs looks as follows Separation - 1m Simple Example (7) ll EERE ERECT 5 Taking now pairs at a separation of 2m gives us: a sa Boe om Separation 2m Simple Example (8) liieeen ea aioe a ee eee @ “Taking now pairs at a separation of 3m gives us: Simple Example (9) Da a al aa > Let us compare the average error nq] peas zi log! Simple Example (10) fe ene URSIN TERT | 1 Let us compare the scatterplots n= 1,2 and 3m ; a, : 7 ae must 3 he 8 aus Simple Example (11) fii mcr orn ETT 5 What can be concluded from the previous experiment? 5 The average square error (a measure of dispersion) generally increases as the lag distance increases © From the scatterplots we see that the cloud around the perfect line gets wider as the lag distance increases 5 Le., as we increase the distance, the samples tend to be 6 less similar Simple Example (12) For the simple example considered, the estimate will be then 1380/1 Keown sample vole ——b__2 , tprown vobe (to be estimated) 138 g/t 290028571 If the data is normally distributed, then we can calculate a confidence interval for this value, so we know that the true grade lies approximately in the interval (with a 95% confidence): (9.171945, 18.42806] c Simple Example (13) Graphic summary of the process Pete marized eS. from drilthole ‘ormaton 138 9/ an 13.89/t Used 6 “uontfy he Serf “Siimaton General Properties — 3 The average error is zero or close to zero © This condition is called unbiasedness « » y+! The variance of estimation is used to calculate the confidence interval 1 As we will see this variance is called Variogram = The variogram reveals the spatial correlation relationships = The variogram can be estimated from the data {experimental variogram) «= A functional form is fitted to this “experimental” variogram (variogram model) ee eee Se eae Extending to more than one sample Pai Paamasoricre roo TCoRRATEW| (0 The previous example was quite simple, in practice we have more than one sample available to estimate values. How do we proceed? Ww Ligh za oft | * . ° 0.5 g/t ° ligt Ol gir eee) —————— Extending ... (2) iid 2 © The natural way to extend is to take a weighted average of the samples available 179 239/ Ligh a : Dan vlad pet *, 1 ° 06 g/t ° * Extending ... (3) Jase ALR RTRSY 5 The question is now how much weight to give to every sample © The answer to this question comes in the form f what is known as Kriging ging ves « model of spatial correlation to determine such weights » All this is good for a given point in space © What happens when we want to estimate at a different support? ‘= This is what is usually called Block Kriging, recall that there is ‘volume variance effec, so the variance needs to be corrected to Toke this into account ae nstalling and Running SGEMS e —_ Introduction a SGeMS stands for Stanford Geostatistical Modeling Software It is based on GsTL (Geostatistical Template Library), also developed at Stanford (by Nicolas Rémy) 3 Ithas been developed further by a group of Tesearchers at Stanford University (including Alexandre Boucher as important contributor) 1 Itis Open Source, so eventually anyone can modify it 1 It provides a user friendly interface © It provides 3-D visualisation SGeMS main window Where do we get SGEMS from? | | NR TPR TTI] 5 SGeMS can be obtained from http: //sgems.sourceforge.net/ © You need to click on download and there you will find several options (about 18Mb each): 0 64 bits with NVIDIA video card 10 64 bits without NVIDIA video card 8 32 Bits with NVIDIA video card © 32 Bits without NVIDIA video card 1D Previous version SGeMS2.1b 32 bits a Al these versions are available for you to install e depending on your machine configuration Installation Process (1) | EES ED 15 Once you click your chosen installer you will get the |@ following window od Sok} ‘Welcome othe SoM -bota-wns2 ‘Satup Ward arn Click on Next> Installation Process (2) | nannies MNS SERSRNEDNEEERSER| «1 Now click on | Agree after reading the license agreement (if you want!) Isc ner ny es "cremate Eo) Installation Process (3) RD AER SEU a ERR REET G Now choose the folder where you oud like to install SGeMS, click on install essa annie - x “Stet ae ete a Installation Process (4) |) ER AE | 3 Something happens in between... az Installation Process (5) | ca eC CREE | cl Until it finishes at which point you click on Next > “Thecoereedcmt a farmstead ome Installation Process (6) pasa 5 Finally a message saying that the installation has been completed is shown in the screen, finally click on Finish ‘Completing the SGaNS-bate-ins2 Setup mars Running SGeMS (1) .. 5 Depending on your version of Windows, you may need to find the program eee ee on Sees eee = ee Running SGeMS (2) | a eA TE TT © Click on the program 5 A welcome screen will be shown co After loading the main window will come up. It is advisable to maximise the main application to get better access the user interface Running SGeMS (3) EE TT Algorithme Data Obiects 3D Visuaisetion Running SGeMS (4) hea © Available algorithms We wil discuss some of these in great detail during the next couple of days ‘The first third contains ilies that allow {you to perform post-processing and most Important, histogram transformations Then there 6 @ Simulation section Finally thore is © Kriging tection (Etimatlon) seeglngpsl yay it i I ! i Hh i i i | i il iH i { I q I H * 3 5 i F i f ? i i ! P mi: Concepts Random Functions and Reg Definitions (1) Jon eR Ta RESET ET EEN Recall that @ random variable is a variable which takes values according to certain probability distribution » A cumulative distribution function (cdf) for the random variable Z is defined as F(a) = Pr{Z < a} Definitions (2) Aa ERNE Te ARES REN CR © Example: Tossing a coin. Events are to get head or to get tail (H,T). Random Variable: Z- 1 if we get head ~\0 if we get tail co We have: PriZ=1}=Priz=}= 1 F,(0)=PriZ <0} ==3F, (I) =Pr{Z Var{Z(%)} = E{(Z(%)—m)}=C(0) VE Fuze +h)-Z@y}=CW)-CH) VE Definitions (9) (en CREE nT RIE RRA RRR «1A stationary process is ergodic if long-term averages along any simple output function of a stationary process converge to the expectations of the corresponding variables n Another view of the same: All possible outcome records of the process have the same statistical properties 3 Not all stationary processes are ergodic Cartesian Grids (1) @ 7 Cartesian Grids are SGeMS object that are specified in terms of a Cartesian system of coordinates 1 We need to specify on origin 1 We need to give the block dimensions along each axis (dimensions can differ) 1 We need to specify a number of cells along each direction Creating the Grid (1) Creating the Grid (2) Name for the Grid needs to be specified Number af ces cn each dimension Calls dimensions on each direction Origin of the coordinate system -—— Example (1) Notes | ia @ © At the moment we have just created an empty shell with no properties yet 5 Properties will be added later by means of geostatistical algorithms 1 You can add as many properties as needed (the only real limitation is computer memory /storage) General 4 SGeMS can read data from files in its own format | @ and can also read files in CSV and GSLIB format lt has to be noted that GSLIB is pretty much a de facto standard format for geostatistical data 3 SGeMS own format is binary and it is only understood by SGeMS ci GSLIB and CSV format can be read with Notepad/EXCEL The Dataset to load | eR SET | 1 Look for the file ZoneA.dat, a GSLIB-format data file containing porosity and permeability data from 85 wells in Zone A of the Big Bean Oil Field (fictitious, but based on characteristics of a real field, 20 x 16 km in extent) 1 The -999.9999's represent missing valves (permeability information is available only at 42 of @ the 85 wells). ZoneA.dat 5100 1900 24.6156 15,1084 1.1407 0.0572 O.aze -0,1696 53500. ¢v00 33.2458 14.3724 -999, 9989 "399.9999 {e800 «800 37-0216 15.0814 -999.9999 995.9999 -999.9999 -969. 9999 Loading ZoneA.dat (1) ) SRS So IT P) sedis bas File Objects Prope Regions Data Anais View Scripls Help ‘Noo New Catesion Grid aun ‘ag. Now Maked id From Catan Gd Cue Loading ZoneA.dat (2) Loading ZoneA.dat (3) For object type please select point st ‘This formot expects North Easting end Eevation coordinates ond additionelly will read the remalning properties You need 1 name the pointset, specity hich colunnshove which coordinates You need to click on Use No-Date-Value ‘ond specify which marker Is being used for absent date in cur case -999.9999) Loading ZoneA.dat (5) | RIMES NR RN RT | 3 Now the object is loaded, you can click on it and select one of the properties for visualization Loading ZoneA.dat (6) | i TS oT | a For example for porosity we have: Additional notes | SERRE sree TTS | 1c We have loaded a GSLIB file, however other formats are available [BYeTre) Saving and Exporting Project with SGeM: Saving a Project Extra Notes | SERRE ST TET | © When you save a Project you are saving all the objects that comprise it in SGeMS binary format, you can see what is inside the folder actually =~ . ‘° is an XML file containing a description of the geometry of ‘the Grid Coordinates SGeMS File | SR This is a binary file (can be decoded but it is out of: | the scope of this short course) Properties files jaa PS 1 Each property is written to a sgems file (binary © format) 1b We note that this structure is logical as we already have another file with the grid coordinates 1 The only thing the program does is then to load properties on top of the already defined grid rn Be careful not to overwrite files in these directories or otherwise you could potentially loose all your work Exporting a dataset (1) he ne © SGeMS offers you the option to export a dataset. If you choose this option you will four different formats: csv, gslib, sgems and sgems_beta Exporting a dataset (2) © You need to chose the object to export and the file format you want this object to export to. - aie ole Be ep Exporting a dataset (3) | SSR er Pe RPE PE PPP eT | 13 Once an object is selected, you need to choose which property(ies) you want to export (allows for multiple selection) ‘Pies of oes, Seer reer rere rere ee J ee ¢ ® Probably @ fabulous thing that wil never happen! (ig q ——————————— Basic Statistics fom 5 Once you have opened a project, or otherwise imported a file or created your own Cartesian grid you may loo at the basic statistics of the data 5 This step should not be overlooked as sometimes digitation errors are present = How do you interpret a grade value for Gold of 5000 ar/tont Available Options 5 SGeMS offer the following options for analysis 41 SGeMS beta File Objects Properties Regions | Data Analyst View Sits. Help Histogram (1) | mms A ER S| nos TE You need to specify the me ‘object of interest a Also the property for that 2 : beet hat you went te conctys0 aoa * 7 —. Finally there are option to | 7 "specify the type of plet you {= | ‘watt fo see 3 err ee . eee) Histogram (2) _ “The histogram describes the overall distribution of grades Hore we observe that Maximum, Minimun, Median, Quarles are informed re Alo Méon and Variance This portiaular graph I called POF (Probability Density Function) Histogram (3) © You can save the histogram as a PNG file (for using it in reports if you wish) The statistics can be added the figure (optioncl step) & Other options for the histogram are Cumulative Density Function (CDF) and PDF + CDF Histogram (5) | AE SS SE | 5 Please feel free to explore other options available in the software 1 For example, change the scale to logarithmic to see how the histogram looks like Hn @ Histogram (6) || RN EE | © You may have noticed a parameter called “number of bins” 1 How many do you need? 1 Extreme cases: Bu 1 bin 200 bins Histogram (7) | ean es ro TEETER] © A simple rule of thumb to remember is to use Vt bins with n the number of data points in your dataset 5 Note that SGeMS will not allow you to use more than 2000 bins (which roughly translates to 4,000,000 points in your database) Q-Q Plot (1) jas 2 A Q-Q Plot is a Quantile-Quantile plot, it is a @ graphical method to compare probability distributions 0 It does so by plotting quantiles of the first dataset against quantiles of the second. 3A quantile is that value that leaves a percentage of the data below him, for example a 0.3 quantile is that value that leaves 30% of the sample below that value 5 The median is the 0.5 quantile of the distribution Q-Q Plot (2) | | oR ER © If two distributions are similar, then they will nicely sit on a line 10 This means that the corresponding proportions of the quantiles remain balanced (thus the values are similarly distributed) Gif they are sitting on a 45 degree line, then they are | | the same ditribution Q-Q Plot (3) Q-Q Plot (4) | SRRERERERSR | © Please prepare a Q-Q Plot of two variables in the dataset 5 What happen if you try Thickness against logperm? = What is your theory of what is going on? = You may benefit from looking at the histogram of logperm... Q-Q Plot (5) iii | © There is another option called P-P plot which does not be confused with the Q-Q@ plot. A P-P plot compares the cumulative empirical distributions It is used to see how closely two datasets agree Scatterplot Variogram | et iba eames ye ARE dal ot NL cM | a This is the most important option (for geoestatistical purposes) available in SGeMS a It will be discussed later in detail... Semi-Variograms eed Semi-Variogram definition © Given a geostatistical model, Z(s), its Variogram ¥(h) is formally defined as 109 = 3 var(zts) = 20) = Zo -200)*6.0asau Experimental Semi-Variogram Din SET A | ci Dissimilarities are measured through the experimental semi-variogram: 1G= 7 dee, +f)-26.)) 1 aa + Relation between semi-variogram and covariances 7(h)=C(0)-C(h) Semi-Variogram Model Sill = overall population variance: sill Samples here not spatially ' 1 1 1 1 1 1 1 correlated | Spatially correlated | Some comments REE Sera NE RE PRI © The experimental variogram needs to be modelled: tis imperfect: the points obtained are subject to imprecision incomplete: it has been calculated on discrete directions at discrete steps © A model is adjusted around the experimental variogram © There are theoretical requirements for such a model @ © This variogram model will be used as if it were the true one Variogram Fitting Requirements for Valid Variogram Models ram is positive: y(h) > 0 ©) The variogram is an even function: y(h) = y(—A) © The variogram is zero at the origin y(0) = 0 © The variogram i concltionally negative definite function Ay € Ry a = 0,VEp uke IF VA, ee Ady (i - %) <0 Semi-variogram Models (1) RR Geraniol e err ea UR TEONRS SS . c caNEY 0 Linear 7h) =¢,+bh 5 Power Model y(t) =c, +bh” 4 Exponential (A) =¢) +¢,(I-e"") o Gaussian 7(h) =) +e, (-e“") 3h 1h, 0 Spherical 74) = ta(32-16) fr0 3 tree ane decor Variogram (12) ) © If you click on next you obtain the following: Variogram (13) @ 2 Let us look at the figure with all variograms © 2000 4000 6000 8000 100001200014000 Variogram (14) | RRR EE ET LE 10 You can save the variogram plots and once fitted you can save the variogram models or otherwise load them from previously saved work 1 Por pony im percent Por porosity in perce (Ui Fie at Window Lead Vogt Mode 1 SaeVutogam Mode ‘Seve primes Vogrs Variogram (15) 1c In order to fit variograms you need a = e to specify the number of structures, —— type of variogram, ranges, nugget effect, etc. 3 SGeMS comes with a menu on the right that allows you to model the variogram 1 If we use the tool we see the theoretical model superimposed on the empirical one. eee Variogram (16) oyebeatocedtd |" {3 Variogram (17) o © For this particular case | “believe” that the variogram is comprised of one spherical structure with the following characteristics: No nugget effect DA sill of about 0.8 5 Omnidirectional with a range of 4200m 9 Variogram modelling is an art of its own and requires experience (or good geological criteria) Variogram (18) fer & Variogram models can be saved (recommended os we will need it later}: 1B Por % porostyin percent vs. Por "pore in pe File Edit. Window A note on angle conventions (1) | antici aesnor is =e eeE | 1D SGeMS used to have its own angle convention however for compatibility issues it has been changed to follow that of GSLIB 1 The logic of the GSLIB convention is that X axis corresponds to the EAST direction, Y corresponds to the NORTH ond Z corresponds to the elevation ©) The goal is to rotate the coordinate system to be along strike, down dip and perpendicular to the structure coordinates A note on angle conventions (2) | © The first rotation angle rotates in a clockwise manner (@) around the Z axis which remains unchanged, © The idea is to orient the Y axis along the strike and then X will orient in the dip direction © This step is called azimuth correction and is performed by the action of the following rotation matrix X\ [cosa -sina 0} /Xx’ ¥,}=|sina cosa o|-(y Z 0 Oo a) 7, _@ A note on angle conventions (3) _ 9 The second rotation angle (8) is a clockwise rotation around the X; axis (so Yp is along the strike and down dip) © This step is called dip correction and is performed by the effect of the following rotation matrix X, 1 0 0 ‘Xy (%)= 0 cosP sing (1) 0 -sinB cosp Zi}, Zz A note on angle conventions (4) ) STE TTC 1 The third rotation is called plunge correction and is a rotation performed at an angle @ around the Yq ‘axis in a clockwise (looking down dip) manner so Xp will coincide with the medium axis and Zp with the minor axis. 10 This is achieved by the action of the following rotation matrix (XR cos 0 —sin@] /Xr (m)=['o 1}. | Zp. sin@ 0 cos@ 2). a A note on angle conventions (5) i ET 9 All the steps can be performed at once with one big. @ rotation matrix (obtained as the product of the three intermediate steps) ¢ | | OQ. forsee -sinacosd-+corasinB sind ~cosp sind sinacosp 8 sing cos esind —sinasinfcos® —sinasin ~cosasinpicos® cos#icas® A note on angle conventions (6) | EE | © The inverse of the previous matrix is just the transpose © In practical terms is quite easy to move from original coordinate system to rotated coordinate system and vice-versa 0 The biggest problem is to identify the angles, that requires good visualization skills @ A note on angle conventions (2) q: z x | ok ne b ne -j ce % x Xp YR Ye Ya lie’ Sampling Brief Introduction to the Theory The Role of Sampling EE i ET RS RRA 5 Sampling must be: Truly representative of the material being sampled © Ne bias introduced in sampling © Rule of “equal probability” must apply, ie. that every particle must have the same probability of being collected in the sample. Idealised Example [ae Heer CREE SR True grade by weight: 1% © Waste © Gold If we sample using 1 block then: In 99% of the cases we will get 0% In 1% of the cases we will get 100% Both results are Incorrect! The way we sample will affect what we communicate. Furthermore, error is unavoidable! Sampling Collection and preparation of samples, to yield analytical data which is: Accurate, where [m(e)]

    You might also like