Three-Dimensional Computer-Based Gold Prospectivity Mapping Using Conventional Geographic Information Systems, Three-Dimensional

Home , Yilgarn Craton

Three-Dimensional Computer-Based Gold Prospectivity Mapping using Conventional Geographic Information systems, Three- Dimensional Mine Visualization Software and Custom Built Spatial Analysis Tools

Eduardo Videla1 and Carl M. Knox-Robinson2 Centre for Teaching and Research in Strategic Mineral Deposits Department of Geology and Geophysics The University of Western Australia NEDLANDS WA 6907, Australia [email protected] [email protected]

Presented at the second annual conference of GeoComputation ‘97 & SIRC ‘97, University of Otago, New Zealand, 26-29 August 1997

Effective mineral exploration requires a detailed knowl- the research aims to identify potential continuations of edge of the factors and processes which result in the for- known ore bodies, and to attempt to locate new prospec- mation of economic deposits. To apply this knowledge, a tive areas for gold mineralisation further to the south of sound three-dimensional understanding of the geology and the present goldfield. structure of a region is required. In most cases, surface The Wiluna goldfield comprises a region approximately geology can be mapped with a high degree of accuracy, 3km x 5km, and is situated in the northern part of the however, the geology at depth has to be inferred from Archaean Yilgarn Block of Western Australia, approximately geophysical methods, or through drilling programs, and is four kilometres south of the Wiluna townsite. The area therefore mapped at much lower spatial resolution. This has been mined for gold since the early 1900s, and pres- anisotropy in spatial-data quality, coupled with the scarcity ently comprises 11 open-cut and underground gold mines. of three dimensional geographic information systems (GIS), Geological information for the region includes detailed makes computer-based exploration at a camp- or district- surface mapping and over 6,000 unevenly distributed drill scale very difficult. holes, totalling in excess of 300km of core samples. The The process of computer-based three-dimensional min- drill-core has been assayed for gold and gold-related ele- eral exploration is being addressed at the Centre for Teach- ments, including arsenic, antinomy, and sulphur. Plans and ing and Research in Strategic Mineral Deposits within the sections from mine construction activities provide detailed Department of Geology and Geophysics at The University three-dimensional information for limited areas. These data of Western Australia. A research project is in progress which have been entered into a 3D mining package, and a model attempts to define a three-dimensional gold prospectivity of the surface and interpolated sub-surface geology and model for the Wiluna goldfield. The aim of this research is structure constructed. This model will be used as a base to gain a better understanding of the factors which spa- on which to conduct a gold prospectivity analysis tially control the location of the known ore bodies, and Several GIS-based methods have been developed to as- especially of high-grade zones within these bodies. Also,

Proceedings of GeoComputation ‘97 & SIRC ‘97 15 sess and map prospectivity on a global to district scale 2- Data collection

using datasets such as geological maps, aeromagnetic and Data available for this project includes: radiometric data, topography, and satellite imagery. a) a detailed geological surface map at scale 1:2 500 pro- Prospectivity mapping methodologies can be split into two duced in a previous research (S. Hagemann, 1992). This broad groups: knowledge driven and data driven. Knowl- map is available in digital format and was updated with edge driven methodologies involve the application of con- more recent information. It includes main ceptual models to appropriate spatial datasets, whereas lithostratigraphic units, first-, second- and third-order data-driven methodologies look for significant spatial rela- structures, and measurements of azimuth and dip of tionships between known sites of mineralisation and sur- faults. rounding geological features. Identified spatial relationships b) an extensive drillhole database including exploration are quantified as mappable criteria and are ultimately inte- and evaluation records. The information contained in grated into a single prospectivity map. Techniques applied the database includes- with varying degrees of success include Boolean logic, in- > 6000 drillholes (RC, diamond, evaluation) dex-overlay, Bayesian statistics, fuzzy logic and artificial > 375 000 meters of drill neural networks. > 200 000 Au assays The majority of assessments to date have been two-di- > 95 000 geochemical assays mensional in nature and normally conducted at a scale > 95 000 magnetic susceptibility of host rock meas- where deposits can be adequately represented as point urements features. This research has the additional complications in > 15 000 rock descriptions that the third dimension must be addressed, and that the detailed geological maps of pit and underground works scale of observation is such that ore-bodies have a definite and interpreted geological cross sections. volume, and cannot be regarded as simple point objects. Although present mining packages are good at visualising 3- Data integration and 3D modelling three-dimensional bodies, and are capable of measuring To achieve the best graphic visualization of the complex lengths, areas and volumes, most packages lack an in-built environment of the geological subsurface, the available in- macro language which would allow a quantitative exami- formation was integrated using a mining visualisation soft- nation of gold prospectivity. Consequently, dedicated data ware. A 3D model was constructed correlating the sur- handling programs are being developed to extract the spa- face map with the drillhole information, underground min- tial information from the mining software and to conduct ing maps and interpreted geological sections. quantitative spatial analysis techniques to identify and quan- A number of problems need to be addressed in the proc- tify significant spatial relationships between high-grade ore ess of interpretation and integration of data. These prob- zones and the surrounding geology. lems are technological as well as inherent to the data. 1- Objectives Present mining visualisation software require high speciali- sation and the process of updating the model according to Through the integration of two-dimensional surface geo- new information is difficult and extremely time consum- logical maps and-three-dimensional subsurface information ing. In terms of the data, good correlation is achieved in of the Wiluna goldfield, the aim of this research is to con- dense sampled areas but an increasing degree of interpo- struct a three-dimensional geological model of the area lation and uncertainty is introduced in poorly sampled ar- and to quantitatively analyse controls on gold mineralisa- eas combined with the inherent anisotropy and complex- tion. Use of these controls to define methodology for re- ity of the geologic subsurface. gional GIS-based gold prospectivity analysis.

16 Proceedings of GeoComputation ‘97 & SIRC ‘97 4- Controls of the mineralisation with fluid-wallrock interaction and physico-chemical con-

It is accepted that the gold mineralisation is late in the ditions in the time of the mineralisation. Extensional veins, tectonic evolution of the Yilgarn craton (Groves, 1993). dilational jogs, shear veins, divergent bends, etc, are all terms The fault system, along with relevant lithological contacts, related with the geometry of the structures formed after is the principal control of the mineralisation. As the faults the application of a directed regional stress to the rockmass. play a critical role in the siting of the ore bodies, an accu- A measure of this deformation is the displacement along a rate spatial representation of this structures is required. fault, the azimuth and dip, the angle formed between faults, veins, joints, the orientation of the schistosity, etc. The traces of the faults on the surface map, their projec- tion in the underground mining maps, and the drillhole In order to find relationships between gold mineralisation intersection of the fault in subsurface, provide the lines and the hosting structures, it is necessary a spatial and points used to create these entities in space. An em- discretisation of the faults into basic components, at a scale pirical spatial resolution of 15 m was adopted for the basic relatively similar to that of the gold assays, and to generate cells that represent three dimensional geological solids. new variables relating the relative spatial position of gold Lines and points were spatially gridded at this resolution, and structure. and a best polynomial algorithm fitted a TIN (triangulated By construction, the fault surface is made from a variable irregular network) to these points. In all cases, control number of ordered triangular facets connected by the sides, points were left without gridding to validate the interpola- they represent the topology of the physical surface. The tion accuracy. The final surfaces look smooth and realistic computation of the centroid (properly called hypocenter) and serve as a basis for further spatial analysis. of every triangular facet generates the points necessaries for the analysis. 5- Data extraction At the same time, operating on the normal vector to every In this study the kind of datasets required for analysis are facet it is possible to calculate its spatial orientation, in dependent primarily on the type of deposit investigated. terms of azimuth and dip. Angular relations between fac- The Wiluna lode gold deposits are predominantly struc- ets or their normals, allow measures of coplanarity, con- turally controlled with relative lithologic control. Conse- cavity, convexity, bends and variability in azimuth and dip quently, the solid 3D representation of faults and struc- of the faults or lithological contacts. The vector equation tures of first to third order is required to identify suitable of the plane for every single facet enables to discriminate relationships with gold mineralisation and specially of high- points in space relative to this plane in terms of above and grade accumulations within these bodies. below, or in geological terms hangingwall and footwall. From the final 3D model, the TIN representation of solid The computation of the distance to the nearest facet in geologic entities can be exported in various formats for the fault for every gold value in space, generates a spatial further analysis. For this project, ASCII files of the format variable that relates gold grade with azimuth, dip and prox- {x1,y1,z1,x2,y2,z2,x3,y3,z3} which represent the spatial imity to the fault. coordinates of the three corners of the basic triangular units, were extracted for each entity. 7- Customised software 6- Spatial analysis of the fault system To conduct quantitative spatial analysis to identify significant relationships between high-grade ore zones and the Considering that the ore bodies are mainly controlled by surrounding geology, dedicated data handling programs faults and are present in determinated sites, but not in were developed. These specific pieces of software were others, it is inferred that particular geometrical features created in Borland C++, to fulfill the necessity of spatial along these faults are responsible for gold deposition along

Proceedings of GeoComputation ‘97 & SIRC ‘97 17 nearest fault. Discrimination between assays in the footwall and hangingwall is made through the sign of the relative distance. A (+) distance indicates points in the hangingwall and (-) in the footwall. LAG-ASSAY: for a selected lag interval h, LAG-ASSAY computes the average and frequency of gold assays within this lag distance relative to the nearest fault position, increasing the searching distance away from the fault surface until all assays are exhausted. The averages are calculated for a normally distributed population of assays as well as a three-parameter lognormally distrib- Figure 1. Spatial representation of a facet and its uted population. In this case the lag h controls the trigonometric attributes amount of smoothing of the distribution and hence is analysis tools other than those which are standard in the called smoothing factor (See Figure 2). mining software. DIP-ASSAY: for a selected dip interval of facets in the fault The name of each module, use and a sample of output surface, DIP-ASSAY computes the average and fre- chart is described below, quency of gold assays associated to these facets. The

FACET-3D: for every facet on a fault surface, FACET-3D averages are calculated for a normally distributed popu- calculates the spatial coordinates of the centroid, azi- lation of assays as well as a three-parameter lognormally muth, dip, dip direction, normal vector, director co- distributed population (See Figure 3). sines, vectorial equation of the plane and fault identifi- AZIM_ASSAY: for a selected azimuth interval of facets in cation (See Figure 1). the fault surface, AZIM_ASSAY computes the average DIST-3D: for a set of spatially distributed gold assays, DIST- and frequency of gold assays associated to these fac- 3D computes the shortest distance to a facet in the ets. The averages are calculated for a normally distrib-

Fig 2 Lag-assay chart: proximity analysis

18 Proceedings of GeoComputation ‘97 & SIRC ‘97 uted population of assays as well as a three-parameter 8- Preliminary results

lognormally distributed population (Figure 4). One important mine “Deposit A”, is examined using these DEPTH-ASSAY: for a selected depth interval, DEPTH-AS- techniques to quantify spatial relationships between gold SAY computes the average and frequency of gold as- mineralisation and structural features. says within this interval relative to the surface level. The averages are calculated for a normally distributed 8.1- Proximity relationships

population of assays as well as a three-parameter For Deposit A, a proximity relationship is identified be- lognormally distributed population (Figure 5). tween high-grade gold mineralisation and the portion of ROCK-ASSAY: for every rock-type present at the miner- the fault hosting that mineralisation. Figure 2 shows that alised site, ROCK-ASSAY computes the average and at a smoothing factor (h) of 10 meters, gold is concen- frequency of gold assays within this rock. The averages trated in economic grades in a narrow corridor around are calculated for a normally distributed population of the hosting fault. The distribution is asymmetric with the assays as well as a three-parameter lognormally dis- highest grades in the hangingwall up to 20 m away from tributed population (Figure 6). the fault surface. In contrast the mineralisation in the STRIKE-BIN: for a selected portion of a fault, STRIKE-BIN footwall is less intense and restricted to the first 10 m,. splits and bins the assays at selected distances from an although the sampling frequency is less abundant in this origin and computes the average and frequency of gold portion of the fault. assays within these bins designed perpendicular to the For the discovery of parallel or secondary mineralised fault strike. The averages are calculated for a normally structures relatives to the main fault, the factor h has to distributed population of assays as well as a three-pa- be related to the sample size n and to the dispersion of rameter lognormally distributed population (Figure 7). the data. The more data available, the more precise is the DIP-AZIM: for selected intervals in dip and azimuth of fac- search for details of the underlying density function. ets in the fault, DIP-AZIM computes the frequency of facets within these intervals for further statistics. Figure 8 demonstrates the effect of the factor (h) on a

Fig 3. Dip-assay chart

Proceedings of GeoComputation ‘97 & SIRC ‘97 19 density estimate of gold-distance to fault. (h = 2.5 m). From for most of the gold in the first 20 m. These peaks corre- Figure 1 it is known that the first 20 m away from the fault, late in depth with two parallel structures hosting high grade in the hangingwall, is highly mineralised, using h = 2.5 m it mineralisation in the south portion of the deposit. These is possible to detect two discrete zones between 5 and minor structures were not incorporated in the model, but 7.5 m and 12.5 and 15 m away from the fault accounting are highlighted using the appropriate h. See fig 7, bins 150

Fig 4. Azimuth-assay chart

Fig 5. Depth-assay chart

20 Proceedings of GeoComputation ‘97 & SIRC ‘97 and 200 for the along-strike extension of this features. Identified a proximity relationship between gold mineralisation and the hosting fault, sections of the fault striking A buffer zone both sides of the fault can be distinguished and dipping in a restricted interval of directions may be according to appropriate cut-off grades to limit the lateral more prospective than others. extension of the mineralised body. Applying an appropriate cut-off for high-grade assays, an i 8.2- Dip-Azimuth relationship x j contingency table of frequency of facets within a par-

Fig 6. Rock-assay chart

Fig 7. Strike-bin chart

Proceedings of GeoComputation ‘97 & SIRC ‘97 21 Fig 8. Effect of choice of smoothing factor h on a density estimate of gold-distance to fault. (h = 2.5 m)

ticular interval of azimuth and dip can be constructed. Based grade related to total facets for the deposit. on the relative proportion of facets in the fault, the ex- If a Dip-Azimuth to high-grade gold relationship is found pected number of facet related to high-grade can be cal- to exist, the Chi-squared component of each Dip-Azimuth culated. These values are the expected if the position of category can be examined to determine which particular the high-grades assays is independent of strike and dip. combination of Dip-Azimuth is more prospective for high- For statistical reasons the expected value Eij for each in- grade values. terval should be greater than 1.0 without endangering the An example of contingency tables for a Chi-square test validity of the test (Conover, 1980). The cells in the contin- for independence between observed and expected facets gency table of expected values with frequencies less than of particular azimuth and dip associated to gold assays > 5 1.0 are condensed into a fewer number of contiguous and ppm, and critical values is shown in Figure 9. logically arranged cells, so that no cell contains less than 1.0 expected facet. The same arrange of cells is then ap- As the statistic is larger than the critical Chi-square value, plied to the observed Oij contingency table. The observed the null hypothesis that both distributions are identical is and expected tables can then be compared using a Chi- rejected at a confidence level of 95%, and consequently a square test for independence with m-1 degrees of free- high-grade dip-azimuth relationship is established. dom, being m the number of condensed cells. Circular or spherical statistical analysis in case of azimuth The test statistic c2 is given by or azimuth and dip data, is required to assess the deviation χ2 Σ Σ ( 2 = Oij-Eij) about the mean direction vector of the fault, of these more Eij prospective sites. These departures can then be used as a where Eij =niCi predictive tool in the search for extensions of present Ni deposits or to target new ones in similar geological condi- ni represents the number of facets for each interval of tions. azimuth-dip in the fault, and Ci/Ni the proportion of high-

22 Proceedings of GeoComputation ‘97 & SIRC ‘97 9- Conclusions ern Australia, Mineralium Deposita 28. pp 366-374.

Quantification of controls on gold mineralisation at camp- Houlding S.W. (1994) 3D Geoscience Modeling. Compu- scale requires a sound three-dimensional understanding ter Techniques for geological characterization, Springer- of the geology and structures involved. Integration of de- Verlag, New York. tailed surface geological maps with subsurface underground Knox-Robinson C.M.(1994) Archeaen lode-gold minerali- mining and direct drilling information, lead to the construc- sation potential of portions of the Yilgarn Block, West- tion of acceptable representations of the three-dimensional ern Australia: development and implementation of geology of the area. These models are used as a base on methodology for the creation of regional-scale which to conduct gold prospectivity analysis. Dedicated prospectivity maps using conventional geological map data handling programs are designed to quantify and ana- data and a geographic information system (GIS). Un- lyse spatial relationships that control known ore bodies. published PhD Thesis, The University of Western Aus- Characteristic features can be identified as more prospec- tralia. tive and consequently used as a predictive tool in the location of new deposits. Mardia K.V. (1972) Statistics of directional data, Academic Press, London, UK 1972. References Rock N.M.S. (1988) Numerical Geology: a source guide, Conover W.J. (1980) Practical non-parametric statistics, 2ed. glossary, and selected bibliography to geological use of John Wiley & Sons, pp 153-168. computers and statistics, Lecture Notes in Earth Sci- Groves D.I. (1993) The crustal continuum model for late- ences 18, Springer-Verlag, New York, 427pp. Archeaen lode-gold deposits of the Yilgarn Block, West-

Fig. 9 Contingency table for Chi-square test of independence

Proceedings of GeoComputation ‘97 & SIRC ‘97 23