Loading ...
Sorry, an error occurred while loading the content.

AI-GEOSTATS: Summary of questions asked

Expand Messages
  • Mark Burnett Deelkraal
    Hi all First off, to all our American colleagues, my condolences following Tuesdays events. Sorry this is so late, however I hope that it will be of some use.
    Message 1 of 2 , Sep 17 3:44 AM
    • 0 Attachment
      Hi all

      First off, to all our American colleagues, my condolences following Tuesdays
      events.

      Sorry this is so late, however I hope that it will be of some use. Note: I
      have precede all of the replies, I can send the entire text if so desired
      (some of the older messages are no longer available).

      Some background, I work on a Witwatersrand gold mine, which has been in
      operation for approximately 20 years. We mine a tabular ore body, which is
      remarkably continuos and flat (max. dip of 25 degrees). Gold is the primary
      mineral that is exploited, though uranium is also present (though in sub
      economic concentrations).

      I have a sampling data set of + 100 000 points. If I include the data from
      the adjoining mines the size of the data set increases to + 300 000 points.
      For comparison the size of the Carbon Leader data set (from all the gold
      mines in the region) will be almost 1 000 000.

      With this wealth of data, I was wondering why most mines sample the primary
      development (on reef) at intervals of 2.5m and stopes on a regular 5x5m
      grid. The information I turned up suggests that these are historical
      hangovers (like the anecdote that explains why the dimensions used by the
      vehicles used to transport the Space Shuttle, were determined by the width
      of a horses ass back in Roman times).

      It would appear that the panels in a stope average a 10m advance per month
      and are 30m long (on average) suggesting a sampling grid that would be a
      multiple of this mining unit. The next logical step would be to sample the
      primary, on reef development at a sub division of the sampling grid used on
      a day to day basis.

      Obviously population variability (reef thickness and gold value) and
      continuity are not taken into account in this practice. The question of what
      the sample spacing should be, rather than what it currently is, is what I
      wanted to know.

      Here are the responses that I had to my questions. I am afraid this may be a
      little long, so copying the text into e.g. Word and reformatting may be an
      option.

      Thanks a lot to Marcel and Jan-Willem, who gave me a lot of leads and things
      to think about.

      The questions that I had were as follows:

      1. What is the optimum number of samples in a block of any particular
      size?
      2. Does anyone know of a good "idiots guide" to GSLIB?
      3. Is there a good declustering program out there?
      4. Does anyone know where I can get a good introductory text on
      Probability kriging, and is there any software that has been designed to run
      this?

      I will list the responses in reverse order to the questions, as 1.
      Elicited the most response.

      Question 4.
      5. Does anyone know where I can get a good introductory text on
      Probability kriging, and is there any software that has been designed to run
      this?

      As Isobel pointed out, probability Kriging is covered in GSLIB (and
      is referenced in the GSLIB manual on page 87. It is commonly known as
      Multiple Indicator Kriging (MIK).

      (Ref: Deutsch and Journel, GSLIB: Geostatistical Software Library
      and User's guide, Oxford University Press, 1998)

      Apcom 87 has an interesting paper by:

      Kim, Zhao and Roditis, "Performance Comparison of Local Recoverable
      Reserve Estimates Using Different Kriging Techniques, p 65-82.

      They also have a useful set of references, esp.

      Verly, The Multigaussian approach and its application to the
      estimate of local reserves. J math. Geol. V 15, 1983.

      The GSLIB Guide has a good reference section and includes e.g.
      Isaaks, MSc, 1984, Stanford, " Risk qualified mappings for hazardous waste
      sites", which apparently covers some of the aspects of MIK.

      I also asked about software, GSLIB is the obvious one, Surpac
      apparently does MIK as well (I have not investigated this. Datamine does not
      have it as an option, though you could probably write a macro to do so.

      Question 3

      Is there a good declustering program out there?

      Being obtuse, declustering is another term for regularization. GSLIB
      does it using the "declus" sub routine.

      This is the start of the whole "change of support issue". In my
      case, for example, sampling is done every 2.5 m in primary development and
      every 5m (5x5m regular grid) when stoping commences. It is a historical fact
      that raise sampling on the VCR is treated with some degree of caution, 30%
      of sub economic raises are found to be economically viable once ledging has
      been commenced.

      Thus in this case regularizing the data into a 5x5m grid would
      suggest itself.

      As Deutsch and Journel state "without a declustering program most
      geostatistical studies would be flawed from the beginning." GSLIB p. 5

      Question 2

      Does anyone know of a good "idiots guide" to GSLIB?

      This question may have been interpreted by some as being insulting;
      it was not intended to be. There is a good introductory guide on the GSLIB
      web site (can't remember it off hand, a search for GSLIB on Gopher.com will
      take you there).

      Question 1

      6. What is the optimum number of samples in a block of any particular
      size?

      This question got the most response. It would appear that this issue is not
      one that is covered in much detail by earth/ life scientists.

      A search of the Internet turned up quite a few sites, however they were all
      under applied mathematics and statistics. (I will post these later).

      Apcom 87 has two interesting papers specifically related to Witwatersrand
      gold mining:

      Krige, Estimating Grade differentials from drilling results on new gold
      properties. PP 31- 41.

      Gershon, Comparisons of Geostatistical Approaches for Drillhole Site
      Selection. Pp 93-100.


      Sampling:

      Don Meyers replied

      1. If nothing is known about the population distribution, then the
      sampling grid/ method will need to be based on other criteria (e.g. cost)
      2
      . Even if there were no spatial correlation and one would only want to
      estimate an average value for the block, you would still need the variance.

      3. Sample locations, compute the sample variance and use this to
      "predict" the sample size.

      4. We need to remember that the data are being used for two different
      things: a. to calculate the covariogram/ variogram. B to Krige. An optimal
      sampling plan for one will not generally be the optimum for the other.


      Jan Willen van Groenigen replied:

      1. Sampling in geostatistics is paradoxical, the more you know about a
      parameter, the better you can optimize a sampling scheme for it.

      2. Webster and McBratey (refs follow) describe an algorithm for
      calculating the optimum grid spacing for a sampling scheme, given the
      maximum allowed krige variance and a variogram.

      3. Jan Willem has developed a simulated annealing based algorithm that
      optimizes the optimal location of individual sampling points rather than
      optimal grid spacing. A semivariogram needs to be assumed.

      4. The use of Krige variance as a measure of interpolation error is
      controversial, as it does not take into account the value of a sample; just
      it's location. This can give problems if the intrinsic hypothesis does not
      hold.

      5. What is the sampling objective... to describe spatial variability, to
      optimize spatial interpolation and to detect "hot spots"?

      Evan Englund replied

      Optimum depends on your objective function and the distribution you are
      sampling. (Ref to Englund paper under references).

      Benjamin Warr replied

      "Optimum" depends on the objective; clearly exhaustive sampling would
      provide the best estimate, but nay not be the most cost effective.

      Generally the size of the block/ grid is fixed, and the number of samples
      that may be taken is also fixed (cost)

      Often the best sampling method is systematic random or stratified random
      sampling.

      Beware of regular grids because of possible periodicity artifacts.

      Graphs to plot: after Webster and Oliver:

      Average sample spacing v's Krige error

      Sample density vs krigging error for various block sizes.


      The pay off between sample density and estimation precision/accuracy is a
      function of the variogram. If the nugget is small you can expect greater
      accuracy for the same sample density. If the range of the variogram is long
      then the predictive capacity of each sample will be greater than if outside
      of the range of spatial dependence.

      Marcel Vallee replied:

      Note: I have used almost all of Marcell's replies as is, as he replies from
      a mining perspective. I have placed all references in the reference spot.

      In my opinion, the final purpose of sampling is
      Estimation leading to mining extraction that achieves
      optimal metal recovery and minimal dilution. Estimating/
      modeling of the variogram and kriging are interim
      objectives. Geological mapping, interpretation and modeling
      are also essential steps with whom sampling methods,
      sampling grid dimensions and geostatistical aspects should
      be integrated in the determination of global and local
      continuity (Sinclair and Vallée, 1994).

      The basic problem of sampling grid size should be viewed in
      three dimensions. Our usual sampling grids are planned for
      efficiency, using a stratified drilling pattern
      perpendicular to the plane of apparent structural /
      geological continuity. So far, so good!

      However, once we have achieved a first delineation and can
      calculate a semivariogram along the drill hole axis,
      too often we neglect or forget to verify if these
      continuity parameters apply are present in the other two
      perpendicular directions.

      This problem has been detected by Michel David and described
      in a sampling paper titled "What happens if?" given at a
      sampling symposium in Australia in 1976 where he describes
      the problem and recommends sampling specifically designed
      to verify continuity parameters in the second and third
      dimensions. Sadly, there is only one sentence in
      "Geostatistical Ore Reserve Estimation (which was already
      in print at the time this paper was prepared and presented)
      that refers to this problem (around page 200).

      Journel and Huijbregts in "Mining Geostatistics" also
      describe this problem (1978) and recommend a few simple
      tools, for instance to lay out a cross of more closely
      spaced drill holes within the main grid.

      I consider, based on my mining experience, that sampling
      of rock in place should be targeted stratified sampling,
      not random sampling. Regionalized variables require
      regionalized sampling

      When the exploration or mining geologist halves the sampling
      grid dimension, he/she is basically using a similar
      strategy. I know the results of this effort are viewed by
      geostatisticians as the "clustering" problem! For the mining
      geologist, the objective is local estimation and close
      determination of ore limits, whether they are grade contacts
      or sharp contacts both for planning and extraction.

      What is the ideal sampling grid dimension for the mining
      geologist and the mining engineer? My answer: the one
      that allows to plan and develop and extract the ore (from
      stopes or open pit) efficiently as described above. I
      understand the student/researcher cannot rely on similar
      amounts of funds for drilling.

      Another important consideration in the Davil paper and in
      Geosatistical Ore Estimaion is that of sample preparation
      and assaying quality control. Quoting from the paper"
      " ... (frequently) it is the sanple preparation procedure
      whick generates the nugget effect rather than the real
      mineralization which generates the nugget effect"
      [underlined in the text]. This subject makes up almost a
      chapter of Geostatistical Ore Reserve Estimation.

      A higher "induced" nugget effect, that is a higher sampling/
      assaying variance of the values used for a selectin decision
      will reduce the accuracy of LOCAL estimation and mine
      selection.

      What is the final grid dimension (the one we call measured
      resource, proven reserve). This will vary depending on
      orebody configuration and mining method selected. In a open
      pit, with a fairly sizeable, not too complex and not too
      nuggetty ore body, you generally can get away with a wider
      grid for proven ore than yo can in the majority of cases
      underground,

      In an open pit you can sample more systematically,
      from bench to bench than you can in most underground mine.
      Also one has more flexibility for selection by adjusting
      blast limits based on test drilling on the pit floor and
      sampling of blast holes.

      The problem of sampling and geostatistics is that
      geostatistics, except for a few exceptions, has taken over
      the indifferent, if I may say gently, sampling methods of a
      majority of geologists and their frequent lack of concern
      for the "measurable reliability" of sampling (W.E. Deming).
      Here is a recap of my evolution, also quoting events of the
      period and repeating some material from my first note.

      Geostatistical Perspectives and Experiments

      ## In 1973, when drifing a drift for bulk sampling at
      the Niobec deposit, 10 kg control samples were taken from
      drift
      rounds, 3-5 kg channels from drift walls, and diamond drill
      holes placed in the walls in slashes to get samples all
      along the drift (carbonative with disseminated pyrochlore,
      20 to 50 mesh, 0.2% to 1.0% Nb2O5. I had this data
      reviewed by Michel David and he found out that diamond
      drilling semi-variograms were close to those of rounds,
      but those from channel samples were more erratic. I don't
      haveexact figures, as all the files were transferred to the
      mine office in 1975 and they lost tract of them.
      Doing the same now, I would be prescribing larger samples.

      # The sampling contents of Geostatistical Ore Reserve
      Estimation was based on verifications and tests I did in
      1975 at SOQUEM, spurred by questions from Michel David
      (as our geostat consultant), regarding possible causes
      of 'anomalies' on the histograms of gold assay results
      from the gold deposit that became the Doyon Mine
      (east of Royn-Noranda). Incidentally this paper, given at
      the 1976 CIM AGM, was reviewed by the Geological Society
      for CIM Bulletin and turned down, "because all this is in
      books on statistics."

      # David 1976 sampling paper touched both sampling
      methods and sampling strategy; I only found this
      paper a year ago, after Michel David's death when reading
      his full CV when preparing a formal submission by CIM for
      his induction into the Canadian Mining Hall of Fame (he did
      not get inducted in 2001, but we tried again this year for
      the 2002 induction.

      # Journel and Huijbrecths 1978 commented on sampling
      strategy, pointed out the usefulness, or need in some cases,
      of getting variograms in the three main dimensions of a
      deposit; I noticed these comments about 5 or 6 years ago.

      # Burn, R.G. 1979, Data reliability in ore reserve
      asesessments. Mining Magazine, Oct. p. 289-299.
      I think this is a major sampling paper that should be read
      by everybody involved in mining, environment, agriculture,
      and geostatistics, despite the fact it is not very
      quantitative.

      # A 1986 case study by Podolsky (INCO's Exploration V.P.)
      was based on the underground and development and pilot plant
      sampling at the Casa Berardi (Québec) gold deposit; this was
      only published in the Proceedings of the CIM Symposium on
      "Ore Reserve Estimation: Methods, Models and Reality."

      # A 1986 sampling case study by Magri and McKeena (S.A.IMM)
      of channel sampling and diamond saw sampling: "conventional
      channel samples and <strictly supervised channels> showed
      virtually the same semivariograms." However, "to achieve
      the same estimation error with conventional channels as with
      rock saw sampling at a sampling interval of 5 m, it would be
      necessary to have a spacing interval of 2 m to 2.5 m with
      chip sampling."

      Similar observations were made by Podolsky (1986)when
      comparing channel samples with round averages obtained from
      35kg rock samples from all trucks load (Vallée 1992, 1998
      Sampling QC).

      I used a tungsten carbide saw to sample a deposit of
      titaniferous magnetite near Chibougamau in 1982 (too hard
      to chip); 20 years or so later, most exploration teams in
      Québec use such a saw instead channel sampling.

      More Recent Practical Applications

      1992 - The Guide to the Evaluation of Gold Deposits has an
      18 p. chapter on sampling that recommends more efforts for
      representivity, for quality control and ends by reminding
      the geologists and engineers involved of their professional
      and
      legal responsibilities.

      1993 - I was involved, with Al Sinclair from UBC, in two
      papers, at the Michel David Forum (Improved Sampling and
      Data
      Gathering for Improved Mineral Inventory and Production
      Control)and at the APCOM XXIV meeting (Quality Management
      Methods for reliable Estimations of Deposits and Reserves).
      The second paper proposed a reference the quality plan
      framework of the ISO 9000 quality assurance system:

      1) targeted objectives,
      2) responsibility and authority,
      3) activities, methods and procedures,
      4) verifications and audits,
      5) & 6) feedback to quality system for continuous
      improvement.

      1997-1998 - I mentioned the CIM Symposium in my first note.
      The paper Sampling Quality Control is an update of the
      1992-1993 work also quoting from concurrent papers. It
      presented a synthesis of what we should could call the
      sampling process sensu lato:

      sampling stragety
      sample collection
      sub-sampling/sample preparation
      assaying
      verification and feedback at all steps.

      Among the other papers at the Symposium, I would
      particularly recommend
      Long's (Practical Quality Control Procedures . ..),
      Sketchley's (Gold Deposits: Establishing Sampling Protocols
      and Quality control),
      François-Bongarçon's(Extensions to the demonstration of Gy's
      formula) and,
      Sinclair and Benzen (Evaluation of Errors in paired
      analytical data by a linear model).

      1997-2001 / Draft Standards
      Recently, I have been expanding and rewriting the parts
      dealing with geological interpretation, sampling,
      interpretation (I have further developed the material in
      Sampling Quality Control).
      There are also updates to make regarding inventory methods
      and feasibility requirements from another paper:

      Vallée, 2000, Mineral resources + engineering, economic and
      legal feasibility = Ore reserve. CIM Bulletin, V. 93, 1038
      March,p.53-61.

      To conclude, I do not always write such long emails, but
      this is a core subject and I do not encounter frequently
      people who take these problems at heart.

      References

      In no specific order!

      (Thanks esp. to Marcel and Jan Willem for all the references)

      Armstrong and Champigny (1989) " A study in mining small blocks", CIM
      Bulletin, V82, 923, pp 128-133.

      Van Groenigen, PhD.

      Englund and Heravi, 1992, Conditional Simulation: Practical Application for
      Sampling design optimization. In Gerostatistics Trioia '92. Soares ed.
      Kluwer Academic Publishers, Dordreeht, pp 613-624. (Thanks to Wevan for
      sending through a pdf copy of the paper).

      Webster and McBratney (1980's) in the Journal of Soil Science

      Yfantis et. Al. (1987) Efficeincy of Kriging Estimation for square,
      triangular and Hexagonal grids. Mathematical Geology 19 (3), 183:205.

      -Burgess, T.M. and Webster, R., 1984. Optimal sampling strategies for
      mapping soil types I. Distribution of boundary spacings. Journal of Soil
      Science, 35: 641-654.

      -Burgess, T.M. and Webster, R., 1984. Optimal sampling strategies for
      mapping soil types II. Risk functions and sampling intervals. Journal of
      Soil Science, 35: 655-665.

      -Burgess, T.M., Webster, R. and McBratney, A.B., 1981. Optimal
      interpolation and isarithmic mapping of soil properties IV. Sampling
      strategy. Journal of Soil Science, 32: 643-659.

      -McBratney, A.B. and Webster, R., 1981. The design of optimal sampling
      schemes for local estimation and mapping of regionalized variables II.
      Program and examples. Computers and Geosciences, 7(4): 335-365.

      -McBratney, A.B. and Webster, R., 1983. Optimal interpolation and
      isarithmic mapping of soil properties V. Co-regionalization and multiple
      sampling strategy. Journal of Soil Science, 34: 137-162.

      -McBratney, A.B., Webster, R. and Burgess, T.M., 1981. The design of
      optimal sampling schemes for local estimation and mapping of regionalized
      variables I. Theory and methods. Computers and Geosciences, 7(4): 331-334.

      -Webster, R. and Burgess, T.M., 1984. Sampling and bulking strategies for
      estimating soil properties in small regions. Journal of Soil Science, 35:
      127-140.

      van Groenigen and Stein 1988, Constraining optomization of saptial sampling
      using continious simulated annealing, J of Enviromnental Quality, 27 (5),
      1078-1086

      van Groenigen, Siderius and Stein, 1999, Constrained optimisation of soil
      sampling for minimising the kriging varience, Geoderma, 87: 239-259

      van Groenigen, Pieters and Stein, 2000, Optimising spatial sampling for
      multivariate contamination in urban areas, Environmetrics, 11, 227-244

      Journel and Huijbregts "Mining Geostatistics"

      David, "Geostatistical Ore Reserve Estimation"

      David M., 1976, What Happens If? A few remarks on Useful
      Geostatistical Concepts for the Design of Sampling Patterns.
      The Aus. I.M.M Melbourne Branch, Sampling Symposium, Sept.
      1976, preprint of proceedings, 16 p.

      Postolski, T. A., Sinclair, A. J. (1998) Geology as a
      Basis for Refining Semivariogram Models for Porphyry-Type
      Deposits. Exploration and Mining Geology, Vol. 7, Nº 1-2,
      p. 45-50.

      Vallée, M, Dagbert, M, & Côte, D. (1993) Quality control
      requirements for more reliable mineral deposit and reserve
      estimates. CIM Bulletin, vol. 86. No 969, p. 65-74.

      Sinclair, A.J. and Vallée, M. (1994) Reviewing continuity:
      An essential element of quality control for deposit and
      reserve estimation. Exploration and Mining Geology, Vol.3,
      Nº 2, pp. 95-108.

      Vallée, M. (1998) Sampling Quality Control. Exploration and
      Mining Geology, Vol, 7, Nº 1-2, p. 107-116. (

      Vallée, M. (1992) Guide to the evaluation of gold deposits.
      CIM Special Volume 45, Canadian Institute of Mining,
      Metallurgy and Petroleum, Montréal, Canada, 299 p.

      Vallée M. Draft Standards for Exploration and
      Resource/Reserve Estimation.

      The Volume 7, No 1-2 issue of Exploration and Mining Geology
      is a special issue on "Quality Assurance, Continuous
      Improvement and Standards in Resource Estimation.

      Jeff Myer "Geostatiscal Error Management,"

      Vallée, 2000, Mineral resources + engineering, economic and
      legal feasibility = Ore reserve. CIM Bulletin, V. 93, 1038
      March,p.53-61.

      Websites:

      http://agronomy.ucdavis.edu/groenigen

      http://www.u.arizona.edu/~donaldm

      Isobel's Kriging Game:

      (as well as a lot of other interesting stuff )

      http://uk.geocities.com/drisobelclarke/briefcase.html


      In conclusion, thanks again to all those people who have replied to my
      questions. Hope the replies posted here help.

      To those whom I still owe stuff, I will get there eventually...

      Mark Burnett



      **********************************************************************
      This email and any files transmitted with it are confidential and
      intended solely for the use of the individual or entity to whom they
      are addressed. If you have received this email in error please notify
      the system manager.

      This footnote also confirms that this email message has been swept by
      MIMEsweeper for the presence of computer viruses.

      www.netcom.co.za
      **********************************************************************

      --
      * To post a message to the list, send it to ai-geostats@...
      * As a general service to the users, please remember to post a summary of any useful responses to your questions.
      * To unsubscribe, send an email to majordomo@... with no subject and "unsubscribe ai-geostats" followed by "end" on the next line in the message body. DO NOT SEND Subscribe/Unsubscribe requests to the list
      * Support to the list is provided at http://www.ai-geostats.org
    • Digby Millikan
      Mark, We and I speak on behalf of Australians that we also offer our sincere condolences for this terrible tragedy in the United States. Regarding your
      Message 2 of 2 , Sep 18 4:13 AM
      • 0 Attachment
        Mark,
        We and I speak on behalf of Australians that we also offer our sincere
        condolences for this terrible tragedy in the United States.

        Regarding your questions;
        You can determine a sample spacing based on what confidence you want in
        estimates e.g. Isobel Clark has pointed out that for measured you may choose
        the kriging variance for your estimates is less than the sample variance.
        There
        are a number of published figures of kriging variance relating to resource
        classification.
        As you say this may be a hangover from sample spacings chosen by D.G. Krige
        determined many years ago which hence have been adopted as a standard on
        assumption of similar gold variability among these mines.

        Hope these comments are of some help to you (at least me!)

        Regards Digby Millikan
        Geolite Mining Systems
        www.ozemail.com.au
        digbym@...



        --
        * To post a message to the list, send it to ai-geostats@...
        * As a general service to the users, please remember to post a summary of any useful responses to your questions.
        * To unsubscribe, send an email to majordomo@... with no subject and "unsubscribe ai-geostats" followed by "end" on the next line in the message body. DO NOT SEND Subscribe/Unsubscribe requests to the list
        * Support to the list is provided at http://www.ai-geostats.org
      Your message has been successfully submitted and would be delivered to recipients shortly.