Title: Message
Gayle,
 
 I can't give you any solution but can inform you of my past experience;
 
    - Gold deposits normally display skewed distributions, so that if the
      assays are used for modelling (usually block modelling) the result
      will be that the grade will be overestimated.
    - Lognormal kriging was developed to overcome this. You krige the
      natural logarithm (not base 10) of the grades then back transform
      the results making an adjustment for the variance.
    - The problem with lognormal kriging is that it is highly sensitive to
       your interpreted variograms, and infact the error in your grade
       estimate is directly proportional to the error in your variogram
       e.g. sill estimation.
       Hence lognormal kriging is best reserved for deposits which show
       very well formed lognormal variograms.
    - Gold data of course is not always strictly lognormal and often 
       shows mixed distribution characteristics (you may use 
       disjunctive kriging for mixed distributions which you may like for a 
       comparison with other estimation methods) which worsens the
       prospects of accurate modelling.
    - In such cases as a non strictly lognormal population and poor
       variograms which may be the case the old hand method is to cut
       the grade population prior to modelling then just use inverse
      distance squared or cubed (experience that cubed is better for
      skewed gold distributions) or ordinary kriging modelling.
    - Determing the cut value is the problem. Some use a cut value
       based on experience or reconciliation of deposits in the region.
       I have used a method whereby I calculate the sichel mean of  
       the dataset (an estimate of the true mean of a lognormally
       distributed population from a sample dataset) then cut the dataset
       until the arithmetic mean of the cut dataset equals the previously
       calculated sichel mean.
       I am currently working on an improved top cut method 
       calculation in association with Frans Manns.
 
   Software used is any generalised mining package for the modelling
   and for calculation of the topcut, if you don't have a macro based
   software or are intimately familiar with a database package,
   spreadsheet software will suffice.
 
   I have not checked or do not know if there is any software in the
  Stanford University GSLIB software library for treatment or 
  processing of mixed populations.
 
  If you would like further details on any of the above processes please
 feel free to contact me.  
 
 
Regards Digby J. Millikan BEng.
* By using the ai-geostats mailing list you agree to follow its rules 
( see http://www.ai-geostats.org/help_ai-geostats.htm )

* To unsubscribe to ai-geostats, send the following in the subject or in the 
body (plain text format) of an email message to [EMAIL PROTECTED]

Signoff ai-geostats

Reply via email to