The answers I have received about the subject Ore Reserves Classification are listed below:

Thanks everyone.

José Quintín

José Quintín Cuador Gil

Computer Department

University of Pinar del Río

Cuba

<cuador@...>

******************************************************************************************************

Question:

Dear list members

The Kriging variance has some uses. In mining, it can be used in the Ore Reserves Classification.

What is the opinion about this in the Geostatistical community?

It is possible to use the Kriging variance for ores reserves classification?, (Yes or No).

Thanks in advances for any opinion.

José Quintín Cuador Gil

Computer Department

University of Pinar del Río

Cuba

<cuador@...>

Answers:

******************************************************************************************************

Form Joao Felipe

yes, have a look at APCOM 2002 proceedings and you'll find a couple a papers in the subject including one I'm a senior authors. If you don't have access to the proceedings I'll be glad to send you an electronic copy

regards

--------------------------------------------------------------------------------------------------------

Joao Felipe Coimbra Leite Costa, Mining Engineer, MSc, PhD

Mining Engineering Department

Federal University of Rio Grande do Sul

Av. Osvaldo Aranha 99/504 90035-190

Porto Alegre, Brazil

+55 +51 33163357

Fax +55 +51 32864343

Home +55 +51 32229321

-------------------------------------------------------------------------------------------------------

******************************************************************************************************

From Marcelo Godoy

Hi José,

Yes, it is possible! But you have to keep in mind the fact that

estimation variance is just a function of sample density. It does

not reflect spatial variability at all. In the latest edition of

the APCOM Symposium several papers have been presented that explore

the issue of Ore Reserve Classification.

Cheers,

Marcelo

-----------------------------------------------------------------

Marcelo C. Godoy, MSc, PhD candidate

University of Queensland Tel: +61 7 3365 1674

W.H. Bryan Mining Geology Research Centre Fax : +61 7 3365 7028

Brisbane, Qld 4072, Australia Home: +61 7 3870 7069

E-mail: m.godoy@...

http://www.minmet.uq.edu.au/~bryan/staff/marcelo.html

-----------------------------------------------------------------

******************************************************************************************************

Form Turkan Kaynak <turkan@...>

Dear Jose,

I'm a mining engineer and I think I can answer your question. We use kriging variance for reserve classification. I think you know, the reserves are classified depend on their error quantity and this quantity can be represented using variance. If we use kriging for reserve estimation you can use kriging variance and classified reserves as possible, probable or proved reserves

******************************************************************************************************

Form Mark Burnett Deelkraal <MBurnett@...>

Dear Jose

Isobel Clarke will probably also comment on this question, however I would

suggest you have a look in the archives at the ai-geostats home page. This

question is generally raised at least once a year.

My personal experience is that using the krige slope of regression (0.6)

works well for ordinary kriging on a Witwatersrand ore body (as long as you

are not trying to estimates more than 10 to 20m ahead of your data. Simple

krige with varying local area means would need a more laborious process,

however here I have found that the Kvar does work.

It still boils down to the quality of your data and a solid understanding of

your ore body.

Hope this helps

Mark

M. Burnett

Ore Reserve Manager

Elandskraal

Production Unit 1 (Deelkraal)

Tel. 018 785 6625

www.Harmony.co.za

******************************************************************************************************

Form Luis Eduardo de Souza <esouza@...>

The estimate and the subsequent classification of the resources in different classes or categories is based on different levels of risk and requires a model able to quantify this risk for evaluation and classification of mineral resources a long time ago.

All classification systems share some common aspects in terms of defining the classes of resources based on distance separating samples and on the degree of confidence or accuracy associated with the results reported. Despite of being very clear in terms of stating sample distances, all the systems of classification do not provide clear definitions on how confidence limits should be calculated.

While the ordinary kriging allows a fast response to determine tonnages, the error calculated requires a series of assumptions which in various cases are difficult to be sustained.

Care must be taken when assigning confidence intervals with a predetermined distribution of the kriging errors. In practice, estimation errors are rarely normally distributed and likewise a lognormal model is just a approximation.

Another drawback of estimation is that the interpolation algorithms tend to smooth out details of the spatial variation of the attribute, where small values are overestimated and large values are underestimated, don´t allowing a realistic evaluation of the uncertainty associated with the estimate.

_________________________________________________________

Luis Eduardo de Souza, Mining Engineer

e-mail: esouza@..., esouzabr@...

Federal University of Rio Grande do Sul - UFRGS

Mining Engineering Department

Mineral Research and Mine Planning Laboratory

Av. Osvaldo Aranha, 99/511

Porto Alegre/RS - Brazil - CEP: 90035-190

Phones:+55 51 3316-3594 (office),+55 51 3333-8229 (home),

+55 51 9905-6587 (cellular)

home-page: http://www.lapes.ufrgs.br/Pessoal/eduardo

_________________________________________________________

******************************************************************************************************

Form Richard Hague <richardh@...>

List Members,

The use of the kriging variance to categorise/classify Mineral (Ore) Resources and/or Ore Reserves is an old chestnut that periodically raises it's ugly head. The kriging variance is related, pure and simply, to the data configuration and has nothing to do with the sample grades/variables being used for interpolation. As an example say a grade was being interpolated into a block which has been sampled on each corner, regardless of what the individual sample grades are, the kriging variance for that block is going to be the same. Example: if all four samples have the same grade of (say) 2.35g/t Au you will get the same kriging variance as the case where the four samples grades are (say) 0.01, 102.9, 0.88 and 3.60 g/t Au. Naturally I would have more confidence in the interpolated grade in the former scenario than the latter; thus the use of the kriging variance to determine confidence (or classification) of an estimate is misleading.

One method of obtaining some feel for the possible error range would be to run a large number of grade simulations into the block, the variance of all simulated grades would give an indication of error - again in the example given above, the variance of the simulated grades using the former case would be much smaller than in the latter case.

Of course classification of Mineral (Ore) Resources and/or Ore Reserves needs to take into account a lot more items (as expounded by the JORC (1999) code) - than just some objective measure of estimation error, it needs to take into consideration, amongst other things, data quality - if you have poor quality data (eg biased/inaccurate), regardless of how good any statistical measure of the estimation error is, you will always have poor estimate.

REFERENCES

JORC; 1999: Australasian code for reporting of mineral resources and ore reserves (the JORC Code). Prepared by the Joint Ore Reserves Committee of the Australasian Institute of Mining and Metallurgy, Australian Institute of Geoscientists and Minerals Council of Australia (JORC).

Richard Hague

Hellman & Schofield Pty Ltd

Brisbane Office

p&f: +61 (0)7 3217 7355

e: richardh@...

w: http://www.hellscho.com.au

******************************************************************************************************

From Isobel Clark <drisobelclark@...>

Richard

Thanks for the clear exposition on the limitation of

the kriging variance as a measure of reliability for

block estimation.

It should, perhaps, be pointed out that the kriging

variance is what we minimise and hence, surely, some

measure of reliability? The whole geometry versus

variability thing has been at issue since Philips and

Watson provided their seminal (sic) paper in 1986.

Given consistent data quality and a Normal (gaussian)

distribution, geometry is what determines likely

error. Under those circumstances, 1000 simulations

will yield an average of the kriged value and a

standard deviation equal to the kriging standard

deviation.

If the data quality is not consistent and the

distribution of values is not Gaussian, then your

comments hold particular force. Since these are the

circumstances under which I labour daily, I would

appreciate any and all suggestions as to what we use

instead. Simulation is not an option when you have

hundreds of thousands of blocks and a limited time to

produce a reserve.

Isobel Clark

http://www.stokos.demon.co.uk

[Non-text portions of this message have been removed]