Search Documents

By J. H. Tu

The technical paper correctly points out that the kriging variance is not a good measure of the uncertainty of the estimated (i.e., kriged) value of individual blocks. The au thors claim that their proposed jackknife method, which is a rekriging of each block by eliminating, in turn, one sample from the original sample set and then taking the average of the rekriged estimates, not only gives good block estimates, but the resulting jackknife kriging standard deviation is a useful indicator of the "true uncertainty associated with block estimates." However, they immediately abandon the idea of using the block-by-block standard deviations, reasoning that these standard deviations are not independent and that there is no easy way to utilize them. There may be another reason for not using them. The jackknife standard deviations for individual blocks given in their example are mostly in the range of 0.004 to 0.005 oz/st (0.14 to 0.17 g/t) with only one block having a high value of 0.012 oz/st (0.41 g/ t). These individual block standard deviations are as low as the jackknife standard deviation for the mean grade of the entire shape, i.e., 0.0041 oz/st (0.14 g/t). Do they represent the "true uncertainty" of the individual block estimates? Could the authors explain this? In a global shape consisting of a large number of blocks, any given sample will affect the kriged estimate of only those few blocks within its vicinity. This is the rationale for the authors' selective rekriging, making the jackknife algorithm more efficient. On second thought, why not do away with jackknifing altogether? Just cumulate and normalize, if necessary, the kriging weights of each sample used during the ordinary block kriging process, and then compute the global variance from these kriging weights and their respective sample grades? After all, isn't the global mean grade nothing but the weighted average of the samples used in the estimation? Reply by G.S. Adisoma and M.G. Hester The jackknife is one of the many tools in a practitioner's toolbox to solve estimation problems. The strengths of the technique lies in its simplicity, i.e., it uses the concept of mean and standard deviation and the fact that it can be easily combined with other tools, in this case kriging. Because the jackknife kriging (JK) estimate is also the mean of the pseudovalues, the JK standard deviation is attractive just as the standard deviation of the mean explains the variability of the data. The difference is that the pseudovalue calculation in jackknife kriging uses the ordinary kriging (OK) weighting scheme instead of simple arithmetic averaging. The data used to illustrate the jackknife technique in the paper consist of high values that are roughly three times the low values. The resulting JK estimate of the block grades show that the highest estimate is roughly twice the grade of the lowest estimate. The contrast between the low and the high estimate is more evident in the JK estimate than in the OK estimate, even though the mean grades of the blocks for the two estimates are very similar. Nonetheless, in this paper, we are concentrating more on the need for a more realistic measure of uncertainty, or precision, for the estimate. Unlike its OK counterpart, the JK standard deviation of the blocks clearly reflects the original data variation. The highest JK standard deviation of the blocks is three times its lowest value. This follows our intuition that, when the samples used to estimate a block is more variable, the resulting estimation variance (or standard deviation) should be higher than the case where the samples are more uniformly valued. However, block-by-block standard deviation or variance is of little practical value in reserve estimation and classification, as well as in mine planning. One is usually more interested in quantifying not the variance of the individual block estimate, but the uncertainties associated with a much larger dimension, such as the minable reserve. Thus, the thrust of the paper is to find a simple way to obtain a single estimation variance or standard deviation associated with the reserve grade estimate. The discussion by J.H. Tu did not mention how one would obtain the global variance from the OK weights and the sample grades. As a technique that offers a data valuebased measure of uncertainty for its estimate, the "leave-one-out" jackknife fills this need nicely through the block kriging shortcut approach described in the paper. Note: The first column and the last two columns of Table 3 in the paper should have contained a single number each, namely, an OK estimate of 0.0317, a JK estimate of 0.0333 and a JK standard deviation of 0.0041 oz/st, respectively, for the shape, as are obvious from the text. ?

Jan 1, 1998

By E. Tulcanaza

M.S. Azun I do not at all agree with the basic points of the author's conclusion. The use of lognormal or normal model to respond to the attribute distribution function should be carefully questioned. If fitting a distribution function of regionalized variables is the main purpose, one may employ one of certain well-known families of distributions, such as Tukey's lambda functions. For instance, the distribution function of gold assay values studied by Krige (Krige, 1978) may fit the lognormal model well, but one should not exclude that the basic idea of using the distribution function of observed data is to have insight of the physical mechanism that gives rise to the regionalized variables. In Applied Geostatistics, there are many examples considering only lognormal or normal model for the distribution function of geostatistical data even if the observed distribution function has more than one mode. Of course, using lognormal or normal model always brings easy computation for the statistical properties of attribute under investigation. Still, I object to adopting any of those models without having any "statistical inference." As seen from the author's paper, lognormal and normal model produce the same average grade (0.746% copper) and almost the same sample variance. Therefore, which model explains the probabilistic behavior of ore deposit is always an important question that should be replied by the geostatistician using not the common sense but the statistical and probabilistic methods. For kriging procedures, such as linear kriging, one should describe either second order stationary properties or bivariate properties of regionalized variables. Second order properties, such as correlogram function or semivariogram function, are the inputs of the so-called kriging equations. Selecting one of any useable models for the observed second order properties, such as spherical model, is not easy since the models are not based on the correlations structure of the regionalized variables. How those models being used in Applied Geostatistics can be distinguished is another important problem. One may develop many fitting models to respond to the observed second order properties of regionalized variables. However, I suggest that the model should be based on the probabilistic behavior of geologic process. The author used spherical model for copper, wolfram, and silver grades from isotropic sector (Fig. 2, 4, 3b) and exponential model for silver grades obtained along raises in isotropic sector (Fig. 3a). It seems that those semivariograms given in Fig. 3b for isotropic sector and Fig. 4 may be described by the random model implying that regionalized variables do belong to renewal process (Azun, 1983). If I knew the total number of data points used in estimation, I may prove that my conclusion is correct. Recalling the test for the first order sample correlogram value, the test procedure is introduced such that the independence is rejected if [ ] where Za/2 is the a/2 percentile of standard normal random variable and N is the total number of samples used in estimation (Azun, 1983). For the other observed semivariogram functions I suggest the author might try to use the so-called "Markovian model" describing not only the correlation structure of regionalized variables, but also the physical mechanism that produces the regionalized variables. The Markovian model for correlogram function is, [ ] The Markovian model for the other second order properties of ReV are also derived. T ve B, in the above equation, show, for example, structural change and mineralizational variation in a considered deposit, respectively (Azun, 1983). In selecting any model, it is not easy to search for the "best" response to the observed second order properties of ReV's. The Markovian model, based on a theoretical understanding of underlying mechanism, gives more information about the occurrence of regionalized variables and respond to all properties. Random model and the so-called hole effect structure can be easily defined as a special form of the Markovian model (Azun, 1983). An estimation of any second order properties of regionalized variables may be computed through (N-1) lag. However, the dependency between the regionalized variables may be deemed to have "died out" after the so-called stationarity width (range). In practice, the estimation is carried out through 25% of the total number samples available. Therefore, the large fluctuations around the zero level and variance for correlogram or covariogram function and semi-variogram function, respectively, cannot be observed. Since the number of pairs involved in the estimation of higher second order properties is small, the estimation variance at those lags is large. There is no need to show higher order estimated second order properties. After using one of any kriging procedure for block estimation, the distribution function of average grades has the same mean as the drilling sample grades but

Jan 3, 1985

By M. S. Azun

I do not at all agree with the basic points of the author's conclusion. The use of lognormal or normal model to respond to the attribute distribution function should be carefully questioned. If fitting a distribution function of regionalized variables is the main purpose, one may employ one of certain well-known families of distributions, such as Tukey's lambda functions. For instance, the distribution function of gold assay values studied by Krige (Krige, 1978) may fit the lognormal model well, but one should not exclude that the basic idea of using the distribution function of observed data is to have insight of the physical mechanism that gives rise to the regionalized variables. In Applied Geostatistics, there are many examples considering only lognormal or normal model for the distribution function of geostatistical data even if the observed distribution function has more than one mode. Of course, using lognormal or normal model always brings easy computation for the statistical properties of attribute under investigation. Still, I object to adopting any of those models without having any "statistical inference." As seen from the author's paper, lognormal and normal model produce the same average grade (0.746% copper) and almost the same sample variance. Therefore, which model explains the probabilistic behavior of ore deposit is always an important question that should be replied by the geostatistician using not the common sense but the statistical and probabilistic methods. For kriging procedures, such as linear kriging, one should describe either second order stationary properties or bivariate properties of regionalized variables. Second order properties, such as correlogram function or semivariogram function, are the inputs of the so-called kriging equations. Selecting one of any useable models for the observed second order properties, such as spherical model, is not easy since the models are not based on the correlations structure of the regionalized variables. How those models being used in Applied Geostatistics can be distinguished is another important problem. One may develop many fitting models to respond to the observed second order properties of regionalized variables. However, I suggest that the model should be based on the probabilistic behavior of geologic process. The author used spherical model for copper, wolfram, and silver grades from isotropic sector (Fig. 2, 4, 3b) and exponential model for silver grades obtained along raises in isotropic sector (Fig. 3a). It seems that those semivariograms given in Fig. 3b for isotropic sector and Fig. 4 may be described by the random model implying that regionalized variables do belong to renewal process (Azun, 1983). If I knew the total number of data points used in estimation, may prove that my conclusion is correct. Recalling the test for the first order sample correlogram value, the test procedure is introduced such that the independence is rejected if [Ir(1) I > Za/2 N1/2] where Za/2 is the a/2 percentile of standard normal random variable and N is the total number of samples used in estimation (Azun, 1983). For the other observed semivariogram functions I suggest the author might try to use the so-called "Markovian model" describing not only the correlation structure of regionalized variables, but also the physical mechanism that produces the regionalized variables. The Markovian model for correlogram function is, [p (h) = T 1ih , h>1 ,] The Markovian model for the other second order properties of ReV are also derived. [T veß], in the above equation, show, for example, structural change and mineralizational variation in a considered deposit, respectively (Azun, 1983). In selecting any model, it is not easy to search for the "best" response to the observed second order properties of ReV's. The Markovian model, based on a theoretical understanding of underlying mechanism, gives more information about the occurrence of regionalized variables and respond to all properties. Random model and the so-called hole effect structure can be easily defined as a special form of the Markovian model (Azun, 1983). An estimation of any second order properties of regionalized variables may be computed through (N-1) lag. However, the dependency between the regionalized variables may be deemed to have "died out" after the so-called stationarity width (range). In practice, the estimation is carried out through 25% of the total number samples available. Therefore, the large fluctuations around the zero level and variance for correlogram or covariogram function and semivariogram function, respectively, cannot be observed. Since the number of pairs involved in the estimation of higher second order properties is small, the estimation variance at those lags is large. There is no need to show higher order estimated second order properties. After using one of any kriging procedure for block estimation, the distribution function of average grades has the same mean as the drilling sample grades but

Jan 1, 1986

By G. Johnston, P. K. Chatterjee, S. Holguin

M. Gershon The article is an excellent one, providing a basis for the conduct of the major tasks in underground mine planning and scheduling. I object to only one word. Since that word is featured in the title, it is a strong objection. The word used erroneously is optimization. There is no optimization performed in the article. By definition, an optimization is an approach that yields the best of all possible solutions. We can limit this definition to its narrowest sense, the optimization techniques of operations research such as linear programming, or broaden it to include less sophisticated search procedures that find the best mine plan or schedule. No such claim is made in the article. No optimization of any kind is performed. The major tool used is a simulation, which, in truth, is probably more appropriate for this problem than any optimization. But simulation does not optimize. It only allows for quick evaluation of many options, leading to the selection of the best plan "of those tried," not "of all possible plans." This article, of course, is not the first mining paper to make this error. It seems that the word optimization is used more to impress people than to refer to any act of optimizing. It is used in a way that implies "try to optimize"or "come close to optimize," but not to really do it. It is usually appropriate to quietly ignore this error. But when Dr. Chatterjee, who has done so much good work in true optimization projects and whose work I have come to admire, makes such an error, it is time to publicly dispute it. Having discussed this very point with him in the past, I am sure that he will agree. If he can support this position in his response, perhaps people will start to listen and we can reduce the number of false claims of optimization. This should be easy for him in this case since the approach he used, simulation, is the more appropriate one anyway.

Jan 8, 1988

By M. Gershon

The article is an excellent one, providing a basis for the conduct of the major tasks in underground mine planning and scheduling. I object to only one word. Since that word is featured in the title, it is a strong objection. The word used erroneously is optimization. There is no optimization performed in the article. By definition, an optimization is an approach that yields the best of all possible solutions. We can limit this definition to its narrowest sense, the optimization techniques of operations research such as linear programming, or broaden it to include less sophisticated search procedures that find the best mine plan or schedule. No such claim is made in the article. No optimization of any kind is performed. The major tool used is a simulation, which, in truth, is probably more appropriate for this problem than any optimization. But simulation does not optimize. It only allows for quick evaluation of many options, leading to the selection of the best plan "of those tried," not "of all possible plans." This article, of course, is not the first mining paper to make this error. It seems that the word optimization is used more to impress people than to refer to any act of optimizing. It is used in a way that implies "try to optimize" or "come close to optimize," but not to really do it. It is usually appropriate to quietly ignore this error. But when Dr. Chatterjee, who has done so much good work in true optimization projects and whose work I have come to admire, makes such an error, it is time to publicly dispute it. Having discussed this very point with him in the past, I am sure that he will agree. If he can support this position in his response, perhaps people will start to listen and we can reduce the number of false claims of optimization. This should be easy for him in this case since the approach he used, simulation, is the more appropriate one anyway.

Jan 1, 1987

By J. H. Tu

The technical paper correctly points out that the kriging variance is not a good measure of the uncertainty of the estimated (i.e., kriged) value of individual blocks. The authors claim that their proposed jack-knife method, which is a rekriging of each block by eliminating, in turn, one sample from m the original sample set and then taking the average of the rekriged estimates, not only gives good block estimates, but the resulting jackknife kriging standard deviation is a useful indicator of the "true uncertainty associated with block estimates." However, they immediately abandon the idea of using the block-by-block standard deviations, reasoning that these standard deviations are not independent and that there is no easy way to utilize them. There may be another reason for not using them. The jackknife standard deviations for individual blocks given in their example are mostly in the range of 0.004 to 0.005 oz/st (0.14 to 0.17 g/t) with only one block having a high value of 0.012 oz/st (0.41 g/ t). These individual block standard deviations are as low as the jackknife standard deviation for the mean grade of the entire shape, i.e., 0.0041 oz/st (0.14 g/t). Do they represent the "true uncertainty" .of the individual block estimates? Could the authors explain this? In a global shape consisting of a large number of blocks, any given sample will affect the kriged estimate of only those few blocks within its vicinity. This is the rationale for the authors' selective rekriging, making the jackknife algorithm more efficient. On second thought, why not do away with jackknifing altogether? Just cumulate and normalize, if necessary, the kriging weights of each sample used during the ordinary block kriging process, and then compute the global variance from these kriging weights and their respective sample grades? After all, isn't the global mean grade nothing but the weighted average of the samples used in the estimation? Reply by G.S. Adisoma and M.G. Hester The jackknife is one of the many tools in a practitioner's toolbox to solve estimation problems. The strengths of the technique lies in its simplicity, i.e., it uses the concept of mean and standard deviation and the fact that it can be easily combined with other tools, in this case kriging. Because the jackknife kriging (JK) estimate is also the mean of the pseudovalues, the JK standard deviation is attractive just as the standard deviation of the mean explains the variability of the data. The difference is that the pseudovalue calculation in jackknife kriging uses the ordinary kriging (OK) weighting scheme instead of simple arithmetic averaging. The data used to illustrate the jackknife technique in the paper con¬sist of high values that are roughly three times the low values. The resulting JK estimate of the block grades show that the highest estimate is roughly twice the grade of the lowest estimate. The contrast between the low and the high estimate is more evident in the JK estimate than in the OK estimate, even though the mean grades of the blocks for the two estimates are very similar. Nonetheless, in this paper, we are concentrating more on the need for a more realistic measure of uncertainty, or precision, for the estimate. Unlike its OK counterpart, the JK standard deviation of the blocks clearly reflects the original data variation. The highest JK standard deviation of the blocks is three times its lowest value. This follows our intuition that, when the samples used to estimate a block is more variable, the resulting estimation variance (or standard deviation) should be higher than the case where the samples are more uniformly valued. However, block-by-block standard deviation or variance is of little practical value in reserve estimation and classification, as well as in mine planning. One is usually more interested in quantifying not the variance of the individual block estimate, but the uncertainties associated with a much larger dimension, such as the minable reserve. Thus, the thrust of the paper is to find a simple way to obtain a single estimation variance or standard deviation associated with the reserve grade estimate. The discussion by J.H. Tu did not mention how one would obtain the global variance from the OK weights and the sample grades. As a technique that offers a data value-based measure of uncertainty for its estimate, the "leave-one-out" jackknife fills this need nicely through the block kriging shortcut approach described in the paper. Note: The first column and the last two columns of Table 3 in the paper should have contained a single number each, namely, an OK estimate of 0.0317, a JK estimate of 0.0333 and a JK standard deviation of 0.0041 oz/st, respectively, for the shape, as are obvious from the text.

Jan 1, 1997

M.S. Azun I have many objections to the content of the author's paper. Before discussing it, however, I would like to repeat the property of semivariogram function. Second order stationary properties of regionalized variables (ReV's) such as semivariogram function y(h) are perfectly known in geostatistics. Also, the kriging equations in the language of mathematical statistics using second order stationary properties are well understood. However, the way to use the sample (estimated) semivariogram function in any one of the kriging procedures is vague. The sample semivariogram function is given as follows: [ ] where N is the total number of samples, Xi is the sample value at the i - th location, Xi+h is the sample value at the i + h - th location, and h is the distance among the samples. An estimation variance of sample semivariogram function of first lag is smaller than that of higher order lag. The theoretical semivariogram function reaches the variance of samples asymptotically. But this is not easily observable because of the larger variation involved in the estimate of semivariogram function. In general, an estimation procedure is done for h = 0, 1, 2, up to the greatest integer less than N/2, even though sample semivariogram function can be computable through N-1. After estimating semivariogram function, the critical question of how to model sample semivariogram function arises. As seen in the above equation, sample semivariogram function is discrete and can be smoothed by the model being selected. Therefore, modeling of sample semivariogram function is the most important step in geostatistics. It not only smoothes a discrete function but also affects the results of the kriging procedure. When the only aim is to model the semivariogram function, which is the basic point of the author's paper, one can employ any fitting techniques, such as curve fitting, or any arbitrary functions, which are called submodels in the paper. The term "arbitrary function" is used rather than "submodel" because there is no basic understanding of developing them. The author suggests that the sum of those submodels can also be used for the modeling of sample semivariogram function. The combination of any arbitrary functions brings many problems instead of giving an insight of the domain structure considered. The author used two arbitrary functions and the nugget effect in response to sample semivariogram function (Fig. 10). For the same example, he stated that the parameters involved in the mixed arbitrary function model can be accepted when the discrepancy between sample semivariogram function and the model is small visually. For verifying the fitting behavior of any selected model, one should not be contented with the visual satisfactory. Some statistical measure such as goodness of fit has to be used. The author's practice is no more than an exercise in curve fitting without any fundamental understanding or conceptualization of the underlying physical mechanism. Furthermore, the selection of any model is not an easy task if the purpose is the search for the "best" response to the observed second order properties of ReV's. I suggest that the Markovian model (Azun, 1983), on the basis of a theoretical understanding of underlying mechanism, which gives more information about the occurrence of regionalized variables, is used to respond all properties of ReV's. There are a lot of problems for modeling of one-dimensional sample semivariogram function. Thus, it is not appropriate to go to higher order dimensional sample semivariogram function modeling. In the meantime, I would recommend that one can connect the values of standardized sample semivariogram function rather than simple values of semivariogram function in the two-dimensional estimation. The standardized values can be computed in dividing the semivariogram function value by the number of sample pairs involved in each lag regardless of the directions. In conclusion, geostatistics is an interdisciplinary area in mining that uses the principles of mathematical statistics. Thus, it should not violate any probabilistic and statistical rules. When Matheron was developing the theory of geostatistical study in the early years of geostatistics, many mining people had a reservation accepting the geostatistical tools. However, this does not mean that we, the geostatisticians, might try to convince those people using some "strange" tools or rules as some authors implied (Baafi and Kim, 1984). Instead, we have to develop and explain the geostatistical tools staying only in the framework of statistical concepts and properties. References Azun, M.S., 1983, "Stochastic Process Modeling of Spatially Distributed Geostatistical Data," Columbia University, Ph.D. Thesis. Baafi, E.Y., and Kim, Y.C., 1984, "Discussion - Comparison of Different Ore Reserve Estimation Methods Using Conditional Simulation," Mining Engineering, Vol. 36, No. 3, p. 280. Reply by J.M. Rendu The interactive method proposed by Rendu allows practitioners to develop semivariogram models that take into account not only the numerical information obtained by sampling, but also highly significant additional information that often cannot be quantified. The geology of the deposit - including hypotheses concerning its genesis, sampling methods, assaying methods, and mathematical methods used to calculate the semivariograms - all have an influence on the numerical results obtained and on how these results should be interpreted. If all the information concerning the spatial distribution of values in a mineral deposit was contained in the sample values, it could be argued that statistical techniques alone would produce optimum models. However, this is rarely, if ever, the case. Methods that allow the user to take into account his experience and his geologic understanding of the deposit should not be rejected for the sake of theoretical statistical purity.

Jan 11, 1985

By M. S. Azun

M.S. Azun I have many objections to the content of the author's paper. Before discussing it, however, I would like to repeat the property of semivariogram function. Second order stationary properties of regionalized variables (ReV's) such as semivariogram function ?(h) are perfectly known in geostatistics. Also, the kriging equations in the language of mathematical statistics using second order stationary properties are well understood. However, the way to use the sample (estimated) semivariogram function in any one of the kriging procedures is vague. The sample semivariogram function is given as follows: [1 N-hy*(h) = 2(N h) i21 {Xi-Xi+h}Z, h=0, 1, N-1] where N is the total number of samples, Xi is the sample value at the i - th location, X i+h is the sample value at the i +h - th location, and h is the distance among the samples. An estimation variance of sample semivariogram function of first lag is smaller than that of higher order lag. The theoretical semivariogram function reaches the variance of samples asymptotically. But this is not easily observable because of the larger variation involved in the estimate of semivariogram function. In general, an estimation procedure is done for h = 0, 1, 2,…., up to the greatest integer less than N/2, even though sample semivariogram function can be computable through N-1. After estimating semivariogram function, the critical question of how to model sample semivariogram function arises. As seen in the above equation, sample semivariogram function is discrete and can be smoothed by the model being selected. Therefore, modeling of sample semivariogram function is the most important step in geostatistics. It not only smoothes a discrete function but also affects the results of the kriging procedure. When the only aim is to model the semivariogram function, which is the basic point of the author's paper, one can employ any fitting techniques, such as curve fitting, or any ar¬bitrary functions, which are called submodels in the paper. The term "arbitrary function" is used rather than "submodel" because there is no basic understanding of developing them. The author suggests that the sum of those submodels can also be used for the modeling of sample semivariogram function. The combination of any arbitrary functions brings many problems instead of giving an insight of the domain structure considered. The author used two arbitrary functions and the nugget effect in response to sample semivariogram function (Fig. 10). For the same example, he stated that the parameters involved in the mixed arbitrary function model can be accepted when the discrepancy between sample semivariogram function and the model is small visually. For verifying the fitting behavior of any selected model, one should not be contented with the visual satisfactory. Some statistical measure such as goodness of fit has to be used. The author's practice is no more than an exercise in curve fitting without any fundamental understanding or conceptualization of the underlying physical mechanism. Furthermore, the selection of any model is not an easy task if the purpose is the search for the "best" response to the observed second order properties of ReV's. I suggest that the Markovian model (Azun, 1983), on the basis of a theoretical understanding of underlying mechanism, which gives more information about the occurrence of regionalized variables, is used to respond all properties of ReV's. There are a lot of problems for modeling of onedimensional sample semivariogram function. Thus, it is not appropriate to go to higher order dimensional sample semivariogram function modeling. In the meantime, I would recommend that one can connect the values of standardized sample semivariogram function rather than simple values of semivariogram function in the two-dimensional estimation. The standardized values can be computed in dividing the semivariogram function value by the number of sample pairs involved in each lag regardless of the directions. In conclusion, geostatistics is an interdisciplinary area in mining that uses the principles of mathematical statistics. Thus, it should not violate any probabilistic and statistical rules. When Matheron was developing the theory of geostatistical study in the early years of geostatistics, many mining people had a reservation accepting the geostatistical tools. However, this does not mean that we, the geostatisticians, might try to convince those people using some "strange" tools or rules as some authors implied (Baafi and Kim, 1984). Instead, we have to develop and explain the geostatistical tools staying only in the framework of statistical concepts and properties. ? References Azun, M.S., 1983, "Stochastic Process Modeling of Spatially Distributed Geostatistical Data," Columbia University, Ph.D. Thesis. Baafi, E.Y., and Kim, Y.C., 1984, "Discussion - Comparison of Different Ore Reserve Estimation Methods Using Conditional Simulation," Mining Engineering, Vol. 36, No. 3, p. 280. Reply by J.M. Rendu The interactive method proposed by Rendu allows practitioners to develop semivariogram models that take into account not only the numerical information obtained by sampling, but also highly significant additional information that often cannot be quantified. The geology of the deposit - including hypotheses concerning its genesis, sampling methods, assaying methods, and mathematical methods used to calculate the semivariograms - all have an influence on the numerical results obtained and on how these results should be interpreted. If all the information concerning the spatial distribution of values in a mineral deposit was contained in the sample values, it could be argued that statistical techniques alone would produce optimum models. However, this is rarely, if ever, the case. Methods that allow the user to take into account his experience and his geologic understanding of the deposit should not be rejected for the sake of theoretical statistical purity. ?

Jan 1, 1986

Jan 1, 1908

JAmes E. Howard, Watertown, Mass.:—In connection with this paper, it is the desire of the Watertown laboratory to receive suggestions as to the lines of work and the particular direction along which the examination of the ingot-metal should proceed, and any suggestions so offered will be noted with care, and the tests conducted so as to embrace such features as it is thought desirable to have investigated. To me the question of structural continuity seems of para-

Jan 1, 1909

Secretary's Note.—M. Beutter's introductory remarks, being simply a resume of Prof. Hone's paper, with a disclaimer of the intention to criticise it, have been omitted, and the remainder of this contribution lias been somewhat condensed, without omitting any essential statements. M. Beutter, St. Etienne, France* (communication to the Secretary†):—Professor Home observes in $ 47 of his paper that the fluid-compression of steel diminishes piping by forcing the liquid and viscous metal, while the pipe is forming, or after it has formed, into the space so produced, and proceeds to study the different systems of compression, alluding to the four principal ones: Whitworth, Illingmorth, Williams, and Harmet. Having now for many years had in use the last-named process, that of M. Henri Harmet, I shall attempt to explain the difference between it and the three other processes named. The Harmet process does not diminish or close the pipe after it has formed, but prevents it from forming; and I may add that it is the only one of the four which produces this result. It is true that the Whitworth process diminishes the interior hollow by localizing it around the central axis of the ingot, which is afterwards removed by trepanning the hollow forg-ings for which this process has been, I believe, always exclusively used. The Illingworth and Williams processes, which, in my opinion, are applicable only to small ingots, of necessity allow the pipe to form, or (to use the expression of Professor Howe) they allow the surface of the lake of liquid steel to sink into the interior of the ingot, since they involve waiting a certain time before compression, so that (in the case of the Williams process) the ingot-mold can be opened without danger, or (in the case of the Illingworth process) so as to avoid the risk that the steel may enter the grooves produced by the removal of the distance-bars. During the time required for

Jan 1, 1908

R. W. Raymond, New York, N. Y,:—Since the presentation of my note on this subject at the New York Meeting, Professor Louis has pointed out an error in my statement of his conception of " pitch "—namely, he understands the pitch, in the restricted sense which he proposes, to be the angle of inclination of the axis of an ore-body, measured on the vertical plane drawn through the body; whereas, I represented him as proposing to define " pitch " as the angle of inclination of an oblique body, measured on the plane of the dip, in the direction of the strike. The difference will appear upon consideration of the accompanying isometric projection, Fig. 1, in which ABCD represents the plane of the inclosing ore-deposit (vein or bed), and AC the axis of the oblique ore-shoot. ABE is the angle of the dip, as universally understood; and ACE

Jan 1, 1909

Jan 1, 1985

By A. D. Hughes

C. W. MERRILL*-Mr. Hughes' paper not only is very well presented but is most timely in that it covers a subject of vital interest to the United States. Tin is one of the strategic metals which has not been found here in appreciable quantities. The total United States tin output since the first recorded production over 100 years ago would not supply current consumption for a fortnight. On the consumption side, tin is extremely important in national defense because of its use in tin plate which, in the form of tin cans, makes it possible to feed properly the armies that the United States recruits from its civilian population in emergencies. Moreover, the internal combustion engines so important in powering tanks, planes, and other war machines, require large quantities of tin alloys for bearings and for soldering. These considerations, along with the fact that Malaya, historically, has been the leading source of tin to the United States, make the kind of information that Mr. Hughes has presented so ably, a matter of vital interest.

Jan 1, 1949

By Brandon Buschmeier, Frederic Masse

"Inclusions are a type of ground improvement that provide an improvement of a soil mass by insertion of a material with better characteristics than the surrounding soils. There are primarily two types of inclusions: “Soft” inclusions which are constructed of granular material (Stone Columns, Rammed Aggregate Piers, VibroPiers...), and “rigid” inclusions which are constructed of grout inserted with or without pressure (VibroConcrete Columns, Controlled Modulus Columns, Soil Mixing Columns).While the central principle of the inclusions is similar (i.e., overall improvement of characteristics of the soil mass below the structure), the load transfer mechanism is radically different for each technique. Therefore, the applicable design methodology is specific to the type of inclusions and there exists no universal method of design which spans across all inclusions.This paper will present an in-depth comparison between the standard design methods used for “soft” granular-type inclusions (Priebe modified, Balaam & Booker, Goughnour & Bayuk...) and the methods used for “rigid” inclusions which mainly rely on finite element analysis. The paper will explain why the design for soft inclusion methods cannot be applied to rigid inclusions and vice-versa. Each design method will be reviewed in detail to better emphasize its limitations and advantages.Two case histories will also be presented; one for each type of inclusions in order to illustrate the concepts developed in the first part of the paper.1 INTRODUCTION1.1 GENERAL OVERVIEW OF GROUND IMPROVEMENT WITH INCLUSIONSThe use of ground improvement methods to provide the appropriate amount of foundation support is a growing and constantly evolving industry. Traditional structural foundation methods including ‘shallow’ and ‘deep’ foundations do not adequately address the wide range of settlement tolerances, soil conditions, and loading considerations inherent to each unique potential project. In some instances, ‘shallow’ foundation solutions are simply not feasible due to soil and loading constraints. Furthermore, ‘deep’ foundation solutions are used to ‘bridge’ the soft compressible soils, ensuring that the entirety of the load in the structure is transmitted through individual elements into competent layers below the ground surface (bedrock, dense residual soils). Highly concentrated structural loads are transferred into the individual elements by means of direct contact requiring the use of structurally rigid surficial components (structural slabs, grade beams, and pile caps).Ground improvement bridges the gap between these two traditional methods, offering adequate foundation support while using more costeffective techniques and design methodologies. Inclusions are a type of ground improvement used to provide a reinforcement of the soil mass through the vertical insertion of a material with better characteristics than the surrounding soils. Much like deep foundations, inclusions are installed through soil layers with high compressibility and/or low bearing capacity and terminated in a suitable bearing stratum. However, unlike deep foundations, ground improvement inclusions rely on a distribution of load between the soil and the inclusions. By doing so, some of the structural load is applied directly to the soil in a proportion equal to its bearing capacity and compressibility limits, while the rest of the load is shed into the vertical inclusions down to the bearing stratum. Load sharing is a critical aspect in the performance and design of the various ground improvement methods."

Jan 1, 2015

Remarks Of Mr. Robert W. Hunt, General Superintendent, Albany And Rensselaer Iron And Steel Company., Troy N. Y. - In discussing Dr. Dudley's two most interesting papers, I feel a natural hesitancy in disagreeing with conclusions formed after such careful and conscientious work. But I think he has made his deductions from too meagre premises. The analyses of but twenty-five samples are entirely too few to justify the establishment of so important a formula. When we consider that he indirectly proposes to regulate an interest amounting, even at the present low prices, to over $20,000,000 per annum, we may well hesitate before accepting a formula based upon the chemical and physical analyses of but twenty-five steel rails. Dr. Dudley touches the keynote in referring to the proposed labors of the Government Test Commission. They based their hopes of attaining success upon the chemical and physical analyses of a very large number of samples, and in the only branch of investigation which the short-sighted policy of Congress has allowed them to complete, i.e., chain cables, we find that a dozen, twenty-five or one hundred tests gave no satisfactory results. It was only after many hundreds that they were enabled to draw conclusions. There are so many circumstances bearing upon the life of a rail, that the task of extracting a correct result seems herculean. I presume I am right in stating that in our Northern latitudes early winter and spring months witness the breaking of the greatest number of rails, thus seeming to prove that an unstable road-bed has much to answer for. Then, again, broken and flattened wheels must be taken into account, but I am convinced that more rails have been broken by the treatment which they received before leaving the rolling-mill * The remarks of the participants in this debate have been written out by the speakers, and are here given without any attempt to retain the order of the dis¬cussion. While the report does not, therefore, preserve the vigor and spontaneity of the debate, it is believed that it gains in value by giving the matured views of the speakers. The continuation of this discussion at the Pittsburgh Meeting, May, 1879 (beginning with remarks of Mr. Cloud on page 401), is appended for convenience of reference.-T. M. D., Sec'y

Jan 1, 1879

William P. Blake, TUCSON, Ariz. (communication to the Secretary): As one who in youth, now nearly half a century ago, had the privilege of participating in the initial explorations which have alreacly resulted, within a single lifetime, in the establishment and Development of a vast continental empire, so well described by Dr. Douglas, I may be permitted to supplement with some additional details the comprehensive sketch given in his paper. Before the War Department, in 1853, sent out the Pacific railroad exploring parties, Fremont had already made his memorable expeditions, exploring the Great Basin and the passes into California; Sitgreaves had traversed, from east to west, the Zuni and Moqui country; and, still later, Emory had examined the region along the boundary-line between the United States and Mexico. The services of these pioneers deserve to be mentioned in even an outline of the history of continental development. Concerning the five practicable railroad-routes, suggested in the Reports of Explorations aid Surveys," etc., as published

Jan 1, 1900

D. HARRINGTON: Owing to lack of funds, the United States Bureau of Mines have made no experiments in the Bruceton mines in the use of mudite. MR. McAULIFFE: I doubt that mudite cam be delivered at fifty cents a ton and that a mile of heading can be treated with mudite for ten dollars. You could not dig a ton of sand out of a sand hank for that money. L. C. CLARE: The tram road leading to Wattis mine crosses a shale bed and it is easy to screen the shale through a %-inch sieve and load it into a mine car. W. J. REID: A slow trip passing through a mudited entry will dislodge practically no dust, a rapid trip will make air disturbance enough to coat the cars with fine dust. MR. PRYDE: We have used mudite on stoppings around a fire in Reliance and for two months thought they were getting good results, but the fire was found to be drawing air and to be spreading. This mudite was smeared over the stoppings with a trowel. After this apparently failed we used dry adobe.

Jan 1, 1925

Discussion at the Midland Institute of Mining, Civil, and Mechanical Engineers' general meeting held in Leeds, March 21st, 1916, with T. Beach, vice-president, in the chair, of a paper published in Trans. Inst. Min. Engrs., 1916, vol. L., page 509, comprised a written contribution by O'Shea L.T.

Dec 1, 1916

Marius R. Campbell, Washington, D. C. (Communication to the Secretary): Having spent considerable time in a systematic study of this field,'both as regards the details of its structure and stratigraphy and the broader question of its relation to the Appalachian basin, I wish to add a word concerning the correlations made by Mr. Hodge in his paper. From A commercial standpoint, the question of correlation, even along the strike, is of little or no importance, for the variability is such that each opening should be tested individually, regardless of the horizon at which it appears; but these correlations assume an entirely different aspect when their scientific bearing is considered; they then become of vital importance, and, before being accepted, should be subjected to the most rigid scrutiny and severe tests. Mr. Hedge states positively that the Imboden seam of coal is equivalent to the Elkhorn coal of Pike and Letcher counties, Kentucky. In the writer's opinion, such equivalency is, at the utmost, only a possibility. This correlation has been generally adopted by the geologists of the Kentucky Geological Survey, but the writer has been unable to find on what grounds they base the statement. The evidence of identity seems to be in the character of the coal, as both seams furnish fine coking coal; also, in the similarity of the interval between them aid the " Bee Rock," or top of the great conglomerate. It is probably an easy matter to determine the thickness of this interval on Elkhorn creek, and we will assume that the measure of 300 to 350 feet is correct. In the Big Stone Gap field, the question of measuring the interval between the Imboden and the " Bee Rock " is exceedingly difficult. The line of outcrop of this mass of strata is along the northern base of Stone mountain, and, in the uprising of the great arch, the strata on its flanks have been thrown into many minor folds, and, in places, have been broken by local faults. This complicated structure, together with the absence of ally well-marked stratum, renders it exceedingly difficult to obtain a correct measure of the intervening sandstones and shales. The interval of 300 to 350 feet, assigned by Mr. Hodge, is mani-

Jan 1, 1893