Multivariate geostatistical analysis: an application to ore body evaluation
محورهای موضوعی : MineralogyMohammad Maleki 1 , Nasser Madani 2
1 - Department of Mining Engineering, University of Chile, Santiago, Chile
2 - Young Researchers Club, South Tehran Branch, Islamic Azad University, Tehran, Iran
کلید واژه: Correlation, Geostatistical simulation, MAF, co-simulation,
چکیده مقاله :
It is now common in the mining industry to deal with several correlated attributes, which need to be jointly simulated in order to reproduce their correlations and assess the multivariate grade risk reasonably. Approaches to multivariate simulation which remove the correlation between attributes of interest prior to simulate and then re-impose the relationship afterward have been gaining popularity over the more common joint simulation methods because of their better accuracy and computational efficiency as the number of attribute being simulated increases. Principal component analysis (PCA) is one of these approaches. However, PCA suffers from some drawbacks such as the factors that are uncorrelated just for collocated locations. Minimum/ maximum autocorrelation factors (MAF) is a modification of PCA approach which the factors are uncorrelated for two lags. As an expectation, when linear co-regionalization model contains only two nested structures, the factors do not have any spatial correlations. The main aim of this research is to compare the results of MAF approach with some traditional approach for multivariate simulation (Co-simulation and independent simulation approach). To this end, two variables have been simulated with three different methods and are then compared together based on some yardsticks such as ability to reproduce the original correlation coefficient between two variables.
It is now a common practice in the mining industry to deal with several correlated exploratory attributes, which need to be jointly simulated to reproduce their correlations and assess the multivariate grade risk. Approaches to multivariate simulation, which remove the correlation between attributes of interest prior to simulation and re-imposing of the relationship afterward, have been gaining popularity over the more common joint simulation methodologies. This is due to their better accuracy and computational efficiency as the number of attributes for simulation increases. Principal component analysis (PCA) is one of these approaches. However, PCA suffers from some drawbacks such as the factors that are uncorrelated just for collocated locations. Minimum/maximum autocorrelation factors (MAF) is a modification of the PCA method where the factors are uncorrelated for two lags. As an expectation, when the linear co-regionalization model contains only two nested structures, the factors do not have any spatial correlations. The main aim of this research is to compare the results of the MAF approach with some traditional approaches for multivariate simulation (co-simulation and independent simulation approaches). To this end, two variables have been simulated with three different methods and are then compared based on some yardsticks such as the ability to reproduce the original correlation coefficient between two variables. The results showed that MAF has the capability to reproduce the intrinsic correlation between the variables.