Robustness Analysis Of Kirging Estimation Of Mineral Resources

Estimation of mineral/resources reserves is one of important tasks in mineralexploration. The precision of reserves estimation has a direct influence on the miningright evaluation. So it is essential to estimating the mineral reserves correct. As one ofthe most excellent method of estimation of mineral/resources reserves, it requires thatsample data must be subordinated the stable distribution. But the exploration data inpractice is not stable, so the result of estimation is not robust. How to improve therobustness of result when the sample is not stable is the key of this study.It can improve the robustness of the estimation of mineral/resources reserves fromtwo fronts: change the patterns of distribution of samples, improve the method ofsemivariogram calculation. This paper makes a procession of both the sample dataand the method of semivariogram calculation in order to obtain a higher robust resultof reserve estimation.The outliers is one of the most important factors that caused to the non-stationarydistribution of sample. The optimal way of outlier approach is the one that can obtainthe most robustness distribution of sample data. This paper shows that the estimatedneighborhood method (ENM) is more better than the3standard error and theinfluence coefficient method. Sample that after outliers treated with ENM, the logtransfer data’s skewness and kurtosis are-0.33and2.59that close to the requirementof normal distribution that skewness is0and the kurtosis is3. Data that treated thatwith3mean error and CIM, the skewness and kurtosis are0.51,4.88and0.91,5.88that far to the requirement of normal distribution.Theoretical model for the semivariogram is the main tool for Kriging estimationand its precision determines the robustness of the result of estimation. So how toobtain a high precision theoretical model for semivariogram is one of the mostimportant task of Kriging estimation. This paper introduced five methods ofsemivariogram calculation: classic, Cressie and Hawkins, Median AbsoluteDeviation,Huber location estimation and Distance weighted processing. The five methods were used to calculate the semivariogram of sample that before and afteroutliers treated. It also used three different standards that are cross validation,estimation error and comprehensive index to measured the precision of differentresults. It shows that the Cressie and Hawkins can obtain a higher precision model forsmooth distribution data while the Huber location estimation can obtain a highermodel for non stationary data. When takes the data interval into consideration,distance weighted processing can obtain a higher semivariogram model.