Presenting the dataset

The Walker Lake dataset (Isaaks and Srivastava, 1989) is derived from a digital elevation model. It consists of the categorical variable T and the numerical variables U and V. The U and V variables will be used in this study, which are preferentially sampled at high-grade zones (Figures 1 and 2) and both present only one population. The exhaustive set has 78,000 measured values.

The statistical parameters for U and V are presented in Table 1.

Table 1 – Statistical parameters for variables U and V.

Results and discussions

Different methods have been tested and their results were compared with a reference model – U (Reference) and V (Reference) – using block size of 5m  5m. Tables 2 and 3 show the statistical parameters for U and V estimates respectively. U (OK) and V (OK) are the results of ordinary kriging of the original values; U (capping Q97) and V (capping Q90) are the results of ordinary kriging after capping the extreme values above the 90^th percentile of the distribution (V = 818.9). The results of median indicator kriging were post processed with postik from GSLIB (Deutsch and Journel, 1998): U (E-type 1.5) and V (E-type 1.5) are the results with a hyperbolic upper tail extrapolation of parameter ω=1.5; and U (E-type 3.0) and V (E-type 3.0) are the results with a hyperbolic upper tail extrapolation of parameter ω=3.0.

Table 2 – Statistical parameters for the U estimates.

Table 3 – Statistical parameters for the V estimates.

For both variables the global average of deposit was best approximated by the FPG technique, especially for the U variable. The correlation coefficients are used as a comparative parameter, as it can be seen in Tables 4 and 5. FPG produces better results, and this improvement is evident when comparing the results with ordinary kriging of the original values.

Table 4 – Correlation coefficients for U.

Table 5 – Correlation coefficients for V.

Conclusion

At this stage FPG provided better results on Walker Lake both globally and locally.

Presenting the dataset

The area containing the deposit is located in the state of Amapá, Brazil. There are three types of mineralisation: sulphides (primary mineralisation), which occur at depth, in fresh rock; saprolite, made of its in situ weathered and oxidised portions; and the colluvial ores, overlaying the saprolite ore as a blanket. The colluvia domain of the deposit was chosen to develop this study, in which the grade distribution reflects the distribution observed in the primary mineralisation. A division of colluvia domain into northern and southern portions was performed, whose statistical parameters are presented in Table 1 and Table 2, respectively.

Table 1 – Statistical parameters of AU in the NORTHERN portion of the deposit.

Table 2 – Statistical parameters of AU in the SOUTHERN portion of the deposit.

Results and discussion

An analysis of the distribution and spatial arrangement shows that two different populations can be recognized: the first ranging from 0.00 ppm to 5.00 ppm and the second, equal or greater than 5.00 ppm, to be treated by FPG using cumulative extensions.

Different methods have been tested and their results were compared. Table 3 and table 4 show the overall statistics of block estimates in northern and southern portions respectively. AU (OK) is the result of ordinary kriging of the original values; AU (FPG) is the result obtained by FPG technique; AU (capping Q95) and AU (capping Q98) are the results of ordinary kriging after capping the extreme values above the 95^th and the 98^th percentile of the distribution (AU = 5.03 ppm and AU = 9.20 ppm, respectively).

Table 3 – Statistical parameters for AU estimates in the NORTHERN portion of the deposit.

Table 4 – Statistical parameters for AU estimates in the SOUTHERN portion of the deposit.

The method yields similar results to classic kriging in the northern part of the deposit. The results show a greater variation in the south, where the variance is greater.

The correlation coefficient is used as a comparative parameter as it can be seen in Table 5 and Table 6. In this case FPG produces results similar to ordinary kriging with a capping in the extreme values, but there is no way to know which top cut is the best.

Table 5 – Correlation coefficients for Amapari NORTH.

Table 6 – Correlation coefficients for Amapari SOUTH.

Conclusion

At this stage FPG provided better results on Walker Lake both globally and locally.