XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3 1049
locations. To estimate this function, one needs to assume
some form of stationarity, that is: that this dependence
structure does not primarily depend on the actual locations
of the samples being compared, but rather on the geo-
graphic distance between them. In a TSF, this assumption
is highly questionable, but necessary here given the absence
of a better model. A model was fitted to the experimen-
tal variogram from the class of power variogram models,
known to be able to account for non-stationarities as those
expected in the case of this TSF. The variogram parameters
were chosen as those producing higher correlations between
observed and estimated values within a leave-one-out cross-
validation routine: in this exercise, every datum is removed
from the data set in turn and predicted using the rest of
the data.
With this variogram model and the available data at all
sampling locations, we use kriging to interpolate the three
log ratios (Eq. 2) at all points in a dense grid within the
TSF. Since Eq. (2) represents an invertible transformation,
we compute the corresponding mass pulls (Msc to Mgt) at
each grid point and calculate the resulting grades and recov-
eries with Eq. (1).
RESULTS
Figure 3A displays the modal composition of the different
composite samples as bulk as well as per size class. Pyrite
and arsenopyrite, the two most abundant sulfide miner-
als in all composite samples, have their lowest and high-
est grades respectively in Family 1 (1.3 &0.3 wt.%) and
Family 3 (7.0 &1.4 wt.%). Sphalerite, chalcopyrite, and
galena occur as minor minerals in all samples (0.5 wt.%).
Figure 3. A) The modal mineralogy of each composite sample, as a bulk and per size class. B) The distribution of a few selected
minerals in the different size classes. Results obtained with the MLA
locations. To estimate this function, one needs to assume
some form of stationarity, that is: that this dependence
structure does not primarily depend on the actual locations
of the samples being compared, but rather on the geo-
graphic distance between them. In a TSF, this assumption
is highly questionable, but necessary here given the absence
of a better model. A model was fitted to the experimen-
tal variogram from the class of power variogram models,
known to be able to account for non-stationarities as those
expected in the case of this TSF. The variogram parameters
were chosen as those producing higher correlations between
observed and estimated values within a leave-one-out cross-
validation routine: in this exercise, every datum is removed
from the data set in turn and predicted using the rest of
the data.
With this variogram model and the available data at all
sampling locations, we use kriging to interpolate the three
log ratios (Eq. 2) at all points in a dense grid within the
TSF. Since Eq. (2) represents an invertible transformation,
we compute the corresponding mass pulls (Msc to Mgt) at
each grid point and calculate the resulting grades and recov-
eries with Eq. (1).
RESULTS
Figure 3A displays the modal composition of the different
composite samples as bulk as well as per size class. Pyrite
and arsenopyrite, the two most abundant sulfide miner-
als in all composite samples, have their lowest and high-
est grades respectively in Family 1 (1.3 &0.3 wt.%) and
Family 3 (7.0 &1.4 wt.%). Sphalerite, chalcopyrite, and
galena occur as minor minerals in all samples (0.5 wt.%).
Figure 3. A) The modal mineralogy of each composite sample, as a bulk and per size class. B) The distribution of a few selected
minerals in the different size classes. Results obtained with the MLA