Geoscience Reference
In-Depth Information
Table 14.8
TCu Kriging plans, by domain, 2003 resource model, final iteration
6HDUFK
5RWDWLRQ
$QJOHV
0LQ1R
2I
&RPSV
0D[1R
2I
&RPSV
0LQ1R
2I
2FWDQWV
0LQ1RRI
&RPSRVLWHV
SHU2FWDQW
0D[1RRI
&RPSRVLWHV
SHU2FWDQW
<¶;¶=¶
VHDUFKUDGLLP
&RPSRVLWHV
XVHG
'
RPDLQ3DVV
holes within a pre-defined volume representing approxi-
mately the last two years of production where averaged into
20 × 20 × 10 m blocks. These same blocks were re-estimated
iteratively, using alternative kriging plans, until a reasonable
match between kriged estimates and block-averaged blast
holes was achieved. It is advisable to validate the kriging
plans any time that a reference grade distribution is avail-
able, which is different than using geostatistical cross vali-
dation methods. Classical cross-validation methods do not
directly validate block estimates, and have been questioned
as tools for validation of kriging parameters (Davis
1987
;
Solow
1990
).
Block kriging was performed by discretizing the block
with 4 × 4 × 1 points. Note that, in the vertical dimension, no
discretization is necessary because 10 m composites were
used to estimate 10-m high blocks. Octant searches were
used, this to aid kriging in the declustering process. The
search ellipsoids were oriented according to the correlogram
models, although the search anisotropies were less marked.
This is done to ensure using composites from directions of
lesser continuity, which contribute useful information and
could be discarded if a very strong anisotropic search is used.
The variable orientations of the search ellipsoids correspond
to the orientations of the correlogram model structures. This
is because it is assumed that the anisotropy observed in the
first correlogram structures is mostly influenced by superim-
posed short-scale controls, such as cross faulting. Therefore,
the kriging passes with smaller search radii are influenced by
the first structure of the correlogram model, while the krig-
ing passes that use longer search radii are mostly influenced
by the second structure.
The estimation parameters used for each Domain for TCu
are shown in Table
14.8
. In most cases, all boundaries be-
tween Domains were treated as hard boundaries, using only
composites from that domain to estimate within the domain
(see last column in Table
14.8
). However, there were some
