Page 192 - Geochemical Anomaly and Mineral Prospectivity Mapping in GIS

194 Chapter 7

cross-validation deposits. The output cross table contains information about number
(Npix) of cross-validation known deposits (au) contained in a class of prospectivity
values (proscl). Values in the Npix column of the cross table are joined to a column
(ndep) in the table histogram of prospectivity values for subsequent calculation of the
cumulative number of deposits (ndepc), total number of deposits (ndept) and the
proportion of deposits per prospectivity class (propdep). Values in the column
propdep are then derived by dividing values in the column ndepc with corresponding
values in the column ndept. Finally, a prediction-rate graph of propdep values versus
proparea values is created.
The prediction-rate curve allows estimation of likelihood of mineral deposit
discovery according to the prospectivity map. Any point along the prediction-rate curve
represents a prediction of prospective zones with a corresponding number of delineated
deposits and number of unit cells or pixels, so the ratio of the former to the latter is
related to the degree of likelihood of mineral deposit occurrence (or discovery) in the
delineated prospective zones. This means that, the higher the value of
propdep÷proparea of predicted prospective zones (Fig. 7-2), the better is the
prediction. It is, therefore, ideal to obtain a mineral prospectivity map with a steep
prediction-rate curve. However, the performance of a mineral prospectivity map is
influenced by (a) the quality of the input spatial data and (b) the way by which evidential
maps are created (i.e., the number of evidential classes per evidential map) and
integrated and, thus, by the modeling technique applied to create a mineral prospectivity
map. We now turn to the concepts of individual modeling techniques that are applicable
to knowledge-driven mapping of mineral prospectivity.

MODELING WITH BINARY EVIDENTIAL MAPS
In this type of modeling, evidential maps representing prospectivity recognition
criteria contain only two classes of evidential scores – maximum evidential score and
minimum evidential score (Figs. 7-1 and 7-3). Maximum evidential score is assigned to
spatial data representing presence of indicative geological features and having optimum
positive spatial association with mineral deposits of the type sought. Minimum evidential
score is assigned to spatial data representing absence of indicative geological features
and lacking positive spatial association with mineral deposits of the type sought. There
are no intermediate evidential scores in modeling with binary evidential maps. This
knowledge-based representation is usually inconsistent with real situations. For example,
whilst certain mineral deposits may actually be associated with certain faults, the
locations of some mineral deposit occurrences indicated in maps are usually, if not
always, the surface projections of their positions in the subsurface 3D-space, whereas the
locations of faults indicated in maps are more-or-less their ‘true’ surface locations (Fig.
7-3). Thus, for locations within the range of distances to such faults where positive
spatial association with mineral deposits is optimal, the evidential scores should not be
uniformly equal to the maximum evidential score. Likewise, for locations beyond the
distance to faults with threshold optimum positive spatial association to the mineral