Data-Driven Modeling of Mineral Prospectivity 285
different classes of data is imperative in knowledge-guided data-driven creation of
predictor maps (cf. Roy et al., 2006), not only via application of the data-driven EBFs
but also via application of the other data-driven methods listed in Tables 8-I and 8-II.
Nevertheless, the examples discussed above demonstrate that, provided that the classes
of evidential data are prudently examined and properly calibrated, the applications of
equations (8.8) to (8.10) for data-driven estimation of EBFs result in geologically
meaningful empirical spatial associations between deposit-type locations and indicative
geological features and, thus, are useful in the creation of predictor maps for mineral
prospectivity mapping.
Integration of data-driven EBFs
Data-driven estimates of EBFs are calculated and then stored usually in attribute
tables associated with the individual
X
i
spatial evidence maps (Figs. 8-11, 8-13 and 8-
14). Attribute maps of EBFs (i.e., predictor maps) for each of the
X
i
spatial evidence
maps are then created. Only attribute maps of
Bel
i
, Dis
i
and Unc
i
are used for integration
of predictor maps according to the application of Dempster’s (1968) rule of combination.
We recall from the introduction to EBFs in Chapter 7 that, according to Walley (1987),
Dempster’s (1968) rule of combination is generally neither suitable for combining
evidence from independent observations nor appropriate for combining prior beliefs with
observational evidence. This means that Dempster’s (1968) rule of combining EBFs is
suitable in modeling of mineral prospectivity because predictor maps used in most, if not
all, cases are conditionally dependent with respect to locations of mineral deposits of the
type sought for at least two reasons. Firstly, many predictor maps of mineral
prospectivity are derived from a common data set (e.g., maps of proximity to individual
sets of faults/fractures are derived from a geological map), which means that they are to
some extents ‘observationally’ dependent on each other. Secondly, predictor maps each
represent Earth processes that, at some periods in the geologic time scale and at some
environments in the Earth’s crust, interacted simultaneously with each other and caused
the formation of mineral deposits. Inferences about the inter-play of geological processes
involved in mineralisation can be represented in the logical (or sequential) integration of
predictor maps portrayed as EBFs.
The formulae for combining maps of EBFs via either an AND or an OR operation
(An et al., 1994a), according to Dempster’s (1968) rule of combination, are given in
Chapter 7 (equations (7.14)-(7.16) and (7.17)-(7.19)) and are not repeated here. An AND
or an OR operation represents a function
f in equation (8.2). An inference network is
applied to logically combine predictor maps representing EBFs of two sets of spatial
evidence at a time. An inference network is a series of logical steps, each of which
represents a hypothesis of inter-relationship between two sets of processes (portrayed in
predictor maps) that represent (a) controls on the occurrence of a geo-object (e.g.,
mineral deposits) and/or (b) spatial features that indicate the presence of the geo-object.
The inference network applied in the knowledge-driven Boolean logic modeling (see
Chapter 7, Fig. 7-4) and in the knowledge-driven evidential belief modeling (Chapter 7)