The idea
Every explorer combines datasets. The usual way is to lay maps over one another and look for places where anomalies coincide. Prospectivity mapping does the same thing with the rules written down: which features count as evidence, how much each counts, and how they are combined. The output is a map in which each cell has a score.
Writing the rules down has two effects. The result can be repeated and updated when new data arrives, and it can be argued with, because anyone can see why a given area scored highly.
Start from the deposit model
Evidence only means something in relation to the kind of deposit being sought. A porphyry copper system, an orogenic gold vein and a sediment-hosted zinc deposit leave different footprints. The mineral-systems approach asks what any deposit of the chosen type needs, namely a source of metal and fluid, a pathway, a trap and a mechanism of deposition, and then asks which mappable features stand for each.
Typical evidence layers are distance to a class of fault or intrusion, favourable host rocks, geochemical anomalies in soil or stream sediment, magnetic or radiometric signatures, and alteration mapped from satellite images. Each is derived from raw data, and the choices made in deriving it matter as much as the method that combines them.
Two families of method
Knowledge-driven methods use expert judgement to set the weights. Index overlay and fuzzy logic are the common ones: each layer is scaled between unfavourable and favourable, and the layers are combined with stated operators. They need no known deposits to learn from, which suits early-stage districts. Their weakness is that they encode the expert's assumptions, including the wrong ones.
Data-driven methods learn the weights from known mineral occurrences. Weights of evidence and logistic regression are the classical ones; random forests and other machine-learning methods are now widely used. They can find relationships an expert would not specify. Their weakness is that they need enough known occurrences, and those occurrences are not a fair sample: they cluster where outcrop is good, where roads run and where people have already looked.
Hybrid approaches use expert knowledge to choose and shape the layers and data to tune the weights. With a handful of known occurrences, the honest choice is usually towards the knowledge-driven end.
How a map is tested
A prospectivity map that has not been tested is an opinion in colour. The standard test is to withhold some known occurrences, build the map without them, and check whether they fall in high-scoring ground. If the top tenth of the area captures most of the withheld occurrences, the map is doing useful work.
Two precautions are needed. Withheld points must be separated in space from the points used to build the map, by holding out whole blocks, because neighbouring points share the same geology and make any method look better than it is. And the test should be reported with its numbers, not summarised as “validated”.
What the score means
A prospectivity score is a ranking, not a probability of discovery. A cell scoring 0.9 has more supporting evidence than a cell scoring 0.5 under the stated model. It does not have a ninety per cent chance of hosting a deposit.
Several other limits follow from how the map is made. It can only reflect the deposit model chosen; a different style of mineralisation will be ranked low however real it is. It inherits the coverage of its inputs, so ground with no geochemical sampling scores low for lack of data, not lack of metal. And it is only as fine as its coarsest important layer.
Using one in practice
For a small company, the practical uses are plain: deciding which part of a large title to work first, choosing between projects, placing a sampling grid, and explaining to investors why the programme starts where it does. In each case the useful output is not the map alone but the map with its reasons: which layers drive each high-ranking area, and which missing data would change the picture most.
Gossan builds prospectivity maps this way as part of multi-source insight, and turns the ranking into holes under drill targeting.