What Is Mineral Prospectivity Mapping?

The core problem in mineral exploration

Finding a mineral deposit is really a prediction problem about place. A geologist and a computer model face the same question. Based on everything we know about the geology of an area, where is mineralisation most likely to be?

The answer is never obvious. Mineral deposits are rare. They only form when several things line up in the right way over millions of years: a heat source, paths for fluids to move through, the right chemistry, and structures that trap the metals. This mix leaves clues across many kinds of data, such as geophysical readings, soil chemistry, rock structures, and altered zones. Reading these clues by hand, one layer at a time, is slow. It depends on expert skill, and one person can only hold so many variables in mind at once.

Prospectivity mapping is the industry's organised answer to this. Instead of looking at one dataset at a time, it brings all the evidence together into a single map. That map shows how likely mineralisation is at every point across the area.

How traditional prospectivity mapping works

In the traditional method, a geoscientist first builds a model of the target deposit. The model lists the geological signs that should appear near that kind of deposit, such as being close to intrusions, the right host rock, certain altered minerals, and a particular geophysical response. The geologist then scores each sign across the map and adds the scores together. This is done by hand or with mapping software (GIS), and the result is a prospectivity map.

This method works, and it holds decades of geological knowledge. But it has limits. It is only as good as the model the geologist starts with. It is hard to use over large areas or with many data layers. And it treats each sign on its own, even though in real geology the signs affect one another. Sometimes three medium clues across three datasets matter more than one strong clue by itself. That kind of link is hard to catch when you combine layers by hand.

What AI prospectivity mapping does differently

AI prospectivity mapping turns this around. The geologist does not decide which signs matter. Instead, the model learns them from the data. It studies the mix of features that shows up near known deposits, or near drill holes that hit mineralisation.

The model is trained on the places where deposits are already known. It learns which patterns tend to appear together with mineralisation, looking at geophysics, soil chemistry, rock structure, rock type, and satellite data all at the same time. It then checks every other point on the map and gives each one a probability score.

The result is not a simple yes or no. It is a smooth map of scores, where a higher score means the spot looks more like the places near known deposits. A geologist can look at the map, find groups of high-score points, see which data layers caused each group, and then decide whether the geology really supports drilling there.

The class imbalance problem

The hardest technical problem here is something called class imbalance. Known deposits are rare, often only one to three in a whole district, while almost all of the ground holds nothing. A simple model trained on this data would just learn to say 'no deposit' everywhere. It would score as highly accurate and still be useless.

The industry has built a few ways to handle this. One is positive-unlabelled learning. It treats the empty ground as 'unknown' rather than 'no deposit', because some of it may hide deposits we have not found yet. Another is to train many models on different samples of the data and combine their answers, which makes the result steadier when known deposits are few. A third is spatial cross-validation. Here the model is tested on separate areas of the map, not on random points. Random points sit too close to the training points, and that would make the model look better than it really is.

These steps are not optional extras. Without them, a prospectivity model can look strong on paper and still fail in the field.

What a prospectivity map looks like in practice

A finished prospectivity map is an image file, usually a GeoTIFF. Each pixel holds a score between 0 and 1 that gives the chance of mineralisation there. The map is smooth, with no sharp lines between high and low areas. High-score pixels tend to group together and form target corridors. A geologist can then compare these corridors with known structures and earlier geological work.

The map comes with an uncertainty map that shows how sure the model is at each point. Uncertainty is high where there is little training data, or where different model runs disagree. The two maps work together. A point with a high score and low uncertainty is a strong drill target you can defend. A point with a high score but high uncertainty is a sign to collect more data first, such as another geophysical line or more soil samples, before you spend money on drilling.

What prospectivity mapping cannot do

AI prospectivity mapping finds statistical patterns. It points to places where the mix of features looks like the places near known deposits. It does not understand how geology works. It cannot reason about rock structures the way a senior geologist can. It does not know that a geophysical reading comes from a mapped intrusion rather than a body of sulfide. And in areas unlike anything in its training data, it performs worse.

The map is a tool for setting priorities, not a decision to drill. A qualified geoscientist should always review it before it guides a drill program. The model shows what the data says, and the geologist decides what it means. Used well, AI prospectivity mapping gives the geologist more reach. It covers more ground, compares more data, and finds patterns that are too subtle to see by hand. The final judgment stays with the expert who understands the geology.