Mineral potential mapping using Cox-type regression for marked point processes

Mineral potential mapping infers the chance of finding a deposit at a location from geological covariables. Usual methods, like weights of evidence, fit a model for the occurrence probability of a deposit in a pixel to a dataset of pixels, where the all geological covariables and presence or absence of the deposit is known.

We propose a new more flexible approach, viewing the locations of deposits as an inhomogeneous marked point process. This allows to model intensity rather than probability. The local occurrence probability is computed from the fitted model. The variation of log intensity is described by a linear model in the covariables like in a Cox-regression model for survival analysis. Applied to similar data the approach gives results similar to the classical approaches.

However it additionally allows for the inclusion of continues covariables, the use of information on smaller uneconomical mineralisation zones as additional information for the model fit by introducing mark distributions, inclusion of incompletely explored areas by reweighting with the probability of finding, and the use of samples rather than complete maps for the region without deposit
using the spatial character of point processes.