Cover thickness uncertainty mapping using Bayesian estimate fusion: leveraging domain knowledge

Abstract

SUMMARY Thickness of cover over crystalline basement is an important consideration for mineral exploration in covered regions. It can be estimated from a variety of geophysical data types using a variety of inference methods. A robust method for combining such estimates to map the cover–basement interface over a region of interest is needed. Due to the large uncertainties involved, these need to be probabilistic maps. Predominantly, interpolation methods are used for this purpose, but these are built on simplifying assumptions about the inputs which are often inappropriate. The Bayesian estimate fusion is an alternative capable of addressing that issue by enabling more extensive use of domain knowledge about all inputs. This study is intended as a first step towards making the Bayesian estimate fusion a practical tool for cover thickness uncertainty mapping. The main contribution is to identify the types of data assumptions that are important for this problem, to demonstrate their importance using synthetic tests and to design a method that enables their use without introducing excessive tedium. We argue that interpolation methods like kriging often cannot achieve this goal and demonstrate that Markov chain Monte Carlo sampling can. This paper focuses on the development of statistical methodology and presents synthetic data tests designed to reflect realistic exploration scenarios on an abstract level. Intended application is for the early stages of exploration where some geophysical data are available while drill hole coverage is poor.