Bayesian Maximum Entropy (BME) methods for Environment and Geosciences"

Tutorial lecture
Geolab, Institute of Geomatics and Analysis of Risk (IGAR)
University of Lausanne
Switzerland
Wednesday 21 March

Prof. Patrick Bogaert, (Université Catholique Louvain, Belgium)

It is a fact that classical geostatistics has been (and still remains) successfully applied to a wide array of problems involving spatial mapping of environmental data. It also emerged as a discipline by rights, i.e. a core of sound and well established results that constitutes the set of widely accepted methods inside a community. In spite of this apparent state of wellness, some inherent limitations have however been obliterated or addressed in a questionable way, e.g. processing of data that are of very different nature or accuracy.

In order to circumvent these limitations, the Bayesian Maximum Entropy (BME) methods have been recently proposed as a new avenue. It consists in a unified and consistent theoretical framework built around the concept of information, thus offering the guarantee of handling in a natural way the processing of spatio-temporal data that are inherently different with respect to their nature and accuracy.

Starting from the basic notions of spatial statistics, this lesson will emphasize the limitations of classical geostatistics. It will be shown how standard results can be extended to encompass a much wider array of applications using BME methods. The fundamental theoretical concepts will be presented, and a special attention will be paid to the concept of data diversity and integration. All concepts will be illustrated using different environmental case studies, in order to emphasize the clear benefits of these methods, especially for environmental risk assessment and risk management.