SPEAKER: Justin Granek (MSc)
TITLE: Computing Geologically Consistent Models from Geophysical Data
DATE: Wednesday, November 28th 2012
TIME: 4:30pm
LOCATION: Room 451, 409 Granville (UK Building at Granville and Hastings)
ABSTRACT: The difficulty of finding economically viable mineral deposits has motivated the development of new exploration methodologies. This has led to greater efforts from the geophysical community to incorporate available sources of geological and geophysical information. Since the suite of available data types is diverse, the synthesis of multiple sources of information into a single coherent model can present many difficulties. In particular, the incorporation of geological constraints in the inversion of geophysical data has been investigated by various researchers.
While valuable information can be gleaned from geological data, a challenge remains due to the disconnect between geological units and geophysical property values. Though descriptive, a distinct geological unit is not always able to uniquely characterize the physical properties of a volume of earth, and vice versa. Interpretation and translation to and from geological and geophysical units can introduce bias based on the expert’s experience.
Current methodologies which exist to incorporate geological and geophysical information into inversion typically suffer from at least one of the following issues: either they require the user to interpret physical property values from geological information, or else they require the user to define some range of influence for each measurement in the model. Imposing constraints on a model which are biased in one of these ways can lead to recovered models unsupported by the data.
The introduced methodology differs from previous attempts at incorporation of a priori information since it applies statistical classification of in situ physical property measurements (as opposed to physical property values inferred from geology) as the basis for constraints. Statistical classification, combined with the iterative nature of the scheme, act to propagate the information from the downhole physical property logs through-out the model with minimum user input required. This automated approach reduces the potential for bias from unsupported constraints, while maximizing the integration of the classification results.