GEO-EPISTEMOLOGY OF SUBSURFACE FLOW SIMULATION: POROUS VERSUS DISCRETE FRACTURE FLOW MODELS

The legacy of Bob Dott extends well beyond the limits of clastic sedimentology. Bob had a strong sense of both the history and the philosophy of geology that he employed extensively in his teaching and research approaches. Specifically, explorations of geo-epistemology – the nature and justification of knowledge– are fundamental to any scientific or engineering activity, and Bob’s lessons provided lasting value to his students whether or not they continued their careers in clastic sedimentology.

Geo-epistemology has a particular importance with the development of computer simulations of natural systems. Increasingly, our society has come to rely on computer simulation. Despite the widespread use of numerical models, few presentations of results consider the irresolvable issues of validation raised by Leonard Konikow and John Bredehoeft (Advances in Water Resources 15:75-83) and Naomi Oreskes (Science, 263:641-646). Models that derive their validity from calibration to past behaviors have inherent problems predicting future performance due to their non-uniqueness. In the epistemological concepts of the philosopher of science, Karl Popper, models derived by calibration or "history matching" fail to meet criteria of "falsifiability".

This presentation will look at the use of discrete fracture network (DFN) models for groundwater and petroleum applications not as simulations of complex systems but as numerical experiments to isolate the effects of parameters and conceptual models and gain insight into system behaviors. Unlike well-established porous media models, DFN models realistically represent the flowing system as networks of planar conduits using data derived from structural mapping and hydraulic testing. Used with porous medium codes, DFN approaches provide “multiple working hypotheses” that help to determine the consequences of uncertainty. Further, in the absence of data, numerical experiments can determine the value of data to guide investments of characterization resources. As an illustration of these concepts, a set of porous and DFN simulations of a well test provide equally good matches to the data, but have significantly different implications of reservoir behavior.