Falsifying Hypotheses for the Origin of a Fractured Bedrock Zone With Geophysical Data

Abstract:

Subsurface heterogeneity – particularly bedrock heterogeneity – is known to be one of the most significant challenges for hydrology and watershed science. Inferences from geophysical data are often used to characterize subsurface heterogeneity, but their data and interpretations have significant uncertainty. In this study we develop a Bayesian framework to capture this uncertainty and test hypotheses regarding the sources of bedrock heterogeneity using geophysical measurements. We demonstrate our approach using electrical resistivity tomography (ERT) data to study a highly-fractured shale bedrock zone within the East River Watershed near Crested Butte, CO. Cores and borehole geophysics clearly show the existence of a hydraulically conductive fracture zone in the area, but there are no mapped geologic features nearby that could cause extensive fracturing. Using the ERT survey we constrain the morphology of this fractured zone to test the hypotheses that the fractures are controlled by: (1) bedding planes, (2) a normal fault, (3) a reverse fault, or (4) post-glacial slumping. The prior model describes the morphology and geophysical properties of the fractured zone with five parameters. Forward geophysical modeling is used to generate synthetic ERT data for many possible fracture zone configurations. Random forest models are trained on the prior realizations to predict the five fracture zone parameters directly from ERT data without inversion. The trained random forests are then applied to the measured ERT data to estimate posterior uncertainties for each of the five prior model parameters. Analysis of the posterior uncertainty falsifies the hypotheses that the fracture zone is controlled by bedding planes or a sub-horizontal reverse fault. The result is that the fracture zone in question is most likely an unmapped, active normal fault that runs parallel to the East River. This interpretation has significant impacts on groundwater modeling strategies throughout the watershed.