Bayesian Integration of Hydrogeological and Geophysical Data for Site Characterization: Theory and Application to the LLNL Superfund Site

A stochastic Bayesian approach for combining well logs and geophysical surveys for enhancing subsurface characterization is presented. The main challenge we face is in creating the bridge to link between ambiguously related geophysical surveys and well data. The second challenge is imposed by the disparity between the scale of the geophysical survey and the scale of the well logs. Our approach intends to integrate and transform the well log data to a form where it can be updated by the geophysical survey and this tends to be a convoluted process. Our approach starts with generating images of the lithology, conditional to well logs. Each lithology image is then used as the basis for generating a series of shaliness images, conditional to well logs data. Shaliness images are converted to resistivity images using a site-specific petrophysical model relating between shaliness, resistivity and lithology, to create the necessary interface with the crosswell resistivity survey. The lithology and resistivity images are then updated using crosswell electromagnetic resistivity surveys. We explored the limits of the approach through synthetic surveys of different resolutions and error levels, employing the relationships between the geophysical and hydrological attributes which are weak or non-linear or both. The synthetic surveys closely mimic the conditions at the LLNL Superfund site. We show that the proposed stochastic Bayesian approach improves hydrogeological site characterization even when using low-resolution resistivity surveys.