DATA FUSION FOR GEOTHERMAL RESERVOIR CHARACTERIZATION

Abstract

This study performs Bayesian inversion of geophysical (gravity) data to generate probabilistic 3D subsurface models. Probabilistic models describe the uncertainty of model predictions, which is beneficial for managing risk during decision making such as well position targeting. We select Krafla geothermal system in Iceland as a case study for the application of three-part framework. The method combines a prior geologic model of the area based on primary geologic data obtained from wells, statistical analysis of petrophysical properties (density and porosity) based on available databases and additional measurements, as well as available gravimetric measurements data. A Markov chain Monte Carlo sampling scheme, implemented in the geological modelling software GeoModeller, is used to invert for subsurface lithology and density. Due to the non-uniqueness of gravity data, many possible models of subsurface density distribution could account for the measured data. The a priori uncertainty of the lithological model largely controls the uncertainty of the posterior results for lithology and density. This shows that reliable prior geological constraints are important in Bayesian inversion. In this case, the lithological model consists of layers of high-density lava flows and low-density hyaloclastites underlain by high-density basement intrusions. Rock density depends also significantly on the extent and type of hydrothermal alteration particularly for hyaloclastites where density tends to be higher with increased extent of alteration. More realistic prior density probability density functions (pdf’s) especially accounting for hydrothermal alteration for hyaloclastites result in more realistic model output.