Estimating Subsurface Permeability with 3D Seismic Attributes: A Neural Net Approach

Abstract

We seek to map subsurface permeability in a geothermal area using 3D seismic attributes extracted from a 14 square mile 3D seismic reflection survey acquired over the Walker Ranch geothermal prospect at Raft River, Idaho. A set of attributes was estimated and a neural net approach used to infer relationships between the attributes and permeability index, as well as other information including resistivity and lithology. The analysis was conducted two ways: using commercial software and an independently developed code. Several types of neural net algorithms were tested. A training set was constructed using an estimated permeability index derived from well data. Several ways of parameterizing the observed permeability distribution inferred from the well data were tested. Validation was conducted using a ‘leave one out’ strategy. The neural net was successful in determining relationships between the training set and the attributes at a high correlation level but results from the validation were inconsistent due largely to problems with false positives. The 3D seismic data was also used to examine the locations of known microseismic events with respect to features in the depth migrated 3D reflection data. Only a general correspondence between the microseismic hypocenters and observed faults in the 3D seismic volume was observed, with no direct relationship with a clearly imaged fault.