Improving Fault Behaviour Predictions using Empirical Rules for QC and Modelling of Fault Populations

Faults are crucial in subsurface processes, affecting hydrocarbon accumulation, well planning, containment risks, and drilling safety. Accurate fault interpretation is essential, as errors can lead to unreliable subsurface predictions. While deep learning has improved fault interpretation, its effectiveness relies on geologically realistic training data and fault surface extractions.

We analyzed 4,754 faults from 44 3D seismic volumes across diverse geologic settings. Statistical compilations of fault dimensions, ambiguity, and density—relative to seismic volume quality—enable three key applications:

1. Quality control of manual interpretations.
2. Creation of realistic datasets for deep learning.
3. Guidance for extracting discrete fault surfaces from inference volumes.

These volume-based tools, independent of horizon interpretations, enhance fault analysis accuracy, supporting safer and more efficient subsurface operations.