Seismic interpretation of the subsurface geologic record has evolved over the years from simple 2D wavelet visualizations on paper to dynamic 3D visualizations in computer programs. Proven 2D manual approaches gave the interpreter ample time to think about and validate geologic relationships as they built their seismic interpretation. But now we can automatically extract and view thousands of seismic reflections as movies in 3D and these accelerated mapping capabilities leave us little time to think about what we are seeing. High-definition monitors and dynamic visualization techniques encourage us to believe seismic images are “photographs” of the geologic record and make it easy to assume that seismic reflections are geologic surfaces. We must not forget that the recorded geophysical response relates to a complex interplay between rock properties, depth and frequency and that these relationships impact interpretive uncertainty. In order to frame this uncertainty, we must understand how frequency loss impacts the accuracy of the seismic to geologic tie. And, in a future where automated seismic mapping is the norm, we must ensure interpretations are validated within the constraints of the data quality.

At what frequency do seismic signals fail to accurately record geologic response? It is well established that seismic signals respond to impedance contrasts associated with variations in rock properties and fluid type in the subsurface, and it is also generally accepted that maximum seismic resolution is approximately ¼ of the seismic wavelet. It is well known that seismic frequency decreases with increasing depth so it should be obvious that deeper seismic reflections can represent a composite of responses from multiple geologic layers. Yet, our accelerated interpretive methods make it easy to overlook this simple reality. In a low-price environment we aim to minimize this geophysical and geologic uncertainty to improve project decisions and insure profitable outcomes.

To address this challenge, Chevron developed CompStrat, a high-resolution 3D computational forward stratigraphic modeling method that allows us to generate reservoir facies models constrained by the physics of clastic depositional process. Additionally, we have developed a convolutional model-to-seismic workflow that can transform these stratigraphic models into 3D synthetic seismic volumes at frequency ranges varying from 10 – 1500 Hz. We begin by generating a CompStrat model (a field-scale, high-resolution depositional model representing the environment of deposition of the targeted reservoir) then we convert it into 3D synthetic seismic volumes at a range of frequencies using a variety of rock-physics relationships. We compare the various seismic results which allows us to measure the scale-dependent relationship between seismic frequency and geologic surfaces and to visualize how frequency loss degrades the accuracy of our interpretation.

When conditioned CompStrat models (using well data, stratigraphic interpretations or conceptual ideas) are converted to synthetic seismic, interpreters gain an interpretive framework in which they may test hypotheses. These integrated models allow interpreters to vary rock property and fluid assumptions, evaluate the observed seismic response, compare the modeled response to field data and to use these results to predict potential reservoir facies complexity. These comparisons allow interpreters to frame seismic uncertainty within the context of geologic possibility and to refine resource estimations in low-resolution scenarios. The ability to convert bed-scale computational models to 3D synthetic seismic models at high frequencies is allowing us to visualize the interpretive limits of seismic data and is substantiating the cautionary statements of our seismic interpretation pioneers.