Improving modelling of seismic images through Point-Spread Function-based convolution
PhD-Candidate Kristian Jensen (associated with Geodynamics and Basin Studies research group)
Main content
Main Supervisor
Associate Professor Isabelle Lecomte, University of Bergen
Co-Supervisors
Professor Børge Arntsen, NTNU
Associate Professor Einar Iversen, University of Bergen
Professor Leiv-J. Gelius, University of Oslo
Principal Research Geophysicist Tina Kaschwich, NORSAR
About the project
Seismic images are generated when a source at the Earth’s surface initiates a pressure wave which travels through the Earth’s subsurface. As the wave hits layer boundaries in the Earth’s interior, energy is reflected back to the Earth’s surface where it is recorded, thus allowing us to obtain an image of the subsurface. The principle behind this approach is exactly the same as in medical ultrasound.
Due to the complexity of the Earth, and the intricate wave physics involved, the raw seismic images usually contain many distortions and artefacts. As such, extensive processing is often needed to obtain proper images of the subsurface. Yet, even after performing a processing workflow, features in the images may still be ambiguous and difficult to identify.
In order to study in greater detail exactly what seismic images will look like for a certain geological setting and survey geometry, seismic modelling may be performed. This involves using a computer to generate a model of the subsurface, and then simulating what the seismic images will look like for a given survey geometry over the model. Proper seismic modelling can be very helpful in identifying exactly why ambiguities occur, and what may be done to improve the quality of the images.
Illustration of how changes in seismic modelling parameters yield different seismic images of the same target area.
Seismic modelling approaches are, however, often computationally expensive. The aim of my PhD is therefore to further develop a method which may perform accurate seismic modelling at a low computational cost. The method estimates the local illumination and resolution at specified points in the model, referred to as point-spread functions (PSFs). The PSFs are computed by tracing the path of the wave through the Earth (a procedure referred to as ray tracing), and not through a full simulation of the wave equation. This allows for a very quick estimation of the PSFs. The PSFs may then be mathematically convolved with the Earth model in order to simulate a processed seismic image. Through this method we can quickly and efficiently study how survey geometry, seismic wave velocity, frequency bandwidth and amplitude effects influence the quality of the seismic image surrounding the point of interest. However, as the method does not solve the full wave equation, some simplifications and limitations exist. My PhD-project therefore aims to improve the quality of the modelled seismic images by attempting to overcome some of the inherent limitations while retaining the low computational cost.
If successfully done, the method may become a valuable tool for any scientist working with seismic data. In addition to function as modelling operators, accurately estimated PSFs have a range of other applications as well, in particular related to improving resolution of seismic images. Furthermore, the theoretical discoveries may also be transferable to other related imaging disciplines, such as medical ultrasound, optical sciences, and radar/telescope technology as used for instance in Mars expeditions and astronomy in general.