Application of multiple reflections inseismic imaging

Shang Huang

Multiple reflections in seismic imaging provide additional information and insights into subsurface structures. However, their practical utilization could be improved due to their complex interaction with the subsurface and interference with primary reflections and noise.This thesis aims to use multiples in seismic imaging in two contexts: classical and machine learning imaging.

For the first context, I attempt to extend aperture illumination in a phase shift plus interpolation (PSPI) migration by adding scattering terms in the phase-shift wavefield propagation operator. This method iteratively adds scattering terms for each reflector as the source and receiver wavefields propagate downward into the subsurface, which helps in efficiently extending the illumination of horizontal reflector edges. In the second context,I consider improving images from reverse time migration (RTM). RTM with multiple reflections(RTMM) can improve illumination but suffers from interference between different orders of multiples. To overcome this limitation, I proposed a method based on a convolutional neural network (CNN) and U-Nets that approximates the inverse of the Hessian similarly to least squares migration, but with less computational cost. The U-NET is trained to learn patterns representing the relation between the reflectivity obtained through RTMM and the true reflectivity. I further developed this by adding a discrete wavelet transform(DWT) input channel which provides an additional constraint that helps to enhance image resolution.

Finally, as a key application of the above techniques, I consider the problem of timelapse seismic monitoring which attempts to detect very weak signal differences produced by changes in reservoirs. This technique is affected by changes in the near-surface noise and insufficient illumination to detect the weak changes. I proposed a novel method that leverages stacked long short-term memory (SD-LSTM) and U-Net neural networks to predict and mitigate noise in monitor data. I test the method in a field dataset, DAS VSP data from the CaMI FRS project. The output provides meaningful information and prediction for CO₂ injection migration within a target area. This result aligns closely with the CaMI FRS project CO₂ injection plan, providing valuable insights for monitoring the CO₂ migration paths for the Basal Belly River Sandstone Formation.