High-frequency wavefield recovery with weighted matrix factorizations

Acquired seismic data is normally not the fully sampled data we would like to work with since traces are missing due to physical constraints or budget limitations. Rank minimization is an effective way to recovering the missing trace data. Unfortunately, this technique's performance may deteriorate at higher frequency because high-frequency data can not necessarily be captured accurately by low-rank matrix factorizations albeit remedies exist such as hierarchical semi-separable matrices. As a result, recovered data often suffers from low signal to noise ratio S/Rs at the higher frequencies. To deal with this situation, we propose a weighted recovery method that improves the performance at the high frequencies by recursively using information from matrix factorizations at neighboring lower frequencies. Essentially, we include prior information from previously reconstructed frequency slices during the wavefield reconstruction. We apply our method to data collected from the Gulf of Suez, which shows that our method performs well compared to the traditional method without weighting.

https://slim.gatech.edu/Publications/Public/Conferences/SEG/2019/zhang20...