Enhancing regional seismic velocity model with higher-resolution local
results using sparse dictionary learning
We use sparse dictionary learning to develop transformations between
seismic velocity models of different resolution and spatial extent.
Starting with results in the common region of both models, the method
can be used to enhance a regional lower-resolution model to match the
style and resolution of local higher-resolution results while preserving
its regional coverage. The method is demonstrated by applying it to
two-dimensional Vs and three-dimensional VP and VS local
and regional velocity models in southern California. The enhanced
reconstructed models exhibit clear visual improvements, especially in
the reconstructed VP/VS ratios, and better correlations with
geological features. We demonstrate the improvements of the
reconstructed model relative to the original velocity model by comparing
waveform simulation results to observations. The improved fitting to
observed waveforms extends beyond the domain of the overlapping region.
The developed dictionary learning approach provides physically
interpretable results and offers a powerful tool for additional
applications for data enhancement in earth sciences.