Effect of Merging Multiscale Models on Seismic Wavefield Predictions Near the Southern San Andreas Fault

Updating Earth models used by the scientific community in geologic studies and hazard assessment has a significant societal impact but is computationally prohibitive due to the large spatial scale. The advent of urban seismology allowed rapid development of local high-resolution models using short-term dense seismic arrays to become conventional. To incorporate the details in these local models in community models, we developed a technique for constructing window taper functions like the cosine taper in arbitrarily shaped spatial domains on regular grids. We apply our algorithm to the problem of low-frequency ground shaking estimation near the southernmost San Andreas fault by creating two hybrid models. These models consist of basin-scale (top 10 km or less) high-resolution models developed using controlled source data embedded into two popular Southern California Earthquake Center community models. We evaluate the models by computing long period (6-30 s) wavefield energy misfits using 11 earthquakes with moment magnitudes between 3.5 and 5.5 not used in developing any of the models under consideration. One of the hybrid models produces an similar to 24% decrease while the other has an similar to 0.6% increase in the overall median misfit relative to their original community models. The overlapping misfit values between the models and variability in waveform fit for different events and stations emphasize the difficulties in model validation. Our approach can merge any type of gridded multiscale and multidimensional datasets, and represents a valuable tool for modeling in the computational sciences.