Global full-waveform inversion using a surface wave orbital

Full-waveform inversion (FWI) has matured over the last decade to produce high-resolution global wavespeed models down to minimum periods of 30 s. While current models agree on large scales, they exhibit structural differences at shorter wavelengths depending on the assimilated data and inversion strategy. REVEAL is a global-scale FWI model utilizing a multi-scale, whole-waveform inversion strategy. To mitigate cycle-skipping, data is incrementally bandpass-filtered and inverted from long to short periods. Currently, REVEAL inverts all phase windows in 1-hour seismograms; this conservatively preserves known body-wave structure while incorporating new surface wave structure.

However, the use of 1-hour seismograms limits the model's sensitivity to the Southern Hemisphere, where receiver density is significantly lower than in the Northern Hemisphere. To address this, we present preliminary work on a next-generation model that incorporates 3.5-hour seismograms. These longer time series capture both minor- and major-arc surface wave arrivals. Including major-arc waves (epicentral distances > 180°) significantly increases sensitivity to structure in the Southern Hemisphere. Starting from the current REVEAL model, we revise the long-period structure (60–200 s) using a multi-scale approach, beginning with 100–200 s and terminating at 60–200 s. We simulate full viscoelastic wavefields using the spectral-element solver Salvus on global fourth-order cubed-sphere meshes. Adjoint sources are calculated using a time-frequency phase misfit, and optimization is performed via mini-batch stochastic trust-region L-BFGS.