AxiSEM and instaseis: Fast simulation of global wavefields across the frequency band

Abstract:

We present our seismic modeling methods AxiSEM and instaseis. These methods exploit recent developments in high-performance computing and suitable numerical methods for seismic wave propagation, while operating efficiently across the vast observable frequency spectrum of global waves in sparse yet realistic structures. AxiSEM (www.axisem.info and geodynamics.org) relies upon axisymmetric (including spherically symmetric) models, thereby satisfying a large fraction of observable data. The benefit of this method lies in the resultant dimensional collapse to two numerical dimensions, whereby the third azimuthal dimension is tackled analytically. For high-frequency wave propagation, this leads to 3-4 orders of magnitude speedup in computational cost compared to 3D domain discretizations. AxiSEM is highly scalable and accommodates efficient implementations of viscoelasticity and anisotropy. We will present benchmarks, data comparisons, a diverse range of applications from inner-core anisotropy to noise modeling and lowermost mantle structures, and wavefields for sensitivity kernels. We also touch upon ongoing efforts for linking computational cost to structural complexity in the vein of Occam's razor, eventually allowing for an adaptive rendition of 1D, 2D and 3D structures at optimally low computational cost, as well as 1D/3D hybrid approaches.

Instaseis (www.instaseis.net) is a methodology to extract full, broadband and accurate waveforms instantaneously from wavefield databases computed with AxiSEM. This "once-and-for-all solution" relies on reciprocity and requires only two AxiSEM simulations to construct the databases, while allowing for arbitrary parameter changes (e.g. source, processing, structure) instantaneously with modest computational cost and storage requirements. The instaseis python package is integrated with ObsPy, contains a graphical user interface, and can be used for source inversion, noise simulations, finite-fault modeling, waveform tomography, uncertainty estimations, as well as teaching.

Both methods are published open-access and freely available via the above website links.