Sequential Data Assimilation for Seismicity: a Proof of Concept

Abstract:

Our physical understanding and probabilistic forecasting ability of earthquakes is significantly hampered by limited indications of the state of stress and strength on faults and their governing parameters. Using the sequential data assimilation framework developed in meteorology and oceanography (e.g., Evensen, JGR, 1994) and a seismic cycle forward model based on Navier-Stokes Partial Differential Equations (van Dinther et al., JGR, 2013), we show that such information with its uncertainties is within reach, at least for laboratory setups. We aim to provide the first, thorough proof of concept for seismicity related PDE applications via a perfect model test of seismic cycles in a simplified wedge-like subduction setup. By evaluating the performance with respect to known numerical input and output, we aim to answer wether there is any probabilistic forecast value for this laboratory-like setup, which and how many parameters can be constrained, and how much data in both space and time would be needed to do so. Thus far our implementation of an Ensemble Kalman Filter demonstrated that probabilistic estimates of both the state of stress and strength on a megathrust fault can be obtained and utilized even when assimilating surface velocity data at a single point in time and space. An ensemble-based error covariance matrix containing velocities, stresses and pressure links surface velocity observations to fault stresses and strengths well enough to update fault coupling accordingly. Depending on what synthetic data show, coseismic events can then be triggered or inhibited.