3:00 PM - 4:00 PM
[S13-P-04] An evolutive quasi-real-time source inversion based on a linear inverse formulation
Finite source inversion is a steppingstone to unveil earthquake rupture. Significant progress has been made on approaches regarding earthquake imaging, thanks to new data acquisition and methodological advances. However, most of these techniques are posterior procedures once seismograms are available. Incorporating source parameters estimation into early warning systems would require to update the source build-up while recording data.
We developed a kinematic source inversion formulated in the time-domain, for which seismograms are linearly related to the slip distribution on the fault through convolutions with Green functions previously estimated and stored. These convolutions are performed in the time-domain as we progressively increase the time window of records at each station specifically. Selected unknowns are the spatio-temporal slip-rate distribution to keep the linearity of the forward problem with respect to unknowns. Through the spatial extension of the expected rupture zone, we progressively build-up the slip-rate when adding new data by assuming rupture causality.
This formulation is based on the adjoint-state method for efficiency. The inverse problem is non-unique and, in most cases, underdetermined. While standard regularization terms are used for stabilizing the inversion, we avoid strategies based on parameter reduction leading to an unwanted non-linear relationship between parameters and seismograms for our progressive build-up. Rise time, rupture velocity and other quantities can be extracted later on as attributes from the slip-rate inversion we perform.
Satisfactory results are obtained on a synthetic example proposed by the Source Inversion Validation project (Mai et al. 2011). The application of this method to the Kumamoto 2016 earthquake is currently being explored. Our specific formulation combined with simple prior information, as well as numerical results obtained so far, yields interesting perspectives for a quasi-real-time implementation.
We developed a kinematic source inversion formulated in the time-domain, for which seismograms are linearly related to the slip distribution on the fault through convolutions with Green functions previously estimated and stored. These convolutions are performed in the time-domain as we progressively increase the time window of records at each station specifically. Selected unknowns are the spatio-temporal slip-rate distribution to keep the linearity of the forward problem with respect to unknowns. Through the spatial extension of the expected rupture zone, we progressively build-up the slip-rate when adding new data by assuming rupture causality.
This formulation is based on the adjoint-state method for efficiency. The inverse problem is non-unique and, in most cases, underdetermined. While standard regularization terms are used for stabilizing the inversion, we avoid strategies based on parameter reduction leading to an unwanted non-linear relationship between parameters and seismograms for our progressive build-up. Rise time, rupture velocity and other quantities can be extracted later on as attributes from the slip-rate inversion we perform.
Satisfactory results are obtained on a synthetic example proposed by the Source Inversion Validation project (Mai et al. 2011). The application of this method to the Kumamoto 2016 earthquake is currently being explored. Our specific formulation combined with simple prior information, as well as numerical results obtained so far, yields interesting perspectives for a quasi-real-time implementation.