← Back to Sessions

A Bayesian Approach for Geodetic Data Inversion and Model Selection of the Mw 8.8 2010 Maule Earthquake Coseismic Slip Distribution

Session: Subduction Processes Along Latin America Subduction Zones [Poster]

Type: Poster

Date: 4/19/2021

Presentation Time: 11:30 AM Pacific

Description: 

We employ Bayesian Inference tools in order to compare two kinematic models for resolving the coseismic slip of subduction zone earthquakes: (i) A finite fault in the megathrust between the subducting and overriding plates and (ii) Two finite faults; an upper plane simulating inverse slip in the megathrust zone and a lower plane with normal slip, located roughly in the elastic limit within the slab. The proposed inverse problem for determining model parameters for each case is nonlinear but can be linearized. We build an algorithm that determines the optimal solution and its uncertainties such that it maximizes the Bayesian Evidence. By using this algorithm in a GPS+InSAR inversion for the Great Mw 8.8 2010 Maule Earthquake, we found that the two-interface model, by depending on a greater number of parameters, fits more precisely the observations than the one interface model, however, this improvement is not significant enough, implying that the selected model by the Bayesian methodology is the simpler, one interface model. The selected slip distribution (Mw 8.76) shows two main patches, north and south of the epicenter, in agreement with past results. Furthermore, our solution indicates high slip uncertainties near the trench, which could be linked to the limitations of our modelling, but also could be interpreted as a more realistic feature of coseismic modelling when no offshore data is used. We conclude that Bayes theory allows to tackle in a good extent the model selection problem, although its viability is conditional to the computation of the Evidence, which becomes complex as the model parameterization also does.

Presenting Author: Sebastian Nunez Jara

Student Presenter: Yes

Authors