Uncertainties in the 2004 Sumatra–Andaman source through nonlinear stochastic inversion of tsunami waves

Abstract

Numerical inversions for earthquake source parameters from tsunami wave data usually incorporate subjective elements to stabilize the search. In addition, noisy and possibly insufficient data result in instability and non-uniqueness in most deterministic inversions, which are barely acknowledged. Here, we employ the satellite altimetry data for the 2004 Sumatra–Andaman tsunami event to invert the source parameters. We also include kinematic parameters that improve the description of tsunami generation and propagation, especially near the source. Using a finite fault model that represents the extent of rupture and the geometry of the trench, we perform a new type of nonlinear joint inversion of the slips, rupture velocities and rise times with minimal a priori constraints. Despite persistently good waveform fits, large uncertainties in the joint parameter distribution constitute a remarkable feature of the inversion. These uncertainties suggest that objective inversion strategies should incorporate more sophisticated physical models of seabed deformation in order to significantly improve the performance of early warning systems.

Response to reviewers' comments

“…As VOLNA-OP2 delivered a qualitative leap in terms of possible uses due to the high performance it can deliver on a variety of hardware architectures, its users have started integrating it into a wide variety of workflows; one of the key uses is for uncertainty quantification; for the stochastic inversion problem of the 2004 Sumatra tsunami in Gopinathan et al (2017), for developing Gaussian process emulators which help reduce the number of simulation runs in Beck and Guillas (2016); Liu and Guillas (2017), applications of stochastic emulators to a submarine slide at the Rockall Bank in Salmanidou et al (2017), a study of run-up behind islands in Stefanakis et al (2014), the durability of oscillating wave surge converters when hit by tsunamis in O'Brien et al (2015), tsunamis in the St. Lawrence estuary in Poncet et al (2010), a study of the generation and inundation phases of tsunamis in Dias et al (2014), and others.…”

“…The active mode is particularly important for tsunamigenic earthquakes with long and slow ruptures, e.g. the 2004 Sumatra-Andaman event described in Lay et al (2005); Gopinathan et al (2017) and submerged landslides in Løvholt et al (2015)), e.g. the Rockall Bank event in Salmanidou et al (2017).…”

“…The vertical component u z (x, t) of the seabed deformation is calculated depending on the physics of tsunami generation, e.g. via co-seismic displacement for finite fault segmentations by Gopinathan et al (2017), submarine sliding by Salmanidou et al (2017Salmanidou et al ( , 2018 etc..…”

The VOLNA-OP2 Tsunami Code (Version 1.0)

“…As VOLNA-OP2 delivered a qualitative leap in terms of possible uses, its users have started integrating it into a wide variety of workflows; one of the key uses is for uncertainty quantification; for the stochastic inversion problem of the 2004 Sumatra 20 tsunami in Gopinathan et al (2017), for developing Gaussian process emulators which help reduce the number of simulation runs (Beck and Guillas (2016); Liu and Guillas (2017)), applications of stochastic emulators to a submarine slide at the Rockall Bank (Salmanidou et al (2017)), a study of run-up behind islands ), the durability of oscillating wave surge converters when hit by tsunamis (O'Brien et al (2015)), tsunamis in the St. Lawrence estuary (Poncet et al (2010)), a study of the generation and inundation phases of tsunamis ), and others. These applications present a 25 number of challenges in integration into the workflow, as well as scalable performance: the need for extracting snapshots of state variables on the full mesh, or at a number of specified locations, capturing the maximum wave elevation or inundationall in the context of distributed memory execution.…”

“…The vertical component u z (x, t) of the seabed deformation is calculated depending on the physics of tsunami generation, e.g. via co-seismic displacement for finite fault segmentations by Gopinathan et al (2017), submarine sliding by Salmanidou et al (2017Salmanidou et al ( , 2018 etc. The time-dependency in u z (x, t) enables the tsunami to be actively generated (Dutykh and Dias (2009)).…”