Probabilistic, high-resolution tsunami predictions in northern Cascadia by exploiting sequential design for efficient emulation

Abstract

Abstract. The potential of a full-margin rupture along the Cascadia subduction zone poses a significant threat over a populous region of North America. Previous probabilistic tsunami hazard assessment studies produced hazard curves based on simulated predictions of tsunami waves, either at low resolution or at high resolution for a local area or under limited ranges of scenarios or at a high computational cost to generate hundreds of scenarios at high resolution. We use the graphics processing unit (GPU)-accelerated tsunami simulator VOLNA-OP2 with a detailed representation of topographic and bathymetric features. We replace the simulator by a Gaussian process emulator at each output location to overcome the large computational burden. The emulators are statistical approximations of the simulator's behaviour. We train the emulators on a set of input–output pairs and use them to generate approximate output values over a six-dimensional scenario parameter space, e.g. uplift/subsidence ratio and maximum uplift, that represent the seabed deformation. We implement an advanced sequential design algorithm for the optimal selection of only 60 simulations. The low cost of emulation provides for additional flexibility in the shape of the deformation, which we illustrate here considering two families – buried rupture and splay-faulting – of 2000 potential scenarios. This approach allows for the first emulation-accelerated computation of probabilistic tsunami hazard in the region of the city of Victoria, British Columbia.

Robust uncertainty quantification of the volume of tsunami ionospheric holes for the 2011 Tohoku-Oki earthquake: towards low-cost satellite-based tsunami warning systems

“…Although, in some cases, tsunamis reach the coast very fast, to apply our method there must be a minimum window of almost 10 min between generation and arrival. However, this is perfectly valid for tsunami hazard assessment over populous regions with larger arrival times, as for example tsunami hazard assessment in the city of Victoria, British Columbia, from a tsunami generated in the Cascadia subduction zone (Salmanidou et al, 2021). Our implementation on the 2011 Tohoku-Oki earthquake in Japan demonstrates that our method works well there.…”

Rapid tsunami force prediction by mode-decomposition-based surrogate modeling

“…Therefore, a surrogate-modeling-based prediction method is employed in this study. Surrogate modeling has been widely accepted for uncertainty quantification and probabilistic risk assessment, such as studies using response surface (e.g., Fukutani et al, 2019;Kotani et al, 2020), Gaussian process (e.g., Sarri et al, 2012;Salmanidou et al, 2017Salmanidou et al, , 2021, polynomial chaos expansion (e.g., Denamiel et al, 2019;Giraldi et al, 2017;Sraj et al, 2017) and multifidelity sparse grids (e.g., de Baar and Roberts, 2017). These works demonstrate the potential of the surrogate-modeling-based approach, while the surrogate model considering spatiotemporal variation has not been well studied.…”

Emulation of high-resolution land surface models using sparse Gaussian processes with application to JULES

“…Gu and Berger (2016) assumed that the same correlation function and parameters are shared across all outputs, enabling huge numbers of emulators to be built in parallel. The "emulate every grid cell" approach has been used in environmental applications including aerosol models (Lee et al, 2012;Johnson et al, 2020), tsunami models (Salmanidou et al, 2021) and volcano models (Spiller et al, 2020). These methods do not directly model spatial and temporal dependence,but instead assume that the training data for each grid cell contain enough information alone, and rely on the fast training and prediction times of standard emulators (though for JULES, the millions of emulators required to follow this approach renders it less feasible).…”