| More Details: |
We present a goal-oriented framework for constructing digital twins with the following properties: (1) they employ discretizations of high-fidelity PDE models governed by autonomous dynamical systems, leading to large-scale forward problems; (2) they solve a linear inverse problem to assimilate observational data to infer uncertain model components followed by a forward prediction of the evolving dynamics; and (3) the entire end-to-end, data-to-inference-to-prediction computation is carried out in real time through a Bayesian framework that rigorously accounts for uncertainties. Realizations of such a framework are faced with several challenges that stem from the extreme scale of forward models and, in some cases, slow eigenvalue decay of the parameter-to-observable map. In this paper, we introduce a methodology to overcome these challenges by exploiting the autonomous structure of the forward model. As a result, we can move the PDE solutions, which dominate the cost for solving the Bayesian inverse problem, to an offline computation and leverage the high-performance dense linear algebra capabilities of GPUs to accelerate the online prediction of quantities of interest. We seek to apply this framework to construct digital twins for the Cascadia subduction zone as a means of providing early warning for tsunamis generated by subduction zone megathrust earthquakes. To that end, we demonstrate how our methodology can be used to employ seafloor pressure observations, along with the coupled acoustic-gravity wave equations, to infer the earthquake-induced seafloor motion (discretized with $O(10^9)$ parameters) and forward predict the tsunami propagation. We present results of an end-to-end inference, prediction, and uncertainty quantification for a representative test problem for which this goal-oriented Bayesian inference is accomplished in real time, that is, in a matter of seconds. |
