Real-time Bayesian inference at extreme scale: A digital twin for tsunami early warning applied to the Cascadia subduction zone

2025-04-23

AI Summary

Early tsunami warning for the Cascadia subduction zone requires real-time Bayesian inversion of seafloor spatiotemporal motion from hydroacoustic sensor data and prediction of tsunami propagation via the 3D acoustic-gravity wave equation—all under strict latency constraints and with rigorous uncertainty quantification. Method: We propose the first billion-parameter-scale real-time Bayesian digital twin system, introducing a novel offline/online decomposition framework grounded in translation-invariant parameter-to-observation mappings, integrated with adjoint wave propagation, high-performance mixed finite element modeling (MFEM), and large-scale GPU parallelization on the El Capitan supercomputer (43,520 GPUs). Contribution/Results: The system achieves sub-0.2-second end-to-end inversion and forecasting—accelerating traditional methods by ten orders of magnitude—while attaining 92% weak scaling efficiency in offline training. It represents the first fully closed-loop, uncertainty-aware tsunami digital twin operating at billion-parameter scale for operational early warning.

Technology Category

• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Machine Learning: Bayesian Learning
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Data management and stream processing for Web, mobile and wireless applications
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines

Abstract

We present a Bayesian inversion-based digital twin that employs acoustic pressure data from seafloor sensors, along with 3D coupled acoustic-gravity wave equations, to infer earthquake-induced spatiotemporal seafloor motion in real time and forecast tsunami propagation toward coastlines for early warning with quantified uncertainties. Our target is the Cascadia subduction zone, with one billion parameters. Computing the posterior mean alone would require 50 years on a 512 GPU machine. Instead, exploiting the shift invariance of the parameter-to-observable map and devising novel parallel algorithms, we induce a fast offline-online decomposition. The offline component requires just one adjoint wave propagation per sensor; using MFEM, we scale this part of the computation to the full El Capitan system (43,520 GPUs) with 92% weak parallel efficiency. Moreover, given real-time data, the online component exactly solves the Bayesian inverse and forecasting problems in 0.2 seconds on a modest GPU system, a ten-billion-fold speedup.

Problem

• Real-time Bayesian inference for tsunami early warning
• Infer earthquake-induced seafloor motion using acoustic data
• Forecast tsunami propagation with quantified uncertainties quickly

Innovation

• Bayesian inversion-based digital twin for real-time analysis
• 3D acoustic-gravity wave equations for tsunami forecasting
• Novel parallel algorithms enabling billion-fold speedup