Reduced Order Modeling for Tsunami Forecasting with Bayesian Hierarchical Pooling

2025

Read Reduced Order Modeling for Tsunami Forecasting with Bayesian Hierarchical Pooling (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-19804.txt.

Introduction

Shane Coffing and three co-authors build a faster way to forecast tsunamis, with honest odds attached. Small models can stand in for the full wave physics. They run far faster than the full sums, but the classic kind follows one fixed path. Their model instead keeps the physics in an initial-value problem. Fixes to its operators tune it so its paths match real wave paths.

A Bayesian pooling step is layered on top of the start values. That yields new paths that work across a range of nearby starts. The result is a stand-in the team calls a randPROM. It gives full, odds-based forecasts of waves. They test it on a made-up Fiji case and on the real 2011 Tohoku tsunami. It cut the number of full runs by a lot, while its calls on arrival times and heights stayed sound.

Catalog record

YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE57 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-19804

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Reduced Order Modeling for Tsunami Forecasting with Bayesian Hierarchical Pooling Abstract Reduced order models (ROM) can represent spatiotemporal processes in significantly fewer dimensions and can be solved many orders faster than their governing partial differential equations (PDEs). For example, using a proper orthogonal decomposition produces a ROM that is a small linear combination of fixed features and weights, but that is constrained to the given process it models. In this work, we explore a new type of ROM that is not constrained to fixed weights, based on neural Galerkin-Projections, which is an initial value problem that encodes the physics of the governing PDEs, calibrated via neural networks to accurately model the trajectory of these weights. Then using a statistical hierarchical pooling technique to learn a distribution on the initial values of the temporal weights, we can create new, statistically interpretable and physically justified weights that are generalized to many similar problems. When recombined with the spatial features, we form a complete physics surrogate, called a randPROM, for generating simulations that are consistent in distribution to a neighborhood of initial conditions close to those used to construct the ROM. We apply the randPROM technique to the study of tsunamis, which are unpredictable, catastrophic, and highly-detailed non-linear problems, modeling both a synthetic case of tsunamis near Fiji and the real-world Tohoku 2011 disaster. We demonstrate that randPROMs may enable us to significantly reduce the number of simulations needed to generate a statistically calibrated and physically defensible prediction model for arrival time and height of tsunami waves. † † journal: Journal of Advances in Modeling Earth Systems (JAMES) † † authors: Shane X. Coffing , John Tipton , Arvind T. Mohan , Darren Engwirda † † corresponding: Shane X. Coffing, [email protected] Computing and Artificial Intelligence Division (CAI-2), Computational Physics and Methods, Los Alamos National Laboratory, Los Alamos, NM, USA Computing and Artificial Intelligence Division (CAI-4), Statistics, Los Alamos National Laboratory, Los Alamos, NM, USA Theoretical Division (T-3), Fluid Dynamics and Solid Mechanics, Los Alamos, Los Alamos, NM, USA keypoints Conventional reduced order modeling (ROM) techniques are not easily generalized to unseen dynamics in many problems in the Earth sciences. We introduce a new neural Galerkin-projection-based ROM that is generalizable and statistically calibrated with Bayesian pooling. We demonstrate that the randPROM framework requires fewer full simulations and provides uncertainty quantification to heights of tsunamis. Plain Language Summary Scientists often must simulate complex sets of equations to explain physical phenomenon, such as tsunami wave propagation. But due to modeling difficulties and long simulation times, these simulations are difficult to deploy for explaining the risks of tsunamis. We often have to rely on reduced order models (ROM), that are faster and simpler, but often fail to explain more general cases or variations in a model. We present a new ROM-based method, called a randPROM, that combines the strengths of neural networks and Bayesian statistics, to create a fast and flexible surrogate that can explain tsunami risks in near real-time. We demonstrate this model by predicting cases in Fiji as well as the real-world Tohoku 2011 disaster. 1 Introduction High-fidelity models of geophysical flows require computational resources, detailed knowledge of initial conditions, and expertise that is likely to exceed the immediate timescales and forecasting needs of an emergency response. This is particularly true of unpredictable events, such as earthquake-induced tsunamis, which require a finely-resolved numerical solution of a system of partial differential equations (PDEs) to evolve the highly non-linear, non-stationary, and local dynamics of a specific tsunami. The shallow water equations (SWEs) are a set of applicable PDEs that alleviate the computational complexity of a full three-dimensional Navier-Stokes based solver and can be used to accurately model the propagation of tsunami waves, in addition to many other atmospheric and oceanic flows. Despite their reduced computational complexity, the SWEs alone may not fully satisfy the aforementioned time constraints to generate a full distribution of possible scenarios. According to the National Oceanic and Atmospheric Administration, ”The main objective of a forecast model is to provide an estimate of wave arrival time, wave height and inundation area immediately after a tsunami event.” – the key being immediacy of the information as to be useful as a decision guiding tool in a crisis. 17 . Tools like the Tsunami Travel Time software ( 22 ) can provide quick estimates for wave arrival times, but may provide only one piece of the information needed. Reduced order models (ROMs) may also assist with the needs to provide reliable, fast, and actionable data to decision makers. These computationally efficient, reduced dimensional representations of data can provide approximations to the key characteristics for the problem of interest. A common mathematically-based ROM comes from proper orthogonal decomposition (POD), also known as principal component analysis, which assumes that the dynamics of a system can be described by orthogonal spatial patterns called modes that are activated via temporal coefficients – this technique has seen extensive use in the ROM literature of hydrodynamics flows since its introduction ( 14 , 5 ). The POD-based ROM is constructed by down-selecting a much smaller set of modes that accurately approximates the full-rank solution. A physically-based ROM can be obtained by projecting those down-selected POD modes back onto the governing PDE, exploiting algebraic rules and orthogonality, thus producing an ODE that describes how the temporal coefficients evolve in time – resulting in what is called a Galerkin-projection ROM (GP-ROM). In the context of tsunami modeling, the GP-ROM has already been used to successfully model atmospheric flows produced by SWE solvers 1 ; 2 , and physics-informed machine learning has been applied to the SWEs 6 to solve standard test cases. Despite their theoretical soundness, GP-ROMs often yield ODEs that are unstable 18 ; 7 , a problem that becomes more severe when solving low-rank versions of the system. The GP-ROM operators may even need to be pre-calibrated (scaled) or pre-conditioned to make them stable, or adding artificial viscosity terms that can dampen instability 4 . It has also been shown that the linear coefficient matrices of GP-ROMs may be optimized, producing highly stable and accurate ODE solutions called a neural Galerkin-projection (or nGP) 16 ; 8 . This work extends the nGP method, additionally optimizing the quadratic coefficient matrices, and demonstrating training success on problems requiring higher numbers of modes. Still, the limitation of the nGP is that it is only designed to accurately reproduce the temporal coefficients of the POD for a particular simulation, by solving the corresponding ODE on the initial values of that POD. A utility is that the nGP can evolve those coefficients much further in time, however, the basis is inherently unaware of dynamics at those later times as the basis is formed from the provided trajectory. For GP-ROMs and nGPs to predict other tsunami trajectories, we must assume that the modes can represent a broader class of scenarios. By applying new coefficients to these modes, the system should be able to approximate different states with acceptable accuracy. Thus, once an nGP is created, we have a stable initial value problem (IVP) and the next task becomes supplying the nGP with physically meaningful initial values. For this, we perform statistical calibration via Bayesian hierarchical pooling 11 .

Read the complete text