Attention-Based Reconstruction of Full-Field Tsunami Waves from Sparse Tsunameter Networks

TL;DR

This paper introduces the Senseiver, an attention-based neural network that reconstructs detailed tsunami wavefields from sparse sensor data, significantly enhancing forecasting accuracy over traditional interpolation methods.

Contribution

The paper presents a novel attention-based neural network architecture for tsunami wavefield reconstruction from sparse data, including unseen epicenters, outperforming existing interpolation techniques.

Findings

Senseiver accurately reconstructs high-resolution tsunami wavefields from sparse observations.

The approach outperforms linear interpolation with Huygens-Fresnel in dense observation network generation.

The method remains effective even when tsunami epicenters are not in the training set.

Abstract

We investigate the potential of an attention-based neural network architecture, the Senseiver, for sparse sensing in tsunami forecasting. Specifically, we focus on the Tsunami Data Assimilation Method, which generates forecasts from tsunameter networks. Our model is used to reconstruct high-resolution tsunami wavefields from extremely sparse observations, including cases where the tsunami epicenters are not represented in the training set. Furthermore, we demonstrate that our approach significantly outperforms the Linear Interpolation with Huygens-Fresnel Principle in generating dense observation networks, achieving markedly improved accuracy.