← Back to Sessions

Real-Time Prediction of Tsunami Amplitude Using Gaussian Process Regression

Description: 

Tsunami Warning Centers depend on coastal tide gauges to measure tsunami wave heights for rapid decision making and alerting. To isolate the tsunami signal from the tidal background, a physics-based harmonic analysis model is often used to predict the tides. This prediction is then subtracted from observed data, ideally leaving a tsunami-only waveform plus noise. However, this model is less accurate in waterways where complex bathymetry and freshwater input introduce complications that are not accounted for in the model. In such cases, determining tsunami wave height is challenging because the tide is poorly modeled, making it difficult for Tsunami Warning Centers to produce accurate measurements.

Data-derived methods such as Gaussian processes may be used to address this need. Gaussian processes are a nonparametric, Bayesian approach that avoids issues with traditional physics-based analyses. This method uses a radial basis kernel function to capture data structure, and its properties are determined directly from tide data through a set of hyperparameters. The study demonstrates improved algorithm performance using tide data recorded in Valdez, AK, during the February 27, 2010 Chilean tsunami; the March 11, 2011 Tohoku tsunami; and the January 15, 2022 Tonga tsunami.

Session: Six Decades of Tsunami Science: From the Source of the 1964 Tsunami to Modern Community Preparedness - I

Type: Oral

Date: 5/2/2024

Presentation Time: 02:45 PM (local time)

Presenting Author: Terry

Student Presenter: No

Invited Presentation: 

Authors