Tsunami Early Warning From Global Navigation Satellite System Data Using Convolutional Neural Networks. Issue 20 (20th October 2022)

• Record Type: Journal Article
• Authors: Rim, Donsub; Baraldi, Robert; Liu, Christopher M.; LeVeque, Randall J.; Terada, Kenjiro
• Is Part Of: Geophysical research letters. Volume 49:Issue 20(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-10-20
• Subjects: tsunami forecasting -- machine learning -- neural network -- GNSS -- synthetic ruptures -- GeoClaw software; Geophysics -- Periodicals; Planets -- Periodicals; Lunar geology -- Periodicals; 550
• DOI: 10.1029/2022GL099511
• Languages: English
• ISSNs: 0094-8276

Abstract

Abstract: We investigate the potential of using Global Navigation Satellite System (GNSS) observations to directly forecast full tsunami waveforms in real time. We train convolutional neural networks to use less than 9 min of GNSS data to forecast the full tsunami waveforms over 6 hr at select locations, and obtain accurate forecasts on a test data set. Our training and test data consists of synthetic earthquakes and associated GNSS data generated for the Cascadia Subduction Zone using the MudPy software, and corresponding tsunami waveforms in Puget Sound computed using GeoClaw. We use the same suite of synthetic earthquakes and waveforms as in earlier work where tsunami waveforms were used for forecasting, and provide a comparison. We also explore varying the number of GNSS stations, their locations, and their observation durations. Plain Language Summary: Producing rapid real‐time forecasts for tsunamis in the first few minutes of an earthquake is a challenging problem. Accurate forecasts often rely on direct measurements of the tsunami, which are only available at sparse locations, and only after the tsunami has passed the sensors. Real‐time numerical modeling of the tsunami is also time consuming. This work attempts to bypass these difficulties by considering a model that can forecast tsunami wave heights based only on Global Navigation Satellite System (GNSS) data, which is available within minutes from an extensive network of stations. We present some initial results using this approach for hypothetical tsunamis originating from the Cascadia Subduction Zone, with forecast locations in Puget Sound. We show that this approach gives comparable results to earlier work based on observing tsunami waveforms for 30 or 60 min, but now using only a few minutes of GNSS data. We explore varying the number of GNSS stations and find that the model yields accurate forecasts when as few as 20 GNSS stations are used, and outperforms our previous model when additional stations are used. The model performs well even when only the initial 4 min of GNSS data is used. Key Points: A deep convolutional neural network (CNN) is trained to predict tsunami waveforms, based only on Global Navigation Satellite System (GNSS) data for hypothetical earthquakes Less than 9 min of data at GNSS stations selected from the existing dense network is used to predict 6 hr of tsunami waveforms Results compare favorably with a previous forecast model based on 30 or 60 min of tsunami waveform data