Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Real-time probabilistic tsunami forecasting in Cascadia from sparse offshore pressure observations

Stefan Henneking, Fabian Kutschera, Sreeram Venkat, Alice-Agnes Gabriel, Omar Ghattas

Abstract

Near-field tsunami early warning in the Cascadia Subduction Zone is limited by sparse offshore observations. We show that a hypothetical network of 175 seafloor pressure sensors can support real-time Bayesian inference of tsunamigenic seafloor motion and probabilistic tsunami forecasts for two fully-coupled Cascadia earthquake dynamic rupture--tsunami scenarios, a partial rupture and a margin-wide rupture. The complex oceanic acoustic, Rayleigh, and tsunami wavefields in both scenarios are similar during the first two minutes and then diverge. Using an acoustic--gravity inversion with offline precomputation and online assimilation of pressure data, tsunami forecasts are obtained in less than a second. We leverage a Bayesian inversion-based framework that splits the computations into an offline precomputation phase performed with large-scale computing facilities, and an online phase that computes forecasts from real-time data and can be executed on a laptop. Forecast errors remain low at 22.1% for the margin-wide rupture and 19.6% for the partial rupture.

Real-time probabilistic tsunami forecasting in Cascadia from sparse offshore pressure observations

Abstract

Near-field tsunami early warning in the Cascadia Subduction Zone is limited by sparse offshore observations. We show that a hypothetical network of 175 seafloor pressure sensors can support real-time Bayesian inference of tsunamigenic seafloor motion and probabilistic tsunami forecasts for two fully-coupled Cascadia earthquake dynamic rupture--tsunami scenarios, a partial rupture and a margin-wide rupture. The complex oceanic acoustic, Rayleigh, and tsunami wavefields in both scenarios are similar during the first two minutes and then diverge. Using an acoustic--gravity inversion with offline precomputation and online assimilation of pressure data, tsunami forecasts are obtained in less than a second. We leverage a Bayesian inversion-based framework that splits the computations into an offline precomputation phase performed with large-scale computing facilities, and an online phase that computes forecasts from real-time data and can be executed on a laptop. Forecast errors remain low at 22.1% for the margin-wide rupture and 19.6% for the partial rupture.
Paper Structure (11 sections, 2 equations, 6 figures, 1 table)

This paper contains 11 sections, 2 equations, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Cascadia fully-coupled dynamic rupture and tsunami simulations
  4. Bayesian inversion framework
  5. Results
  6. Fully-coupled margin-wide and partial dynamic rupture models
  7. Inference of seafloor motion and forecast of tsunami wave heights
  8. Discussion
  9. Implications for near-field tsunami forecasting
  10. Implications for real-time tsunami forecasting
  11. Conclusions

Figures (6)

  • Figure 1: (a) Overview of the Pacific Northwest coast and major plate boundaries Bird_2003_UpdatedDigitalModel. The blue polygon outlines the domain for the 3D fully-coupled earthquake dynamic rupture and tsunami simulations performed with SeisSol Gabriel_2025_SeisSol. The dashed cyan line marks a densely spaced virtual receiver profile located at the sea surface, following the Juan de Fuca (JF) and North American (NA) plate boundary from 50°N to 42°N. (b) Earthquake slip distributions for the margin-wide and partial rupture fully-coupled simulations, closely matching the Solid Earth-only dynamic rupture scenarios of Glehman_2025_PartialRupturesGoverned. (c) Sea surface height anomaly (ssha) for the margin-wide rupture scenario at 90 s, during tsunami generation, and at 420 s, at the end of the fully-coupled simulation. (d) Space-time evolution of sea surface vertical velocities (ssvv) along the trench profile for the margin-wide and partial rupture scenarios. Seismo-acoustic wave amplitudes dominate during the earthquake rupture and early tsunami generation phase. After rupture arrest, oceanic Rayleigh wave signals attenuate rapidly, whereas acoustic waves remain visible until the end of the modeled time window for both cases.
  • Figure 2: Snapshots of the true (left column) and inferred (middle and right columns) seafloor normal displacements at the final time ($t$ = 420 seconds) for a margin-wide rupture (top row) and partial rupture (middle row) scenario, and their associated uncertainties given as pointwise standard deviations (bottom row). Results in the middle and right columns respectively correspond to inversion using synthetic data from 600 and 175 hypothesized ocean bottom pressure sensors, with the sensor locations marked in the corresponding uncertainty plots of the bottom row. The bottom left plot depicts ocean bathymetry with the 21 tsunami wave height forecasting locations.
  • Figure 3: Tsunami wave height forecasts and their uncertainties (depicted as 95% credible intervals) at a subset of the 21 target locations (labeled as QoI #1--#21, with locations as shown in the bottom left plot of Figure \ref{['fig:inversion-results']}), obtained with 600 sensors (left column) and 175 sensors (right column), for the margin-wide rupture (top row) and partial rupture scenarios (bottom row). The relative errors of the tsunami forecasts are 18.6% (600 sensors) and 22.1% (175 sensors) for the margin-wide rupture, showing only a modest degradation of the forecast quality with the sparser sensor network. Similar tsunami forecast errors of 18.1% (600 sensors) and 19.6% (175 sensors) are obtained for the partial rupture scenario.
  • Figure 4: Snapshots of the slip rate every 30 s for (a) the margin-wide rupture and (b) the partial rupture scenario. The respective final moment magnitude is annotated at the top right. Note the earlier rupture arrest for the partial rupture scenario compared to the margin-wide rupture. Animations of the slip rate are uploaded as additional, supplementary movies.
  • Figure 5: Snapshots of the seismo-acoustic and gravity wave (i.e., tsunami) excitation at the sea surface as indicated by the sea surface height anomaly (ssha) every 60 s for (a) the margin-wide rupture and (b) the partial rupture scenario. Animations of the time-dependent ssha are uploaded as additional, supplementary movies.
  • ...and 1 more figures