Image showing the tsunami recorded by the DART buoys and SWOT satellite. (Image Credit: The Seismic Record)
On July 29, 2025, a magnitude 8.8 earthquake struck the Kuril-Kamchatka region, triggering a huge tsunami that rolled across the Pacific. Luckily, the SWOT satellite passed overhead, capturing the event in high resolution. It revealed a complex, spreading pattern rather than a single wave crest.
DART stations recorded the tsunami. Each DART system features a seafloor bottom pressure recorder (BPR) and a surface buoy. The BPR continuously measures water-column pressure with sub-centimeter precision, enabling it to detect tsunami waves as low as a few millimeters high. Data is transmitted acoustically to the surface buoy before being relayed to the shore via satellite. Multiple DART stations captured the Kamchatka tsunami, recording the wave’s propagation.
Researchers examined three sites close to the tsunami source. All data were lightly processed, resampled to 15-second intervals, and went through a high-pass filter (2-hour threshold applied) to remove tides. With this refined dataset, the team performed a tsunami source inversion to estimate the seafloor deformation that generated the recorded signals.
The study also incorporated sea surface height (SSH) measurements from the SWOT satellite, providing extremely detailed views of the tsunami. Their analysis used data collected along the ascending path of SWOT’s orbit.
SWOT maps sea surface height across a wide swath of approximately 120 km at 2-km resolution. These measurements come from the KaRIn low-rate ocean measurements produced by NASA/JPL and CNES. Level three processing removes SWOT’s instrument errors, including roll and phase. It also filters out points affected by atmospheric disturbances. Researchers used the denoised sea level anomaly field, generated with a U-net-based convolutional neural network, to suppress high-frequency noise while preserving geophysical patterns. They also incorporated SWOT-Nadir altimeter observations along the center of the swath to support their interpretation.
This study uses the U.S. Geological Survey (USGS) finite-fault Version 3 source model. It combines the teleseismic body and surface waves, and ground-deformation measurements derived from InSAR. Using both datasets enhances how well the slip pattern on the fault can be resolved. Teleseismic waveforms help constrain the rupture timing and the total seismic moment released, while the InSAR data provide detailed information on the deformation near the fault. This USGS model represents the earthquake rupture on a segmented planar fault following the geometry of the Kuril-Kamchatka subduction zone. Slip, rake, rupture onset time, and rise time for each subfault are determined via simulated annealing optimization.
SWOT nadir gauges show tsunami waves and dispersive patterns, with SRTM bathymetry indicating the track and sample trajectories. (Image Credit: The Seismic Record)
To map the seafloor deformation produced by the earthquake, the team modeled displacements using an elastic layered half-space representation of the Earth. The model uses the UTHO1.0 crust over a PREM-based mantle, so the calculations can incorporate realistic variations in seismic velocity and elastic properties. The vertical displacement field is the starting point for tsunami generation.
However, since the USGS model’s tsunami prediction didn’t match the DART observations, the team produced an estimation of the tsunami source. In this case, they inverted deep-ocean records from three DART stations: 21415, 21416, and 21419, which helped determine the sea-surface displacement for this event. They used a “Gaussian lump” parameterization to represent the surface disturbance as a sum of localized Gaussian-shaped patches. Each Gaussian has a 15 km deviation and is placed on a grid spaced 0.3” (latitude and longitude). This technique details a smooth, compact description of the source while keeping the parameters manageable.
The key advantage of this formulation is that it doesn’t require assumptions about the earthquake’s fault geometry, rake, or slip distribution. Instead, tsunami data provide the spatial pattern of uplift and subsidence. The inversion assumes a linear response, and while GeoClaw solves the nonlinear shallow-water equation, the early tsunami waves in the inversion are small enough to keep the linear approximation precise.
They performed the inversion over a domain surrounding the USGS V3 model. As a result, the solution remained physically consistent with independent seismic and geodetic data. The team used GeoClaw to calculate the tsunami response for each Gaussian patch.
Each Gaussian had a sea-surface displacement applied and was propagated through the model to the DART stations, generating unit-amplitude synthetic waveforms. The tsunami at the buoys is represented as a linear combination of the precomputed responses, with the inversion solving for the best-fitting amplitude of each Gaussian to reproduce the recorded signals.
DART time series were band-pass filtered between two minutes and two hours to isolate the tsunami band and reduce high-frequency noise and long-period background variations like tides or infragravity waves. Initial DART record portions are impacted by Rayleigh waves from the earthquake that generate vertical accelerations appearing as pressure fluctuations in the water column. To minimize contamination and maintain inversion linearity, the researchers manually selected the first clear wavelength of the tsunami signal for each buoy and restricted the inversion to that interval. The inversion creates a sea-surface displacement field from the tsunami observations.
With GeoClaw and the team’s initial condition model, the tsunami was propagated across the open ocean at high resolution via the Shuttle Radar Topography dataset. While passing, SWOT moved from south to north along its polar orbit, sampling the tsunami at slightly different times rather than a single snapshot. To produce a synthetic SWOT track, the team extracted simulated tsunami amplitudes at the coordinates of each SWOT pixel at the relevant time. They achieved this by generating output at over 60,000 synthetic tide gauges and interpolating the results to produce a continuous synthetic product.
For better full wavefield reproductions, the team incorporated dispersive wave effects using the Boussinesq-type solvers in GeoClaw. Dispersive corrections are interpolated across the adaptive mesh refinement levels, allowing GeoClaw to model dispersive propagation efficiently while maintaining stability near shore.
Have a story tip? Message me here at element14.
