The wave under the seafloor is now being asked to give its own countdown
the wave under the seafloor is now being asked to give its own countdown: a generative system called gentew infers inland inundation from offshore waveforms with calibrated probabilities even when sensors are sparse, a gravity signal read by ai sizes the quake before the seismic waves arrive, and a fourier neural operator replaces hours of tsunami simulations with instant basin scale forecasts.
the coast now files its own arrival estimate.
Context
GenTEW (arXiv, August 2026) is a conditional latent diffusion model that predicts onshore tsunami inundation from offshore waveforms. According to a third-party summary it encodes 150 waveform series from the S-net and NOWPHAS networks, is trained on about 2,000 simulated scenarios, and makes 51 ensemble members in about 38 seconds on one GPU. The abstract says it was validated with 2011 Tohoku-oki data and that its uncertainty falls over time as the earthquake data come in.
The gravity result is a 2022 Nature paper (Licciardi and others). A deep learning model reads prompt elastogravity signals recorded by regional broadband seismometers in Japan before seismic waves arrive and tracks the earthquake source time function. It was trained on synthetic waveforms with real noise added, and the authors say it works once the event reaches a certain magnitude.
The tsunami propagation surrogate (EGUsphere preprint, May 2026, Korean authors) uses a Factorized Fourier Neural Operator trained on COMCOT simulations for the East Sea (Sea of Japan). On held-out cases it reports surface elevation errors of 2 to 8 cm and first-arrival timing errors of about 8 to 12 minutes, at 8.5 to 12 seconds per scenario against 87.9 to 95.7 seconds for the physics simulation.
The note says GenTEW gives calibrated probabilities 'even when sensors are sparse'. The abstract does not say that, and a third-party summary lists performance under severely sparse sensing as needing further validation, and says its evaluation leans on simulated scenarios plus one historical case. The gravity study is from 2022 and was trained on synthetic data, so 'sizes the quake before the seismic waves arrive' holds for large events in Japan once a magnitude threshold is passed, not for every quake. The note's 'instant basin scale forecasts replacing hours of simulations' overstates the neural operator paper: the simulation it replaces takes about 90 seconds per scenario there, the surrogate takes about 10 seconds, and timing errors of 8 to 12 minutes are not small for warnings. Three separate studies, two of them preprints.
Related work
- Real-time probabilistic tsunami forecasting via generative AI (arXiv) ↗Source for GenTEW, its Tohoku validation and calibrated ensemble approach.
- Real-time probabilistic tsunami forecasting via generative AI (arXivLens summary) ↗Third-party summary for the 150 waveforms, 2,000 scenarios, 51 members and listed limitations.
- Instantaneous tracking of earthquake growth with elastogravity signals (Nature) ↗Source for the elastogravity signal deep learning model.
- Monitoring Earthquakes at the Speed of Light (Eos) ↗Plain language explainer of the elastogravity work.
- A Factorized Fourier Neural Operator Surrogate for Basin-Scale Tsunami Propagation (EGUsphere) ↗Source for the Fourier neural operator timing and error numbers.
Watch next
- Look for tests of GenTEW on real sparse or failed sensor cases. Check whether the elastogravity model has been tested on real large events since 2022.
Sources
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 4 October 2026 at 00:49 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →