The fault is now being asked to predict its own sequence
the fault is now being asked to predict its own sequence: a diffusion model called quakegen generates aftershock rate fields from recent seismicity, a physics optimized network learns the gutenberg richter and omori utsu laws as constraints, and an lstm estimates future magnitudes with 0.07 bits of information gain per event.
the ground now files its own rupture forecast.
Context
The QuakeGen paper (Weiqiang Zhu, UC Berkeley) describes a conditional diffusion model that forecasts future seismicity as spatiotemporal fields of earthquake frequency and maximum magnitude, conditioned on the seismicity already observed. Sampling it repeatedly gives an ensemble whose mean or median is the forecast and whose spread reflects uncertainty. POSEIDON is a physics-informed energy-based model that embeds the Gutenberg-Richter relation and the Omori-Utsu aftershock decay as learnable constraints.
MAGNET, in Scientific Reports, uses a multi-encoder network with LSTM units on past hypocenter locations, times and magnitudes. It reports an information gain of approximately 0.07 bit per earthquake on average over the Gutenberg-Richter benchmark in Southern California, Japan and New Zealand catalogs.
The note joins three separate studies. QuakeGen forecasts earthquake frequency and maximum magnitude fields, POSEIDON uses physical laws as constraints, and the 0.07 bits is from MAGNET, which forecasts the magnitude of future events and not aftershock rates. The 0.07 bit is an average over three regions. The paper also reports about 0.15 bits in one Japanese region, and says the gain is over a Gutenberg-Richter baseline.
The paper says its gain turns negative for some events, such as after the 2011 Tohoku earthquake, so 'forecast' here means a modest average improvement over a baseline, not reliable prediction of single events.
'The ground now files its own rupture forecast' is the author's line. The pages read describe statistical and learned models of catalogs, and none claims to forecast a specific rupture. The QuakeGen preprint date and any peer review were not confirmed in the page read, so its status is unsupported here, not refuted.
Related work
- Earthquake Aftershock Forecasting using Conditional Generative Models (Weiqiang Zhu, UC Berkeley, arXiv 2607.24109) ↗Source for QuakeGen's diffusion design and what it forecasts.
- POSEIDON: Physics-Optimized Seismic Energy Inference and Detection Operating Network (arXiv 2601.02264) ↗Source for the physics-informed constraints.
- Earthquake magnitudes depend on seismic history, as revealed by a neural network analysis (Scientific Reports, 2026) ↗Source for MAGNET, its LSTM encoders and the 0.07 bit average.
Watch next
- Read how QuakeGen is scored against ETAS. Check whether POSEIDON reports results on held-out earthquake sequences.
Sources
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 3 October 2026 at 20:31 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 →