SeisLM
Learns a general representation of seismic waveforms, fine-tuned for event detection, phase picking and event classification.
- Field
- Geophysics
- Method
- Foundation model
- Data it takes
- waveform
- License
- No published license — restricted use
SeisLM is an encoder for single-station, three-component seismic waveforms sampled at 100 Hz, pretrained without labels on eight SeisBench datasets. It comes in two sizes, 11.4 million and 90.7 million parameters, and is fine-tuned for event detection, phase identification and onset picking, or foreshock–aftershock classification. In the paper, its advantage over PhaseNet is largest when labels are scarce.
The released weights are the pretrained encoder only, so every task needs fine-tuning first. SeisBench provides pickers that are already trained.
The lab offers SeisLM to Washington University groups only. Upstream publishes no license file for the code or the weights; its package metadata names MIT but ships no license text. The lab does not redistribute the weights.
Catalog entry last checked 2026-10-03. All models →
Run SeisLM on your data
The lab runs this on its own compute, in a container, with the inputs and parameters recorded alongside the result. Initial scoping conversations are free.
Contact the Lab