Physics-Aware Machine Learning for Seismic and Volcanic Signal Interpretation

William Thorossian

cs.LGMar 18, 2026

Physics-Aware Machine Learning for Seismic and Volcanic Signal Interpretation

Authors: William Thorossian

Abstract

Modern seismic and volcanic monitoring is increasingly shaped by continuous, multi-sensor observations and by the need to extract actionable information from nonstationary, noisy wavefields. In this context, machine learning has moved from a research curiosity to a practical ingredient of processing chains for detection, phase picking, classification, denoising, and anomaly tracking. However, improved accuracy on a fixed dataset is not sufficient for operational use. Models must remain reliable under domain shift (new stations, changing noise, evolving volcanic activity), provide uncertainty that supports decision-making, and connect their outputs to physically meaningful constraints. This paper surveys and organizes recent ML approaches for seismic and volcanic signal analysis, highlighting where classical signal processing provides indispensable inductive bias, how self-supervision and generative modeling can reduce dependence on labels, and which evaluation protocols best reflect transfer across regions. We conclude with open challenges for robust, interpretable, and maintainable AI-assisted monitoring.

Figures & tables

Task	Input	Model	Strengths	Limits	Shift robust.	Interpret- ability	Ops readiness
Detection & picking	3C windows	CNN / Transformer	High recall with mature tooling	Station bias without strict split design	Medium	Medium	High
Association	Picks + metadata	Sequence / BGMM	Event-level consistency in swarms	Upstream pick errors propagate	Medium– High	High	High
Transfer / adaptation	Source + target + metadata	Transfer + DANN	Label-efficient adaptation to new regions	Negative transfer; tuning overhead	High †	Medium	Medium
Uncertainty-aware classification	Scores + calibration split	Ensemble + conformal	Risk-aware thresholding and coverage control	Extra computation under severe shift	Medium– High	High	Medium– High
Foundation representations	Unlabeled waveform archives	SeisLM-like encoders	Strong few-label transfer across tasks	Expensive pretraining; limited common benchmarks	Promis- ing	Low– Medium	Medium
Volcanic multi-site classification	Seismic / infrasound spectrograms	VOISS-Net-like models	Broad cross-volcano detection coverage	Confusions on unseen transient classes	Medium	Medium	Medium– High

Table 1: Comparative synthesis of representative ML approaches for seismic and volcanic monitoring. Ratings are qualitative and intended as survey-level guidance rather than leaderboard claims. † High when validated with strict out-of-domain protocols.

Challenge	Operational failure mode	Recommended strategies	Monitoring metrics
Domain shift (station/time/region)	Recall drop on unseen stations or new noise regimes	Leave-one-station-out and time-blocked validation; metadata-aware normalization; targeted local fine-tuning	POD at fixed FAR by station, drift indices
Uncalibrated confidence	Overconfident false alerts or missed events in rare regimes	Temperature scaling/ensembles; conformal prediction sets; “unknown” rejection policy	ECE, reliability slope, conformal coverage/set size
Noisy labels and ambiguous taxonomy	Brittle boundaries and poor cross-observatory transfer	Soft/hierarchical labels; cross-analyst audits; weak-label robust losses	Inter-annotator agreement, class-wise calibration
Telemetry artifacts and missing data	False triggers from glitches, silent failures on clipped/gappy streams	Explicit masking; corruption stress tests; conservative censoring over aggressive interpolation	Corruption risk curves, missing-channel robustness
Weak physical consistency	Plausible class scores but implausible picks/locations	Hybrid losses with travel-time constraints; post hoc association consistency checks	Travel-time residual tails, association consistency rate
Lifecycle and maintainability	KPI decay after network/configuration updates	Versioned data/models; periodic revalidation; rollback-ready deployment policy	Rolling OOD score, monthly KPI deltas

Table 2: Challenges-versus-strategies summary for operational ML in seismic and volcanic monitoring.

Physics-Informed ML

Physics-informed machine learning combines data-driven models with physical laws to solve, predict, and infer scientific systems.

See all 833 papers

Learning Earthquake Wave Arrival Time Picking from Labels with Inaccuracies
Jun 13, 2026Sen Li, Xu Yang, S. Mostafa Mousavi +5Contrastive Learning Seismology
Department of Earth and Planetary Sciences, Harvard University, Cambridge, MA, USA. · School of Computer Science and Technology, China University of Mining and Technology, Xuzhou, Jiangsu, China · School of Mines, China University of Mining and Technology, Xuzhou, Jiangsu, China +1
Evaluating PhaseNet on Teleseismic Data with MsPASS
May 10, 2026Jinxin Ma, Yinzhi Wang, Gary L. Pavlis +1Efficient Neural Network Inference Seismology
Texas Advanced Computing Center, The University of Texas at Austin · Department of Earth and Atmospheric Sciences, Indiana University, Bloomington, IN 47405
OpenSeisML: Open Large-Scale Real Seismic and well-log Dataset for Generative AI
May 19, 2026Ipsita Bhar, Huseyin Tuna Erdinc, Thales Souza +2Synthetic Data Generation Training Data Curation
Georgia Institute of Technology, Atlanta, USA · Osokey Ltd, Henley-on-Thames, UK

Jun 13, 2026J/K move · Enter open · S save

Learning Earthquake Wave Arrival Time Picking from Labels with Inaccuracies

Sen Li, Xu Yang, S. Mostafa Mousavi, Anye Cao+4

Department of Earth and Planetary Sciences, Harvard University, Cambridge, MA, USA. · School of Computer Science and Technology, China University of Mining and Technology, Xuzhou, Jiangsu, China · School of Mines, China University of Mining and Technology, Xuzhou, Jiangsu, China+1

Contrastive Learning Seismology Learning with Noisy Labels

Open paper AskPDF arXiv

Inaccurately labeled training data, or "label noise", poses a significant threat to the integrity of supervised machine learning models. This corruption directly degrades performance by teaching the model erroneous mappings between features and labels, which leads to poor generalization and reduced accuracy on properly labeled validation and test data. Current seismological applications mainly rely on large-scale training sets or data augmentation to reduce the label-noise impact, which can be labor-intensive and costly. Here, we introduce a Label Noise-Contrastive Robust Learning (LaNCoR) approach that can effectively handle noisy labels in seismic signal processing tasks, without requiring large-scale training datasets. In this approach, the input waveform feature and label representation distributions are aligned in the feature space to correct mislabeling and reduce its impact on the training process. We present LaNCoR's performance on the task of P-phase arrival-time picking of real microseismic data using two baseline models and training approaches. Our results indicate that LaNCoR can improve performance by up to 28.8% across performance metrics. This approach holds great promise for model training in seismology and geosciences.

Abstract

Physics-Informed ML

Explore similar work

Learning Earthquake Wave Arrival Time Picking from Labels with Inaccuracies

Evaluating PhaseNet on Teleseismic Data with MsPASS

OpenSeisML: Open Large-Scale Real Seismic and well-log Dataset for Generative AI