Lingsen Meng

Research

How earthquakes start, grow and stop

We image earthquake ruptures with seismic arrays and joint seismic–geodetic inversion, compare what we see with fracture mechanics, and use the results for tsunami and hazard assessment. Most of our work starts from a large earthquake that has just happened; the rest builds the methods those studies need. The six topics below each have a page with a summary of the main results, interactive figures drawn from published data, and the key papers.

Rupture physics & supershear earthquakes

Rupture speed controls how strongly a fault shakes the ground. We measure it with slowness-enhanced back-projection and far-field Mach waves. A systematic search of large strike-slip earthquakes since 2000 found oceanic supershear ruptures to be as frequent as continental ones, and at least 14% of Mw ≥ 6.7 shallow strike-slip events to be supershear. The 2025 Mandalay earthquake ran ~450 km at supershear speed.

Supershear explainer and data
  • Xu et al. 2025, Science: Mandalay supershear rupture
  • Bao et al. 2022, Nat. Geosci.: global frequency of supershear
  • Bao et al. 2019, Nat. Geosci.: early and persistent supershear at Palu

Back-projection imaging

Back-projection stacks array recordings to locate where high-frequency waves are radiated during a rupture, second by second. We introduced MUSIC into back-projection, the reference-window strategy against the “swimming” artefact, aftershock-based slowness calibration (SEBP), 3-D back-projection with depth phases, and PKIKP back-projection for Southern-Hemisphere events.

How back-projection works
  • Meng et al. 2012, Science: 2012 Sumatra rupture in a maze
  • Meng et al. 2016, GRL: aftershock slowness calibration
  • Xu et al. 2025, JGR: PKIKP back-projection

Finite-fault & joint inversion

Slip models from teleseismic, strong-motion, GNSS and InSAR data on multi-segment faults, combined with back-projection to constrain rupture timing: bifurcating faults at Maduo 2021, a multi-segment rupture in a millennia supercycle in Türkiye 2023, dual initiation at Noto 2024, and a conjugate fault system at Hualien 2024.

3-D slip models
  • Xu et al. 2024, Science: Noto dual initiation
  • Xu et al. 2023, Commun. Earth Environ.: Kahramanmaraş
  • Xu et al. 2023, JGR: Maduo bilateral rupture

Tsunami source imaging & warning

Seismic waves reach coastal stations minutes before the tsunami. We use that head start with back-projection-based warning, and we image the initial sea surface directly from tide-gauge and pressure records by time-reversal and adjoint-state inversion, which does not need a fault model and resolved a submarine landslide source after the 2024 Noto earthquake.

Tsunami page
  • Mohanna et al. 2025, EPS: Noto tsunami and landslide source
  • Zhou et al. 2019, JGR: adjoint-state tsunami inversion
  • Xie & Meng 2020, GRL: multi-array BP for tsunami warning

Swarms, slow slip & fibre sensing

High-resolution catalogues from template matching and machine-learning picking show how fluids, afterslip and fault geometry drive swarms (Noto, Tengchong) and how slow unlocking precedes megathrust events (Iquique, Illapel). We also develop processing for distributed acoustic sensing (DAS) on optical fibres.

Swarms and DAS
  • Mohanna et al. 2026, EPSL: Noto swarm cessation
  • Ma et al. 2026, Tectonophysics: Tengchong swarms
  • Ma, Meng & Lin 2026, GJI: DAS compression

AI for geoscience

Machine learning now detects far more earthquakes than analysts can review. We build pickers (EdgePhase), measure how many machine detections a seismologist would endorse, and run agent-based pipelines that take raw waveforms to relocated catalogues with every step recorded. Related work on AI-assisted mathematics is on the AI for Math page.

AI for geoscience
  • Meng et al. 2026, submitted: can seismologists verify ML-detected earthquakes?
  • Feng et al. 2022, G-Cubed: EdgePhase picker

Data behind the figures

All downloads

Every interactive figure on these pages is drawn from a released dataset (mostly the Zenodo archives of the papers) and links to the JSON or CSV file it reads. The data page lists each file with its source and citation.