Skip to main content
Geophysics LaboratoryCUHK

Research

Our research

The long-term goals of our research are to advance the understanding of earthquake, volcanic, and Earth surface processes to aid in hazard mitigation and sustainable development.

We seek to quantify how stress perturbations, fluid migration, and subsurface dynamics drive seismic and volcanic activity. We are best known for applying machine learning to produce high-resolution earthquake catalogues, which are especially powerful when paired with advanced statistical methods to reveal otherwise inaccessible features of fault systems and volcanic plumbing.

Our group integrates seismology (dense arrays, ocean-bottom seismometers), geophysical data analysis (catalogue construction, waveform classification), and computation to study hazard processes and to improve early warning and forecasting. Group members move fluidly between data and models, often inventing new methods to answer their questions.

We work on many different tectonic and volcanic systems because we find ourselves drawn, again and again, to the power of high-resolution seismic data: as catalogue completeness improves, pattern recognition becomes easier — until the resolution gets too good, and then the spatiotemporal complexity begins to reveal itself and the interesting challenges begin.

Interview with Prof. Yen Joe Tan (陳衍佐教授) —watch on YouTube (opens in a new tab)

Theme 01

Machine learning applications in geophysics

We have applied machine learning to identify eruption precursors, to improve detection of different types of seismic events, and to produce high-resolution earthquake catalogues that map fault structures and constrain earthquake interactions.

Those catalogues have let us characterise foreshock sequences and swarms, including along oceanic transform faults where conventional detection thresholds leave most of the seismicity unseen.

Key publications

  • Tan et al. (2021) (opens in a new tab) — Machine-learning-based high-resolution earthquake catalog reveals how complex fault structures were activated during the 2016–2017 central Italy sequence. The Seismic Record
  • Liu et al. (2024) (opens in a new tab) — Intersection between tectonic faults and magmatic systems promotes swarms with large-magnitude earthquakes around the Tengchong volcanic field, southeastern Tibetan Plateau. Geology
  • Zhong et al. (2024) (opens in a new tab) — Deep-learning-based phase picking for volcano-tectonic and long-period earthquakes. Geophysical Research Letters
  • Liu et al. (2025) (opens in a new tab) — Fluids and fault structures underlying the complex foreshock sequence of the 2021 Mw 6.1 Yangbi earthquake. Earth and Planetary Science Letters
  • Liu et al. (2025) (opens in a new tab) — Evaluating the performance of machine-learning-based phase pickers when applied to ocean bottom seismic data: Blanco oceanic transform fault as a case study. Geophysical Journal International
  • Liu et al. (2026) (opens in a new tab) — Large earthquakes along the Mendocino oceanic transform fault hardly have any foreshocks. Geophysical Research Letters
Recall, precision and F1 score for P- and S-wave picking by PhaseNet and EQTransformer as a function of frequency index, comparing the published models with versions retrained in this study, together with the frequency-index and signal-to-noise distributions of the training data. — open the full-size figure in a new tab
Phase-picking performance against frequency index, for published models and models retrained here.

Theme 02

Fault response to stress changes

We recognised that external stress perturbations can modulate fault slip behaviour and earthquake nucleation. In both systematic surveys and isolated case studies, we have demonstrated how tidal stress affects earthquake rates and frequency–magnitude distributions, constraining frictional properties.

We have also identified how dynamic triggering by teleseismic waves, reservoir-induced seismicity, and landslide-dammed lakes reveal the state of stress on faults.

Key publications

Schematic block of a mountain valley where landslides have dammed a river to form a lake. Arrows show direct loading from the impounded water and fluid diffusion into the rock beneath, triggering an earthquake below the dam. — open the full-size figure in a new tab
How a landslide dam loads the rock beneath it, through both direct loading and fluid diffusion.

Theme 03

Dynamics of volcanic systems

We are interested in the structural dynamics and hazard assessment of submarine and subaerial volcanic systems. Using ocean-bottom seismometer (OBS) data, we analyse microearthquakes, mixed-frequency earthquakes, tremors, and seismo-acoustic events to characterise magmatic processes and seismic velocity changes.

With machine learning platforms, we develop models for detecting long-period and volcano-tectonic earthquakes, and we continue to push the limits of what OBS networks can resolve about an active volcanic system.

Key publications

Three-dimensional block diagram of a submarine volcano. Bathymetry is shown above; below it, two tremor phases migrate upward from the axial melt lens through the lower and upper crust at roughly 1.8 and 1.1 kilometres per hour in the days before an eruption. — open the full-size figure in a new tab
Tremor migrating upward from the axial melt lens in the days before a submarine eruption.

Theme 04

Low-frequency and long-period earthquakes

We are intrigued by the spatiotemporal clustering and scaling properties of low-frequency earthquakes (LFEs), and we apply these insights to large-scale studies of slow-slip events and fault slip velocities.

We investigate LFE source processes to track fluid and magma movement and to support eruption forecasting. We like to bring a statistical seismology perspective to the table, to understand the relative importance of tectonic and volcanic loading in shaping the fault systems we observe today.

Key publications

Seismic moment plotted against rupture duration for two event populations, coloured by frequency index, beside three-component waveforms at station SN07 comparing a regular event with a low-frequency one. — open the full-size figure in a new tab
Moment against rupture duration for two event populations, and how their waveforms differ.

Theme 05

Earth surface processes and hazard cascades

A major challenge in geohazard science is the inability to directly monitor and quantify rapid mass movements and their cascading impacts. We use seismic methods to quantify the dynamics of large landslides, dam breaches, and outburst floods.

In recent work we discovered that landslide hazard cascades can trigger earthquakes, revealing previously unrecognised linkages between surface and subsurface hazards. Our long-term goal is to increase the information content of every seismic record, to reveal the full chain of geophysical events and improve multi-hazard early warning.

Key publications

  • Zhang et al. (2024) (opens in a new tab) — Seismic monitoring and geomorphic impacts of the catastrophic 2018 Baige landslide hazard cascades in the Tibetan plateau. Journal of Geophysical Research: Earth Surface
  • Zhang et al. (2024) (opens in a new tab) — Landslide hazard cascades can trigger earthquakes. Nature Communications
  • Ho et al. (2025) (opens in a new tab) — Analysing dynamics of the 2000 Yigong landslide in the Tibetan plateau using seismic observations. Landslides
  • Zhang et al. — Citizen seismology enables low-cost early detection of catastrophic mass movements. PNAS (under review)
Map of broadband seismic stations across Asia that recorded a single event near Lhasa, with the aligned waveform section, the vertical-component velocity trace at station LSA and its spectrogram. — open the full-size figure in a new tab
A large mass movement recorded at regional and teleseismic distance, from station map to spectrogram.

A complete, numbered list of the group's work — with peer reviews where they are public — is on the publications page.