Research projects
Research projects
Embedded Files
1. Dynamics of caldera collapse earthquakes
Earthquake dynamics in caldera environment can be extremely complex due to the elevated temperature at depths, the heterogeneous rheology on the ring fault, and the mechanical coupling with the underlying magma chamber. During the caldera collapse sequence of Kilauea volcano in 2018, thousands of volcano (Mw < 4) tectonic earthquakes per day occurred in between Mw > 5 caldera collapse earthquakes. A key question, from a hazard perspective, is what drives these earthquakes? In Wang et al., 2023, we examine at two hypotheses: the earthquakes are driven by 1) localized stressing due to rapid (meter/day) ring fault creep in between collapse earthquakes (as recorded by GPS). 2) bulk stressing of the caldera roof due to magma chamber depressurization. We evaluate the hypotheses using rate-and-state friction based seismicity models, and investigate the ring fault rheology required to permit meter-per day creep in between collapse earthquakes.
Compared to earthquakes on tectonic faults, caldera collapse earthquakes are longer in duration (~ 10 s) and involves larger fault slip (> 2 m) for its magnitude (~ Mw 5). The answer to this puzzling question lies in the fact that, for caldera collapse earthquakes, the rupture phase (propagation of rupture on the ring fault) is followed by a much longer-duration collapse phase, (relatively uniform ring fault slip), the latter of which is characterized by the mechanical coupling between the caldera block and the underlying magma chamber. In this study, we quantify the effects of seismic wave radiation and magma viscosity on fault slip with 3D dynamic rupture simulations. We use the model to simulate the 2018 Kilauea caldera collapse. Three stages of collapse, characterized by ring fault rupture initiation and propagation, deceleration of the downward-moving caldera block and magma column, and post-collapse resonant oscillations, in addition to chamber pressurization, are identified in simulated and observed (unfiltered) near-field seismograms. We interpret each stage of collapse in terms of equivalent seismic force and moment. A detailed comparison between simulated and observed displacement waveforms reveals a complex nucleation phase for earthquakes initiated on the northwest (Invited talk at USGS Earthquake Seminar; Published on JGR: Solid Earth Wang et al., 2024 ).
2. Hindcasting aseismic slip at enhanced geothermal systems
Aseismic fault slip during fluid injection has only in recent decades been recognized. In reservoir stimulation operations, natural fault systems act as fluid conduits and can slip aseismically when fluid injection reduces effective normal stress. Using a customized 3D rate-and-state fault model, we constrain the amount of aseismic slip that occurred at Cooper Basin Enhanced Geothermal project and distinguish between seismicity loaded by fluid injection-resulted pressure change versus aseismic slip on a major fault zone (Wang et al., 2022) .
3. Constrain subsurface magma transport and magma reservoir geometries using geodesy and Neural Network based emulators
It is increasingly recognized that magma reservoirs are not isolated, melt dominated region in the crust. Rather, magma reservoirs are comprised of semi-continuous magma storage regions distributed over depths ("transcrustal magmatic systems"). The caldera collapse of Kīlauea volcano in 2018 results in a large pressure deficit in its magmatic reservoirs, and provided an unprecedented opportunity to constrain the geometry of and connectivity between the reservoirs (Wang et al., 2021 ). These constraints on reservoirs can potentially enhance our ability to estimate, in real time, the pressure evolution inside the magma chambers, and forecast eruptions (Segall et al., 2022 ).
A main challenge in quantitatively constraining magma reservoir geometry is the high computational cost for computing surface deformations, especially when the magma reservoir geometry is complex and when Bayesian inversions are preferred. To tackle this problem, Ian McBrearty and I developed Graph Neural Network (GNN) based emulators to compute surface deformation for arbitrarily complex reservoir geometries, which are parameterized with spherical harmonics (Wang et al., 2025 ). We show that GNN emulators (illustrated in figure below), once trained, provide relatively accurate surface deformation predictions (given complex reservoir geometry), for a computational cost three orders of magnitude lower than the benchmark numerical method.
Page updated
Google Sites
Report abuse