Jul 2025 · Mw 8.8
2025 Kamchatka megathrust earthquake
A magnitude 8.8 earthquake struck off Kamchatka about 50 km from the start of the 1952 magnitude 9.0 earthquake. Combining global seismic waves with tsunami records, PhD student Junpeng Li and I found that most slip was at moderate to greater depth in the southwest, where few earthquakes had occurred since 1952. Peak slip of about 14 m exceeds the roughly 6 m of slip deficit built up since 1952, so the 2025 earthquake released a deeper patch that likely stayed locked in 1952.
Li, J., & Jia, Z. (2026), Geophysical Research Letters. Paper
Sep 2022 · Mw 6.6
2022 Luding, China, earthquake
Published depths of the deadliest earthquake in mainland China in eight years ranged from 2 to 18.4 km. Regional data favor about 6 km and teleseismic data about 17 km. Both fit a two-subevent model: a larger shallow subevent at about 6 km and, 7 seconds later, a smaller one at about 17 km.
Han, H., Ni, S., Meng, Q., Gu, W., Jia, Z., et al. (2026), Bulletin of the Seismological Society of America. Paper
3D geodynamic model
Slab dynamics and earthquakes in Izu-Bonin-Mariana
A 3D model of the sinking Pacific slab predicts temperature, stress, and energy dissipation. Earthquakes occur mostly in rock cooler than about 900 °C, where dissipation is high, and the modeled stresses explain megathrust, outer-rise, and deep bending earthquakes. Shallow events favor a weak slab and deep events a stronger slab with a 10 to 100 fold viscosity increase at 660 km.
Li, Y., Ribe, N. M., & Jia, Z. (2026), Earth and Planetary Science Letters. Paper
Tonga–Fiji · mantle flow models
The Tonga slab and a relic slab
About two-thirds of the world's deep earthquakes occur in the Tonga slab. Mantle flow models that follow the region's plate history show a relic slab from the Vanuatu trench colliding with the Tonga slab, folding over it, and steepening it, which reproduces the stresses seen in deep earthquakes. The models imply that the 2018 Fiji doublet broke warm slab rock, which supports local slab temperature as the control on deep ruptures.
Liu, H., Gurnis, M., Leng, W., Jia, Z., & Zhan, Z. (2021), Geophysical Research Letters. Paper · PDF
Mar–Aug 2011 · Mw 3.9–4.8
2011 Hawthorne, Nevada, earthquake swarm
A swarm of 71 earthquakes shook an area near a Navy base and young volcanic features. Waveform modeling put the nine largest at 2 to 4 km depth with normal faulting, and satellite radar showed up to 10 cm of slip at 1 to 2 km depth. The radar implies about 15% more slip than the earthquakes released, and the swarm likely came from slow fault slip under regional stretching, with no sign of volcanic activity.
Zha, X., Jia, Z., Dai, Z., & Lu, Z. (2019), Journal of Geodesy. Paper · PDF
U-Net · Long Beach dense array
Deep learning for ambient-noise dispersion curves
Dense arrays produce thousands of dispersion images that analysts trace by hand. A U-Net trained on 100,000 synthetic curves and fine-tuned on 1,000 hand-labeled examples labels every pixel as fundamental mode, first overtone, or noise. On 4,340 held-out examples its picks match a human analyst's, with median precision of at least 98% and recall of at least 94%, and it cuts human labor by about 80%.
Zhang, X., Jia, Z., Ross, Z. E., & Clayton, R. W. (2020), IEEE Transactions on Geoscience and Remote Sensing. Paper · PDF
2008 Wells, Nevada · Mw 6.0
Earthquake depth from sparse stations
With few nearby stations, waveform fits can match almost equally well over a wide range of depths. Rayleigh waves recorded 5 to 15 degrees away have spectral dips that shift with source depth. Fitting them together with local waveforms gave depths of 8 km for the 2008 Wells, Nevada, earthquake and 32 km for a 2013 outer-rise earthquake off Japan, both confirmed by teleseismic depth phases.
Jia, Z., Ni, S., Chu, R., & Zhan, Z. (2017), Pure and Applied Geophysics. Paper · PDF
Software · local and teleseismic waveforms
gCAPjoint full moment tensor software
About 10% of earthquakes in the Global CMT catalog have large non-double-couple components. gCAPjoint solves for the full moment tensor and depth by fitting local and teleseismic waveforms together. Tests on the 2017 Jiuzhaigou, 2010 Jiashian, and 2016 Pamir earthquakes gave a shallow 6 km source for Jiuzhaigou with a non-double-couple share of up to 25%.
Bai, Q., Ni, S., Chu, R., & Jia, Z. (2020), Seismological Research Letters. Paper · PDF
Software · moderate earthquakes
CAPjoint source inversion software
CAPjoint fits local and teleseismic waveforms together to find the fault mechanism, depth, and magnitude of moderate earthquakes recorded by few stations. For the 2008 Wells, Nevada, earthquake it found Mw 6.06 at 8.6 km depth. Adding a single local station to four teleseismic stations brought the result close to the solution from many stations.
Chen, W., Ni, S., Kanamori, H., Wei, S., Jia, Z., & Zhu, L. (2015), Seismological Research Letters. Paper · PDF
Apr 2013 · Ms 7.0 · Sichuan
Early aftershocks of the 2013 Lushan earthquake
Catalogs miss many aftershocks in the first hours after a large earthquake. Detecting them in 24 hours of continuous data from one station 30 km away found 734 aftershocks, against 372 in the national catalog. The aftershock rate stayed nearly flat for about 8,000 seconds before the usual Omori decay began.
Jia, Z., Chen, W., & Chu, R. (2013), Earthquake Science. Paper · PDF