---

Detection and waveform-based source imaging of small-magnitude events using unsupervised machine learning and grouped time-reversals.

Dalai B, Kumar P.

ACTA GEOPHYSICA

https://doi.org/10.1007/s11600-025-01779-z 

Accurate location of small magnitude earthquakes is essential for understanding fault activity and assessing seismic hazards, yet it remains challenging due to weak signals, noise, and sparse networks. A framework has been developed that combines unsupervised machine learning with waveform-based, time-reversal (RTM-inspired) imaging to address these limitations. The approach detects weak arrivals directly from data and images seismic sources using waveform coherence, resulting in improved depth and location resolution and reduced uncertainty compared to the conventional travel-time–based inversion methods in complex tectonic regions.

Fig: Field data from the NW- Himalayan region. The velocity model (a) of the region is derived by the ray tracing-based inversion approach using the local seismicity. The vertical component of the waveforms is retrieved for the seismic stations (gray inverted triangle) shown on the top of subplot (b). The gray portion on the waveforms indicates the signal of the P waveforms that are used as input for grouped time-reversal imaging. The focused source image as derived using the present methodology is depicted in (c). For this combination, the derived source depth is 5.85 ± 0.5 km, whereas the traditional travel-time inversion approach yields the source depth of 6.5±3.5 km (P and S) and 6.7±3.8 km (P-only).

---