Induced seismicity associated with shale-gas development has become an increasing concern for seismic hazard assessment in Southwest China. This study develops a machine-learning–based framework to distinguish natural earthquakes from shale-gas-induced seismicity in Chongqing using seismic waveform records archived in the JOPENS system. A suite of physically interpretable spectral and time-domain features was extracted from preprocessed P- and S-wave windows, with emphasis on frequency-dependent characteristics and P/S spectral amplitude ratios. Induced earthquakes were identified using independent time-based labeling and subsequently evaluated through spatial proximity analysis and waveform characterization. The results show that induced events exhibit pronounced high-frequency spectral enrichment and strong spatial clustering around hydraulic fracturing platforms, with most events occurring within approximately 3 km and a median distance of 2.1 km from the nearest platform. An optimized Support Vector Machine classifier achieves an overall accuracy of 0.96 on independent test data, correctly identifying 15 of 16 induced earthquakes and all 12 natural earthquakes, with precision and recall exceeding 0.92 for both classes. These findings demonstrate that combining physically interpretable waveform features with spatial constraints provides a robust and transparent approach for distinguishing induced from natural seismicity, offering practical value for real-time monitoring and seismic hazard mitigation in shale-gas development regions.
Zhou et al. (2026) studied this question.