Deep learning forecasting of induced earthquakes through the analysis of precursory signals

The current limited knowledge about Earth system prevents deterministic earthquake prediction. This will probably continue for the foreseeable future. However, the improved capability of identifying reliable precursory phenomena can allow geoscientists to comprehend if the monitored system is evolving toward an unstable state. Among the premonitory phenomena preceding earthquakes, foreshocks represent the most promising candidate. Physically, two hand-member mechanisms have been proposed to interpret foreshocks. The first considers the failing of populations of small patches of fault that eventually but not necessarily become large earthquakes whereas the second assumes that foreshocks are a part of the nucleation process which ultimately leads to the mainshock. The prompt identification of foreshocks with respect to background seismicity is an issue and the task is worsened when dealing with low-magnitude earthquakes. However, the use of Artificial Intelligence (AI) can help real-time seismology to effectively discriminate precursory signals.In the present study, we propose a deep learning method named PreD-Net (Precursor Detection Network) to address the precursory signal identification of induced earthquakes through the analysis of several statistical features. PreD-Net has been trained on data related to two induced seismicity areas, namely The Geysers, located in California, USA, and Hengill in Iceland. Notably, the network shows a suitable model generalization, providing considerable results on samples that were excluded from the training dataset of all the sites. The performed tests on related samples of induced relatively large events demonstrate the possibility of setting up a real-time warning strategy to be used to avoid adverse consequences during field operations.This work is supported by project D.I.R.E.C.T.I.O.N.S. - Deep learning aIded foReshock deteCTIOn Of iNduced mainShocks, project code: P20229KB4F - - Next Generation EU (PRIN-PNRR 2022, CUP D53D23022800001)