AI-assisted assessment of low-magnitude seismicity in the area of Kavala-Prinos (Greece)

For the transition to a low-carbon future, carbon capture and storage (CCS) is a field of intense research worldwide. However, the process needs to be monitored closely for induced seismicity, and this in turn requires a clear picture of the background seismicity of the immediate area around the storage site. This study focuses on assessing the background seismicity of the Gulf of Kavala in Greece, where an offshore CCS pilot is deployed within EU project COREu. However, a major challenge associated with this area is the scarcity of catalogued events due to its low seismicity, as well as the sparse seismic station distribution, owing also to geography. To overcome this challenge, in this study, state-of-the-art AI-based seismic detectors (EQTransformer and PhaseNet) are used to re-evaluate existing recordings and enrich the area’s catalogue with low-magnitude events. A two-stage approach is considered to take into account and resolve the particularities of this involved task. In the first stage, we evaluate the baseline performance of the adopted pre-trained AI-based detectors, using data from the Corinth Gulf area, selected because the seismic network there is significantly denser and the seismicity higher. Specifically, for our purposes, we used data from a microseismic sequence of more than 400 events recorded in the first half of 2021, with magnitudes ranging from 0.1 to 1. In the experiment we utilize recordings of the selected events from a total of 50 stations located around the seismic sequence, with distances out to several tens of kilometers, to build a dataset with a wide and representative range of recording SNRs. To assess the detectability of the events, for each event/station pair, we measure the output of the detectors in a time-window of 5 seconds around the event arrival, forming a (detector) response magnitude vs SNR curve. This is used as a guideline for determining a detection threshold that strikes a good balance between true and false positives. Through this successful application of the method in the Corinth Gulf area, we gained significant knowledge about the limitations and the necessary configuration of the methods. In the second stage of the study, we conduct a preliminary detection experiment on continuous recordings from Prinos, utilizing data from stations surrounding the target area. The outcome of this experiment is evaluated by expert seismologists, using a specially created visualization tool for assisting the evaluation process. The adopted two-stage approach leads to the detection of a considerable number of low-magnitude, previously undetected events, constituting a significant first step towards assessing the implementation of CCS  in the Prinos area.