Automatic detection and classification of Nanoseismicity in Distributed Acoustic Sensing data

The detection of nanoseismicity (very tiny earthquakes sometimes associated with small cracks in rock, also called acoustic emissions) is an important area of research aiding in the understanding of geophysical processes, hazard detection, material failure and human-driven nanoseismicity. The high frequency and attenuation of nanoseismicity require high-frequency monitoring within metres of the source to capture the event. This has made them difficult to monitor in conditions outside of small-scale lab experiments, in which failure is intentionally induced. The development of distributed acoustic sensing (DAS) as a new tool for seismic monitoring, however, has increased the feasibility of investigating such signals in the field due to its high temporal and spatial resolution. Manual picking of these events, while possible, is impractical for long-term deployments and for time-critical applications such as stability monitoring, which limits the utility of the technology. Automation of the detection of nanoseismic events within such data is therefore essential for the long-term processing of DAS data and real-time processing of data for use in stability monitoring.  

We have developed a pipeline for the automated extraction of nanoseismic events from DAS data, using a new, simple ratio technique called Spatial Short-Term Average (SSTA). The pipeline takes an input of DAS data and generates a series of windows within the data containing information about high amplitude signals relating to nanoseismicity.  

Using the automatically detected events, we labelled the windows to train a series of machine learning models to classify the different signals. Once trained, we evaluated the performance of the various models to select the most effective method for processing the collected data. The best performing models will then be tested at scale with the resulting classified dataset being plotted spatially along the length of the deployment to identify patterns of activity across space and time.