Back to Calendar
Dr. Karianne Bergen, Harvard Data Science Initiative Fellow at Harvard U., presents “Big data for small earthquakes: a data mining approach to large-scale earthquake detection” at the MIT Earth Resources Laboratory on September 28, 2018.
Live video of this talk will be available on our Youtube channel.
“Earthquake detection, the problem of extracting weak earthquake signals from continuous waveform data recorded by sensors in a seismic network, is a critical and challenging task in seismology. New algorithmic advances in “big data” and artificial intelligence have created opportunities to advance the state-of-the-art in earthquake detection algorithms. In this talk, I will present Fingerprint and Similarity Thresholding (FAST; Yoon et al, 2015), a data mining approach to large-scale earthquake detection, inspired by technology for rapid audio identification. FAST leverages locality sensitive hashing (LSH), a technique for efficiently identifying similar items in large data sets, to detect new candidate earthquakes without template waveforms (“training data”). I will present recent algorithmic extensions to FAST that enable detection over a seismic network and limit false detections due to local correlated noise (Bergen & Beroza, 2018). Using the foreshock sequence prior to the 2014 Mw 8.2 Iquique earthquake as a test case, we demonstrate that our approach is sensitive and maintains a low false detections rate, identifying five times as many events as the local seismicity catalog with a false discovery rate of less than 1%. We show that our new optimized FAST software is capable of discovering new events with unknown sources in 10 years of continuous data (Rong et al, 2018). I will end the talk with recommendations, based on our experience developing the FAST detector, for how the solid Earth geoscience community can leverage machine learning and data mining to enable data-driven discovery. “