Regional Earthquake Monitoring with Deep Learning
Albert Leonardo Aguilar Suarez
Stanford University
- Date & Time
- Location
- Hybrid In-Person and Online seminar via Microsoft Teams
- Host
- Justin Rubinstein
- Summary
Recent developments in machine learning and deep learning for observational seismology, the basis for new boosted earthquake catalogs have been focused on small earthquakes recorded at short local distances, mainly in densely instrumented regions. However, most of the world is sparsely instrumented, making earthquake monitoring rely on recordings at regional distances ( beyond ~100 km). I will describe the process of assembling the Curated Regional Earthquake Waveforms (CREW) dataset of earthquake waveforms and labeled arrivals between 1 and 20 degrees of source to receiver distance through semi-supervised machine learning . This dataset has been leveraged to train deep learning models for the task of seismic phase picking, helping find more earthquakes than previously reported in regions far from instrumentation, like the Gorda Ridge in offshore Northern California, where the closest land seismometer is 300 km away. Furthermore, these models are powering our new earthquake catalog for Colombia, where average inter station spacing is 98 km, compared to 16 km in California. This catalog consists of over 1 million earthquakes, 4.5 times more than the Colombian Geological Survey routine catalog.